1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 132 "struct pf_krule *", "struct mbuf *", "int"); 133 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 134 "struct pf_sctp_source *"); 135 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 136 "struct pf_kstate *", "struct pf_sctp_source *"); 137 138 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 139 "struct mbuf *"); 140 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 141 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 142 "int", "struct pf_keth_rule *", "char *"); 143 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 145 "int", "struct pf_keth_rule *"); 146 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 147 148 /* 149 * Global variables 150 */ 151 152 /* state tables */ 153 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 154 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 157 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 158 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 159 VNET_DEFINE(struct pf_kstatus, pf_status); 160 161 VNET_DEFINE(u_int32_t, ticket_altqs_active); 162 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 163 VNET_DEFINE(int, altqs_inactive_open); 164 VNET_DEFINE(u_int32_t, ticket_pabuf); 165 166 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 167 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 168 VNET_DEFINE(u_char, pf_tcp_secret[16]); 169 #define V_pf_tcp_secret VNET(pf_tcp_secret) 170 VNET_DEFINE(int, pf_tcp_secret_init); 171 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 172 VNET_DEFINE(int, pf_tcp_iss_off); 173 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 174 VNET_DECLARE(int, pf_vnet_active); 175 #define V_pf_vnet_active VNET(pf_vnet_active) 176 177 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 178 #define V_pf_purge_idx VNET(pf_purge_idx) 179 180 #ifdef PF_WANT_32_TO_64_COUNTER 181 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 182 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 183 184 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 185 VNET_DEFINE(size_t, pf_allrulecount); 186 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 187 #endif 188 189 struct pf_sctp_endpoint; 190 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 191 struct pf_sctp_source { 192 sa_family_t af; 193 struct pf_addr addr; 194 TAILQ_ENTRY(pf_sctp_source) entry; 195 }; 196 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 197 struct pf_sctp_endpoint 198 { 199 uint32_t v_tag; 200 struct pf_sctp_sources sources; 201 RB_ENTRY(pf_sctp_endpoint) entry; 202 }; 203 static int 204 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 205 { 206 return (a->v_tag - b->v_tag); 207 } 208 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 209 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 210 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 211 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 212 static struct mtx_padalign pf_sctp_endpoints_mtx; 213 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 214 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 215 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 216 217 /* 218 * Queue for pf_intr() sends. 219 */ 220 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 221 struct pf_send_entry { 222 STAILQ_ENTRY(pf_send_entry) pfse_next; 223 struct mbuf *pfse_m; 224 enum { 225 PFSE_IP, 226 PFSE_IP6, 227 PFSE_ICMP, 228 PFSE_ICMP6, 229 } pfse_type; 230 struct { 231 int type; 232 int code; 233 int mtu; 234 } icmpopts; 235 }; 236 237 STAILQ_HEAD(pf_send_head, pf_send_entry); 238 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 239 #define V_pf_sendqueue VNET(pf_sendqueue) 240 241 static struct mtx_padalign pf_sendqueue_mtx; 242 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 243 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 244 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 245 246 /* 247 * Queue for pf_overload_task() tasks. 248 */ 249 struct pf_overload_entry { 250 SLIST_ENTRY(pf_overload_entry) next; 251 struct pf_addr addr; 252 sa_family_t af; 253 uint8_t dir; 254 struct pf_krule *rule; 255 }; 256 257 SLIST_HEAD(pf_overload_head, pf_overload_entry); 258 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 259 #define V_pf_overloadqueue VNET(pf_overloadqueue) 260 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 261 #define V_pf_overloadtask VNET(pf_overloadtask) 262 263 static struct mtx_padalign pf_overloadqueue_mtx; 264 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 265 "pf overload/flush queue", MTX_DEF); 266 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 267 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 268 269 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 270 struct mtx_padalign pf_unlnkdrules_mtx; 271 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 272 MTX_DEF); 273 274 struct sx pf_config_lock; 275 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 276 277 struct mtx_padalign pf_table_stats_lock; 278 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 279 MTX_DEF); 280 281 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 282 #define V_pf_sources_z VNET(pf_sources_z) 283 uma_zone_t pf_mtag_z; 284 VNET_DEFINE(uma_zone_t, pf_state_z); 285 VNET_DEFINE(uma_zone_t, pf_state_key_z); 286 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 287 288 VNET_DEFINE(struct unrhdr64, pf_stateid); 289 290 static void pf_src_tree_remove_state(struct pf_kstate *); 291 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 292 u_int32_t); 293 static void pf_add_threshold(struct pf_threshold *); 294 static int pf_check_threshold(struct pf_threshold *); 295 296 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 297 u_int16_t *, u_int16_t *, struct pf_addr *, 298 u_int16_t, u_int8_t, sa_family_t); 299 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 300 struct tcphdr *, struct pf_state_peer *); 301 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 302 int *, u_int16_t *, u_int16_t *); 303 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 304 struct pf_addr *, struct pf_addr *, u_int16_t, 305 u_int16_t *, u_int16_t *, u_int16_t *, 306 u_int16_t *, u_int8_t, sa_family_t); 307 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 308 sa_family_t, struct pf_krule *, int); 309 static void pf_detach_state(struct pf_kstate *); 310 static int pf_state_key_attach(struct pf_state_key *, 311 struct pf_state_key *, struct pf_kstate *); 312 static void pf_state_key_detach(struct pf_kstate *, int); 313 static int pf_state_key_ctor(void *, int, void *, int); 314 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 315 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 316 struct pf_mtag *pf_mtag); 317 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 318 struct pf_krule *, struct mbuf **); 319 static int pf_dummynet_route(struct pf_pdesc *, 320 struct pf_kstate *, struct pf_krule *, 321 struct ifnet *, struct sockaddr *, struct mbuf **); 322 static int pf_test_eth_rule(int, struct pfi_kkif *, 323 struct mbuf **); 324 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 325 struct pfi_kkif *, struct mbuf *, int, 326 struct pf_pdesc *, struct pf_krule **, 327 struct pf_kruleset **, struct inpcb *, int); 328 static int pf_create_state(struct pf_krule *, struct pf_krule *, 329 struct pf_krule *, struct pf_pdesc *, 330 struct pf_ksrc_node *, struct pf_state_key *, 331 struct pf_state_key *, struct mbuf *, int, 332 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 333 struct pf_kstate **, int, u_int16_t, u_int16_t, 334 int, struct pf_krule_slist *, struct pf_udp_mapping *); 335 static int pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *, 336 int, struct pf_state_key_cmp *, int, struct pf_addr *, 337 int, struct pf_addr *, int); 338 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 339 struct mbuf *, void *, struct pf_pdesc *, 340 struct pf_krule **, struct pf_kruleset **); 341 static int pf_tcp_track_full(struct pf_kstate **, 342 struct pfi_kkif *, struct mbuf *, int, 343 struct pf_pdesc *, u_short *, int *); 344 static int pf_tcp_track_sloppy(struct pf_kstate **, 345 struct pf_pdesc *, u_short *); 346 static int pf_test_state_tcp(struct pf_kstate **, 347 struct pfi_kkif *, struct mbuf *, int, 348 void *, struct pf_pdesc *, u_short *); 349 static int pf_test_state_udp(struct pf_kstate **, 350 struct pfi_kkif *, struct mbuf *, int, 351 void *, struct pf_pdesc *); 352 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 353 struct pf_pdesc *, struct pf_kstate **, struct mbuf *, 354 int, int, struct pfi_kkif *, u_int16_t, u_int16_t, 355 int, int *, int, int); 356 static int pf_test_state_icmp(struct pf_kstate **, 357 struct pfi_kkif *, struct mbuf *, int, 358 void *, struct pf_pdesc *, u_short *); 359 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 360 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 361 struct pfi_kkif *, struct pf_kstate *, int); 362 static int pf_test_state_sctp(struct pf_kstate **, 363 struct pfi_kkif *, struct mbuf *, int, 364 void *, struct pf_pdesc *, u_short *); 365 static int pf_test_state_other(struct pf_kstate **, 366 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 367 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 368 int, u_int16_t); 369 static int pf_check_proto_cksum(struct mbuf *, int, int, 370 u_int8_t, sa_family_t); 371 static void pf_print_state_parts(struct pf_kstate *, 372 struct pf_state_key *, struct pf_state_key *); 373 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 374 bool, u_int8_t); 375 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 376 const struct pf_state_key_cmp *, u_int); 377 static int pf_src_connlimit(struct pf_kstate **); 378 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 379 static void pf_counters_inc(int, 380 struct pf_pdesc *, struct pfi_kkif *, 381 struct pf_kstate *, struct pf_krule *, 382 struct pf_krule *); 383 static void pf_overload_task(void *v, int pending); 384 static u_short pf_insert_src_node(struct pf_ksrc_node **, 385 struct pf_krule *, struct pf_addr *, sa_family_t); 386 static u_int pf_purge_expired_states(u_int, int); 387 static void pf_purge_unlinked_rules(void); 388 static int pf_mtag_uminit(void *, int, int); 389 static void pf_mtag_free(struct m_tag *); 390 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 391 int, struct pf_state_key *); 392 #ifdef INET 393 static void pf_route(struct mbuf **, struct pf_krule *, 394 struct ifnet *, struct pf_kstate *, 395 struct pf_pdesc *, struct inpcb *); 396 #endif /* INET */ 397 #ifdef INET6 398 static void pf_change_a6(struct pf_addr *, u_int16_t *, 399 struct pf_addr *, u_int8_t); 400 static void pf_route6(struct mbuf **, struct pf_krule *, 401 struct ifnet *, struct pf_kstate *, 402 struct pf_pdesc *, struct inpcb *); 403 #endif /* INET6 */ 404 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 405 406 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 407 408 extern int pf_end_threads; 409 extern struct proc *pf_purge_proc; 410 411 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 412 413 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 414 415 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 416 do { \ 417 struct pf_state_key *nk; \ 418 if ((pd->dir) == PF_OUT) \ 419 nk = (_s)->key[PF_SK_STACK]; \ 420 else \ 421 nk = (_s)->key[PF_SK_WIRE]; \ 422 pf_packet_rework_nat(_m, _pd, _off, nk); \ 423 } while (0) 424 425 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 426 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 427 428 #define STATE_LOOKUP(i, k, s, pd) \ 429 do { \ 430 (s) = pf_find_state((i), (k), (pd->dir)); \ 431 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 432 if ((s) == NULL) \ 433 return (PF_DROP); \ 434 if (PACKET_LOOPED(pd)) \ 435 return (PF_PASS); \ 436 } while (0) 437 438 static struct pfi_kkif * 439 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 440 { 441 SDT_PROBE2(pf, ip, , bound_iface, st, k); 442 443 /* Floating unless otherwise specified. */ 444 if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND)) 445 return (V_pfi_all); 446 447 /* 448 * Initially set to all, because we don't know what interface we'll be 449 * sending this out when we create the state. 450 */ 451 if (st->rule.ptr->rt == PF_REPLYTO) 452 return (V_pfi_all); 453 454 /* Don't overrule the interface for states created on incoming packets. */ 455 if (st->direction == PF_IN) 456 return (k); 457 458 /* No route-to, so don't overrule. */ 459 if (st->rt != PF_ROUTETO) 460 return (k); 461 462 /* Bind to the route-to interface. */ 463 return (st->rt_kif); 464 } 465 466 #define STATE_INC_COUNTERS(s) \ 467 do { \ 468 struct pf_krule_item *mrm; \ 469 counter_u64_add(s->rule.ptr->states_cur, 1); \ 470 counter_u64_add(s->rule.ptr->states_tot, 1); \ 471 if (s->anchor.ptr != NULL) { \ 472 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 473 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 474 } \ 475 if (s->nat_rule.ptr != NULL) { \ 476 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 477 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 478 } \ 479 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 480 counter_u64_add(mrm->r->states_cur, 1); \ 481 counter_u64_add(mrm->r->states_tot, 1); \ 482 } \ 483 } while (0) 484 485 #define STATE_DEC_COUNTERS(s) \ 486 do { \ 487 struct pf_krule_item *mrm; \ 488 if (s->nat_rule.ptr != NULL) \ 489 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 490 if (s->anchor.ptr != NULL) \ 491 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 492 counter_u64_add(s->rule.ptr->states_cur, -1); \ 493 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 494 counter_u64_add(mrm->r->states_cur, -1); \ 495 } while (0) 496 497 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 498 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 499 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 500 VNET_DEFINE(struct pf_idhash *, pf_idhash); 501 VNET_DEFINE(struct pf_srchash *, pf_srchash); 502 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 503 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 504 505 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 506 "pf(4)"); 507 508 VNET_DEFINE(u_long, pf_hashmask); 509 VNET_DEFINE(u_long, pf_srchashmask); 510 VNET_DEFINE(u_long, pf_udpendpointhashmask); 511 VNET_DEFINE_STATIC(u_long, pf_hashsize); 512 #define V_pf_hashsize VNET(pf_hashsize) 513 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 514 #define V_pf_srchashsize VNET(pf_srchashsize) 515 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 516 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 517 u_long pf_ioctl_maxcount = 65535; 518 519 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 520 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 521 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 522 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 523 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 524 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 525 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 526 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 527 528 VNET_DEFINE(void *, pf_swi_cookie); 529 VNET_DEFINE(struct intr_event *, pf_swi_ie); 530 531 VNET_DEFINE(uint32_t, pf_hashseed); 532 #define V_pf_hashseed VNET(pf_hashseed) 533 534 static void 535 pf_sctp_checksum(struct mbuf *m, int off) 536 { 537 uint32_t sum = 0; 538 539 /* Zero out the checksum, to enable recalculation. */ 540 m_copyback(m, off + offsetof(struct sctphdr, checksum), 541 sizeof(sum), (caddr_t)&sum); 542 543 sum = sctp_calculate_cksum(m, off); 544 545 m_copyback(m, off + offsetof(struct sctphdr, checksum), 546 sizeof(sum), (caddr_t)&sum); 547 } 548 549 int 550 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 551 { 552 553 switch (af) { 554 #ifdef INET 555 case AF_INET: 556 if (a->addr32[0] > b->addr32[0]) 557 return (1); 558 if (a->addr32[0] < b->addr32[0]) 559 return (-1); 560 break; 561 #endif /* INET */ 562 #ifdef INET6 563 case AF_INET6: 564 if (a->addr32[3] > b->addr32[3]) 565 return (1); 566 if (a->addr32[3] < b->addr32[3]) 567 return (-1); 568 if (a->addr32[2] > b->addr32[2]) 569 return (1); 570 if (a->addr32[2] < b->addr32[2]) 571 return (-1); 572 if (a->addr32[1] > b->addr32[1]) 573 return (1); 574 if (a->addr32[1] < b->addr32[1]) 575 return (-1); 576 if (a->addr32[0] > b->addr32[0]) 577 return (1); 578 if (a->addr32[0] < b->addr32[0]) 579 return (-1); 580 break; 581 #endif /* INET6 */ 582 default: 583 panic("%s: unknown address family %u", __func__, af); 584 } 585 return (0); 586 } 587 588 static void 589 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 590 struct pf_state_key *nk) 591 { 592 593 switch (pd->proto) { 594 case IPPROTO_TCP: { 595 struct tcphdr *th = &pd->hdr.tcp; 596 597 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 598 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 599 &th->th_sum, &nk->addr[pd->sidx], 600 nk->port[pd->sidx], 0, pd->af); 601 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 602 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 603 &th->th_sum, &nk->addr[pd->didx], 604 nk->port[pd->didx], 0, pd->af); 605 m_copyback(m, off, sizeof(*th), (caddr_t)th); 606 break; 607 } 608 case IPPROTO_UDP: { 609 struct udphdr *uh = &pd->hdr.udp; 610 611 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 612 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 613 &uh->uh_sum, &nk->addr[pd->sidx], 614 nk->port[pd->sidx], 1, pd->af); 615 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 616 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 617 &uh->uh_sum, &nk->addr[pd->didx], 618 nk->port[pd->didx], 1, pd->af); 619 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 620 break; 621 } 622 case IPPROTO_SCTP: { 623 struct sctphdr *sh = &pd->hdr.sctp; 624 uint16_t checksum = 0; 625 626 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 627 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 628 &checksum, &nk->addr[pd->sidx], 629 nk->port[pd->sidx], 1, pd->af); 630 } 631 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 632 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 633 &checksum, &nk->addr[pd->didx], 634 nk->port[pd->didx], 1, pd->af); 635 } 636 637 break; 638 } 639 case IPPROTO_ICMP: { 640 struct icmp *ih = &pd->hdr.icmp; 641 642 if (nk->port[pd->sidx] != ih->icmp_id) { 643 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 644 ih->icmp_cksum, ih->icmp_id, 645 nk->port[pd->sidx], 0); 646 ih->icmp_id = nk->port[pd->sidx]; 647 pd->sport = &ih->icmp_id; 648 649 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 650 } 651 /* FALLTHROUGH */ 652 } 653 default: 654 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 655 switch (pd->af) { 656 case AF_INET: 657 pf_change_a(&pd->src->v4.s_addr, 658 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 659 0); 660 break; 661 case AF_INET6: 662 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 663 break; 664 } 665 } 666 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 667 switch (pd->af) { 668 case AF_INET: 669 pf_change_a(&pd->dst->v4.s_addr, 670 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 671 0); 672 break; 673 case AF_INET6: 674 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 675 break; 676 } 677 } 678 break; 679 } 680 } 681 682 static __inline uint32_t 683 pf_hashkey(const struct pf_state_key *sk) 684 { 685 uint32_t h; 686 687 h = murmur3_32_hash32((const uint32_t *)sk, 688 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 689 V_pf_hashseed); 690 691 return (h & V_pf_hashmask); 692 } 693 694 static __inline uint32_t 695 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 696 { 697 uint32_t h; 698 699 switch (af) { 700 case AF_INET: 701 h = murmur3_32_hash32((uint32_t *)&addr->v4, 702 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 703 break; 704 case AF_INET6: 705 h = murmur3_32_hash32((uint32_t *)&addr->v6, 706 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 707 break; 708 default: 709 panic("%s: unknown address family %u", __func__, af); 710 } 711 712 return (h & V_pf_srchashmask); 713 } 714 715 static inline uint32_t 716 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 717 { 718 uint32_t h; 719 720 h = murmur3_32_hash32((uint32_t *)endpoint, 721 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 722 V_pf_hashseed); 723 return (h & V_pf_udpendpointhashmask); 724 } 725 726 #ifdef ALTQ 727 static int 728 pf_state_hash(struct pf_kstate *s) 729 { 730 u_int32_t hv = (intptr_t)s / sizeof(*s); 731 732 hv ^= crc32(&s->src, sizeof(s->src)); 733 hv ^= crc32(&s->dst, sizeof(s->dst)); 734 if (hv == 0) 735 hv = 1; 736 return (hv); 737 } 738 #endif 739 740 static __inline void 741 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 742 { 743 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 744 s->dst.state = newstate; 745 if (which == PF_PEER_DST) 746 return; 747 if (s->src.state == newstate) 748 return; 749 if (s->creatorid == V_pf_status.hostid && 750 s->key[PF_SK_STACK] != NULL && 751 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 752 !(TCPS_HAVEESTABLISHED(s->src.state) || 753 s->src.state == TCPS_CLOSED) && 754 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 755 atomic_add_32(&V_pf_status.states_halfopen, -1); 756 757 s->src.state = newstate; 758 } 759 760 #ifdef INET6 761 void 762 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 763 { 764 switch (af) { 765 #ifdef INET 766 case AF_INET: 767 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 768 break; 769 #endif /* INET */ 770 case AF_INET6: 771 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 772 break; 773 } 774 } 775 #endif /* INET6 */ 776 777 static void 778 pf_init_threshold(struct pf_threshold *threshold, 779 u_int32_t limit, u_int32_t seconds) 780 { 781 threshold->limit = limit * PF_THRESHOLD_MULT; 782 threshold->seconds = seconds; 783 threshold->count = 0; 784 threshold->last = time_uptime; 785 } 786 787 static void 788 pf_add_threshold(struct pf_threshold *threshold) 789 { 790 u_int32_t t = time_uptime, diff = t - threshold->last; 791 792 if (diff >= threshold->seconds) 793 threshold->count = 0; 794 else 795 threshold->count -= threshold->count * diff / 796 threshold->seconds; 797 threshold->count += PF_THRESHOLD_MULT; 798 threshold->last = t; 799 } 800 801 static int 802 pf_check_threshold(struct pf_threshold *threshold) 803 { 804 return (threshold->count > threshold->limit); 805 } 806 807 static int 808 pf_src_connlimit(struct pf_kstate **state) 809 { 810 struct pf_overload_entry *pfoe; 811 int bad = 0; 812 813 PF_STATE_LOCK_ASSERT(*state); 814 /* 815 * XXXKS: The src node is accessed unlocked! 816 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 817 */ 818 819 (*state)->src_node->conn++; 820 (*state)->src.tcp_est = 1; 821 pf_add_threshold(&(*state)->src_node->conn_rate); 822 823 if ((*state)->rule.ptr->max_src_conn && 824 (*state)->rule.ptr->max_src_conn < 825 (*state)->src_node->conn) { 826 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 827 bad++; 828 } 829 830 if ((*state)->rule.ptr->max_src_conn_rate.limit && 831 pf_check_threshold(&(*state)->src_node->conn_rate)) { 832 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 833 bad++; 834 } 835 836 if (!bad) 837 return (0); 838 839 /* Kill this state. */ 840 (*state)->timeout = PFTM_PURGE; 841 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 842 843 if ((*state)->rule.ptr->overload_tbl == NULL) 844 return (1); 845 846 /* Schedule overloading and flushing task. */ 847 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 848 if (pfoe == NULL) 849 return (1); /* too bad :( */ 850 851 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 852 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 853 pfoe->rule = (*state)->rule.ptr; 854 pfoe->dir = (*state)->direction; 855 PF_OVERLOADQ_LOCK(); 856 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 857 PF_OVERLOADQ_UNLOCK(); 858 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 859 860 return (1); 861 } 862 863 static void 864 pf_overload_task(void *v, int pending) 865 { 866 struct pf_overload_head queue; 867 struct pfr_addr p; 868 struct pf_overload_entry *pfoe, *pfoe1; 869 uint32_t killed = 0; 870 871 CURVNET_SET((struct vnet *)v); 872 873 PF_OVERLOADQ_LOCK(); 874 queue = V_pf_overloadqueue; 875 SLIST_INIT(&V_pf_overloadqueue); 876 PF_OVERLOADQ_UNLOCK(); 877 878 bzero(&p, sizeof(p)); 879 SLIST_FOREACH(pfoe, &queue, next) { 880 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 881 if (V_pf_status.debug >= PF_DEBUG_MISC) { 882 printf("%s: blocking address ", __func__); 883 pf_print_host(&pfoe->addr, 0, pfoe->af); 884 printf("\n"); 885 } 886 887 p.pfra_af = pfoe->af; 888 switch (pfoe->af) { 889 #ifdef INET 890 case AF_INET: 891 p.pfra_net = 32; 892 p.pfra_ip4addr = pfoe->addr.v4; 893 break; 894 #endif 895 #ifdef INET6 896 case AF_INET6: 897 p.pfra_net = 128; 898 p.pfra_ip6addr = pfoe->addr.v6; 899 break; 900 #endif 901 } 902 903 PF_RULES_WLOCK(); 904 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 905 PF_RULES_WUNLOCK(); 906 } 907 908 /* 909 * Remove those entries, that don't need flushing. 910 */ 911 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 912 if (pfoe->rule->flush == 0) { 913 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 914 free(pfoe, M_PFTEMP); 915 } else 916 counter_u64_add( 917 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 918 919 /* If nothing to flush, return. */ 920 if (SLIST_EMPTY(&queue)) { 921 CURVNET_RESTORE(); 922 return; 923 } 924 925 for (int i = 0; i <= V_pf_hashmask; i++) { 926 struct pf_idhash *ih = &V_pf_idhash[i]; 927 struct pf_state_key *sk; 928 struct pf_kstate *s; 929 930 PF_HASHROW_LOCK(ih); 931 LIST_FOREACH(s, &ih->states, entry) { 932 sk = s->key[PF_SK_WIRE]; 933 SLIST_FOREACH(pfoe, &queue, next) 934 if (sk->af == pfoe->af && 935 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 936 pfoe->rule == s->rule.ptr) && 937 ((pfoe->dir == PF_OUT && 938 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 939 (pfoe->dir == PF_IN && 940 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 941 s->timeout = PFTM_PURGE; 942 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 943 killed++; 944 } 945 } 946 PF_HASHROW_UNLOCK(ih); 947 } 948 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 949 free(pfoe, M_PFTEMP); 950 if (V_pf_status.debug >= PF_DEBUG_MISC) 951 printf("%s: %u states killed", __func__, killed); 952 953 CURVNET_RESTORE(); 954 } 955 956 /* 957 * Can return locked on failure, so that we can consistently 958 * allocate and insert a new one. 959 */ 960 struct pf_ksrc_node * 961 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 962 struct pf_srchash **sh, bool returnlocked) 963 { 964 struct pf_ksrc_node *n; 965 966 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 967 968 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 969 PF_HASHROW_LOCK(*sh); 970 LIST_FOREACH(n, &(*sh)->nodes, entry) 971 if (n->rule.ptr == rule && n->af == af && 972 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 973 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 974 break; 975 976 if (n != NULL) { 977 n->states++; 978 PF_HASHROW_UNLOCK(*sh); 979 } else if (returnlocked == false) 980 PF_HASHROW_UNLOCK(*sh); 981 982 return (n); 983 } 984 985 static void 986 pf_free_src_node(struct pf_ksrc_node *sn) 987 { 988 989 for (int i = 0; i < 2; i++) { 990 counter_u64_free(sn->bytes[i]); 991 counter_u64_free(sn->packets[i]); 992 } 993 uma_zfree(V_pf_sources_z, sn); 994 } 995 996 static u_short 997 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 998 struct pf_addr *src, sa_family_t af) 999 { 1000 u_short reason = 0; 1001 struct pf_srchash *sh = NULL; 1002 1003 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1004 rule->rpool.opts & PF_POOL_STICKYADDR), 1005 ("%s for non-tracking rule %p", __func__, rule)); 1006 1007 if (*sn == NULL) 1008 *sn = pf_find_src_node(src, rule, af, &sh, true); 1009 1010 if (*sn == NULL) { 1011 PF_HASHROW_ASSERT(sh); 1012 1013 if (rule->max_src_nodes && 1014 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1015 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1016 PF_HASHROW_UNLOCK(sh); 1017 reason = PFRES_SRCLIMIT; 1018 goto done; 1019 } 1020 1021 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1022 if ((*sn) == NULL) { 1023 PF_HASHROW_UNLOCK(sh); 1024 reason = PFRES_MEMORY; 1025 goto done; 1026 } 1027 1028 for (int i = 0; i < 2; i++) { 1029 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1030 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1031 1032 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1033 pf_free_src_node(*sn); 1034 PF_HASHROW_UNLOCK(sh); 1035 reason = PFRES_MEMORY; 1036 goto done; 1037 } 1038 } 1039 1040 pf_init_threshold(&(*sn)->conn_rate, 1041 rule->max_src_conn_rate.limit, 1042 rule->max_src_conn_rate.seconds); 1043 1044 MPASS((*sn)->lock == NULL); 1045 (*sn)->lock = &sh->lock; 1046 1047 (*sn)->af = af; 1048 (*sn)->rule.ptr = rule; 1049 PF_ACPY(&(*sn)->addr, src, af); 1050 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1051 (*sn)->creation = time_uptime; 1052 (*sn)->ruletype = rule->action; 1053 (*sn)->states = 1; 1054 if ((*sn)->rule.ptr != NULL) 1055 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 1056 PF_HASHROW_UNLOCK(sh); 1057 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1058 } else { 1059 if (rule->max_src_states && 1060 (*sn)->states >= rule->max_src_states) { 1061 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1062 1); 1063 reason = PFRES_SRCLIMIT; 1064 goto done; 1065 } 1066 } 1067 done: 1068 return (reason); 1069 } 1070 1071 void 1072 pf_unlink_src_node(struct pf_ksrc_node *src) 1073 { 1074 PF_SRC_NODE_LOCK_ASSERT(src); 1075 1076 LIST_REMOVE(src, entry); 1077 if (src->rule.ptr) 1078 counter_u64_add(src->rule.ptr->src_nodes, -1); 1079 } 1080 1081 u_int 1082 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1083 { 1084 struct pf_ksrc_node *sn, *tmp; 1085 u_int count = 0; 1086 1087 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1088 pf_free_src_node(sn); 1089 count++; 1090 } 1091 1092 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1093 1094 return (count); 1095 } 1096 1097 void 1098 pf_mtag_initialize(void) 1099 { 1100 1101 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1102 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1103 UMA_ALIGN_PTR, 0); 1104 } 1105 1106 /* Per-vnet data storage structures initialization. */ 1107 void 1108 pf_initialize(void) 1109 { 1110 struct pf_keyhash *kh; 1111 struct pf_idhash *ih; 1112 struct pf_srchash *sh; 1113 struct pf_udpendpointhash *uh; 1114 u_int i; 1115 1116 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1117 V_pf_hashsize = PF_HASHSIZ; 1118 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1119 V_pf_srchashsize = PF_SRCHASHSIZ; 1120 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1121 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1122 1123 V_pf_hashseed = arc4random(); 1124 1125 /* States and state keys storage. */ 1126 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1127 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1128 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1129 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1130 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1131 1132 V_pf_state_key_z = uma_zcreate("pf state keys", 1133 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1134 UMA_ALIGN_PTR, 0); 1135 1136 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1137 M_PFHASH, M_NOWAIT | M_ZERO); 1138 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1139 M_PFHASH, M_NOWAIT | M_ZERO); 1140 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1141 printf("pf: Unable to allocate memory for " 1142 "state_hashsize %lu.\n", V_pf_hashsize); 1143 1144 free(V_pf_keyhash, M_PFHASH); 1145 free(V_pf_idhash, M_PFHASH); 1146 1147 V_pf_hashsize = PF_HASHSIZ; 1148 V_pf_keyhash = mallocarray(V_pf_hashsize, 1149 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1150 V_pf_idhash = mallocarray(V_pf_hashsize, 1151 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1152 } 1153 1154 V_pf_hashmask = V_pf_hashsize - 1; 1155 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1156 i++, kh++, ih++) { 1157 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1158 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1159 } 1160 1161 /* Source nodes. */ 1162 V_pf_sources_z = uma_zcreate("pf source nodes", 1163 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1164 0); 1165 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1166 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1167 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1168 1169 V_pf_srchash = mallocarray(V_pf_srchashsize, 1170 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1171 if (V_pf_srchash == NULL) { 1172 printf("pf: Unable to allocate memory for " 1173 "source_hashsize %lu.