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/random.h> 58 #include <sys/refcount.h> 59 #include <sys/sdt.h> 60 #include <sys/socket.h> 61 #include <sys/sysctl.h> 62 #include <sys/taskqueue.h> 63 #include <sys/ucred.h> 64 65 #include <crypto/sha2/sha512.h> 66 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/if_private.h> 70 #include <net/if_types.h> 71 #include <net/if_vlan_var.h> 72 #include <net/route.h> 73 #include <net/route/nhop.h> 74 #include <net/vnet.h> 75 76 #include <net/pfil.h> 77 #include <net/pfvar.h> 78 #include <net/if_pflog.h> 79 #include <net/if_pfsync.h> 80 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_var.h> 83 #include <netinet/in_fib.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_fw.h> 86 #include <netinet/ip_icmp.h> 87 #include <netinet/icmp_var.h> 88 #include <netinet/ip_var.h> 89 #include <netinet/tcp.h> 90 #include <netinet/tcp_fsm.h> 91 #include <netinet/tcp_seq.h> 92 #include <netinet/tcp_timer.h> 93 #include <netinet/tcp_var.h> 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 97 /* dummynet */ 98 #include <netinet/ip_dummynet.h> 99 #include <netinet/ip_fw.h> 100 #include <netpfil/ipfw/dn_heap.h> 101 #include <netpfil/ipfw/ip_fw_private.h> 102 #include <netpfil/ipfw/ip_dn_private.h> 103 104 #ifdef INET6 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet6/in6_fib.h> 111 #include <netinet6/scope6_var.h> 112 #endif /* INET6 */ 113 114 #include <netinet/sctp_header.h> 115 #include <netinet/sctp_crc32.h> 116 117 #include <netipsec/ah.h> 118 119 #include <machine/in_cksum.h> 120 #include <security/mac/mac_framework.h> 121 122 SDT_PROVIDER_DEFINE(pf); 123 SDT_PROBE_DEFINE2(pf, , test, reason_set, "int", "int"); 124 SDT_PROBE_DEFINE4(pf, ip, test, 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, ip, route_to, entry, "struct mbuf *", 132 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 133 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 134 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 135 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 136 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 137 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 138 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 139 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 140 "struct pf_krule *", "struct mbuf *", "int"); 141 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 142 "struct pf_sctp_source *"); 143 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 144 "struct pf_kstate *", "struct pf_sctp_source *"); 145 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 146 "int", "struct pf_pdesc *", "int"); 147 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 148 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 149 "int"); 150 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 151 "int"); 152 153 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 154 "struct mbuf *"); 155 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 156 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 157 "int", "struct pf_keth_rule *", "char *"); 158 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 159 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 160 "int", "struct pf_keth_rule *"); 161 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 162 SDT_PROBE_DEFINE2(pf, , log, log, "int", "const char *"); 163 164 /* 165 * Global variables 166 */ 167 168 /* state tables */ 169 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 170 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 173 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 174 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 175 VNET_DEFINE(struct pf_kstatus, pf_status); 176 177 VNET_DEFINE(u_int32_t, ticket_altqs_active); 178 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 179 VNET_DEFINE(int, altqs_inactive_open); 180 VNET_DEFINE(u_int32_t, ticket_pabuf); 181 182 static const int PF_HDR_LIMIT = 20; /* arbitrary limit */ 183 184 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 185 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 186 VNET_DEFINE(u_char, pf_tcp_secret[16]); 187 #define V_pf_tcp_secret VNET(pf_tcp_secret) 188 VNET_DEFINE(int, pf_tcp_secret_init); 189 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 190 VNET_DEFINE(int, pf_tcp_iss_off); 191 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 192 VNET_DECLARE(int, pf_vnet_active); 193 #define V_pf_vnet_active VNET(pf_vnet_active) 194 195 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 196 #define V_pf_purge_idx VNET(pf_purge_idx) 197 198 #ifdef PF_WANT_32_TO_64_COUNTER 199 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 200 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 201 202 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 203 VNET_DEFINE(size_t, pf_allrulecount); 204 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 205 #endif 206 207 #define PF_SCTP_MAX_ENDPOINTS 8 208 209 struct pf_sctp_endpoint; 210 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 211 struct pf_sctp_source { 212 sa_family_t af; 213 struct pf_addr addr; 214 TAILQ_ENTRY(pf_sctp_source) entry; 215 }; 216 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 217 struct pf_sctp_endpoint 218 { 219 uint32_t v_tag; 220 struct pf_sctp_sources sources; 221 RB_ENTRY(pf_sctp_endpoint) entry; 222 }; 223 static int 224 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 225 { 226 return (a->v_tag - b->v_tag); 227 } 228 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 229 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 230 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 231 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 232 static struct mtx_padalign pf_sctp_endpoints_mtx; 233 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 234 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 235 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 236 237 /* 238 * Queue for pf_intr() sends. 239 */ 240 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 241 struct pf_send_entry { 242 STAILQ_ENTRY(pf_send_entry) pfse_next; 243 struct mbuf *pfse_m; 244 enum { 245 PFSE_IP, 246 PFSE_IP6, 247 PFSE_ICMP, 248 PFSE_ICMP6, 249 } pfse_type; 250 struct { 251 int type; 252 int code; 253 int mtu; 254 } icmpopts; 255 }; 256 257 STAILQ_HEAD(pf_send_head, pf_send_entry); 258 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 259 #define V_pf_sendqueue VNET(pf_sendqueue) 260 261 static struct mtx_padalign pf_sendqueue_mtx; 262 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 263 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 264 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 265 266 /* 267 * Queue for pf_overload_task() tasks. 268 */ 269 struct pf_overload_entry { 270 SLIST_ENTRY(pf_overload_entry) next; 271 struct pf_addr addr; 272 sa_family_t af; 273 uint8_t dir; 274 struct pf_krule *rule; 275 }; 276 277 SLIST_HEAD(pf_overload_head, pf_overload_entry); 278 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 279 #define V_pf_overloadqueue VNET(pf_overloadqueue) 280 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 281 #define V_pf_overloadtask VNET(pf_overloadtask) 282 283 static struct mtx_padalign pf_overloadqueue_mtx; 284 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 285 "pf overload/flush queue", MTX_DEF); 286 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 287 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 288 289 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 290 struct mtx_padalign pf_unlnkdrules_mtx; 291 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 292 MTX_DEF); 293 294 struct sx pf_config_lock; 295 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 296 297 struct mtx_padalign pf_table_stats_lock; 298 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 299 MTX_DEF); 300 301 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 302 #define V_pf_sources_z VNET(pf_sources_z) 303 uma_zone_t pf_mtag_z; 304 VNET_DEFINE(uma_zone_t, pf_state_z); 305 VNET_DEFINE(uma_zone_t, pf_state_key_z); 306 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 307 308 VNET_DEFINE(struct unrhdr64, pf_stateid); 309 310 static void pf_src_tree_remove_state(struct pf_kstate *); 311 static int pf_check_threshold(struct pf_kthreshold *); 312 313 static void pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *, 314 struct pf_addr *, u_int16_t); 315 static int pf_modulate_sack(struct pf_pdesc *, 316 struct tcphdr *, struct pf_state_peer *); 317 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 318 u_int16_t *, u_int16_t *); 319 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 320 struct pf_addr *, struct pf_addr *, u_int16_t, 321 u_int16_t *, u_int16_t *, u_int16_t *, 322 u_int16_t *, u_int8_t, sa_family_t); 323 int pf_change_icmp_af(struct mbuf *, int, 324 struct pf_pdesc *, struct pf_pdesc *, 325 struct pf_addr *, struct pf_addr *, sa_family_t, 326 sa_family_t); 327 int pf_translate_icmp_af(int, void *); 328 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 329 int, sa_family_t, struct pf_krule *, int); 330 static void pf_detach_state(struct pf_kstate *); 331 static int pf_state_key_attach(struct pf_state_key *, 332 struct pf_state_key *, struct pf_kstate *); 333 static void pf_state_key_detach(struct pf_kstate *, int); 334 static int pf_state_key_ctor(void *, int, void *, int); 335 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 336 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 337 struct pf_mtag *pf_mtag); 338 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 339 struct pf_krule *, struct mbuf **); 340 static int pf_dummynet_route(struct pf_pdesc *, 341 struct pf_kstate *, struct pf_krule *, 342 struct ifnet *, const struct sockaddr *, struct mbuf **); 343 static int pf_test_eth_rule(int, struct pfi_kkif *, 344 struct mbuf **); 345 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 346 struct pf_pdesc *, struct pf_krule **, 347 struct pf_kruleset **, u_short *, struct inpcb *, 348 struct pf_krule_slist *); 349 static int pf_create_state(struct pf_krule *, 350 struct pf_test_ctx *, 351 struct pf_kstate **, u_int16_t, u_int16_t, 352 struct pf_krule_slist *match_rules); 353 static int pf_state_key_addr_setup(struct pf_pdesc *, 354 struct pf_state_key_cmp *, int); 355 static int pf_tcp_track_full(struct pf_kstate *, 356 struct pf_pdesc *, u_short *, int *, 357 struct pf_state_peer *, struct pf_state_peer *, 358 u_int8_t, u_int8_t); 359 static int pf_tcp_track_sloppy(struct pf_kstate *, 360 struct pf_pdesc *, u_short *, 361 struct pf_state_peer *, struct pf_state_peer *, 362 u_int8_t, u_int8_t); 363 static int pf_test_state(struct pf_kstate **, struct pf_pdesc *, 364 u_short *); 365 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 366 struct pf_pdesc *, struct pf_kstate **, 367 u_int16_t, u_int16_t, int, int *, int, int); 368 static int pf_test_state_icmp(struct pf_kstate **, 369 struct pf_pdesc *, u_short *); 370 static int pf_sctp_track(struct pf_kstate *, struct pf_pdesc *, 371 u_short *); 372 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 373 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 374 struct pfi_kkif *, struct pf_kstate *, int); 375 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 376 int, u_int16_t); 377 static int pf_check_proto_cksum(struct mbuf *, int, int, 378 u_int8_t, sa_family_t); 379 static int pf_walk_option(struct pf_pdesc *, struct ip *, 380 int, int, u_short *); 381 static int pf_walk_header(struct pf_pdesc *, struct ip *, u_short *); 382 #ifdef INET6 383 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 384 int, int, u_short *); 385 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 386 u_short *); 387 #endif 388 static void pf_print_state_parts(struct pf_kstate *, 389 struct pf_state_key *, struct pf_state_key *); 390 static int pf_patch_8(struct pf_pdesc *, u_int8_t *, u_int8_t, 391 bool); 392 static int pf_find_state(struct pf_pdesc *, 393 const struct pf_state_key_cmp *, struct pf_kstate **); 394 static bool pf_src_connlimit(struct pf_kstate *); 395 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 396 static void pf_counters_inc(int, struct pf_pdesc *, 397 struct pf_kstate *, struct pf_krule *, 398 struct pf_krule *, struct pf_krule_slist *); 399 static void pf_log_matches(struct pf_pdesc *, struct pf_krule *, 400 struct pf_krule *, struct pf_kruleset *, 401 struct pf_krule_slist *); 402 static void pf_overload_task(void *v, int pending); 403 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 404 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 405 struct pf_addr *, sa_family_t, struct pf_addr *, 406 struct pfi_kkif *, sa_family_t, pf_sn_types_t); 407 static u_int pf_purge_expired_states(u_int, int); 408 static void pf_purge_unlinked_rules(void); 409 static int pf_mtag_uminit(void *, int, int); 410 static void pf_mtag_free(struct m_tag *); 411 static void pf_packet_rework_nat(struct pf_pdesc *, int, 412 struct pf_state_key *); 413 #ifdef INET 414 static int pf_route(struct pf_krule *, 415 struct ifnet *, struct pf_kstate *, 416 struct pf_pdesc *, struct inpcb *); 417 #endif /* INET */ 418 #ifdef INET6 419 static void pf_change_a6(struct pf_addr *, u_int16_t *, 420 struct pf_addr *, u_int8_t); 421 static int pf_route6(struct pf_krule *, 422 struct ifnet *, struct pf_kstate *, 423 struct pf_pdesc *, struct inpcb *); 424 #endif /* INET6 */ 425 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 426 427 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 428 429 extern int pf_end_threads; 430 extern struct proc *pf_purge_proc; 431 432 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 433 434 #define PACKET_UNDO_NAT(_pd, _off, _s) \ 435 do { \ 436 struct pf_state_key *nk; \ 437 if ((pd->dir) == PF_OUT) \ 438 nk = (_s)->key[PF_SK_STACK]; \ 439 else \ 440 nk = (_s)->key[PF_SK_WIRE]; \ 441 pf_packet_rework_nat(_pd, _off, nk); \ 442 } while (0) 443 444 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 445 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 446 447 static struct pfi_kkif * 448 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 449 { 450 struct pfi_kkif *k = pd->kif; 451 452 SDT_PROBE2(pf, ip, , bound_iface, st, k); 453 454 /* Floating unless otherwise specified. */ 455 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 456 return (V_pfi_all); 457 458 /* 459 * Initially set to all, because we don't know what interface we'll be 460 * sending this out when we create the state. 461 */ 462 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN)) 463 return (V_pfi_all); 464 465 /* 466 * If this state is created based on another state (e.g. SCTP 467 * multihome) always set it floating initially. We can't know for sure 468 * what interface the actual traffic for this state will come in on. 469 */ 470 if (pd->related_rule) 471 return (V_pfi_all); 472 473 /* Don't overrule the interface for states created on incoming packets. */ 474 if (st->direction == PF_IN) 475 return (k); 476 477 /* No route-to, so don't overrule. */ 478 if (st->act.rt != PF_ROUTETO) 479 return (k); 480 481 /* Bind to the route-to interface. */ 482 return (st->act.rt_kif); 483 } 484 485 #define STATE_INC_COUNTERS(s) \ 486 do { \ 487 struct pf_krule_item *mrm; \ 488 counter_u64_add(s->rule->states_cur, 1); \ 489 counter_u64_add(s->rule->states_tot, 1); \ 490 if (s->anchor != NULL) { \ 491 counter_u64_add(s->anchor->states_cur, 1); \ 492 counter_u64_add(s->anchor->states_tot, 1); \ 493 } \ 494 if (s->nat_rule != NULL && s->nat_rule != s->rule) { \ 495 counter_u64_add(s->nat_rule->states_cur, 1); \ 496 counter_u64_add(s->nat_rule->states_tot, 1); \ 497 } \ 498 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 499 if (s->nat_rule != mrm->r) { \ 500 counter_u64_add(mrm->r->states_cur, 1); \ 501 counter_u64_add(mrm->r->states_tot, 1); \ 502 } \ 503 } \ 504 } while (0) 505 506 #define STATE_DEC_COUNTERS(s) \ 507 do { \ 508 struct pf_krule_item *mrm; \ 509 counter_u64_add(s->rule->states_cur, -1); \ 510 if (s->anchor != NULL) \ 511 counter_u64_add(s->anchor->states_cur, -1); \ 512 if (s->nat_rule != NULL && s->nat_rule != s->rule) \ 513 counter_u64_add(s->nat_rule->states_cur, -1); \ 514 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 515 if (s->nat_rule != mrm->r) { \ 516 counter_u64_add(mrm->r->states_cur, -1);\ 517 } \ 518 } while (0) 519 520 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 521 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 522 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 523 VNET_DEFINE(struct pf_idhash *, pf_idhash); 524 VNET_DEFINE(struct pf_srchash *, pf_srchash); 525 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 526 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 527 528 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 529 "pf(4)"); 530 531 VNET_DEFINE(u_long, pf_hashmask); 532 VNET_DEFINE(u_long, pf_srchashmask); 533 VNET_DEFINE(u_long, pf_udpendpointhashmask); 534 VNET_DEFINE_STATIC(u_long, pf_hashsize); 535 #define V_pf_hashsize VNET(pf_hashsize) 536 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 537 #define V_pf_srchashsize VNET(pf_srchashsize) 538 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 539 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 540 u_long pf_ioctl_maxcount = 65535; 541 542 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 543 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 544 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 545 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 546 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 547 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 548 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 549 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 550 551 VNET_DEFINE(void *, pf_swi_cookie); 552 VNET_DEFINE(struct intr_event *, pf_swi_ie); 553 554 VNET_DEFINE(uint32_t, pf_hashseed); 555 #define V_pf_hashseed VNET(pf_hashseed) 556 557 static void 558 pf_sctp_checksum(struct mbuf *m, int off) 559 { 560 uint32_t sum = 0; 561 562 /* Zero out the checksum, to enable recalculation. */ 563 m_copyback(m, off + offsetof(struct sctphdr, checksum), 564 sizeof(sum), (caddr_t)&sum); 565 566 sum = sctp_calculate_cksum(m, off); 567 568 m_copyback(m, off + offsetof(struct sctphdr, checksum), 569 sizeof(sum), (caddr_t)&sum); 570 } 571 572 int 573 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 574 { 575 576 switch (af) { 577 #ifdef INET 578 case AF_INET: 579 if (a->addr32[0] > b->addr32[0]) 580 return (1); 581 if (a->addr32[0] < b->addr32[0]) 582 return (-1); 583 break; 584 #endif /* INET */ 585 #ifdef INET6 586 case AF_INET6: 587 if (a->addr32[3] > b->addr32[3]) 588 return (1); 589 if (a->addr32[3] < b->addr32[3]) 590 return (-1); 591 if (a->addr32[2] > b->addr32[2]) 592 return (1); 593 if (a->addr32[2] < b->addr32[2]) 594 return (-1); 595 if (a->addr32[1] > b->addr32[1]) 596 return (1); 597 if (a->addr32[1] < b->addr32[1]) 598 return (-1); 599 if (a->addr32[0] > b->addr32[0]) 600 return (1); 601 if (a->addr32[0] < b->addr32[0]) 602 return (-1); 603 break; 604 #endif /* INET6 */ 605 default: 606 unhandled_af(af); 607 } 608 return (0); 609 } 610 611 static bool 612 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 613 { 614 switch (af) { 615 #ifdef INET 616 case AF_INET: 617 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 618 #endif /* INET */ 619 case AF_INET6: 620 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 621 default: 622 unhandled_af(af); 623 } 624 } 625 626 static void 627 pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk) 628 { 629 630 switch (pd->virtual_proto) { 631 case IPPROTO_TCP: { 632 struct tcphdr *th = &pd->hdr.tcp; 633 634 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 635 pf_change_ap(pd, pd->src, &th->th_sport, 636 &nk->addr[pd->sidx], nk->port[pd->sidx]); 637 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 638 pf_change_ap(pd, pd->dst, &th->th_dport, 639 &nk->addr[pd->didx], nk->port[pd->didx]); 640 m_copyback(pd->m, off, sizeof(*th), (caddr_t)th); 641 break; 642 } 643 case IPPROTO_UDP: { 644 struct udphdr *uh = &pd->hdr.udp; 645 646 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 647 pf_change_ap(pd, pd->src, &uh->uh_sport, 648 &nk->addr[pd->sidx], nk->port[pd->sidx]); 649 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 650 pf_change_ap(pd, pd->dst, &uh->uh_dport, 651 &nk->addr[pd->didx], nk->port[pd->didx]); 652 m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh); 653 break; 654 } 655 case IPPROTO_SCTP: { 656 struct sctphdr *sh = &pd->hdr.sctp; 657 658 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 659 pf_change_ap(pd, pd->src, &sh->src_port, 660 &nk->addr[pd->sidx], nk->port[pd->sidx]); 661 } 662 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 663 pf_change_ap(pd, pd->dst, &sh->dest_port, 664 &nk->addr[pd->didx], nk->port[pd->didx]); 665 } 666 667 break; 668 } 669 case IPPROTO_ICMP: { 670 struct icmp *ih = &pd->hdr.icmp; 671 672 if (nk->port[pd->sidx] != ih->icmp_id) { 673 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 674 ih->icmp_cksum, ih->icmp_id, 675 nk->port[pd->sidx], 0); 676 ih->icmp_id = nk->port[pd->sidx]; 677 pd->sport = &ih->icmp_id; 678 679 m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih); 680 } 681 /* FALLTHROUGH */ 682 } 683 default: 684 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 685 switch (pd->af) { 686 case AF_INET: 687 pf_change_a(&pd->src->v4.s_addr, 688 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 689 0); 690 break; 691 case AF_INET6: 692 pf_addrcpy(pd->src, &nk->addr[pd->sidx], 693 pd->af); 694 break; 695 default: 696 unhandled_af(pd->af); 697 } 698 } 699 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 700 switch (pd->af) { 701 case AF_INET: 702 pf_change_a(&pd->dst->v4.s_addr, 703 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 704 0); 705 break; 706 case AF_INET6: 707 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 708 pd->af); 709 break; 710 default: 711 unhandled_af(pd->af); 712 } 713 } 714 break; 715 } 716 } 717 718 static __inline uint32_t 719 pf_hashkey(const struct pf_state_key *sk) 720 { 721 uint32_t h; 722 723 h = murmur3_32_hash32((const uint32_t *)sk, 724 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 725 V_pf_hashseed); 726 727 return (h & V_pf_hashmask); 728 } 729 730 __inline uint32_t 731 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 732 { 733 uint32_t h; 734 735 switch (af) { 736 case AF_INET: 737 h = murmur3_32_hash32((uint32_t *)&addr->v4, 738 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 739 break; 740 case AF_INET6: 741 h = murmur3_32_hash32((uint32_t *)&addr->v6, 742 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 743 break; 744 default: 745 unhandled_af(af); 746 } 747 748 return (h & V_pf_srchashmask); 749 } 750 751 static inline uint32_t 752 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 753 { 754 uint32_t h; 755 756 h = murmur3_32_hash32((uint32_t *)endpoint, 757 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 758 V_pf_hashseed); 759 return (h & V_pf_udpendpointhashmask); 760 } 761 762 #ifdef ALTQ 763 static int 764 pf_state_hash(struct pf_kstate *s) 765 { 766 u_int32_t hv = (intptr_t)s / sizeof(*s); 767 768 hv ^= crc32(&s->src, sizeof(s->src)); 769 hv ^= crc32(&s->dst, sizeof(s->dst)); 770 if (hv == 0) 771 hv = 1; 772 return (hv); 773 } 774 #endif /* ALTQ */ 775 776 static __inline void 777 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 778 { 779 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 780 s->dst.state = newstate; 781 if (which == PF_PEER_DST) 782 return; 783 if (s->src.state == newstate) 784 return; 785 if (s->creatorid == V_pf_status.hostid && 786 s->key[PF_SK_STACK] != NULL && 787 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 788 !