\n", V_pf_srchashsize); 1174 1175 V_pf_srchashsize = PF_SRCHASHSIZ; 1176 V_pf_srchash = mallocarray(V_pf_srchashsize, 1177 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1178 } 1179 1180 V_pf_srchashmask = V_pf_srchashsize - 1; 1181 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1182 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1183 1184 1185 /* UDP endpoint mappings. */ 1186 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1187 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1188 UMA_ALIGN_PTR, 0); 1189 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1190 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1191 if (V_pf_udpendpointhash == NULL) { 1192 printf("pf: Unable to allocate memory for " 1193 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1194 1195 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1196 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1197 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1198 } 1199 1200 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1201 for (i = 0, uh = V_pf_udpendpointhash; 1202 i <= V_pf_udpendpointhashmask; 1203 i++, uh++) { 1204 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1205 MTX_DEF | MTX_DUPOK); 1206 } 1207 1208 /* ALTQ */ 1209 TAILQ_INIT(&V_pf_altqs[0]); 1210 TAILQ_INIT(&V_pf_altqs[1]); 1211 TAILQ_INIT(&V_pf_altqs[2]); 1212 TAILQ_INIT(&V_pf_altqs[3]); 1213 TAILQ_INIT(&V_pf_pabuf); 1214 V_pf_altqs_active = &V_pf_altqs[0]; 1215 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1216 V_pf_altqs_inactive = &V_pf_altqs[2]; 1217 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1218 1219 /* Send & overload+flush queues. */ 1220 STAILQ_INIT(&V_pf_sendqueue); 1221 SLIST_INIT(&V_pf_overloadqueue); 1222 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1223 1224 /* Unlinked, but may be referenced rules. */ 1225 TAILQ_INIT(&V_pf_unlinked_rules); 1226 } 1227 1228 void 1229 pf_mtag_cleanup(void) 1230 { 1231 1232 uma_zdestroy(pf_mtag_z); 1233 } 1234 1235 void 1236 pf_cleanup(void) 1237 { 1238 struct pf_keyhash *kh; 1239 struct pf_idhash *ih; 1240 struct pf_srchash *sh; 1241 struct pf_udpendpointhash *uh; 1242 struct pf_send_entry *pfse, *next; 1243 u_int i; 1244 1245 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1246 i <= V_pf_hashmask; 1247 i++, kh++, ih++) { 1248 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1249 __func__)); 1250 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1251 __func__)); 1252 mtx_destroy(&kh->lock); 1253 mtx_destroy(&ih->lock); 1254 } 1255 free(V_pf_keyhash, M_PFHASH); 1256 free(V_pf_idhash, M_PFHASH); 1257 1258 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1259 KASSERT(LIST_EMPTY(&sh->nodes), 1260 ("%s: source node hash not empty", __func__)); 1261 mtx_destroy(&sh->lock); 1262 } 1263 free(V_pf_srchash, M_PFHASH); 1264 1265 for (i = 0, uh = V_pf_udpendpointhash; 1266 i <= V_pf_udpendpointhashmask; 1267 i++, uh++) { 1268 KASSERT(LIST_EMPTY(&uh->endpoints), 1269 ("%s: udp endpoint hash not empty", __func__)); 1270 mtx_destroy(&uh->lock); 1271 } 1272 free(V_pf_udpendpointhash, M_PFHASH); 1273 1274 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1275 m_freem(pfse->pfse_m); 1276 free(pfse, M_PFTEMP); 1277 } 1278 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1279 1280 uma_zdestroy(V_pf_sources_z); 1281 uma_zdestroy(V_pf_state_z); 1282 uma_zdestroy(V_pf_state_key_z); 1283 uma_zdestroy(V_pf_udp_mapping_z); 1284 } 1285 1286 static int 1287 pf_mtag_uminit(void *mem, int size, int how) 1288 { 1289 struct m_tag *t; 1290 1291 t = (struct m_tag *)mem; 1292 t->m_tag_cookie = MTAG_ABI_COMPAT; 1293 t->m_tag_id = PACKET_TAG_PF; 1294 t->m_tag_len = sizeof(struct pf_mtag); 1295 t->m_tag_free = pf_mtag_free; 1296 1297 return (0); 1298 } 1299 1300 static void 1301 pf_mtag_free(struct m_tag *t) 1302 { 1303 1304 uma_zfree(pf_mtag_z, t); 1305 } 1306 1307 struct pf_mtag * 1308 pf_get_mtag(struct mbuf *m) 1309 { 1310 struct m_tag *mtag; 1311 1312 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1313 return ((struct pf_mtag *)(mtag + 1)); 1314 1315 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1316 if (mtag == NULL) 1317 return (NULL); 1318 bzero(mtag + 1, sizeof(struct pf_mtag)); 1319 m_tag_prepend(m, mtag); 1320 1321 return ((struct pf_mtag *)(mtag + 1)); 1322 } 1323 1324 static int 1325 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1326 struct pf_kstate *s) 1327 { 1328 struct pf_keyhash *khs, *khw, *kh; 1329 struct pf_state_key *sk, *cur; 1330 struct pf_kstate *si, *olds = NULL; 1331 int idx; 1332 1333 NET_EPOCH_ASSERT(); 1334 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1335 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1336 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1337 1338 /* 1339 * We need to lock hash slots of both keys. To avoid deadlock 1340 * we always lock the slot with lower address first. Unlock order 1341 * isn't important. 1342 * 1343 * We also need to lock ID hash slot before dropping key 1344 * locks. On success we return with ID hash slot locked. 1345 */ 1346 1347 if (skw == sks) { 1348 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1349 PF_HASHROW_LOCK(khs); 1350 } else { 1351 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1352 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1353 if (khs == khw) { 1354 PF_HASHROW_LOCK(khs); 1355 } else if (khs < khw) { 1356 PF_HASHROW_LOCK(khs); 1357 PF_HASHROW_LOCK(khw); 1358 } else { 1359 PF_HASHROW_LOCK(khw); 1360 PF_HASHROW_LOCK(khs); 1361 } 1362 } 1363 1364 #define KEYS_UNLOCK() do { \ 1365 if (khs != khw) { \ 1366 PF_HASHROW_UNLOCK(khs); \ 1367 PF_HASHROW_UNLOCK(khw); \ 1368 } else \ 1369 PF_HASHROW_UNLOCK(khs); \ 1370 } while (0) 1371 1372 /* 1373 * First run: start with wire key. 1374 */ 1375 sk = skw; 1376 kh = khw; 1377 idx = PF_SK_WIRE; 1378 1379 MPASS(s->lock == NULL); 1380 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1381 1382 keyattach: 1383 LIST_FOREACH(cur, &kh->keys, entry) 1384 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1385 break; 1386 1387 if (cur != NULL) { 1388 /* Key exists. Check for same kif, if none, add to key. */ 1389 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1390 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1391 1392 PF_HASHROW_LOCK(ih); 1393 if (si->kif == s->kif && 1394 si->direction == s->direction) { 1395 if (sk->proto == IPPROTO_TCP && 1396 si->src.state >= TCPS_FIN_WAIT_2 && 1397 si->dst.state >= TCPS_FIN_WAIT_2) { 1398 /* 1399 * New state matches an old >FIN_WAIT_2 1400 * state. We can't drop key hash locks, 1401 * thus we can't unlink it properly. 1402 * 1403 * As a workaround we drop it into 1404 * TCPS_CLOSED state, schedule purge 1405 * ASAP and push it into the very end 1406 * of the slot TAILQ, so that it won't 1407 * conflict with our new state. 1408 */ 1409 pf_set_protostate(si, PF_PEER_BOTH, 1410 TCPS_CLOSED); 1411 si->timeout = PFTM_PURGE; 1412 olds = si; 1413 } else { 1414 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1415 printf("pf: %s key attach " 1416 "failed on %s: ", 1417 (idx == PF_SK_WIRE) ? 1418 "wire" : "stack", 1419 s->kif->pfik_name); 1420 pf_print_state_parts(s, 1421 (idx == PF_SK_WIRE) ? 1422 sk : NULL, 1423 (idx == PF_SK_STACK) ? 1424 sk : NULL); 1425 printf(", existing: "); 1426 pf_print_state_parts(si, 1427 (idx == PF_SK_WIRE) ? 1428 sk : NULL, 1429 (idx == PF_SK_STACK) ? 1430 sk : NULL); 1431 printf("\n"); 1432 } 1433 s->timeout = PFTM_UNLINKED; 1434 PF_HASHROW_UNLOCK(ih); 1435 KEYS_UNLOCK(); 1436 uma_zfree(V_pf_state_key_z, sk); 1437 if (idx == PF_SK_STACK) 1438 pf_detach_state(s); 1439 return (EEXIST); /* collision! */ 1440 } 1441 } 1442 PF_HASHROW_UNLOCK(ih); 1443 } 1444 uma_zfree(V_pf_state_key_z, sk); 1445 s->key[idx] = cur; 1446 } else { 1447 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1448 s->key[idx] = sk; 1449 } 1450 1451 stateattach: 1452 /* List is sorted, if-bound states before floating. */ 1453 if (s->kif == V_pfi_all) 1454 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1455 else 1456 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1457 1458 if (olds) { 1459 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1460 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1461 key_list[idx]); 1462 olds = NULL; 1463 } 1464 1465 /* 1466 * Attach done. See how should we (or should not?) 1467 * attach a second key. 1468 */ 1469 if (sks == skw) { 1470 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1471 idx = PF_SK_STACK; 1472 sks = NULL; 1473 goto stateattach; 1474 } else if (sks != NULL) { 1475 /* 1476 * Continue attaching with stack key. 1477 */ 1478 sk = sks; 1479 kh = khs; 1480 idx = PF_SK_STACK; 1481 sks = NULL; 1482 goto keyattach; 1483 } 1484 1485 PF_STATE_LOCK(s); 1486 KEYS_UNLOCK(); 1487 1488 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1489 ("%s failure", __func__)); 1490 1491 return (0); 1492 #undef KEYS_UNLOCK 1493 } 1494 1495 static void 1496 pf_detach_state(struct pf_kstate *s) 1497 { 1498 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1499 struct pf_keyhash *kh; 1500 1501 NET_EPOCH_ASSERT(); 1502 MPASS(s->timeout >= PFTM_MAX); 1503 1504 pf_sctp_multihome_detach_addr(s); 1505 1506 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1507 V_pflow_export_state_ptr(s); 1508 1509 if (sks != NULL) { 1510 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1511 PF_HASHROW_LOCK(kh); 1512 if (s->key[PF_SK_STACK] != NULL) 1513 pf_state_key_detach(s, PF_SK_STACK); 1514 /* 1515 * If both point to same key, then we are done. 1516 */ 1517 if (sks == s->key[PF_SK_WIRE]) { 1518 pf_state_key_detach(s, PF_SK_WIRE); 1519 PF_HASHROW_UNLOCK(kh); 1520 return; 1521 } 1522 PF_HASHROW_UNLOCK(kh); 1523 } 1524 1525 if (s->key[PF_SK_WIRE] != NULL) { 1526 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1527 PF_HASHROW_LOCK(kh); 1528 if (s->key[PF_SK_WIRE] != NULL) 1529 pf_state_key_detach(s, PF_SK_WIRE); 1530 PF_HASHROW_UNLOCK(kh); 1531 } 1532 } 1533 1534 static void 1535 pf_state_key_detach(struct pf_kstate *s, int idx) 1536 { 1537 struct pf_state_key *sk = s->key[idx]; 1538 #ifdef INVARIANTS 1539 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1540 1541 PF_HASHROW_ASSERT(kh); 1542 #endif 1543 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1544 s->key[idx] = NULL; 1545 1546 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1547 LIST_REMOVE(sk, entry); 1548 uma_zfree(V_pf_state_key_z, sk); 1549 } 1550 } 1551 1552 static int 1553 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1554 { 1555 struct pf_state_key *sk = mem; 1556 1557 bzero(sk, sizeof(struct pf_state_key_cmp)); 1558 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1559 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1560 1561 return (0); 1562 } 1563 1564 static int 1565 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1566 struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr, 1567 int didx, struct pf_addr *daddr, int multi) 1568 { 1569 #ifdef INET6 1570 struct nd_neighbor_solicit nd; 1571 struct pf_addr *target; 1572 u_short action, reason; 1573 1574 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1575 goto copy; 1576 1577 switch (pd->hdr.icmp6.icmp6_type) { 1578 case ND_NEIGHBOR_SOLICIT: 1579 if (multi) 1580 return (-1); 1581 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1582 return (-1); 1583 target = (struct pf_addr *)&nd.nd_ns_target; 1584 daddr = target; 1585 break; 1586 case ND_NEIGHBOR_ADVERT: 1587 if (multi) 1588 return (-1); 1589 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1590 return (-1); 1591 target = (struct pf_addr *)&nd.nd_ns_target; 1592 saddr = target; 1593 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1594 key->addr[didx].addr32[0] = 0; 1595 key->addr[didx].addr32[1] = 0; 1596 key->addr[didx].addr32[2] = 0; 1597 key->addr[didx].addr32[3] = 0; 1598 daddr = NULL; /* overwritten */ 1599 } 1600 break; 1601 default: 1602 if (multi == PF_ICMP_MULTI_LINK) { 1603 key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1604 key->addr[sidx].addr32[1] = 0; 1605 key->addr[sidx].addr32[2] = 0; 1606 key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1607 saddr = NULL; /* overwritten */ 1608 } 1609 } 1610 copy: 1611 #endif 1612 if (saddr) 1613 PF_ACPY(&key->addr[sidx], saddr, pd->af); 1614 if (daddr) 1615 PF_ACPY(&key->addr[didx], daddr, pd->af); 1616 1617 return (0); 1618 } 1619 1620 struct pf_state_key * 1621 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1622 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1623 u_int16_t dport) 1624 { 1625 struct pf_state_key *sk; 1626 1627 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1628 if (sk == NULL) 1629 return (NULL); 1630 1631 if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk, 1632 pd->sidx, pd->src, pd->didx, pd->dst, 0)) { 1633 uma_zfree(V_pf_state_key_z, sk); 1634 return (NULL); 1635 } 1636 1637 sk->port[pd->sidx] = sport; 1638 sk->port[pd->didx] = dport; 1639 sk->proto = pd->proto; 1640 sk->af = pd->af; 1641 1642 return (sk); 1643 } 1644 1645 struct pf_state_key * 1646 pf_state_key_clone(const struct pf_state_key *orig) 1647 { 1648 struct pf_state_key *sk; 1649 1650 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1651 if (sk == NULL) 1652 return (NULL); 1653 1654 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1655 1656 return (sk); 1657 } 1658 1659 int 1660 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1661 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1662 { 1663 struct pf_idhash *ih; 1664 struct pf_kstate *cur; 1665 int error; 1666 1667 NET_EPOCH_ASSERT(); 1668 1669 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1670 ("%s: sks not pristine", __func__)); 1671 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1672 ("%s: skw not pristine", __func__)); 1673 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1674 1675 s->kif = kif; 1676 s->orig_kif = orig_kif; 1677 1678 if (s->id == 0 && s->creatorid == 0) { 1679 s->id = alloc_unr64(&V_pf_stateid); 1680 s->id = htobe64(s->id); 1681 s->creatorid = V_pf_status.hostid; 1682 } 1683 1684 /* Returns with ID locked on success. */ 1685 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1686 return (error); 1687 1688 ih = &V_pf_idhash[PF_IDHASH(s)]; 1689 PF_HASHROW_ASSERT(ih); 1690 LIST_FOREACH(cur, &ih->states, entry) 1691 if (cur->id == s->id && cur->creatorid == s->creatorid) 1692 break; 1693 1694 if (cur != NULL) { 1695 s->timeout = PFTM_UNLINKED; 1696 PF_HASHROW_UNLOCK(ih); 1697 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1698 printf("pf: state ID collision: " 1699 "id: %016llx creatorid: %08x\n", 1700 (unsigned long long)be64toh(s->id), 1701 ntohl(s->creatorid)); 1702 } 1703 pf_detach_state(s); 1704 return (EEXIST); 1705 } 1706 LIST_INSERT_HEAD(&ih->states, s, entry); 1707 /* One for keys, one for ID hash. */ 1708 refcount_init(&s->refs, 2); 1709 1710 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1711 if (V_pfsync_insert_state_ptr != NULL) 1712 V_pfsync_insert_state_ptr(s); 1713 1714 /* Returns locked. */ 1715 return (0); 1716 } 1717 1718 /* 1719 * Find state by ID: returns with locked row on success. 1720 */ 1721 struct pf_kstate * 1722 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1723 { 1724 struct pf_idhash *ih; 1725 struct pf_kstate *s; 1726 1727 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1728 1729 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1730 1731 PF_HASHROW_LOCK(ih); 1732 LIST_FOREACH(s, &ih->states, entry) 1733 if (s->id == id && s->creatorid == creatorid) 1734 break; 1735 1736 if (s == NULL) 1737 PF_HASHROW_UNLOCK(ih); 1738 1739 return (s); 1740 } 1741 1742 /* 1743 * Find state by key. 1744 * Returns with ID hash slot locked on success. 1745 */ 1746 static struct pf_kstate * 1747 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1748 u_int dir) 1749 { 1750 struct pf_keyhash *kh; 1751 struct pf_state_key *sk; 1752 struct pf_kstate *s; 1753 int idx; 1754 1755 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1756 1757 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1758 1759 PF_HASHROW_LOCK(kh); 1760 LIST_FOREACH(sk, &kh->keys, entry) 1761 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1762 break; 1763 if (sk == NULL) { 1764 PF_HASHROW_UNLOCK(kh); 1765 return (NULL); 1766 } 1767 1768 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1769 1770 /* List is sorted, if-bound states before floating ones. */ 1771 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1772 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1773 PF_STATE_LOCK(s); 1774 PF_HASHROW_UNLOCK(kh); 1775 if (__predict_false(s->timeout >= PFTM_MAX)) { 1776 /* 1777 * State is either being processed by 1778 * pf_unlink_state() in an other thread, or 1779 * is scheduled for immediate expiry. 1780 */ 1781 PF_STATE_UNLOCK(s); 1782 return (NULL); 1783 } 1784 return (s); 1785 } 1786 PF_HASHROW_UNLOCK(kh); 1787 1788 return (NULL); 1789 } 1790 1791 /* 1792 * Returns with ID hash slot locked on success. 1793 */ 1794 struct pf_kstate * 1795 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1796 { 1797 struct pf_keyhash *kh; 1798 struct pf_state_key *sk; 1799 struct pf_kstate *s, *ret = NULL; 1800 int idx, inout = 0; 1801 1802 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1803 1804 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1805 1806 PF_HASHROW_LOCK(kh); 1807 LIST_FOREACH(sk, &kh->keys, entry) 1808 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1809 break; 1810 if (sk == NULL) { 1811 PF_HASHROW_UNLOCK(kh); 1812 return (NULL); 1813 } 1814 switch (dir) { 1815 case PF_IN: 1816 idx = PF_SK_WIRE; 1817 break; 1818 case PF_OUT: 1819 idx = PF_SK_STACK; 1820 break; 1821 case PF_INOUT: 1822 idx = PF_SK_WIRE; 1823 inout = 1; 1824 break; 1825 default: 1826 panic("%s: dir %u", __func__, dir); 1827 } 1828 second_run: 1829 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1830 if (more == NULL) { 1831 PF_STATE_LOCK(s); 1832 PF_HASHROW_UNLOCK(kh); 1833 return (s); 1834 } 1835 1836 if (ret) 1837 (*more)++; 1838 else { 1839 ret = s; 1840 PF_STATE_LOCK(s); 1841 } 1842 } 1843 if (inout == 1) { 1844 inout = 0; 1845 idx = PF_SK_STACK; 1846 goto second_run; 1847 } 1848 PF_HASHROW_UNLOCK(kh); 1849 1850 return (ret); 1851 } 1852 1853 /* 1854 * FIXME 1855 * This routine is inefficient -- locks the state only to unlock immediately on 1856 * return. 1857 * It is racy -- after the state is unlocked nothing stops other threads from 1858 * removing it. 1859 */ 1860 bool 1861 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1862 { 1863 struct pf_kstate *s; 1864 1865 s = pf_find_state_all(key, dir, NULL); 1866 if (s != NULL) { 1867 PF_STATE_UNLOCK(s); 1868 return (true); 1869 } 1870 return (false); 1871 } 1872 1873 struct pf_udp_mapping * 1874 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1875 struct pf_addr *nat_addr, uint16_t nat_port) 1876 { 1877 struct pf_udp_mapping *mapping; 1878 1879 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1880 if (mapping == NULL) 1881 return (NULL); 1882 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1883 mapping->endpoints[0].port = src_port; 1884 mapping->endpoints[0].af = af; 1885 mapping->endpoints[0].mapping = mapping; 1886 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1887 mapping->endpoints[1].port = nat_port; 1888 mapping->endpoints[1].af = af; 1889 mapping->endpoints[1].mapping = mapping; 1890 refcount_init(&mapping->refs, 1); 1891 return (mapping); 1892 } 1893 1894 int 1895 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1896 { 1897 struct pf_udpendpointhash *h0, *h1; 1898 struct pf_udp_endpoint *endpoint; 1899 int ret = EEXIST; 1900 1901 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1902 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1903 if (h0 == h1) { 1904 PF_HASHROW_LOCK(h0); 1905 } else if (h0 < h1) { 1906 PF_HASHROW_LOCK(h0); 1907 PF_HASHROW_LOCK(h1); 1908 } else { 1909 PF_HASHROW_LOCK(h1); 1910 PF_HASHROW_LOCK(h0); 1911 } 1912 1913 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1914 if (bcmp(endpoint, &mapping->endpoints[0], 1915 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1916 break; 1917 } 1918 if (endpoint != NULL) 1919 goto cleanup; 1920 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1921 if (bcmp(endpoint, &mapping->endpoints[1], 1922 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1923 break; 1924 } 1925 if (endpoint != NULL) 1926 goto cleanup; 1927 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1928 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1929 ret = 0; 1930 1931 cleanup: 1932 if (h0 != h1) { 1933 PF_HASHROW_UNLOCK(h0); 1934 PF_HASHROW_UNLOCK(h1); 1935 } else { 1936 PF_HASHROW_UNLOCK(h0); 1937 } 1938 return (ret); 1939 } 1940 1941 void 1942 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1943 { 1944 /* refcount is synchronized on the source endpoint's row lock */ 1945 struct pf_udpendpointhash *h0, *h1; 1946 1947 if (mapping == NULL) 1948 return; 1949 1950 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1951 PF_HASHROW_LOCK(h0); 1952 if (refcount_release(&mapping->refs)) { 1953 LIST_REMOVE(&mapping->endpoints[0], entry); 1954 PF_HASHROW_UNLOCK(h0); 1955 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1956 PF_HASHROW_LOCK(h1); 1957 LIST_REMOVE(&mapping->endpoints[1], entry); 1958 PF_HASHROW_UNLOCK(h1); 1959 1960 uma_zfree(V_pf_udp_mapping_z, mapping); 1961 } else { 1962 PF_HASHROW_UNLOCK(h0); 1963 } 1964 } 1965 1966 1967 struct pf_udp_mapping * 1968 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1969 { 1970 struct pf_udpendpointhash *uh; 1971 struct pf_udp_endpoint *endpoint; 1972 1973 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1974 1975 PF_HASHROW_LOCK(uh); 1976 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1977 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1978 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1979 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1980 break; 1981 } 1982 if (endpoint == NULL) { 1983 PF_HASHROW_UNLOCK(uh); 1984 return (NULL); 1985 } 1986 refcount_acquire(&endpoint->mapping->refs); 1987 PF_HASHROW_UNLOCK(uh); 1988 return (endpoint->mapping); 1989 } 1990 /* END state table stuff */ 1991 1992 static void 1993 pf_send(struct pf_send_entry *pfse) 1994 { 1995 1996 PF_SENDQ_LOCK(); 1997 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1998 PF_SENDQ_UNLOCK(); 1999 swi_sched(V_pf_swi_cookie, 0); 2000 } 2001 2002 static bool 2003 pf_isforlocal(struct mbuf *m, int af) 2004 { 2005 switch (af) { 2006 #ifdef INET 2007 case AF_INET: { 2008 struct ip *ip = mtod(m, struct ip *); 2009 2010 return (in_localip(ip->ip_dst)); 2011 } 2012 #endif 2013 #ifdef INET6 2014 case AF_INET6: { 2015 struct ip6_hdr *ip6; 2016 struct in6_ifaddr *ia; 2017 ip6 = mtod(m, struct ip6_hdr *); 2018 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2019 if (ia == NULL) 2020 return (false); 2021 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2022 } 2023 #endif 2024 default: 2025 panic("Unsupported af %d", af); 2026 } 2027 2028 return (false); 2029 } 2030 2031 int 2032 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2033 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2034 { 2035 /* 2036 * ICMP types marked with PF_OUT are typically responses to 2037 * PF_IN, and will match states in the opposite direction. 2038 * PF_IN ICMP types need to match a state with that type. 2039 */ 2040 *icmp_dir = PF_OUT; 2041 *multi = PF_ICMP_MULTI_LINK; 2042 /* Queries (and responses) */ 2043 switch (pd->af) { 2044 #ifdef INET 2045 case AF_INET: 2046 switch (type) { 2047 case ICMP_ECHO: 2048 *icmp_dir = PF_IN; 2049 case ICMP_ECHOREPLY: 2050 *virtual_type = ICMP_ECHO; 2051 *virtual_id = pd->hdr.icmp.icmp_id; 2052 break; 2053 2054 case ICMP_TSTAMP: 2055 *icmp_dir = PF_IN; 2056 case ICMP_TSTAMPREPLY: 2057 *virtual_type = ICMP_TSTAMP; 2058 *virtual_id = pd->hdr.icmp.icmp_id; 2059 break; 2060 2061 case ICMP_IREQ: 2062 *icmp_dir = PF_IN; 2063 case ICMP_IREQREPLY: 2064 *virtual_type = ICMP_IREQ; 2065 *virtual_id = pd->hdr.icmp.icmp_id; 2066 break; 2067 2068 case ICMP_MASKREQ: 2069 *icmp_dir = PF_IN; 2070 case ICMP_MASKREPLY: 2071 *virtual_type = ICMP_MASKREQ; 2072 *virtual_id = pd->hdr.icmp.icmp_id; 2073 break; 2074 2075 case ICMP_IPV6_WHEREAREYOU: 2076 *icmp_dir = PF_IN; 2077 case ICMP_IPV6_IAMHERE: 2078 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2079 *virtual_id = 0; /* Nothing sane to match on! */ 2080 break; 2081 2082 case ICMP_MOBILE_REGREQUEST: 2083 *icmp_dir = PF_IN; 2084 case ICMP_MOBILE_REGREPLY: 2085 *virtual_type = ICMP_MOBILE_REGREQUEST; 2086 *virtual_id = 0; /* Nothing sane to match on! */ 2087 break; 2088 2089 case ICMP_ROUTERSOLICIT: 2090 *icmp_dir = PF_IN; 2091 case ICMP_ROUTERADVERT: 2092 *virtual_type = ICMP_ROUTERSOLICIT; 2093 *virtual_id = 0; /* Nothing sane to match on! */ 2094 break; 2095 2096 /* These ICMP types map to other connections */ 2097 case ICMP_UNREACH: 2098 case ICMP_SOURCEQUENCH: 2099 case ICMP_REDIRECT: 2100 case ICMP_TIMXCEED: 2101 case ICMP_PARAMPROB: 2102 /* These will not be used, but set them anyway */ 2103 *icmp_dir = PF_IN; 2104 *virtual_type = type; 2105 *virtual_id = 0; 2106 HTONS(*virtual_type); 2107 return (1); /* These types match to another state */ 2108 2109 /* 2110 * All remaining ICMP types get their own states, 2111 * and will only match in one direction. 2112 */ 2113 default: 2114 *icmp_dir = PF_IN; 2115 *virtual_type = type; 2116 *virtual_id = 0; 2117 break; 2118 } 2119 break; 2120 #endif /* INET */ 2121 #ifdef INET6 2122 case AF_INET6: 2123 switch (type) { 2124 case ICMP6_ECHO_REQUEST: 2125 *icmp_dir = PF_IN; 2126 case ICMP6_ECHO_REPLY: 2127 *virtual_type = ICMP6_ECHO_REQUEST; 2128 *virtual_id = pd->hdr.icmp6.icmp6_id; 2129 break; 2130 2131 case MLD_LISTENER_QUERY: 2132 case MLD_LISTENER_REPORT: { 2133 /* 2134 * Listener Report can be sent by clients 2135 * without an associated Listener Query. 2136 * In addition to that, when Report is sent as a 2137 * reply to a Query its source and destination 2138 * address are different. 2139 */ 2140 *icmp_dir = PF_IN; 2141 *virtual_type = MLD_LISTENER_QUERY; 2142 *virtual_id = 0; 2143 break; 2144 } 2145 case MLD_MTRACE: 2146 *icmp_dir = PF_IN; 2147 case MLD_MTRACE_RESP: 2148 *virtual_type = MLD_MTRACE; 2149 *virtual_id = 0; /* Nothing sane to match on! */ 2150 break; 2151 2152 case ND_NEIGHBOR_SOLICIT: 2153 *icmp_dir = PF_IN; 2154 case ND_NEIGHBOR_ADVERT: { 2155 *virtual_type = ND_NEIGHBOR_SOLICIT; 2156 *virtual_id = 0; 2157 break; 2158 } 2159 2160 /* 2161 * These ICMP types map to other connections. 2162 * ND_REDIRECT can't be in this list because the triggering 2163 * packet header is optional. 2164 */ 2165 case ICMP6_DST_UNREACH: 2166 case ICMP6_PACKET_TOO_BIG: 2167 case ICMP6_TIME_EXCEEDED: 2168 case ICMP6_PARAM_PROB: 2169 /* These will not be used, but set them anyway */ 2170 *icmp_dir = PF_IN; 2171 *virtual_type = type; 2172 *virtual_id = 0; 2173 HTONS(*virtual_type); 2174 return (1); /* These types match to another state */ 2175 /* 2176 * All remaining ICMP6 types get their own states, 2177 * and will only match in one direction. 2178 */ 2179 default: 2180 *icmp_dir = PF_IN; 2181 *virtual_type = type; 2182 *virtual_id = 0; 2183 break; 2184 } 2185 break; 2186 #endif /* INET6 */ 2187 default: 2188 *icmp_dir = PF_IN; 2189 *virtual_type = type; 2190 *virtual_id = 0; 2191 break; 2192 } 2193 HTONS(*virtual_type); 2194 return (0); /* These types match to their own state */ 2195 } 2196 2197 void 2198 pf_intr(void *v) 2199 { 2200 struct epoch_tracker et; 2201 struct pf_send_head queue; 2202 struct pf_send_entry *pfse, *next; 2203 2204 CURVNET_SET((struct vnet *)v); 2205 2206 PF_SENDQ_LOCK(); 2207 queue = V_pf_sendqueue; 2208 STAILQ_INIT(&V_pf_sendqueue); 2209 PF_SENDQ_UNLOCK(); 2210 2211 NET_EPOCH_ENTER(et); 2212 2213 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2214 switch (pfse->pfse_type) { 2215 #ifdef INET 2216 case PFSE_IP: { 2217 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2218 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2219 pfse->pfse_m->m_pkthdr.csum_flags |= 2220 CSUM_IP_VALID | CSUM_IP_CHECKED; 2221 ip_input(pfse->pfse_m); 2222 } else { 2223 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2224 NULL); 2225 } 2226 break; 2227 } 2228 case PFSE_ICMP: 2229 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2230 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2231 break; 2232 #endif /* INET */ 2233 #ifdef INET6 2234 case PFSE_IP6: 2235 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2236 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2237 ip6_input(pfse->pfse_m); 2238 } else { 2239 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2240 NULL, NULL); 2241 } 2242 break; 2243 case PFSE_ICMP6: 2244 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2245 pfse->icmpopts.code, pfse->icmpopts.mtu); 2246 break; 2247 #endif /* INET6 */ 2248 default: 2249 panic("%s: unknown type", __func__); 2250 } 2251 free(pfse, M_PFTEMP); 2252 } 2253 NET_EPOCH_EXIT(et); 2254 CURVNET_RESTORE(); 2255 } 2256 2257 #define pf_purge_thread_period (hz / 10) 2258 2259 #ifdef PF_WANT_32_TO_64_COUNTER 2260 static void 2261 pf_status_counter_u64_periodic(void) 2262 { 2263 2264 PF_RULES_RASSERT(); 2265 2266 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2267 return; 2268 } 2269 2270 for (int i = 0; i < FCNT_MAX; i++) { 2271 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2272 } 2273 } 2274 2275 static void 2276 pf_kif_counter_u64_periodic(void) 2277 { 2278 struct pfi_kkif *kif; 2279 size_t r, run; 2280 2281 PF_RULES_RASSERT(); 2282 2283 if (__predict_false(V_pf_allkifcount == 0)) { 2284 return; 2285 } 2286 2287 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2288 return; 2289 } 2290 2291 run = V_pf_allkifcount / 10; 2292 if (run < 5) 2293 run = 5; 2294 2295 for (r = 0; r < run; r++) { 2296 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2297 if (kif == NULL) { 2298 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2299 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2300 break; 2301 } 2302 2303 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2304 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2305 2306 for (int i = 0; i < 2; i++) { 2307 for (int j = 0; j < 2; j++) { 2308 for (int k = 0; k < 2; k++) { 2309 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2310 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2311 } 2312 } 2313 } 2314 } 2315 } 2316 2317 static void 2318 pf_rule_counter_u64_periodic(void) 2319 { 2320 struct pf_krule *rule; 2321 size_t r, run; 2322 2323 PF_RULES_RASSERT(); 2324 2325 if (__predict_false(V_pf_allrulecount == 0)) { 2326 return; 2327 } 2328 2329 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2330 return; 2331 } 2332 2333 run = V_pf_allrulecount / 10; 2334 if (run < 5) 2335 run = 5; 2336 2337 for (r = 0; r < run; r++) { 2338 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2339 if (rule == NULL) { 2340 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2341 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2342 break; 2343 } 2344 2345 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2346 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2347 2348 pf_counter_u64_periodic(&rule->evaluations); 2349 for (int i = 0; i < 2; i++) { 2350 pf_counter_u64_periodic(&rule->packets[i]); 2351 pf_counter_u64_periodic(&rule->bytes[i]); 2352 } 2353 } 2354 } 2355 2356 static void 2357 pf_counter_u64_periodic_main(void) 2358 { 2359 PF_RULES_RLOCK_TRACKER; 2360 2361 V_pf_counter_periodic_iter++; 2362 2363 PF_RULES_RLOCK(); 2364 pf_counter_u64_critical_enter(); 2365 pf_status_counter_u64_periodic(); 2366 pf_kif_counter_u64_periodic(); 2367 pf_rule_counter_u64_periodic(); 2368 pf_counter_u64_critical_exit(); 2369 PF_RULES_RUNLOCK(); 2370 } 2371 #else 2372 #define pf_counter_u64_periodic_main() do { } while (0) 2373 #endif 2374 2375 void 2376 pf_purge_thread(void *unused __unused) 2377 { 2378 struct epoch_tracker et; 2379 2380 VNET_ITERATOR_DECL(vnet_iter); 2381 2382 sx_xlock(&pf_end_lock); 2383 while (pf_end_threads == 0) { 2384 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2385 2386 VNET_LIST_RLOCK(); 2387 NET_EPOCH_ENTER(et); 2388 VNET_FOREACH(vnet_iter) { 2389 CURVNET_SET(vnet_iter); 2390 2391 /* Wait until V_pf_default_rule is initialized. */ 2392 if (V_pf_vnet_active == 0) { 2393 CURVNET_RESTORE(); 2394 continue; 2395 } 2396 2397 pf_counter_u64_periodic_main(); 2398 2399 /* 2400 * Process 1/interval fraction of the state 2401 * table every run. 2402 */ 2403 V_pf_purge_idx = 2404 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2405 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2406 2407 /* 2408 * Purge other expired types every 2409 * PFTM_INTERVAL seconds. 2410 */ 2411 if (V_pf_purge_idx == 0) { 2412 /* 2413 * Order is important: 2414 * - states and src nodes reference rules 2415 * - states and rules reference kifs 2416 */ 2417 pf_purge_expired_fragments(); 2418 pf_purge_expired_src_nodes(); 2419 pf_purge_unlinked_rules(); 2420 pfi_kkif_purge(); 2421 } 2422 CURVNET_RESTORE(); 2423 } 2424 NET_EPOCH_EXIT(et); 2425 VNET_LIST_RUNLOCK(); 2426 } 2427 2428 pf_end_threads++; 2429 sx_xunlock(&pf_end_lock); 2430 kproc_exit(0); 2431 } 2432 2433 void 2434 pf_unload_vnet_purge(void) 2435 { 2436 2437 /* 2438 * To cleanse up all kifs and rules we need 2439 * two runs: first one clears reference flags, 2440 * then pf_purge_expired_states() doesn't 2441 * raise them, and then second run frees. 2442 */ 2443 pf_purge_unlinked_rules(); 2444 pfi_kkif_purge(); 2445 2446 /* 2447 * Now purge everything. 2448 */ 2449 pf_purge_expired_states(0, V_pf_hashmask); 2450 pf_purge_fragments(UINT_MAX); 2451 pf_purge_expired_src_nodes(); 2452 2453 /* 2454 * Now all kifs & rules should be unreferenced, 2455 * thus should be successfully freed. 2456 */ 2457 pf_purge_unlinked_rules(); 2458 pfi_kkif_purge(); 2459 } 2460 2461 u_int32_t 2462 pf_state_expires(const struct pf_kstate *state) 2463 { 2464 u_int32_t timeout; 2465 u_int32_t start; 2466 u_int32_t end; 2467 u_int32_t states; 2468 2469 /* handle all PFTM_* > PFTM_MAX here */ 2470 if (state->timeout == PFTM_PURGE) 2471 return (time_uptime); 2472 KASSERT(state->timeout != PFTM_UNLINKED, 2473 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2474 KASSERT((state->timeout < PFTM_MAX), 2475 ("pf_state_expires: timeout > PFTM_MAX")); 2476 timeout = state->rule.ptr->timeout[state->timeout]; 2477 if (!timeout) 2478 timeout = V_pf_default_rule.timeout[state->timeout]; 2479 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2480 if (start && state->rule.ptr != &V_pf_default_rule) { 2481 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2482 states = counter_u64_fetch(state->rule.ptr->states_cur); 2483 } else { 2484 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2485 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2486 states = V_pf_status.states; 2487 } 2488 if (end && states > start && start < end) { 2489 if (states < end) { 2490 timeout = (u_int64_t)timeout * (end - states) / 2491 (end - start); 2492 return ((state->expire / 1000) + timeout); 2493 } 2494 else 2495 return (time_uptime); 2496 } 2497 return ((state->expire / 1000) + timeout); 2498 } 2499 2500 void 2501 pf_purge_expired_src_nodes(void) 2502 { 2503 struct pf_ksrc_node_list freelist; 2504 struct pf_srchash *sh; 2505 struct pf_ksrc_node *cur, *next; 2506 int i; 2507 2508 LIST_INIT(&freelist); 2509 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2510 PF_HASHROW_LOCK(sh); 2511 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2512 if (cur->states == 0 && cur->expire <= time_uptime) { 2513 pf_unlink_src_node(cur); 2514 LIST_INSERT_HEAD(&freelist, cur, entry); 2515 } else if (cur->rule.ptr != NULL) 2516 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2517 PF_HASHROW_UNLOCK(sh); 2518 } 2519 2520 pf_free_src_nodes(&freelist); 2521 2522 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2523 } 2524 2525 static void 2526 pf_src_tree_remove_state(struct pf_kstate *s) 2527 { 2528 struct pf_ksrc_node *sn; 2529 uint32_t timeout; 2530 2531 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2532 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2533 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2534 2535 if (s->src_node != NULL) { 2536 sn = s->src_node; 2537 PF_SRC_NODE_LOCK(sn); 2538 if (s->src.tcp_est) 2539 --sn->conn; 2540 if (--sn->states == 0) 2541 sn->expire = time_uptime + timeout; 2542 PF_SRC_NODE_UNLOCK(sn); 2543 } 2544 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2545 sn = s->nat_src_node; 2546 PF_SRC_NODE_LOCK(sn); 2547 if (--sn->states == 0) 2548 sn->expire = time_uptime + timeout; 2549 PF_SRC_NODE_UNLOCK(sn); 2550 } 2551 s->src_node = s->nat_src_node = NULL; 2552 } 2553 2554 /* 2555 * Unlink and potentilly free a state. Function may be 2556 * called with ID hash row locked, but always returns 2557 * unlocked, since it needs to go through key hash locking. 2558 */ 2559 int 2560 pf_unlink_state(struct pf_kstate *s) 2561 { 2562 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2563 2564 NET_EPOCH_ASSERT(); 2565 PF_HASHROW_ASSERT(ih); 2566 2567 if (s->timeout == PFTM_UNLINKED) { 2568 /* 2569 * State is being processed 2570 * by pf_unlink_state() in 2571 * an other thread. 2572 */ 2573 PF_HASHROW_UNLOCK(ih); 2574 return (0); /* XXXGL: undefined actually */ 2575 } 2576 2577 if (s->src.state == PF_TCPS_PROXY_DST) { 2578 /* XXX wire key the right one? */ 2579 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2580 &s->key[PF_SK_WIRE]->addr[1], 2581 &s->key[PF_SK_WIRE]->addr[0], 2582 s->key[PF_SK_WIRE]->port[1], 2583 s->key[PF_SK_WIRE]->port[0], 2584 s->src.seqhi, s->src.seqlo + 1, 2585 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2586 } 2587 2588 LIST_REMOVE(s, entry); 2589 pf_src_tree_remove_state(s); 2590 2591 if (V_pfsync_delete_state_ptr != NULL) 2592 V_pfsync_delete_state_ptr(s); 2593 2594 STATE_DEC_COUNTERS(s); 2595 2596 s->timeout = PFTM_UNLINKED; 2597 2598 /* Ensure we remove it from the list of halfopen states, if needed. */ 2599 if (s->key[PF_SK_STACK] != NULL && 2600 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2601 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2602 2603 PF_HASHROW_UNLOCK(ih); 2604 2605 pf_detach_state(s); 2606 2607 pf_udp_mapping_release(s->udp_mapping); 2608 2609 /* pf_state_insert() initialises refs to 2 */ 2610 return (pf_release_staten(s, 2)); 2611 } 2612 2613 struct pf_kstate * 2614 pf_alloc_state(int flags) 2615 { 2616 2617 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2618 } 2619 2620 void 2621 pf_free_state(struct pf_kstate *cur) 2622 { 2623 struct pf_krule_item *ri; 2624 2625 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2626 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2627 cur->timeout)); 2628 2629 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2630 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2631 free(ri, M_PF_RULE_ITEM); 2632 } 2633 2634 pf_normalize_tcp_cleanup(cur); 2635 uma_zfree(V_pf_state_z, cur); 2636 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2637 } 2638 2639 /* 2640 * Called only from pf_purge_thread(), thus serialized. 2641 */ 2642 static u_int 2643 pf_purge_expired_states(u_int i, int maxcheck) 2644 { 2645 struct pf_idhash *ih; 2646 struct pf_kstate *s; 2647 struct pf_krule_item *mrm; 2648 size_t count __unused; 2649 2650 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2651 2652 /* 2653 * Go through hash and unlink states that expire now. 2654 */ 2655 while (maxcheck > 0) { 2656 count = 0; 2657 ih = &V_pf_idhash[i]; 2658 2659 /* only take the lock if we expect to do work */ 2660 if (!LIST_EMPTY(&ih->states)) { 2661 relock: 2662 PF_HASHROW_LOCK(ih); 2663 LIST_FOREACH(s, &ih->states, entry) { 2664 if (pf_state_expires(s) <= time_uptime) { 2665 V_pf_status.states -= 2666 pf_unlink_state(s); 2667 goto relock; 2668 } 2669 s->rule.ptr->rule_ref |= PFRULE_REFS; 2670 if (s->nat_rule.ptr != NULL) 2671 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2672 if (s->anchor.ptr != NULL) 2673 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2674 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2675 SLIST_FOREACH(mrm, &s->match_rules, entry) 2676 mrm->r->rule_ref |= PFRULE_REFS; 2677 if (s->rt_kif) 2678 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2679 count++; 2680 } 2681 PF_HASHROW_UNLOCK(ih); 2682 } 2683 2684 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2685 2686 /* Return when we hit end of hash. */ 2687 if (++i > V_pf_hashmask) { 2688 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2689 return (0); 2690 } 2691 2692 maxcheck--; 2693 } 2694 2695 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2696 2697 return (i); 2698 } 2699 2700 static void 2701 pf_purge_unlinked_rules(void) 2702 { 2703 struct pf_krulequeue tmpq; 2704 struct pf_krule *r, *r1; 2705 2706 /* 2707 * If we have overloading task pending, then we'd 2708 * better skip purging this time. There is a tiny 2709 * probability that overloading task references 2710 * an already unlinked rule. 2711 */ 2712 PF_OVERLOADQ_LOCK(); 2713 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2714 PF_OVERLOADQ_UNLOCK(); 2715 return; 2716 } 2717 PF_OVERLOADQ_UNLOCK(); 2718 2719 /* 2720 * Do naive mark-and-sweep garbage collecting of old rules. 2721 * Reference flag is raised by pf_purge_expired_states() 2722 * and pf_purge_expired_src_nodes(). 2723 * 2724 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2725 * use a temporary queue. 2726 */ 2727 TAILQ_INIT(&tmpq); 2728 PF_UNLNKDRULES_LOCK(); 2729 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2730 if (!(r->rule_ref & PFRULE_REFS)) { 2731 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2732 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2733 } else 2734 r->rule_ref &= ~PFRULE_REFS; 2735 } 2736 PF_UNLNKDRULES_UNLOCK(); 2737 2738 if (!TAILQ_EMPTY(&tmpq)) { 2739 PF_CONFIG_LOCK(); 2740 PF_RULES_WLOCK(); 2741 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2742 TAILQ_REMOVE(&tmpq, r, entries); 2743 pf_free_rule(r); 2744 } 2745 PF_RULES_WUNLOCK(); 2746 PF_CONFIG_UNLOCK(); 2747 } 2748 } 2749 2750 void 2751 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2752 { 2753 switch (af) { 2754 #ifdef INET 2755 case AF_INET: { 2756 u_int32_t a = ntohl(addr->addr32[0]); 2757 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2758 (a>>8)&255, a&255); 2759 if (p) { 2760 p = ntohs(p); 2761 printf(":%u", p); 2762 } 2763 break; 2764 } 2765 #endif /* INET */ 2766 #ifdef INET6 2767 case AF_INET6: { 2768 u_int16_t b; 2769 u_int8_t i, curstart, curend, maxstart, maxend; 2770 curstart = curend = maxstart = maxend = 255; 2771 for (i = 0; i < 8; i++) { 2772 if (!addr->addr16[i]) { 2773 if (curstart == 255) 2774 curstart = i; 2775 curend = i; 2776 } else { 2777 if ((curend - curstart) > 2778 (maxend - maxstart)) { 2779 maxstart = curstart; 2780 maxend = curend; 2781 } 2782 curstart = curend = 255; 2783 } 2784 } 2785 if ((curend - curstart) > 2786 (maxend - maxstart)) { 2787 maxstart = curstart; 2788 maxend = curend; 2789 } 2790 for (i = 0; i < 8; i++) { 2791 if (i >= maxstart && i <= maxend) { 2792 if (i == 0) 2793 printf(":"); 2794 if (i == maxend) 2795 printf(":"); 2796 } else { 2797 b = ntohs(addr->addr16[i]); 2798 printf("%x", b); 2799 if (i < 7) 2800 printf(":"); 2801 } 2802 } 2803 if (p) { 2804 p = ntohs(p); 2805 printf("[%u]", p); 2806 } 2807 break; 2808 } 2809 #endif /* INET6 */ 2810 } 2811 } 2812 2813 void 2814 pf_print_state(struct pf_kstate *s) 2815 { 2816 pf_print_state_parts(s, NULL, NULL); 2817 } 2818 2819 static void 2820 pf_print_state_parts(struct pf_kstate *s, 2821 struct pf_state_key *skwp, struct pf_state_key *sksp) 2822 { 2823 struct pf_state_key *skw, *sks; 2824 u_int8_t proto, dir; 2825 2826 /* Do our best to fill these, but they're skipped if NULL */ 2827 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2828 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2829 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2830 dir = s ? s->direction : 0; 2831 2832 switch (proto) { 2833 case IPPROTO_IPV4: 2834 printf("IPv4"); 2835 break; 2836 case IPPROTO_IPV6: 2837 printf("IPv6"); 2838 break; 2839 case IPPROTO_TCP: 2840 printf("TCP"); 2841 break; 2842 case IPPROTO_UDP: 2843 printf("UDP"); 2844 break; 2845 case IPPROTO_ICMP: 2846 printf("ICMP"); 2847 break; 2848 case IPPROTO_ICMPV6: 2849 printf("ICMPv6"); 2850 break; 2851 default: 2852 printf("%u", proto); 2853 break; 2854 } 2855 switch (dir) { 2856 case PF_IN: 2857 printf(" in"); 2858 break; 2859 case PF_OUT: 2860 printf(" out"); 2861 break; 2862 } 2863 if (skw) { 2864 printf(" wire: "); 2865 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2866 printf(" "); 2867 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2868 } 2869 if (sks) { 2870 printf(" stack: "); 2871 if (sks != skw) { 2872 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2873 printf(" "); 2874 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2875 } else 2876 printf("-"); 2877 } 2878 if (s) { 2879 if (proto == IPPROTO_TCP) { 2880 printf(" [lo=%u high=%u win=%u modulator=%u", 2881 s->src.seqlo, s->src.seqhi, 2882 s->src.max_win, s->src.seqdiff); 2883 if (s->src.wscale && s->dst.wscale) 2884 printf(" wscale=%u", 2885 s->src.wscale & PF_WSCALE_MASK); 2886 printf("]"); 2887 printf(" [lo=%u high=%u win=%u modulator=%u", 2888 s->dst.seqlo, s->dst.seqhi, 2889 s->dst.max_win, s->dst.seqdiff); 2890 if (s->src.wscale && s->dst.wscale) 2891 printf(" wscale=%u", 2892 s->dst.wscale & PF_WSCALE_MASK); 2893 printf("]"); 2894 } 2895 printf(" %u:%u", s->src.state, s->dst.state); 2896 if (s->rule.ptr) 2897 printf(" @%d", s->rule.ptr->nr); 2898 } 2899 } 2900 2901 void 2902 pf_print_flags(u_int8_t f) 2903 { 2904 if (f) 2905 printf(" "); 2906 if (f & TH_FIN) 2907 printf("F"); 2908 if (f & TH_SYN) 2909 printf("S"); 2910 if (f & TH_RST) 2911 printf("R"); 2912 if (f & TH_PUSH) 2913 printf("P"); 2914 if (f & TH_ACK) 2915 printf("A"); 2916 if (f & TH_URG) 2917 printf("U"); 2918 if (f & TH_ECE) 2919 printf("E"); 2920 if (f & TH_CWR) 2921 printf("W"); 2922 } 2923 2924 #define PF_SET_SKIP_STEPS(i) \ 2925 do { \ 2926 while (head[i] != cur) { \ 2927 head[i]->skip[i].ptr = cur; \ 2928 head[i] = TAILQ_NEXT(head[i], entries); \ 2929 } \ 2930 } while (0) 2931 2932 void 2933 pf_calc_skip_steps(struct pf_krulequeue *rules) 2934 { 2935 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2936 int i; 2937 2938 cur = TAILQ_FIRST(rules); 2939 prev = cur; 2940 for (i = 0; i < PF_SKIP_COUNT; ++i) 2941 head[i] = cur; 2942 while (cur != NULL) { 2943 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2944 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2945 if (cur->direction != prev->direction) 2946 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2947 if (cur->af != prev->af) 2948 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2949 if (cur->proto != prev->proto) 2950 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2951 if (cur->src.neg != prev->src.neg || 2952 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2953 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2954 if (cur->src.port[0] != prev->src.port[0] || 2955 cur->src.port[1] != prev->src.port[1] || 2956 cur->src.port_op != prev->src.port_op) 2957 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2958 if (cur->dst.neg != prev->dst.neg || 2959 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2960 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2961 if (cur->dst.port[0] != prev->dst.port[0] || 2962 cur->dst.port[1] != prev->dst.port[1] || 2963 cur->dst.port_op != prev->dst.port_op) 2964 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2965 2966 prev = cur; 2967 cur = TAILQ_NEXT(cur, entries); 2968 } 2969 for (i = 0; i < PF_SKIP_COUNT; ++i) 2970 PF_SET_SKIP_STEPS(i); 2971 } 2972 2973 int 2974 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2975 { 2976 if (aw1->type != aw2->type) 2977 return (1); 2978 switch (aw1->type) { 2979 case PF_ADDR_ADDRMASK: 2980 case PF_ADDR_RANGE: 2981 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2982 return (1); 2983 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2984 return (1); 2985 return (0); 2986 case PF_ADDR_DYNIFTL: 2987 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2988 case PF_ADDR_NOROUTE: 2989 case PF_ADDR_URPFFAILED: 2990 return (0); 2991 case PF_ADDR_TABLE: 2992 return (aw1->p.tbl != aw2->p.tbl); 2993 default: 2994 printf("invalid address type: %d\n", aw1->type); 2995 return (1); 2996 } 2997 } 2998 2999 /** 3000 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3001 * header isn't always a full checksum. In some cases (i.e. output) it's a 3002 * pseudo-header checksum, which is a partial checksum over src/dst IP 3003 * addresses, protocol number and length. 3004 * 3005 * That means we have the following cases: 3006 * * Input or forwarding: we don't have TSO, the checksum fields are full 3007 * checksums, we need to update the checksum whenever we change anything. 3008 * * Output (i.e. the checksum is a pseudo-header checksum): 3009 * x The field being updated is src/dst address or affects the length of 3010 * the packet. We need to update the pseudo-header checksum (note that this 3011 * checksum is not ones' complement). 3012 * x Some other field is being modified (e.g. src/dst port numbers): We 3013 * don't have to update anything. 3014 **/ 3015 u_int16_t 3016 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3017 { 3018 u_int32_t x; 3019 3020 x = cksum + old - new; 3021 x = (x + (x >> 16)) & 0xffff; 3022 3023 /* optimise: eliminate a branch when not udp */ 3024 if (udp && cksum == 0x0000) 3025 return cksum; 3026 if (udp && x == 0x0000) 3027 x = 0xffff; 3028 3029 return (u_int16_t)(x); 3030 } 3031 3032 static void 3033 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3034 u_int8_t udp) 3035 { 3036 u_int16_t old = htons(hi ? (*f << 8) : *f); 3037 u_int16_t new = htons(hi ? ( v << 8) : v); 3038 3039 if (*f == v) 3040 return; 3041 3042 *f = v; 3043 3044 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3045 return; 3046 3047 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3048 } 3049 3050 void 3051 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3052 bool hi, u_int8_t udp) 3053 { 3054 u_int8_t *fb = (u_int8_t *)f; 3055 u_int8_t *vb = (u_int8_t *)&v; 3056 3057 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3058 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3059 } 3060 3061 void 3062 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3063 bool hi, u_int8_t udp) 3064 { 3065 u_int8_t *fb = (u_int8_t *)f; 3066 u_int8_t *vb = (u_int8_t *)&v; 3067 3068 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3069 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3070 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3071 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3072 } 3073 3074 u_int16_t 3075 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3076 u_int16_t new, u_int8_t udp) 3077 { 3078 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3079 return (cksum); 3080 3081 return (pf_cksum_fixup(cksum, old, new, udp)); 3082 } 3083 3084 static void 3085 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3086 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3087 sa_family_t af) 3088 { 3089 struct pf_addr ao; 3090 u_int16_t po = *p; 3091 3092 PF_ACPY(&ao, a, af); 3093 PF_ACPY(a, an, af); 3094 3095 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3096 *pc = ~*pc; 3097 3098 *p = pn; 3099 3100 switch (af) { 3101 #ifdef INET 3102 case AF_INET: 3103 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3104 ao.addr16[0], an->addr16[0], 0), 3105 ao.addr16[1], an->addr16[1], 0); 3106 *p = pn; 3107 3108 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3109 ao.addr16[0], an->addr16[0], u), 3110 ao.addr16[1], an->addr16[1], u); 3111 3112 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3113 break; 3114 #endif /* INET */ 3115 #ifdef INET6 3116 case AF_INET6: 3117 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3118 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3119 pf_cksum_fixup(pf_cksum_fixup(*pc, 3120 ao.addr16[0], an->addr16[0], u), 3121 ao.addr16[1], an->addr16[1], u), 3122 ao.addr16[2], an->addr16[2], u), 3123 ao.addr16[3], an->addr16[3], u), 3124 ao.addr16[4], an->addr16[4], u), 3125 ao.addr16[5], an->addr16[5], u), 3126 ao.addr16[6], an->addr16[6], u), 3127 ao.addr16[7], an->addr16[7], u); 3128 3129 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3130 break; 3131 #endif /* INET6 */ 3132 } 3133 3134 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3135 CSUM_DELAY_DATA_IPV6)) { 3136 *pc = ~*pc; 3137 if (! *pc) 3138 *pc = 0xffff; 3139 } 3140 } 3141 3142 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3143 void 3144 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3145 { 3146 u_int32_t ao; 3147 3148 memcpy(&ao, a, sizeof(ao)); 3149 memcpy(a, &an, sizeof(u_int32_t)); 3150 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3151 ao % 65536, an % 65536, u); 3152 } 3153 3154 void 3155 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3156 { 3157 u_int32_t ao; 3158 3159 memcpy(&ao, a, sizeof(ao)); 3160 memcpy(a, &an, sizeof(u_int32_t)); 3161 3162 *c = pf_proto_cksum_fixup(m, 3163 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3164 ao % 65536, an % 65536, udp); 3165 } 3166 3167 #ifdef INET6 3168 static void 3169 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3170 { 3171 struct pf_addr ao; 3172 3173 PF_ACPY(&ao, a, AF_INET6); 3174 PF_ACPY(a, an, AF_INET6); 3175 3176 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3177 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3178 pf_cksum_fixup(pf_cksum_fixup(*c, 3179 ao.addr16[0], an->addr16[0], u), 3180 ao.addr16[1], an->addr16[1], u), 3181 ao.addr16[2], an->addr16[2], u), 3182 ao.addr16[3], an->addr16[3], u), 3183 ao.addr16[4], an->addr16[4], u), 3184 ao.addr16[5], an->addr16[5], u), 3185 ao.addr16[6], an->addr16[6], u), 3186 ao.addr16[7], an->addr16[7], u); 3187 } 3188 #endif /* INET6 */ 3189 3190 static void 3191 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3192 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3193 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3194 { 3195 struct pf_addr oia, ooa; 3196 3197 PF_ACPY(&oia, ia, af); 3198 if (oa) 3199 PF_ACPY(&ooa, oa, af); 3200 3201 /* Change inner protocol port, fix inner protocol checksum. */ 3202 if (ip != NULL) { 3203 u_int16_t oip = *ip; 3204 u_int32_t opc; 3205 3206 if (pc != NULL) 3207 opc = *pc; 3208 *ip = np; 3209 if (pc != NULL) 3210 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3211 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3212 if (pc != NULL) 3213 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3214 } 3215 /* Change inner ip address, fix inner ip and icmp checksums. */ 3216 PF_ACPY(ia, na, af); 3217 switch (af) { 3218 #ifdef INET 3219 case AF_INET: { 3220 u_int32_t oh2c = *h2c; 3221 3222 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3223 oia.addr16[0], ia->addr16[0], 0), 3224 oia.addr16[1], ia->addr16[1], 0); 3225 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3226 oia.addr16[0], ia->addr16[0], 0), 3227 oia.addr16[1], ia->addr16[1], 0); 3228 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3229 break; 3230 } 3231 #endif /* INET */ 3232 #ifdef INET6 3233 case AF_INET6: 3234 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3235 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3236 pf_cksum_fixup(pf_cksum_fixup(*ic, 3237 oia.addr16[0], ia->addr16[0], u), 3238 oia.addr16[1], ia->addr16[1], u), 3239 oia.addr16[2], ia->addr16[2], u), 3240 oia.addr16[3], ia->addr16[3], u), 3241 oia.addr16[4], ia->addr16[4], u), 3242 oia.addr16[5], ia->addr16[5], u), 3243 oia.addr16[6], ia->addr16[6], u), 3244 oia.addr16[7], ia->addr16[7], u); 3245 break; 3246 #endif /* INET6 */ 3247 } 3248 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3249 if (oa) { 3250 PF_ACPY(oa, na, af); 3251 switch (af) { 3252 #ifdef INET 3253 case AF_INET: 3254 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3255 ooa.addr16[0], oa->addr16[0], 0), 3256 ooa.addr16[1], oa->addr16[1], 0); 3257 break; 3258 #endif /* INET */ 3259 #ifdef INET6 3260 case AF_INET6: 3261 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3262 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3263 pf_cksum_fixup(pf_cksum_fixup(*ic, 3264 ooa.addr16[0], oa->addr16[0], u), 3265 ooa.addr16[1], oa->addr16[1], u), 3266 ooa.addr16[2], oa->addr16[2], u), 3267 ooa.addr16[3], oa->addr16[3], u), 3268 ooa.addr16[4], oa->addr16[4], u), 3269 ooa.addr16[5], oa->addr16[5], u), 3270 ooa.addr16[6], oa->addr16[6], u), 3271 ooa.addr16[7], oa->addr16[7], u); 3272 break; 3273 #endif /* INET6 */ 3274 } 3275 } 3276 } 3277 3278 /* 3279 * Need to modulate the sequence numbers in the TCP SACK option 3280 * (credits to Krzysztof Pfaff for report and patch) 3281 */ 3282 static int 3283 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 3284 struct tcphdr *th, struct pf_state_peer *dst) 3285 { 3286 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3287 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3288 int copyback = 0, i, olen; 3289 struct sackblk sack; 3290 3291 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3292 if (hlen < TCPOLEN_SACKLEN || 3293 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3294 return 0; 3295 3296 while (hlen >= TCPOLEN_SACKLEN) { 3297 size_t startoff = opt - opts; 3298 olen = opt[1]; 3299 switch (*opt) { 3300 case TCPOPT_EOL: /* FALLTHROUGH */ 3301 case TCPOPT_NOP: 3302 opt++; 3303 hlen--; 3304 break; 3305 case TCPOPT_SACK: 3306 if (olen > hlen) 3307 olen = hlen; 3308 if (olen >= TCPOLEN_SACKLEN) { 3309 for (i = 2; i + TCPOLEN_SACK <= olen; 3310 i += TCPOLEN_SACK) { 3311 memcpy(&sack, &opt[i], sizeof(sack)); 3312 pf_patch_32_unaligned(m, 3313 &th->th_sum, &sack.start, 3314 htonl(ntohl(sack.start) - dst->seqdiff), 3315 PF_ALGNMNT(startoff), 3316 0); 3317 pf_patch_32_unaligned(m, &th->th_sum, 3318 &sack.end, 3319 htonl(ntohl(sack.end) - dst->seqdiff), 3320 PF_ALGNMNT(startoff), 3321 0); 3322 memcpy(&opt[i], &sack, sizeof(sack)); 3323 } 3324 copyback = 1; 3325 } 3326 /* FALLTHROUGH */ 3327 default: 3328 if (olen < 2) 3329 olen = 2; 3330 hlen -= olen; 3331 opt += olen; 3332 } 3333 } 3334 3335 if (copyback) 3336 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 3337 return (copyback); 3338 } 3339 3340 struct mbuf * 3341 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3342 const struct pf_addr *saddr, const struct pf_addr *daddr, 3343 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3344 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3345 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3346 { 3347 struct mbuf *m; 3348 int len, tlen; 3349 #ifdef INET 3350 struct ip *h = NULL; 3351 #endif /* INET */ 3352 #ifdef INET6 3353 struct ip6_hdr *h6 = NULL; 3354 #endif /* INET6 */ 3355 struct tcphdr *th; 3356 char *opt; 3357 struct pf_mtag *pf_mtag; 3358 3359 len = 0; 3360 th = NULL; 3361 3362 /* maximum segment size tcp option */ 3363 tlen = sizeof(struct tcphdr); 3364 if (mss) 3365 tlen += 4; 3366 3367 switch (af) { 3368 #ifdef INET 3369 case AF_INET: 3370 len = sizeof(struct ip) + tlen; 3371 break; 3372 #endif /* INET */ 3373 #ifdef INET6 3374 case AF_INET6: 3375 len = sizeof(struct ip6_hdr) + tlen; 3376 break; 3377 #endif /* INET6 */ 3378 default: 3379 panic("%s: unsupported af %d", __func__, af); 3380 } 3381 3382 m = m_gethdr(M_NOWAIT, MT_DATA); 3383 if (m == NULL) 3384 return (NULL); 3385 3386 #ifdef MAC 3387 mac_netinet_firewall_send(m); 3388 #endif 3389 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3390 m_freem(m); 3391 return (NULL); 3392 } 3393 if (skip_firewall) 3394 m->m_flags |= M_SKIP_FIREWALL; 3395 pf_mtag->tag = mtag_tag; 3396 pf_mtag->flags = mtag_flags; 3397 3398 if (rtableid >= 0) 3399 M_SETFIB(m, rtableid); 3400 3401 #ifdef ALTQ 3402 if (r != NULL && r->qid) { 3403 pf_mtag->qid = r->qid; 3404 3405 /* add hints for ecn */ 3406 pf_mtag->hdr = mtod(m, struct ip *); 3407 } 3408 #endif /* ALTQ */ 3409 m->m_data += max_linkhdr; 3410 m->m_pkthdr.len = m->m_len = len; 3411 /* The rest of the stack assumes a rcvif, so provide one. 3412 * This is a locally generated packet, so .. close enough. */ 3413 m->m_pkthdr.rcvif = V_loif; 3414 bzero(m->m_data, len); 3415 switch (af) { 3416 #ifdef INET 3417 case AF_INET: 3418 h = mtod(m, struct ip *); 3419 3420 /* IP header fields included in the TCP checksum */ 3421 h->ip_p = IPPROTO_TCP; 3422 h->ip_len = htons(tlen); 3423 h->ip_src.s_addr = saddr->v4.s_addr; 3424 h->ip_dst.s_addr = daddr->v4.s_addr; 3425 3426 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3427 break; 3428 #endif /* INET */ 3429 #ifdef INET6 3430 case AF_INET6: 3431 h6 = mtod(m, struct ip6_hdr *); 3432 3433 /* IP header fields included in the TCP checksum */ 3434 h6->ip6_nxt = IPPROTO_TCP; 3435 h6->ip6_plen = htons(tlen); 3436 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3437 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3438 3439 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3440 break; 3441 #endif /* INET6 */ 3442 } 3443 3444 /* TCP header */ 3445 th->th_sport = sport; 3446 th->th_dport = dport; 3447 th->th_seq = htonl(seq); 3448 th->th_ack = htonl(ack); 3449 th->th_off = tlen >> 2; 3450 th->th_flags = tcp_flags; 3451 th->th_win = htons(win); 3452 3453 if (mss) { 3454 opt = (char *)(th + 1); 3455 opt[0] = TCPOPT_MAXSEG; 3456 opt[1] = 4; 3457 HTONS(mss); 3458 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3459 } 3460 3461 switch (af) { 3462 #ifdef INET 3463 case AF_INET: 3464 /* TCP checksum */ 3465 th->th_sum = in_cksum(m, len); 3466 3467 /* Finish the IP header */ 3468 h->ip_v = 4; 3469 h->ip_hl = sizeof(*h) >> 2; 3470 h->ip_tos = IPTOS_LOWDELAY; 3471 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3472 h->ip_len = htons(len); 3473 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3474 h->ip_sum = 0; 3475 break; 3476 #endif /* INET */ 3477 #ifdef INET6 3478 case AF_INET6: 3479 /* TCP checksum */ 3480 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3481 sizeof(struct ip6_hdr), tlen); 3482 3483 h6->ip6_vfc |= IPV6_VERSION; 3484 h6->ip6_hlim = IPV6_DEFHLIM; 3485 break; 3486 #endif /* INET6 */ 3487 } 3488 3489 return (m); 3490 } 3491 3492 static void 3493 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3494 uint8_t ttl, int rtableid) 3495 { 3496 struct mbuf *m; 3497 #ifdef INET 3498 struct ip *h = NULL; 3499 #endif /* INET */ 3500 #ifdef INET6 3501 struct ip6_hdr *h6 = NULL; 3502 #endif /* INET6 */ 3503 struct sctphdr *hdr; 3504 struct sctp_chunkhdr *chunk; 3505 struct pf_send_entry *pfse; 3506 int off = 0; 3507 3508 MPASS(af == pd->af); 3509 3510 m = m_gethdr(M_NOWAIT, MT_DATA); 3511 if (m == NULL) 3512 return; 3513 3514 m->m_data += max_linkhdr; 3515 m->m_flags |= M_SKIP_FIREWALL; 3516 /* The rest of the stack assumes a rcvif, so provide one. 3517 * This is a locally generated packet, so .. close enough. */ 3518 m->m_pkthdr.rcvif = V_loif; 3519 3520 /* IPv4|6 header */ 3521 switch (af) { 3522 #ifdef INET 3523 case AF_INET: 3524 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3525 3526 h = mtod(m, struct ip *); 3527 3528 /* IP header fields included in the TCP checksum */ 3529 3530 h->ip_p = IPPROTO_SCTP; 3531 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3532 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3533 h->ip_src = pd->dst->v4; 3534 h->ip_dst = pd->src->v4; 3535 3536 off += sizeof(struct ip); 3537 break; 3538 #endif /* INET */ 3539 #ifdef INET6 3540 case AF_INET6: 3541 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3542 3543 h6 = mtod(m, struct ip6_hdr *); 3544 3545 /* IP header fields included in the TCP checksum */ 3546 h6->ip6_vfc |= IPV6_VERSION; 3547 h6->ip6_nxt = IPPROTO_SCTP; 3548 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3549 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3550 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3551 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3552 3553 off += sizeof(struct ip6_hdr); 3554 break; 3555 #endif /* INET6 */ 3556 } 3557 3558 /* SCTP header */ 3559 hdr = mtodo(m, off); 3560 3561 hdr->src_port = pd->hdr.sctp.dest_port; 3562 hdr->dest_port = pd->hdr.sctp.src_port; 3563 hdr->v_tag = pd->sctp_initiate_tag; 3564 hdr->checksum = 0; 3565 3566 /* Abort chunk. */ 3567 off += sizeof(struct sctphdr); 3568 chunk = mtodo(m, off); 3569 3570 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3571 chunk->chunk_length = htons(sizeof(*chunk)); 3572 3573 /* SCTP checksum */ 3574 off += sizeof(*chunk); 3575 m->m_pkthdr.len = m->m_len = off; 3576 3577 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3578 3579 if (rtableid >= 0) 3580 M_SETFIB(m, rtableid); 3581 3582 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3583 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3584 if (pfse == NULL) { 3585 m_freem(m); 3586 return; 3587 } 3588 3589 switch (af) { 3590 #ifdef INET 3591 case AF_INET: 3592 pfse->pfse_type = PFSE_IP; 3593 break; 3594 #endif /* INET */ 3595 #ifdef INET6 3596 case AF_INET6: 3597 pfse->pfse_type = PFSE_IP6; 3598 break; 3599 #endif /* INET6 */ 3600 } 3601 3602 pfse->pfse_m = m; 3603 pf_send(pfse); 3604 } 3605 3606 void 3607 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3608 const struct pf_addr *saddr, const struct pf_addr *daddr, 3609 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3610 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3611 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3612 { 3613 struct pf_send_entry *pfse; 3614 struct mbuf *m; 3615 3616 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3617 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3618 if (m == NULL) 3619 return; 3620 3621 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3622 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3623 if (pfse == NULL) { 3624 m_freem(m); 3625 return; 3626 } 3627 3628 switch (af) { 3629 #ifdef INET 3630 case AF_INET: 3631 pfse->pfse_type = PFSE_IP; 3632 break; 3633 #endif /* INET */ 3634 #ifdef INET6 3635 case AF_INET6: 3636 pfse->pfse_type = PFSE_IP6; 3637 break; 3638 #endif /* INET6 */ 3639 } 3640 3641 pfse->pfse_m = m; 3642 pf_send(pfse); 3643 } 3644 3645 static void 3646 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3647 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3648 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3649 u_short *reason, int rtableid) 3650 { 3651 struct pf_addr * const saddr = pd->src; 3652 struct pf_addr * const daddr = pd->dst; 3653 sa_family_t af = pd->af; 3654 3655 /* undo NAT changes, if they have taken place */ 3656 if (nr != NULL) { 3657 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3658 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3659 if (pd->sport) 3660 *pd->sport = sk->port[pd->sidx]; 3661 if (pd->dport) 3662 *pd->dport = sk->port[pd->didx]; 3663 if (pd->proto_sum) 3664 *pd->proto_sum = bproto_sum; 3665 if (pd->ip_sum) 3666 *pd->ip_sum = bip_sum; 3667 m_copyback(m, off, hdrlen, pd->hdr.any); 3668 } 3669 if (pd->proto == IPPROTO_TCP && 3670 ((r->rule_flag & PFRULE_RETURNRST) || 3671 (r->rule_flag & PFRULE_RETURN)) && 3672 !(th->th_flags & TH_RST)) { 3673 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3674 int len = 0; 3675 #ifdef INET 3676 struct ip *h4; 3677 #endif 3678 #ifdef INET6 3679 struct ip6_hdr *h6; 3680 #endif 3681 3682 switch (af) { 3683 #ifdef INET 3684 case AF_INET: 3685 h4 = mtod(m, struct ip *); 3686 len = ntohs(h4->ip_len) - off; 3687 break; 3688 #endif 3689 #ifdef INET6 3690 case AF_INET6: 3691 h6 = mtod(m, struct ip6_hdr *); 3692 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3693 break; 3694 #endif 3695 } 3696 3697 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3698 REASON_SET(reason, PFRES_PROTCKSUM); 3699 else { 3700 if (th->th_flags & TH_SYN) 3701 ack++; 3702 if (th->th_flags & TH_FIN) 3703 ack++; 3704 pf_send_tcp(r, af, pd->dst, 3705 pd->src, th->th_dport, th->th_sport, 3706 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3707 r->return_ttl, true, 0, 0, rtableid); 3708 } 3709 } else if (pd->proto == IPPROTO_SCTP && 3710 (r->rule_flag & PFRULE_RETURN)) { 3711 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3712 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3713 r->return_icmp) 3714 pf_send_icmp(m, r->return_icmp >> 8, 3715 r->return_icmp & 255, af, r, rtableid); 3716 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3717 r->return_icmp6) 3718 pf_send_icmp(m, r->return_icmp6 >> 8, 3719 r->return_icmp6 & 255, af, r, rtableid); 3720 } 3721 3722 static int 3723 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3724 { 3725 struct m_tag *mtag; 3726 u_int8_t mpcp; 3727 3728 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3729 if (mtag == NULL) 3730 return (0); 3731 3732 if (prio == PF_PRIO_ZERO) 3733 prio = 0; 3734 3735 mpcp = *(uint8_t *)(mtag + 1); 3736 3737 return (mpcp == prio); 3738 } 3739 3740 static int 3741 pf_icmp_to_bandlim(uint8_t type) 3742 { 3743 switch (type) { 3744 case ICMP_ECHO: 3745 case ICMP_ECHOREPLY: 3746 return (BANDLIM_ICMP_ECHO); 3747 case ICMP_TSTAMP: 3748 case ICMP_TSTAMPREPLY: 3749 return (BANDLIM_ICMP_TSTAMP); 3750 case ICMP_UNREACH: 3751 default: 3752 return (BANDLIM_ICMP_UNREACH); 3753 } 3754 } 3755 3756 static void 3757 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3758 struct pf_krule *r, int rtableid) 3759 { 3760 struct pf_send_entry *pfse; 3761 struct mbuf *m0; 3762 struct pf_mtag *pf_mtag; 3763 3764 /* ICMP packet rate limitation. */ 3765 #ifdef INET6 3766 if (af == AF_INET6) { 3767 if (icmp6_ratelimit(NULL, type, code)) 3768 return; 3769 } 3770 #endif 3771 #ifdef INET 3772 if (af == AF_INET) { 3773 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3774 return; 3775 } 3776 #endif 3777 3778 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3779 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3780 if (pfse == NULL) 3781 return; 3782 3783 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3784 free(pfse, M_PFTEMP); 3785 return; 3786 } 3787 3788 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3789 free(pfse, M_PFTEMP); 3790 return; 3791 } 3792 /* XXX: revisit */ 3793 m0->m_flags |= M_SKIP_FIREWALL; 3794 3795 if (rtableid >= 0) 3796 M_SETFIB(m0, rtableid); 3797 3798 #ifdef ALTQ 3799 if (r->qid) { 3800 pf_mtag->qid = r->qid; 3801 /* add hints for ecn */ 3802 pf_mtag->hdr = mtod(m0, struct ip *); 3803 } 3804 #endif /* ALTQ */ 3805 3806 switch (af) { 3807 #ifdef INET 3808 case AF_INET: 3809 pfse->pfse_type = PFSE_ICMP; 3810 break; 3811 #endif /* INET */ 3812 #ifdef INET6 3813 case AF_INET6: 3814 pfse->pfse_type = PFSE_ICMP6; 3815 break; 3816 #endif /* INET6 */ 3817 } 3818 pfse->pfse_m = m0; 3819 pfse->icmpopts.type = type; 3820 pfse->icmpopts.code = code; 3821 pf_send(pfse); 3822 } 3823 3824 /* 3825 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3826 * If n is 0, they match if they are equal. If n is != 0, they match if they 3827 * are different. 