(TCPS_HAVEESTABLISHED(s->src.state) || 789 s->src.state == TCPS_CLOSED) && 790 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 791 atomic_add_32(&V_pf_status.states_halfopen, -1); 792 793 s->src.state = newstate; 794 } 795 796 bool 797 pf_init_threshold(struct pf_kthreshold *threshold, 798 u_int32_t limit, u_int32_t seconds) 799 { 800 threshold->limit = limit; 801 threshold->seconds = seconds; 802 threshold->cr = counter_rate_alloc(M_NOWAIT, seconds); 803 804 return (threshold->cr != NULL); 805 } 806 807 static int 808 pf_check_threshold(struct pf_kthreshold *threshold) 809 { 810 return (counter_ratecheck(threshold->cr, threshold->limit) < 0); 811 } 812 813 static bool 814 pf_src_connlimit(struct pf_kstate *state) 815 { 816 struct pf_overload_entry *pfoe; 817 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 818 bool limited = false; 819 820 PF_STATE_LOCK_ASSERT(state); 821 PF_SRC_NODE_LOCK(src_node); 822 823 src_node->conn++; 824 state->src.tcp_est = 1; 825 826 if (state->rule->max_src_conn && 827 state->rule->max_src_conn < 828 src_node->conn) { 829 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 830 limited = true; 831 } 832 833 if (state->rule->max_src_conn_rate.limit && 834 pf_check_threshold(&src_node->conn_rate)) { 835 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 836 limited = true; 837 } 838 839 if (!limited) 840 goto done; 841 842 /* Kill this state. */ 843 state->timeout = PFTM_PURGE; 844 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 845 846 if (state->rule->overload_tbl == NULL) 847 goto done; 848 849 /* Schedule overloading and flushing task. */ 850 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 851 if (pfoe == NULL) 852 goto done; /* too bad :( */ 853 854 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 855 pfoe->af = state->key[PF_SK_WIRE]->af; 856 pfoe->rule = state->rule; 857 pfoe->dir = state->direction; 858 PF_OVERLOADQ_LOCK(); 859 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 860 PF_OVERLOADQ_UNLOCK(); 861 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 862 863 done: 864 PF_SRC_NODE_UNLOCK(src_node); 865 return (limited); 866 } 867 868 static void 869 pf_overload_task(void *v, int pending) 870 { 871 struct pf_overload_head queue; 872 struct pfr_addr p; 873 struct pf_overload_entry *pfoe, *pfoe1; 874 uint32_t killed = 0; 875 876 CURVNET_SET((struct vnet *)v); 877 878 PF_OVERLOADQ_LOCK(); 879 queue = V_pf_overloadqueue; 880 SLIST_INIT(&V_pf_overloadqueue); 881 PF_OVERLOADQ_UNLOCK(); 882 883 bzero(&p, sizeof(p)); 884 SLIST_FOREACH(pfoe, &queue, next) { 885 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 886 if (V_pf_status.debug >= PF_DEBUG_MISC) { 887 printf("%s: blocking address ", __func__); 888 pf_print_host(&pfoe->addr, 0, pfoe->af); 889 printf("\n"); 890 } 891 892 p.pfra_af = pfoe->af; 893 switch (pfoe->af) { 894 #ifdef INET 895 case AF_INET: 896 p.pfra_net = 32; 897 p.pfra_ip4addr = pfoe->addr.v4; 898 break; 899 #endif /* INET */ 900 #ifdef INET6 901 case AF_INET6: 902 p.pfra_net = 128; 903 p.pfra_ip6addr = pfoe->addr.v6; 904 break; 905 #endif /* INET6 */ 906 default: 907 unhandled_af(pfoe->af); 908 } 909 910 PF_RULES_WLOCK(); 911 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 912 PF_RULES_WUNLOCK(); 913 } 914 915 /* 916 * Remove those entries, that don't need flushing. 917 */ 918 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 919 if (pfoe->rule->flush == 0) { 920 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 921 free(pfoe, M_PFTEMP); 922 } else 923 counter_u64_add( 924 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 925 926 /* If nothing to flush, return. */ 927 if (SLIST_EMPTY(&queue)) { 928 CURVNET_RESTORE(); 929 return; 930 } 931 932 for (int i = 0; i <= V_pf_hashmask; i++) { 933 struct pf_idhash *ih = &V_pf_idhash[i]; 934 struct pf_state_key *sk; 935 struct pf_kstate *s; 936 937 PF_HASHROW_LOCK(ih); 938 LIST_FOREACH(s, &ih->states, entry) { 939 sk = s->key[PF_SK_WIRE]; 940 SLIST_FOREACH(pfoe, &queue, next) 941 if (sk->af == pfoe->af && 942 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 943 pfoe->rule == s->rule) && 944 ((pfoe->dir == PF_OUT && 945 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 946 (pfoe->dir == PF_IN && 947 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 948 s->timeout = PFTM_PURGE; 949 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 950 killed++; 951 } 952 } 953 PF_HASHROW_UNLOCK(ih); 954 } 955 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 956 free(pfoe, M_PFTEMP); 957 if (V_pf_status.debug >= PF_DEBUG_MISC) 958 printf("%s: %u states killed", __func__, killed); 959 960 CURVNET_RESTORE(); 961 } 962 963 /* 964 * On node found always returns locked. On not found its configurable. 965 */ 966 struct pf_ksrc_node * 967 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 968 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 969 { 970 struct pf_ksrc_node *n; 971 972 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 973 974 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 975 PF_HASHROW_LOCK(*sh); 976 LIST_FOREACH(n, &(*sh)->nodes, entry) 977 if (n->rule == rule && n->af == af && n->type == sn_type && 978 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 979 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 980 break; 981 982 if (n == NULL && !returnlocked) 983 PF_HASHROW_UNLOCK(*sh); 984 985 return (n); 986 } 987 988 bool 989 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 990 { 991 struct pf_ksrc_node *cur; 992 993 if ((*sn) == NULL) 994 return (false); 995 996 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 997 998 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 999 PF_HASHROW_LOCK(sh); 1000 LIST_FOREACH(cur, &(sh->nodes), entry) { 1001 if (cur == (*sn) && 1002 cur->expire != 1) /* Ignore nodes being killed */ 1003 return (true); 1004 } 1005 PF_HASHROW_UNLOCK(sh); 1006 (*sn) = NULL; 1007 return (false); 1008 } 1009 1010 void 1011 pf_free_src_node(struct pf_ksrc_node *sn) 1012 { 1013 1014 for (int i = 0; i < 2; i++) { 1015 counter_u64_free(sn->bytes[i]); 1016 counter_u64_free(sn->packets[i]); 1017 } 1018 counter_rate_free(sn->conn_rate.cr); 1019 uma_zfree(V_pf_sources_z, sn); 1020 } 1021 1022 static u_short 1023 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1024 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1025 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1026 struct pfi_kkif *rkif, sa_family_t raf, pf_sn_types_t sn_type) 1027 { 1028 u_short reason = 0; 1029 struct pf_krule *r_track = rule; 1030 struct pf_ksrc_node **sn = &(sns[sn_type]); 1031 struct pf_srchash **sh = &(snhs[sn_type]); 1032 1033 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1034 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1035 1036 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1037 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1038 1039 /* 1040 * XXX: There could be a KASSERT for 1041 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1042 * but we'd need to pass pool *only* for this KASSERT. 1043 */ 1044 1045 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1046 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1047 r_track = &V_pf_default_rule; 1048 1049 /* 1050 * Request the sh to always be locked, as we might insert a new sn. 1051 */ 1052 if (*sn == NULL) 1053 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1054 1055 if (*sn == NULL) { 1056 PF_HASHROW_ASSERT(*sh); 1057 1058 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1059 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1060 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1061 reason = PFRES_SRCLIMIT; 1062 goto done; 1063 } 1064 1065 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1066 if ((*sn) == NULL) { 1067 reason = PFRES_MEMORY; 1068 goto done; 1069 } 1070 1071 for (int i = 0; i < 2; i++) { 1072 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1073 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1074 1075 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1076 pf_free_src_node(*sn); 1077 reason = PFRES_MEMORY; 1078 goto done; 1079 } 1080 } 1081 1082 if (sn_type == PF_SN_LIMIT) 1083 if (! pf_init_threshold(&(*sn)->conn_rate, 1084 rule->max_src_conn_rate.limit, 1085 rule->max_src_conn_rate.seconds)) { 1086 pf_free_src_node(*sn); 1087 reason = PFRES_MEMORY; 1088 goto done; 1089 } 1090 1091 MPASS((*sn)->lock == NULL); 1092 (*sn)->lock = &(*sh)->lock; 1093 1094 (*sn)->af = af; 1095 (*sn)->rule = r_track; 1096 pf_addrcpy(&(*sn)->addr, src, af); 1097 if (raddr != NULL) 1098 pf_addrcpy(&(*sn)->raddr, raddr, raf); 1099 (*sn)->rkif = rkif; 1100 (*sn)->raf = raf; 1101 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1102 (*sn)->creation = time_uptime; 1103 (*sn)->ruletype = rule->action; 1104 (*sn)->type = sn_type; 1105 counter_u64_add(r_track->src_nodes[sn_type], 1); 1106 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1107 } else { 1108 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1109 (*sn)->states >= rule->max_src_states) { 1110 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1111 1); 1112 reason = PFRES_SRCLIMIT; 1113 goto done; 1114 } 1115 } 1116 done: 1117 if (reason == 0) 1118 (*sn)->states++; 1119 else 1120 (*sn) = NULL; 1121 1122 PF_HASHROW_UNLOCK(*sh); 1123 return (reason); 1124 } 1125 1126 void 1127 pf_unlink_src_node(struct pf_ksrc_node *src) 1128 { 1129 PF_SRC_NODE_LOCK_ASSERT(src); 1130 1131 LIST_REMOVE(src, entry); 1132 if (src->rule) 1133 counter_u64_add(src->rule->src_nodes[src->type], -1); 1134 } 1135 1136 u_int 1137 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1138 { 1139 struct pf_ksrc_node *sn, *tmp; 1140 u_int count = 0; 1141 1142 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1143 pf_free_src_node(sn); 1144 count++; 1145 } 1146 1147 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1148 1149 return (count); 1150 } 1151 1152 void 1153 pf_mtag_initialize(void) 1154 { 1155 1156 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1157 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1158 UMA_ALIGN_PTR, 0); 1159 } 1160 1161 /* Per-vnet data storage structures initialization. */ 1162 void 1163 pf_initialize(void) 1164 { 1165 struct pf_keyhash *kh; 1166 struct pf_idhash *ih; 1167 struct pf_srchash *sh; 1168 struct pf_udpendpointhash *uh; 1169 u_int i; 1170 1171 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1172 V_pf_hashsize = PF_HASHSIZ; 1173 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1174 V_pf_srchashsize = PF_SRCHASHSIZ; 1175 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1176 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1177 1178 V_pf_hashseed = arc4random(); 1179 1180 /* States and state keys storage. */ 1181 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1182 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1183 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1184 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1185 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1186 1187 V_pf_state_key_z = uma_zcreate("pf state keys", 1188 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1189 UMA_ALIGN_PTR, 0); 1190 1191 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1192 M_PFHASH, M_NOWAIT | M_ZERO); 1193 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1194 M_PFHASH, M_NOWAIT | M_ZERO); 1195 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1196 printf("pf: Unable to allocate memory for " 1197 "state_hashsize %lu.\n", V_pf_hashsize); 1198 1199 free(V_pf_keyhash, M_PFHASH); 1200 free(V_pf_idhash, M_PFHASH); 1201 1202 V_pf_hashsize = PF_HASHSIZ; 1203 V_pf_keyhash = mallocarray(V_pf_hashsize, 1204 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1205 V_pf_idhash = mallocarray(V_pf_hashsize, 1206 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1207 } 1208 1209 V_pf_hashmask = V_pf_hashsize - 1; 1210 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1211 i++, kh++, ih++) { 1212 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1213 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1214 } 1215 1216 /* Source nodes. */ 1217 V_pf_sources_z = uma_zcreate("pf source nodes", 1218 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1219 0); 1220 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1221 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1222 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1223 1224 V_pf_srchash = mallocarray(V_pf_srchashsize, 1225 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1226 if (V_pf_srchash == NULL) { 1227 printf("pf: Unable to allocate memory for " 1228 "source_hashsize %lu.\n", V_pf_srchashsize); 1229 1230 V_pf_srchashsize = PF_SRCHASHSIZ; 1231 V_pf_srchash = mallocarray(V_pf_srchashsize, 1232 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1233 } 1234 1235 V_pf_srchashmask = V_pf_srchashsize - 1; 1236 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1237 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1238 1239 1240 /* UDP endpoint mappings. */ 1241 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1242 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1243 UMA_ALIGN_PTR, 0); 1244 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1245 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1246 if (V_pf_udpendpointhash == NULL) { 1247 printf("pf: Unable to allocate memory for " 1248 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1249 1250 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1251 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1252 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1253 } 1254 1255 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1256 for (i = 0, uh = V_pf_udpendpointhash; 1257 i <= V_pf_udpendpointhashmask; 1258 i++, uh++) { 1259 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1260 MTX_DEF | MTX_DUPOK); 1261 } 1262 1263 /* Anchors */ 1264 V_pf_anchor_z = uma_zcreate("pf anchors", 1265 sizeof(struct pf_kanchor), NULL, NULL, NULL, NULL, 1266 UMA_ALIGN_PTR, 0); 1267 V_pf_limits[PF_LIMIT_ANCHORS].zone = V_pf_anchor_z; 1268 uma_zone_set_max(V_pf_anchor_z, PF_ANCHOR_HIWAT); 1269 uma_zone_set_warning(V_pf_anchor_z, "PF anchor limit reached"); 1270 1271 V_pf_eth_anchor_z = uma_zcreate("pf Ethernet anchors", 1272 sizeof(struct pf_keth_anchor), NULL, NULL, NULL, NULL, 1273 UMA_ALIGN_PTR, 0); 1274 V_pf_limits[PF_LIMIT_ETH_ANCHORS].zone = V_pf_eth_anchor_z; 1275 uma_zone_set_max(V_pf_eth_anchor_z, PF_ANCHOR_HIWAT); 1276 uma_zone_set_warning(V_pf_eth_anchor_z, "PF Ethernet anchor limit reached"); 1277 1278 /* ALTQ */ 1279 TAILQ_INIT(&V_pf_altqs[0]); 1280 TAILQ_INIT(&V_pf_altqs[1]); 1281 TAILQ_INIT(&V_pf_altqs[2]); 1282 TAILQ_INIT(&V_pf_altqs[3]); 1283 TAILQ_INIT(&V_pf_pabuf[0]); 1284 TAILQ_INIT(&V_pf_pabuf[1]); 1285 TAILQ_INIT(&V_pf_pabuf[2]); 1286 V_pf_altqs_active = &V_pf_altqs[0]; 1287 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1288 V_pf_altqs_inactive = &V_pf_altqs[2]; 1289 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1290 1291 /* Send & overload+flush queues. */ 1292 STAILQ_INIT(&V_pf_sendqueue); 1293 SLIST_INIT(&V_pf_overloadqueue); 1294 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1295 1296 /* Unlinked, but may be referenced rules. */ 1297 TAILQ_INIT(&V_pf_unlinked_rules); 1298 } 1299 1300 void 1301 pf_mtag_cleanup(void) 1302 { 1303 1304 uma_zdestroy(pf_mtag_z); 1305 } 1306 1307 void 1308 pf_cleanup(void) 1309 { 1310 struct pf_keyhash *kh; 1311 struct pf_idhash *ih; 1312 struct pf_srchash *sh; 1313 struct pf_udpendpointhash *uh; 1314 struct pf_send_entry *pfse, *next; 1315 u_int i; 1316 1317 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1318 i <= V_pf_hashmask; 1319 i++, kh++, ih++) { 1320 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1321 __func__)); 1322 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1323 __func__)); 1324 mtx_destroy(&kh->lock); 1325 mtx_destroy(&ih->lock); 1326 } 1327 free(V_pf_keyhash, M_PFHASH); 1328 free(V_pf_idhash, M_PFHASH); 1329 1330 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1331 KASSERT(LIST_EMPTY(&sh->nodes), 1332 ("%s: source node hash not empty", __func__)); 1333 mtx_destroy(&sh->lock); 1334 } 1335 free(V_pf_srchash, M_PFHASH); 1336 1337 for (i = 0, uh = V_pf_udpendpointhash; 1338 i <= V_pf_udpendpointhashmask; 1339 i++, uh++) { 1340 KASSERT(LIST_EMPTY(&uh->endpoints), 1341 ("%s: udp endpoint hash not empty", __func__)); 1342 mtx_destroy(&uh->lock); 1343 } 1344 free(V_pf_udpendpointhash, M_PFHASH); 1345 1346 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1347 m_freem(pfse->pfse_m); 1348 free(pfse, M_PFTEMP); 1349 } 1350 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1351 1352 uma_zdestroy(V_pf_sources_z); 1353 uma_zdestroy(V_pf_state_z); 1354 uma_zdestroy(V_pf_state_key_z); 1355 uma_zdestroy(V_pf_udp_mapping_z); 1356 uma_zdestroy(V_pf_anchor_z); 1357 uma_zdestroy(V_pf_eth_anchor_z); 1358 } 1359 1360 static int 1361 pf_mtag_uminit(void *mem, int size, int how) 1362 { 1363 struct m_tag *t; 1364 1365 t = (struct m_tag *)mem; 1366 t->m_tag_cookie = MTAG_ABI_COMPAT; 1367 t->m_tag_id = PACKET_TAG_PF; 1368 t->m_tag_len = sizeof(struct pf_mtag); 1369 t->m_tag_free = pf_mtag_free; 1370 1371 return (0); 1372 } 1373 1374 static void 1375 pf_mtag_free(struct m_tag *t) 1376 { 1377 1378 uma_zfree(pf_mtag_z, t); 1379 } 1380 1381 struct pf_mtag * 1382 pf_get_mtag(struct mbuf *m) 1383 { 1384 struct m_tag *mtag; 1385 1386 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1387 return ((struct pf_mtag *)(mtag + 1)); 1388 1389 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1390 if (mtag == NULL) 1391 return (NULL); 1392 bzero(mtag + 1, sizeof(struct pf_mtag)); 1393 m_tag_prepend(m, mtag); 1394 1395 return ((struct pf_mtag *)(mtag + 1)); 1396 } 1397 1398 static int 1399 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1400 struct pf_kstate *s) 1401 { 1402 struct pf_keyhash *khs, *khw, *kh; 1403 struct pf_state_key *sk, *cur; 1404 struct pf_kstate *si, *olds = NULL; 1405 int idx; 1406 1407 NET_EPOCH_ASSERT(); 1408 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1409 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1410 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1411 1412 /* 1413 * We need to lock hash slots of both keys. To avoid deadlock 1414 * we always lock the slot with lower address first. Unlock order 1415 * isn't important. 1416 * 1417 * We also need to lock ID hash slot before dropping key 1418 * locks. On success we return with ID hash slot locked. 1419 */ 1420 1421 if (skw == sks) { 1422 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1423 PF_HASHROW_LOCK(khs); 1424 } else { 1425 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1426 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1427 if (khs == khw) { 1428 PF_HASHROW_LOCK(khs); 1429 } else if (khs < khw) { 1430 PF_HASHROW_LOCK(khs); 1431 PF_HASHROW_LOCK(khw); 1432 } else { 1433 PF_HASHROW_LOCK(khw); 1434 PF_HASHROW_LOCK(khs); 1435 } 1436 } 1437 1438 #define KEYS_UNLOCK() do { \ 1439 if (khs != khw) { \ 1440 PF_HASHROW_UNLOCK(khs); \ 1441 PF_HASHROW_UNLOCK(khw); \ 1442 } else \ 1443 PF_HASHROW_UNLOCK(khs); \ 1444 } while (0) 1445 1446 /* 1447 * First run: start with wire key. 1448 */ 1449 sk = skw; 1450 kh = khw; 1451 idx = PF_SK_WIRE; 1452 1453 MPASS(s->lock == NULL); 1454 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1455 1456 keyattach: 1457 LIST_FOREACH(cur, &kh->keys, entry) 1458 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1459 break; 1460 1461 if (cur != NULL) { 1462 /* Key exists. Check for same kif, if none, add to key. */ 1463 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1464 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1465 1466 PF_HASHROW_LOCK(ih); 1467 if (si->kif == s->kif && 1468 ((si->key[PF_SK_WIRE]->af == sk->af && 1469 si->direction == s->direction) || 1470 (si->key[PF_SK_WIRE]->af != 1471 si->key[PF_SK_STACK]->af && 1472 sk->af == si->key[PF_SK_STACK]->af && 1473 si->direction != s->direction))) { 1474 bool reuse = false; 1475 1476 if (sk->proto == IPPROTO_TCP && 1477 si->src.state >= TCPS_FIN_WAIT_2 && 1478 si->dst.state >= TCPS_FIN_WAIT_2) 1479 reuse = true; 1480 1481 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1482 printf("pf: %s key attach " 1483 "%s on %s: ", 1484 (idx == PF_SK_WIRE) ? 1485 "wire" : "stack", 1486 reuse ? "reuse" : "failed", 1487 s->kif->pfik_name); 1488 pf_print_state_parts(s, 1489 (idx == PF_SK_WIRE) ? 1490 sk : NULL, 1491 (idx == PF_SK_STACK) ? 1492 sk : NULL); 1493 printf(", existing: "); 1494 pf_print_state_parts(si, 1495 (idx == PF_SK_WIRE) ? 1496 sk : NULL, 1497 (idx == PF_SK_STACK) ? 1498 sk : NULL); 1499 printf("\n"); 1500 } 1501 1502 if (reuse) { 1503 /* 1504 * New state matches an old >FIN_WAIT_2 1505 * state. We can't drop key hash locks, 1506 * thus we can't unlink it properly. 1507 * 1508 * As a workaround we drop it into 1509 * TCPS_CLOSED state, schedule purge 1510 * ASAP and push it into the very end 1511 * of the slot TAILQ, so that it won't 1512 * conflict with our new state. 1513 */ 1514 pf_set_protostate(si, PF_PEER_BOTH, 1515 TCPS_CLOSED); 1516 si->timeout = PFTM_PURGE; 1517 olds = si; 1518 } else { 1519 s->timeout = PFTM_UNLINKED; 1520 if (idx == PF_SK_STACK) 1521 /* 1522 * Remove the wire key from 1523 * the hash. Other threads 1524 * can't be referencing it 1525 * because we still hold the 1526 * hash lock. 1527 */ 1528 pf_state_key_detach(s, 1529 PF_SK_WIRE); 1530 PF_HASHROW_UNLOCK(ih); 1531 KEYS_UNLOCK(); 1532 if (idx == PF_SK_WIRE) 1533 /* 1534 * We've not inserted either key. 1535 * Free both. 1536 */ 1537 uma_zfree(V_pf_state_key_z, skw); 1538 if (skw != sks) 1539 uma_zfree( 1540 V_pf_state_key_z, 1541 sks); 1542 return (EEXIST); /* collision! */ 1543 } 1544 } 1545 PF_HASHROW_UNLOCK(ih); 1546 } 1547 uma_zfree(V_pf_state_key_z, sk); 1548 s->key[idx] = cur; 1549 } else { 1550 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1551 s->key[idx] = sk; 1552 } 1553 1554 stateattach: 1555 /* List is sorted, if-bound states before floating. */ 1556 if (s->kif == V_pfi_all) 1557 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1558 else 1559 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1560 1561 if (olds) { 1562 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1563 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1564 key_list[idx]); 1565 olds = NULL; 1566 } 1567 1568 /* 1569 * Attach done. See how should we (or should not?) 1570 * attach a second key. 1571 */ 1572 if (sks == skw) { 1573 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1574 idx = PF_SK_STACK; 1575 sks = NULL; 1576 goto stateattach; 1577 } else if (sks != NULL) { 1578 /* 1579 * Continue attaching with stack key. 1580 */ 1581 sk = sks; 1582 kh = khs; 1583 idx = PF_SK_STACK; 1584 sks = NULL; 1585 goto keyattach; 1586 } 1587 1588 PF_STATE_LOCK(s); 1589 KEYS_UNLOCK(); 1590 1591 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1592 ("%s failure", __func__)); 1593 1594 return (0); 1595 #undef KEYS_UNLOCK 1596 } 1597 1598 static void 1599 pf_detach_state(struct pf_kstate *s) 1600 { 1601 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1602 struct pf_keyhash *kh; 1603 1604 NET_EPOCH_ASSERT(); 1605 MPASS(s->timeout >= PFTM_MAX); 1606 1607 pf_sctp_multihome_detach_addr(s); 1608 1609 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1610 V_pflow_export_state_ptr(s); 1611 1612 if (sks != NULL) { 1613 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1614 PF_HASHROW_LOCK(kh); 1615 if (s->key[PF_SK_STACK] != NULL) 1616 pf_state_key_detach(s, PF_SK_STACK); 1617 /* 1618 * If both point to same key, then we are done. 1619 */ 1620 if (sks == s->key[PF_SK_WIRE]) { 1621 pf_state_key_detach(s, PF_SK_WIRE); 1622 PF_HASHROW_UNLOCK(kh); 1623 return; 1624 } 1625 PF_HASHROW_UNLOCK(kh); 1626 } 1627 1628 if (s->key[PF_SK_WIRE] != NULL) { 1629 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1630 PF_HASHROW_LOCK(kh); 1631 if (s->key[PF_SK_WIRE] != NULL) 1632 pf_state_key_detach(s, PF_SK_WIRE); 1633 PF_HASHROW_UNLOCK(kh); 1634 } 1635 } 1636 1637 static void 1638 pf_state_key_detach(struct pf_kstate *s, int idx) 1639 { 1640 struct pf_state_key *sk = s->key[idx]; 1641 #ifdef INVARIANTS 1642 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1643 1644 PF_HASHROW_ASSERT(kh); 1645 #endif /* INVARIANTS */ 1646 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1647 s->key[idx] = NULL; 1648 1649 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1650 LIST_REMOVE(sk, entry); 1651 uma_zfree(V_pf_state_key_z, sk); 1652 } 1653 } 1654 1655 static int 1656 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1657 { 1658 struct pf_state_key *sk = mem; 1659 1660 bzero(sk, sizeof(struct pf_state_key_cmp)); 1661 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1662 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1663 1664 return (0); 1665 } 1666 1667 static int 1668 pf_state_key_addr_setup(struct pf_pdesc *pd, 1669 struct pf_state_key_cmp *key, int multi) 1670 { 1671 struct pf_addr *saddr = pd->src; 1672 struct pf_addr *daddr = pd->dst; 1673 #ifdef INET6 1674 struct nd_neighbor_solicit nd; 1675 struct pf_addr *target; 1676 1677 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1678 goto copy; 1679 1680 switch (pd->hdr.icmp6.icmp6_type) { 1681 case ND_NEIGHBOR_SOLICIT: 1682 if (multi) 1683 return (-1); 1684 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL, 1685 pd->af)) 1686 return (-1); 1687 target = (struct pf_addr *)&nd.nd_ns_target; 1688 daddr = target; 1689 break; 1690 case ND_NEIGHBOR_ADVERT: 1691 if (multi) 1692 return (-1); 1693 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL, 1694 pd->af)) 1695 return (-1); 1696 target = (struct pf_addr *)&nd.nd_ns_target; 1697 saddr = target; 1698 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1699 key->addr[pd->didx].addr32[0] = 0; 1700 key->addr[pd->didx].addr32[1] = 0; 1701 key->addr[pd->didx].addr32[2] = 0; 1702 key->addr[pd->didx].addr32[3] = 0; 1703 daddr = NULL; /* overwritten */ 1704 } 1705 break; 1706 default: 1707 if (multi) { 1708 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1709 key->addr[pd->sidx].addr32[1] = 0; 1710 key->addr[pd->sidx].addr32[2] = 0; 1711 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1712 saddr = NULL; /* overwritten */ 1713 } 1714 } 1715 copy: 1716 #endif /* INET6 */ 1717 if (saddr) 1718 pf_addrcpy(&key->addr[pd->sidx], saddr, pd->af); 1719 if (daddr) 1720 pf_addrcpy(&key->addr[pd->didx], daddr, pd->af); 1721 1722 return (0); 1723 } 1724 1725 int 1726 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1727 struct pf_state_key **sk, struct pf_state_key **nk) 1728 { 1729 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1730 if (*sk == NULL) 1731 return (ENOMEM); 1732 1733 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1734 0)) { 1735 uma_zfree(V_pf_state_key_z, *sk); 1736 *sk = NULL; 1737 return (ENOMEM); 1738 } 1739 1740 (*sk)->port[pd->sidx] = sport; 1741 (*sk)->port[pd->didx] = dport; 1742 (*sk)->proto = pd->proto; 1743 (*sk)->af = pd->af; 1744 1745 *nk = pf_state_key_clone(*sk); 1746 if (*nk == NULL) { 1747 uma_zfree(V_pf_state_key_z, *sk); 1748 *sk = NULL; 1749 return (ENOMEM); 1750 } 1751 1752 if (pd->af != pd->naf) { 1753 (*sk)->port[pd->sidx] = pd->osport; 1754 (*sk)->port[pd->didx] = pd->odport; 1755 1756 (*nk)->af = pd->naf; 1757 1758 /* 1759 * We're overwriting an address here, so potentially there's bits of an IPv6 1760 * address left in here. Clear that out first. 