3828 */ 3829 int 3830 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3831 struct pf_addr *b, sa_family_t af) 3832 { 3833 int match = 0; 3834 3835 switch (af) { 3836 #ifdef INET 3837 case AF_INET: 3838 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3839 match++; 3840 break; 3841 #endif /* INET */ 3842 #ifdef INET6 3843 case AF_INET6: 3844 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3845 match++; 3846 break; 3847 #endif /* INET6 */ 3848 } 3849 if (match) { 3850 if (n) 3851 return (0); 3852 else 3853 return (1); 3854 } else { 3855 if (n) 3856 return (1); 3857 else 3858 return (0); 3859 } 3860 } 3861 3862 /* 3863 * Return 1 if b <= a <= e, otherwise return 0. 3864 */ 3865 int 3866 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3867 struct pf_addr *a, sa_family_t af) 3868 { 3869 switch (af) { 3870 #ifdef INET 3871 case AF_INET: 3872 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3873 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3874 return (0); 3875 break; 3876 #endif /* INET */ 3877 #ifdef INET6 3878 case AF_INET6: { 3879 int i; 3880 3881 /* check a >= b */ 3882 for (i = 0; i < 4; ++i) 3883 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3884 break; 3885 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3886 return (0); 3887 /* check a <= e */ 3888 for (i = 0; i < 4; ++i) 3889 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3890 break; 3891 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3892 return (0); 3893 break; 3894 } 3895 #endif /* INET6 */ 3896 } 3897 return (1); 3898 } 3899 3900 static int 3901 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3902 { 3903 switch (op) { 3904 case PF_OP_IRG: 3905 return ((p > a1) && (p < a2)); 3906 case PF_OP_XRG: 3907 return ((p < a1) || (p > a2)); 3908 case PF_OP_RRG: 3909 return ((p >= a1) && (p <= a2)); 3910 case PF_OP_EQ: 3911 return (p == a1); 3912 case PF_OP_NE: 3913 return (p != a1); 3914 case PF_OP_LT: 3915 return (p < a1); 3916 case PF_OP_LE: 3917 return (p <= a1); 3918 case PF_OP_GT: 3919 return (p > a1); 3920 case PF_OP_GE: 3921 return (p >= a1); 3922 } 3923 return (0); /* never reached */ 3924 } 3925 3926 int 3927 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3928 { 3929 NTOHS(a1); 3930 NTOHS(a2); 3931 NTOHS(p); 3932 return (pf_match(op, a1, a2, p)); 3933 } 3934 3935 static int 3936 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3937 { 3938 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3939 return (0); 3940 return (pf_match(op, a1, a2, u)); 3941 } 3942 3943 static int 3944 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3945 { 3946 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3947 return (0); 3948 return (pf_match(op, a1, a2, g)); 3949 } 3950 3951 int 3952 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3953 { 3954 if (*tag == -1) 3955 *tag = mtag; 3956 3957 return ((!r->match_tag_not && r->match_tag == *tag) || 3958 (r->match_tag_not && r->match_tag != *tag)); 3959 } 3960 3961 static int 3962 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3963 { 3964 struct ifnet *ifp = m->m_pkthdr.rcvif; 3965 struct pfi_kkif *kif; 3966 3967 if (ifp == NULL) 3968 return (0); 3969 3970 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3971 3972 if (kif == NULL) { 3973 DPFPRINTF(PF_DEBUG_URGENT, 3974 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3975 r->rcv_ifname)); 3976 return (0); 3977 } 3978 3979 return (pfi_kkif_match(r->rcv_kif, kif)); 3980 } 3981 3982 int 3983 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3984 { 3985 3986 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3987 3988 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3989 return (ENOMEM); 3990 3991 pd->pf_mtag->tag = tag; 3992 3993 return (0); 3994 } 3995 3996 #define PF_ANCHOR_STACKSIZE 32 3997 struct pf_kanchor_stackframe { 3998 struct pf_kruleset *rs; 3999 struct pf_krule *r; /* XXX: + match bit */ 4000 struct pf_kanchor *child; 4001 }; 4002 4003 /* 4004 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4005 */ 4006 #define PF_ANCHORSTACK_MATCH 0x00000001 4007 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4008 4009 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4010 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4011 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4012 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4013 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4014 } while (0) 4015 4016 void 4017 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4018 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4019 int *match) 4020 { 4021 struct pf_kanchor_stackframe *f; 4022 4023 PF_RULES_RASSERT(); 4024 4025 if (match) 4026 *match = 0; 4027 if (*depth >= PF_ANCHOR_STACKSIZE) { 4028 printf("%s: anchor stack overflow on %s\n", 4029 __func__, (*r)->anchor->name); 4030 *r = TAILQ_NEXT(*r, entries); 4031 return; 4032 } else if (*depth == 0 && a != NULL) 4033 *a = *r; 4034 f = stack + (*depth)++; 4035 f->rs = *rs; 4036 f->r = *r; 4037 if ((*r)->anchor_wildcard) { 4038 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4039 4040 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4041 *r = NULL; 4042 return; 4043 } 4044 *rs = &f->child->ruleset; 4045 } else { 4046 f->child = NULL; 4047 *rs = &(*r)->anchor->ruleset; 4048 } 4049 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4050 } 4051 4052 int 4053 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4054 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4055 int *match) 4056 { 4057 struct pf_kanchor_stackframe *f; 4058 struct pf_krule *fr; 4059 int quick = 0; 4060 4061 PF_RULES_RASSERT(); 4062 4063 do { 4064 if (*depth <= 0) 4065 break; 4066 f = stack + *depth - 1; 4067 fr = PF_ANCHOR_RULE(f); 4068 if (f->child != NULL) { 4069 /* 4070 * This block traverses through 4071 * a wildcard anchor. 4072 */ 4073 if (match != NULL && *match) { 4074 /* 4075 * If any of "*" matched, then 4076 * "foo/ *" matched, mark frame 4077 * appropriately. 4078 */ 4079 PF_ANCHOR_SET_MATCH(f); 4080 *match = 0; 4081 } 4082 f->child = RB_NEXT(pf_kanchor_node, 4083 &fr->anchor->children, f->child); 4084 if (f->child != NULL) { 4085 *rs = &f->child->ruleset; 4086 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4087 if (*r == NULL) 4088 continue; 4089 else 4090 break; 4091 } 4092 } 4093 (*depth)--; 4094 if (*depth == 0 && a != NULL) 4095 *a = NULL; 4096 *rs = f->rs; 4097 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4098 quick = fr->quick; 4099 *r = TAILQ_NEXT(fr, entries); 4100 } while (*r == NULL); 4101 4102 return (quick); 4103 } 4104 4105 struct pf_keth_anchor_stackframe { 4106 struct pf_keth_ruleset *rs; 4107 struct pf_keth_rule *r; /* XXX: + match bit */ 4108 struct pf_keth_anchor *child; 4109 }; 4110 4111 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4112 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4113 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4114 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4115 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4116 } while (0) 4117 4118 void 4119 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4120 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4121 struct pf_keth_rule **a, int *match) 4122 { 4123 struct pf_keth_anchor_stackframe *f; 4124 4125 NET_EPOCH_ASSERT(); 4126 4127 if (match) 4128 *match = 0; 4129 if (*depth >= PF_ANCHOR_STACKSIZE) { 4130 printf("%s: anchor stack overflow on %s\n", 4131 __func__, (*r)->anchor->name); 4132 *r = TAILQ_NEXT(*r, entries); 4133 return; 4134 } else if (*depth == 0 && a != NULL) 4135 *a = *r; 4136 f = stack + (*depth)++; 4137 f->rs = *rs; 4138 f->r = *r; 4139 if ((*r)->anchor_wildcard) { 4140 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4141 4142 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4143 *r = NULL; 4144 return; 4145 } 4146 *rs = &f->child->ruleset; 4147 } else { 4148 f->child = NULL; 4149 *rs = &(*r)->anchor->ruleset; 4150 } 4151 *r = TAILQ_FIRST((*rs)->active.rules); 4152 } 4153 4154 int 4155 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4156 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4157 struct pf_keth_rule **a, int *match) 4158 { 4159 struct pf_keth_anchor_stackframe *f; 4160 struct pf_keth_rule *fr; 4161 int quick = 0; 4162 4163 NET_EPOCH_ASSERT(); 4164 4165 do { 4166 if (*depth <= 0) 4167 break; 4168 f = stack + *depth - 1; 4169 fr = PF_ETH_ANCHOR_RULE(f); 4170 if (f->child != NULL) { 4171 /* 4172 * This block traverses through 4173 * a wildcard anchor. 4174 */ 4175 if (match != NULL && *match) { 4176 /* 4177 * If any of "*" matched, then 4178 * "foo/ *" matched, mark frame 4179 * appropriately. 4180 */ 4181 PF_ETH_ANCHOR_SET_MATCH(f); 4182 *match = 0; 4183 } 4184 f->child = RB_NEXT(pf_keth_anchor_node, 4185 &fr->anchor->children, f->child); 4186 if (f->child != NULL) { 4187 *rs = &f->child->ruleset; 4188 *r = TAILQ_FIRST((*rs)->active.rules); 4189 if (*r == NULL) 4190 continue; 4191 else 4192 break; 4193 } 4194 } 4195 (*depth)--; 4196 if (*depth == 0 && a != NULL) 4197 *a = NULL; 4198 *rs = f->rs; 4199 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4200 quick = fr->quick; 4201 *r = TAILQ_NEXT(fr, entries); 4202 } while (*r == NULL); 4203 4204 return (quick); 4205 } 4206 4207 #ifdef INET6 4208 void 4209 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4210 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4211 { 4212 switch (af) { 4213 #ifdef INET 4214 case AF_INET: 4215 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4216 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4217 break; 4218 #endif /* INET */ 4219 case AF_INET6: 4220 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4221 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4222 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4223 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4224 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4225 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4226 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4227 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4228 break; 4229 } 4230 } 4231 4232 void 4233 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4234 { 4235 switch (af) { 4236 #ifdef INET 4237 case AF_INET: 4238 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4239 break; 4240 #endif /* INET */ 4241 case AF_INET6: 4242 if (addr->addr32[3] == 0xffffffff) { 4243 addr->addr32[3] = 0; 4244 if (addr->addr32[2] == 0xffffffff) { 4245 addr->addr32[2] = 0; 4246 if (addr->addr32[1] == 0xffffffff) { 4247 addr->addr32[1] = 0; 4248 addr->addr32[0] = 4249 htonl(ntohl(addr->addr32[0]) + 1); 4250 } else 4251 addr->addr32[1] = 4252 htonl(ntohl(addr->addr32[1]) + 1); 4253 } else 4254 addr->addr32[2] = 4255 htonl(ntohl(addr->addr32[2]) + 1); 4256 } else 4257 addr->addr32[3] = 4258 htonl(ntohl(addr->addr32[3]) + 1); 4259 break; 4260 } 4261 } 4262 #endif /* INET6 */ 4263 4264 void 4265 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4266 { 4267 /* 4268 * Modern rules use the same flags in rules as they do in states. 4269 */ 4270 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4271 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4272 4273 /* 4274 * Old-style scrub rules have different flags which need to be translated. 4275 */ 4276 if (r->rule_flag & PFRULE_RANDOMID) 4277 a->flags |= PFSTATE_RANDOMID; 4278 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4279 a->flags |= PFSTATE_SETTOS; 4280 a->set_tos = r->set_tos; 4281 } 4282 4283 if (r->qid) 4284 a->qid = r->qid; 4285 if (r->pqid) 4286 a->pqid = r->pqid; 4287 if (r->rtableid >= 0) 4288 a->rtableid = r->rtableid; 4289 a->log |= r->log; 4290 if (r->min_ttl) 4291 a->min_ttl = r->min_ttl; 4292 if (r->max_mss) 4293 a->max_mss = r->max_mss; 4294 if (r->dnpipe) 4295 a->dnpipe = r->dnpipe; 4296 if (r->dnrpipe) 4297 a->dnrpipe = r->dnrpipe; 4298 if (r->dnpipe || r->dnrpipe) { 4299 if (r->free_flags & PFRULE_DN_IS_PIPE) 4300 a->flags |= PFSTATE_DN_IS_PIPE; 4301 else 4302 a->flags &= ~PFSTATE_DN_IS_PIPE; 4303 } 4304 if (r->scrub_flags & PFSTATE_SETPRIO) { 4305 a->set_prio[0] = r->set_prio[0]; 4306 a->set_prio[1] = r->set_prio[1]; 4307 } 4308 } 4309 4310 int 4311 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 4312 { 4313 struct pf_addr *saddr, *daddr; 4314 u_int16_t sport, dport; 4315 struct inpcbinfo *pi; 4316 struct inpcb *inp; 4317 4318 pd->lookup.uid = UID_MAX; 4319 pd->lookup.gid = GID_MAX; 4320 4321 switch (pd->proto) { 4322 case IPPROTO_TCP: 4323 sport = pd->hdr.tcp.th_sport; 4324 dport = pd->hdr.tcp.th_dport; 4325 pi = &V_tcbinfo; 4326 break; 4327 case IPPROTO_UDP: 4328 sport = pd->hdr.udp.uh_sport; 4329 dport = pd->hdr.udp.uh_dport; 4330 pi = &V_udbinfo; 4331 break; 4332 default: 4333 return (-1); 4334 } 4335 if (pd->dir == PF_IN) { 4336 saddr = pd->src; 4337 daddr = pd->dst; 4338 } else { 4339 u_int16_t p; 4340 4341 p = sport; 4342 sport = dport; 4343 dport = p; 4344 saddr = pd->dst; 4345 daddr = pd->src; 4346 } 4347 switch (pd->af) { 4348 #ifdef INET 4349 case AF_INET: 4350 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4351 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4352 if (inp == NULL) { 4353 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4354 daddr->v4, dport, INPLOOKUP_WILDCARD | 4355 INPLOOKUP_RLOCKPCB, NULL, m); 4356 if (inp == NULL) 4357 return (-1); 4358 } 4359 break; 4360 #endif /* INET */ 4361 #ifdef INET6 4362 case AF_INET6: 4363 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4364 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4365 if (inp == NULL) { 4366 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4367 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4368 INPLOOKUP_RLOCKPCB, NULL, m); 4369 if (inp == NULL) 4370 return (-1); 4371 } 4372 break; 4373 #endif /* INET6 */ 4374 4375 default: 4376 return (-1); 4377 } 4378 INP_RLOCK_ASSERT(inp); 4379 pd->lookup.uid = inp->inp_cred->cr_uid; 4380 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4381 INP_RUNLOCK(inp); 4382 4383 return (1); 4384 } 4385 4386 u_int8_t 4387 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4388 { 4389 int hlen; 4390 u_int8_t hdr[60]; 4391 u_int8_t *opt, optlen; 4392 u_int8_t wscale = 0; 4393 4394 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4395 if (hlen <= sizeof(struct tcphdr)) 4396 return (0); 4397 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4398 return (0); 4399 opt = hdr + sizeof(struct tcphdr); 4400 hlen -= sizeof(struct tcphdr); 4401 while (hlen >= 3) { 4402 switch (*opt) { 4403 case TCPOPT_EOL: 4404 case TCPOPT_NOP: 4405 ++opt; 4406 --hlen; 4407 break; 4408 case TCPOPT_WINDOW: 4409 wscale = opt[2]; 4410 if (wscale > TCP_MAX_WINSHIFT) 4411 wscale = TCP_MAX_WINSHIFT; 4412 wscale |= PF_WSCALE_FLAG; 4413 /* FALLTHROUGH */ 4414 default: 4415 optlen = opt[1]; 4416 if (optlen < 2) 4417 optlen = 2; 4418 hlen -= optlen; 4419 opt += optlen; 4420 break; 4421 } 4422 } 4423 return (wscale); 4424 } 4425 4426 u_int16_t 4427 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4428 { 4429 int hlen; 4430 u_int8_t hdr[60]; 4431 u_int8_t *opt, optlen; 4432 u_int16_t mss = V_tcp_mssdflt; 4433 4434 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4435 if (hlen <= sizeof(struct tcphdr)) 4436 return (0); 4437 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4438 return (0); 4439 opt = hdr + sizeof(struct tcphdr); 4440 hlen -= sizeof(struct tcphdr); 4441 while (hlen >= TCPOLEN_MAXSEG) { 4442 switch (*opt) { 4443 case TCPOPT_EOL: 4444 case TCPOPT_NOP: 4445 ++opt; 4446 --hlen; 4447 break; 4448 case TCPOPT_MAXSEG: 4449 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4450 NTOHS(mss); 4451 /* FALLTHROUGH */ 4452 default: 4453 optlen = opt[1]; 4454 if (optlen < 2) 4455 optlen = 2; 4456 hlen -= optlen; 4457 opt += optlen; 4458 break; 4459 } 4460 } 4461 return (mss); 4462 } 4463 4464 static u_int16_t 4465 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4466 { 4467 struct nhop_object *nh; 4468 #ifdef INET6 4469 struct in6_addr dst6; 4470 uint32_t scopeid; 4471 #endif /* INET6 */ 4472 int hlen = 0; 4473 uint16_t mss = 0; 4474 4475 NET_EPOCH_ASSERT(); 4476 4477 switch (af) { 4478 #ifdef INET 4479 case AF_INET: 4480 hlen = sizeof(struct ip); 4481 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4482 if (nh != NULL) 4483 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4484 break; 4485 #endif /* INET */ 4486 #ifdef INET6 4487 case AF_INET6: 4488 hlen = sizeof(struct ip6_hdr); 4489 in6_splitscope(&addr->v6, &dst6, &scopeid); 4490 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4491 if (nh != NULL) 4492 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4493 break; 4494 #endif /* INET6 */ 4495 } 4496 4497 mss = max(V_tcp_mssdflt, mss); 4498 mss = min(mss, offer); 4499 mss = max(mss, 64); /* sanity - at least max opt space */ 4500 return (mss); 4501 } 4502 4503 static u_int32_t 4504 pf_tcp_iss(struct pf_pdesc *pd) 4505 { 4506 MD5_CTX ctx; 4507 u_int32_t digest[4]; 4508 4509 if (V_pf_tcp_secret_init == 0) { 4510 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4511 MD5Init(&V_pf_tcp_secret_ctx); 4512 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4513 sizeof(V_pf_tcp_secret)); 4514 V_pf_tcp_secret_init = 1; 4515 } 4516 4517 ctx = V_pf_tcp_secret_ctx; 4518 4519 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4520 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4521 if (pd->af == AF_INET6) { 4522 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4523 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4524 } else { 4525 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4526 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4527 } 4528 MD5Final((u_char *)digest, &ctx); 4529 V_pf_tcp_iss_off += 4096; 4530 #define ISN_RANDOM_INCREMENT (4096 - 1) 4531 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4532 V_pf_tcp_iss_off); 4533 #undef ISN_RANDOM_INCREMENT 4534 } 4535 4536 static bool 4537 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4538 { 4539 bool match = true; 4540 4541 /* Always matches if not set */ 4542 if (! r->isset) 4543 return (!r->neg); 4544 4545 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4546 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4547 match = false; 4548 break; 4549 } 4550 } 4551 4552 return (match ^ r->neg); 4553 } 4554 4555 static int 4556 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4557 { 4558 if (*tag == -1) 4559 *tag = mtag; 4560 4561 return ((!r->match_tag_not && r->match_tag == *tag) || 4562 (r->match_tag_not && r->match_tag != *tag)); 4563 } 4564 4565 static void 4566 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4567 { 4568 /* If we don't have the interface drop the packet. */ 4569 if (ifp == NULL) { 4570 m_freem(m); 4571 return; 4572 } 4573 4574 switch (ifp->if_type) { 4575 case IFT_ETHER: 4576 case IFT_XETHER: 4577 case IFT_L2VLAN: 4578 case IFT_BRIDGE: 4579 case IFT_IEEE8023ADLAG: 4580 break; 4581 default: 4582 m_freem(m); 4583 return; 4584 } 4585 4586 ifp->if_transmit(ifp, m); 4587 } 4588 4589 static int 4590 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4591 { 4592 #ifdef INET 4593 struct ip ip; 4594 #endif 4595 #ifdef INET6 4596 struct ip6_hdr ip6; 4597 #endif 4598 struct mbuf *m = *m0; 4599 struct ether_header *e; 4600 struct pf_keth_rule *r, *rm, *a = NULL; 4601 struct pf_keth_ruleset *ruleset = NULL; 4602 struct pf_mtag *mtag; 4603 struct pf_keth_ruleq *rules; 4604 struct pf_addr *src = NULL, *dst = NULL; 4605 struct pfi_kkif *bridge_to; 4606 sa_family_t af = 0; 4607 uint16_t proto; 4608 int asd = 0, match = 0; 4609 int tag = -1; 4610 uint8_t action; 4611 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4612 4613 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4614 NET_EPOCH_ASSERT(); 4615 4616 PF_RULES_RLOCK_TRACKER; 4617 4618 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4619 4620 mtag = pf_find_mtag(m); 4621 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4622 /* Dummynet re-injects packets after they've 4623 * completed their delay. We've already 4624 * processed them, so pass unconditionally. */ 4625 4626 /* But only once. We may see the packet multiple times (e.g. 4627 * PFIL_IN/PFIL_OUT). */ 4628 pf_dummynet_flag_remove(m, mtag); 4629 4630 return (PF_PASS); 4631 } 4632 4633 ruleset = V_pf_keth; 4634 rules = ck_pr_load_ptr(&ruleset->active.rules); 4635 r = TAILQ_FIRST(rules); 4636 rm = NULL; 4637 4638 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4639 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4640 DPFPRINTF(PF_DEBUG_URGENT, 4641 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4642 ", pullup failed\n")); 4643 return (PF_DROP); 4644 } 4645 e = mtod(m, struct ether_header *); 4646 proto = ntohs(e->ether_type); 4647 4648 switch (proto) { 4649 #ifdef INET 4650 case ETHERTYPE_IP: { 4651 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4652 sizeof(ip))) 4653 return (PF_DROP); 4654 4655 af = AF_INET; 4656 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4657 (caddr_t)&ip); 4658 src = (struct pf_addr *)&ip.ip_src; 4659 dst = (struct pf_addr *)&ip.ip_dst; 4660 break; 4661 } 4662 #endif /* INET */ 4663 #ifdef INET6 4664 case ETHERTYPE_IPV6: { 4665 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4666 sizeof(ip6))) 4667 return (PF_DROP); 4668 4669 af = AF_INET6; 4670 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4671 (caddr_t)&ip6); 4672 src = (struct pf_addr *)&ip6.ip6_src; 4673 dst = (struct pf_addr *)&ip6.ip6_dst; 4674 break; 4675 } 4676 #endif /* INET6 */ 4677 } 4678 4679 PF_RULES_RLOCK(); 4680 4681 while (r != NULL) { 4682 counter_u64_add(r->evaluations, 1); 4683 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4684 4685 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4686 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4687 "kif"); 4688 r = r->skip[PFE_SKIP_IFP].ptr; 4689 } 4690 else if (r->direction && r->direction != dir) { 4691 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4692 "dir"); 4693 r = r->skip[PFE_SKIP_DIR].ptr; 4694 } 4695 else if (r->proto && r->proto != proto) { 4696 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4697 "proto"); 4698 r = r->skip[PFE_SKIP_PROTO].ptr; 4699 } 4700 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4701 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4702 "src"); 4703 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4704 } 4705 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4706 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4707 "dst"); 4708 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4709 } 4710 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4711 r->ipsrc.neg, kif, M_GETFIB(m))) { 4712 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4713 "ip_src"); 4714 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4715 } 4716 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4717 r->ipdst.neg, kif, M_GETFIB(m))) { 4718 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4719 "ip_dst"); 4720 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4721 } 4722 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4723 mtag ? mtag->tag : 0)) { 4724 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4725 "match_tag"); 4726 r = TAILQ_NEXT(r, entries); 4727 } 4728 else { 4729 if (r->tag) 4730 tag = r->tag; 4731 if (r->anchor == NULL) { 4732 /* Rule matches */ 4733 rm = r; 4734 4735 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4736 4737 if (r->quick) 4738 break; 4739 4740 r = TAILQ_NEXT(r, entries); 4741 } else { 4742 pf_step_into_keth_anchor(anchor_stack, &asd, 4743 &ruleset, &r, &a, &match); 4744 } 4745 } 4746 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4747 &ruleset, &r, &a, &match)) 4748 break; 4749 } 4750 4751 r = rm; 4752 4753 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4754 4755 /* Default to pass. */ 4756 if (r == NULL) { 4757 PF_RULES_RUNLOCK(); 4758 return (PF_PASS); 4759 } 4760 4761 /* Execute action. */ 4762 counter_u64_add(r->packets[dir == PF_OUT], 1); 4763 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4764 pf_update_timestamp(r); 4765 4766 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4767 if (r->action == PF_DROP) { 4768 PF_RULES_RUNLOCK(); 4769 return (PF_DROP); 4770 } 4771 4772 if (tag > 0) { 4773 if (mtag == NULL) 4774 mtag = pf_get_mtag(m); 4775 if (mtag == NULL) { 4776 PF_RULES_RUNLOCK(); 4777 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4778 return (PF_DROP); 4779 } 4780 mtag->tag = tag; 4781 } 4782 4783 if (r->qid != 0) { 4784 if (mtag == NULL) 4785 mtag = pf_get_mtag(m); 4786 if (mtag == NULL) { 4787 PF_RULES_RUNLOCK(); 4788 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4789 return (PF_DROP); 4790 } 4791 mtag->qid = r->qid; 4792 } 4793 4794 action = r->action; 4795 bridge_to = r->bridge_to; 4796 4797 /* Dummynet */ 4798 if (r->dnpipe) { 4799 struct ip_fw_args dnflow; 4800 4801 /* Drop packet if dummynet is not loaded. */ 4802 if (ip_dn_io_ptr == NULL) { 4803 PF_RULES_RUNLOCK(); 4804 m_freem(m); 4805 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4806 return (PF_DROP); 4807 } 4808 if (mtag == NULL) 4809 mtag = pf_get_mtag(m); 4810 if (mtag == NULL) { 4811 PF_RULES_RUNLOCK(); 4812 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4813 return (PF_DROP); 4814 } 4815 4816 bzero(&dnflow, sizeof(dnflow)); 4817 4818 /* We don't have port numbers here, so we set 0. That means 4819 * that we'll be somewhat limited in distinguishing flows (i.e. 4820 * only based on IP addresses, not based on port numbers), but 4821 * it's better than nothing. */ 4822 dnflow.f_id.dst_port = 0; 4823 dnflow.f_id.src_port = 0; 4824 dnflow.f_id.proto = 0; 4825 4826 dnflow.rule.info = r->dnpipe; 4827 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4828 if (r->dnflags & PFRULE_DN_IS_PIPE) 4829 dnflow.rule.info |= IPFW_IS_PIPE; 4830 4831 dnflow.f_id.extra = dnflow.rule.info; 4832 4833 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4834 dnflow.flags |= IPFW_ARGS_ETHER; 4835 dnflow.ifp = kif->pfik_ifp; 4836 4837 switch (af) { 4838 case AF_INET: 4839 dnflow.f_id.addr_type = 4; 4840 dnflow.f_id.src_ip = src->v4.s_addr; 4841 dnflow.f_id.dst_ip = dst->v4.s_addr; 4842 break; 4843 case AF_INET6: 4844 dnflow.flags |= IPFW_ARGS_IP6; 4845 dnflow.f_id.addr_type = 6; 4846 dnflow.f_id.src_ip6 = src->v6; 4847 dnflow.f_id.dst_ip6 = dst->v6; 4848 break; 4849 } 4850 4851 PF_RULES_RUNLOCK(); 4852 4853 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4854 ip_dn_io_ptr(m0, &dnflow); 4855 if (*m0 != NULL) 4856 pf_dummynet_flag_remove(m, mtag); 4857 } else { 4858 PF_RULES_RUNLOCK(); 4859 } 4860 4861 if (action == PF_PASS && bridge_to) { 4862 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4863 *m0 = NULL; /* We've eaten the packet. */ 4864 } 4865 4866 return (action); 4867 } 4868 4869 static int 4870 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4871 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4872 struct pf_kruleset **rsm, struct inpcb *inp, int hdrlen) 4873 { 4874 struct pf_krule *nr = NULL; 4875 struct pf_addr * const saddr = pd->src; 4876 struct pf_addr * const daddr = pd->dst; 4877 sa_family_t af = pd->af; 4878 struct pf_krule *r, *a = NULL; 4879 struct pf_kruleset *ruleset = NULL; 4880 struct pf_krule_slist match_rules; 4881 struct pf_krule_item *ri; 4882 struct pf_ksrc_node *nsn = NULL; 4883 struct tcphdr *th = &pd->hdr.tcp; 4884 struct pf_state_key *sk = NULL, *nk = NULL; 4885 u_short reason, transerror; 4886 int rewrite = 0; 4887 int tag = -1; 4888 int asd = 0; 4889 int match = 0; 4890 int state_icmp = 0, icmp_dir, multi; 4891 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4892 u_int16_t bproto_sum = 0, bip_sum = 0; 4893 u_int8_t icmptype = 0, icmpcode = 0; 4894 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4895 struct pf_udp_mapping *udp_mapping = NULL; 4896 4897 PF_RULES_RASSERT(); 4898 4899 SLIST_INIT(&match_rules); 4900 4901 if (inp != NULL) { 4902 INP_LOCK_ASSERT(inp); 4903 pd->lookup.uid = inp->inp_cred->cr_uid; 4904 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4905 pd->lookup.done = 1; 4906 } 4907 4908 switch (pd->proto) { 4909 case IPPROTO_TCP: 4910 sport = th->th_sport; 4911 dport = th->th_dport; 4912 break; 4913 case IPPROTO_UDP: 4914 sport = pd->hdr.udp.uh_sport; 4915 dport = pd->hdr.udp.uh_dport; 4916 break; 4917 case IPPROTO_SCTP: 4918 sport = pd->hdr.sctp.src_port; 4919 dport = pd->hdr.sctp.dest_port; 4920 break; 4921 #ifdef INET 4922 case IPPROTO_ICMP: 4923 MPASS(af == AF_INET); 4924 icmptype = pd->hdr.icmp.icmp_type; 4925 icmpcode = pd->hdr.icmp.icmp_code; 4926 state_icmp = pf_icmp_mapping(pd, icmptype, 4927 &icmp_dir, &multi, &virtual_id, &virtual_type); 4928 if (icmp_dir == PF_IN) { 4929 sport = virtual_id; 4930 dport = virtual_type; 4931 } else { 4932 sport = virtual_type; 4933 dport = virtual_id; 4934 } 4935 break; 4936 #endif /* INET */ 4937 #ifdef INET6 4938 case IPPROTO_ICMPV6: 4939 MPASS(af == AF_INET6); 4940 icmptype = pd->hdr.icmp6.icmp6_type; 4941 icmpcode = pd->hdr.icmp6.icmp6_code; 4942 state_icmp = pf_icmp_mapping(pd, icmptype, 4943 &icmp_dir, &multi, &virtual_id, &virtual_type); 4944 if (icmp_dir == PF_IN) { 4945 sport = virtual_id; 4946 dport = virtual_type; 4947 } else { 4948 sport = virtual_type; 4949 dport = virtual_id; 4950 } 4951 4952 break; 4953 #endif /* INET6 */ 4954 default: 4955 sport = dport = 0; 4956 break; 4957 } 4958 4959 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4960 4961 /* check packet for BINAT/NAT/RDR */ 4962 transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4963 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4964 switch (transerror) { 4965 default: 4966 /* A translation error occurred. */ 4967 REASON_SET(&reason, transerror); 4968 goto cleanup; 4969 case PFRES_MAX: 4970 /* No match. */ 4971 break; 4972 case PFRES_MATCH: 4973 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4974 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4975 4976 if (nr->log) { 4977 PFLOG_PACKET(kif, m, PF_PASS, PFRES_MATCH, nr, a, 4978 ruleset, pd, 1); 4979 } 4980 4981 if (pd->ip_sum) 4982 bip_sum = *pd->ip_sum; 4983 4984 switch (pd->proto) { 4985 case IPPROTO_TCP: 4986 bproto_sum = th->th_sum; 4987 pd->proto_sum = &th->th_sum; 4988 4989 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4990 nk->port[pd->sidx] != sport) { 4991 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4992 &th->th_sum, &nk->addr[pd->sidx], 4993 nk->port[pd->sidx], 0, af); 4994 pd->sport = &th->th_sport; 4995 sport = th->th_sport; 4996 } 4997 4998 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4999 nk->port[pd->didx] != dport) { 5000 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 5001 &th->th_sum, &nk->addr[pd->didx], 5002 nk->port[pd->didx], 0, af); 5003 dport = th->th_dport; 5004 pd->dport = &th->th_dport; 5005 } 5006 rewrite++; 5007 break; 5008 case IPPROTO_UDP: 5009 bproto_sum = pd->hdr.udp.uh_sum; 5010 pd->proto_sum = &pd->hdr.udp.uh_sum; 5011 5012 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5013 nk->port[pd->sidx] != sport) { 5014 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 5015 pd->ip_sum, &pd->hdr.udp.uh_sum, 5016 &nk->addr[pd->sidx], 5017 nk->port[pd->sidx], 1, af); 5018 sport = pd->hdr.udp.uh_sport; 5019 pd->sport = &pd->hdr.udp.uh_sport; 5020 } 5021 5022 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5023 nk->port[pd->didx] != dport) { 5024 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 5025 pd->ip_sum, &pd->hdr.udp.uh_sum, 5026 &nk->addr[pd->didx], 5027 nk->port[pd->didx], 1, af); 5028 dport = pd->hdr.udp.uh_dport; 5029 pd->dport = &pd->hdr.udp.uh_dport; 5030 } 5031 rewrite++; 5032 break; 5033 case IPPROTO_SCTP: { 5034 uint16_t checksum = 0; 5035 5036 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5037 nk->port[pd->sidx] != sport) { 5038 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 5039 pd->ip_sum, &checksum, 5040 &nk->addr[pd->sidx], 5041 nk->port[pd->sidx], 1, af); 5042 } 5043 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5044 nk->port[pd->didx] != dport) { 5045 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 5046 pd->ip_sum, &checksum, 5047 &nk->addr[pd->didx], 5048 nk->port[pd->didx], 1, af); 5049 } 5050 break; 5051 } 5052 #ifdef INET 5053 case IPPROTO_ICMP: 5054 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5055 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5056 nk->addr[pd->sidx].v4.s_addr, 0); 5057 5058 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5059 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5060 nk->addr[pd->didx].v4.s_addr, 0); 5061 5062 if (virtual_type == htons(ICMP_ECHO) && 5063 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5064 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5065 pd->hdr.icmp.icmp_cksum, sport, 5066 nk->port[pd->sidx], 0); 5067 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5068 pd->sport = &pd->hdr.icmp.icmp_id; 5069 } 5070 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5071 break; 5072 #endif /* INET */ 5073 #ifdef INET6 5074 case IPPROTO_ICMPV6: 5075 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5076 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5077 &nk->addr[pd->sidx], 0); 5078 5079 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5080 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5081 &nk->addr[pd->didx], 0); 5082 rewrite++; 5083 break; 5084 #endif /* INET */ 5085 default: 5086 switch (af) { 5087 #ifdef INET 5088 case AF_INET: 5089 if (PF_ANEQ(saddr, 5090 &nk->addr[pd->sidx], AF_INET)) 5091 pf_change_a(&saddr->v4.s_addr, 5092 pd->ip_sum, 5093 nk->addr[pd->sidx].v4.s_addr, 0); 5094 5095 if (PF_ANEQ(daddr, 5096 &nk->addr[pd->didx], AF_INET)) 5097 pf_change_a(&daddr->v4.s_addr, 5098 pd->ip_sum, 5099 nk->addr[pd->didx].v4.s_addr, 0); 5100 break; 5101 #endif /* INET */ 5102 #ifdef INET6 5103 case AF_INET6: 5104 if (PF_ANEQ(saddr, 5105 &nk->addr[pd->sidx], AF_INET6)) 5106 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 5107 5108 if (PF_ANEQ(daddr, 5109 &nk->addr[pd->didx], AF_INET6)) 5110 PF_ACPY(daddr, &nk->addr[pd->didx], af); 5111 break; 5112 #endif /* INET */ 5113 } 5114 break; 5115 } 5116 if (nr->natpass) 5117 r = NULL; 5118 pd->nat_rule = nr; 5119 } 5120 5121 while (r != NULL) { 5122 pf_counter_u64_add(&r->evaluations, 1); 5123 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5124 r = r->skip[PF_SKIP_IFP].ptr; 5125 else if (r->direction && r->direction != pd->dir) 5126 r = r->skip[PF_SKIP_DIR].ptr; 5127 else if (r->af && r->af != af) 5128 r = r->skip[PF_SKIP_AF].ptr; 5129 else if (r->proto && r->proto != pd->proto) 5130 r = r->skip[PF_SKIP_PROTO].ptr; 5131 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 5132 r->src.neg, kif, M_GETFIB(m))) 5133 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5134 /* tcp/udp only. port_op always 0 in other cases */ 5135 else if (r->src.port_op && !pf_match_port(r->src.port_op, 5136 r->src.port[0], r->src.port[1], sport)) 5137 r = r->skip[PF_SKIP_SRC_PORT].ptr; 5138 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 5139 r->dst.neg, NULL, M_GETFIB(m))) 5140 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5141 /* tcp/udp only. port_op always 0 in other cases */ 5142 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 5143 r->dst.port[0], r->dst.port[1], dport)) 5144 r = r->skip[PF_SKIP_DST_PORT].ptr; 5145 /* icmp only. type always 0 in other cases */ 5146 else if (r->type && r->type != icmptype + 1) 5147 r = TAILQ_NEXT(r, entries); 5148 /* icmp only. type always 0 in other cases */ 5149 else if (r->code && r->code != icmpcode + 1) 5150 r = TAILQ_NEXT(r, entries); 5151 else if (r->tos && !(r->tos == pd->tos)) 5152 r = TAILQ_NEXT(r, entries); 5153 else if (r->rule_flag & PFRULE_FRAGMENT) 5154 r = TAILQ_NEXT(r, entries); 5155 else if (pd->proto == IPPROTO_TCP && 5156 (r->flagset & th->th_flags) != r->flags) 5157 r = TAILQ_NEXT(r, entries); 5158 /* tcp/udp only. uid.op always 0 in other cases */ 5159 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 5160 pf_socket_lookup(pd, m), 1)) && 5161 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5162 pd->lookup.uid)) 5163 r = TAILQ_NEXT(r, entries); 5164 /* tcp/udp only. gid.op always 0 in other cases */ 5165 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 5166 pf_socket_lookup(pd, m), 1)) && 5167 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5168 pd->lookup.gid)) 5169 r = TAILQ_NEXT(r, entries); 5170 else if (r->prio && 5171 !pf_match_ieee8021q_pcp(r->prio, m)) 5172 r = TAILQ_NEXT(r, entries); 5173 else if (r->prob && 5174 r->prob <= arc4random()) 5175 r = TAILQ_NEXT(r, entries); 5176 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5177 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5178 r = TAILQ_NEXT(r, entries); 5179 else if (r->rcv_kif && !pf_match_rcvif(m, r)) 5180 r = TAILQ_NEXT(r, entries); 5181 else if (r->os_fingerprint != PF_OSFP_ANY && 5182 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 5183 pf_osfp_fingerprint(pd, m, off, th), 5184 r->os_fingerprint))) 5185 r = TAILQ_NEXT(r, entries); 5186 else { 5187 if (r->tag) 5188 tag = r->tag; 5189 if (r->anchor == NULL) { 5190 if (r->action == PF_MATCH) { 5191 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5192 if (ri == NULL) { 5193 REASON_SET(&reason, PFRES_MEMORY); 5194 goto cleanup; 5195 } 5196 ri->r = r; 5197 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5198 pf_counter_u64_critical_enter(); 5199 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5200 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5201 pf_counter_u64_critical_exit(); 5202 pf_rule_to_actions(r, &pd->act); 5203 if (r->log || pd->act.log & PF_LOG_MATCHES) 5204 PFLOG_PACKET(kif, m, 5205 r->action, PFRES_MATCH, r, 5206 a, ruleset, pd, 1); 5207 } else { 5208 match = 1; 5209 *rm = r; 5210 *am = a; 5211 *rsm = ruleset; 5212 if (pd->act.log & PF_LOG_MATCHES) 5213 PFLOG_PACKET(kif, m, 5214 r->action, PFRES_MATCH, r, 5215 a, ruleset, pd, 1); 5216 } 5217 if ((*rm)->quick) 5218 break; 5219 r = TAILQ_NEXT(r, entries); 5220 } else 5221 pf_step_into_anchor(anchor_stack, &asd, 5222 &ruleset, PF_RULESET_FILTER, &r, &a, 5223 &match); 5224 } 5225 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5226 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5227 break; 5228 } 5229 r = *rm; 5230 a = *am; 5231 ruleset = *rsm; 5232 5233 REASON_SET(&reason, PFRES_MATCH); 5234 5235 /* apply actions for last matching pass/block rule */ 5236 pf_rule_to_actions(r, &pd->act); 5237 5238 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5239 if (rewrite) 5240 m_copyback(m, off, hdrlen, pd->hdr.any); 5241 PFLOG_PACKET(kif, m, r->action, reason, r, a, ruleset, pd, 1); 5242 } 5243 5244 if ((r->action == PF_DROP) && 5245 ((r->rule_flag & PFRULE_RETURNRST) || 5246 (r->rule_flag & PFRULE_RETURNICMP) || 5247 (r->rule_flag & PFRULE_RETURN))) { 5248 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 5249 bip_sum, hdrlen, &reason, r->rtableid); 5250 } 5251 5252 if (r->action == PF_DROP) 5253 goto cleanup; 5254 5255 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5256 REASON_SET(&reason, PFRES_MEMORY); 5257 goto cleanup; 5258 } 5259 if (pd->act.rtableid >= 0) 5260 M_SETFIB(m, pd->act.rtableid); 5261 5262 if (!state_icmp && (r->keep_state || nr != NULL || 5263 (pd->flags & PFDESC_TCP_NORM))) { 5264 int action; 5265 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 5266 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 5267 hdrlen, &match_rules, udp_mapping); 5268 if (action != PF_PASS) { 5269 pf_udp_mapping_release(udp_mapping); 5270 if (action == PF_DROP && 5271 (r->rule_flag & PFRULE_RETURN)) 5272 pf_return(r, nr, pd, sk, off, m, th, kif, 5273 bproto_sum, bip_sum, hdrlen, &reason, 5274 pd->act.rtableid); 5275 return (action); 5276 } 5277 } else { 5278 while ((ri = SLIST_FIRST(&match_rules))) { 5279 SLIST_REMOVE_HEAD(&match_rules, entry); 5280 free(ri, M_PF_RULE_ITEM); 5281 } 5282 5283 uma_zfree(V_pf_state_key_z, sk); 5284 uma_zfree(V_pf_state_key_z, nk); 5285 pf_udp_mapping_release(udp_mapping); 5286 } 5287 5288 /* copy back packet headers if we performed NAT operations */ 5289 if (rewrite) 5290 m_copyback(m, off, hdrlen, pd->hdr.any); 5291 5292 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5293 pd->dir == PF_OUT && 5294 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 5295 /* 5296 * We want the state created, but we dont 5297 * want to send this in case a partner 5298 * firewall has to know about it to allow 5299 * replies through it. 5300 */ 5301 return (PF_DEFER); 5302 5303 return (PF_PASS); 5304 5305 cleanup: 5306 while ((ri = SLIST_FIRST(&match_rules))) { 5307 SLIST_REMOVE_HEAD(&match_rules, entry); 5308 free(ri, M_PF_RULE_ITEM); 5309 } 5310 5311 uma_zfree(V_pf_state_key_z, sk); 5312 uma_zfree(V_pf_state_key_z, nk); 5313 pf_udp_mapping_release(udp_mapping); 5314 5315 return (PF_DROP); 5316 } 5317 5318 static int 5319 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5320 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5321 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 5322 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 5323 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 5324 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5325 { 5326 struct pf_kstate *s = NULL; 5327 struct pf_ksrc_node *sn = NULL; 5328 struct tcphdr *th = &pd->hdr.tcp; 5329 u_int16_t mss = V_tcp_mssdflt; 5330 u_short reason, sn_reason; 5331 struct pf_krule_item *ri; 5332 5333 /* check maximums */ 5334 if (r->max_states && 5335 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5336 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5337 REASON_SET(&reason, PFRES_MAXSTATES); 5338 goto csfailed; 5339 } 5340 /* src node for filter rule */ 5341 if ((r->rule_flag & PFRULE_SRCTRACK || 5342 r->rpool.opts & PF_POOL_STICKYADDR) && 5343 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5344 REASON_SET(&reason, sn_reason); 5345 goto csfailed; 5346 } 5347 /* src node for translation rule */ 5348 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5349 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5350 pd->af)) != 0 ) { 5351 REASON_SET(&reason, sn_reason); 5352 goto csfailed; 5353 } 5354 s = pf_alloc_state(M_NOWAIT); 5355 if (s == NULL) { 5356 REASON_SET(&reason, PFRES_MEMORY); 5357 goto csfailed; 5358 } 5359 s->rule.ptr = r; 5360 s->nat_rule.ptr = nr; 5361 s->anchor.ptr = a; 5362 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5363 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5364 5365 STATE_INC_COUNTERS(s); 5366 if (r->allow_opts) 5367 s->state_flags |= PFSTATE_ALLOWOPTS; 5368 if (r->rule_flag & PFRULE_STATESLOPPY) 5369 s->state_flags |= PFSTATE_SLOPPY; 5370 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5371 s->state_flags |= PFSTATE_SCRUB_TCP; 5372 if ((r->rule_flag & PFRULE_PFLOW) || 5373 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5374 s->state_flags |= PFSTATE_PFLOW; 5375 5376 s->act.log = pd->act.log & PF_LOG_ALL; 5377 s->sync_state = PFSYNC_S_NONE; 5378 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5379 5380 if (nr != NULL) 5381 s->act.log |= nr->log & PF_LOG_ALL; 5382 switch (pd->proto) { 5383 case IPPROTO_TCP: 5384 s->src.seqlo = ntohl(th->th_seq); 5385 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5386 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5387 r->keep_state == PF_STATE_MODULATE) { 5388 /* Generate sequence number modulator */ 5389 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5390 0) 5391 s->src.seqdiff = 1; 5392 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 5393 htonl(s->src.seqlo + s->src.seqdiff), 0); 5394 *rewrite = 1; 5395 } else 5396 s->src.seqdiff = 0; 5397 if (th->th_flags & TH_SYN) { 5398 s->src.seqhi++; 5399 s->src.wscale = pf_get_wscale(m, off, 5400 th->th_off, pd->af); 5401 } 5402 s->src.max_win = MAX(ntohs(th->th_win), 1); 5403 if (s->src.wscale & PF_WSCALE_MASK) { 5404 /* Remove scale factor from initial window */ 5405 int win = s->src.max_win; 5406 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5407 s->src.max_win = (win - 1) >> 5408 (s->src.wscale & PF_WSCALE_MASK); 5409 } 5410 if (th->th_flags & TH_FIN) 5411 s->src.seqhi++; 5412 s->dst.seqhi = 1; 5413 s->dst.max_win = 1; 5414 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5415 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5416 s->timeout = PFTM_TCP_FIRST_PACKET; 5417 atomic_add_32(&V_pf_status.states_halfopen, 1); 5418 break; 5419 case IPPROTO_UDP: 5420 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5421 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5422 s->timeout = PFTM_UDP_FIRST_PACKET; 5423 break; 5424 case IPPROTO_SCTP: 5425 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5426 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5427 s->timeout = PFTM_SCTP_FIRST_PACKET; 5428 break; 5429 case IPPROTO_ICMP: 5430 #ifdef INET6 5431 case IPPROTO_ICMPV6: 5432 #endif 5433 s->timeout = PFTM_ICMP_FIRST_PACKET; 5434 break; 5435 default: 5436 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5437 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5438 s->timeout = PFTM_OTHER_FIRST_PACKET; 5439 } 5440 5441 if (r->rt) { 5442 /* pf_map_addr increases the reason counters */ 5443 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 5444 &s->rt_kif, NULL, &sn)) != 0) 5445 goto csfailed; 5446 s->rt = r->rt; 5447 } 5448 5449 s->creation = s->expire = pf_get_uptime(); 5450 5451 if (sn != NULL) 5452 s->src_node = sn; 5453 if (nsn != NULL) { 5454 /* XXX We only modify one side for now. */ 5455 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5456 s->nat_src_node = nsn; 5457 } 5458 if (pd->proto == IPPROTO_TCP) { 5459 if (s->state_flags & PFSTATE_SCRUB_TCP && 5460 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 5461 REASON_SET(&reason, PFRES_MEMORY); 5462 goto csfailed; 5463 } 5464 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5465 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 5466 &s->src, &s->dst, rewrite)) { 5467 /* This really shouldn't happen!!! */ 5468 DPFPRINTF(PF_DEBUG_URGENT, 5469 ("pf_normalize_tcp_stateful failed on first " 5470 "pkt\n")); 5471 goto csfailed; 5472 } 5473 } else if (pd->proto == IPPROTO_SCTP) { 5474 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5475 goto csfailed; 5476 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5477 goto csfailed; 5478 } 5479 s->direction = pd->dir; 5480 5481 /* 5482 * sk/nk could already been setup by pf_get_translation(). 5483 */ 5484 if (nr == NULL) { 5485 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5486 __func__, nr, sk, nk)); 5487 sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport); 5488 if (sk == NULL) 5489 goto csfailed; 5490 nk = sk; 5491 } else 5492 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5493 __func__, nr, sk, nk)); 5494 5495 /* Swap sk/nk for PF_OUT. */ 5496 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5497 (pd->dir == PF_IN) ? sk : nk, 5498 (pd->dir == PF_IN) ? nk : sk, s)) { 5499 REASON_SET(&reason, PFRES_STATEINS); 5500 goto drop; 5501 } else 5502 *sm = s; 5503 5504 if (tag > 0) 5505 s->tag = tag; 5506 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5507 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5508 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5509 /* undo NAT changes, if they have taken place */ 5510 if (nr != NULL) { 5511 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5512 if (pd->dir == PF_OUT) 5513 skt = s->key[PF_SK_STACK]; 5514 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5515 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5516 if (pd->sport) 5517 *pd->sport = skt->port[pd->sidx]; 5518 if (pd->dport) 5519 *pd->dport = skt->port[pd->didx]; 5520 if (pd->proto_sum) 5521 *pd->proto_sum = bproto_sum; 5522 if (pd->ip_sum) 5523 *pd->ip_sum = bip_sum; 5524 m_copyback(m, off, hdrlen, pd->hdr.any); 5525 } 5526 s->src.seqhi = htonl(arc4random()); 5527 /* Find mss option */ 5528 int rtid = M_GETFIB(m); 5529 mss = pf_get_mss(m, off, th->th_off, pd->af); 5530 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5531 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5532 s->src.mss = mss; 5533 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5534 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5535 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5536 pd->act.rtableid); 5537 REASON_SET(&reason, PFRES_SYNPROXY); 5538 return (PF_SYNPROXY_DROP); 5539 } 5540 5541 s->udp_mapping = udp_mapping; 5542 5543 return (PF_PASS); 5544 5545 csfailed: 5546 while ((ri = SLIST_FIRST(match_rules))) { 5547 SLIST_REMOVE_HEAD(match_rules, entry); 5548 free(ri, M_PF_RULE_ITEM); 5549 } 5550 5551 uma_zfree(V_pf_state_key_z, sk); 5552 uma_zfree(V_pf_state_key_z, nk); 5553 5554 if (sn != NULL) { 5555 PF_SRC_NODE_LOCK(sn); 5556 if (--sn->states == 0 && sn->expire == 0) { 5557 pf_unlink_src_node(sn); 5558 uma_zfree(V_pf_sources_z, sn); 5559 counter_u64_add( 5560 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5561 } 5562 PF_SRC_NODE_UNLOCK(sn); 5563 } 5564 5565 if (nsn != sn && nsn != NULL) { 5566 PF_SRC_NODE_LOCK(nsn); 5567 if (--nsn->states == 0 && nsn->expire == 0) { 5568 pf_unlink_src_node(nsn); 5569 uma_zfree(V_pf_sources_z, nsn); 5570 counter_u64_add( 5571 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5572 } 5573 PF_SRC_NODE_UNLOCK(nsn); 5574 } 5575 5576 drop: 5577 if (s != NULL) { 5578 pf_src_tree_remove_state(s); 5579 s->timeout = PFTM_UNLINKED; 5580 STATE_DEC_COUNTERS(s); 5581 pf_free_state(s); 5582 } 5583 5584 return (PF_DROP); 5585 } 5586 5587 static int 5588 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5589 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5590 struct pf_kruleset **rsm) 5591 { 5592 struct pf_krule *r, *a = NULL; 5593 struct pf_kruleset *ruleset = NULL; 5594 struct pf_krule_slist match_rules; 5595 struct pf_krule_item *ri; 5596 sa_family_t af = pd->af; 5597 u_short reason; 5598 int tag = -1; 5599 int asd = 0; 5600 int match = 0; 5601 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5602 5603 PF_RULES_RASSERT(); 5604 5605 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5606 SLIST_INIT(&match_rules); 5607 while (r != NULL) { 5608 pf_counter_u64_add(&r->evaluations, 1); 5609 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5610 r = r->skip[PF_SKIP_IFP].ptr; 5611 else if (r->direction && r->direction != pd->dir) 5612 r = r->skip[PF_SKIP_DIR].ptr; 5613 else if (r->af && r->af != af) 5614 r = r->skip[PF_SKIP_AF].ptr; 5615 else if (r->proto && r->proto != pd->proto) 5616 r = r->skip[PF_SKIP_PROTO].ptr; 5617 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5618 r->src.neg, kif, M_GETFIB(m))) 5619 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5620 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5621 r->dst.neg, NULL, M_GETFIB(m))) 5622 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5623 else if (r->tos && !(r->tos == pd->tos)) 5624 r = TAILQ_NEXT(r, entries); 5625 else if (r->os_fingerprint != PF_OSFP_ANY) 5626 r = TAILQ_NEXT(r, entries); 5627 else if (pd->proto == IPPROTO_UDP && 5628 (r->src.port_op || r->dst.port_op)) 5629 r = TAILQ_NEXT(r, entries); 5630 else if (pd->proto == IPPROTO_TCP && 5631 (r->src.port_op || r->dst.port_op || r->flagset)) 5632 r = TAILQ_NEXT(r, entries); 5633 else if ((pd->proto == IPPROTO_ICMP || 5634 pd->proto == IPPROTO_ICMPV6) && 5635 (r->type || r->code)) 5636 r = TAILQ_NEXT(r, entries); 5637 else if (r->prio && 5638 !pf_match_ieee8021q_pcp(r->prio, m)) 5639 r = TAILQ_NEXT(r, entries); 5640 else if (r->prob && r->prob <= 5641 (arc4random() % (UINT_MAX - 1) + 1)) 5642 r = TAILQ_NEXT(r, entries); 5643 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5644 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5645 r = TAILQ_NEXT(r, entries); 5646 else { 5647 if (r->anchor == NULL) { 5648 if (r->action == PF_MATCH) { 5649 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5650 if (ri == NULL) { 5651 REASON_SET(&reason, PFRES_MEMORY); 5652 goto cleanup; 5653 } 5654 ri->r = r; 5655 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5656 pf_counter_u64_critical_enter(); 5657 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5658 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5659 pf_counter_u64_critical_exit(); 5660 pf_rule_to_actions(r, &pd->act); 5661 if (r->log) 5662 PFLOG_PACKET(kif, m, 5663 r->action, PFRES_MATCH, r, 5664 a, ruleset, pd, 1); 5665 } else { 5666 match = 1; 5667 *rm = r; 5668 *am = a; 5669 *rsm = ruleset; 5670 } 5671 if ((*rm)->quick) 5672 break; 5673 r = TAILQ_NEXT(r, entries); 5674 } else 5675 pf_step_into_anchor(anchor_stack, &asd, 5676 &ruleset, PF_RULESET_FILTER, &r, &a, 5677 &match); 5678 } 5679 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5680 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5681 break; 5682 } 5683 r = *rm; 5684 a = *am; 5685 ruleset = *rsm; 5686 5687 REASON_SET(&reason, PFRES_MATCH); 5688 5689 /* apply actions for last matching pass/block rule */ 5690 pf_rule_to_actions(r, &pd->act); 5691 5692 if (r->log) 5693 PFLOG_PACKET(kif, m, r->action, reason, r, a, ruleset, pd, 1); 5694 5695 if (r->action != PF_PASS) 5696 return (PF_DROP); 5697 5698 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5699 REASON_SET(&reason, PFRES_MEMORY); 5700 goto cleanup; 5701 } 5702 5703 return (PF_PASS); 5704 5705 cleanup: 5706 while ((ri = SLIST_FIRST(&match_rules))) { 5707 SLIST_REMOVE_HEAD(&match_rules, entry); 5708 free(ri, M_PF_RULE_ITEM); 5709 } 5710 5711 return (PF_DROP); 5712 } 5713 5714 static int 5715 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5716 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5717 int *copyback) 5718 { 5719 struct tcphdr *th = &pd->hdr.tcp; 5720 struct pf_state_peer *src, *dst; 5721 u_int16_t win = ntohs(th->th_win); 5722 u_int32_t ack, end, data_end, seq, orig_seq; 5723 u_int8_t sws, dws, psrc, pdst; 5724 int ackskew; 5725 5726 if (pd->dir == (*state)->direction) { 5727 src = &(*state)->src; 5728 dst = &(*state)->dst; 5729 psrc = PF_PEER_SRC; 5730 pdst = PF_PEER_DST; 5731 } else { 5732 src = &(*state)->dst; 5733 dst = &(*state)->src; 5734 psrc = PF_PEER_DST; 5735 pdst = PF_PEER_SRC; 5736 } 5737 5738 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5739 sws = src->wscale & PF_WSCALE_MASK; 5740 dws = dst->wscale & PF_WSCALE_MASK; 5741 } else 5742 sws = dws = 0; 5743 5744 /* 5745 * Sequence tracking algorithm from Guido van Rooij's paper: 5746 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5747 * tcp_filtering.ps 5748 */ 5749 5750 orig_seq = seq = ntohl(th->th_seq); 5751 if (src->seqlo == 0) { 5752 /* First packet from this end. Set its state */ 5753 5754 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5755 src->scrub == NULL) { 5756 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5757 REASON_SET(reason, PFRES_MEMORY); 5758 return (PF_DROP); 5759 } 5760 } 5761 5762 /* Deferred generation of sequence number modulator */ 5763 if (dst->seqdiff && !src->seqdiff) { 5764 /* use random iss for the TCP server */ 5765 while ((src->seqdiff = arc4random() - seq) == 0) 5766 ; 5767 ack = ntohl(th->th_ack) - dst->seqdiff; 5768 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5769 src->seqdiff), 0); 5770 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5771 *copyback = 1; 5772 } else { 5773 ack = ntohl(th->th_ack); 5774 } 5775 5776 end = seq + pd->p_len; 5777 if (th->th_flags & TH_SYN) { 5778 end++; 5779 if (dst->wscale & PF_WSCALE_FLAG) { 5780 src->wscale = pf_get_wscale(m, off, th->th_off, 5781 pd->af); 5782 if (src->wscale & PF_WSCALE_FLAG) { 5783 /* Remove scale factor from initial 5784 * window */ 5785 sws = src->wscale & PF_WSCALE_MASK; 5786 win = ((u_int32_t)win + (1 << sws) - 1) 5787 >> sws; 5788 dws = dst->wscale & PF_WSCALE_MASK; 5789 } else { 5790 /* fixup other window */ 5791 dst->max_win = MIN(TCP_MAXWIN, 5792 (u_int32_t)dst->max_win << 5793 (dst->wscale & PF_WSCALE_MASK)); 5794 /* in case of a retrans SYN|ACK */ 5795 dst->wscale = 0; 5796 } 5797 } 5798 } 5799 data_end = end; 5800 if (th->th_flags & TH_FIN) 5801 end++; 5802 5803 src->seqlo = seq; 5804 if (src->state < TCPS_SYN_SENT) 5805 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5806 5807 /* 5808 * May need to slide the window (seqhi may have been set by 5809 * the crappy stack check or if we picked up the connection 5810 * after establishment) 5811 */ 5812 if (src->seqhi == 1 || 5813 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5814 src->seqhi = end + MAX(1, dst->max_win << dws); 5815 if (win > src->max_win) 5816 src->max_win = win; 5817 5818 } else { 5819 ack = ntohl(th->th_ack) - dst->seqdiff; 5820 if (src->seqdiff) { 5821 /* Modulate sequence numbers */ 5822 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5823 src->seqdiff), 0); 5824 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5825 *copyback = 1; 5826 } 5827 end = seq + pd->p_len; 5828 if (th->th_flags & TH_SYN) 5829 end++; 5830 data_end = end; 5831 if (th->th_flags & TH_FIN) 5832 end++; 5833 } 5834 5835 if ((th->th_flags & TH_ACK) == 0) { 5836 /* Let it pass through the ack skew check */ 5837 ack = dst->seqlo; 5838 } else if ((ack == 0 && 5839 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5840 /* broken tcp stacks do not set ack */ 5841 (dst->state < TCPS_SYN_SENT)) { 5842 /* 5843 * Many stacks (ours included) will set the ACK number in an 5844 * FIN|ACK if the SYN times out -- no sequence to ACK. 5845 */ 5846 ack = dst->seqlo; 5847 } 5848 5849 if (seq == end) { 5850 /* Ease sequencing restrictions on no data packets */ 5851 seq = src->seqlo; 5852 data_end = end = seq; 5853 } 5854 5855 ackskew = dst->seqlo - ack; 5856 5857 /* 5858 * Need to demodulate the sequence numbers in any TCP SACK options 5859 * (Selective ACK). We could optionally validate the SACK values 5860 * against the current ACK window, either forwards or backwards, but 5861 * I'm not confident that SACK has been implemented properly 5862 * everywhere. It wouldn't surprise me if several stacks accidentally 5863 * SACK too far backwards of previously ACKed data. There really aren't 5864 * any security implications of bad SACKing unless the target stack 5865 * doesn't validate the option length correctly. Someone trying to 5866 * spoof into a TCP connection won't bother blindly sending SACK 5867 * options anyway. 5868 */ 5869 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5870 if (pf_modulate_sack(m, off, pd, th, dst)) 5871 *copyback = 1; 5872 } 5873 5874 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5875 if (SEQ_GEQ(src->seqhi, data_end) && 5876 /* Last octet inside other's window space */ 5877 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5878 /* Retrans: not more than one window back */ 5879 (ackskew >= -MAXACKWINDOW) && 5880 /* Acking not more than one reassembled fragment backwards */ 5881 (ackskew <= (MAXACKWINDOW << sws)) && 5882 /* Acking not more than one window forward */ 5883 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5884 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5885 /* Require an exact/+1 sequence match on resets when possible */ 5886 5887 if (dst->scrub || src->scrub) { 5888 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5889 *state, src, dst, copyback)) 5890 return (PF_DROP); 5891 } 5892 5893 /* update max window */ 5894 if (src->max_win < win) 5895 src->max_win = win; 5896 /* synchronize sequencing */ 5897 if (SEQ_GT(end, src->seqlo)) 5898 src->seqlo = end; 5899 /* slide the window of what the other end can send */ 5900 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5901 dst->seqhi = ack + MAX((win << sws), 1); 5902 5903 /* update states */ 5904 if (th->th_flags & TH_SYN) 5905 if (src->state < TCPS_SYN_SENT) 5906 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5907 if (th->th_flags & TH_FIN) 5908 if (src->state < TCPS_CLOSING) 5909 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5910 if (th->th_flags & TH_ACK) { 5911 if (dst->state == TCPS_SYN_SENT) { 5912 pf_set_protostate(*state, pdst, 5913 TCPS_ESTABLISHED); 5914 if (src->state == TCPS_ESTABLISHED && 5915 (*state)->src_node != NULL && 5916 pf_src_connlimit(state)) { 5917 REASON_SET(reason, PFRES_SRCLIMIT); 5918 return (PF_DROP); 5919 } 5920 } else if (dst->state == TCPS_CLOSING) 5921 pf_set_protostate(*state, pdst, 5922 TCPS_FIN_WAIT_2); 5923 } 5924 if (th->th_flags & TH_RST) 5925 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5926 5927 /* update expire time */ 5928 (*state)->expire = pf_get_uptime(); 5929 if (src->state >= TCPS_FIN_WAIT_2 && 5930 dst->state >= TCPS_FIN_WAIT_2) 5931 (*state)->timeout = PFTM_TCP_CLOSED; 5932 else if (src->state >= TCPS_CLOSING && 5933 dst->state >= TCPS_CLOSING) 5934 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5935 else if (src->state < TCPS_ESTABLISHED || 5936 dst->state < TCPS_ESTABLISHED) 5937 (*state)->timeout = PFTM_TCP_OPENING; 5938 else if (src->state >= TCPS_CLOSING || 5939 dst->state >= TCPS_CLOSING) 5940 (*state)->timeout = PFTM_TCP_CLOSING; 5941 else 5942 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5943 5944 /* Fall through to PASS packet */ 5945 5946 } else if ((dst->state < TCPS_SYN_SENT || 5947 dst->state >= TCPS_FIN_WAIT_2 || 5948 src->state >= TCPS_FIN_WAIT_2) && 5949 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5950 /* Within a window forward of the originating packet */ 5951 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5952 /* Within a window backward of the originating packet */ 5953 5954 /* 5955 * This currently handles three situations: 5956 * 1) Stupid stacks will shotgun SYNs before their peer 5957 * replies. 5958 * 2) When PF catches an already established stream (the 5959 * firewall rebooted, the state table was flushed, routes 5960 * changed...) 5961 * 3) Packets get funky immediately after the connection 5962 * closes (this should catch Solaris spurious ACK|FINs 5963 * that web servers like to spew after a close) 5964 * 5965 * This must be a little more careful than the above code 5966 * since packet floods will also be caught here. We don't 5967 * update the TTL here to mitigate the damage of a packet 5968 * flood and so the same code can handle awkward establishment 5969 * and a loosened connection close. 5970 * In the establishment case, a correct peer response will 5971 * validate the connection, go through the normal state code 5972 * and keep updating the state TTL. 5973 */ 5974 5975 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5976 printf("pf: loose state match: "); 5977 pf_print_state(*state); 5978 pf_print_flags(th->th_flags); 5979 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5980 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5981 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5982 (unsigned long long)(*state)->packets[1], 5983 pd->dir == PF_IN ? "in" : "out", 5984 pd->dir == (*state)->direction ? "fwd" : "rev"); 5985 } 5986 5987 if (dst->scrub || src->scrub) { 5988 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5989 *state, src, dst, copyback)) 5990 return (PF_DROP); 5991 } 5992 5993 /* update max window */ 5994 if (src->max_win < win) 5995 src->max_win = win; 5996 /* synchronize sequencing */ 5997 if (SEQ_GT(end, src->seqlo)) 5998 src->seqlo = end; 5999 /* slide the window of what the other end can send */ 6000 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6001 dst->seqhi = ack + MAX((win << sws), 1); 6002 6003 /* 6004 * Cannot set dst->seqhi here since this could be a shotgunned 6005 * SYN and not an already established connection. 6006 */ 6007 6008 if (th->th_flags & TH_FIN) 6009 if (src->state < TCPS_CLOSING) 6010 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6011 if (th->th_flags & TH_RST) 6012 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6013 6014 /* Fall through to PASS packet */ 6015 6016 } else { 6017 if ((*state)->dst.state == TCPS_SYN_SENT && 6018 (*state)->src.state == TCPS_SYN_SENT) { 6019 /* Send RST for state mismatches during handshake */ 6020 if (!(th->th_flags & TH_RST)) 6021 pf_send_tcp((*state)->rule.ptr, pd->af, 6022 pd->dst, pd->src, th->th_dport, 6023 th->th_sport, ntohl(th->th_ack), 0, 6024 TH_RST, 0, 0, 6025 (*state)->rule.ptr->return_ttl, true, 0, 0, 6026 (*state)->act.rtableid); 6027 src->seqlo = 0; 6028 src->seqhi = 1; 6029 src->max_win = 1; 6030 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6031 printf("pf: BAD state: "); 6032 pf_print_state(*state); 6033 pf_print_flags(th->th_flags); 6034 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6035 "pkts=%llu:%llu dir=%s,%s\n", 6036 seq, orig_seq, ack, pd->p_len, ackskew, 6037 (unsigned long long)(*state)->packets[0], 6038 (unsigned long long)(*state)->packets[1], 6039 pd->dir == PF_IN ? "in" : "out", 6040 pd->dir == (*state)->direction ? "fwd" : "rev"); 6041 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6042 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6043 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6044 ' ': '2', 6045 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6046 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6047 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6048 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6049 } 6050 REASON_SET(reason, PFRES_BADSTATE); 6051 return (PF_DROP); 6052 } 6053 6054 return (PF_PASS); 6055 } 6056 6057 static int 6058 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6059 { 6060 struct tcphdr *th = &pd->hdr.tcp; 6061 struct pf_state_peer *src, *dst; 6062 u_int8_t psrc, pdst; 6063 6064 if (pd->dir == (*state)->direction) { 6065 src = &(*state)->src; 6066 dst = &(*state)->dst; 6067 psrc = PF_PEER_SRC; 6068 pdst = PF_PEER_DST; 6069 } else { 6070 src = &(*state)->dst; 6071 dst = &(*state)->src; 6072 psrc = PF_PEER_DST; 6073 pdst = PF_PEER_SRC; 6074 } 6075 6076 if (th->th_flags & TH_SYN) 6077 if (src->state < TCPS_SYN_SENT) 6078 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6079 if (th->th_flags & TH_FIN) 6080 if (src->state < TCPS_CLOSING) 6081 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6082 if (th->th_flags & TH_ACK) { 6083 if (dst->state == TCPS_SYN_SENT) { 6084 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6085 if (src->state == TCPS_ESTABLISHED && 6086 (*state)->src_node != NULL && 6087 pf_src_connlimit(state)) { 6088 REASON_SET(reason, PFRES_SRCLIMIT); 6089 return (PF_DROP); 6090 } 6091 } else if (dst->state == TCPS_CLOSING) { 6092 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6093 } else if (src->state == TCPS_SYN_SENT && 6094 dst->state < TCPS_SYN_SENT) { 6095 /* 6096 * Handle a special sloppy case where we only see one 6097 * half of the connection. If there is a ACK after 6098 * the initial SYN without ever seeing a packet from 6099 * the destination, set the connection to established. 6100 */ 6101 pf_set_protostate(*state, PF_PEER_BOTH, 6102 TCPS_ESTABLISHED); 6103 dst->state = src->state = TCPS_ESTABLISHED; 6104 if ((*state)->src_node != NULL && 6105 pf_src_connlimit(state)) { 6106 REASON_SET(reason, PFRES_SRCLIMIT); 6107 return (PF_DROP); 6108 } 6109 } else if (src->state == TCPS_CLOSING && 6110 dst->state == TCPS_ESTABLISHED && 6111 dst->seqlo == 0) { 6112 /* 6113 * Handle the closing of half connections where we 6114 * don't see the full bidirectional FIN/ACK+ACK 6115 * handshake. 