1761 */ 1762 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1763 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1764 if (pd->dir == PF_IN) { 1765 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->nsaddr, 1766 pd->naf); 1767 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->ndaddr, 1768 pd->naf); 1769 (*nk)->port[pd->didx] = pd->nsport; 1770 (*nk)->port[pd->sidx] = pd->ndport; 1771 } else { 1772 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->nsaddr, 1773 pd->naf); 1774 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->ndaddr, 1775 pd->naf); 1776 (*nk)->port[pd->sidx] = pd->nsport; 1777 (*nk)->port[pd->didx] = pd->ndport; 1778 } 1779 1780 switch (pd->proto) { 1781 case IPPROTO_ICMP: 1782 (*nk)->proto = IPPROTO_ICMPV6; 1783 break; 1784 case IPPROTO_ICMPV6: 1785 (*nk)->proto = IPPROTO_ICMP; 1786 break; 1787 default: 1788 (*nk)->proto = pd->proto; 1789 } 1790 } 1791 1792 return (0); 1793 } 1794 1795 struct pf_state_key * 1796 pf_state_key_clone(const struct pf_state_key *orig) 1797 { 1798 struct pf_state_key *sk; 1799 1800 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1801 if (sk == NULL) 1802 return (NULL); 1803 1804 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1805 1806 return (sk); 1807 } 1808 1809 int 1810 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1811 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1812 { 1813 struct pf_idhash *ih; 1814 struct pf_kstate *cur; 1815 int error; 1816 1817 NET_EPOCH_ASSERT(); 1818 1819 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1820 ("%s: sks not pristine", __func__)); 1821 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1822 ("%s: skw not pristine", __func__)); 1823 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1824 1825 s->kif = kif; 1826 s->orig_kif = orig_kif; 1827 1828 if (s->id == 0 && s->creatorid == 0) { 1829 s->id = alloc_unr64(&V_pf_stateid); 1830 s->id = htobe64(s->id); 1831 s->creatorid = V_pf_status.hostid; 1832 } 1833 1834 /* Returns with ID locked on success. */ 1835 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1836 return (error); 1837 skw = sks = NULL; 1838 1839 ih = &V_pf_idhash[PF_IDHASH(s)]; 1840 PF_HASHROW_ASSERT(ih); 1841 LIST_FOREACH(cur, &ih->states, entry) 1842 if (cur->id == s->id && cur->creatorid == s->creatorid) 1843 break; 1844 1845 if (cur != NULL) { 1846 s->timeout = PFTM_UNLINKED; 1847 PF_HASHROW_UNLOCK(ih); 1848 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1849 printf("pf: state ID collision: " 1850 "id: %016llx creatorid: %08x\n", 1851 (unsigned long long)be64toh(s->id), 1852 ntohl(s->creatorid)); 1853 } 1854 pf_detach_state(s); 1855 return (EEXIST); 1856 } 1857 LIST_INSERT_HEAD(&ih->states, s, entry); 1858 /* One for keys, one for ID hash. */ 1859 refcount_init(&s->refs, 2); 1860 1861 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1862 if (V_pfsync_insert_state_ptr != NULL) 1863 V_pfsync_insert_state_ptr(s); 1864 1865 /* Returns locked. */ 1866 return (0); 1867 } 1868 1869 /* 1870 * Find state by ID: returns with locked row on success. 1871 */ 1872 struct pf_kstate * 1873 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1874 { 1875 struct pf_idhash *ih; 1876 struct pf_kstate *s; 1877 1878 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1879 1880 ih = &V_pf_idhash[PF_IDHASHID(id)]; 1881 1882 PF_HASHROW_LOCK(ih); 1883 LIST_FOREACH(s, &ih->states, entry) 1884 if (s->id == id && s->creatorid == creatorid) 1885 break; 1886 1887 if (s == NULL) 1888 PF_HASHROW_UNLOCK(ih); 1889 1890 return (s); 1891 } 1892 1893 /* 1894 * Find state by key. 1895 * Returns with ID hash slot locked on success. 1896 */ 1897 static int 1898 pf_find_state(struct pf_pdesc *pd, const struct pf_state_key_cmp *key, 1899 struct pf_kstate **state) 1900 { 1901 struct pf_keyhash *kh; 1902 struct pf_state_key *sk; 1903 struct pf_kstate *s; 1904 int idx; 1905 1906 *state = NULL; 1907 1908 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1909 1910 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1911 1912 PF_HASHROW_LOCK(kh); 1913 LIST_FOREACH(sk, &kh->keys, entry) 1914 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1915 break; 1916 if (sk == NULL) { 1917 PF_HASHROW_UNLOCK(kh); 1918 return (PF_DROP); 1919 } 1920 1921 idx = (pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1922 1923 /* List is sorted, if-bound states before floating ones. */ 1924 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1925 if (s->kif == V_pfi_all || s->kif == pd->kif || 1926 s->orig_kif == pd->kif) { 1927 PF_STATE_LOCK(s); 1928 PF_HASHROW_UNLOCK(kh); 1929 if (__predict_false(s->timeout >= PFTM_MAX)) { 1930 /* 1931 * State is either being processed by 1932 * pf_remove_state() in an other thread, or 1933 * is scheduled for immediate expiry. 1934 */ 1935 PF_STATE_UNLOCK(s); 1936 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 1937 key, (pd->dir), pd, *state); 1938 return (PF_DROP); 1939 } 1940 goto out; 1941 } 1942 1943 /* Look through the other list, in case of AF-TO */ 1944 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1945 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1946 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1947 continue; 1948 if (s->kif == V_pfi_all || s->kif == pd->kif || 1949 s->orig_kif == pd->kif) { 1950 PF_STATE_LOCK(s); 1951 PF_HASHROW_UNLOCK(kh); 1952 if (__predict_false(s->timeout >= PFTM_MAX)) { 1953 /* 1954 * State is either being processed by 1955 * pf_remove_state() in an other thread, or 1956 * is scheduled for immediate expiry. 1957 */ 1958 PF_STATE_UNLOCK(s); 1959 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 1960 key, (pd->dir), pd, NULL); 1961 return (PF_DROP); 1962 } 1963 goto out; 1964 } 1965 } 1966 1967 PF_HASHROW_UNLOCK(kh); 1968 1969 out: 1970 SDT_PROBE5(pf, ip, state, lookup, pd->kif, key, (pd->dir), pd, *state); 1971 1972 if (s == NULL || s->timeout == PFTM_PURGE) { 1973 if (s) 1974 PF_STATE_UNLOCK(s); 1975 return (PF_DROP); 1976 } 1977 1978 if ((s)->rule->pktrate.limit && pd->dir == (s)->direction) { 1979 if (pf_check_threshold(&(s)->rule->pktrate)) { 1980 PF_STATE_UNLOCK(s); 1981 return (PF_DROP); 1982 } 1983 } 1984 if (PACKET_LOOPED(pd)) { 1985 PF_STATE_UNLOCK(s); 1986 return (PF_PASS); 1987 } 1988 1989 *state = s; 1990 1991 return (PF_MATCH); 1992 } 1993 1994 /* 1995 * Returns with ID hash slot locked on success. 1996 */ 1997 struct pf_kstate * 1998 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1999 { 2000 struct pf_keyhash *kh; 2001 struct pf_state_key *sk; 2002 struct pf_kstate *s, *ret = NULL; 2003 int idx, inout = 0; 2004 2005 if (more != NULL) 2006 *more = 0; 2007 2008 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 2009 2010 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 2011 2012 PF_HASHROW_LOCK(kh); 2013 LIST_FOREACH(sk, &kh->keys, entry) 2014 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 2015 break; 2016 if (sk == NULL) { 2017 PF_HASHROW_UNLOCK(kh); 2018 return (NULL); 2019 } 2020 switch (dir) { 2021 case PF_IN: 2022 idx = PF_SK_WIRE; 2023 break; 2024 case PF_OUT: 2025 idx = PF_SK_STACK; 2026 break; 2027 case PF_INOUT: 2028 idx = PF_SK_WIRE; 2029 inout = 1; 2030 break; 2031 default: 2032 panic("%s: dir %u", __func__, dir); 2033 } 2034 second_run: 2035 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 2036 if (more == NULL) { 2037 PF_STATE_LOCK(s); 2038 PF_HASHROW_UNLOCK(kh); 2039 return (s); 2040 } 2041 2042 if (ret) 2043 (*more)++; 2044 else { 2045 ret = s; 2046 PF_STATE_LOCK(s); 2047 } 2048 } 2049 if (inout == 1) { 2050 inout = 0; 2051 idx = PF_SK_STACK; 2052 goto second_run; 2053 } 2054 PF_HASHROW_UNLOCK(kh); 2055 2056 return (ret); 2057 } 2058 2059 /* 2060 * FIXME 2061 * This routine is inefficient -- locks the state only to unlock immediately on 2062 * return. 2063 * It is racy -- after the state is unlocked nothing stops other threads from 2064 * removing it. 2065 */ 2066 bool 2067 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2068 { 2069 struct pf_kstate *s; 2070 2071 s = pf_find_state_all(key, dir, NULL); 2072 if (s != NULL) { 2073 PF_STATE_UNLOCK(s); 2074 return (true); 2075 } 2076 return (false); 2077 } 2078 2079 void 2080 pf_state_peer_hton(const struct pf_state_peer *s, struct pf_state_peer_export *d) 2081 { 2082 d->seqlo = htonl(s->seqlo); 2083 d->seqhi = htonl(s->seqhi); 2084 d->seqdiff = htonl(s->seqdiff); 2085 d->max_win = htons(s->max_win); 2086 d->mss = htons(s->mss); 2087 d->state = s->state; 2088 d->wscale = s->wscale; 2089 if (s->scrub) { 2090 d->scrub.pfss_flags = htons( 2091 s->scrub->pfss_flags & PFSS_TIMESTAMP); 2092 d->scrub.pfss_ttl = (s)->scrub->pfss_ttl; 2093 d->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod); 2094 d->scrub.scrub_flag = PF_SCRUB_FLAG_VALID; 2095 } 2096 } 2097 2098 void 2099 pf_state_peer_ntoh(const struct pf_state_peer_export *s, struct pf_state_peer *d) 2100 { 2101 d->seqlo = ntohl(s->seqlo); 2102 d->seqhi = ntohl(s->seqhi); 2103 d->seqdiff = ntohl(s->seqdiff); 2104 d->max_win = ntohs(s->max_win); 2105 d->mss = ntohs(s->mss); 2106 d->state = s->state; 2107 d->wscale = s->wscale; 2108 if (s->scrub.scrub_flag == PF_SCRUB_FLAG_VALID && 2109 d->scrub != NULL) { 2110 d->scrub->pfss_flags = ntohs(s->scrub.pfss_flags) & 2111 PFSS_TIMESTAMP; 2112 d->scrub->pfss_ttl = s->scrub.pfss_ttl; 2113 d->scrub->pfss_ts_mod = ntohl(s->scrub.pfss_ts_mod); 2114 } 2115 } 2116 2117 struct pf_udp_mapping * 2118 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2119 struct pf_addr *nat_addr, uint16_t nat_port) 2120 { 2121 struct pf_udp_mapping *mapping; 2122 2123 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2124 if (mapping == NULL) 2125 return (NULL); 2126 pf_addrcpy(&mapping->endpoints[0].addr, src_addr, af); 2127 mapping->endpoints[0].port = src_port; 2128 mapping->endpoints[0].af = af; 2129 mapping->endpoints[0].mapping = mapping; 2130 pf_addrcpy(&mapping->endpoints[1].addr, nat_addr, af); 2131 mapping->endpoints[1].port = nat_port; 2132 mapping->endpoints[1].af = af; 2133 mapping->endpoints[1].mapping = mapping; 2134 refcount_init(&mapping->refs, 1); 2135 return (mapping); 2136 } 2137 2138 int 2139 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2140 { 2141 struct pf_udpendpointhash *h0, *h1; 2142 struct pf_udp_endpoint *endpoint; 2143 int ret = EEXIST; 2144 2145 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2146 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2147 if (h0 == h1) { 2148 PF_HASHROW_LOCK(h0); 2149 } else if (h0 < h1) { 2150 PF_HASHROW_LOCK(h0); 2151 PF_HASHROW_LOCK(h1); 2152 } else { 2153 PF_HASHROW_LOCK(h1); 2154 PF_HASHROW_LOCK(h0); 2155 } 2156 2157 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2158 if (bcmp(endpoint, &mapping->endpoints[0], 2159 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2160 break; 2161 } 2162 if (endpoint != NULL) 2163 goto cleanup; 2164 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2165 if (bcmp(endpoint, &mapping->endpoints[1], 2166 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2167 break; 2168 } 2169 if (endpoint != NULL) 2170 goto cleanup; 2171 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2172 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2173 ret = 0; 2174 2175 cleanup: 2176 if (h0 != h1) { 2177 PF_HASHROW_UNLOCK(h0); 2178 PF_HASHROW_UNLOCK(h1); 2179 } else { 2180 PF_HASHROW_UNLOCK(h0); 2181 } 2182 return (ret); 2183 } 2184 2185 void 2186 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2187 { 2188 /* refcount is synchronized on the source endpoint's row lock */ 2189 struct pf_udpendpointhash *h0, *h1; 2190 2191 if (mapping == NULL) 2192 return; 2193 2194 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2195 PF_HASHROW_LOCK(h0); 2196 if (refcount_release(&mapping->refs)) { 2197 LIST_REMOVE(&mapping->endpoints[0], entry); 2198 PF_HASHROW_UNLOCK(h0); 2199 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2200 PF_HASHROW_LOCK(h1); 2201 LIST_REMOVE(&mapping->endpoints[1], entry); 2202 PF_HASHROW_UNLOCK(h1); 2203 2204 uma_zfree(V_pf_udp_mapping_z, mapping); 2205 } else { 2206 PF_HASHROW_UNLOCK(h0); 2207 } 2208 } 2209 2210 2211 struct pf_udp_mapping * 2212 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2213 { 2214 struct pf_udpendpointhash *uh; 2215 struct pf_udp_endpoint *endpoint; 2216 2217 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2218 2219 PF_HASHROW_LOCK(uh); 2220 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2221 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2222 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2223 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2224 break; 2225 } 2226 if (endpoint == NULL) { 2227 PF_HASHROW_UNLOCK(uh); 2228 return (NULL); 2229 } 2230 refcount_acquire(&endpoint->mapping->refs); 2231 PF_HASHROW_UNLOCK(uh); 2232 return (endpoint->mapping); 2233 } 2234 /* END state table stuff */ 2235 2236 static void 2237 pf_send(struct pf_send_entry *pfse) 2238 { 2239 2240 PF_SENDQ_LOCK(); 2241 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2242 PF_SENDQ_UNLOCK(); 2243 swi_sched(V_pf_swi_cookie, 0); 2244 } 2245 2246 static bool 2247 pf_isforlocal(struct mbuf *m, int af) 2248 { 2249 switch (af) { 2250 #ifdef INET 2251 case AF_INET: { 2252 struct ip *ip = mtod(m, struct ip *); 2253 2254 return (in_localip(ip->ip_dst)); 2255 } 2256 #endif /* INET */ 2257 #ifdef INET6 2258 case AF_INET6: { 2259 struct ip6_hdr *ip6; 2260 struct in6_ifaddr *ia; 2261 ip6 = mtod(m, struct ip6_hdr *); 2262 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2263 if (ia == NULL) 2264 return (false); 2265 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2266 } 2267 #endif /* INET6 */ 2268 default: 2269 unhandled_af(af); 2270 } 2271 2272 return (false); 2273 } 2274 2275 int 2276 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2277 int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type) 2278 { 2279 /* 2280 * ICMP types marked with PF_OUT are typically responses to 2281 * PF_IN, and will match states in the opposite direction. 2282 * PF_IN ICMP types need to match a state with that type. 2283 */ 2284 *icmp_dir = PF_OUT; 2285 2286 /* Queries (and responses) */ 2287 switch (pd->af) { 2288 #ifdef INET 2289 case AF_INET: 2290 switch (type) { 2291 case ICMP_ECHO: 2292 *icmp_dir = PF_IN; 2293 /* FALLTHROUGH */ 2294 case ICMP_ECHOREPLY: 2295 *virtual_type = ICMP_ECHO; 2296 *virtual_id = pd->hdr.icmp.icmp_id; 2297 break; 2298 2299 case ICMP_TSTAMP: 2300 *icmp_dir = PF_IN; 2301 /* FALLTHROUGH */ 2302 case ICMP_TSTAMPREPLY: 2303 *virtual_type = ICMP_TSTAMP; 2304 *virtual_id = pd->hdr.icmp.icmp_id; 2305 break; 2306 2307 case ICMP_IREQ: 2308 *icmp_dir = PF_IN; 2309 /* FALLTHROUGH */ 2310 case ICMP_IREQREPLY: 2311 *virtual_type = ICMP_IREQ; 2312 *virtual_id = pd->hdr.icmp.icmp_id; 2313 break; 2314 2315 case ICMP_MASKREQ: 2316 *icmp_dir = PF_IN; 2317 /* FALLTHROUGH */ 2318 case ICMP_MASKREPLY: 2319 *virtual_type = ICMP_MASKREQ; 2320 *virtual_id = pd->hdr.icmp.icmp_id; 2321 break; 2322 2323 case ICMP_IPV6_WHEREAREYOU: 2324 *icmp_dir = PF_IN; 2325 /* FALLTHROUGH */ 2326 case ICMP_IPV6_IAMHERE: 2327 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2328 *virtual_id = 0; /* Nothing sane to match on! */ 2329 break; 2330 2331 case ICMP_MOBILE_REGREQUEST: 2332 *icmp_dir = PF_IN; 2333 /* FALLTHROUGH */ 2334 case ICMP_MOBILE_REGREPLY: 2335 *virtual_type = ICMP_MOBILE_REGREQUEST; 2336 *virtual_id = 0; /* Nothing sane to match on! */ 2337 break; 2338 2339 case ICMP_ROUTERSOLICIT: 2340 *icmp_dir = PF_IN; 2341 /* FALLTHROUGH */ 2342 case ICMP_ROUTERADVERT: 2343 *virtual_type = ICMP_ROUTERSOLICIT; 2344 *virtual_id = 0; /* Nothing sane to match on! */ 2345 break; 2346 2347 /* These ICMP types map to other connections */ 2348 case ICMP_UNREACH: 2349 case ICMP_SOURCEQUENCH: 2350 case ICMP_REDIRECT: 2351 case ICMP_TIMXCEED: 2352 case ICMP_PARAMPROB: 2353 /* These will not be used, but set them anyway */ 2354 *icmp_dir = PF_IN; 2355 *virtual_type = type; 2356 *virtual_id = 0; 2357 *virtual_type = htons(*virtual_type); 2358 return (1); /* These types match to another state */ 2359 2360 /* 2361 * All remaining ICMP types get their own states, 2362 * and will only match in one direction. 2363 */ 2364 default: 2365 *icmp_dir = PF_IN; 2366 *virtual_type = type; 2367 *virtual_id = 0; 2368 break; 2369 } 2370 break; 2371 #endif /* INET */ 2372 #ifdef INET6 2373 case AF_INET6: 2374 switch (type) { 2375 case ICMP6_ECHO_REQUEST: 2376 *icmp_dir = PF_IN; 2377 /* FALLTHROUGH */ 2378 case ICMP6_ECHO_REPLY: 2379 *virtual_type = ICMP6_ECHO_REQUEST; 2380 *virtual_id = pd->hdr.icmp6.icmp6_id; 2381 break; 2382 2383 case MLD_LISTENER_QUERY: 2384 case MLD_LISTENER_REPORT: { 2385 /* 2386 * Listener Report can be sent by clients 2387 * without an associated Listener Query. 2388 * In addition to that, when Report is sent as a 2389 * reply to a Query its source and destination 2390 * address are different. 2391 */ 2392 *icmp_dir = PF_IN; 2393 *virtual_type = MLD_LISTENER_QUERY; 2394 *virtual_id = 0; 2395 break; 2396 } 2397 case MLD_MTRACE: 2398 *icmp_dir = PF_IN; 2399 /* FALLTHROUGH */ 2400 case MLD_MTRACE_RESP: 2401 *virtual_type = MLD_MTRACE; 2402 *virtual_id = 0; /* Nothing sane to match on! */ 2403 break; 2404 2405 case ND_NEIGHBOR_SOLICIT: 2406 *icmp_dir = PF_IN; 2407 /* FALLTHROUGH */ 2408 case ND_NEIGHBOR_ADVERT: { 2409 *virtual_type = ND_NEIGHBOR_SOLICIT; 2410 *virtual_id = 0; 2411 break; 2412 } 2413 2414 /* 2415 * These ICMP types map to other connections. 2416 * ND_REDIRECT can't be in this list because the triggering 2417 * packet header is optional. 2418 */ 2419 case ICMP6_DST_UNREACH: 2420 case ICMP6_PACKET_TOO_BIG: 2421 case ICMP6_TIME_EXCEEDED: 2422 case ICMP6_PARAM_PROB: 2423 /* These will not be used, but set them anyway */ 2424 *icmp_dir = PF_IN; 2425 *virtual_type = type; 2426 *virtual_id = 0; 2427 *virtual_type = htons(*virtual_type); 2428 return (1); /* These types match to another state */ 2429 /* 2430 * All remaining ICMP6 types get their own states, 2431 * and will only match in one direction. 2432 */ 2433 default: 2434 *icmp_dir = PF_IN; 2435 *virtual_type = type; 2436 *virtual_id = 0; 2437 break; 2438 } 2439 break; 2440 #endif /* INET6 */ 2441 default: 2442 unhandled_af(pd->af); 2443 } 2444 *virtual_type = htons(*virtual_type); 2445 return (0); /* These types match to their own state */ 2446 } 2447 2448 void 2449 pf_intr(void *v) 2450 { 2451 struct epoch_tracker et; 2452 struct pf_send_head queue; 2453 struct pf_send_entry *pfse, *next; 2454 2455 CURVNET_SET((struct vnet *)v); 2456 2457 PF_SENDQ_LOCK(); 2458 queue = V_pf_sendqueue; 2459 STAILQ_INIT(&V_pf_sendqueue); 2460 PF_SENDQ_UNLOCK(); 2461 2462 NET_EPOCH_ENTER(et); 2463 2464 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2465 switch (pfse->pfse_type) { 2466 #ifdef INET 2467 case PFSE_IP: { 2468 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2469 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2470 ("%s: rcvif != loif", __func__)); 2471 2472 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2473 pfse->pfse_m->m_pkthdr.csum_flags |= 2474 CSUM_IP_VALID | CSUM_IP_CHECKED | 2475 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2476 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2477 ip_input(pfse->pfse_m); 2478 } else { 2479 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2480 NULL); 2481 } 2482 break; 2483 } 2484 case PFSE_ICMP: 2485 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2486 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2487 break; 2488 #endif /* INET */ 2489 #ifdef INET6 2490 case PFSE_IP6: 2491 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2492 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2493 ("%s: rcvif != loif", __func__)); 2494 2495 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2496 M_LOOP; 2497 pfse->pfse_m->m_pkthdr.csum_flags |= 2498 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2499 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2500 ip6_input(pfse->pfse_m); 2501 } else { 2502 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2503 NULL, NULL); 2504 } 2505 break; 2506 case PFSE_ICMP6: 2507 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2508 pfse->icmpopts.code, pfse->icmpopts.mtu); 2509 break; 2510 #endif /* INET6 */ 2511 default: 2512 panic("%s: unknown type", __func__); 2513 } 2514 free(pfse, M_PFTEMP); 2515 } 2516 NET_EPOCH_EXIT(et); 2517 CURVNET_RESTORE(); 2518 } 2519 2520 #define pf_purge_thread_period (hz / 10) 2521 2522 #ifdef PF_WANT_32_TO_64_COUNTER 2523 static void 2524 pf_status_counter_u64_periodic(void) 2525 { 2526 2527 PF_RULES_RASSERT(); 2528 2529 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2530 return; 2531 } 2532 2533 for (int i = 0; i < FCNT_MAX; i++) { 2534 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2535 } 2536 } 2537 2538 static void 2539 pf_kif_counter_u64_periodic(void) 2540 { 2541 struct pfi_kkif *kif; 2542 size_t r, run; 2543 2544 PF_RULES_RASSERT(); 2545 2546 if (__predict_false(V_pf_allkifcount == 0)) { 2547 return; 2548 } 2549 2550 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2551 return; 2552 } 2553 2554 run = V_pf_allkifcount / 10; 2555 if (run < 5) 2556 run = 5; 2557 2558 for (r = 0; r < run; r++) { 2559 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2560 if (kif == NULL) { 2561 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2562 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2563 break; 2564 } 2565 2566 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2567 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2568 2569 for (int i = 0; i < 2; i++) { 2570 for (int j = 0; j < 2; j++) { 2571 for (int k = 0; k < 2; k++) { 2572 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2573 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2574 } 2575 } 2576 } 2577 } 2578 } 2579 2580 static void 2581 pf_rule_counter_u64_periodic(void) 2582 { 2583 struct pf_krule *rule; 2584 size_t r, run; 2585 2586 PF_RULES_RASSERT(); 2587 2588 if (__predict_false(V_pf_allrulecount == 0)) { 2589 return; 2590 } 2591 2592 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2593 return; 2594 } 2595 2596 run = V_pf_allrulecount / 10; 2597 if (run < 5) 2598 run = 5; 2599 2600 for (r = 0; r < run; r++) { 2601 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2602 if (rule == NULL) { 2603 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2604 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2605 break; 2606 } 2607 2608 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2609 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2610 2611 pf_counter_u64_periodic(&rule->evaluations); 2612 for (int i = 0; i < 2; i++) { 2613 pf_counter_u64_periodic(&rule->packets[i]); 2614 pf_counter_u64_periodic(&rule->bytes[i]); 2615 } 2616 } 2617 } 2618 2619 static void 2620 pf_counter_u64_periodic_main(void) 2621 { 2622 PF_RULES_RLOCK_TRACKER; 2623 2624 V_pf_counter_periodic_iter++; 2625 2626 PF_RULES_RLOCK(); 2627 pf_counter_u64_critical_enter(); 2628 pf_status_counter_u64_periodic(); 2629 pf_kif_counter_u64_periodic(); 2630 pf_rule_counter_u64_periodic(); 2631 pf_counter_u64_critical_exit(); 2632 PF_RULES_RUNLOCK(); 2633 } 2634 #else 2635 #define pf_counter_u64_periodic_main() do { } while (0) 2636 #endif 2637 2638 void 2639 pf_purge_thread(void *unused __unused) 2640 { 2641 struct epoch_tracker et; 2642 2643 VNET_ITERATOR_DECL(vnet_iter); 2644 2645 sx_xlock(&pf_end_lock); 2646 while (pf_end_threads == 0) { 2647 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2648 2649 VNET_LIST_RLOCK(); 2650 NET_EPOCH_ENTER(et); 2651 VNET_FOREACH(vnet_iter) { 2652 CURVNET_SET(vnet_iter); 2653 2654 /* Wait until V_pf_default_rule is initialized. */ 2655 if (V_pf_vnet_active == 0) { 2656 CURVNET_RESTORE(); 2657 continue; 2658 } 2659 2660 pf_counter_u64_periodic_main(); 2661 2662 /* 2663 * Process 1/interval fraction of the state 2664 * table every run. 2665 */ 2666 V_pf_purge_idx = 2667 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2668 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2669 2670 /* 2671 * Purge other expired types every 2672 * PFTM_INTERVAL seconds. 2673 */ 2674 if (V_pf_purge_idx == 0) { 2675 /* 2676 * Order is important: 2677 * - states and src nodes reference rules 2678 * - states and rules reference kifs 2679 */ 2680 pf_purge_expired_fragments(); 2681 pf_purge_expired_src_nodes(); 2682 pf_purge_unlinked_rules(); 2683 pfi_kkif_purge(); 2684 } 2685 CURVNET_RESTORE(); 2686 } 2687 NET_EPOCH_EXIT(et); 2688 VNET_LIST_RUNLOCK(); 2689 } 2690 2691 pf_end_threads++; 2692 sx_xunlock(&pf_end_lock); 2693 kproc_exit(0); 2694 } 2695 2696 void 2697 pf_unload_vnet_purge(void) 2698 { 2699 2700 /* 2701 * To cleanse up all kifs and rules we need 2702 * two runs: first one clears reference flags, 2703 * then pf_purge_expired_states() doesn't 2704 * raise them, and then second run frees. 