6116 */ 6117 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6118 } 6119 } 6120 if (th->th_flags & TH_RST) 6121 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6122 6123 /* update expire time */ 6124 (*state)->expire = pf_get_uptime(); 6125 if (src->state >= TCPS_FIN_WAIT_2 && 6126 dst->state >= TCPS_FIN_WAIT_2) 6127 (*state)->timeout = PFTM_TCP_CLOSED; 6128 else if (src->state >= TCPS_CLOSING && 6129 dst->state >= TCPS_CLOSING) 6130 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6131 else if (src->state < TCPS_ESTABLISHED || 6132 dst->state < TCPS_ESTABLISHED) 6133 (*state)->timeout = PFTM_TCP_OPENING; 6134 else if (src->state >= TCPS_CLOSING || 6135 dst->state >= TCPS_CLOSING) 6136 (*state)->timeout = PFTM_TCP_CLOSING; 6137 else 6138 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6139 6140 return (PF_PASS); 6141 } 6142 6143 static int 6144 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6145 { 6146 struct pf_state_key *sk = (*state)->key[pd->didx]; 6147 struct tcphdr *th = &pd->hdr.tcp; 6148 6149 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6150 if (pd->dir != (*state)->direction) { 6151 REASON_SET(reason, PFRES_SYNPROXY); 6152 return (PF_SYNPROXY_DROP); 6153 } 6154 if (th->th_flags & TH_SYN) { 6155 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6156 REASON_SET(reason, PFRES_SYNPROXY); 6157 return (PF_DROP); 6158 } 6159 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 6160 pd->src, th->th_dport, th->th_sport, 6161 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6162 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 6163 (*state)->act.rtableid); 6164 REASON_SET(reason, PFRES_SYNPROXY); 6165 return (PF_SYNPROXY_DROP); 6166 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6167 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6168 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6169 REASON_SET(reason, PFRES_SYNPROXY); 6170 return (PF_DROP); 6171 } else if ((*state)->src_node != NULL && 6172 pf_src_connlimit(state)) { 6173 REASON_SET(reason, PFRES_SRCLIMIT); 6174 return (PF_DROP); 6175 } else 6176 pf_set_protostate(*state, PF_PEER_SRC, 6177 PF_TCPS_PROXY_DST); 6178 } 6179 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6180 if (pd->dir == (*state)->direction) { 6181 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6182 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6183 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6184 REASON_SET(reason, PFRES_SYNPROXY); 6185 return (PF_DROP); 6186 } 6187 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6188 if ((*state)->dst.seqhi == 1) 6189 (*state)->dst.seqhi = htonl(arc4random()); 6190 pf_send_tcp((*state)->rule.ptr, pd->af, 6191 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6192 sk->port[pd->sidx], sk->port[pd->didx], 6193 (*state)->dst.seqhi, 0, TH_SYN, 0, 6194 (*state)->src.mss, 0, false, (*state)->tag, 0, 6195 (*state)->act.rtableid); 6196 REASON_SET(reason, PFRES_SYNPROXY); 6197 return (PF_SYNPROXY_DROP); 6198 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6199 (TH_SYN|TH_ACK)) || 6200 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6201 REASON_SET(reason, PFRES_SYNPROXY); 6202 return (PF_DROP); 6203 } else { 6204 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6205 (*state)->dst.seqlo = ntohl(th->th_seq); 6206 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 6207 pd->src, th->th_dport, th->th_sport, 6208 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6209 TH_ACK, (*state)->src.max_win, 0, 0, false, 6210 (*state)->tag, 0, (*state)->act.rtableid); 6211 pf_send_tcp((*state)->rule.ptr, pd->af, 6212 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6213 sk->port[pd->sidx], sk->port[pd->didx], 6214 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6215 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 6216 (*state)->act.rtableid); 6217 (*state)->src.seqdiff = (*state)->dst.seqhi - 6218 (*state)->src.seqlo; 6219 (*state)->dst.seqdiff = (*state)->src.seqhi - 6220 (*state)->dst.seqlo; 6221 (*state)->src.seqhi = (*state)->src.seqlo + 6222 (*state)->dst.max_win; 6223 (*state)->dst.seqhi = (*state)->dst.seqlo + 6224 (*state)->src.max_win; 6225 (*state)->src.wscale = (*state)->dst.wscale = 0; 6226 pf_set_protostate(*state, PF_PEER_BOTH, 6227 TCPS_ESTABLISHED); 6228 REASON_SET(reason, PFRES_SYNPROXY); 6229 return (PF_SYNPROXY_DROP); 6230 } 6231 } 6232 6233 return (PF_PASS); 6234 } 6235 6236 static int 6237 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 6238 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 6239 u_short *reason) 6240 { 6241 struct pf_state_key_cmp key; 6242 struct tcphdr *th = &pd->hdr.tcp; 6243 int copyback = 0; 6244 int action; 6245 struct pf_state_peer *src, *dst; 6246 6247 bzero(&key, sizeof(key)); 6248 key.af = pd->af; 6249 key.proto = IPPROTO_TCP; 6250 if (pd->dir == PF_IN) { /* wire side, straight */ 6251 PF_ACPY(&key.addr[0], pd->src, key.af); 6252 PF_ACPY(&key.addr[1], pd->dst, key.af); 6253 key.port[0] = th->th_sport; 6254 key.port[1] = th->th_dport; 6255 } else { /* stack side, reverse */ 6256 PF_ACPY(&key.addr[1], pd->src, key.af); 6257 PF_ACPY(&key.addr[0], pd->dst, key.af); 6258 key.port[1] = th->th_sport; 6259 key.port[0] = th->th_dport; 6260 } 6261 6262 STATE_LOOKUP(kif, &key, *state, pd); 6263 6264 if (pd->dir == (*state)->direction) { 6265 src = &(*state)->src; 6266 dst = &(*state)->dst; 6267 } else { 6268 src = &(*state)->dst; 6269 dst = &(*state)->src; 6270 } 6271 6272 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6273 return (action); 6274 6275 if (dst->state >= TCPS_FIN_WAIT_2 && 6276 src->state >= TCPS_FIN_WAIT_2 && 6277 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6278 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6279 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6280 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6281 printf("pf: state reuse "); 6282 pf_print_state(*state); 6283 pf_print_flags(th->th_flags); 6284 printf("\n"); 6285 } 6286 /* XXX make sure it's the same direction ?? */ 6287 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6288 pf_unlink_state(*state); 6289 *state = NULL; 6290 return (PF_DROP); 6291 } 6292 6293 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6294 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6295 return (PF_DROP); 6296 } else { 6297 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 6298 ©back) == PF_DROP) 6299 return (PF_DROP); 6300 } 6301 6302 /* translate source/destination address, if necessary */ 6303 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6304 struct pf_state_key *nk = (*state)->key[pd->didx]; 6305 6306 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6307 nk->port[pd->sidx] != th->th_sport) 6308 pf_change_ap(m, pd->src, &th->th_sport, 6309 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6310 nk->port[pd->sidx], 0, pd->af); 6311 6312 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6313 nk->port[pd->didx] != th->th_dport) 6314 pf_change_ap(m, pd->dst, &th->th_dport, 6315 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6316 nk->port[pd->didx], 0, pd->af); 6317 copyback = 1; 6318 } 6319 6320 /* Copyback sequence modulation or stateful scrub changes if needed */ 6321 if (copyback) 6322 m_copyback(m, off, sizeof(*th), (caddr_t)th); 6323 6324 return (PF_PASS); 6325 } 6326 6327 static int 6328 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 6329 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 6330 { 6331 struct pf_state_peer *src, *dst; 6332 struct pf_state_key_cmp key; 6333 struct udphdr *uh = &pd->hdr.udp; 6334 uint8_t psrc, pdst; 6335 6336 bzero(&key, sizeof(key)); 6337 key.af = pd->af; 6338 key.proto = IPPROTO_UDP; 6339 if (pd->dir == PF_IN) { /* wire side, straight */ 6340 PF_ACPY(&key.addr[0], pd->src, key.af); 6341 PF_ACPY(&key.addr[1], pd->dst, key.af); 6342 key.port[0] = uh->uh_sport; 6343 key.port[1] = uh->uh_dport; 6344 } else { /* stack side, reverse */ 6345 PF_ACPY(&key.addr[1], pd->src, key.af); 6346 PF_ACPY(&key.addr[0], pd->dst, key.af); 6347 key.port[1] = uh->uh_sport; 6348 key.port[0] = uh->uh_dport; 6349 } 6350 6351 STATE_LOOKUP(kif, &key, *state, pd); 6352 6353 if (pd->dir == (*state)->direction) { 6354 src = &(*state)->src; 6355 dst = &(*state)->dst; 6356 psrc = PF_PEER_SRC; 6357 pdst = PF_PEER_DST; 6358 } else { 6359 src = &(*state)->dst; 6360 dst = &(*state)->src; 6361 psrc = PF_PEER_DST; 6362 pdst = PF_PEER_SRC; 6363 } 6364 6365 /* update states */ 6366 if (src->state < PFUDPS_SINGLE) 6367 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6368 if (dst->state == PFUDPS_SINGLE) 6369 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6370 6371 /* update expire time */ 6372 (*state)->expire = pf_get_uptime(); 6373 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6374 (*state)->timeout = PFTM_UDP_MULTIPLE; 6375 else 6376 (*state)->timeout = PFTM_UDP_SINGLE; 6377 6378 /* translate source/destination address, if necessary */ 6379 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6380 struct pf_state_key *nk = (*state)->key[pd->didx]; 6381 6382 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6383 nk->port[pd->sidx] != uh->uh_sport) 6384 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 6385 &uh->uh_sum, &nk->addr[pd->sidx], 6386 nk->port[pd->sidx], 1, pd->af); 6387 6388 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6389 nk->port[pd->didx] != uh->uh_dport) 6390 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 6391 &uh->uh_sum, &nk->addr[pd->didx], 6392 nk->port[pd->didx], 1, pd->af); 6393 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 6394 } 6395 6396 return (PF_PASS); 6397 } 6398 6399 static int 6400 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 6401 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6402 { 6403 struct pf_state_key_cmp key; 6404 struct pf_state_peer *src, *dst; 6405 struct sctphdr *sh = &pd->hdr.sctp; 6406 u_int8_t psrc; //, pdst; 6407 6408 bzero(&key, sizeof(key)); 6409 key.af = pd->af; 6410 key.proto = IPPROTO_SCTP; 6411 if (pd->dir == PF_IN) { /* wire side, straight */ 6412 PF_ACPY(&key.addr[0], pd->src, key.af); 6413 PF_ACPY(&key.addr[1], pd->dst, key.af); 6414 key.port[0] = sh->src_port; 6415 key.port[1] = sh->dest_port; 6416 } else { /* stack side, reverse */ 6417 PF_ACPY(&key.addr[1], pd->src, key.af); 6418 PF_ACPY(&key.addr[0], pd->dst, key.af); 6419 key.port[1] = sh->src_port; 6420 key.port[0] = sh->dest_port; 6421 } 6422 6423 STATE_LOOKUP(kif, &key, *state, pd); 6424 6425 if (pd->dir == (*state)->direction) { 6426 src = &(*state)->src; 6427 dst = &(*state)->dst; 6428 psrc = PF_PEER_SRC; 6429 } else { 6430 src = &(*state)->dst; 6431 dst = &(*state)->src; 6432 psrc = PF_PEER_DST; 6433 } 6434 6435 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6436 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6437 pd->sctp_flags & PFDESC_SCTP_INIT) { 6438 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6439 pf_unlink_state(*state); 6440 *state = NULL; 6441 return (PF_DROP); 6442 } 6443 6444 /* Track state. */ 6445 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6446 if (src->state < SCTP_COOKIE_WAIT) { 6447 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6448 (*state)->timeout = PFTM_SCTP_OPENING; 6449 } 6450 } 6451 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6452 MPASS(dst->scrub != NULL); 6453 if (dst->scrub->pfss_v_tag == 0) 6454 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6455 } 6456 6457 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6458 if (src->state < SCTP_ESTABLISHED) { 6459 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6460 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6461 } 6462 } 6463 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6464 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6465 if (src->state < SCTP_SHUTDOWN_PENDING) { 6466 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6467 (*state)->timeout = PFTM_SCTP_CLOSING; 6468 } 6469 } 6470 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6471 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6472 (*state)->timeout = PFTM_SCTP_CLOSED; 6473 } 6474 6475 if (src->scrub != NULL) { 6476 if (src->scrub->pfss_v_tag == 0) { 6477 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6478 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6479 return (PF_DROP); 6480 } 6481 6482 (*state)->expire = pf_get_uptime(); 6483 6484 /* translate source/destination address, if necessary */ 6485 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6486 uint16_t checksum = 0; 6487 struct pf_state_key *nk = (*state)->key[pd->didx]; 6488 6489 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6490 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6491 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6492 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6493 nk->port[pd->sidx], 1, pd->af); 6494 } 6495 6496 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6497 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6498 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6499 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6500 nk->port[pd->didx], 1, pd->af); 6501 } 6502 } 6503 6504 return (PF_PASS); 6505 } 6506 6507 static void 6508 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6509 { 6510 struct pf_sctp_endpoint key; 6511 struct pf_sctp_endpoint *ep; 6512 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6513 struct pf_sctp_source *i, *tmp; 6514 6515 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6516 return; 6517 6518 PF_SCTP_ENDPOINTS_LOCK(); 6519 6520 key.v_tag = s->dst.scrub->pfss_v_tag; 6521 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6522 if (ep != NULL) { 6523 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6524 if (pf_addr_cmp(&i->addr, 6525 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6526 s->key[PF_SK_WIRE]->af) == 0) { 6527 SDT_PROBE3(pf, sctp, multihome, remove, 6528 key.v_tag, s, i); 6529 TAILQ_REMOVE(&ep->sources, i, entry); 6530 free(i, M_PFTEMP); 6531 break; 6532 } 6533 } 6534 6535 if (TAILQ_EMPTY(&ep->sources)) { 6536 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6537 free(ep, M_PFTEMP); 6538 } 6539 } 6540 6541 /* Other direction. */ 6542 key.v_tag = s->src.scrub->pfss_v_tag; 6543 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6544 if (ep != NULL) { 6545 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6546 if (pf_addr_cmp(&i->addr, 6547 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6548 s->key[PF_SK_WIRE]->af) == 0) { 6549 SDT_PROBE3(pf, sctp, multihome, remove, 6550 key.v_tag, s, i); 6551 TAILQ_REMOVE(&ep->sources, i, entry); 6552 free(i, M_PFTEMP); 6553 break; 6554 } 6555 } 6556 6557 if (TAILQ_EMPTY(&ep->sources)) { 6558 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6559 free(ep, M_PFTEMP); 6560 } 6561 } 6562 6563 PF_SCTP_ENDPOINTS_UNLOCK(); 6564 } 6565 6566 static void 6567 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6568 { 6569 struct pf_sctp_endpoint key = { 6570 .v_tag = v_tag, 6571 }; 6572 struct pf_sctp_source *i; 6573 struct pf_sctp_endpoint *ep; 6574 6575 PF_SCTP_ENDPOINTS_LOCK(); 6576 6577 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6578 if (ep == NULL) { 6579 ep = malloc(sizeof(struct pf_sctp_endpoint), 6580 M_PFTEMP, M_NOWAIT); 6581 if (ep == NULL) { 6582 PF_SCTP_ENDPOINTS_UNLOCK(); 6583 return; 6584 } 6585 6586 ep->v_tag = v_tag; 6587 TAILQ_INIT(&ep->sources); 6588 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6589 } 6590 6591 /* Avoid inserting duplicates. */ 6592 TAILQ_FOREACH(i, &ep->sources, entry) { 6593 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6594 PF_SCTP_ENDPOINTS_UNLOCK(); 6595 return; 6596 } 6597 } 6598 6599 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6600 if (i == NULL) { 6601 PF_SCTP_ENDPOINTS_UNLOCK(); 6602 return; 6603 } 6604 6605 i->af = pd->af; 6606 memcpy(&i->addr, a, sizeof(*a)); 6607 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6608 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6609 6610 PF_SCTP_ENDPOINTS_UNLOCK(); 6611 } 6612 6613 static void 6614 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6615 struct pf_kstate *s, int action) 6616 { 6617 struct pf_sctp_multihome_job *j, *tmp; 6618 struct pf_sctp_source *i; 6619 int ret __unused; 6620 struct pf_kstate *sm = NULL; 6621 struct pf_krule *ra = NULL; 6622 struct pf_krule *r = &V_pf_default_rule; 6623 struct pf_kruleset *rs = NULL; 6624 bool do_extra = true; 6625 6626 PF_RULES_RLOCK_TRACKER; 6627 6628 again: 6629 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6630 if (s == NULL || action != PF_PASS) 6631 goto free; 6632 6633 /* Confirm we don't recurse here. */ 6634 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6635 6636 switch (j->op) { 6637 case SCTP_ADD_IP_ADDRESS: { 6638 uint32_t v_tag = pd->sctp_initiate_tag; 6639 6640 if (v_tag == 0) { 6641 if (s->direction == pd->dir) 6642 v_tag = s->src.scrub->pfss_v_tag; 6643 else 6644 v_tag = s->dst.scrub->pfss_v_tag; 6645 } 6646 6647 /* 6648 * Avoid duplicating states. We'll already have 6649 * created a state based on the source address of 6650 * the packet, but SCTP endpoints may also list this 6651 * address again in the INIT(_ACK) parameters. 6652 */ 6653 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6654 break; 6655 } 6656 6657 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6658 PF_RULES_RLOCK(); 6659 sm = NULL; 6660 /* 6661 * New connections need to be floating, because 6662 * we cannot know what interfaces it will use. 6663 * That's why we pass V_pfi_all rather than kif. 6664 */ 6665 ret = pf_test_rule(&r, &sm, V_pfi_all, 6666 j->m, off, &j->pd, &ra, &rs, NULL, 6667 sizeof(j->pd.hdr.sctp)); 6668 PF_RULES_RUNLOCK(); 6669 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6670 if (ret != PF_DROP && sm != NULL) { 6671 /* Inherit v_tag values. */ 6672 if (sm->direction == s->direction) { 6673 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6674 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6675 } else { 6676 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6677 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6678 } 6679 PF_STATE_UNLOCK(sm); 6680 } else { 6681 /* If we try duplicate inserts? */ 6682 break; 6683 } 6684 6685 /* Only add the address if we've actually allowed the state. */ 6686 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6687 6688 if (! do_extra) { 6689 break; 6690 } 6691 /* 6692 * We need to do this for each of our source addresses. 6693 * Find those based on the verification tag. 6694 */ 6695 struct pf_sctp_endpoint key = { 6696 .v_tag = pd->hdr.sctp.v_tag, 6697 }; 6698 struct pf_sctp_endpoint *ep; 6699 6700 PF_SCTP_ENDPOINTS_LOCK(); 6701 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6702 if (ep == NULL) { 6703 PF_SCTP_ENDPOINTS_UNLOCK(); 6704 break; 6705 } 6706 MPASS(ep != NULL); 6707 6708 TAILQ_FOREACH(i, &ep->sources, entry) { 6709 struct pf_sctp_multihome_job *nj; 6710 6711 /* SCTP can intermingle IPv4 and IPv6. */ 6712 if (i->af != pd->af) 6713 continue; 6714 6715 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6716 if (! nj) { 6717 continue; 6718 } 6719 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6720 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6721 nj->pd.src = &nj->src; 6722 // New destination address! 6723 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6724 nj->pd.dst = &nj->dst; 6725 nj->m = j->m; 6726 nj->op = j->op; 6727 6728 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6729 } 6730 PF_SCTP_ENDPOINTS_UNLOCK(); 6731 6732 break; 6733 } 6734 case SCTP_DEL_IP_ADDRESS: { 6735 struct pf_state_key_cmp key; 6736 uint8_t psrc; 6737 6738 bzero(&key, sizeof(key)); 6739 key.af = j->pd.af; 6740 key.proto = IPPROTO_SCTP; 6741 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6742 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6743 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6744 key.port[0] = j->pd.hdr.sctp.src_port; 6745 key.port[1] = j->pd.hdr.sctp.dest_port; 6746 } else { /* stack side, reverse */ 6747 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6748 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6749 key.port[1] = j->pd.hdr.sctp.src_port; 6750 key.port[0] = j->pd.hdr.sctp.dest_port; 6751 } 6752 6753 sm = pf_find_state(kif, &key, j->pd.dir); 6754 if (sm != NULL) { 6755 PF_STATE_LOCK_ASSERT(sm); 6756 if (j->pd.dir == sm->direction) { 6757 psrc = PF_PEER_SRC; 6758 } else { 6759 psrc = PF_PEER_DST; 6760 } 6761 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6762 sm->timeout = PFTM_SCTP_CLOSING; 6763 PF_STATE_UNLOCK(sm); 6764 } 6765 break; 6766 default: 6767 panic("Unknown op %#x", j->op); 6768 } 6769 } 6770 6771 free: 6772 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6773 free(j, M_PFTEMP); 6774 } 6775 6776 /* We may have inserted extra work while processing the list. */ 6777 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6778 do_extra = false; 6779 goto again; 6780 } 6781 } 6782 6783 static int 6784 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6785 struct pfi_kkif *kif, int op) 6786 { 6787 int off = 0; 6788 struct pf_sctp_multihome_job *job; 6789 6790 while (off < len) { 6791 struct sctp_paramhdr h; 6792 6793 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6794 pd->af)) 6795 return (PF_DROP); 6796 6797 /* Parameters are at least 4 bytes. */ 6798 if (ntohs(h.param_length) < 4) 6799 return (PF_DROP); 6800 6801 switch (ntohs(h.param_type)) { 6802 case SCTP_IPV4_ADDRESS: { 6803 struct in_addr t; 6804 6805 if (ntohs(h.param_length) != 6806 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6807 return (PF_DROP); 6808 6809 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6810 NULL, NULL, pd->af)) 6811 return (PF_DROP); 6812 6813 if (in_nullhost(t)) 6814 t.s_addr = pd->src->v4.s_addr; 6815 6816 /* 6817 * We hold the state lock (idhash) here, which means 6818 * that we can't acquire the keyhash, or we'll get a 6819 * LOR (and potentially double-lock things too). We also 6820 * can't release the state lock here, so instead we'll 6821 * enqueue this for async handling. 6822 * There's a relatively small race here, in that a 6823 * packet using the new addresses could arrive already, 6824 * but that's just though luck for it. 6825 */ 6826 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6827 if (! job) 6828 return (PF_DROP); 6829 6830 memcpy(&job->pd, pd, sizeof(*pd)); 6831 6832 // New source address! 6833 memcpy(&job->src, &t, sizeof(t)); 6834 job->pd.src = &job->src; 6835 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6836 job->pd.dst = &job->dst; 6837 job->m = m; 6838 job->op = op; 6839 6840 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6841 break; 6842 } 6843 #ifdef INET6 6844 case SCTP_IPV6_ADDRESS: { 6845 struct in6_addr t; 6846 6847 if (ntohs(h.param_length) != 6848 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6849 return (PF_DROP); 6850 6851 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6852 NULL, NULL, pd->af)) 6853 return (PF_DROP); 6854 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6855 break; 6856 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6857 memcpy(&t, &pd->src->v6, sizeof(t)); 6858 6859 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6860 if (! job) 6861 return (PF_DROP); 6862 6863 memcpy(&job->pd, pd, sizeof(*pd)); 6864 memcpy(&job->src, &t, sizeof(t)); 6865 job->pd.src = &job->src; 6866 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6867 job->pd.dst = &job->dst; 6868 job->m = m; 6869 job->op = op; 6870 6871 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6872 break; 6873 } 6874 #endif 6875 case SCTP_ADD_IP_ADDRESS: { 6876 int ret; 6877 struct sctp_asconf_paramhdr ah; 6878 6879 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6880 NULL, NULL, pd->af)) 6881 return (PF_DROP); 6882 6883 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6884 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6885 SCTP_ADD_IP_ADDRESS); 6886 if (ret != PF_PASS) 6887 return (ret); 6888 break; 6889 } 6890 case SCTP_DEL_IP_ADDRESS: { 6891 int ret; 6892 struct sctp_asconf_paramhdr ah; 6893 6894 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6895 NULL, NULL, pd->af)) 6896 return (PF_DROP); 6897 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6898 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6899 SCTP_DEL_IP_ADDRESS); 6900 if (ret != PF_PASS) 6901 return (ret); 6902 break; 6903 } 6904 default: 6905 break; 6906 } 6907 6908 off += roundup(ntohs(h.param_length), 4); 6909 } 6910 6911 return (PF_PASS); 6912 } 6913 int 6914 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6915 struct pfi_kkif *kif) 6916 { 6917 start += sizeof(struct sctp_init_chunk); 6918 len -= sizeof(struct sctp_init_chunk); 6919 6920 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6921 } 6922 6923 int 6924 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6925 struct pf_pdesc *pd, struct pfi_kkif *kif) 6926 { 6927 start += sizeof(struct sctp_asconf_chunk); 6928 len -= sizeof(struct sctp_asconf_chunk); 6929 6930 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6931 } 6932 6933 int 6934 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6935 struct pf_kstate **state, struct mbuf *m, int off, int direction, 6936 struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir, 6937 int *iidx, int multi, int inner) 6938 { 6939 key->af = pd->af; 6940 key->proto = pd->proto; 6941 if (icmp_dir == PF_IN) { 6942 *iidx = pd->sidx; 6943 key->port[pd->sidx] = icmpid; 6944 key->port[pd->didx] = type; 6945 } else { 6946 *iidx = pd->didx; 6947 key->port[pd->sidx] = type; 6948 key->port[pd->didx] = icmpid; 6949 } 6950 if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src, 6951 pd->didx, pd->dst, multi)) 6952 return (PF_DROP); 6953 6954 STATE_LOOKUP(kif, key, *state, pd); 6955 6956 if ((*state)->state_flags & PFSTATE_SLOPPY) 6957 return (-1); 6958 6959 /* Is this ICMP message flowing in right direction? */ 6960 if ((*state)->rule.ptr->type && 6961 (((!inner && (*state)->direction == direction) || 6962 (inner && (*state)->direction != direction)) ? 6963 PF_IN : PF_OUT) != icmp_dir) { 6964 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6965 printf("pf: icmp type %d in wrong direction (%d): ", 6966 ntohs(type), icmp_dir); 6967 pf_print_state(*state); 6968 printf("\n"); 6969 } 6970 PF_STATE_UNLOCK(*state); 6971 *state = NULL; 6972 return (PF_DROP); 6973 } 6974 return (-1); 6975 } 6976 6977 static int 6978 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6979 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6980 { 6981 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6982 u_int16_t *icmpsum, virtual_id, virtual_type; 6983 u_int8_t icmptype, icmpcode; 6984 int icmp_dir, iidx, ret, multi; 6985 struct pf_state_key_cmp key; 6986 #ifdef INET 6987 u_int16_t icmpid; 6988 #endif 6989 6990 MPASS(*state == NULL); 6991 6992 bzero(&key, sizeof(key)); 6993 switch (pd->proto) { 6994 #ifdef INET 6995 case IPPROTO_ICMP: 6996 icmptype = pd->hdr.icmp.icmp_type; 6997 icmpcode = pd->hdr.icmp.icmp_code; 6998 icmpid = pd->hdr.icmp.icmp_id; 6999 icmpsum = &pd->hdr.icmp.icmp_cksum; 7000 break; 7001 #endif /* INET */ 7002 #ifdef INET6 7003 case IPPROTO_ICMPV6: 7004 icmptype = pd->hdr.icmp6.icmp6_type; 7005 icmpcode = pd->hdr.icmp6.icmp6_code; 7006 #ifdef INET 7007 icmpid = pd->hdr.icmp6.icmp6_id; 7008 #endif 7009 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7010 break; 7011 #endif /* INET6 */ 7012 } 7013 7014 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 7015 &virtual_id, &virtual_type) == 0) { 7016 /* 7017 * ICMP query/reply message not related to a TCP/UDP packet. 7018 * Search for an ICMP state. 7019 */ 7020 ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir, 7021 kif, virtual_id, virtual_type, icmp_dir, &iidx, 7022 PF_ICMP_MULTI_NONE, 0); 7023 if (ret >= 0) { 7024 MPASS(*state == NULL); 7025 if (ret == PF_DROP && pd->af == AF_INET6 && 7026 icmp_dir == PF_OUT) { 7027 ret = pf_icmp_state_lookup(&key, pd, state, m, off, 7028 pd->dir, kif, virtual_id, virtual_type, 7029 icmp_dir, &iidx, multi, 0); 7030 if (ret >= 0) { 7031 MPASS(*state == NULL); 7032 return (ret); 7033 } 7034 } else 7035 return (ret); 7036 } 7037 7038 (*state)->expire = pf_get_uptime(); 7039 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7040 7041 /* translate source/destination address, if necessary */ 7042 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7043 struct pf_state_key *nk = (*state)->key[pd->didx]; 7044 7045 switch (pd->af) { 7046 #ifdef INET 7047 case AF_INET: 7048 if (PF_ANEQ(pd->src, 7049 &nk->addr[pd->sidx], AF_INET)) 7050 pf_change_a(&saddr->v4.s_addr, 7051 pd->ip_sum, 7052 nk->addr[pd->sidx].v4.s_addr, 0); 7053 7054 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 7055 AF_INET)) 7056 pf_change_a(&daddr->v4.s_addr, 7057 pd->ip_sum, 7058 nk->addr[pd->didx].v4.s_addr, 0); 7059 7060 if (nk->port[iidx] != 7061 pd->hdr.icmp.icmp_id) { 7062 pd->hdr.icmp.icmp_cksum = 7063 pf_cksum_fixup( 7064 pd->hdr.icmp.icmp_cksum, icmpid, 7065 nk->port[iidx], 0); 7066 pd->hdr.icmp.icmp_id = 7067 nk->port[iidx]; 7068 } 7069 7070 m_copyback(m, off, ICMP_MINLEN, 7071 (caddr_t )&pd->hdr.icmp); 7072 break; 7073 #endif /* INET */ 7074 #ifdef INET6 7075 case AF_INET6: 7076 if (PF_ANEQ(pd->src, 7077 &nk->addr[pd->sidx], AF_INET6)) 7078 pf_change_a6(saddr, 7079 &pd->hdr.icmp6.icmp6_cksum, 7080 &nk->addr[pd->sidx], 0); 7081 7082 if (PF_ANEQ(pd->dst, 7083 &nk->addr[pd->didx], AF_INET6)) 7084 pf_change_a6(daddr, 7085 &pd->hdr.icmp6.icmp6_cksum, 7086 &nk->addr[pd->didx], 0); 7087 7088 m_copyback(m, off, sizeof(struct icmp6_hdr), 7089 (caddr_t )&pd->hdr.icmp6); 7090 break; 7091 #endif /* INET6 */ 7092 } 7093 } 7094 return (PF_PASS); 7095 7096 } else { 7097 /* 7098 * ICMP error message in response to a TCP/UDP packet. 7099 * Extract the inner TCP/UDP header and search for that state. 7100 */ 7101 7102 struct pf_pdesc pd2; 7103 bzero(&pd2, sizeof pd2); 7104 #ifdef INET 7105 struct ip h2; 7106 #endif /* INET */ 7107 #ifdef INET6 7108 struct ip6_hdr h2_6; 7109 int terminal = 0; 7110 #endif /* INET6 */ 7111 int ipoff2 = 0; 7112 int off2 = 0; 7113 7114 pd2.af = pd->af; 7115 pd2.dir = pd->dir; 7116 /* Payload packet is from the opposite direction. */ 7117 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7118 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7119 switch (pd->af) { 7120 #ifdef INET 7121 case AF_INET: 7122 /* offset of h2 in mbuf chain */ 7123 ipoff2 = off + ICMP_MINLEN; 7124 7125 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 7126 NULL, reason, pd2.af)) { 7127 DPFPRINTF(PF_DEBUG_MISC, 7128 ("pf: ICMP error message too short " 7129 "(ip)\n")); 7130 return (PF_DROP); 7131 } 7132 /* 7133 * ICMP error messages don't refer to non-first 7134 * fragments 7135 */ 7136 if (h2.ip_off & htons(IP_OFFMASK)) { 7137 REASON_SET(reason, PFRES_FRAG); 7138 return (PF_DROP); 7139 } 7140 7141 /* offset of protocol header that follows h2 */ 7142 off2 = ipoff2 + (h2.ip_hl << 2); 7143 7144 pd2.proto = h2.ip_p; 7145 pd2.src = (struct pf_addr *)&h2.ip_src; 7146 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7147 pd2.ip_sum = &h2.ip_sum; 7148 break; 7149 #endif /* INET */ 7150 #ifdef INET6 7151 case AF_INET6: 7152 ipoff2 = off + sizeof(struct icmp6_hdr); 7153 7154 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 7155 NULL, reason, pd2.af)) { 7156 DPFPRINTF(PF_DEBUG_MISC, 7157 ("pf: ICMP error message too short " 7158 "(ip6)\n")); 7159 return (PF_DROP); 7160 } 7161 pd2.proto = h2_6.ip6_nxt; 7162 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7163 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7164 pd2.ip_sum = NULL; 7165 off2 = ipoff2 + sizeof(h2_6); 7166 do { 7167 switch (pd2.proto) { 7168 case IPPROTO_FRAGMENT: 7169 /* 7170 * ICMPv6 error messages for 7171 * non-first fragments 7172 */ 7173 REASON_SET(reason, PFRES_FRAG); 7174 return (PF_DROP); 7175 case IPPROTO_AH: 7176 case IPPROTO_HOPOPTS: 7177 case IPPROTO_ROUTING: 7178 case IPPROTO_DSTOPTS: { 7179 /* get next header and header length */ 7180 struct ip6_ext opt6; 7181 7182 if (!pf_pull_hdr(m, off2, &opt6, 7183 sizeof(opt6), NULL, reason, 7184 pd2.af)) { 7185 DPFPRINTF(PF_DEBUG_MISC, 7186 ("pf: ICMPv6 short opt\n")); 7187 return (PF_DROP); 7188 } 7189 if (pd2.proto == IPPROTO_AH) 7190 off2 += (opt6.ip6e_len + 2) * 4; 7191 else 7192 off2 += (opt6.ip6e_len + 1) * 8; 7193 pd2.proto = opt6.ip6e_nxt; 7194 /* goto the next header */ 7195 break; 7196 } 7197 default: 7198 terminal++; 7199 break; 7200 } 7201 } while (!terminal); 7202 break; 7203 #endif /* INET6 */ 7204 } 7205 7206 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7207 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7208 printf("pf: BAD ICMP %d:%d outer dst: ", 7209 icmptype, icmpcode); 7210 pf_print_host(pd->src, 0, pd->af); 7211 printf(" -> "); 7212 pf_print_host(pd->dst, 0, pd->af); 7213 printf(" inner src: "); 7214 pf_print_host(pd2.src, 0, pd2.af); 7215 printf(" -> "); 7216 pf_print_host(pd2.dst, 0, pd2.af); 7217 printf("\n"); 7218 } 7219 REASON_SET(reason, PFRES_BADSTATE); 7220 return (PF_DROP); 7221 } 7222 7223 switch (pd2.proto) { 7224 case IPPROTO_TCP: { 7225 struct tcphdr th; 7226 u_int32_t seq; 7227 struct pf_state_peer *src, *dst; 7228 u_int8_t dws; 7229 int copyback = 0; 7230 7231 /* 7232 * Only the first 8 bytes of the TCP header can be 7233 * expected. Don't access any TCP header fields after 7234 * th_seq, an ackskew test is not possible. 7235 */ 7236 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 7237 pd2.af)) { 7238 DPFPRINTF(PF_DEBUG_MISC, 7239 ("pf: ICMP error message too short " 7240 "(tcp)\n")); 7241 return (PF_DROP); 7242 } 7243 7244 key.af = pd2.af; 7245 key.proto = IPPROTO_TCP; 7246 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7247 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7248 key.port[pd2.sidx] = th.th_sport; 7249 key.port[pd2.didx] = th.th_dport; 7250 7251 STATE_LOOKUP(kif, &key, *state, pd); 7252 7253 if (pd->dir == (*state)->direction) { 7254 src = &(*state)->dst; 7255 dst = &(*state)->src; 7256 } else { 7257 src = &(*state)->src; 7258 dst = &(*state)->dst; 7259 } 7260 7261 if (src->wscale && dst->wscale) 7262 dws = dst->wscale & PF_WSCALE_MASK; 7263 else 7264 dws = 0; 7265 7266 /* Demodulate sequence number */ 7267 seq = ntohl(th.th_seq) - src->seqdiff; 7268 if (src->seqdiff) { 7269 pf_change_a(&th.th_seq, icmpsum, 7270 htonl(seq), 0); 7271 copyback = 1; 7272 } 7273 7274 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7275 (!SEQ_GEQ(src->seqhi, seq) || 7276 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7277 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7278 printf("pf: BAD ICMP %d:%d ", 7279 icmptype, icmpcode); 7280 pf_print_host(pd->src, 0, pd->af); 7281 printf(" -> "); 7282 pf_print_host(pd->dst, 0, pd->af); 7283 printf(" state: "); 7284 pf_print_state(*state); 7285 printf(" seq=%u\n", seq); 7286 } 7287 REASON_SET(reason, PFRES_BADSTATE); 7288 return (PF_DROP); 7289 } else { 7290 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7291 printf("pf: OK ICMP %d:%d ", 7292 icmptype, icmpcode); 7293 pf_print_host(pd->src, 0, pd->af); 7294 printf(" -> "); 7295 pf_print_host(pd->dst, 0, pd->af); 7296 printf(" state: "); 7297 pf_print_state(*state); 7298 printf(" seq=%u\n", seq); 7299 } 7300 } 7301 7302 /* translate source/destination address, if necessary */ 7303 if ((*state)->key[PF_SK_WIRE] != 7304 (*state)->key[PF_SK_STACK]) { 7305 struct pf_state_key *nk = 7306 (*state)->key[pd->didx]; 7307 7308 if (PF_ANEQ(pd2.src, 7309 &nk->addr[pd2.sidx], pd2.af) || 7310 nk->port[pd2.sidx] != th.th_sport) 7311 pf_change_icmp(pd2.src, &th.th_sport, 7312 daddr, &nk->addr[pd2.sidx], 7313 nk->port[pd2.sidx], NULL, 7314 pd2.ip_sum, icmpsum, 7315 pd->ip_sum, 0, pd2.af); 7316 7317 if (PF_ANEQ(pd2.dst, 7318 &nk->addr[pd2.didx], pd2.af) || 7319 nk->port[pd2.didx] != th.th_dport) 7320 pf_change_icmp(pd2.dst, &th.th_dport, 7321 saddr, &nk->addr[pd2.didx], 7322 nk->port[pd2.didx], NULL, 7323 pd2.ip_sum, icmpsum, 7324 pd->ip_sum, 0, pd2.af); 7325 copyback = 1; 7326 } 7327 7328 if (copyback) { 7329 switch (pd2.af) { 7330 #ifdef INET 7331 case AF_INET: 7332 m_copyback(m, off, ICMP_MINLEN, 7333 (caddr_t )&pd->hdr.icmp); 7334 m_copyback(m, ipoff2, sizeof(h2), 7335 (caddr_t )&h2); 7336 break; 7337 #endif /* INET */ 7338 #ifdef INET6 7339 case AF_INET6: 7340 m_copyback(m, off, 7341 sizeof(struct icmp6_hdr), 7342 (caddr_t )&pd->hdr.icmp6); 7343 m_copyback(m, ipoff2, sizeof(h2_6), 7344 (caddr_t )&h2_6); 7345 break; 7346 #endif /* INET6 */ 7347 } 7348 m_copyback(m, off2, 8, (caddr_t)&th); 7349 } 7350 7351 return (PF_PASS); 7352 break; 7353 } 7354 case IPPROTO_UDP: { 7355 struct udphdr uh; 7356 7357 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 7358 NULL, reason, pd2.af)) { 7359 DPFPRINTF(PF_DEBUG_MISC, 7360 ("pf: ICMP error message too short " 7361 "(udp)\n")); 7362 return (PF_DROP); 7363 } 7364 7365 key.af = pd2.af; 7366 key.proto = IPPROTO_UDP; 7367 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7368 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7369 key.port[pd2.sidx] = uh.uh_sport; 7370 key.port[pd2.didx] = uh.uh_dport; 7371 7372 STATE_LOOKUP(kif, &key, *state, pd); 7373 7374 /* translate source/destination address, if necessary */ 7375 if ((*state)->key[PF_SK_WIRE] != 7376 (*state)->key[PF_SK_STACK]) { 7377 struct pf_state_key *nk = 7378 (*state)->key[pd->didx]; 7379 7380 if (PF_ANEQ(pd2.src, 7381 &nk->addr[pd2.sidx], pd2.af) || 7382 nk->port[pd2.sidx] != uh.uh_sport) 7383 pf_change_icmp(pd2.src, &uh.uh_sport, 7384 daddr, &nk->addr[pd2.sidx], 7385 nk->port[pd2.sidx], &uh.uh_sum, 7386 pd2.ip_sum, icmpsum, 7387 pd->ip_sum, 1, pd2.af); 7388 7389 if (PF_ANEQ(pd2.dst, 7390 &nk->addr[pd2.didx], pd2.af) || 7391 nk->port[pd2.didx] != uh.uh_dport) 7392 pf_change_icmp(pd2.dst, &uh.uh_dport, 7393 saddr, &nk->addr[pd2.didx], 7394 nk->port[pd2.didx], &uh.uh_sum, 7395 pd2.ip_sum, icmpsum, 7396 pd->ip_sum, 1, pd2.af); 7397 7398 switch (pd2.af) { 7399 #ifdef INET 7400 case AF_INET: 7401 m_copyback(m, off, ICMP_MINLEN, 7402 (caddr_t )&pd->hdr.icmp); 7403 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7404 break; 7405 #endif /* INET */ 7406 #ifdef INET6 7407 case AF_INET6: 7408 m_copyback(m, off, 7409 sizeof(struct icmp6_hdr), 7410 (caddr_t )&pd->hdr.icmp6); 7411 m_copyback(m, ipoff2, sizeof(h2_6), 7412 (caddr_t )&h2_6); 7413 break; 7414 #endif /* INET6 */ 7415 } 7416 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 7417 } 7418 return (PF_PASS); 7419 break; 7420 } 7421 #ifdef INET 7422 case IPPROTO_ICMP: { 7423 struct icmp *iih = &pd2.hdr.icmp; 7424 7425 if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN, 7426 NULL, reason, pd2.af)) { 7427 DPFPRINTF(PF_DEBUG_MISC, 7428 ("pf: ICMP error message too short i" 7429 "(icmp)\n")); 7430 return (PF_DROP); 7431 } 7432 7433 icmpid = iih->icmp_id; 7434 pf_icmp_mapping(&pd2, iih->icmp_type, 7435 &icmp_dir, &multi, &virtual_id, &virtual_type); 7436 7437 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7438 pd2.dir, kif, virtual_id, virtual_type, 7439 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7440 if (ret >= 0) { 7441 MPASS(*state == NULL); 7442 return (ret); 7443 } 7444 7445 /* translate source/destination address, if necessary */ 7446 if ((*state)->key[PF_SK_WIRE] != 7447 (*state)->key[PF_SK_STACK]) { 7448 struct pf_state_key *nk = 7449 (*state)->key[pd->didx]; 7450 7451 if (PF_ANEQ(pd2.src, 7452 &nk->addr[pd2.sidx], pd2.af) || 7453 (virtual_type == htons(ICMP_ECHO) && 7454 nk->port[iidx] != iih->icmp_id)) 7455 pf_change_icmp(pd2.src, 7456 (virtual_type == htons(ICMP_ECHO)) ? 7457 &iih->icmp_id : NULL, 7458 daddr, &nk->addr[pd2.sidx], 7459 (virtual_type == htons(ICMP_ECHO)) ? 7460 nk->port[iidx] : 0, NULL, 7461 pd2.ip_sum, icmpsum, 7462 pd->ip_sum, 0, AF_INET); 7463 7464 if (PF_ANEQ(pd2.dst, 7465 &nk->addr[pd2.didx], pd2.af)) 7466 pf_change_icmp(pd2.dst, NULL, NULL, 7467 &nk->addr[pd2.didx], 0, NULL, 7468 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7469 AF_INET); 7470 7471 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7472 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7473 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih); 7474 } 7475 return (PF_PASS); 7476 break; 7477 } 7478 #endif /* INET */ 7479 #ifdef INET6 7480 case IPPROTO_ICMPV6: { 7481 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7482 7483 if (!pf_pull_hdr(m, off2, iih, 7484 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7485 DPFPRINTF(PF_DEBUG_MISC, 7486 ("pf: ICMP error message too short " 7487 "(icmp6)\n")); 7488 return (PF_DROP); 7489 } 7490 7491 pf_icmp_mapping(&pd2, iih->icmp6_type, 7492 &icmp_dir, &multi, &virtual_id, &virtual_type); 7493 7494 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7495 pd->dir, kif, virtual_id, virtual_type, 7496 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7497 if (ret >= 0) { 7498 MPASS(*state == NULL); 7499 if (ret == PF_DROP && pd2.af == AF_INET6 && 7500 icmp_dir == PF_OUT) { 7501 ret = pf_icmp_state_lookup(&key, &pd2, 7502 state, m, off, pd->dir, kif, 7503 virtual_id, virtual_type, 7504 icmp_dir, &iidx, multi, 1); 7505 if (ret >= 0) { 7506 MPASS(*state == NULL); 7507 return (ret); 7508 } 7509 } else 7510 return (ret); 7511 } 7512 7513 /* translate source/destination address, if necessary */ 7514 if ((*state)->key[PF_SK_WIRE] != 7515 (*state)->key[PF_SK_STACK]) { 7516 struct pf_state_key *nk = 7517 (*state)->key[pd->didx]; 7518 7519 if (PF_ANEQ(pd2.src, 7520 &nk->addr[pd2.sidx], pd2.af) || 7521 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7522 nk->port[pd2.sidx] != iih->icmp6_id)) 7523 pf_change_icmp(pd2.src, 7524 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7525 ? &iih->icmp6_id : NULL, 7526 daddr, &nk->addr[pd2.sidx], 7527 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7528 ? nk->port[iidx] : 0, NULL, 7529 pd2.ip_sum, icmpsum, 7530 pd->ip_sum, 0, AF_INET6); 7531 7532 if (PF_ANEQ(pd2.dst, 7533 &nk->addr[pd2.didx], pd2.af)) 7534 pf_change_icmp(pd2.dst, NULL, NULL, 7535 &nk->addr[pd2.didx], 0, NULL, 7536 pd2.ip_sum, icmpsum, 7537 pd->ip_sum, 0, AF_INET6); 7538 7539 m_copyback(m, off, sizeof(struct icmp6_hdr), 7540 (caddr_t)&pd->hdr.icmp6); 7541 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7542 m_copyback(m, off2, sizeof(struct icmp6_hdr), 7543 (caddr_t)iih); 7544 } 7545 return (PF_PASS); 7546 break; 7547 } 7548 #endif /* INET6 */ 7549 default: { 7550 key.af = pd2.af; 7551 key.proto = pd2.proto; 7552 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7553 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7554 key.port[0] = key.port[1] = 0; 7555 7556 STATE_LOOKUP(kif, &key, *state, pd); 7557 7558 /* translate source/destination address, if necessary */ 7559 if ((*state)->key[PF_SK_WIRE] != 7560 (*state)->key[PF_SK_STACK]) { 7561 struct pf_state_key *nk = 7562 (*state)->key[pd->didx]; 7563 7564 if (PF_ANEQ(pd2.src, 7565 &nk->addr[pd2.sidx], pd2.af)) 7566 pf_change_icmp(pd2.src, NULL, daddr, 7567 &nk->addr[pd2.sidx], 0, NULL, 7568 pd2.ip_sum, icmpsum, 7569 pd->ip_sum, 0, pd2.af); 7570 7571 if (PF_ANEQ(pd2.dst, 7572 &nk->addr[pd2.didx], pd2.af)) 7573 pf_change_icmp(pd2.dst, NULL, saddr, 7574 &nk->addr[pd2.didx], 0, NULL, 7575 pd2.ip_sum, icmpsum, 7576 pd->ip_sum, 0, pd2.af); 7577 7578 switch (pd2.af) { 7579 #ifdef INET 7580 case AF_INET: 7581 m_copyback(m, off, ICMP_MINLEN, 7582 (caddr_t)&pd->hdr.icmp); 7583 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7584 break; 7585 #endif /* INET */ 7586 #ifdef INET6 7587 case AF_INET6: 7588 m_copyback(m, off, 7589 sizeof(struct icmp6_hdr), 7590 (caddr_t )&pd->hdr.icmp6); 7591 m_copyback(m, ipoff2, sizeof(h2_6), 7592 (caddr_t )&h2_6); 7593 break; 7594 #endif /* INET6 */ 7595 } 7596 } 7597 return (PF_PASS); 7598 break; 7599 } 7600 } 7601 } 7602 } 7603 7604 static int 7605 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7606 struct mbuf *m, struct pf_pdesc *pd) 7607 { 7608 struct pf_state_peer *src, *dst; 7609 struct pf_state_key_cmp key; 7610 uint8_t psrc, pdst; 7611 7612 bzero(&key, sizeof(key)); 7613 key.af = pd->af; 7614 key.proto = pd->proto; 7615 if (pd->dir == PF_IN) { 7616 PF_ACPY(&key.addr[0], pd->src, key.af); 7617 PF_ACPY(&key.addr[1], pd->dst, key.af); 7618 key.port[0] = key.port[1] = 0; 7619 } else { 7620 PF_ACPY(&key.addr[1], pd->src, key.af); 7621 PF_ACPY(&key.addr[0], pd->dst, key.af); 7622 key.port[1] = key.port[0] = 0; 7623 } 7624 7625 STATE_LOOKUP(kif, &key, *state, pd); 7626 7627 if (pd->dir == (*state)->direction) { 7628 src = &(*state)->src; 7629 dst = &(*state)->dst; 7630 psrc = PF_PEER_SRC; 7631 pdst = PF_PEER_DST; 7632 } else { 7633 src = &(*state)->dst; 7634 dst = &(*state)->src; 7635 psrc = PF_PEER_DST; 7636 pdst = PF_PEER_SRC; 7637 } 7638 7639 /* update states */ 7640 if (src->state < PFOTHERS_SINGLE) 7641 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7642 if (dst->state == PFOTHERS_SINGLE) 7643 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7644 7645 /* update expire time */ 7646 (*state)->expire = pf_get_uptime(); 7647 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7648 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7649 else 7650 (*state)->timeout = PFTM_OTHER_SINGLE; 7651 7652 /* translate source/destination address, if necessary */ 7653 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7654 struct pf_state_key *nk = (*state)->key[pd->didx]; 7655 7656 KASSERT(nk, ("%s: nk is null", __func__)); 7657 KASSERT(pd, ("%s: pd is null", __func__)); 7658 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7659 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7660 switch (pd->af) { 7661 #ifdef INET 7662 case AF_INET: 7663 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7664 pf_change_a(&pd->src->v4.s_addr, 7665 pd->ip_sum, 7666 nk->addr[pd->sidx].v4.s_addr, 7667 0); 7668 7669 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7670 pf_change_a(&pd->dst->v4.s_addr, 7671 pd->ip_sum, 7672 nk->addr[pd->didx].v4.s_addr, 7673 0); 7674 7675 break; 7676 #endif /* INET */ 7677 #ifdef INET6 7678 case AF_INET6: 7679 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7680 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7681 7682 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7683 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7684 #endif /* INET6 */ 7685 } 7686 } 7687 return (PF_PASS); 7688 } 7689 7690 /* 7691 * ipoff and off are measured from the start of the mbuf chain. 7692 * h must be at "ipoff" on the mbuf chain. 7693 */ 7694 void * 7695 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7696 u_short *actionp, u_short *reasonp, sa_family_t af) 7697 { 7698 switch (af) { 7699 #ifdef INET 7700 case AF_INET: { 7701 const struct ip *h = mtod(m, struct ip *); 7702 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7703 7704 if (fragoff) { 7705 if (fragoff >= len) 7706 ACTION_SET(actionp, PF_PASS); 7707 else { 7708 ACTION_SET(actionp, PF_DROP); 7709 REASON_SET(reasonp, PFRES_FRAG); 7710 } 7711 return (NULL); 7712 } 7713 if (m->m_pkthdr.len < off + len || 7714 ntohs(h->ip_len) < off + len) { 7715 ACTION_SET(actionp, PF_DROP); 7716 REASON_SET(reasonp, PFRES_SHORT); 7717 return (NULL); 7718 } 7719 break; 7720 } 7721 #endif /* INET */ 7722 #ifdef INET6 7723 case AF_INET6: { 7724 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7725 7726 if (m->m_pkthdr.len < off + len || 7727 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7728 (unsigned)(off + len)) { 7729 ACTION_SET(actionp, PF_DROP); 7730 REASON_SET(reasonp, PFRES_SHORT); 7731 return (NULL); 7732 } 7733 break; 7734 } 7735 #endif /* INET6 */ 7736 } 7737 m_copydata(m, off, len, p); 7738 return (p); 7739 } 7740 7741 int 7742 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7743 int rtableid) 7744 { 7745 struct ifnet *ifp; 7746 7747 /* 7748 * Skip check for addresses with embedded interface scope, 7749 * as they would always match anyway. 7750 */ 7751 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7752 return (1); 7753 7754 if (af != AF_INET && af != AF_INET6) 7755 return (0); 7756 7757 if (kif == V_pfi_all) 7758 return (1); 7759 7760 /* Skip checks for ipsec interfaces */ 7761 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7762 return (1); 7763 7764 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7765 7766 switch (af) { 7767 #ifdef INET6 7768 case AF_INET6: 7769 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7770 ifp)); 7771 #endif 7772 #ifdef INET 7773 case AF_INET: 7774 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7775 ifp)); 7776 #endif 7777 } 7778 7779 return (0); 7780 } 7781 7782 #ifdef INET 7783 static void 7784 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7785 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7786 { 7787 struct mbuf *m0, *m1, *md; 7788 struct sockaddr_in dst; 7789 struct ip *ip; 7790 struct pfi_kkif *nkif = NULL; 7791 struct ifnet *ifp = NULL; 7792 struct pf_addr naddr; 7793 struct pf_ksrc_node *sn = NULL; 7794 int error = 0; 7795 uint16_t ip_len, ip_off; 7796 uint16_t tmp; 7797 int r_rt, r_dir; 7798 7799 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7800 7801 if (s) { 7802 r_rt = s->rt; 7803 r_dir = s->direction; 7804 } else { 7805 r_rt = r->rt; 7806 r_dir = r->direction; 7807 } 7808 7809 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7810 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7811 __func__)); 7812 7813 if ((pd->pf_mtag == NULL && 7814 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7815 pd->pf_mtag->routed++ > 3) { 7816 m0 = *m; 7817 *m = NULL; 7818 goto bad_locked; 7819 } 7820 7821 if (r_rt == PF_DUPTO) { 7822 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7823 if (s == NULL) { 7824 ifp = r->rpool.cur->kif ? 7825 r->rpool.cur->kif->pfik_ifp : NULL; 7826 } else { 7827 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7828 /* If pfsync'd */ 7829 if (ifp == NULL && r->rpool.cur != NULL) 7830 ifp = r->rpool.cur->kif ? 7831 r->rpool.cur->kif->pfik_ifp : NULL; 7832 PF_STATE_UNLOCK(s); 7833 } 7834 if (ifp == oifp) { 7835 /* When the 2nd interface is not skipped */ 7836 return; 7837 } else { 7838 m0 = *m; 7839 *m = NULL; 7840 goto bad; 7841 } 7842 } else { 7843 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7844 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7845 if (s) 7846 PF_STATE_UNLOCK(s); 7847 return; 7848 } 7849 } 7850 } else { 7851 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7852 pf_dummynet(pd, s, r, m); 7853 if (s) 7854 PF_STATE_UNLOCK(s); 7855 return; 7856 } 7857 m0 = *m; 7858 } 7859 7860 ip = mtod(m0, struct ip *); 7861 7862 bzero(&dst, sizeof(dst)); 7863 dst.sin_family = AF_INET; 7864 dst.sin_len = sizeof(dst); 7865 dst.sin_addr = ip->ip_dst; 7866 7867 bzero(&naddr, sizeof(naddr)); 7868 7869 if (s == NULL) { 7870 if (TAILQ_EMPTY(&r->rpool.list)) { 7871 DPFPRINTF(PF_DEBUG_URGENT, 7872 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7873 goto bad_locked; 7874 } 7875 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7876 &naddr, &nkif, NULL, &sn); 7877 if (!PF_AZERO(&naddr, AF_INET)) 7878 dst.sin_addr.s_addr = naddr.v4.s_addr; 7879 ifp = nkif ? nkif->pfik_ifp : NULL; 7880 } else { 7881 struct pfi_kkif *kif; 7882 7883 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7884 dst.sin_addr.s_addr = 7885 s->rt_addr.v4.s_addr; 7886 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7887 kif = s->rt_kif; 7888 /* If pfsync'd */ 7889 if (ifp == NULL && r->rpool.cur != NULL) { 7890 ifp = r->rpool.cur->kif ? 7891 r->rpool.cur->kif->pfik_ifp : NULL; 7892 kif = r->rpool.cur->kif; 7893 } 7894 if (ifp != NULL && kif != NULL && 7895 r->rule_flag & PFRULE_IFBOUND && 7896 r->rt == PF_REPLYTO && 7897 s->kif == V_pfi_all) { 7898 s->kif = kif; 7899 s->orig_kif = oifp->if_pf_kif; 7900 } 7901 7902 PF_STATE_UNLOCK(s); 7903 } 7904 7905 if (ifp == NULL) 7906 goto bad; 7907 7908 if (pd->dir == PF_IN) { 7909 if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7910 goto bad; 7911 else if (m0 == NULL) 7912 goto done; 7913 if (m0->m_len < sizeof(struct ip)) { 7914 DPFPRINTF(PF_DEBUG_URGENT, 7915 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7916 goto bad; 7917 } 7918 ip = mtod(m0, struct ip *); 7919 } 7920 7921 if (ifp->if_flags & IFF_LOOPBACK) 7922 m0->m_flags |= M_SKIP_FIREWALL; 7923 7924 ip_len = ntohs(ip->ip_len); 7925 ip_off = ntohs(ip->ip_off); 7926 7927 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7928 m0->m_pkthdr.csum_flags |= CSUM_IP; 7929 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7930 in_delayed_cksum(m0); 7931 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7932 } 7933 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7934 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7935 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7936 } 7937 7938 if (pd->dir == PF_IN) { 7939 /* 7940 * Make sure dummynet gets the correct direction, in case it needs to 7941 * re-inject later. 7942 */ 7943 pd->dir = PF_OUT; 7944 7945 /* 7946 * The following processing is actually the rest of the inbound processing, even 7947 * though we've marked it as outbound (so we don't look through dummynet) and it 7948 * happens after the outbound processing (pf_test(PF_OUT) above). 7949 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7950 * conclusion about what direction it's processing, and we can't fix it or it 7951 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7952 * decision will pick the right pipe, and everything will mostly work as expected. 7953 */ 7954 tmp = pd->act.dnrpipe; 7955 pd->act.dnrpipe = pd->act.dnpipe; 7956 pd->act.dnpipe = tmp; 7957 } 7958 7959 /* 7960 * If small enough for interface, or the interface will take 7961 * care of the fragmentation for us, we can just send directly. 7962 */ 7963 if (ip_len <= ifp->if_mtu || 7964 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7965 ip->ip_sum = 0; 7966 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7967 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7968 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7969 } 7970 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7971 7972 md = m0; 7973 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7974 if (md != NULL) 7975 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7976 goto done; 7977 } 7978 7979 /* Balk when DF bit is set or the interface didn't support TSO. */ 7980 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7981 error = EMSGSIZE; 7982 KMOD_IPSTAT_INC(ips_cantfrag); 7983 if (r_rt != PF_DUPTO) { 7984 if (s && pd->nat_rule != NULL) 7985 PACKET_UNDO_NAT(m0, pd, 7986 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7987 s); 7988 7989 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7990 ifp->if_mtu); 7991 goto done; 7992 } else 7993 goto bad; 7994 } 7995 7996 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7997 if (error) 7998 goto bad; 7999 8000 for (; m0; m0 = m1) { 8001 m1 = m0->m_nextpkt; 8002 m0->m_nextpkt = NULL; 8003 if (error == 0) { 8004 m_clrprotoflags(m0); 8005 md = m0; 8006 pd->pf_mtag = pf_find_mtag(md); 8007 error = pf_dummynet_route(pd, s, r, ifp, 8008 sintosa(&dst), &md); 8009 if (md != NULL) 8010 error = (*ifp->if_output)(ifp, md, 8011 sintosa(&dst), NULL); 8012 } else 8013 m_freem(m0); 8014 } 8015 8016 if (error == 0) 8017 KMOD_IPSTAT_INC(ips_fragmented); 8018 8019 done: 8020 if (r_rt != PF_DUPTO) 8021 *m = NULL; 8022 return; 8023 8024 bad_locked: 8025 if (s) 8026 PF_STATE_UNLOCK(s); 8027 bad: 8028 m_freem(m0); 8029 goto done; 8030 } 8031 #endif /* INET */ 8032 8033 #ifdef INET6 8034 static void 8035 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8036 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8037 { 8038 struct mbuf *m0, *md; 8039 struct sockaddr_in6 dst; 8040 struct ip6_hdr *ip6; 8041 struct pfi_kkif *nkif = NULL; 8042 struct ifnet *ifp = NULL; 8043 struct pf_addr naddr; 8044 struct pf_ksrc_node *sn = NULL; 8045 int r_rt, r_dir; 8046 8047 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8048 8049 if (s) { 8050 r_rt = s->rt; 8051 r_dir = s->direction; 8052 } else { 8053 r_rt = r->rt; 8054 r_dir = r->direction; 8055 } 8056 8057 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8058 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8059 __func__)); 8060 8061 if ((pd->pf_mtag == NULL && 8062 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8063 pd->pf_mtag->routed++ > 3) { 8064 m0 = *m; 8065 *m = NULL; 8066 goto bad_locked; 8067 } 8068 8069 if (r_rt == PF_DUPTO) { 8070 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8071 if (s == NULL) { 8072 ifp = r->rpool.cur->kif ? 8073 r->rpool.cur->kif->pfik_ifp : NULL; 8074 } else { 8075 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8076 /* If pfsync'd */ 8077 if (ifp == NULL && r->rpool.cur != NULL) 8078 ifp = r->rpool.cur->kif ? 8079 r->rpool.cur->kif->pfik_ifp : NULL; 8080 PF_STATE_UNLOCK(s); 8081 } 8082 if (ifp == oifp) { 8083 /* When the 2nd interface is not skipped */ 8084 return; 8085 } else { 8086 m0 = *m; 8087 *m = NULL; 8088 goto bad; 8089 } 8090 } else { 8091 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8092 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8093 if (s) 8094 PF_STATE_UNLOCK(s); 8095 return; 8096 } 8097 } 8098 } else { 8099 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8100 pf_dummynet(pd, s, r, m); 8101 if (s) 8102 PF_STATE_UNLOCK(s); 8103 return; 8104 } 8105 m0 = *m; 8106 } 8107 8108 ip6 = mtod(m0, struct ip6_hdr *); 8109 8110 bzero(&dst, sizeof(dst)); 8111 dst.sin6_family = AF_INET6; 8112 dst.sin6_len = sizeof(dst); 8113 dst.sin6_addr = ip6->ip6_dst; 8114 8115 bzero(&naddr, sizeof(naddr)); 8116 8117 if (s == NULL) { 8118 if (TAILQ_EMPTY(&r->rpool.list)) { 8119 DPFPRINTF(PF_DEBUG_URGENT, 8120 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 8121 goto bad_locked; 8122 } 8123 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 8124 &naddr, &nkif, NULL, &sn); 8125 if (!PF_AZERO(&naddr, AF_INET6)) 8126 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8127 &naddr, AF_INET6); 8128 ifp = nkif ? nkif->pfik_ifp : NULL; 8129 } else { 8130 struct pfi_kkif *kif; 8131 8132 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 8133 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8134 &s->rt_addr, AF_INET6); 8135 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8136 kif = s->rt_kif; 8137 /* If pfsync'd */ 8138 if (ifp == NULL && r->rpool.cur != NULL) { 8139 ifp = r->rpool.cur->kif ? 8140 r->rpool.cur->kif->pfik_ifp : NULL; 8141 kif = r->rpool.cur->kif; 8142 } 8143 if (ifp != NULL && kif != NULL && 8144 r->rule_flag & PFRULE_IFBOUND && 8145 r->rt == PF_REPLYTO && 8146 s->kif == V_pfi_all) { 8147 s->kif = kif; 8148 s->orig_kif = oifp->if_pf_kif; 8149 } 8150 } 8151 8152 if (s) 8153 PF_STATE_UNLOCK(s); 8154 8155 if (ifp == NULL) 8156 goto bad; 8157 8158 if (pd->dir == PF_IN) { 8159 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 8160 goto bad; 8161 else if (m0 == NULL) 8162 goto done; 8163 if (m0->m_len < sizeof(struct ip6_hdr)) { 8164 DPFPRINTF(PF_DEBUG_URGENT, 8165 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8166 __func__)); 8167 goto bad; 8168 } 8169 ip6 = mtod(m0, struct ip6_hdr *); 8170 } 8171 8172 if (ifp->if_flags & IFF_LOOPBACK) 8173 m0->m_flags |= M_SKIP_FIREWALL; 8174 8175 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8176 ~ifp->if_hwassist) { 8177 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8178 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8179 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8180 } 8181 8182 /* 8183 * If the packet is too large for the outgoing interface, 8184 * send back an icmp6 error. 8185 */ 8186 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8187 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8188 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8189 md = m0; 8190 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8191 if (md != NULL) 8192 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8193 } 8194 else { 8195 in6_ifstat_inc(ifp, ifs6_in_toobig); 8196 if (r_rt != PF_DUPTO) { 8197 if (s && pd->nat_rule != NULL) 8198 PACKET_UNDO_NAT(m0, pd, 8199 ((caddr_t)ip6 - m0->m_data) + 8200 sizeof(struct ip6_hdr), s); 8201 8202 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8203 } else 8204 goto bad; 8205 } 8206 8207 done: 8208 if (r_rt != PF_DUPTO) 8209 *m = NULL; 8210 return; 8211 8212 bad_locked: 8213 if (s) 8214 PF_STATE_UNLOCK(s); 8215 bad: 8216 m_freem(m0); 8217 goto done; 8218 } 8219 #endif /* INET6 */ 8220 8221 /* 8222 * FreeBSD supports cksum offloads for the following drivers. 8223 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8224 * 8225 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8226 * network driver performed cksum including pseudo header, need to verify 8227 * csum_data 8228 * CSUM_DATA_VALID : 8229 * network driver performed cksum, needs to additional pseudo header 8230 * cksum computation with partial csum_data(i.e. lack of H/W support for 8231 * pseudo header, for instance sk(4) and possibly gem(4)) 8232 * 8233 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8234 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8235 * TCP/UDP layer. 8236 * Also, set csum_data to 0xffff to force cksum validation. 8237 */ 8238 static int 8239 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8240 { 8241 u_int16_t sum = 0; 8242 int hw_assist = 0; 8243 struct ip *ip; 8244 8245 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8246 return (1); 8247 if (m->m_pkthdr.len < off + len) 8248 return (1); 8249 8250 switch (p) { 8251 case IPPROTO_TCP: 8252 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8253 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8254 sum = m->m_pkthdr.csum_data; 8255 } else { 8256 ip = mtod(m, struct ip *); 8257 sum = in_pseudo(ip->ip_src.s_addr, 8258 ip->ip_dst.s_addr, htonl((u_short)len + 8259 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8260 } 8261 sum ^= 0xffff; 8262 ++hw_assist; 8263 } 8264 break; 8265 case IPPROTO_UDP: 8266 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8267 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8268 sum = m->m_pkthdr.csum_data; 8269 } else { 8270 ip = mtod(m, struct ip *); 8271 sum = in_pseudo(ip->ip_src.s_addr, 8272 ip->ip_dst.s_addr, htonl((u_short)len + 8273 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8274 } 8275 sum ^= 0xffff; 8276 ++hw_assist; 8277 } 8278 break; 8279 case IPPROTO_ICMP: 8280 #ifdef INET6 8281 case IPPROTO_ICMPV6: 8282 #endif /* INET6 */ 8283 break; 8284 default: 8285 return (1); 8286 } 8287 8288 if (!hw_assist) { 8289 switch (af) { 8290 case AF_INET: 8291 if (p == IPPROTO_ICMP) { 8292 if (m->m_len < off) 8293 return (1); 8294 m->m_data += off; 8295 m->m_len -= off; 8296 sum = in_cksum(m, len); 8297 m->m_data -= off; 8298 m->m_len += off; 8299 } else { 8300 if (m->m_len < sizeof(struct ip)) 8301 return (1); 8302 sum = in4_cksum(m, p, off, len); 8303 } 8304 break; 8305 #ifdef INET6 8306 case AF_INET6: 8307 if (m->m_len < sizeof(struct ip6_hdr)) 8308 return (1); 8309 sum = in6_cksum(m, p, off, len); 8310 break; 8311 #endif /* INET6 */ 8312 default: 8313 return (1); 8314 } 8315 } 8316 if (sum) { 8317 switch (p) { 8318 case IPPROTO_TCP: 8319 { 8320 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8321 break; 8322 } 8323 case IPPROTO_UDP: 8324 { 8325 KMOD_UDPSTAT_INC(udps_badsum); 8326 break; 8327 } 8328 #ifdef INET 8329 case IPPROTO_ICMP: 8330 { 8331 KMOD_ICMPSTAT_INC(icps_checksum); 8332 break; 8333 } 8334 #endif 8335 #ifdef INET6 8336 case IPPROTO_ICMPV6: 8337 { 8338 KMOD_ICMP6STAT_INC(icp6s_checksum); 8339 break; 8340 } 8341 #endif /* INET6 */ 8342 } 8343 return (1); 8344 } else { 8345 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8346 m->m_pkthdr.csum_flags |= 8347 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8348 m->m_pkthdr.csum_data = 0xffff; 8349 } 8350 } 8351 return (0); 8352 } 8353 8354 static bool 8355 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8356 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8357 { 8358 int dndir = r->direction; 8359 8360 if (s && dndir == PF_INOUT) { 8361 dndir = s->direction; 8362 } else if (dndir == PF_INOUT) { 8363 /* Assume primary direction. Happens when we've set dnpipe in 8364 * the ethernet level code. */ 8365 dndir = pd->dir; 8366 } 8367 8368 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8369 return (false); 8370 8371 memset(dnflow, 0, sizeof(*dnflow)); 8372 8373 if (pd->dport != NULL) 8374 dnflow->f_id.dst_port = ntohs(*pd->dport); 8375 if (pd->sport != NULL) 8376 dnflow->f_id.src_port = ntohs(*pd->sport); 8377 8378 if (pd->dir == PF_IN) 8379 dnflow->flags |= IPFW_ARGS_IN; 8380 else 8381 dnflow->flags |= IPFW_ARGS_OUT; 8382 8383 if (pd->dir != dndir && pd->act.dnrpipe) { 8384 dnflow->rule.info = pd->act.dnrpipe; 8385 } 8386 else if (pd->dir == dndir && pd->act.dnpipe) { 8387 dnflow->rule.info = pd->act.dnpipe; 8388 } 8389 else { 8390 return (false); 8391 } 8392 8393 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8394 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8395 dnflow->rule.info |= IPFW_IS_PIPE; 8396 8397 dnflow->f_id.proto = pd->proto; 8398 dnflow->f_id.extra = dnflow->rule.info; 8399 switch (pd->af) { 8400 case AF_INET: 8401 dnflow->f_id.addr_type = 4; 8402 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8403 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8404 break; 8405 case AF_INET6: 8406 dnflow->flags |= IPFW_ARGS_IP6; 8407 dnflow->f_id.addr_type = 6; 8408 dnflow->f_id.src_ip6 = pd->src->v6; 8409 dnflow->f_id.dst_ip6 = pd->dst->v6; 8410 break; 8411 default: 8412 panic("Invalid AF"); 8413 break; 8414 } 8415 8416 return (true); 8417 } 8418 8419 int 8420 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8421 struct inpcb *inp) 8422 { 8423 struct pfi_kkif *kif; 8424 struct mbuf *m = *m0; 8425 8426 M_ASSERTPKTHDR(m); 8427 MPASS(ifp->if_vnet == curvnet); 8428 NET_EPOCH_ASSERT(); 8429 8430 if (!V_pf_status.running) 8431 return (PF_PASS); 8432 8433 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8434 8435 if (kif == NULL) { 8436 DPFPRINTF(PF_DEBUG_URGENT, 8437 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8438 return (PF_DROP); 8439 } 8440 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8441 return (PF_PASS); 8442 8443 if (m->m_flags & M_SKIP_FIREWALL) 8444 return (PF_PASS); 8445 8446 if (__predict_false(! M_WRITABLE(*m0))) { 8447 m = *m0 = m_unshare(*m0, M_NOWAIT); 8448 if (*m0 == NULL) 8449 return (PF_DROP); 8450 } 8451 8452 /* Stateless! */ 8453 return (pf_test_eth_rule(dir, kif, m0)); 8454 } 8455 8456 static __inline void 8457 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8458 { 8459 struct m_tag *mtag; 8460 8461 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8462 8463 /* dummynet adds this tag, but pf does not need it, 8464 * and keeping it creates unexpected behavior, 8465 * e.g. in case of divert(4) usage right after dummynet. */ 8466 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8467 if (mtag != NULL) 8468 m_tag_delete(m, mtag); 8469 } 8470 8471 static int 8472 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8473 struct pf_krule *r, struct mbuf **m0) 8474 { 8475 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8476 } 8477 8478 static int 8479 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8480 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8481 struct mbuf **m0) 8482 { 8483 NET_EPOCH_ASSERT(); 8484 8485 if (pd->act.dnpipe || pd->act.dnrpipe) { 8486 struct ip_fw_args dnflow; 8487 if (ip_dn_io_ptr == NULL) { 8488 m_freem(*m0); 8489 *m0 = NULL; 8490 return (ENOMEM); 8491 } 8492 8493 if (pd->pf_mtag == NULL && 8494 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8495 m_freem(*m0); 8496 *m0 = NULL; 8497 return (ENOMEM); 8498 } 8499 8500 if (ifp != NULL) { 8501 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8502 8503 pd->pf_mtag->if_index = ifp->if_index; 8504 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8505 8506 MPASS(sa != NULL); 8507 8508 if (pd->af == AF_INET) 8509 memcpy(&pd->pf_mtag->dst, sa, 8510 sizeof(struct sockaddr_in)); 8511 else 8512 memcpy(&pd->pf_mtag->dst, sa, 8513 sizeof(struct sockaddr_in6)); 8514 } 8515 8516 if (s != NULL && s->nat_rule.ptr != NULL && 8517 s->nat_rule.ptr->action == PF_RDR && 8518 ( 8519 #ifdef INET 8520 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8521 #endif 8522 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8523 /* 8524 * If we're redirecting to loopback mark this packet 8525 * as being local. Otherwise it might get dropped 8526 * if dummynet re-injects. 8527 */ 8528 (*m0)->m_pkthdr.rcvif = V_loif; 8529 } 8530 8531 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8532 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8533 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8534 ip_dn_io_ptr(m0, &dnflow); 8535 if (*m0 != NULL) { 8536 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8537 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8538 } 8539 } 8540 } 8541 8542 return (0); 8543 } 8544 8545 int 8546 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf *m, 8547 u_short *action, u_short *reason, struct pfi_kkif *kif, struct pf_krule **a, 8548 struct pf_krule **r, struct pf_kruleset **ruleset, int *off, int *hdrlen, 8549 struct pf_rule_actions *default_actions) 8550 { 8551 8552 TAILQ_INIT(&pd->sctp_multihome_jobs); 8553 if (default_actions != NULL) 8554 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8555 pd->pf_mtag = pf_find_mtag(m); 8556 8557 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8558 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8559 pd->act.flags = pd->pf_mtag->dnflags; 8560 } 8561 8562 pd->af = af; 8563 8564 switch (af) { 8565 #ifdef INET 8566 case AF_INET: { 8567 struct ip *h; 8568 8569 h = mtod(m, struct ip *); 8570 *off = h->ip_hl << 2; 8571 if (*off < (int)sizeof(*h)) { 8572 *action = PF_DROP; 8573 REASON_SET(reason, PFRES_SHORT); 8574 return (-1); 8575 } 8576 pd->src = (struct pf_addr *)&h->ip_src; 8577 pd->dst = (struct pf_addr *)&h->ip_dst; 8578 pd->sport = pd->dport = NULL; 8579 pd->ip_sum = &h->ip_sum; 8580 pd->proto_sum = NULL; 8581 pd->proto = h->ip_p; 8582 pd->dir = dir; 8583 pd->sidx = (dir == PF_IN) ? 