2705 */ 2706 pf_purge_unlinked_rules(); 2707 pfi_kkif_purge(); 2708 2709 /* 2710 * Now purge everything. 2711 */ 2712 pf_purge_expired_states(0, V_pf_hashmask); 2713 pf_purge_fragments(UINT_MAX); 2714 pf_purge_expired_src_nodes(); 2715 2716 /* 2717 * Now all kifs & rules should be unreferenced, 2718 * thus should be successfully freed. 2719 */ 2720 pf_purge_unlinked_rules(); 2721 pfi_kkif_purge(); 2722 } 2723 2724 u_int32_t 2725 pf_state_expires(const struct pf_kstate *state) 2726 { 2727 u_int32_t timeout; 2728 u_int32_t start; 2729 u_int32_t end; 2730 u_int32_t states; 2731 2732 /* handle all PFTM_* > PFTM_MAX here */ 2733 if (state->timeout == PFTM_PURGE) 2734 return (time_uptime); 2735 KASSERT(state->timeout != PFTM_UNLINKED, 2736 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2737 KASSERT((state->timeout < PFTM_MAX), 2738 ("pf_state_expires: timeout > PFTM_MAX")); 2739 timeout = state->rule->timeout[state->timeout]; 2740 if (!timeout) 2741 timeout = V_pf_default_rule.timeout[state->timeout]; 2742 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2743 if (start && state->rule != &V_pf_default_rule) { 2744 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2745 states = counter_u64_fetch(state->rule->states_cur); 2746 } else { 2747 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2748 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2749 states = V_pf_status.states; 2750 } 2751 if (end && states > start && start < end) { 2752 if (states < end) { 2753 timeout = (u_int64_t)timeout * (end - states) / 2754 (end - start); 2755 return ((state->expire / 1000) + timeout); 2756 } 2757 else 2758 return (time_uptime); 2759 } 2760 return ((state->expire / 1000) + timeout); 2761 } 2762 2763 void 2764 pf_purge_expired_src_nodes(void) 2765 { 2766 struct pf_ksrc_node_list freelist; 2767 struct pf_srchash *sh; 2768 struct pf_ksrc_node *cur, *next; 2769 int i; 2770 2771 LIST_INIT(&freelist); 2772 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2773 PF_HASHROW_LOCK(sh); 2774 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2775 if (cur->states == 0 && cur->expire <= time_uptime) { 2776 pf_unlink_src_node(cur); 2777 LIST_INSERT_HEAD(&freelist, cur, entry); 2778 } else if (cur->rule != NULL) 2779 cur->rule->rule_ref |= PFRULE_REFS; 2780 PF_HASHROW_UNLOCK(sh); 2781 } 2782 2783 pf_free_src_nodes(&freelist); 2784 2785 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2786 } 2787 2788 static void 2789 pf_src_tree_remove_state(struct pf_kstate *s) 2790 { 2791 uint32_t timeout; 2792 2793 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2794 s->rule->timeout[PFTM_SRC_NODE] : 2795 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2796 2797 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 2798 if (s->sns[sn_type] == NULL) 2799 continue; 2800 PF_SRC_NODE_LOCK(s->sns[sn_type]); 2801 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 2802 --(s->sns[sn_type]->conn); 2803 if (--(s->sns[sn_type]->states) == 0) 2804 s->sns[sn_type]->expire = time_uptime + timeout; 2805 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 2806 s->sns[sn_type] = NULL; 2807 } 2808 2809 } 2810 2811 /* 2812 * Unlink and potentilly free a state. Function may be 2813 * called with ID hash row locked, but always returns 2814 * unlocked, since it needs to go through key hash locking. 2815 */ 2816 int 2817 pf_remove_state(struct pf_kstate *s) 2818 { 2819 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2820 2821 NET_EPOCH_ASSERT(); 2822 PF_HASHROW_ASSERT(ih); 2823 2824 if (s->timeout == PFTM_UNLINKED) { 2825 /* 2826 * State is being processed 2827 * by pf_remove_state() in 2828 * an other thread. 2829 */ 2830 PF_HASHROW_UNLOCK(ih); 2831 return (0); /* XXXGL: undefined actually */ 2832 } 2833 2834 if (s->src.state == PF_TCPS_PROXY_DST) { 2835 /* XXX wire key the right one? */ 2836 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2837 &s->key[PF_SK_WIRE]->addr[1], 2838 &s->key[PF_SK_WIRE]->addr[0], 2839 s->key[PF_SK_WIRE]->port[1], 2840 s->key[PF_SK_WIRE]->port[0], 2841 s->src.seqhi, s->src.seqlo + 1, 2842 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2843 s->act.rtableid, NULL); 2844 } 2845 2846 LIST_REMOVE(s, entry); 2847 pf_src_tree_remove_state(s); 2848 2849 if (V_pfsync_delete_state_ptr != NULL) 2850 V_pfsync_delete_state_ptr(s); 2851 2852 STATE_DEC_COUNTERS(s); 2853 2854 s->timeout = PFTM_UNLINKED; 2855 2856 /* Ensure we remove it from the list of halfopen states, if needed. */ 2857 if (s->key[PF_SK_STACK] != NULL && 2858 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2859 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2860 2861 PF_HASHROW_UNLOCK(ih); 2862 2863 pf_detach_state(s); 2864 2865 pf_udp_mapping_release(s->udp_mapping); 2866 2867 /* pf_state_insert() initialises refs to 2 */ 2868 return (pf_release_staten(s, 2)); 2869 } 2870 2871 struct pf_kstate * 2872 pf_alloc_state(int flags) 2873 { 2874 2875 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2876 } 2877 2878 static __inline void 2879 pf_free_match_rules(struct pf_krule_slist *match_rules) { 2880 struct pf_krule_item *ri; 2881 2882 while ((ri = SLIST_FIRST(match_rules))) { 2883 SLIST_REMOVE_HEAD(match_rules, entry); 2884 free(ri, M_PF_RULE_ITEM); 2885 } 2886 } 2887 2888 void 2889 pf_free_state(struct pf_kstate *cur) 2890 { 2891 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2892 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2893 cur->timeout)); 2894 2895 pf_free_match_rules(&(cur->match_rules)); 2896 pf_normalize_tcp_cleanup(cur); 2897 uma_zfree(V_pf_state_z, cur); 2898 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2899 } 2900 2901 /* 2902 * Called only from pf_purge_thread(), thus serialized. 2903 */ 2904 static u_int 2905 pf_purge_expired_states(u_int i, int maxcheck) 2906 { 2907 struct pf_idhash *ih; 2908 struct pf_kstate *s; 2909 struct pf_krule_item *mrm; 2910 size_t count __unused; 2911 2912 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2913 2914 /* 2915 * Go through hash and unlink states that expire now. 2916 */ 2917 while (maxcheck > 0) { 2918 count = 0; 2919 ih = &V_pf_idhash[i]; 2920 2921 /* only take the lock if we expect to do work */ 2922 if (!LIST_EMPTY(&ih->states)) { 2923 relock: 2924 PF_HASHROW_LOCK(ih); 2925 LIST_FOREACH(s, &ih->states, entry) { 2926 if (pf_state_expires(s) <= time_uptime) { 2927 V_pf_status.states -= 2928 pf_remove_state(s); 2929 goto relock; 2930 } 2931 s->rule->rule_ref |= PFRULE_REFS; 2932 if (s->nat_rule != NULL) 2933 s->nat_rule->rule_ref |= PFRULE_REFS; 2934 if (s->anchor != NULL) 2935 s->anchor->rule_ref |= PFRULE_REFS; 2936 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2937 SLIST_FOREACH(mrm, &s->match_rules, entry) 2938 mrm->r->rule_ref |= PFRULE_REFS; 2939 if (s->act.rt_kif) 2940 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2941 count++; 2942 } 2943 PF_HASHROW_UNLOCK(ih); 2944 } 2945 2946 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2947 2948 /* Return when we hit end of hash. */ 2949 if (++i > V_pf_hashmask) { 2950 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2951 return (0); 2952 } 2953 2954 maxcheck--; 2955 } 2956 2957 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2958 2959 return (i); 2960 } 2961 2962 static void 2963 pf_purge_unlinked_rules(void) 2964 { 2965 struct pf_krulequeue tmpq; 2966 struct pf_krule *r, *r1; 2967 2968 /* 2969 * If we have overloading task pending, then we'd 2970 * better skip purging this time. There is a tiny 2971 * probability that overloading task references 2972 * an already unlinked rule. 2973 */ 2974 PF_OVERLOADQ_LOCK(); 2975 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2976 PF_OVERLOADQ_UNLOCK(); 2977 return; 2978 } 2979 PF_OVERLOADQ_UNLOCK(); 2980 2981 /* 2982 * Do naive mark-and-sweep garbage collecting of old rules. 2983 * Reference flag is raised by pf_purge_expired_states() 2984 * and pf_purge_expired_src_nodes(). 2985 * 2986 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2987 * use a temporary queue. 2988 */ 2989 TAILQ_INIT(&tmpq); 2990 PF_UNLNKDRULES_LOCK(); 2991 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2992 if (!(r->rule_ref & PFRULE_REFS)) { 2993 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2994 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2995 } else 2996 r->rule_ref &= ~PFRULE_REFS; 2997 } 2998 PF_UNLNKDRULES_UNLOCK(); 2999 3000 if (!TAILQ_EMPTY(&tmpq)) { 3001 PF_CONFIG_LOCK(); 3002 PF_RULES_WLOCK(); 3003 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 3004 TAILQ_REMOVE(&tmpq, r, entries); 3005 pf_free_rule(r); 3006 } 3007 PF_RULES_WUNLOCK(); 3008 PF_CONFIG_UNLOCK(); 3009 } 3010 } 3011 3012 void 3013 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 3014 { 3015 switch (af) { 3016 #ifdef INET 3017 case AF_INET: { 3018 u_int32_t a = ntohl(addr->addr32[0]); 3019 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 3020 (a>>8)&255, a&255); 3021 if (p) { 3022 p = ntohs(p); 3023 printf(":%u", p); 3024 } 3025 break; 3026 } 3027 #endif /* INET */ 3028 #ifdef INET6 3029 case AF_INET6: { 3030 u_int16_t b; 3031 u_int8_t i, curstart, curend, maxstart, maxend; 3032 curstart = curend = maxstart = maxend = 255; 3033 for (i = 0; i < 8; i++) { 3034 if (!addr->addr16[i]) { 3035 if (curstart == 255) 3036 curstart = i; 3037 curend = i; 3038 } else { 3039 if ((curend - curstart) > 3040 (maxend - maxstart)) { 3041 maxstart = curstart; 3042 maxend = curend; 3043 } 3044 curstart = curend = 255; 3045 } 3046 } 3047 if ((curend - curstart) > 3048 (maxend - maxstart)) { 3049 maxstart = curstart; 3050 maxend = curend; 3051 } 3052 for (i = 0; i < 8; i++) { 3053 if (i >= maxstart && i <= maxend) { 3054 if (i == 0) 3055 printf(":"); 3056 if (i == maxend) 3057 printf(":"); 3058 } else { 3059 b = ntohs(addr->addr16[i]); 3060 printf("%x", b); 3061 if (i < 7) 3062 printf(":"); 3063 } 3064 } 3065 if (p) { 3066 p = ntohs(p); 3067 printf("[%u]", p); 3068 } 3069 break; 3070 } 3071 #endif /* INET6 */ 3072 default: 3073 unhandled_af(af); 3074 } 3075 } 3076 3077 void 3078 pf_print_state(struct pf_kstate *s) 3079 { 3080 pf_print_state_parts(s, NULL, NULL); 3081 } 3082 3083 static void 3084 pf_print_state_parts(struct pf_kstate *s, 3085 struct pf_state_key *skwp, struct pf_state_key *sksp) 3086 { 3087 struct pf_state_key *skw, *sks; 3088 u_int8_t proto, dir; 3089 3090 /* Do our best to fill these, but they're skipped if NULL */ 3091 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 3092 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 3093 proto = skw ? skw->proto : (sks ? sks->proto : 0); 3094 dir = s ? s->direction : 0; 3095 3096 switch (proto) { 3097 case IPPROTO_IPV4: 3098 printf("IPv4"); 3099 break; 3100 case IPPROTO_IPV6: 3101 printf("IPv6"); 3102 break; 3103 case IPPROTO_TCP: 3104 printf("TCP"); 3105 break; 3106 case IPPROTO_UDP: 3107 printf("UDP"); 3108 break; 3109 case IPPROTO_ICMP: 3110 printf("ICMP"); 3111 break; 3112 case IPPROTO_ICMPV6: 3113 printf("ICMPv6"); 3114 break; 3115 default: 3116 printf("%u", proto); 3117 break; 3118 } 3119 switch (dir) { 3120 case PF_IN: 3121 printf(" in"); 3122 break; 3123 case PF_OUT: 3124 printf(" out"); 3125 break; 3126 } 3127 if (skw) { 3128 printf(" wire: "); 3129 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3130 printf(" "); 3131 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3132 } 3133 if (sks) { 3134 printf(" stack: "); 3135 if (sks != skw) { 3136 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3137 printf(" "); 3138 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3139 } else 3140 printf("-"); 3141 } 3142 if (s) { 3143 if (proto == IPPROTO_TCP) { 3144 printf(" [lo=%u high=%u win=%u modulator=%u", 3145 s->src.seqlo, s->src.seqhi, 3146 s->src.max_win, s->src.seqdiff); 3147 if (s->src.wscale && s->dst.wscale) 3148 printf(" wscale=%u", 3149 s->src.wscale & PF_WSCALE_MASK); 3150 printf("]"); 3151 printf(" [lo=%u high=%u win=%u modulator=%u", 3152 s->dst.seqlo, s->dst.seqhi, 3153 s->dst.max_win, s->dst.seqdiff); 3154 if (s->src.wscale && s->dst.wscale) 3155 printf(" wscale=%u", 3156 s->dst.wscale & PF_WSCALE_MASK); 3157 printf("]"); 3158 } 3159 printf(" %u:%u", s->src.state, s->dst.state); 3160 if (s->rule) 3161 printf(" @%d", s->rule->nr); 3162 } 3163 } 3164 3165 void 3166 pf_print_flags(uint16_t f) 3167 { 3168 if (f) 3169 printf(" "); 3170 if (f & TH_FIN) 3171 printf("F"); 3172 if (f & TH_SYN) 3173 printf("S"); 3174 if (f & TH_RST) 3175 printf("R"); 3176 if (f & TH_PUSH) 3177 printf("P"); 3178 if (f & TH_ACK) 3179 printf("A"); 3180 if (f & TH_URG) 3181 printf("U"); 3182 if (f & TH_ECE) 3183 printf("E"); 3184 if (f & TH_CWR) 3185 printf("W"); 3186 if (f & TH_AE) 3187 printf("e"); 3188 } 3189 3190 #define PF_SET_SKIP_STEPS(i) \ 3191 do { \ 3192 while (head[i] != cur) { \ 3193 head[i]->skip[i] = cur; \ 3194 head[i] = TAILQ_NEXT(head[i], entries); \ 3195 } \ 3196 } while (0) 3197 3198 void 3199 pf_calc_skip_steps(struct pf_krulequeue *rules) 3200 { 3201 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3202 int i; 3203 3204 cur = TAILQ_FIRST(rules); 3205 prev = cur; 3206 for (i = 0; i < PF_SKIP_COUNT; ++i) 3207 head[i] = cur; 3208 while (cur != NULL) { 3209 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3210 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3211 if (cur->direction != prev->direction) 3212 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3213 if (cur->af != prev->af) 3214 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3215 if (cur->proto != prev->proto) 3216 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3217 if (cur->src.neg != prev->src.neg || 3218 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3219 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3220 if (cur->dst.neg != prev->dst.neg || 3221 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3222 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3223 if (cur->src.port[0] != prev->src.port[0] || 3224 cur->src.port[1] != prev->src.port[1] || 3225 cur->src.port_op != prev->src.port_op) 3226 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3227 if (cur->dst.port[0] != prev->dst.port[0] || 3228 cur->dst.port[1] != prev->dst.port[1] || 3229 cur->dst.port_op != prev->dst.port_op) 3230 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3231 3232 prev = cur; 3233 cur = TAILQ_NEXT(cur, entries); 3234 } 3235 for (i = 0; i < PF_SKIP_COUNT; ++i) 3236 PF_SET_SKIP_STEPS(i); 3237 } 3238 3239 int 3240 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3241 { 3242 if (aw1->type != aw2->type) 3243 return (1); 3244 switch (aw1->type) { 3245 case PF_ADDR_ADDRMASK: 3246 case PF_ADDR_RANGE: 3247 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3248 return (1); 3249 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3250 return (1); 3251 return (0); 3252 case PF_ADDR_DYNIFTL: 3253 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3254 case PF_ADDR_NONE: 3255 case PF_ADDR_NOROUTE: 3256 case PF_ADDR_URPFFAILED: 3257 return (0); 3258 case PF_ADDR_TABLE: 3259 return (aw1->p.tbl != aw2->p.tbl); 3260 default: 3261 printf("invalid address type: %d\n", aw1->type); 3262 return (1); 3263 } 3264 } 3265 3266 /** 3267 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3268 * header isn't always a full checksum. In some cases (i.e. output) it's a 3269 * pseudo-header checksum, which is a partial checksum over src/dst IP 3270 * addresses, protocol number and length. 3271 * 3272 * That means we have the following cases: 3273 * * Input or forwarding: we don't have TSO, the checksum fields are full 3274 * checksums, we need to update the checksum whenever we change anything. 3275 * * Output (i.e. the checksum is a pseudo-header checksum): 3276 * x The field being updated is src/dst address or affects the length of 3277 * the packet. We need to update the pseudo-header checksum (note that this 3278 * checksum is not ones' complement). 3279 * x Some other field is being modified (e.g. src/dst port numbers): We 3280 * don't have to update anything. 3281 **/ 3282 u_int16_t 3283 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3284 { 3285 u_int32_t x; 3286 3287 x = cksum + old - new; 3288 x = (x + (x >> 16)) & 0xffff; 3289 3290 /* optimise: eliminate a branch when not udp */ 3291 if (udp && cksum == 0x0000) 3292 return cksum; 3293 if (udp && x == 0x0000) 3294 x = 0xffff; 3295 3296 return (u_int16_t)(x); 3297 } 3298 3299 static int 3300 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 3301 { 3302 int rewrite = 0; 3303 3304 if (*f != v) { 3305 uint16_t old = htons(hi ? (*f << 8) : *f); 3306 uint16_t new = htons(hi ? ( v << 8) : v); 3307 3308 *f = v; 3309 3310 if (! (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3311 CSUM_DELAY_DATA_IPV6))) 3312 *pd->pcksum = pf_cksum_fixup(*pd->pcksum, old, new, 3313 pd->proto == IPPROTO_UDP); 3314 3315 rewrite = 1; 3316 } 3317 3318 return (rewrite); 3319 } 3320 3321 int 3322 pf_patch_16(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 3323 { 3324 int rewrite = 0; 3325 u_int8_t *fb = (u_int8_t *)f; 3326 u_int8_t *vb = (u_int8_t *)&v; 3327 3328 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3329 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3330 3331 return (rewrite); 3332 } 3333 3334 int 3335 pf_patch_32(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 3336 { 3337 int rewrite = 0; 3338 u_int8_t *fb = (u_int8_t *)f; 3339 u_int8_t *vb = (u_int8_t *)&v; 3340 3341 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3342 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3343 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3344 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3345 3346 return (rewrite); 3347 } 3348 3349 u_int16_t 3350 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3351 u_int16_t new, u_int8_t udp) 3352 { 3353 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3354 return (cksum); 3355 3356 return (pf_cksum_fixup(cksum, old, new, udp)); 3357 } 3358 3359 static void 3360 pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p, 3361 struct pf_addr *an, u_int16_t pn) 3362 { 3363 struct pf_addr ao; 3364 u_int16_t po; 3365 uint8_t u = pd->virtual_proto == IPPROTO_UDP; 3366 3367 MPASS(pd->pcksum != NULL); 3368 if (pd->af == AF_INET) { 3369 MPASS(pd->ip_sum); 3370 } 3371 3372 pf_addrcpy(&ao, a, pd->af); 3373 if (pd->af == pd->naf) 3374 pf_addrcpy(a, an, pd->af); 3375 3376 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3377 *pd->pcksum = ~*pd->pcksum; 3378 3379 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3380 return; 3381 po = *p; 3382 *p = pn; 3383 3384 switch (pd->af) { 3385 #ifdef INET 3386 case AF_INET: 3387 switch (pd->naf) { 3388 case AF_INET: 3389 *pd->ip_sum = pf_cksum_fixup(pf_cksum_fixup(*pd->ip_sum, 3390 ao.addr16[0], an->addr16[0], 0), 3391 ao.addr16[1], an->addr16[1], 0); 3392 *p = pn; 3393 3394 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3395 ao.addr16[0], an->addr16[0], u), 3396 ao.addr16[1], an->addr16[1], u); 3397 3398 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3399 break; 3400 #ifdef INET6 3401 case AF_INET6: 3402 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3403 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3404 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3405 ao.addr16[0], an->addr16[0], u), 3406 ao.addr16[1], an->addr16[1], u), 3407 0, an->addr16[2], u), 3408 0, an->addr16[3], u), 3409 0, an->addr16[4], u), 3410 0, an->addr16[5], u), 3411 0, an->addr16[6], u), 3412 0, an->addr16[7], u), 3413 po, pn, u); 3414 break; 3415 #endif /* INET6 */ 3416 default: 3417 unhandled_af(pd->naf); 3418 } 3419 break; 3420 #endif /* INET */ 3421 #ifdef INET6 3422 case AF_INET6: 3423 switch (pd->naf) { 3424 #ifdef INET 3425 case AF_INET: 3426 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3427 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3428 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3429 ao.addr16[0], an->addr16[0], u), 3430 ao.addr16[1], an->addr16[1], u), 3431 ao.addr16[2], 0, u), 3432 ao.addr16[3], 0, u), 3433 ao.addr16[4], 0, u), 3434 ao.addr16[5], 0, u), 3435 ao.addr16[6], 0, u), 3436 ao.addr16[7], 0, u), 3437 po, pn, u); 3438 break; 3439 #endif /* INET */ 3440 case AF_INET6: 3441 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3442 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3443 pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3444 ao.addr16[0], an->addr16[0], u), 3445 ao.addr16[1], an->addr16[1], u), 3446 ao.addr16[2], an->addr16[2], u), 3447 ao.addr16[3], an->addr16[3], u), 3448 ao.addr16[4], an->addr16[4], u), 3449 ao.addr16[5], an->addr16[5], u), 3450 ao.addr16[6], an->addr16[6], u), 3451 ao.addr16[7], an->addr16[7], u); 3452 3453 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3454 break; 3455 default: 3456 unhandled_af(pd->naf); 3457 } 3458 break; 3459 #endif /* INET6 */ 3460 default: 3461 unhandled_af(pd->af); 3462 } 3463 3464 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3465 CSUM_DELAY_DATA_IPV6)) { 3466 *pd->pcksum = ~*pd->pcksum; 3467 if (! *pd->pcksum) 3468 *pd->pcksum = 0xffff; 3469 } 3470 } 3471 3472 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3473 void 3474 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3475 { 3476 u_int32_t ao; 3477 3478 memcpy(&ao, a, sizeof(ao)); 3479 memcpy(a, &an, sizeof(u_int32_t)); 3480 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3481 ao % 65536, an % 65536, u); 3482 } 3483 3484 void 3485 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3486 { 3487 u_int32_t ao; 3488 3489 memcpy(&ao, a, sizeof(ao)); 3490 memcpy(a, &an, sizeof(u_int32_t)); 3491 3492 *c = pf_proto_cksum_fixup(m, 3493 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3494 ao % 65536, an % 65536, udp); 3495 } 3496 3497 #ifdef INET6 3498 static void 3499 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3500 { 3501 struct pf_addr ao; 3502 3503 pf_addrcpy(&ao, a, AF_INET6); 3504 pf_addrcpy(a, an, AF_INET6); 3505 3506 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3507 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3508 pf_cksum_fixup(pf_cksum_fixup(*c, 3509 ao.addr16[0], an->addr16[0], u), 3510 ao.addr16[1], an->addr16[1], u), 3511 ao.addr16[2], an->addr16[2], u), 3512 ao.addr16[3], an->addr16[3], u), 3513 ao.addr16[4], an->addr16[4], u), 3514 ao.addr16[5], an->addr16[5], u), 3515 ao.addr16[6], an->addr16[6], u), 3516 ao.addr16[7], an->addr16[7], u); 3517 } 3518 #endif /* INET6 */ 3519 3520 static void 3521 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3522 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3523 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3524 { 3525 struct pf_addr oia, ooa; 3526 3527 pf_addrcpy(&oia, ia, af); 3528 if (oa) 3529 pf_addrcpy(&ooa, oa, af); 3530 3531 /* Change inner protocol port, fix inner protocol checksum. */ 3532 if (ip != NULL) { 3533 u_int16_t oip = *ip; 3534 u_int16_t opc; 3535 3536 if (pc != NULL) 3537 opc = *pc; 3538 *ip = np; 3539 if (pc != NULL) 3540 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3541 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3542 if (pc != NULL) 3543 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3544 } 3545 /* Change inner ip address, fix inner ip and icmp checksums. */ 3546 pf_addrcpy(ia, na, af); 3547 switch (af) { 3548 #ifdef INET 3549 case AF_INET: { 3550 u_int16_t oh2c = *h2c; 3551 3552 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3553 oia.addr16[0], ia->addr16[0], 0), 3554 oia.addr16[1], ia->addr16[1], 0); 3555 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3556 oia.addr16[0], ia->addr16[0], 0), 3557 oia.addr16[1], ia->addr16[1], 0); 3558 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3559 break; 3560 } 3561 #endif /* INET */ 3562 #ifdef INET6 3563 case AF_INET6: 3564 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3565 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3566 pf_cksum_fixup(pf_cksum_fixup(*ic, 3567 oia.addr16[0], ia->addr16[0], u), 3568 oia.addr16[1], ia->addr16[1], u), 3569 oia.addr16[2], ia->addr16[2], u), 3570 oia.addr16[3], ia->addr16[3], u), 3571 oia.addr16[4], ia->addr16[4], u), 3572 oia.addr16[5], ia->addr16[5], u), 3573 oia.addr16[6], ia->addr16[6], u), 3574 oia.addr16[7], ia->addr16[7], u); 3575 break; 3576 #endif /* INET6 */ 3577 } 3578 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3579 if (oa) { 3580 pf_addrcpy(oa, na, af); 3581 switch (af) { 3582 #ifdef INET 3583 case AF_INET: 3584 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3585 ooa.addr16[0], oa->addr16[0], 0), 3586 ooa.addr16[1], oa->addr16[1], 0); 3587 break; 3588 #endif /* INET */ 3589 #ifdef INET6 3590 case AF_INET6: 3591 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3592 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3593 pf_cksum_fixup(pf_cksum_fixup(*ic, 3594 ooa.addr16[0], oa->addr16[0], u), 3595 ooa.addr16[1], oa->addr16[1], u), 3596 ooa.addr16[2], oa->addr16[2], u), 3597 ooa.addr16[3], oa->addr16[3], u), 3598 ooa.addr16[4], oa->addr16[4], u), 3599 ooa.addr16[5], oa->addr16[5], u), 3600 ooa.addr16[6], oa->addr16[6], u), 3601 ooa.addr16[7], oa->addr16[7], u); 3602 break; 3603 #endif /* INET6 */ 3604 } 3605 } 3606 } 3607 3608 static int 3609 pf_translate_af(struct pf_pdesc *pd, struct pf_krule *r) 3610 { 3611 #if defined(INET) && defined(INET6) 3612 struct mbuf *mp; 3613 struct ip *ip4; 3614 struct ip6_hdr *ip6; 3615 struct icmp6_hdr *icmp; 3616 struct m_tag *mtag; 3617 struct pf_fragment_tag *ftag; 3618 int hlen; 3619 3620 if (pd->ttl == 1) { 3621 /* We'd generate an ICMP error. Do so now rather than after af translation. */ 3622 if (pd->af == AF_INET) { 3623 pf_send_icmp(pd->m, ICMP_TIMXCEED, 3624 ICMP_TIMXCEED_INTRANS, 0, pd->af, r, 3625 pd->act.rtableid); 3626 } else { 3627 pf_send_icmp(pd->m, ICMP6_TIME_EXCEEDED, 3628 ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r, 3629 pd->act.rtableid); 3630 } 3631 3632 return (-1); 3633 } 3634 3635 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3636 3637 /* trim the old header */ 3638 m_adj(pd->m, pd->off); 3639 3640 /* prepend a new one */ 3641 M_PREPEND(pd->m, hlen, M_NOWAIT); 3642 if (pd->m == NULL) 3643 return (-1); 3644 3645 switch (pd->naf) { 3646 case AF_INET: 3647 ip4 = mtod(pd->m, struct ip *); 3648 bzero(ip4, hlen); 3649 ip4->ip_v = IPVERSION; 3650 ip4->ip_hl = hlen >> 2; 3651 ip4->ip_tos = pd->tos; 3652 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3653 ip_fillid(ip4, V_ip_random_id); 3654 ip4->ip_ttl = pd->ttl; 3655 ip4->ip_p = pd->proto; 3656 ip4->ip_src = pd->nsaddr.v4; 3657 ip4->ip_dst = pd->ndaddr.v4; 3658 pd->src = (struct pf_addr *)&ip4->ip_src; 3659 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3660 pd->off = sizeof(struct ip); 3661 if (pd->m->m_pkthdr.csum_flags & CSUM_TCP_IPV6) { 3662 pd->m->m_pkthdr.csum_flags &= ~CSUM_TCP_IPV6; 3663 pd->m->m_pkthdr.csum_flags |= CSUM_TCP; 3664 } 3665 if (pd->m->m_pkthdr.csum_flags & CSUM_UDP_IPV6) { 3666 pd->m->m_pkthdr.csum_flags &= ~CSUM_UDP_IPV6; 3667 pd->m->m_pkthdr.csum_flags |= CSUM_UDP; 3668 } 3669 if (pd->m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) { 3670 pd->m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; 3671 pd->m->m_pkthdr.csum_flags |= CSUM_SCTP; 3672 } 3673 break; 3674 case AF_INET6: 3675 ip6 = mtod(pd->m, struct ip6_hdr *); 3676 bzero(ip6, hlen); 3677 ip6->ip6_vfc = IPV6_VERSION; 3678 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3679 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3680 ip6->ip6_nxt = pd->proto; 3681 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3682 ip6->ip6_hlim = IPV6_DEFHLIM; 3683 else 3684 ip6->ip6_hlim = pd->ttl; 3685 ip6->ip6_src = pd->nsaddr.v6; 3686 ip6->ip6_dst = pd->ndaddr.v6; 3687 pd->src = (struct pf_addr *)&ip6->ip6_src; 3688 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3689 pd->off = sizeof(struct ip6_hdr); 3690 if (pd->m->m_pkthdr.csum_flags & CSUM_TCP) { 3691 pd->m->m_pkthdr.csum_flags &= ~CSUM_TCP; 3692 pd->m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 3693 } 3694 if (pd->m->m_pkthdr.csum_flags & CSUM_UDP) { 3695 pd->m->m_pkthdr.csum_flags &= ~CSUM_UDP; 3696 pd->m->m_pkthdr.csum_flags |= CSUM_UDP_IPV6; 3697 } 3698 if (pd->m->m_pkthdr.csum_flags & CSUM_SCTP) { 3699 pd->m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 3700 pd->m->m_pkthdr.csum_flags |= CSUM_SCTP_IPV6; 3701 } 3702 3703 /* 3704 * If we're dealing with a reassembled packet we need to adjust 3705 * the header length from the IPv4 header size to IPv6 header 3706 * size. 3707 */ 3708 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3709 if (mtag) { 3710 ftag = (struct pf_fragment_tag *)(mtag + 1); 3711 ftag->ft_hdrlen = sizeof(*ip6); 3712 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3713 sizeof(struct ip) + sizeof(struct ip6_frag); 3714 } 3715 break; 3716 default: 3717 return (-1); 3718 } 3719 3720 /* recalculate icmp/icmp6 checksums */ 3721 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3722 int off; 3723 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3724 NULL) { 3725 pd->m = NULL; 3726 return (-1); 3727 } 3728 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3729 icmp->icmp6_cksum = 0; 3730 icmp->icmp6_cksum = pd->naf == AF_INET ? 