0 : 1; 8584 pd->didx = (dir == PF_IN) ? 1 : 0; 8585 pd->tos = h->ip_tos; 8586 pd->tot_len = ntohs(h->ip_len); 8587 pd->act.rtableid = -1; 8588 8589 /* fragments not reassembled handled later */ 8590 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 8591 return (0); 8592 8593 switch (h->ip_p) { 8594 case IPPROTO_TCP: { 8595 struct tcphdr *th = &pd->hdr.tcp; 8596 8597 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8598 reason, AF_INET)) { 8599 *action = PF_DROP; 8600 REASON_SET(reason, PFRES_SHORT); 8601 return (-1); 8602 } 8603 *hdrlen = sizeof(*th); 8604 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8605 pd->sport = &th->th_sport; 8606 pd->dport = &th->th_dport; 8607 break; 8608 } 8609 case IPPROTO_UDP: { 8610 struct udphdr *uh = &pd->hdr.udp; 8611 8612 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8613 reason, AF_INET)) { 8614 *action = PF_DROP; 8615 REASON_SET(reason, PFRES_SHORT); 8616 return (-1); 8617 } 8618 *hdrlen = sizeof(*uh); 8619 if (uh->uh_dport == 0 || 8620 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8621 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8622 *action = PF_DROP; 8623 REASON_SET(reason, PFRES_SHORT); 8624 return (-1); 8625 } 8626 pd->sport = &uh->uh_sport; 8627 pd->dport = &uh->uh_dport; 8628 break; 8629 } 8630 case IPPROTO_SCTP: { 8631 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8632 action, reason, AF_INET)) { 8633 *action = PF_DROP; 8634 REASON_SET(reason, PFRES_SHORT); 8635 return (-1); 8636 } 8637 *hdrlen = sizeof(pd->hdr.sctp); 8638 pd->p_len = pd->tot_len - *off; 8639 8640 pd->sport = &pd->hdr.sctp.src_port; 8641 pd->dport = &pd->hdr.sctp.dest_port; 8642 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8643 *action = PF_DROP; 8644 REASON_SET(reason, PFRES_SHORT); 8645 return (-1); 8646 } 8647 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8648 *action = PF_DROP; 8649 REASON_SET(reason, PFRES_SHORT); 8650 return (-1); 8651 } 8652 break; 8653 } 8654 case IPPROTO_ICMP: { 8655 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp, ICMP_MINLEN, 8656 action, reason, AF_INET)) { 8657 *action = PF_DROP; 8658 REASON_SET(reason, PFRES_SHORT); 8659 return (-1); 8660 } 8661 *hdrlen = ICMP_MINLEN; 8662 break; 8663 } 8664 } 8665 break; 8666 } 8667 #endif 8668 #ifdef INET6 8669 case AF_INET6: { 8670 struct ip6_hdr *h; 8671 int terminal = 0; 8672 8673 h = mtod(m, struct ip6_hdr *); 8674 pd->src = (struct pf_addr *)&h->ip6_src; 8675 pd->dst = (struct pf_addr *)&h->ip6_dst; 8676 pd->sport = pd->dport = NULL; 8677 pd->ip_sum = NULL; 8678 pd->proto_sum = NULL; 8679 pd->dir = dir; 8680 pd->sidx = (dir == PF_IN) ? 0 : 1; 8681 pd->didx = (dir == PF_IN) ? 1 : 0; 8682 pd->tos = IPV6_DSCP(h); 8683 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8684 *off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8685 pd->proto = h->ip6_nxt; 8686 pd->act.rtableid = -1; 8687 8688 do { 8689 switch (pd->proto) { 8690 case IPPROTO_FRAGMENT: 8691 if (kif == NULL || r == NULL) /* pflog */ 8692 *action = PF_DROP; 8693 else 8694 *action = pf_test_fragment(r, kif, 8695 m, h, pd, a, ruleset); 8696 if (*action == PF_DROP) 8697 REASON_SET(reason, PFRES_FRAG); 8698 return (-1); 8699 case IPPROTO_ROUTING: { 8700 struct ip6_rthdr rthdr; 8701 8702 if (pd->rh_cnt++) { 8703 DPFPRINTF(PF_DEBUG_MISC, 8704 ("pf: IPv6 more than one rthdr")); 8705 *action = PF_DROP; 8706 REASON_SET(reason, PFRES_IPOPTIONS); 8707 return (-1); 8708 } 8709 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), 8710 NULL, reason, pd->af)) { 8711 DPFPRINTF(PF_DEBUG_MISC, 8712 ("pf: IPv6 short rthdr")); 8713 *action = PF_DROP; 8714 REASON_SET(reason, PFRES_SHORT); 8715 return (-1); 8716 } 8717 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8718 DPFPRINTF(PF_DEBUG_MISC, 8719 ("pf: IPv6 rthdr0")); 8720 *action = PF_DROP; 8721 REASON_SET(reason, PFRES_IPOPTIONS); 8722 return (-1); 8723 } 8724 /* FALLTHROUGH */ 8725 } 8726 case IPPROTO_AH: 8727 case IPPROTO_HOPOPTS: 8728 case IPPROTO_DSTOPTS: { 8729 /* get next header and header length */ 8730 struct ip6_ext opt6; 8731 8732 if (!pf_pull_hdr(m, *off, &opt6, sizeof(opt6), 8733 NULL, reason, pd->af)) { 8734 DPFPRINTF(PF_DEBUG_MISC, 8735 ("pf: IPv6 short opt")); 8736 *action = PF_DROP; 8737 return (-1); 8738 } 8739 if (pd->proto == IPPROTO_AH) 8740 *off += (opt6.ip6e_len + 2) * 4; 8741 else 8742 *off += (opt6.ip6e_len + 1) * 8; 8743 pd->proto = opt6.ip6e_nxt; 8744 /* goto the next header */ 8745 break; 8746 } 8747 default: 8748 terminal++; 8749 break; 8750 } 8751 } while (!terminal); 8752 8753 switch (pd->proto) { 8754 case IPPROTO_TCP: { 8755 struct tcphdr *th = &pd->hdr.tcp; 8756 8757 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8758 reason, AF_INET6)) { 8759 *action = PF_DROP; 8760 REASON_SET(reason, PFRES_SHORT); 8761 return (-1); 8762 } 8763 *hdrlen = sizeof(*th); 8764 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8765 pd->sport = &th->th_sport; 8766 pd->dport = &th->th_dport; 8767 break; 8768 } 8769 case IPPROTO_UDP: { 8770 struct udphdr *uh = &pd->hdr.udp; 8771 8772 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8773 reason, AF_INET6)) { 8774 *action = PF_DROP; 8775 REASON_SET(reason, PFRES_SHORT); 8776 return (-1); 8777 } 8778 *hdrlen = sizeof(*uh); 8779 if (uh->uh_dport == 0 || 8780 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8781 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8782 *action = PF_DROP; 8783 REASON_SET(reason, PFRES_SHORT); 8784 return (-1); 8785 } 8786 pd->sport = &uh->uh_sport; 8787 pd->dport = &uh->uh_dport; 8788 break; 8789 } 8790 case IPPROTO_SCTP: { 8791 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8792 action, reason, AF_INET6)) { 8793 *action = PF_DROP; 8794 REASON_SET(reason, PFRES_SHORT); 8795 return (-1); 8796 } 8797 *hdrlen = sizeof(pd->hdr.sctp); 8798 pd->p_len = pd->tot_len - *off; 8799 8800 pd->sport = &pd->hdr.sctp.src_port; 8801 pd->dport = &pd->hdr.sctp.dest_port; 8802 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8803 *action = PF_DROP; 8804 REASON_SET(reason, PFRES_SHORT); 8805 return (-1); 8806 } 8807 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8808 *action = PF_DROP; 8809 REASON_SET(reason, PFRES_SHORT); 8810 return (-1); 8811 } 8812 break; 8813 } 8814 case IPPROTO_ICMPV6: { 8815 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8816 8817 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8818 action, reason, AF_INET6)) { 8819 *action = PF_DROP; 8820 REASON_SET(reason, PFRES_SHORT); 8821 return (-1); 8822 } 8823 /* ICMP headers we look further into to match state */ 8824 switch (pd->hdr.icmp6.icmp6_type) { 8825 case MLD_LISTENER_QUERY: 8826 case MLD_LISTENER_REPORT: 8827 icmp_hlen = sizeof(struct mld_hdr); 8828 break; 8829 case ND_NEIGHBOR_SOLICIT: 8830 case ND_NEIGHBOR_ADVERT: 8831 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8832 break; 8833 } 8834 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8835 !pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8836 action, reason, AF_INET6)) { 8837 *action = PF_DROP; 8838 REASON_SET(reason, PFRES_SHORT); 8839 return (-1); 8840 } 8841 *hdrlen = icmp_hlen; 8842 break; 8843 } 8844 } 8845 break; 8846 } 8847 #endif 8848 default: 8849 panic("pf_setup_pdesc called with illegal af %u", af); 8850 } 8851 return (0); 8852 } 8853 8854 static void 8855 pf_counters_inc(int action, struct pf_pdesc *pd, 8856 struct pfi_kkif *kif, struct pf_kstate *s, 8857 struct pf_krule *r, struct pf_krule *a) 8858 { 8859 struct pf_krule *tr, *nr; 8860 int dir = pd->dir; 8861 int dirndx; 8862 8863 pf_counter_u64_critical_enter(); 8864 pf_counter_u64_add_protected( 8865 &kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8866 pd->tot_len); 8867 pf_counter_u64_add_protected( 8868 &kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8869 1); 8870 8871 if (action == PF_PASS || r->action == PF_DROP) { 8872 dirndx = (dir == PF_OUT); 8873 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8874 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8875 pf_update_timestamp(r); 8876 8877 if (a != NULL) { 8878 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8879 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8880 } 8881 if (s != NULL) { 8882 struct pf_krule_item *ri; 8883 8884 if (s->nat_rule.ptr != NULL) { 8885 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8886 1); 8887 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8888 pd->tot_len); 8889 } 8890 if (s->src_node != NULL) { 8891 counter_u64_add(s->src_node->packets[dirndx], 8892 1); 8893 counter_u64_add(s->src_node->bytes[dirndx], 8894 pd->tot_len); 8895 } 8896 if (s->nat_src_node != NULL) { 8897 counter_u64_add(s->nat_src_node->packets[dirndx], 8898 1); 8899 counter_u64_add(s->nat_src_node->bytes[dirndx], 8900 pd->tot_len); 8901 } 8902 dirndx = (dir == s->direction) ? 0 : 1; 8903 s->packets[dirndx]++; 8904 s->bytes[dirndx] += pd->tot_len; 8905 8906 SLIST_FOREACH(ri, &s->match_rules, entry) { 8907 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8908 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8909 } 8910 } 8911 tr = r; 8912 nr = (s != NULL) ? s->nat_rule.ptr : pd->nat_rule; 8913 if (nr != NULL && r == &V_pf_default_rule) 8914 tr = nr; 8915 if (tr->src.addr.type == PF_ADDR_TABLE) 8916 pfr_update_stats(tr->src.addr.p.tbl, 8917 (s == NULL) ? pd->src : 8918 &s->key[(s->direction == PF_IN)]-> 8919 addr[(s->direction == PF_OUT)], 8920 pd->af, pd->tot_len, dir == PF_OUT, 8921 r->action == PF_PASS, tr->src.neg); 8922 if (tr->dst.addr.type == PF_ADDR_TABLE) 8923 pfr_update_stats(tr->dst.addr.p.tbl, 8924 (s == NULL) ? pd->dst : 8925 &s->key[(s->direction == PF_IN)]-> 8926 addr[(s->direction == PF_IN)], 8927 pd->af, pd->tot_len, dir == PF_OUT, 8928 r->action == PF_PASS, tr->dst.neg); 8929 } 8930 pf_counter_u64_critical_exit(); 8931 } 8932 8933 #ifdef INET 8934 int 8935 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8936 struct inpcb *inp, struct pf_rule_actions *default_actions) 8937 { 8938 struct pfi_kkif *kif; 8939 u_short action, reason = 0; 8940 struct mbuf *m = *m0; 8941 struct ip *h = NULL; 8942 struct m_tag *mtag; 8943 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8944 struct pf_kstate *s = NULL; 8945 struct pf_kruleset *ruleset = NULL; 8946 struct pf_pdesc pd; 8947 int off, hdrlen, use_2nd_queue = 0; 8948 uint16_t tag; 8949 uint8_t rt; 8950 8951 PF_RULES_RLOCK_TRACKER; 8952 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8953 M_ASSERTPKTHDR(m); 8954 8955 if (!V_pf_status.running) 8956 return (PF_PASS); 8957 8958 PF_RULES_RLOCK(); 8959 8960 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8961 8962 if (__predict_false(kif == NULL)) { 8963 DPFPRINTF(PF_DEBUG_URGENT, 8964 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8965 PF_RULES_RUNLOCK(); 8966 return (PF_DROP); 8967 } 8968 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8969 PF_RULES_RUNLOCK(); 8970 return (PF_PASS); 8971 } 8972 8973 if (m->m_flags & M_SKIP_FIREWALL) { 8974 PF_RULES_RUNLOCK(); 8975 return (PF_PASS); 8976 } 8977 8978 if (__predict_false(! M_WRITABLE(*m0))) { 8979 m = *m0 = m_unshare(*m0, M_NOWAIT); 8980 if (*m0 == NULL) 8981 return (PF_DROP); 8982 } 8983 8984 if (__predict_false(m->m_len < sizeof(struct ip)) && 8985 (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8986 DPFPRINTF(PF_DEBUG_URGENT, 8987 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8988 PF_RULES_RUNLOCK(); 8989 return (PF_DROP); 8990 } 8991 8992 memset(&pd, 0, sizeof(pd)); 8993 pd.dir = dir; 8994 8995 if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 8996 /* We do IP header normalization and packet reassembly here */ 8997 action = PF_DROP; 8998 goto done; 8999 } 9000 m = *m0; /* pf_normalize messes with m0 */ 9001 h = mtod(m, struct ip *); 9002 9003 if (pf_setup_pdesc(AF_INET, dir, &pd, m, &action, &reason, kif, &a, &r, 9004 &ruleset, &off, &hdrlen, default_actions) == -1) { 9005 if (action != PF_PASS) 9006 pd.act.log |= PF_LOG_FORCE; 9007 goto done; 9008 } 9009 9010 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9011 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9012 9013 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9014 pd.pf_mtag->if_idxgen); 9015 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9016 PF_RULES_RUNLOCK(); 9017 m_freem(*m0); 9018 *m0 = NULL; 9019 return (PF_PASS); 9020 } 9021 PF_RULES_RUNLOCK(); 9022 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 9023 *m0 = NULL; 9024 return (PF_PASS); 9025 } 9026 9027 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9028 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9029 /* Dummynet re-injects packets after they've 9030 * completed their delay. We've already 9031 * processed them, so pass unconditionally. */ 9032 9033 /* But only once. We may see the packet multiple times (e.g. 9034 * PFIL_IN/PFIL_OUT). */ 9035 pf_dummynet_flag_remove(m, pd.pf_mtag); 9036 PF_RULES_RUNLOCK(); 9037 9038 return (PF_PASS); 9039 } 9040 9041 if (__predict_false(ip_divert_ptr != NULL) && 9042 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 9043 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 9044 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 9045 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 9046 if (pd.pf_mtag == NULL && 9047 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9048 action = PF_DROP; 9049 goto done; 9050 } 9051 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 9052 } 9053 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 9054 m->m_flags |= M_FASTFWD_OURS; 9055 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9056 } 9057 m_tag_delete(m, mtag); 9058 9059 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9060 if (mtag != NULL) 9061 m_tag_delete(m, mtag); 9062 } 9063 9064 /* handle fragments that didn't get reassembled by normalization */ 9065 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 9066 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 9067 goto done; 9068 } 9069 9070 switch (h->ip_p) { 9071 case IPPROTO_TCP: { 9072 9073 /* Respond to SYN with a syncookie. */ 9074 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9075 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9076 pf_syncookie_send(m, off, &pd); 9077 action = PF_DROP; 9078 break; 9079 } 9080 9081 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9082 use_2nd_queue = 1; 9083 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 9084 if (action == PF_DROP) 9085 goto done; 9086 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 9087 if (action == PF_PASS) { 9088 if (V_pfsync_update_state_ptr != NULL) 9089 V_pfsync_update_state_ptr(s); 9090 r = s->rule.ptr; 9091 a = s->anchor.ptr; 9092 } else if (s == NULL) { 9093 /* Validate remote SYN|ACK, re-create original SYN if 9094 * valid. */ 9095 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9096 TH_ACK && pf_syncookie_validate(&pd) && 9097 pd.dir == PF_IN) { 9098 struct mbuf *msyn; 9099 9100 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 9101 &pd); 9102 if (msyn == NULL) { 9103 action = PF_DROP; 9104 break; 9105 } 9106 9107 action = pf_test(dir, pflags, ifp, &msyn, inp, 9108 &pd.act); 9109 m_freem(msyn); 9110 if (action != PF_PASS) 9111 break; 9112 9113 action = pf_test_state_tcp(&s, kif, m, off, h, 9114 &pd, &reason); 9115 if (action != PF_PASS || s == NULL) { 9116 action = PF_DROP; 9117 break; 9118 } 9119 9120 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9121 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9122 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9123 action = pf_synproxy(&pd, &s, &reason); 9124 break; 9125 } else { 9126 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9127 &a, &ruleset, inp, hdrlen); 9128 } 9129 } 9130 break; 9131 } 9132 9133 case IPPROTO_UDP: { 9134 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 9135 if (action == PF_PASS) { 9136 if (V_pfsync_update_state_ptr != NULL) 9137 V_pfsync_update_state_ptr(s); 9138 r = s->rule.ptr; 9139 a = s->anchor.ptr; 9140 } else if (s == NULL) 9141 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9142 &a, &ruleset, inp, hdrlen); 9143 break; 9144 } 9145 9146 case IPPROTO_SCTP: { 9147 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 9148 if (action == PF_DROP) 9149 goto done; 9150 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 9151 &reason); 9152 if (action == PF_PASS) { 9153 if (V_pfsync_update_state_ptr != NULL) 9154 V_pfsync_update_state_ptr(s); 9155 r = s->rule.ptr; 9156 a = s->anchor.ptr; 9157 } else if (s == NULL) { 9158 action = pf_test_rule(&r, &s, kif, m, off, 9159 &pd, &a, &ruleset, inp, hdrlen); 9160 } 9161 break; 9162 } 9163 9164 case IPPROTO_ICMP: { 9165 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 9166 if (action == PF_PASS) { 9167 if (V_pfsync_update_state_ptr != NULL) 9168 V_pfsync_update_state_ptr(s); 9169 r = s->rule.ptr; 9170 a = s->anchor.ptr; 9171 } else if (s == NULL) 9172 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9173 &a, &ruleset, inp, hdrlen); 9174 break; 9175 } 9176 9177 case IPPROTO_ICMPV6: { 9178 action = PF_DROP; 9179 DPFPRINTF(PF_DEBUG_MISC, 9180 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9181 goto done; 9182 } 9183 9184 default: 9185 action = pf_test_state_other(&s, kif, m, &pd); 9186 if (action == PF_PASS) { 9187 if (V_pfsync_update_state_ptr != NULL) 9188 V_pfsync_update_state_ptr(s); 9189 r = s->rule.ptr; 9190 a = s->anchor.ptr; 9191 } else if (s == NULL) 9192 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9193 &a, &ruleset, inp, hdrlen); 9194 break; 9195 } 9196 9197 done: 9198 PF_RULES_RUNLOCK(); 9199 if (action == PF_PASS && h->ip_hl > 5 && 9200 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9201 action = PF_DROP; 9202 REASON_SET(&reason, PFRES_IPOPTIONS); 9203 pd.act.log = PF_LOG_FORCE; 9204 DPFPRINTF(PF_DEBUG_MISC, 9205 ("pf: dropping packet with ip options\n")); 9206 } 9207 9208 if (s) { 9209 uint8_t log = pd.act.log; 9210 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9211 pd.act.log |= log; 9212 tag = s->tag; 9213 rt = s->rt; 9214 } else { 9215 tag = r->tag; 9216 rt = r->rt; 9217 } 9218 9219 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9220 action = PF_DROP; 9221 REASON_SET(&reason, PFRES_MEMORY); 9222 } 9223 9224 pf_scrub_ip(m, &pd); 9225 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9226 pf_normalize_mss(m, off, &pd); 9227 9228 if (pd.act.rtableid >= 0) 9229 M_SETFIB(m, pd.act.rtableid); 9230 9231 if (pd.act.flags & PFSTATE_SETPRIO) { 9232 if (pd.tos & IPTOS_LOWDELAY) 9233 use_2nd_queue = 1; 9234 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9235 action = PF_DROP; 9236 REASON_SET(&reason, PFRES_MEMORY); 9237 pd.act.log = PF_LOG_FORCE; 9238 DPFPRINTF(PF_DEBUG_MISC, 9239 ("pf: failed to allocate 802.1q mtag\n")); 9240 } 9241 } 9242 9243 #ifdef ALTQ 9244 if (action == PF_PASS && pd.act.qid) { 9245 if (pd.pf_mtag == NULL && 9246 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9247 action = PF_DROP; 9248 REASON_SET(&reason, PFRES_MEMORY); 9249 } else { 9250 if (s != NULL) 9251 pd.pf_mtag->qid_hash = pf_state_hash(s); 9252 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9253 pd.pf_mtag->qid = pd.act.pqid; 9254 else 9255 pd.pf_mtag->qid = pd.act.qid; 9256 /* Add hints for ecn. */ 9257 pd.pf_mtag->hdr = h; 9258 } 9259 } 9260 #endif /* ALTQ */ 9261 9262 /* 9263 * connections redirected to loopback should not match sockets 9264 * bound specifically to loopback due to security implications, 9265 * see tcp_input() and in_pcblookup_listen(). 9266 */ 9267 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9268 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9269 (s->nat_rule.ptr->action == PF_RDR || 9270 s->nat_rule.ptr->action == PF_BINAT) && 9271 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 9272 m->m_flags |= M_SKIP_FIREWALL; 9273 9274 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 9275 r->divert.port && !PACKET_LOOPED(&pd)) { 9276 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9277 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9278 if (mtag != NULL) { 9279 ((struct pf_divert_mtag *)(mtag+1))->port = 9280 ntohs(r->divert.port); 9281 ((struct pf_divert_mtag *)(mtag+1))->idir = 9282 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9283 PF_DIVERT_MTAG_DIR_OUT; 9284 9285 if (s) 9286 PF_STATE_UNLOCK(s); 9287 9288 m_tag_prepend(m, mtag); 9289 if (m->m_flags & M_FASTFWD_OURS) { 9290 if (pd.pf_mtag == NULL && 9291 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9292 action = PF_DROP; 9293 REASON_SET(&reason, PFRES_MEMORY); 9294 pd.act.log = PF_LOG_FORCE; 9295 DPFPRINTF(PF_DEBUG_MISC, 9296 ("pf: failed to allocate tag\n")); 9297 } else { 9298 pd.pf_mtag->flags |= 9299 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9300 m->m_flags &= ~M_FASTFWD_OURS; 9301 } 9302 } 9303 ip_divert_ptr(*m0, dir == PF_IN); 9304 *m0 = NULL; 9305 9306 return (action); 9307 } else { 9308 /* XXX: ipfw has the same behaviour! */ 9309 action = PF_DROP; 9310 REASON_SET(&reason, PFRES_MEMORY); 9311 pd.act.log = PF_LOG_FORCE; 9312 DPFPRINTF(PF_DEBUG_MISC, 9313 ("pf: failed to allocate divert tag\n")); 9314 } 9315 } 9316 /* this flag will need revising if the pkt is forwarded */ 9317 if (pd.pf_mtag) 9318 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9319 9320 if (pd.act.log) { 9321 struct pf_krule *lr; 9322 struct pf_krule_item *ri; 9323 9324 if (s != NULL && s->nat_rule.ptr != NULL && 9325 s->nat_rule.ptr->log & PF_LOG_ALL) 9326 lr = s->nat_rule.ptr; 9327 else 9328 lr = r; 9329 9330 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9331 PFLOG_PACKET(kif, m, action, reason, lr, a, 9332 ruleset, &pd, (s == NULL)); 9333 if (s) { 9334 SLIST_FOREACH(ri, &s->match_rules, entry) 9335 if (ri->r->log & PF_LOG_ALL) 9336 PFLOG_PACKET(kif, m, action, 9337 reason, ri->r, a, ruleset, &pd, 0); 9338 } 9339 } 9340 9341 pf_counters_inc(action, &pd, kif, s, r, a); 9342 9343 switch (action) { 9344 case PF_SYNPROXY_DROP: 9345 m_freem(*m0); 9346 case PF_DEFER: 9347 *m0 = NULL; 9348 action = PF_PASS; 9349 break; 9350 case PF_DROP: 9351 m_freem(*m0); 9352 *m0 = NULL; 9353 break; 9354 default: 9355 /* pf_route() returns unlocked. */ 9356 if (rt) { 9357 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9358 goto out; 9359 } 9360 if (pf_dummynet(&pd, s, r, m0) != 0) { 9361 action = PF_DROP; 9362 REASON_SET(&reason, PFRES_MEMORY); 9363 } 9364 break; 9365 } 9366 9367 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9368 9369 if (s && action != PF_DROP) { 9370 if (!s->if_index_in && dir == PF_IN) 9371 s->if_index_in = ifp->if_index; 9372 else if (!s->if_index_out && dir == PF_OUT) 9373 s->if_index_out = ifp->if_index; 9374 } 9375 9376 if (s) 9377 PF_STATE_UNLOCK(s); 9378 9379 out: 9380 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9381 9382 return (action); 9383 } 9384 #endif /* INET */ 9385 9386 #ifdef INET6 9387 int 9388 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 9389 struct pf_rule_actions *default_actions) 9390 { 9391 struct pfi_kkif *kif; 9392 u_short action, reason = 0; 9393 struct mbuf *m = *m0, *n = NULL; 9394 struct m_tag *mtag; 9395 struct ip6_hdr *h = NULL; 9396 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 9397 struct pf_kstate *s = NULL; 9398 struct pf_kruleset *ruleset = NULL; 9399 struct pf_pdesc pd; 9400 int off, hdrlen, use_2nd_queue = 0; 9401 uint16_t tag; 9402 uint8_t rt; 9403 9404 PF_RULES_RLOCK_TRACKER; 9405 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 9406 M_ASSERTPKTHDR(m); 9407 9408 if (!V_pf_status.running) 9409 return (PF_PASS); 9410 9411 PF_RULES_RLOCK(); 9412 9413 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9414 if (__predict_false(kif == NULL)) { 9415 DPFPRINTF(PF_DEBUG_URGENT, 9416 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 9417 PF_RULES_RUNLOCK(); 9418 return (PF_DROP); 9419 } 9420 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 9421 PF_RULES_RUNLOCK(); 9422 return (PF_PASS); 9423 } 9424 9425 if (m->m_flags & M_SKIP_FIREWALL) { 9426 PF_RULES_RUNLOCK(); 9427 return (PF_PASS); 9428 } 9429 9430 /* 9431 * If we end up changing IP addresses (e.g. binat) the stack may get 9432 * confused and fail to send the icmp6 packet too big error. Just send 9433 * it here, before we do any NAT. 9434 */ 9435 if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 9436 PF_RULES_RUNLOCK(); 9437 *m0 = NULL; 9438 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 9439 return (PF_DROP); 9440 } 9441 9442 if (__predict_false(! M_WRITABLE(*m0))) { 9443 m = *m0 = m_unshare(*m0, M_NOWAIT); 9444 if (*m0 == NULL) 9445 return (PF_DROP); 9446 } 9447 9448 if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) && 9449 (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 9450 DPFPRINTF(PF_DEBUG_URGENT, 9451 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 9452 ", pullup failed\n")); 9453 PF_RULES_RUNLOCK(); 9454 return (PF_DROP); 9455 } 9456 9457 memset(&pd, 0, sizeof(pd)); 9458 pd.dir = dir; 9459 9460 /* We do IP header normalization and packet reassembly here */ 9461 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 9462 action = PF_DROP; 9463 goto done; 9464 } 9465 m = *m0; /* pf_normalize messes with m0 */ 9466 h = mtod(m, struct ip6_hdr *); 9467 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 9468 9469 if (pf_setup_pdesc(AF_INET6, dir, &pd, m, &action, &reason, kif, &a, &r, 9470 &ruleset, &off, &hdrlen, default_actions) == -1) { 9471 if (action != PF_PASS) 9472 pd.act.log |= PF_LOG_FORCE; 9473 goto done; 9474 } 9475 9476 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9477 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9478 9479 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9480 pd.pf_mtag->if_idxgen); 9481 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9482 PF_RULES_RUNLOCK(); 9483 m_freem(*m0); 9484 *m0 = NULL; 9485 return (PF_PASS); 9486 } 9487 PF_RULES_RUNLOCK(); 9488 nd6_output_ifp(ifp, ifp, m, 9489 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 9490 *m0 = NULL; 9491 return (PF_PASS); 9492 } 9493 9494 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9495 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9496 pf_dummynet_flag_remove(m, pd.pf_mtag); 9497 /* Dummynet re-injects packets after they've 9498 * completed their delay. We've already 9499 * processed them, so pass unconditionally. */ 9500 PF_RULES_RUNLOCK(); 9501 return (PF_PASS); 9502 } 9503 9504 /* 9505 * we do not support jumbogram. if we keep going, zero ip6_plen 9506 * will do something bad, so drop the packet for now. 9507 */ 9508 if (htons(h->ip6_plen) == 0) { 9509 action = PF_DROP; 9510 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 9511 goto done; 9512 } 9513 9514 /* if there's no routing header, use unmodified mbuf for checksumming */ 9515 if (!n) 9516 n = m; 9517 9518 switch (pd.proto) { 9519 case IPPROTO_TCP: { 9520 /* Respond to SYN with a syncookie. */ 9521 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9522 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9523 pf_syncookie_send(m, off, &pd); 9524 action = PF_DROP; 9525 break; 9526 } 9527 9528 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9529 use_2nd_queue = 1; 9530 9531 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 9532 if (action == PF_DROP) 9533 goto done; 9534 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 9535 if (action == PF_PASS) { 9536 if (V_pfsync_update_state_ptr != NULL) 9537 V_pfsync_update_state_ptr(s); 9538 r = s->rule.ptr; 9539 a = s->anchor.ptr; 9540 } else if (s == NULL) { 9541 /* Validate remote SYN|ACK, re-create original SYN if 9542 * valid. */ 9543 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9544 TH_ACK && pf_syncookie_validate(&pd) && 9545 pd.dir == PF_IN) { 9546 struct mbuf *msyn; 9547 9548 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 9549 off, &pd); 9550 if (msyn == NULL) { 9551 action = PF_DROP; 9552 break; 9553 } 9554 9555 action = pf_test6(dir, pflags, ifp, &msyn, inp, 9556 &pd.act); 9557 m_freem(msyn); 9558 if (action != PF_PASS) 9559 break; 9560 9561 action = pf_test_state_tcp(&s, kif, m, off, h, 9562 &pd, &reason); 9563 if (action != PF_PASS || s == NULL) { 9564 action = PF_DROP; 9565 break; 9566 } 9567 9568 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9569 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9570 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9571 9572 action = pf_synproxy(&pd, &s, &reason); 9573 break; 9574 } else { 9575 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9576 &a, &ruleset, inp, hdrlen); 9577 } 9578 } 9579 break; 9580 } 9581 9582 case IPPROTO_UDP: { 9583 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 9584 if (action == PF_PASS) { 9585 if (V_pfsync_update_state_ptr != NULL) 9586 V_pfsync_update_state_ptr(s); 9587 r = s->rule.ptr; 9588 a = s->anchor.ptr; 9589 } else if (s == NULL) 9590 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9591 &a, &ruleset, inp, hdrlen); 9592 break; 9593 } 9594 9595 case IPPROTO_SCTP: { 9596 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 9597 if (action == PF_DROP) 9598 goto done; 9599 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 9600 &reason); 9601 if (action == PF_PASS) { 9602 if (V_pfsync_update_state_ptr != NULL) 9603 V_pfsync_update_state_ptr(s); 9604 r = s->rule.ptr; 9605 a = s->anchor.ptr; 9606 } else if (s == NULL) { 9607 action = pf_test_rule(&r, &s, kif, m, off, 9608 &pd, &a, &ruleset, inp, hdrlen); 9609 } 9610 break; 9611 } 9612 9613 case IPPROTO_ICMP: { 9614 action = PF_DROP; 9615 DPFPRINTF(PF_DEBUG_MISC, 9616 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 9617 goto done; 9618 } 9619 9620 case IPPROTO_ICMPV6: { 9621 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 9622 if (action == PF_PASS) { 9623 if (V_pfsync_update_state_ptr != NULL) 9624 V_pfsync_update_state_ptr(s); 9625 r = s->rule.ptr; 9626 a = s->anchor.ptr; 9627 } else if (s == NULL) 9628 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9629 &a, &ruleset, inp, hdrlen); 9630 break; 9631 } 9632 9633 default: 9634 action = pf_test_state_other(&s, kif, m, &pd); 9635 if (action == PF_PASS) { 9636 if (V_pfsync_update_state_ptr != NULL) 9637 V_pfsync_update_state_ptr(s); 9638 r = s->rule.ptr; 9639 a = s->anchor.ptr; 9640 } else if (s == NULL) 9641 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9642 &a, &ruleset, inp, hdrlen); 9643 break; 9644 } 9645 9646 done: 9647 PF_RULES_RUNLOCK(); 9648 if (n != m) { 9649 m_freem(n); 9650 n = NULL; 9651 } 9652 9653 /* handle dangerous IPv6 extension headers. */ 9654 if (action == PF_PASS && pd.rh_cnt && 9655 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9656 action = PF_DROP; 9657 REASON_SET(&reason, PFRES_IPOPTIONS); 9658 pd.act.log = r->log; 9659 DPFPRINTF(PF_DEBUG_MISC, 9660 ("pf: dropping packet with dangerous v6 headers\n")); 9661 } 9662 9663 if (s) { 9664 uint8_t log = pd.act.log; 9665 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9666 pd.act.log |= log; 9667 tag = s->tag; 9668 rt = s->rt; 9669 } else { 9670 tag = r->tag; 9671 rt = r->rt; 9672 } 9673 9674 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9675 action = PF_DROP; 9676 REASON_SET(&reason, PFRES_MEMORY); 9677 } 9678 9679 pf_scrub_ip6(m, &pd); 9680 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9681 pf_normalize_mss(m, off, &pd); 9682 9683 if (pd.act.rtableid >= 0) 9684 M_SETFIB(m, pd.act.rtableid); 9685 9686 if (pd.act.flags & PFSTATE_SETPRIO) { 9687 if (pd.tos & IPTOS_LOWDELAY) 9688 use_2nd_queue = 1; 9689 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9690 action = PF_DROP; 9691 REASON_SET(&reason, PFRES_MEMORY); 9692 pd.act.log = PF_LOG_FORCE; 9693 DPFPRINTF(PF_DEBUG_MISC, 9694 ("pf: failed to allocate 802.1q mtag\n")); 9695 } 9696 } 9697 9698 #ifdef ALTQ 9699 if (action == PF_PASS && pd.act.qid) { 9700 if (pd.pf_mtag == NULL && 9701 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9702 action = PF_DROP; 9703 REASON_SET(&reason, PFRES_MEMORY); 9704 } else { 9705 if (s != NULL) 9706 pd.pf_mtag->qid_hash = pf_state_hash(s); 9707 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9708 pd.pf_mtag->qid = pd.act.pqid; 9709 else 9710 pd.pf_mtag->qid = pd.act.qid; 9711 /* Add hints for ecn. */ 9712 pd.pf_mtag->hdr = h; 9713 } 9714 } 9715 #endif /* ALTQ */ 9716 9717 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9718 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9719 (s->nat_rule.ptr->action == PF_RDR || 9720 s->nat_rule.ptr->action == PF_BINAT) && 9721 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 9722 m->m_flags |= M_SKIP_FIREWALL; 9723 9724 /* XXX: Anybody working on it?! */ 9725 if (r->divert.port) 9726 printf("pf: divert(9) is not supported for IPv6\n"); 9727 9728 if (pd.act.log) { 9729 struct pf_krule *lr; 9730 struct pf_krule_item *ri; 9731 9732 if (s != NULL && s->nat_rule.ptr != NULL && 9733 s->nat_rule.ptr->log & PF_LOG_ALL) 9734 lr = s->nat_rule.ptr; 9735 else 9736 lr = r; 9737 9738 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9739 PFLOG_PACKET(kif, m, action, reason, lr, a, ruleset, 9740 &pd, (s == NULL)); 9741 if (s) { 9742 SLIST_FOREACH(ri, &s->match_rules, entry) 9743 if (ri->r->log & PF_LOG_ALL) 9744 PFLOG_PACKET(kif, m, action, reason, 9745 ri->r, a, ruleset, &pd, 0); 9746 } 9747 } 9748 9749 pf_counters_inc(action, &pd, kif, s, r, a); 9750 9751 switch (action) { 9752 case PF_SYNPROXY_DROP: 9753 m_freem(*m0); 9754 case PF_DEFER: 9755 *m0 = NULL; 9756 action = PF_PASS; 9757 break; 9758 case PF_DROP: 9759 m_freem(*m0); 9760 *m0 = NULL; 9761 break; 9762 default: 9763 /* pf_route6() returns unlocked. */ 9764 if (rt) { 9765 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9766 goto out; 9767 } 9768 if (pf_dummynet(&pd, s, r, m0) != 0) { 9769 action = PF_DROP; 9770 REASON_SET(&reason, PFRES_MEMORY); 9771 } 9772 break; 9773 } 9774 9775 if (s && action != PF_DROP) { 9776 if (!s->if_index_in && dir == PF_IN) 9777 s->if_index_in = ifp->if_index; 9778 else if (!s->if_index_out && dir == PF_OUT) 9779 s->if_index_out = ifp->if_index; 9780 } 9781 9782 if (s) 9783 PF_STATE_UNLOCK(s); 9784 9785 /* If reassembled packet passed, create new fragments. */ 9786 if (action == PF_PASS && *m0 && dir == PF_OUT && 9787 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9788 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9789 9790 out: 9791 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 9792 9793 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9794 9795 return (action); 9796 } 9797 #endif /* INET6 */ 9798