3731 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3732 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3733 ntohs(ip6->ip6_plen)); 3734 } 3735 #endif /* INET && INET6 */ 3736 3737 return (0); 3738 } 3739 3740 int 3741 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3742 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3743 sa_family_t af, sa_family_t naf) 3744 { 3745 #if defined(INET) && defined(INET6) 3746 struct mbuf *n = NULL; 3747 struct ip *ip4; 3748 struct ip6_hdr *ip6; 3749 int hlen, olen, mlen; 3750 3751 if (af == naf || (af != AF_INET && af != AF_INET6) || 3752 (naf != AF_INET && naf != AF_INET6)) 3753 return (-1); 3754 3755 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3756 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3757 return (-1); 3758 3759 /* old header */ 3760 olen = pd2->off - off; 3761 /* new header */ 3762 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3763 3764 /* trim old header */ 3765 m_adj(n, olen); 3766 3767 /* prepend a new one */ 3768 M_PREPEND(n, hlen, M_NOWAIT); 3769 if (n == NULL) 3770 return (-1); 3771 3772 /* translate inner ip/ip6 header */ 3773 switch (naf) { 3774 case AF_INET: 3775 ip4 = mtod(n, struct ip *); 3776 bzero(ip4, sizeof(*ip4)); 3777 ip4->ip_v = IPVERSION; 3778 ip4->ip_hl = sizeof(*ip4) >> 2; 3779 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 3780 ip_fillid(ip4, V_ip_random_id); 3781 ip4->ip_off = htons(IP_DF); 3782 ip4->ip_ttl = pd2->ttl; 3783 if (pd2->proto == IPPROTO_ICMPV6) 3784 ip4->ip_p = IPPROTO_ICMP; 3785 else 3786 ip4->ip_p = pd2->proto; 3787 ip4->ip_src = src->v4; 3788 ip4->ip_dst = dst->v4; 3789 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3790 break; 3791 case AF_INET6: 3792 ip6 = mtod(n, struct ip6_hdr *); 3793 bzero(ip6, sizeof(*ip6)); 3794 ip6->ip6_vfc = IPV6_VERSION; 3795 ip6->ip6_plen = htons(pd2->tot_len - olen); 3796 if (pd2->proto == IPPROTO_ICMP) 3797 ip6->ip6_nxt = IPPROTO_ICMPV6; 3798 else 3799 ip6->ip6_nxt = pd2->proto; 3800 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3801 ip6->ip6_hlim = IPV6_DEFHLIM; 3802 else 3803 ip6->ip6_hlim = pd2->ttl; 3804 ip6->ip6_src = src->v6; 3805 ip6->ip6_dst = dst->v6; 3806 break; 3807 default: 3808 unhandled_af(naf); 3809 } 3810 3811 /* adjust payload offset and total packet length */ 3812 pd2->off += hlen - olen; 3813 pd->tot_len += hlen - olen; 3814 3815 /* merge modified inner packet with the original header */ 3816 mlen = n->m_pkthdr.len; 3817 m_cat(m, n); 3818 m->m_pkthdr.len += mlen; 3819 #endif /* INET && INET6 */ 3820 3821 return (0); 3822 } 3823 3824 #define PTR_IP(field) (offsetof(struct ip, field)) 3825 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3826 3827 int 3828 pf_translate_icmp_af(int af, void *arg) 3829 { 3830 #if defined(INET) && defined(INET6) 3831 struct icmp *icmp4; 3832 struct icmp6_hdr *icmp6; 3833 u_int32_t mtu; 3834 int32_t ptr = -1; 3835 u_int8_t type; 3836 u_int8_t code; 3837 3838 switch (af) { 3839 case AF_INET: 3840 icmp6 = arg; 3841 type = icmp6->icmp6_type; 3842 code = icmp6->icmp6_code; 3843 mtu = ntohl(icmp6->icmp6_mtu); 3844 3845 switch (type) { 3846 case ICMP6_ECHO_REQUEST: 3847 type = ICMP_ECHO; 3848 break; 3849 case ICMP6_ECHO_REPLY: 3850 type = ICMP_ECHOREPLY; 3851 break; 3852 case ICMP6_DST_UNREACH: 3853 type = ICMP_UNREACH; 3854 switch (code) { 3855 case ICMP6_DST_UNREACH_NOROUTE: 3856 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3857 case ICMP6_DST_UNREACH_ADDR: 3858 code = ICMP_UNREACH_HOST; 3859 break; 3860 case ICMP6_DST_UNREACH_ADMIN: 3861 code = ICMP_UNREACH_HOST_PROHIB; 3862 break; 3863 case ICMP6_DST_UNREACH_NOPORT: 3864 code = ICMP_UNREACH_PORT; 3865 break; 3866 default: 3867 return (-1); 3868 } 3869 break; 3870 case ICMP6_PACKET_TOO_BIG: 3871 type = ICMP_UNREACH; 3872 code = ICMP_UNREACH_NEEDFRAG; 3873 mtu -= 20; 3874 break; 3875 case ICMP6_TIME_EXCEEDED: 3876 type = ICMP_TIMXCEED; 3877 break; 3878 case ICMP6_PARAM_PROB: 3879 switch (code) { 3880 case ICMP6_PARAMPROB_HEADER: 3881 type = ICMP_PARAMPROB; 3882 code = ICMP_PARAMPROB_ERRATPTR; 3883 ptr = ntohl(icmp6->icmp6_pptr); 3884 3885 if (ptr == PTR_IP6(ip6_vfc)) 3886 ; /* preserve */ 3887 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3888 ptr = PTR_IP(ip_tos); 3889 else if (ptr == PTR_IP6(ip6_plen) || 3890 ptr == PTR_IP6(ip6_plen) + 1) 3891 ptr = PTR_IP(ip_len); 3892 else if (ptr == PTR_IP6(ip6_nxt)) 3893 ptr = PTR_IP(ip_p); 3894 else if (ptr == PTR_IP6(ip6_hlim)) 3895 ptr = PTR_IP(ip_ttl); 3896 else if (ptr >= PTR_IP6(ip6_src) && 3897 ptr < PTR_IP6(ip6_dst)) 3898 ptr = PTR_IP(ip_src); 3899 else if (ptr >= PTR_IP6(ip6_dst) && 3900 ptr < sizeof(struct ip6_hdr)) 3901 ptr = PTR_IP(ip_dst); 3902 else { 3903 return (-1); 3904 } 3905 break; 3906 case ICMP6_PARAMPROB_NEXTHEADER: 3907 type = ICMP_UNREACH; 3908 code = ICMP_UNREACH_PROTOCOL; 3909 break; 3910 default: 3911 return (-1); 3912 } 3913 break; 3914 default: 3915 return (-1); 3916 } 3917 if (icmp6->icmp6_type != type) { 3918 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3919 icmp6->icmp6_type, type, 0); 3920 icmp6->icmp6_type = type; 3921 } 3922 if (icmp6->icmp6_code != code) { 3923 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3924 icmp6->icmp6_code, code, 0); 3925 icmp6->icmp6_code = code; 3926 } 3927 if (icmp6->icmp6_mtu != htonl(mtu)) { 3928 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3929 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3930 /* aligns well with a icmpv4 nextmtu */ 3931 icmp6->icmp6_mtu = htonl(mtu); 3932 } 3933 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3934 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3935 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3936 /* icmpv4 pptr is a one most significant byte */ 3937 icmp6->icmp6_pptr = htonl(ptr << 24); 3938 } 3939 break; 3940 case AF_INET6: 3941 icmp4 = arg; 3942 type = icmp4->icmp_type; 3943 code = icmp4->icmp_code; 3944 mtu = ntohs(icmp4->icmp_nextmtu); 3945 3946 switch (type) { 3947 case ICMP_ECHO: 3948 type = ICMP6_ECHO_REQUEST; 3949 break; 3950 case ICMP_ECHOREPLY: 3951 type = ICMP6_ECHO_REPLY; 3952 break; 3953 case ICMP_UNREACH: 3954 type = ICMP6_DST_UNREACH; 3955 switch (code) { 3956 case ICMP_UNREACH_NET: 3957 case ICMP_UNREACH_HOST: 3958 case ICMP_UNREACH_NET_UNKNOWN: 3959 case ICMP_UNREACH_HOST_UNKNOWN: 3960 case ICMP_UNREACH_ISOLATED: 3961 case ICMP_UNREACH_TOSNET: 3962 case ICMP_UNREACH_TOSHOST: 3963 code = ICMP6_DST_UNREACH_NOROUTE; 3964 break; 3965 case ICMP_UNREACH_PORT: 3966 code = ICMP6_DST_UNREACH_NOPORT; 3967 break; 3968 case ICMP_UNREACH_NET_PROHIB: 3969 case ICMP_UNREACH_HOST_PROHIB: 3970 case ICMP_UNREACH_FILTER_PROHIB: 3971 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3972 code = ICMP6_DST_UNREACH_ADMIN; 3973 break; 3974 case ICMP_UNREACH_PROTOCOL: 3975 type = ICMP6_PARAM_PROB; 3976 code = ICMP6_PARAMPROB_NEXTHEADER; 3977 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3978 break; 3979 case ICMP_UNREACH_NEEDFRAG: 3980 type = ICMP6_PACKET_TOO_BIG; 3981 code = 0; 3982 mtu += 20; 3983 break; 3984 default: 3985 return (-1); 3986 } 3987 break; 3988 case ICMP_TIMXCEED: 3989 type = ICMP6_TIME_EXCEEDED; 3990 break; 3991 case ICMP_PARAMPROB: 3992 type = ICMP6_PARAM_PROB; 3993 switch (code) { 3994 case ICMP_PARAMPROB_ERRATPTR: 3995 code = ICMP6_PARAMPROB_HEADER; 3996 break; 3997 case ICMP_PARAMPROB_LENGTH: 3998 code = ICMP6_PARAMPROB_HEADER; 3999 break; 4000 default: 4001 return (-1); 4002 } 4003 4004 ptr = icmp4->icmp_pptr; 4005 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 4006 ; /* preserve */ 4007 else if (ptr == PTR_IP(ip_len) || 4008 ptr == PTR_IP(ip_len) + 1) 4009 ptr = PTR_IP6(ip6_plen); 4010 else if (ptr == PTR_IP(ip_ttl)) 4011 ptr = PTR_IP6(ip6_hlim); 4012 else if (ptr == PTR_IP(ip_p)) 4013 ptr = PTR_IP6(ip6_nxt); 4014 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 4015 ptr = PTR_IP6(ip6_src); 4016 else if (ptr >= PTR_IP(ip_dst) && 4017 ptr < sizeof(struct ip)) 4018 ptr = PTR_IP6(ip6_dst); 4019 else { 4020 return (-1); 4021 } 4022 break; 4023 default: 4024 return (-1); 4025 } 4026 if (icmp4->icmp_type != type) { 4027 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4028 icmp4->icmp_type, type, 0); 4029 icmp4->icmp_type = type; 4030 } 4031 if (icmp4->icmp_code != code) { 4032 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4033 icmp4->icmp_code, code, 0); 4034 icmp4->icmp_code = code; 4035 } 4036 if (icmp4->icmp_nextmtu != htons(mtu)) { 4037 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4038 icmp4->icmp_nextmtu, htons(mtu), 0); 4039 icmp4->icmp_nextmtu = htons(mtu); 4040 } 4041 if (ptr >= 0 && icmp4->icmp_void != ptr) { 4042 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4043 htons(icmp4->icmp_pptr), htons(ptr), 0); 4044 icmp4->icmp_void = htonl(ptr); 4045 } 4046 break; 4047 default: 4048 unhandled_af(af); 4049 } 4050 #endif /* INET && INET6 */ 4051 4052 return (0); 4053 } 4054 4055 /* 4056 * Need to modulate the sequence numbers in the TCP SACK option 4057 * (credits to Krzysztof Pfaff for report and patch) 4058 */ 4059 static int 4060 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 4061 struct pf_state_peer *dst) 4062 { 4063 struct sackblk sack; 4064 int copyback = 0, i; 4065 int olen, optsoff; 4066 uint8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 4067 4068 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 4069 optsoff = pd->off + sizeof(struct tcphdr); 4070 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 4071 if (olen < TCPOLEN_MINSACK || 4072 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, pd->af)) 4073 return (0); 4074 4075 eoh = opts + olen; 4076 opt = opts; 4077 while ((opt = pf_find_tcpopt(opt, opts, olen, 4078 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 4079 { 4080 size_t safelen = MIN(opt[1], (eoh - opt)); 4081 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 4082 size_t startoff = (opt + i) - opts; 4083 memcpy(&sack, &opt[i], sizeof(sack)); 4084 pf_patch_32(pd, &sack.start, 4085 htonl(ntohl(sack.start) - dst->seqdiff), 4086 PF_ALGNMNT(startoff)); 4087 pf_patch_32(pd, &sack.end, 4088 htonl(ntohl(sack.end) - dst->seqdiff), 4089 PF_ALGNMNT(startoff + sizeof(sack.start))); 4090 memcpy(&opt[i], &sack, sizeof(sack)); 4091 } 4092 copyback = 1; 4093 opt += opt[1]; 4094 } 4095 4096 if (copyback) 4097 m_copyback(pd->m, optsoff, olen, (caddr_t)opts); 4098 4099 return (copyback); 4100 } 4101 4102 struct mbuf * 4103 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 4104 const struct pf_addr *saddr, const struct pf_addr *daddr, 4105 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4106 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4107 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, u_int sack, 4108 int rtableid, u_short *reason) 4109 { 4110 struct mbuf *m; 4111 int len, tlen; 4112 #ifdef INET 4113 struct ip *h = NULL; 4114 #endif /* INET */ 4115 #ifdef INET6 4116 struct ip6_hdr *h6 = NULL; 4117 #endif /* INET6 */ 4118 struct tcphdr *th; 4119 char *opt; 4120 struct pf_mtag *pf_mtag; 4121 4122 len = 0; 4123 th = NULL; 4124 4125 /* maximum segment size tcp option */ 4126 tlen = sizeof(struct tcphdr); 4127 if (mss) 4128 tlen += 4; 4129 if (sack) 4130 tlen += 2; 4131 4132 switch (af) { 4133 #ifdef INET 4134 case AF_INET: 4135 len = sizeof(struct ip) + tlen; 4136 break; 4137 #endif /* INET */ 4138 #ifdef INET6 4139 case AF_INET6: 4140 len = sizeof(struct ip6_hdr) + tlen; 4141 break; 4142 #endif /* INET6 */ 4143 default: 4144 unhandled_af(af); 4145 } 4146 4147 m = m_gethdr(M_NOWAIT, MT_DATA); 4148 if (m == NULL) { 4149 REASON_SET(reason, PFRES_MEMORY); 4150 return (NULL); 4151 } 4152 4153 #ifdef MAC 4154 mac_netinet_firewall_send(m); 4155 #endif 4156 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 4157 REASON_SET(reason, PFRES_MEMORY); 4158 m_freem(m); 4159 return (NULL); 4160 } 4161 m->m_flags |= mbuf_flags; 4162 pf_mtag->tag = mtag_tag; 4163 pf_mtag->flags = mtag_flags; 4164 4165 if (rtableid >= 0) 4166 M_SETFIB(m, rtableid); 4167 4168 #ifdef ALTQ 4169 if (r != NULL && r->qid) { 4170 pf_mtag->qid = r->qid; 4171 4172 /* add hints for ecn */ 4173 pf_mtag->hdr = mtod(m, struct ip *); 4174 } 4175 #endif /* ALTQ */ 4176 m->m_data += max_linkhdr; 4177 m->m_pkthdr.len = m->m_len = len; 4178 /* The rest of the stack assumes a rcvif, so provide one. 4179 * This is a locally generated packet, so .. close enough. */ 4180 m->m_pkthdr.rcvif = V_loif; 4181 bzero(m->m_data, len); 4182 switch (af) { 4183 #ifdef INET 4184 case AF_INET: 4185 m->m_pkthdr.csum_flags |= CSUM_TCP; 4186 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4187 4188 h = mtod(m, struct ip *); 4189 4190 h->ip_p = IPPROTO_TCP; 4191 h->ip_len = htons(tlen); 4192 h->ip_v = 4; 4193 h->ip_hl = sizeof(*h) >> 2; 4194 h->ip_tos = IPTOS_LOWDELAY; 4195 h->ip_len = htons(len); 4196 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4197 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4198 h->ip_sum = 0; 4199 h->ip_src.s_addr = saddr->v4.s_addr; 4200 h->ip_dst.s_addr = daddr->v4.s_addr; 4201 4202 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4203 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4204 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4205 break; 4206 #endif /* INET */ 4207 #ifdef INET6 4208 case AF_INET6: 4209 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4210 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4211 4212 h6 = mtod(m, struct ip6_hdr *); 4213 4214 /* IP header fields included in the TCP checksum */ 4215 h6->ip6_nxt = IPPROTO_TCP; 4216 h6->ip6_plen = htons(tlen); 4217 h6->ip6_vfc |= IPV6_VERSION; 4218 h6->ip6_hlim = V_ip6_defhlim; 4219 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4220 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4221 4222 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4223 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4224 IPPROTO_TCP, 0); 4225 break; 4226 #endif /* INET6 */ 4227 } 4228 4229 /* TCP header */ 4230 th->th_sport = sport; 4231 th->th_dport = dport; 4232 th->th_seq = htonl(seq); 4233 th->th_ack = htonl(ack); 4234 th->th_off = tlen >> 2; 4235 tcp_set_flags(th, tcp_flags); 4236 th->th_win = htons(win); 4237 4238 opt = (char *)(th + 1); 4239 if (mss) { 4240 opt = (char *)(th + 1); 4241 opt[0] = TCPOPT_MAXSEG; 4242 opt[1] = 4; 4243 mss = htons(mss); 4244 memcpy((opt + 2), &mss, 2); 4245 opt += 4; 4246 } 4247 if (sack) { 4248 opt[0] = TCPOPT_SACK_PERMITTED; 4249 opt[1] = 2; 4250 opt += 2; 4251 } 4252 4253 return (m); 4254 } 4255 4256 static void 4257 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4258 uint8_t ttl, int rtableid) 4259 { 4260 struct mbuf *m; 4261 #ifdef INET 4262 struct ip *h = NULL; 4263 #endif /* INET */ 4264 #ifdef INET6 4265 struct ip6_hdr *h6 = NULL; 4266 #endif /* INET6 */ 4267 struct sctphdr *hdr; 4268 struct sctp_chunkhdr *chunk; 4269 struct pf_send_entry *pfse; 4270 int off = 0; 4271 4272 MPASS(af == pd->af); 4273 4274 m = m_gethdr(M_NOWAIT, MT_DATA); 4275 if (m == NULL) 4276 return; 4277 4278 m->m_data += max_linkhdr; 4279 m->m_flags |= M_SKIP_FIREWALL; 4280 /* The rest of the stack assumes a rcvif, so provide one. 4281 * This is a locally generated packet, so .. close enough. */ 4282 m->m_pkthdr.rcvif = V_loif; 4283 4284 /* IPv4|6 header */ 4285 switch (af) { 4286 #ifdef INET 4287 case AF_INET: 4288 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4289 4290 h = mtod(m, struct ip *); 4291 4292 /* IP header fields included in the TCP checksum */ 4293 4294 h->ip_p = IPPROTO_SCTP; 4295 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4296 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4297 h->ip_src = pd->dst->v4; 4298 h->ip_dst = pd->src->v4; 4299 4300 off += sizeof(struct ip); 4301 break; 4302 #endif /* INET */ 4303 #ifdef INET6 4304 case AF_INET6: 4305 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4306 4307 h6 = mtod(m, struct ip6_hdr *); 4308 4309 /* IP header fields included in the TCP checksum */ 4310 h6->ip6_vfc |= IPV6_VERSION; 4311 h6->ip6_nxt = IPPROTO_SCTP; 4312 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4313 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4314 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4315 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4316 4317 off += sizeof(struct ip6_hdr); 4318 break; 4319 #endif /* INET6 */ 4320 default: 4321 unhandled_af(af); 4322 } 4323 4324 /* SCTP header */ 4325 hdr = mtodo(m, off); 4326 4327 hdr->src_port = pd->hdr.sctp.dest_port; 4328 hdr->dest_port = pd->hdr.sctp.src_port; 4329 hdr->v_tag = pd->sctp_initiate_tag; 4330 hdr->checksum = 0; 4331 4332 /* Abort chunk. */ 4333 off += sizeof(struct sctphdr); 4334 chunk = mtodo(m, off); 4335 4336 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4337 chunk->chunk_length = htons(sizeof(*chunk)); 4338 4339 /* SCTP checksum */ 4340 off += sizeof(*chunk); 4341 m->m_pkthdr.len = m->m_len = off; 4342 4343 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4344 4345 if (rtableid >= 0) 4346 M_SETFIB(m, rtableid); 4347 4348 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4349 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4350 if (pfse == NULL) { 4351 m_freem(m); 4352 return; 4353 } 4354 4355 switch (af) { 4356 #ifdef INET 4357 case AF_INET: 4358 pfse->pfse_type = PFSE_IP; 4359 break; 4360 #endif /* INET */ 4361 #ifdef INET6 4362 case AF_INET6: 4363 pfse->pfse_type = PFSE_IP6; 4364 break; 4365 #endif /* INET6 */ 4366 } 4367 4368 pfse->pfse_m = m; 4369 pf_send(pfse); 4370 } 4371 4372 void 4373 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4374 const struct pf_addr *saddr, const struct pf_addr *daddr, 4375 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4376 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4377 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid, 4378 u_short *reason) 4379 { 4380 struct pf_send_entry *pfse; 4381 struct mbuf *m; 4382 4383 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4384 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, 0, rtableid, reason); 4385 if (m == NULL) 4386 return; 4387 4388 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4389 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4390 if (pfse == NULL) { 4391 m_freem(m); 4392 REASON_SET(reason, PFRES_MEMORY); 4393 return; 4394 } 4395 4396 switch (af) { 4397 #ifdef INET 4398 case AF_INET: 4399 pfse->pfse_type = PFSE_IP; 4400 break; 4401 #endif /* INET */ 4402 #ifdef INET6 4403 case AF_INET6: 4404 pfse->pfse_type = PFSE_IP6; 4405 break; 4406 #endif /* INET6 */ 4407 default: 4408 unhandled_af(af); 4409 } 4410 4411 pfse->pfse_m = m; 4412 pf_send(pfse); 4413 } 4414 4415 static void 4416 pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum) 4417 { 4418 /* undo NAT changes, if they have taken place */ 4419 if (nr != NULL) { 4420 pf_addrcpy(pd->src, &pd->osrc, pd->af); 4421 pf_addrcpy(pd->dst, &pd->odst, pd->af); 4422 if (pd->sport) 4423 *pd->sport = pd->osport; 4424 if (pd->dport) 4425 *pd->dport = pd->odport; 4426 if (pd->ip_sum) 4427 *pd->ip_sum = bip_sum; 4428 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4429 } 4430 } 4431 4432 static void 4433 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4434 struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum, 4435 u_short *reason, int rtableid) 4436 { 4437 pf_undo_nat(nr, pd, bip_sum); 4438 4439 if (pd->proto == IPPROTO_TCP && 4440 ((r->rule_flag & PFRULE_RETURNRST) || 4441 (r->rule_flag & PFRULE_RETURN)) && 4442 !(tcp_get_flags(th) & TH_RST)) { 4443 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4444 4445 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4446 IPPROTO_TCP, pd->af)) 4447 REASON_SET(reason, PFRES_PROTCKSUM); 4448 else { 4449 if (tcp_get_flags(th) & TH_SYN) 4450 ack++; 4451 if (tcp_get_flags(th) & TH_FIN) 4452 ack++; 4453 pf_send_tcp(r, pd->af, pd->dst, 4454 pd->src, th->th_dport, th->th_sport, 4455 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4456 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid, 4457 reason); 4458 } 4459 } else if (pd->proto == IPPROTO_SCTP && 4460 (r->rule_flag & PFRULE_RETURN)) { 4461 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4462 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4463 r->return_icmp) 4464 pf_send_icmp(pd->m, r->return_icmp >> 8, 4465 r->return_icmp & 255, 0, pd->af, r, rtableid); 4466 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4467 r->return_icmp6) 4468 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4469 r->return_icmp6 & 255, 0, pd->af, r, rtableid); 4470 } 4471 4472 static int 4473 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4474 { 4475 struct m_tag *mtag; 4476 u_int8_t mpcp; 4477 4478 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4479 if (mtag == NULL) 4480 return (0); 4481 4482 if (prio == PF_PRIO_ZERO) 4483 prio = 0; 4484 4485 mpcp = *(uint8_t *)(mtag + 1); 4486 4487 return (mpcp == prio); 4488 } 4489 4490 static int 4491 pf_icmp_to_bandlim(uint8_t type) 4492 { 4493 switch (type) { 4494 case ICMP_ECHO: 4495 case ICMP_ECHOREPLY: 4496 return (BANDLIM_ICMP_ECHO); 4497 case ICMP_TSTAMP: 4498 case ICMP_TSTAMPREPLY: 4499 return (BANDLIM_ICMP_TSTAMP); 4500 case ICMP_UNREACH: 4501 default: 4502 return (BANDLIM_ICMP_UNREACH); 4503 } 4504 } 4505 4506 static void 4507 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_kstate *s, 4508 struct pf_state_peer *src, struct pf_state_peer *dst, 4509 u_short *reason) 4510 { 4511 /* 4512 * We are sending challenge ACK as a response to SYN packet, which 4513 * matches existing state (modulo TCP window check). Therefore packet 4514 * must be sent on behalf of destination. 4515 * 4516 * We expect sender to remain either silent, or send RST packet 4517 * so both, firewall and remote peer, can purge dead state from 4518 * memory. 4519 */ 4520 pf_send_tcp(s->rule, pd->af, pd->dst, pd->src, 4521 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 4522 src->seqlo, TH_ACK, 0, 0, s->rule->return_ttl, 0, 0, 0, 4523 s->rule->rtableid, reason); 4524 } 4525 4526 static void 4527 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int mtu, 4528 sa_family_t af, struct pf_krule *r, int rtableid) 4529 { 4530 struct pf_send_entry *pfse; 4531 struct mbuf *m0; 4532 struct pf_mtag *pf_mtag; 4533 4534 /* ICMP packet rate limitation. */ 4535 switch (af) { 4536 #ifdef INET6 4537 case AF_INET6: 4538 if (icmp6_ratelimit(NULL, type, code)) 4539 return; 4540 break; 4541 #endif /* INET6 */ 4542 #ifdef INET 4543 case AF_INET: 4544 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4545 return; 4546 break; 4547 #endif /* INET */ 4548 } 4549 4550 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4551 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4552 if (pfse == NULL) 4553 return; 4554 4555 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4556 free(pfse, M_PFTEMP); 4557 return; 4558 } 4559 4560 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4561 free(pfse, M_PFTEMP); 4562 return; 4563 } 4564 /* XXX: revisit */ 4565 m0->m_flags |= M_SKIP_FIREWALL; 4566 4567 if (rtableid >= 0) 4568 M_SETFIB(m0, rtableid); 4569 4570 #ifdef ALTQ 4571 if (r->qid) { 4572 pf_mtag->qid = r->qid; 4573 /* add hints for ecn */ 4574 pf_mtag->hdr = mtod(m0, struct ip *); 4575 } 4576 #endif /* ALTQ */ 4577 4578 switch (af) { 4579 #ifdef INET 4580 case AF_INET: 4581 pfse->pfse_type = PFSE_ICMP; 4582 break; 4583 #endif /* INET */ 4584 #ifdef INET6 4585 case AF_INET6: 4586 pfse->pfse_type = PFSE_ICMP6; 4587 break; 4588 #endif /* INET6 */ 4589 } 4590 pfse->pfse_m = m0; 4591 pfse->icmpopts.type = type; 4592 pfse->icmpopts.code = code; 4593 pfse->icmpopts.mtu = mtu; 4594 pf_send(pfse); 4595 } 4596 4597 /* 4598 * Return ((n = 0) == (a = b [with mask m])) 4599 * Note: n != 0 => returns (a != b [with mask m]) 4600 */ 4601 int 4602 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4603 const struct pf_addr *b, sa_family_t af) 4604 { 4605 switch (af) { 4606 #ifdef INET 4607 case AF_INET: 4608 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4609 return (n == 0); 4610 break; 4611 #endif /* INET */ 4612 #ifdef INET6 4613 case AF_INET6: 4614 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4615 return (n == 0); 4616 break; 4617 #endif /* INET6 */ 4618 } 4619 4620 return (n != 0); 4621 } 4622 4623 /* 4624 * Return 1 if b <= a <= e, otherwise return 0. 4625 */ 4626 int 4627 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4628 const struct pf_addr *a, sa_family_t af) 4629 { 4630 switch (af) { 4631 #ifdef INET 4632 case AF_INET: 4633 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4634 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4635 return (0); 4636 break; 4637 #endif /* INET */ 4638 #ifdef INET6 4639 case AF_INET6: { 4640 int i; 4641 4642 /* check a >= b */ 4643 for (i = 0; i < 4; ++i) 4644 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4645 break; 4646 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4647 return (0); 4648 /* check a <= e */ 4649 for (i = 0; i < 4; ++i) 4650 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4651 break; 4652 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4653 return (0); 4654 break; 4655 } 4656 #endif /* INET6 */ 4657 } 4658 return (1); 4659 } 4660 4661 static int 4662 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4663 { 4664 switch (op) { 4665 case PF_OP_IRG: 4666 return ((p > a1) && (p < a2)); 4667 case PF_OP_XRG: 4668 return ((p < a1) || (p > a2)); 4669 case PF_OP_RRG: 4670 return ((p >= a1) && (p <= a2)); 4671 case PF_OP_EQ: 4672 return (p == a1); 4673 case PF_OP_NE: 4674 return (p != a1); 4675 case PF_OP_LT: 4676 return (p < a1); 4677 case PF_OP_LE: 4678 return (p <= a1); 4679 case PF_OP_GT: 4680 return (p > a1); 4681 case PF_OP_GE: 4682 return (p >= a1); 4683 } 4684 return (0); /* never reached */ 4685 } 4686 4687 int 4688 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4689 { 4690 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 4691 } 4692 4693 static int 4694 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4695 { 4696 if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE) 4697 return (0); 4698 return (pf_match(op, a1, a2, u)); 4699 } 4700 4701 static int 4702 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4703 { 4704 if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE) 4705 return (0); 4706 return (pf_match(op, a1, a2, g)); 4707 } 4708 4709 int 4710 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4711 { 4712 if (*tag == -1) 4713 *tag = mtag; 4714 4715 return ((!r->match_tag_not && r->match_tag == *tag) || 4716 (r->match_tag_not && r->match_tag != *tag)); 4717 } 4718 4719 static int 4720 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4721 { 4722 struct ifnet *ifp = m->m_pkthdr.rcvif; 4723 struct pfi_kkif *kif; 4724 4725 if (ifp == NULL) 4726 return (0); 4727 4728 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4729 4730 if (kif == NULL) { 4731 DPFPRINTF(PF_DEBUG_URGENT, 4732 "%s: kif == NULL, @%d via %s", __func__, r->nr, 4733 r->rcv_ifname); 4734 return (0); 4735 } 4736 4737 return (pfi_kkif_match(r->rcv_kif, kif)); 4738 } 4739 4740 int 4741 pf_tag_packet(struct pf_pdesc *pd, int tag) 4742 { 4743 4744 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4745 4746 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4747 return (ENOMEM); 4748 4749 pd->pf_mtag->tag = tag; 4750 4751 return (0); 4752 } 4753 4754 /* 4755 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4756 */ 4757 #define PF_ANCHORSTACK_MATCH 0x00000001 4758 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4759 4760 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4761 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4762 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4763 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4764 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4765 } while (0) 4766 4767 enum pf_test_status 4768 pf_step_into_anchor(struct pf_test_ctx *ctx, struct pf_krule *r, 4769 struct pf_krule_slist *match_rules) 4770 { 4771 enum pf_test_status rv; 4772 4773 PF_RULES_RASSERT(); 4774 4775 if (ctx->depth >= PF_ANCHOR_STACK_MAX) { 4776 printf("%s: anchor stack overflow on %s\n", 4777 __func__, r->anchor->name); 4778 return (PF_TEST_FAIL); 4779 } 4780 4781 ctx->depth++; 4782 4783 if (r->anchor_wildcard) { 4784 struct pf_kanchor *child; 4785 rv = PF_TEST_OK; 4786 RB_FOREACH(child, pf_kanchor_node, &r->anchor->children) { 4787 rv = pf_match_rule(ctx, &child->ruleset, match_rules); 4788 if ((rv == PF_TEST_QUICK) || (rv == PF_TEST_FAIL)) { 4789 /* 4790 * we either hit a rule with quick action 4791 * (more likely), or hit some runtime 4792 * error (e.g. pool_get() failure). 4793 */ 4794 break; 4795 } 4796 } 4797 } else { 4798 rv = pf_match_rule(ctx, &r->anchor->ruleset, match_rules); 4799 /* 4800 * Unless errors occured, stop iff any rule matched 4801 * within quick anchors. 4802 */ 4803 if (rv != PF_TEST_FAIL && r->quick == PF_TEST_QUICK && 4804 *ctx->am == r) 4805 rv = PF_TEST_QUICK; 4806 } 4807 4808 ctx->depth--; 4809 4810 return (rv); 4811 } 4812 4813 struct pf_keth_anchor_stackframe { 4814 struct pf_keth_ruleset *rs; 4815 struct pf_keth_rule *r; /* XXX: + match bit */ 4816 struct pf_keth_anchor *child; 4817 }; 4818 4819 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4820 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4821 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4822 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4823 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4824 } while (0) 4825 4826 void 4827 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4828 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4829 struct pf_keth_rule **a, int *match) 4830 { 4831 struct pf_keth_anchor_stackframe *f; 4832 4833 NET_EPOCH_ASSERT(); 4834 4835 if (match) 4836 *match = 0; 4837 if (*depth >= PF_ANCHOR_STACK_MAX) { 4838 printf("%s: anchor stack overflow on %s\n", 4839 __func__, (*r)->anchor->name); 4840 *r = TAILQ_NEXT(*r, entries); 4841 return; 4842 } else if (*depth == 0 && a != NULL) 4843 *a = *r; 4844 f = stack + (*depth)++; 4845 f->rs = *rs; 4846 f->r = *r; 4847 if ((*r)->anchor_wildcard) { 4848 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4849 4850 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4851 *r = NULL; 4852 return; 4853 } 4854 *rs = &f->child->ruleset; 4855 } else { 4856 f->child = NULL; 4857 *rs = &(*r)->anchor->ruleset; 4858 } 4859 *r = TAILQ_FIRST((*rs)->active.rules); 4860 } 4861 4862 int 4863 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4864 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4865 struct pf_keth_rule **a, int *match) 4866 { 4867 struct pf_keth_anchor_stackframe *f; 4868 struct pf_keth_rule *fr; 4869 int quick = 0; 4870 4871 NET_EPOCH_ASSERT(); 4872 4873 do { 4874 if (*depth <= 0) 4875 break; 4876 f = stack + *depth - 1; 4877 fr = PF_ETH_ANCHOR_RULE(f); 4878 if (f->child != NULL) { 4879 /* 4880 * This block traverses through 4881 * a wildcard anchor. 4882 */ 4883 if (match != NULL && *match) { 4884 /* 4885 * If any of "*" matched, then 4886 * "foo/ *" matched, mark frame 4887 * appropriately. 4888 */ 4889 PF_ETH_ANCHOR_SET_MATCH(f); 4890 *match = 0; 4891 } 4892 f->child = RB_NEXT(pf_keth_anchor_node, 4893 &fr->anchor->children, f->child); 4894 if (f->child != NULL) { 4895 *rs = &f->child->ruleset; 4896 *r = TAILQ_FIRST((*rs)->active.rules); 4897 if (*r == NULL) 4898 continue; 4899 else 4900 break; 4901 } 4902 } 4903 (*depth)--; 4904 if (*depth == 0 && a != NULL) 4905 *a = NULL; 4906 *rs = f->rs; 4907 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4908 quick = fr->quick; 4909 *r = TAILQ_NEXT(fr, entries); 4910 } while (*r == NULL); 4911 4912 return (quick); 4913 } 4914 4915 void 4916 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4917 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4918 { 4919 switch (af) { 4920 #ifdef INET 4921 case AF_INET: 4922 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4923 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4924 break; 4925 #endif /* INET */ 4926 #ifdef INET6 4927 case AF_INET6: 4928 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4929 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4930 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4931 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4932 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4933 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4934 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4935 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4936 break; 4937 #endif /* INET6 */ 4938 } 4939 } 4940 4941 void 4942 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4943 { 4944 switch (af) { 4945 #ifdef INET 4946 case AF_INET: 4947 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4948 break; 4949 #endif /* INET */ 4950 #ifdef INET6 4951 case AF_INET6: 4952 if (addr->addr32[3] == 0xffffffff) { 4953 addr->addr32[3] = 0; 4954 if (addr->addr32[2] == 0xffffffff) { 4955 addr->addr32[2] = 0; 4956 if (addr->addr32[1] == 0xffffffff) { 4957 addr->addr32[1] = 0; 4958 addr->addr32[0] = 4959 htonl(ntohl(addr->addr32[0]) + 1); 4960 } else 4961 addr->addr32[1] = 4962 htonl(ntohl(addr->addr32[1]) + 1); 4963 } else 4964 addr->addr32[2] = 4965 htonl(ntohl(addr->addr32[2]) + 1); 4966 } else 4967 addr->addr32[3] = 4968 htonl(ntohl(addr->addr32[3]) + 1); 4969 break; 4970 #endif /* INET6 */ 4971 } 4972 } 4973 4974 void 4975 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4976 { 4977 /* 4978 * Modern rules use the same flags in rules as they do in states. 4979 */ 4980 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4981 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4982 4983 /* 4984 * Old-style scrub rules have different flags which need to be translated. 4985 */ 4986 if (r->rule_flag & PFRULE_RANDOMID) 4987 a->flags |= PFSTATE_RANDOMID; 4988 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4989 a->flags |= PFSTATE_SETTOS; 4990 a->set_tos = r->set_tos; 4991 } 4992 4993 if (r->qid) 4994 a->qid = r->qid; 4995 if (r->pqid) 4996 a->pqid = r->pqid; 4997 if (r->rtableid >= 0) 4998 a->rtableid = r->rtableid; 4999 a->log |= r->log; 5000 if (r->min_ttl) 5001 a->min_ttl = r->min_ttl; 5002 if (r->max_mss) 5003 a->max_mss = r->max_mss; 5004 if (r->dnpipe) 5005 a->dnpipe = r->dnpipe; 5006 if (r->dnrpipe) 5007 a->dnrpipe = r->dnrpipe; 5008 if (r->dnpipe || r->dnrpipe) { 5009 if (r->free_flags & PFRULE_DN_IS_PIPE) 5010 a->flags |= PFSTATE_DN_IS_PIPE; 5011 else 5012 a->flags &= ~PFSTATE_DN_IS_PIPE; 5013 } 5014 if (r->scrub_flags & PFSTATE_SETPRIO) { 5015 a->set_prio[0] = r->set_prio[0]; 5016 a->set_prio[1] = r->set_prio[1]; 5017 } 5018 if (r->allow_opts) 5019 a->allow_opts = r->allow_opts; 5020 if (r->max_pkt_size) 5021 a->max_pkt_size = r->max_pkt_size; 5022 } 5023 5024 int 5025 pf_socket_lookup(struct pf_pdesc *pd) 5026 { 5027 struct pf_addr *saddr, *daddr; 5028 u_int16_t sport, dport; 5029 struct inpcbinfo *pi; 5030 struct inpcb *inp; 5031 5032 pd->lookup.uid = -1; 5033 pd->lookup.gid = -1; 5034 5035 switch (pd->proto) { 5036 case IPPROTO_TCP: 5037 sport = pd->hdr.tcp.th_sport; 5038 dport = pd->hdr.tcp.th_dport; 5039 pi = &V_tcbinfo; 5040 break; 5041 case IPPROTO_UDP: 5042 sport = pd->hdr.udp.uh_sport; 5043 dport = pd->hdr.udp.uh_dport; 5044 pi = &V_udbinfo; 5045 break; 5046 default: 5047 return (-1); 5048 } 5049 if (pd->dir == PF_IN) { 5050 saddr = pd->src; 5051 daddr = pd->dst; 5052 } else { 5053 u_int16_t p; 5054 5055 p = sport; 5056 sport = dport; 5057 dport = p; 5058 saddr = pd->dst; 5059 daddr = pd->src; 5060 } 5061 switch (pd->af) { 5062 #ifdef INET 5063 case AF_INET: 5064 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 5065 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 5066 if (inp == NULL) { 5067 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 5068 daddr->v4, dport, INPLOOKUP_WILDCARD | 5069 INPLOOKUP_RLOCKPCB, NULL, pd->m); 5070 if (inp == NULL) 5071 return (-1); 5072 } 5073 break; 5074 #endif /* INET */ 5075 #ifdef INET6 5076 case AF_INET6: 5077 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 5078 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 5079 if (inp == NULL) { 5080 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 5081 &daddr->v6, dport, INPLOOKUP_WILDCARD | 5082 INPLOOKUP_RLOCKPCB, NULL, pd->m); 5083 if (inp == NULL) 5084 return (-1); 5085 } 5086 break; 5087 #endif /* INET6 */ 5088 default: 5089 unhandled_af(pd->af); 5090 } 5091 INP_RLOCK_ASSERT(inp); 5092 pd->lookup.uid = inp->inp_cred->cr_uid; 5093 pd->lookup.gid = inp->inp_cred->cr_gid; 5094 INP_RUNLOCK(inp); 5095 5096 return (1); 5097 } 5098 5099 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 5100 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 5101 * 5102 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 5103 */ 5104 uint8_t* 5105 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 5106 u_int8_t min_typelen) 5107 { 5108 uint8_t *eoh = opts + hlen; 5109 5110 if (min_typelen < 2) 5111 return (NULL); 5112 5113 while ((eoh - opt) >= min_typelen) { 5114 switch (*opt) { 5115 case TCPOPT_EOL: 5116 /* FALLTHROUGH - Workaround the failure of some 5117 systems to NOP-pad their bzero'd option buffers, 5118 producing spurious EOLs */ 5119 case TCPOPT_NOP: 5120 opt++; 5121 continue; 5122 default: 5123 if (opt[0] == type && 5124 opt[1] >= min_typelen) 5125 return (opt); 5126 } 5127 5128 opt += MAX(opt[1], 2); /* evade infinite loops */ 5129 } 5130 5131 return (NULL); 5132 } 5133 5134 u_int8_t 5135 pf_get_wscale(struct pf_pdesc *pd) 5136 { 5137 int olen; 5138 uint8_t opts[MAX_TCPOPTLEN], *opt; 5139 uint8_t wscale = 0; 5140 5141 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5142 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 5143 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 5144 return (0); 5145 5146 opt = opts; 5147 while ((opt = pf_find_tcpopt(opt, opts, olen, 5148 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 5149 wscale = opt[2]; 5150 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 5151 wscale |= PF_WSCALE_FLAG; 5152 5153 opt += opt[1]; 5154 } 5155 5156 return (wscale); 5157 } 5158 5159 u_int16_t 5160 pf_get_mss(struct pf_pdesc *pd) 5161 { 5162 int olen; 5163 uint8_t opts[MAX_TCPOPTLEN], *opt; 5164 u_int16_t mss = V_tcp_mssdflt; 5165 5166 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5167 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 5168 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 5169 return (0); 5170 5171 opt = opts; 5172 while ((opt = pf_find_tcpopt(opt, opts, olen, 5173 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 5174 memcpy(&mss, (opt + 2), 2); 5175 mss = ntohs(mss); 5176 opt += opt[1]; 5177 } 5178 5179 return (mss); 5180 } 5181 5182 static u_int16_t 5183 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5184 { 5185 struct nhop_object *nh; 5186 #ifdef INET6 5187 struct in6_addr dst6; 5188 uint32_t scopeid; 5189 #endif /* INET6 */ 5190 int hlen = 0; 5191 uint16_t mss = 0; 5192 5193 NET_EPOCH_ASSERT(); 5194 5195 switch (af) { 5196 #ifdef INET 5197 case AF_INET: 5198 hlen = sizeof(struct ip); 5199 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5200 if (nh != NULL) 5201 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5202 break; 5203 #endif /* INET */ 5204 #ifdef INET6 5205 case AF_INET6: 5206 hlen = sizeof(struct ip6_hdr); 5207 in6_splitscope(&addr->v6, &dst6, &scopeid); 5208 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5209 if (nh != NULL) 5210 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5211 break; 5212 #endif /* INET6 */ 5213 } 5214 5215 mss = max(V_tcp_mssdflt, mss); 5216 mss = min(mss, offer); 5217 mss = max(mss, 64); /* sanity - at least max opt space */ 5218 return (mss); 5219 } 5220 5221 static u_int32_t 5222 pf_tcp_iss(struct pf_pdesc *pd) 5223 { 5224 SHA512_CTX ctx; 5225 union { 5226 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5227 uint32_t words[1]; 5228 } digest; 5229 5230 if (V_pf_tcp_secret_init == 0) { 5231 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5232 SHA512_Init(&V_pf_tcp_secret_ctx); 5233 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5234 sizeof(V_pf_tcp_secret)); 5235 V_pf_tcp_secret_init = 1; 5236 } 5237 5238 ctx = V_pf_tcp_secret_ctx; 5239 5240 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5241 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5242 switch (pd->af) { 5243 case AF_INET6: 5244 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5245 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5246 break; 5247 case AF_INET: 5248 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5249 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5250 break; 5251 } 5252 SHA512_Final(digest.bytes, &ctx); 5253 V_pf_tcp_iss_off += 4096; 5254 #define ISN_RANDOM_INCREMENT (4096 - 1) 5255 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5256 V_pf_tcp_iss_off); 5257 #undef ISN_RANDOM_INCREMENT 5258 } 5259 5260 static bool 5261 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5262 { 5263 bool match = true; 5264 5265 /* Always matches if not set */ 5266 if (! r->isset) 5267 return (!r->neg); 5268 5269 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5270 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5271 match = false; 5272 break; 5273 } 5274 } 5275 5276 return (match ^ r->neg); 5277 } 5278 5279 static int 5280 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5281 { 5282 if (*tag == -1) 5283 *tag = mtag; 5284 5285 return ((!r->match_tag_not && r->match_tag == *tag) || 5286 (r->match_tag_not && r->match_tag != *tag)); 5287 } 5288 5289 static void 5290 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5291 { 5292 /* If we don't have the interface drop the packet. */ 5293 if (ifp == NULL) { 5294 m_freem(m); 5295 return; 5296 } 5297 5298 switch (ifp->if_type) { 5299 case IFT_ETHER: 5300 case IFT_XETHER: 5301 case IFT_L2VLAN: 5302 case IFT_BRIDGE: 5303 case IFT_IEEE8023ADLAG: 5304 break; 5305 default: 5306 m_freem(m); 5307 return; 5308 } 5309 5310 ifp->if_transmit(ifp, m); 5311 } 5312 5313 static int 5314 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5315 { 5316 #ifdef INET 5317 struct ip ip; 5318 #endif /* INET */ 5319 #ifdef INET6 5320 struct ip6_hdr ip6; 5321 #endif /* INET6 */ 5322 struct mbuf *m = *m0; 5323 struct ether_header *e; 5324 struct pf_keth_rule *r, *rm, *a = NULL; 5325 struct pf_keth_ruleset *ruleset = NULL; 5326 struct pf_mtag *mtag; 5327 struct pf_keth_ruleq *rules; 5328 struct pf_addr *src = NULL, *dst = NULL; 5329 struct pfi_kkif *bridge_to; 5330 sa_family_t af = 0; 5331 uint16_t proto; 5332 int asd = 0, match = 0; 5333 int tag = -1; 5334 uint8_t action; 5335 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACK_MAX]; 5336 5337 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5338 NET_EPOCH_ASSERT(); 5339 5340 PF_RULES_RLOCK_TRACKER; 5341 5342 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5343 5344 mtag = pf_find_mtag(m); 5345 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5346 /* Dummynet re-injects packets after they've 5347 * completed their delay. We've already 5348 * processed them, so pass unconditionally. */ 5349 5350 /* But only once. We may see the packet multiple times (e.g. 5351 * PFIL_IN/PFIL_OUT). */ 5352 pf_dummynet_flag_remove(m, mtag); 5353 5354 return (PF_PASS); 5355 } 5356 5357 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5358 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5359 DPFPRINTF(PF_DEBUG_URGENT, 5360 "%s: m_len < sizeof(struct ether_header)" 5361 ", pullup failed", __func__); 5362 return (PF_DROP); 5363 } 5364 e = mtod(m, struct ether_header *); 5365 proto = ntohs(e->ether_type); 5366 5367 switch (proto) { 5368 #ifdef INET 5369 case ETHERTYPE_IP: { 5370 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5371 sizeof(ip))) 5372 return (PF_DROP); 5373 5374 af = AF_INET; 5375 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5376 (caddr_t)&ip); 5377 src = (struct pf_addr *)&ip.ip_src; 5378 dst = (struct pf_addr *)&ip.ip_dst; 5379 break; 5380 } 5381 #endif /* INET */ 5382 #ifdef INET6 5383 case ETHERTYPE_IPV6: { 5384 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5385 sizeof(ip6))) 5386 return (PF_DROP); 5387 5388 af = AF_INET6; 5389 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5390 (caddr_t)&ip6); 5391 src = (struct pf_addr *)&ip6.ip6_src; 5392 dst = (struct pf_addr *)&ip6.ip6_dst; 5393 break; 5394 } 5395 #endif /* INET6 */ 5396 } 5397 5398 PF_RULES_RLOCK(); 5399 5400 ruleset = V_pf_keth; 5401 rules = atomic_load_ptr(&ruleset->active.rules); 5402 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5403 counter_u64_add(r->evaluations, 1); 5404 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5405 5406 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5407 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5408 "kif"); 5409 r = r->skip[PFE_SKIP_IFP].ptr; 5410 } 5411 else if (r->direction && r->direction != dir) { 5412 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5413 "dir"); 5414 r = r->skip[PFE_SKIP_DIR].ptr; 5415 } 5416 else if (r->proto && r->proto != proto) { 5417 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5418 "proto"); 5419 r = r->skip[PFE_SKIP_PROTO].ptr; 5420 } 5421 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5422 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5423 "src"); 5424 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5425 } 5426 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5427 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5428 "dst"); 5429 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5430 } 5431 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5432 r->ipsrc.neg, kif, M_GETFIB(m))) { 5433 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5434 "ip_src"); 5435 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5436 } 5437 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5438 r->ipdst.neg, kif, M_GETFIB(m))) { 5439 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5440 "ip_dst"); 5441 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5442 } 5443 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5444 mtag ? mtag->tag : 0)) { 5445 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5446 "match_tag"); 5447 r = TAILQ_NEXT(r, entries); 5448 } 5449 else { 5450 if (r->tag) 5451 tag = r->tag; 5452 if (r->anchor == NULL) { 5453 /* Rule matches */ 5454 rm = r; 5455 5456 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5457 5458 if (r->quick) 5459 break; 5460 5461 r = TAILQ_NEXT(r, entries); 5462 } else { 5463 pf_step_into_keth_anchor(anchor_stack, &asd, 5464 &ruleset, &r, &a, &match); 5465 } 5466 } 5467 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5468 &ruleset, &r, &a, &match)) 5469 break; 5470 } 5471 5472 r = rm; 5473 5474 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5475 5476 /* Default to pass. */ 5477 if (r == NULL) { 5478 PF_RULES_RUNLOCK(); 5479 return (PF_PASS); 5480 } 5481 5482 /* Execute action. */ 5483 counter_u64_add(r->packets[dir == PF_OUT], 1); 5484 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5485 pf_update_timestamp(r); 5486 5487 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5488 if (r->action == PF_DROP) { 5489 PF_RULES_RUNLOCK(); 5490 return (PF_DROP); 5491 } 5492 5493 if (tag > 0) { 5494 if (mtag == NULL) 5495 mtag = pf_get_mtag(m); 5496 if (mtag == NULL) { 5497 PF_RULES_RUNLOCK(); 5498 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5499 return (PF_DROP); 5500 } 5501 mtag->tag = tag; 5502 } 5503 5504 if (r->qid != 0) { 5505 if (mtag == NULL) 5506 mtag = pf_get_mtag(m); 5507 if (mtag == NULL) { 5508 PF_RULES_RUNLOCK(); 5509 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5510 return (PF_DROP); 5511 } 5512 mtag->qid = r->qid; 5513 } 5514 5515 action = r->action; 5516 bridge_to = r->bridge_to; 5517 5518 /* Dummynet */ 5519 if (r->dnpipe) { 5520 struct ip_fw_args dnflow; 5521 5522 /* Drop packet if dummynet is not loaded. */ 5523 if (ip_dn_io_ptr == NULL) { 5524 PF_RULES_RUNLOCK(); 5525 m_freem(m); 5526 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5527 return (PF_DROP); 5528 } 5529 if (mtag == NULL) 5530 mtag = pf_get_mtag(m); 5531 if (mtag == NULL) { 5532 PF_RULES_RUNLOCK(); 5533 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5534 return (PF_DROP); 5535 } 5536 5537 bzero(&dnflow, sizeof(dnflow)); 5538 5539 /* We don't have port numbers here, so we set 0. That means 5540 * that we'll be somewhat limited in distinguishing flows (i.e. 5541 * only based on IP addresses, not based on port numbers), but 5542 * it's better than nothing. */ 5543 dnflow.f_id.dst_port = 0; 5544 dnflow.f_id.src_port = 0; 5545 dnflow.f_id.proto = 0; 5546 5547 dnflow.rule.info = r->dnpipe; 5548 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5549 if (r->dnflags & PFRULE_DN_IS_PIPE) 5550 dnflow.rule.info |= IPFW_IS_PIPE; 5551 5552 dnflow.f_id.extra = dnflow.rule.info; 5553 5554 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5555 dnflow.flags |= IPFW_ARGS_ETHER; 5556 dnflow.ifp = kif->pfik_ifp; 5557 5558 switch (af) { 5559 case AF_INET: 5560 dnflow.f_id.addr_type = 4; 5561 dnflow.f_id.src_ip = src->v4.s_addr; 5562 dnflow.f_id.dst_ip = dst->v4.s_addr; 5563 break; 5564 case AF_INET6: 5565 dnflow.flags |= IPFW_ARGS_IP6; 5566 dnflow.f_id.addr_type = 6; 5567 dnflow.f_id.src_ip6 = src->v6; 5568 dnflow.f_id.dst_ip6 = dst->v6; 5569 break; 5570 } 5571 5572 PF_RULES_RUNLOCK(); 5573 5574 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5575 ip_dn_io_ptr(m0, &dnflow); 5576 if (*m0 != NULL) 5577 pf_dummynet_flag_remove(m, mtag); 5578 } else { 5579 PF_RULES_RUNLOCK(); 5580 } 5581 5582 if (action == PF_PASS && bridge_to) { 5583 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5584 *m0 = NULL; /* We've eaten the packet. */ 5585 } 5586 5587 return (action); 5588 } 5589 5590 #define PF_TEST_ATTRIB(t, a) \ 5591 if (t) { \ 5592 r = a; \ 5593 continue; \ 5594 } else do { \ 5595 } while (0) 5596 5597 static __inline u_short 5598 pf_rule_apply_nat(struct pf_test_ctx *ctx, struct pf_krule *r) 5599 { 5600 struct pf_pdesc *pd = ctx->pd; 5601 u_short transerror; 5602 u_int8_t nat_action; 5603 5604 if (r->rule_flag & PFRULE_AFTO) { 5605 /* Don't translate if there was an old style NAT rule */ 5606 if (ctx->nr != NULL) 5607 return (PFRES_TRANSLATE); 5608 5609 /* pass af-to rules, unsupported on match rules */ 5610 KASSERT(r->action != PF_MATCH, ("%s: af-to on match rule", __func__)); 5611 /* XXX I can imagine scenarios where we have both NAT and RDR source tracking */ 5612 ctx->nat_pool = &(r->nat); 5613 ctx->nr = r; 5614 pd->naf = r->naf; 5615 if (pf_get_transaddr_af(ctx->nr, pd) == -1) { 5616 return (PFRES_TRANSLATE); 5617 } 5618 return (PFRES_MATCH); 5619 } else if (r->rdr.cur || r->nat.cur) { 5620 /* Don't translate if there was an old style NAT rule */ 5621 if (ctx->nr != NULL) 5622 return (PFRES_TRANSLATE); 5623 5624 /* match/pass nat-to/rdr-to rules */ 5625 ctx->nr = r; 5626 if (r->nat.cur) { 5627 nat_action = PF_NAT; 5628 ctx->nat_pool = &(r->nat); 5629 } else { 5630 nat_action = PF_RDR; 5631 ctx->nat_pool = &(r->rdr); 5632 } 5633 5634 transerror = pf_get_transaddr(ctx, ctx->nr, 5635 nat_action, ctx->nat_pool); 5636 if (transerror == PFRES_MATCH) { 5637 ctx->rewrite += pf_translate_compat(ctx); 5638 return(PFRES_MATCH); 5639 } 5640 return (transerror); 5641 } 5642 5643 return (PFRES_MAX); 5644 } 5645 5646 enum pf_test_status 5647 pf_match_rule(struct pf_test_ctx *ctx, struct pf_kruleset *ruleset, 5648 struct pf_krule_slist *match_rules) 5649 { 5650 struct pf_krule_item *ri, *rt; 5651 struct pf_krule *r; 5652 struct pf_krule *save_a; 5653 struct pf_kruleset *save_aruleset; 5654 struct pf_pdesc *pd = ctx->pd; 5655 u_short transerror; 5656 5657 r = TAILQ_FIRST(ruleset->rules[PF_RULESET_FILTER].active.ptr); 5658 while (r != NULL) { 5659 if (ctx->pd->related_rule) { 5660 *ctx->rm = ctx->pd->related_rule; 5661 break; 5662 } 5663 PF_TEST_ATTRIB(r->rule_flag & PFRULE_EXPIRED, 5664 TAILQ_NEXT(r, entries)); 5665 /* Don't count expired rule evaluations. */ 5666 pf_counter_u64_add(&r->evaluations, 1); 5667 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5668 r->skip[PF_SKIP_IFP]); 5669 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5670 r->skip[PF_SKIP_DIR]); 5671 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5672 r->skip[PF_SKIP_AF]); 5673 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5674 r->skip[PF_SKIP_PROTO]); 5675 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5676 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5677 r->skip[PF_SKIP_SRC_ADDR]); 5678 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5679 r->dst.neg, NULL, M_GETFIB(pd->m)), 5680 r->skip[PF_SKIP_DST_ADDR]); 5681 switch (pd->virtual_proto) { 5682 case PF_VPROTO_FRAGMENT: 5683 /* tcp/udp only. port_op always 0 in other cases */ 5684 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5685 TAILQ_NEXT(r, entries)); 5686 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5687 TAILQ_NEXT(r, entries)); 5688 /* icmp only. type/code always 0 in other cases */ 5689 PF_TEST_ATTRIB((r->type || r->code), 5690 TAILQ_NEXT(r, entries)); 5691 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5692 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5693 TAILQ_NEXT(r, entries)); 5694 break; 5695 5696 case IPPROTO_TCP: 5697 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(ctx->th)) 5698 != r->flags, 5699 TAILQ_NEXT(r, entries)); 5700 /* FALLTHROUGH */ 5701 case IPPROTO_SCTP: 5702 case IPPROTO_UDP: 5703 /* tcp/udp only. port_op always 0 in other cases */ 5704 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5705 r->src.port[0], r->src.port[1], pd->nsport), 5706 r->skip[PF_SKIP_SRC_PORT]); 5707 /* tcp/udp only. port_op always 0 in other cases */ 5708 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5709 r->dst.port[0], r->dst.port[1], pd->ndport), 5710 r->skip[PF_SKIP_DST_PORT]); 5711 /* tcp/udp only. uid.op always 0 in other cases */ 5712 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5713 pf_socket_lookup(pd), 1)) && 5714 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5715 pd->lookup.uid), 5716 TAILQ_NEXT(r, entries)); 5717 /* tcp/udp only. gid.op always 0 in other cases */ 5718 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5719 pf_socket_lookup(pd), 1)) && 5720 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5721 pd->lookup.gid), 5722 TAILQ_NEXT(r, entries)); 5723 break; 5724 5725 case IPPROTO_ICMP: 5726 case IPPROTO_ICMPV6: 5727 /* icmp only. type always 0 in other cases */ 5728 PF_TEST_ATTRIB(r->type && r->type != ctx->icmptype + 1, 5729 TAILQ_NEXT(r, entries)); 5730 /* icmp only. type always 0 in other cases */ 5731 PF_TEST_ATTRIB(r->code && r->code != ctx->icmpcode + 1, 5732 TAILQ_NEXT(r, entries)); 5733 break; 5734 5735 default: 5736 break; 5737 } 5738 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5739 TAILQ_NEXT(r, entries)); 5740 PF_TEST_ATTRIB(r->prio && 5741 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5742 TAILQ_NEXT(r, entries)); 5743 PF_TEST_ATTRIB(r->prob && 5744 r->prob <= arc4random(), 5745 TAILQ_NEXT(r, entries)); 5746 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, 5747 &ctx->tag, pd->pf_mtag ? pd->pf_mtag->tag : 0), 5748 TAILQ_NEXT(r, entries)); 5749 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 5750 r->rcvifnot), 5751 TAILQ_NEXT(r, entries)); 5752 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5753 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5754 TAILQ_NEXT(r, entries)); 5755 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5756 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5757 pf_osfp_fingerprint(pd, ctx->th), 5758 r->os_fingerprint)), 5759 TAILQ_NEXT(r, entries)); 5760 /* must be last! */ 5761 if (r->pktrate.limit) { 5762 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 5763 TAILQ_NEXT(r, entries)); 5764 } 5765 /* FALLTHROUGH */ 5766 if (r->tag) 5767 ctx->tag = r->tag; 5768 if (r->anchor == NULL) { 5769 5770 if (r->rule_flag & PFRULE_ONCE) { 5771 uint32_t rule_flag; 5772 5773 rule_flag = r->rule_flag; 5774 if ((rule_flag & PFRULE_EXPIRED) == 0 && 5775 atomic_cmpset_int(&r->rule_flag, rule_flag, 5776 rule_flag | PFRULE_EXPIRED)) { 5777 r->exptime = time_uptime; 5778 } else { 5779 r = TAILQ_NEXT(r, entries); 5780 continue; 5781 } 5782 } 5783 5784 if (r->action == PF_MATCH) { 5785 /* 5786 * Apply translations before increasing counters, 5787 * in case it fails. 5788 */ 5789 transerror = pf_rule_apply_nat(ctx, r); 5790 switch (transerror) { 5791 case PFRES_MATCH: 5792 /* Translation action found in rule and applied successfully */ 5793 case PFRES_MAX: 5794 /* No translation action found in rule */ 5795 break; 5796 default: 5797 /* Translation action found in rule but failed to apply */ 5798 REASON_SET(&ctx->reason, transerror); 5799 return (PF_TEST_FAIL); 5800 } 5801 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5802 if (ri == NULL) { 5803 REASON_SET(&ctx->reason, PFRES_MEMORY); 5804 return (PF_TEST_FAIL); 5805 } 5806 ri->r = r; 5807 5808 if (SLIST_EMPTY(match_rules)) { 5809 SLIST_INSERT_HEAD(match_rules, ri, entry); 5810 } else { 5811 SLIST_INSERT_AFTER(rt, ri, entry); 5812 } 5813 rt = ri; 5814 5815 pf_rule_to_actions(r, &pd->act); 5816 if (r->log) 5817 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5818 ctx->a, ruleset, pd, 1, NULL); 5819 } else { 5820 /* 5821 * found matching r 5822 */ 5823 *ctx->rm = r; 5824 /* 5825 * anchor, with ruleset, where r belongs to 5826 */ 5827 *ctx->am = ctx->a; 5828 /* 5829 * ruleset where r belongs to 5830 */ 5831 *ctx->rsm = ruleset; 5832 /* 5833 * ruleset, where anchor belongs to. 5834 */ 5835 ctx->arsm = ctx->aruleset; 5836 } 5837 if (pd->act.log & PF_LOG_MATCHES) 5838 pf_log_matches(pd, r, ctx->a, ruleset, match_rules); 5839 if (r->quick) { 5840 ctx->test_status = PF_TEST_QUICK; 5841 break; 5842 } 5843 } else { 5844 save_a = ctx->a; 5845 save_aruleset = ctx->aruleset; 5846 5847 ctx->a = r; /* remember anchor */ 5848 ctx->aruleset = ruleset; /* and its ruleset */ 5849 if (ctx->a->quick) 5850 ctx->test_status = PF_TEST_QUICK; 5851 /* 5852 * Note: we don't need to restore if we are not going 5853 * to continue with ruleset evaluation. 5854 */ 5855 if (pf_step_into_anchor(ctx, r, match_rules) != PF_TEST_OK) { 5856 break; 5857 } 5858 ctx->a = save_a; 5859 ctx->aruleset = save_aruleset; 5860 } 5861 r = TAILQ_NEXT(r, entries); 5862 } 5863 5864 5865 return (ctx->test_status); 5866 } 5867 5868 static int 5869 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5870 struct pf_pdesc *pd, struct pf_krule **am, 5871 struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp, 5872 struct pf_krule_slist *match_rules) 5873 { 5874 struct pf_krule *r = NULL; 5875 struct pf_kruleset *ruleset = NULL; 5876 struct pf_test_ctx ctx; 5877 u_short transerror; 5878 int action = PF_PASS; 5879 u_int16_t bproto_sum = 0, bip_sum = 0; 5880 enum pf_test_status rv; 5881 5882 PF_RULES_RASSERT(); 5883 5884 bzero(&ctx, sizeof(ctx)); 5885 ctx.tag = -1; 5886 ctx.pd = pd; 5887 ctx.rm = rm; 5888 ctx.am = am; 5889 ctx.rsm = rsm; 5890 ctx.th = &pd->hdr.tcp; 5891 ctx.reason = *reason; 5892 5893 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 5894 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 5895 5896 if (inp != NULL) { 5897 INP_LOCK_ASSERT(inp); 5898 pd->lookup.uid = inp->inp_cred->cr_uid; 5899 pd->lookup.gid = inp->inp_cred->cr_gid; 5900 pd->lookup.done = 1; 5901 } 5902 5903 if (pd->ip_sum) 5904 bip_sum = *pd->ip_sum; 5905 5906 switch (pd->virtual_proto) { 5907 case IPPROTO_TCP: 5908 bproto_sum = ctx.th->th_sum; 5909 pd->nsport = ctx.th->th_sport; 5910 pd->ndport = ctx.th->th_dport; 5911 break; 5912 case IPPROTO_UDP: 5913 bproto_sum = pd->hdr.udp.uh_sum; 5914 pd->nsport = pd->hdr.udp.uh_sport; 5915 pd->ndport = pd->hdr.udp.uh_dport; 5916 break; 5917 case IPPROTO_SCTP: 5918 pd->nsport = pd->hdr.sctp.src_port; 5919 pd->ndport = pd->hdr.sctp.dest_port; 5920 break; 5921 #ifdef INET 5922 case IPPROTO_ICMP: 5923 MPASS(pd->af == AF_INET); 5924 ctx.icmptype = pd->hdr.icmp.icmp_type; 5925 ctx.icmpcode = pd->hdr.icmp.icmp_code; 5926 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 5927 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 5928 if (ctx.icmp_dir == PF_IN) { 5929 pd->nsport = ctx.virtual_id; 5930 pd->ndport = ctx.virtual_type; 5931 } else { 5932 pd->nsport = ctx.virtual_type; 5933 pd->ndport = ctx.virtual_id; 5934 } 5935 break; 5936 #endif /* INET */ 5937 #ifdef INET6 5938 case IPPROTO_ICMPV6: 5939 MPASS(pd->af == AF_INET6); 5940 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 5941 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 5942 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 5943 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 5944 if (ctx.icmp_dir == PF_IN) { 5945 pd->nsport = ctx.virtual_id; 5946 pd->ndport = ctx.virtual_type; 5947 } else { 5948 pd->nsport = ctx.virtual_type; 5949 pd->ndport = ctx.virtual_id; 5950 } 5951 5952 break; 5953 #endif /* INET6 */ 5954 default: 5955 pd->nsport = pd->ndport = 0; 5956 break; 5957 } 5958 pd->osport = pd->nsport; 5959 pd->odport = pd->ndport; 5960 5961 /* check packet for BINAT/NAT/RDR */ 5962 transerror = pf_get_translation(&ctx); 5963 switch (transerror) { 5964 default: 5965 /* A translation error occurred. */ 5966 REASON_SET(&ctx.reason, transerror); 5967 goto cleanup; 5968 case PFRES_MAX: 5969 /* No match. */ 5970 break; 5971 case PFRES_MATCH: 5972 KASSERT(ctx.sk != NULL, ("%s: null sk", __func__)); 5973 KASSERT(ctx.nk != NULL, ("%s: null nk", __func__)); 5974 if (ctx.nr->log) { 5975 PFLOG_PACKET(ctx.nr->action, PFRES_MATCH, ctx.nr, ctx.a, 5976 ruleset, pd, 1, NULL); 5977 } 5978 5979 ctx.rewrite += pf_translate_compat(&ctx); 5980 ctx.nat_pool = &(ctx.nr->rdr); 5981 } 5982 5983 *ctx.rm = &V_pf_default_rule; 5984 if (ctx.nr && ctx.nr->natpass) { 5985 r = ctx.nr; 5986 ruleset = *ctx.rsm; 5987 } else { 5988 ruleset = &pf_main_ruleset; 5989 rv = pf_match_rule(&ctx, ruleset, match_rules); 5990 if (rv == PF_TEST_FAIL) { 5991 /* 5992 * Reason has been set in pf_match_rule() already. 5993 */ 5994 goto cleanup; 5995 } 5996 5997 r = *ctx.rm; /* matching rule */ 5998 ctx.a = *ctx.am; /* rule that defines an anchor containing 'r' */ 5999 ruleset = *ctx.rsm; /* ruleset of the anchor defined by the rule 'a' */ 6000 ctx.aruleset = ctx.arsm; /* ruleset of the 'a' rule itself */ 6001 6002 /* apply actions for last matching pass/block rule */ 6003 pf_rule_to_actions(r, &pd->act); 6004 transerror = pf_rule_apply_nat(&ctx, r); 6005 switch (transerror) { 6006 case PFRES_MATCH: 6007 /* Translation action found in rule and applied successfully */ 6008 case PFRES_MAX: 6009 /* No translation action found in rule */ 6010 break; 6011 default: 6012 /* Translation action found in rule but failed to apply */ 6013 REASON_SET(&ctx.reason, transerror); 6014 goto cleanup; 6015 } 6016 } 6017 6018 REASON_SET(&ctx.reason, PFRES_MATCH); 6019 6020 if (r->log) { 6021 if (ctx.rewrite) 6022 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6023 PFLOG_PACKET(r->action, ctx.reason, r, ctx.a, ruleset, pd, 1, NULL); 6024 } 6025 if (pd->act.log & PF_LOG_MATCHES) 6026 pf_log_matches(pd, r, ctx.a, ruleset, match_rules); 6027 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 6028 (r->action == PF_DROP) && 6029 ((r->rule_flag & PFRULE_RETURNRST) || 6030 (r->rule_flag & PFRULE_RETURNICMP) || 6031 (r->rule_flag & PFRULE_RETURN))) { 6032 pf_return(r, ctx.nr, pd, ctx.th, bproto_sum, 6033 bip_sum, &ctx.reason, r->rtableid); 6034 } 6035 6036 if (r->action == PF_DROP) 6037 goto cleanup; 6038 6039 if (ctx.tag > 0 && pf_tag_packet(pd, ctx.tag)) { 6040 REASON_SET(&ctx.reason, PFRES_MEMORY); 6041 goto cleanup; 6042 } 6043 if (pd->act.rtableid >= 0) 6044 M_SETFIB(pd->m, pd->act.rtableid); 6045 6046 if (r->rt) { 6047 /* 6048 * Set act.rt here instead of in pf_rule_to_actions() because 6049 * it is applied only from the last pass rule. For rules 6050 * with the prefer-ipv6-nexthop option act.rt_af is a hint 6051 * about AF of the forwarded packet and might be changed. 6052 */ 6053 pd->act.rt = r->rt; 6054 if (r->rt == PF_REPLYTO) 6055 pd->act.rt_af = pd->af; 6056 else 6057 pd->act.rt_af = pd->naf; 6058 if ((transerror = pf_map_addr_sn(pd->af, r, pd->src, 6059 &pd->act.rt_addr, &pd->act.rt_af, &pd->act.rt_kif, NULL, 6060 &(r->route), PF_SN_ROUTE)) != PFRES_MATCH) { 6061 REASON_SET(&ctx.reason, transerror); 6062 goto cleanup; 6063 } 6064 } 6065 6066 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 6067 (!ctx.state_icmp && (r->keep_state || ctx.nr != NULL || 6068 (pd->flags & PFDESC_TCP_NORM)))) { 6069 bool nat64; 6070 6071 action = pf_create_state(r, &ctx, sm, bproto_sum, bip_sum, 6072 match_rules); 6073 ctx.sk = ctx.nk = NULL; 6074 if (action != PF_PASS) { 6075 pf_udp_mapping_release(ctx.udp_mapping); 6076 if (r->log || (ctx.nr != NULL && ctx.nr->log) || 6077 ctx.reason == PFRES_MEMORY) 6078 pd->act.log |= PF_LOG_FORCE; 6079 if (action == PF_DROP && 6080 (r->rule_flag & PFRULE_RETURN)) 6081 pf_return(r, ctx.nr, pd, ctx.th, 6082 bproto_sum, bip_sum, &ctx.reason, 6083 pd->act.rtableid); 6084 *reason = ctx.reason; 6085 return (action); 6086 } 6087 6088 nat64 = pd->af != pd->naf; 6089 if (nat64) { 6090 int ret; 6091 6092 if (ctx.sk == NULL) 6093 ctx.sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 6094 if (ctx.nk == NULL) 6095 ctx.nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 6096 6097 if (pd->dir == PF_IN) { 6098 ret = pf_translate(pd, &ctx.sk->addr[pd->didx], 6099 ctx.sk->port[pd->didx], &ctx.sk->addr[pd->sidx], 6100 ctx.sk->port[pd->sidx], ctx.virtual_type, 6101 ctx.icmp_dir); 6102 } else { 6103 ret = pf_translate(pd, &ctx.sk->addr[pd->sidx], 6104 ctx.sk->port[pd->sidx], &ctx.sk->addr[pd->didx], 6105 ctx.sk->port[pd->didx], ctx.virtual_type, 6106 ctx.icmp_dir); 6107 } 6108 6109 if (ret < 0) 6110 goto cleanup; 6111 6112 ctx.rewrite += ret; 6113 6114 if (ctx.rewrite && ctx.sk->af != ctx.nk->af) 6115 action = PF_AFRT; 6116 } 6117 } else { 6118 uma_zfree(V_pf_state_key_z, ctx.sk); 6119 uma_zfree(V_pf_state_key_z, ctx.nk); 6120 ctx.sk = ctx.nk = NULL; 6121 pf_udp_mapping_release(ctx.udp_mapping); 6122 } 6123 6124 /* copy back packet headers if we performed NAT operations */ 6125 if (ctx.rewrite) 6126 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6127 6128 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 6129 pd->dir == PF_OUT && 6130 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) { 6131 /* 6132 * We want the state created, but we dont 6133 * want to send this in case a partner 6134 * firewall has to know about it to allow 6135 * replies through it. 6136 */ 6137 *reason = ctx.reason; 6138 return (PF_DEFER); 6139 } 6140 6141 *reason = ctx.reason; 6142 return (action); 6143 6144 cleanup: 6145 uma_zfree(V_pf_state_key_z, ctx.sk); 6146 uma_zfree(V_pf_state_key_z, ctx.nk); 6147 pf_udp_mapping_release(ctx.udp_mapping); 6148 *reason = ctx.reason; 6149 6150 return (PF_DROP); 6151 } 6152 6153 static int 6154 pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx, 6155 struct pf_kstate **sm, u_int16_t bproto_sum, u_int16_t bip_sum, 6156 struct pf_krule_slist *match_rules) 6157 { 6158 struct pf_pdesc *pd = ctx->pd; 6159 struct pf_kstate *s = NULL; 6160 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6161 /* 6162 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6163 * but for PF_SN_NAT it is different. Don't try optimizing it, 6164 * just store all 3 hashes. 6165 */ 6166 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6167 struct tcphdr *th = &pd->hdr.tcp; 6168 u_int16_t mss = V_tcp_mssdflt; 6169 u_short sn_reason; 6170 6171 /* check maximums */ 6172 if (r->max_states && 6173 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6174 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6175 REASON_SET(&ctx->reason, PFRES_MAXSTATES); 6176 goto csfailed; 6177 } 6178 /* src node for limits */ 6179 if ((r->rule_flag & PFRULE_SRCTRACK) && 6180 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6181 NULL, NULL, pd->af, PF_SN_LIMIT)) != 0) { 6182 REASON_SET(&ctx->reason, sn_reason); 6183 goto csfailed; 6184 } 6185 /* src node for route-to rule */ 6186 if (r->rt) { 6187 if ((r->route.opts & PF_POOL_STICKYADDR) && 6188 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, 6189 pd->af, &pd->act.rt_addr, pd->act.rt_kif, pd->act.rt_af, 6190 PF_SN_ROUTE)) != 0) { 6191 REASON_SET(&ctx->reason, sn_reason); 6192 goto csfailed; 6193 } 6194 } 6195 /* src node for translation rule */ 6196 if (ctx->nr != NULL) { 6197 KASSERT(ctx->nat_pool != NULL, ("%s: nat_pool is NULL", __func__)); 6198 /* 6199 * The NAT addresses are chosen during ruleset parsing. 6200 * The new afto code stores post-nat addresses in nsaddr. 6201 * The old nat code (also used for new nat-to rules) creates 6202 * state keys and stores addresses in them. 6203 */ 6204 if ((ctx->nat_pool->opts & PF_POOL_STICKYADDR) && 6205 (sn_reason = pf_insert_src_node(sns, snhs, ctx->nr, 6206 ctx->sk ? &(ctx->sk->addr[pd->sidx]) : pd->src, pd->af, 6207 ctx->nk ? &(ctx->nk->addr[1]) : &(pd->nsaddr), NULL, 6208 pd->naf, PF_SN_NAT)) != 0 ) { 6209 REASON_SET(&ctx->reason, sn_reason); 6210 goto csfailed; 6211 } 6212 } 6213 s = pf_alloc_state(M_NOWAIT); 6214 if (s == NULL) { 6215 REASON_SET(&ctx->reason, PFRES_MEMORY); 6216 goto csfailed; 6217 } 6218 s->rule = r; 6219 s->nat_rule = ctx->nr; 6220 s->anchor = ctx->a; 6221 s->match_rules = *match_rules; 6222 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6223 6224 if (pd->act.allow_opts) 6225 s->state_flags |= PFSTATE_ALLOWOPTS; 6226 if (r->rule_flag & PFRULE_STATESLOPPY) 6227 s->state_flags |= PFSTATE_SLOPPY; 6228 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6229 s->state_flags |= PFSTATE_SCRUB_TCP; 6230 if ((r->rule_flag & PFRULE_PFLOW) || 6231 (ctx->nr != NULL && ctx->nr->rule_flag & PFRULE_PFLOW)) 6232 s->state_flags |= PFSTATE_PFLOW; 6233 6234 s->act.log = pd->act.log & PF_LOG_ALL; 6235 s->sync_state = PFSYNC_S_NONE; 6236 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6237 6238 if (ctx->nr != NULL) 6239 s->act.log |= ctx->nr->log & PF_LOG_ALL; 6240 switch (pd->proto) { 6241 case IPPROTO_TCP: 6242 s->src.seqlo = ntohl(th->th_seq); 6243 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6244 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6245 r->keep_state == PF_STATE_MODULATE) { 6246 /* Generate sequence number modulator */ 6247 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6248 0) 6249 s->src.seqdiff = 1; 6250 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6251 htonl(s->src.seqlo + s->src.seqdiff), 0); 6252 ctx->rewrite = 1; 6253 } else 6254 s->src.seqdiff = 0; 6255 if (tcp_get_flags(th) & TH_SYN) { 6256 s->src.seqhi++; 6257 s->src.wscale = pf_get_wscale(pd); 6258 } 6259 s->src.max_win = MAX(ntohs(th->th_win), 1); 6260 if (s->src.wscale & PF_WSCALE_MASK) { 6261 /* Remove scale factor from initial window */ 6262 int win = s->src.max_win; 6263 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6264 s->src.max_win = (win - 1) >> 6265 (s->src.wscale & PF_WSCALE_MASK); 6266 } 6267 if (tcp_get_flags(th) & TH_FIN) 6268 s->src.seqhi++; 6269 s->dst.seqhi = 1; 6270 s->dst.max_win = 1; 6271 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6272 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6273 s->timeout = PFTM_TCP_FIRST_PACKET; 6274 atomic_add_32(&V_pf_status.states_halfopen, 1); 6275 break; 6276 case IPPROTO_UDP: 6277 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6278 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6279 s->timeout = PFTM_UDP_FIRST_PACKET; 6280 break; 6281 case IPPROTO_SCTP: 6282 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6283 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6284 s->timeout = PFTM_SCTP_FIRST_PACKET; 6285 break; 6286 case IPPROTO_ICMP: 6287 #ifdef INET6 6288 case IPPROTO_ICMPV6: 6289 #endif /* INET6 */ 6290 s->timeout = PFTM_ICMP_FIRST_PACKET; 6291 break; 6292 default: 6293 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6294 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6295 s->timeout = PFTM_OTHER_FIRST_PACKET; 6296 } 6297 6298 s->creation = s->expire = pf_get_uptime(); 6299 6300 if (pd->proto == IPPROTO_TCP) { 6301 if (s->state_flags & PFSTATE_SCRUB_TCP && 6302 pf_normalize_tcp_init(pd, th, &s->src)) { 6303 REASON_SET(&ctx->reason, PFRES_MEMORY); 6304 goto csfailed; 6305 } 6306 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6307 pf_normalize_tcp_stateful(pd, &ctx->reason, th, s, 6308 &s->src, &s->dst, &ctx->rewrite)) { 6309 /* This really shouldn't happen!!! */ 6310 DPFPRINTF(PF_DEBUG_URGENT, 6311 "%s: tcp normalize failed on first " 6312 "pkt", __func__); 6313 goto csfailed; 6314 } 6315 } else if (pd->proto == IPPROTO_SCTP) { 6316 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6317 goto csfailed; 6318 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6319 goto csfailed; 6320 } 6321 s->direction = pd->dir; 6322 6323 /* 6324 * sk/nk could already been setup by pf_get_translation(). 6325 */ 6326 if (ctx->sk == NULL && ctx->nk == NULL) { 6327 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6328 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6329 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, 6330 &ctx->sk, &ctx->nk)) { 6331 goto csfailed; 6332 } 6333 } else 6334 KASSERT((ctx->sk != NULL && ctx->nk != NULL), ("%s: nr %p sk %p, nk %p", 6335 __func__, ctx->nr, ctx->sk, ctx->nk)); 6336 6337 /* Swap sk/nk for PF_OUT. */ 6338 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6339 (pd->dir == PF_IN) ? ctx->sk : ctx->nk, 6340 (pd->dir == PF_IN) ? ctx->nk : ctx->sk, s)) { 6341 REASON_SET(&ctx->reason, PFRES_STATEINS); 6342 goto drop; 6343 } else 6344 *sm = s; 6345 ctx->sk = ctx->nk = NULL; 6346 6347 STATE_INC_COUNTERS(s); 6348 6349 /* 6350 * Lock order is important: first state, then source node. 6351 */ 6352 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6353 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6354 s->sns[sn_type] = sns[sn_type]; 6355 PF_HASHROW_UNLOCK(snhs[sn_type]); 6356 } 6357 } 6358 6359 if (ctx->tag > 0) 6360 s->tag = ctx->tag; 6361 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6362 TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 6363 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6364 pf_undo_nat(ctx->nr, pd, bip_sum); 6365 s->src.seqhi = arc4random(); 6366 /* Find mss option */ 6367 int rtid = M_GETFIB(pd->m); 6368 mss = pf_get_mss(pd); 6369 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6370 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6371 s->src.mss = mss; 6372 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6373 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6374 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6375 pd->act.rtableid, &ctx->reason); 6376 REASON_SET(&ctx->reason, PFRES_SYNPROXY); 6377 return (PF_SYNPROXY_DROP); 6378 } 6379 6380 s->udp_mapping = ctx->udp_mapping; 6381 6382 return (PF_PASS); 6383 6384 csfailed: 6385 uma_zfree(V_pf_state_key_z, ctx->sk); 6386 uma_zfree(V_pf_state_key_z, ctx->nk); 6387 6388 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6389 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6390 if (--sns[sn_type]->states == 0 && 6391 sns[sn_type]->expire == 0) { 6392 pf_unlink_src_node(sns[sn_type]); 6393 pf_free_src_node(sns[sn_type]); 6394 counter_u64_add( 6395 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6396 } 6397 PF_HASHROW_UNLOCK(snhs[sn_type]); 6398 } 6399 } 6400 6401 drop: 6402 if (s != NULL) { 6403 pf_src_tree_remove_state(s); 6404 s->timeout = PFTM_UNLINKED; 6405 pf_free_state(s); 6406 } 6407 6408 return (PF_DROP); 6409 } 6410 6411 int 6412 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6413 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6414 int icmp_dir) 6415 { 6416 /* 6417 * pf_translate() implements OpenBSD's "new" NAT approach. 6418 * We don't follow it, because it involves a breaking syntax change 6419 * (removing nat/rdr rules, moving it into regular pf rules.) 6420 * It also moves NAT processing to be done after normal rules evaluation 6421 * whereas in FreeBSD that's done before rules processing. 6422 * 6423 * We adopt the function only for nat64, and keep other NAT processing 6424 * before rules processing. 6425 */ 6426 int rewrite = 0; 6427 int afto = pd->af != pd->naf; 6428 6429 MPASS(afto); 6430 6431 switch (pd->proto) { 6432 case IPPROTO_TCP: 6433 case IPPROTO_UDP: 6434 case IPPROTO_SCTP: 6435 if (afto || *pd->sport != sport) { 6436 pf_change_ap(pd, pd->src, pd->sport, 6437 saddr, sport); 6438 rewrite = 1; 6439 } 6440 if (afto || *pd->dport != dport) { 6441 pf_change_ap(pd, pd->dst, pd->dport, 6442 daddr, dport); 6443 rewrite = 1; 6444 } 6445 break; 6446 6447 #ifdef INET 6448 case IPPROTO_ICMP: 6449 /* pf_translate() is also used when logging invalid packets */ 6450 if (pd->af != AF_INET) 6451 return (0); 6452 6453 if (afto) { 6454 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6455 return (-1); 6456 pd->proto = IPPROTO_ICMPV6; 6457 rewrite = 1; 6458 } 6459 if (virtual_type == htons(ICMP_ECHO)) { 6460 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6461 6462 if (icmpid != pd->hdr.icmp.icmp_id) { 6463 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6464 pd->hdr.icmp.icmp_cksum, 6465 pd->hdr.icmp.icmp_id, icmpid, 0); 6466 pd->hdr.icmp.icmp_id = icmpid; 6467 /* XXX TODO copyback. */ 6468 rewrite = 1; 6469 } 6470 } 6471 break; 6472 #endif /* INET */ 6473 6474 #ifdef INET6 6475 case IPPROTO_ICMPV6: 6476 /* pf_translate() is also used when logging invalid packets */ 6477 if (pd->af != AF_INET6) 6478 return (0); 6479 6480 if (afto) { 6481 /* ip_sum will be recalculated in pf_translate_af */ 6482 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6483 return (0); 6484 pd->proto = IPPROTO_ICMP; 6485 rewrite = 1; 6486 } 6487 break; 6488 #endif /* INET6 */ 6489 6490 default: 6491 break; 6492 } 6493 6494 return (rewrite); 6495 } 6496 6497 int 6498 pf_translate_compat(struct pf_test_ctx *ctx) 6499 { 6500 struct pf_pdesc *pd = ctx->pd; 6501 struct pf_state_key *nk = ctx->nk; 6502 struct tcphdr *th = &pd->hdr.tcp; 6503 int rewrite = 0; 6504 6505 KASSERT(ctx->sk != NULL, ("%s: null sk", __func__)); 6506 KASSERT(ctx->nk != NULL, ("%s: null nk", __func__)); 6507 6508 switch (pd->virtual_proto) { 6509 case IPPROTO_TCP: 6510 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6511 nk->port[pd->sidx] != pd->nsport) { 6512 pf_change_ap(pd, pd->src, &th->th_sport, 6513 &nk->addr[pd->sidx], nk->port[pd->sidx]); 6514 pd->sport = &th->th_sport; 6515 pd->nsport = th->th_sport; 6516 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6517 } 6518 6519 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6520 nk->port[pd->didx] != pd->ndport) { 6521 pf_change_ap(pd, pd->dst, &th->th_dport, 6522 &nk->addr[pd->didx], nk->port[pd->didx]); 6523 pd->dport = &th->th_dport; 6524 pd->ndport = th->th_dport; 6525 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6526 } 6527 rewrite++; 6528 break; 6529 case IPPROTO_UDP: 6530 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6531 nk->port[pd->sidx] != pd->nsport) { 6532 pf_change_ap(pd, pd->src, 6533 &pd->hdr.udp.uh_sport, 6534 &nk->addr[pd->sidx], 6535 nk->port[pd->sidx]); 6536 pd->sport = &pd->hdr.udp.uh_sport; 6537 pd->nsport = pd->hdr.udp.uh_sport; 6538 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6539 } 6540 6541 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6542 nk->port[pd->didx] != pd->ndport) { 6543 pf_change_ap(pd, pd->dst, 6544 &pd->hdr.udp.uh_dport, 6545 &nk->addr[pd->didx], 6546 nk->port[pd->didx]); 6547 pd->dport = &pd->hdr.udp.uh_dport; 6548 pd->ndport = pd->hdr.udp.uh_dport; 6549 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6550 } 6551 rewrite++; 6552 break; 6553 case IPPROTO_SCTP: { 6554 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6555 nk->port[pd->sidx] != pd->nsport) { 6556 pf_change_ap(pd, pd->src, 6557 &pd->hdr.sctp.src_port, 6558 &nk->addr[pd->sidx], 6559 nk->port[pd->sidx]); 6560 pd->sport = &pd->hdr.sctp.src_port; 6561 pd->nsport = pd->hdr.sctp.src_port; 6562 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6563 } 6564 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6565 nk->port[pd->didx] != pd->ndport) { 6566 pf_change_ap(pd, pd->dst, 6567 &pd->hdr.sctp.dest_port, 6568 &nk->addr[pd->didx], 6569 nk->port[pd->didx]); 6570 pd->dport = &pd->hdr.sctp.dest_port; 6571 pd->ndport = pd->hdr.sctp.dest_port; 6572 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6573 } 6574 break; 6575 } 6576 #ifdef INET 6577 case IPPROTO_ICMP: 6578 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 6579 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 6580 nk->addr[pd->sidx].v4.s_addr, 0); 6581 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6582 } 6583 6584 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 6585 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 6586 nk->addr[pd->didx].v4.s_addr, 0); 6587 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6588 } 6589 6590 if (ctx->virtual_type == htons(ICMP_ECHO) && 6591 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 6592 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6593 pd->hdr.icmp.icmp_cksum, pd->nsport, 6594 nk->port[pd->sidx], 0); 6595 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 6596 pd->sport = &pd->hdr.icmp.icmp_id; 6597 } 6598 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6599 break; 6600 #endif /* INET */ 6601 #ifdef INET6 6602 case IPPROTO_ICMPV6: 6603 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 6604 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 6605 &nk->addr[pd->sidx], 0); 6606 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6607 } 6608 6609 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 6610 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 6611 &nk->addr[pd->didx], 0); 6612 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6613 } 6614 rewrite++; 6615 break; 6616 #endif /* INET */ 6617 default: 6618 switch (pd->af) { 6619 #ifdef INET 6620 case AF_INET: 6621 if (PF_ANEQ(&pd->nsaddr, 6622 &nk->addr[pd->sidx], AF_INET)) { 6623 pf_change_a(&pd->src->v4.s_addr, 6624 pd->ip_sum, 6625 nk->addr[pd->sidx].v4.s_addr, 0); 6626 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6627 } 6628 6629 if (PF_ANEQ(&pd->ndaddr, 6630 &nk->addr[pd->didx], AF_INET)) { 6631 pf_change_a(&pd->dst->v4.s_addr, 6632 pd->ip_sum, 6633 nk->addr[pd->didx].v4.s_addr, 0); 6634 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6635 } 6636 break; 6637 #endif /* INET */ 6638 #ifdef INET6 6639 case AF_INET6: 6640 if (PF_ANEQ(&pd->nsaddr, 6641 &nk->addr[pd->sidx], AF_INET6)) { 6642 pf_addrcpy(&pd->nsaddr, &nk->addr[pd->sidx], 6643 pd->af); 6644 pf_addrcpy(pd->src, &nk->addr[pd->sidx], pd->af); 6645 } 6646 6647 if (PF_ANEQ(&pd->ndaddr, 6648 &nk->addr[pd->didx], AF_INET6)) { 6649 pf_addrcpy(&pd->ndaddr, &nk->addr[pd->didx], 6650 pd->af); 6651 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 6652 pd->af); 6653 } 6654 break; 6655 #endif /* INET6 */ 6656 } 6657 break; 6658 } 6659 return (rewrite); 6660 } 6661 6662 static int 6663 pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd, 6664 u_short *reason, int *copyback, struct pf_state_peer *src, 6665 struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst) 6666 { 6667 struct tcphdr *th = &pd->hdr.tcp; 6668 u_int16_t win = ntohs(th->th_win); 6669 u_int32_t ack, end, data_end, seq, orig_seq; 6670 u_int8_t sws, dws; 6671 int ackskew; 6672 6673 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6674 sws = src->wscale & PF_WSCALE_MASK; 6675 dws = dst->wscale & PF_WSCALE_MASK; 6676 } else 6677 sws = dws = 0; 6678 6679 /* 6680 * Sequence tracking algorithm from Guido van Rooij's paper: 6681 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6682 * tcp_filtering.ps 6683 */ 6684 6685 orig_seq = seq = ntohl(th->th_seq); 6686 if (src->seqlo == 0) { 6687 /* First packet from this end. Set its state */ 6688 6689 if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6690 src->scrub == NULL) { 6691 if (pf_normalize_tcp_init(pd, th, src)) { 6692 REASON_SET(reason, PFRES_MEMORY); 6693 return (PF_DROP); 6694 } 6695 } 6696 6697 /* Deferred generation of sequence number modulator */ 6698 if (dst->seqdiff && !src->seqdiff) { 6699 /* use random iss for the TCP server */ 6700 while ((src->seqdiff = arc4random() - seq) == 0) 6701 ; 6702 ack = ntohl(th->th_ack) - dst->seqdiff; 6703 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6704 src->seqdiff), 0); 6705 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6706 *copyback = 1; 6707 } else { 6708 ack = ntohl(th->th_ack); 6709 } 6710 6711 end = seq + pd->p_len; 6712 if (tcp_get_flags(th) & TH_SYN) { 6713 end++; 6714 if (dst->wscale & PF_WSCALE_FLAG) { 6715 src->wscale = pf_get_wscale(pd); 6716 if (src->wscale & PF_WSCALE_FLAG) { 6717 /* Remove scale factor from initial 6718 * window */ 6719 sws = src->wscale & PF_WSCALE_MASK; 6720 win = ((u_int32_t)win + (1 << sws) - 1) 6721 >> sws; 6722 dws = dst->wscale & PF_WSCALE_MASK; 6723 } else { 6724 /* fixup other window */ 6725 dst->max_win = MIN(TCP_MAXWIN, 6726 (u_int32_t)dst->max_win << 6727 (dst->wscale & PF_WSCALE_MASK)); 6728 /* in case of a retrans SYN|ACK */ 6729 dst->wscale = 0; 6730 } 6731 } 6732 } 6733 data_end = end; 6734 if (tcp_get_flags(th) & TH_FIN) 6735 end++; 6736 6737 src->seqlo = seq; 6738 if (src->state < TCPS_SYN_SENT) 6739 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6740 6741 /* 6742 * May need to slide the window (seqhi may have been set by 6743 * the crappy stack check or if we picked up the connection 6744 * after establishment) 6745 */ 6746 if (src->seqhi == 1 || 6747 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6748 src->seqhi = end + MAX(1, dst->max_win << dws); 6749 if (win > src->max_win) 6750 src->max_win = win; 6751 6752 } else { 6753 ack = ntohl(th->th_ack) - dst->seqdiff; 6754 if (src->seqdiff) { 6755 /* Modulate sequence numbers */ 6756 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6757 src->seqdiff), 0); 6758 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6759 *copyback = 1; 6760 } 6761 end = seq + pd->p_len; 6762 if (tcp_get_flags(th) & TH_SYN) 6763 end++; 6764 data_end = end; 6765 if (tcp_get_flags(th) & TH_FIN) 6766 end++; 6767 } 6768 6769 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6770 /* Let it pass through the ack skew check */ 6771 ack = dst->seqlo; 6772 } else if ((ack == 0 && 6773 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6774 /* broken tcp stacks do not set ack */ 6775 (dst->state < TCPS_SYN_SENT)) { 6776 /* 6777 * Many stacks (ours included) will set the ACK number in an 6778 * FIN|ACK if the SYN times out -- no sequence to ACK. 6779 */ 6780 ack = dst->seqlo; 6781 } 6782 6783 if (seq == end) { 6784 /* Ease sequencing restrictions on no data packets */ 6785 seq = src->seqlo; 6786 data_end = end = seq; 6787 } 6788 6789 ackskew = dst->seqlo - ack; 6790 6791 /* 6792 * Need to demodulate the sequence numbers in any TCP SACK options 6793 * (Selective ACK). We could optionally validate the SACK values 6794 * against the current ACK window, either forwards or backwards, but 6795 * I'm not confident that SACK has been implemented properly 6796 * everywhere. It wouldn't surprise me if several stacks accidentally 6797 * SACK too far backwards of previously ACKed data. There really aren't 6798 * any security implications of bad SACKing unless the target stack 6799 * doesn't validate the option length correctly. Someone trying to 6800 * spoof into a TCP connection won't bother blindly sending SACK 6801 * options anyway. 6802 */ 6803 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6804 if (pf_modulate_sack(pd, th, dst)) 6805 *copyback = 1; 6806 } 6807 6808 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6809 if (SEQ_GEQ(src->seqhi, data_end) && 6810 /* Last octet inside other's window space */ 6811 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6812 /* Retrans: not more than one window back */ 6813 (ackskew >= -MAXACKWINDOW) && 6814 /* Acking not more than one reassembled fragment backwards */ 6815 (ackskew <= (MAXACKWINDOW << sws)) && 6816 /* Acking not more than one window forward */ 6817 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6818 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 6819 /* Require an exact/+1 sequence match on resets when possible */ 6820 (SEQ_GEQ(orig_seq, src->seqlo - (dst->max_win << dws)) && 6821 SEQ_LEQ(orig_seq, src->seqlo + 1) && ackskew == 0 && 6822 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)))) { 6823 /* Allow resets to match sequence window if ack is perfect match */ 6824 6825 if (dst->scrub || src->scrub) { 6826 if (pf_normalize_tcp_stateful(pd, reason, th, 6827 state, src, dst, copyback)) 6828 return (PF_DROP); 6829 } 6830 6831 /* update max window */ 6832 if (src->max_win < win) 6833 src->max_win = win; 6834 /* synchronize sequencing */ 6835 if (SEQ_GT(end, src->seqlo)) 6836 src->seqlo = end; 6837 /* slide the window of what the other end can send */ 6838 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6839 dst->seqhi = ack + MAX((win << sws), 1); 6840 6841 /* update states */ 6842 if (tcp_get_flags(th) & TH_SYN) 6843 if (src->state < TCPS_SYN_SENT) 6844 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6845 if (tcp_get_flags(th) & TH_FIN) 6846 if (src->state < TCPS_CLOSING) 6847 pf_set_protostate(state, psrc, TCPS_CLOSING); 6848 if (tcp_get_flags(th) & TH_ACK) { 6849 if (dst->state == TCPS_SYN_SENT) { 6850 pf_set_protostate(state, pdst, 6851 TCPS_ESTABLISHED); 6852 if (src->state == TCPS_ESTABLISHED && 6853 state->sns[PF_SN_LIMIT] != NULL && 6854 pf_src_connlimit(state)) { 6855 REASON_SET(reason, PFRES_SRCLIMIT); 6856 return (PF_DROP); 6857 } 6858 } else if (dst->state == TCPS_CLOSING) 6859 pf_set_protostate(state, pdst, 6860 TCPS_FIN_WAIT_2); 6861 } 6862 if (tcp_get_flags(th) & TH_RST) 6863 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6864 6865 /* update expire time */ 6866 state->expire = pf_get_uptime(); 6867 if (src->state >= TCPS_FIN_WAIT_2 && 6868 dst->state >= TCPS_FIN_WAIT_2) 6869 state->timeout = PFTM_TCP_CLOSED; 6870 else if (src->state >= TCPS_CLOSING && 6871 dst->state >= TCPS_CLOSING) 6872 state->timeout = PFTM_TCP_FIN_WAIT; 6873 else if (src->state < TCPS_ESTABLISHED || 6874 dst->state < TCPS_ESTABLISHED) 6875 state->timeout = PFTM_TCP_OPENING; 6876 else if (src->state >= TCPS_CLOSING || 6877 dst->state >= TCPS_CLOSING) 6878 state->timeout = PFTM_TCP_CLOSING; 6879 else 6880 state->timeout = PFTM_TCP_ESTABLISHED; 6881 6882 /* Fall through to PASS packet */ 6883 6884 } else if ((dst->state < TCPS_SYN_SENT || 6885 dst->state >= TCPS_FIN_WAIT_2 || 6886 src->state >= TCPS_FIN_WAIT_2) && 6887 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6888 /* Within a window forward of the originating packet */ 6889 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6890 /* Within a window backward of the originating packet */ 6891 6892 /* 6893 * This currently handles three situations: 6894 * 1) Stupid stacks will shotgun SYNs before their peer 6895 * replies. 6896 * 2) When PF catches an already established stream (the 6897 * firewall rebooted, the state table was flushed, routes 6898 * changed...) 6899 * 3) Packets get funky immediately after the connection 6900 * closes (this should catch Solaris spurious ACK|FINs 6901 * that web servers like to spew after a close) 6902 * 6903 * This must be a little more careful than the above code 6904 * since packet floods will also be caught here. We don't 6905 * update the TTL here to mitigate the damage of a packet 6906 * flood and so the same code can handle awkward establishment 6907 * and a loosened connection close. 6908 * In the establishment case, a correct peer response will 6909 * validate the connection, go through the normal state code 6910 * and keep updating the state TTL. 6911 */ 6912 6913 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6914 printf("pf: loose state match: "); 6915 pf_print_state(state); 6916 pf_print_flags(tcp_get_flags(th)); 6917 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6918 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6919 pd->p_len, ackskew, (unsigned long long)state->packets[0], 6920 (unsigned long long)state->packets[1], 6921 pd->dir == PF_IN ? "in" : "out", 6922 pd->dir == state->direction ? "fwd" : "rev"); 6923 } 6924 6925 if (dst->scrub || src->scrub) { 6926 if (pf_normalize_tcp_stateful(pd, reason, th, 6927 state, src, dst, copyback)) 6928 return (PF_DROP); 6929 } 6930 6931 /* update max window */ 6932 if (src->max_win < win) 6933 src->max_win = win; 6934 /* synchronize sequencing */ 6935 if (SEQ_GT(end, src->seqlo)) 6936 src->seqlo = end; 6937 /* slide the window of what the other end can send */ 6938 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6939 dst->seqhi = ack + MAX((win << sws), 1); 6940 6941 /* 6942 * Cannot set dst->seqhi here since this could be a shotgunned 6943 * SYN and not an already established connection. 6944 */ 6945 6946 if (tcp_get_flags(th) & TH_FIN) 6947 if (src->state < TCPS_CLOSING) 6948 pf_set_protostate(state, psrc, TCPS_CLOSING); 6949 if (tcp_get_flags(th) & TH_RST) 6950 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6951 6952 /* Fall through to PASS packet */ 6953 6954 } else { 6955 if (state->dst.state == TCPS_SYN_SENT && 6956 state->src.state == TCPS_SYN_SENT) { 6957 /* Send RST for state mismatches during handshake */ 6958 if (!(tcp_get_flags(th) & TH_RST)) 6959 pf_send_tcp(state->rule, pd->af, 6960 pd->dst, pd->src, th->th_dport, 6961 th->th_sport, ntohl(th->th_ack), 0, 6962 TH_RST, 0, 0, 6963 state->rule->return_ttl, M_SKIP_FIREWALL, 6964 0, 0, state->act.rtableid, reason); 6965 src->seqlo = 0; 6966 src->seqhi = 1; 6967 src->max_win = 1; 6968 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6969 printf("pf: BAD state: "); 6970 pf_print_state(state); 6971 pf_print_flags(tcp_get_flags(th)); 6972 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6973 "pkts=%llu:%llu dir=%s,%s\n", 6974 seq, orig_seq, ack, pd->p_len, ackskew, 6975 (unsigned long long)state->packets[0], 6976 (unsigned long long)state->packets[1], 6977 pd->dir == PF_IN ? "in" : "out", 6978 pd->dir == state->direction ? "fwd" : "rev"); 6979 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6980 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6981 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6982 ' ': '2', 6983 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6984 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6985 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6986 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6987 } 6988 REASON_SET(reason, PFRES_BADSTATE); 6989 return (PF_DROP); 6990 } 6991 6992 return (PF_PASS); 6993 } 6994 6995 static int 6996 pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd, 6997 u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst, 6998 u_int8_t psrc, u_int8_t pdst) 6999 { 7000 struct tcphdr *th = &pd->hdr.tcp; 7001 7002 if (tcp_get_flags(th) & TH_SYN) 7003 if (src->state < TCPS_SYN_SENT) 7004 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 7005 if (tcp_get_flags(th) & TH_FIN) 7006 if (src->state < TCPS_CLOSING) 7007 pf_set_protostate(state, psrc, TCPS_CLOSING); 7008 if (tcp_get_flags(th) & TH_ACK) { 7009 if (dst->state == TCPS_SYN_SENT) { 7010 pf_set_protostate(state, pdst, TCPS_ESTABLISHED); 7011 if (src->state == TCPS_ESTABLISHED && 7012 state->sns[PF_SN_LIMIT] != NULL && 7013 pf_src_connlimit(state)) { 7014 REASON_SET(reason, PFRES_SRCLIMIT); 7015 return (PF_DROP); 7016 } 7017 } else if (dst->state == TCPS_CLOSING) { 7018 pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2); 7019 } else if (src->state == TCPS_SYN_SENT && 7020 dst->state < TCPS_SYN_SENT) { 7021 /* 7022 * Handle a special sloppy case where we only see one 7023 * half of the connection. If there is a ACK after 7024 * the initial SYN without ever seeing a packet from 7025 * the destination, set the connection to established. 7026 */ 7027 pf_set_protostate(state, PF_PEER_BOTH, 7028 TCPS_ESTABLISHED); 7029 dst->state = src->state = TCPS_ESTABLISHED; 7030 if (state->sns[PF_SN_LIMIT] != NULL && 7031 pf_src_connlimit(state)) { 7032 REASON_SET(reason, PFRES_SRCLIMIT); 7033 return (PF_DROP); 7034 } 7035 } else if (src->state == TCPS_CLOSING && 7036 dst->state == TCPS_ESTABLISHED && 7037 dst->seqlo == 0) { 7038 /* 7039 * Handle the closing of half connections where we 7040 * don't see the full bidirectional FIN/ACK+ACK 7041 * handshake. 7042 */ 7043 pf_set_protostate(state, pdst, TCPS_CLOSING); 7044 } 7045 } 7046 if (tcp_get_flags(th) & TH_RST) 7047 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 7048 7049 /* update expire time */ 7050 state->expire = pf_get_uptime(); 7051 if (src->state >= TCPS_FIN_WAIT_2 && 7052 dst->state >= TCPS_FIN_WAIT_2) 7053 state->timeout = PFTM_TCP_CLOSED; 7054 else if (src->state >= TCPS_CLOSING && 7055 dst->state >= TCPS_CLOSING) 7056 state->timeout = PFTM_TCP_FIN_WAIT; 7057 else if (src->state < TCPS_ESTABLISHED || 7058 dst->state < TCPS_ESTABLISHED) 7059 state->timeout = PFTM_TCP_OPENING; 7060 else if (src->state >= TCPS_CLOSING || 7061 dst->state >= TCPS_CLOSING) 7062 state->timeout = PFTM_TCP_CLOSING; 7063 else 7064 state->timeout = PFTM_TCP_ESTABLISHED; 7065 7066 return (PF_PASS); 7067 } 7068 7069 static int 7070 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason) 7071 { 7072 struct pf_state_key *sk = state->key[pd->didx]; 7073 struct tcphdr *th = &pd->hdr.tcp; 7074 7075 if (state->src.state == PF_TCPS_PROXY_SRC) { 7076 if (pd->dir != state->direction) { 7077 REASON_SET(reason, PFRES_SYNPROXY); 7078 return (PF_SYNPROXY_DROP); 7079 } 7080 if (tcp_get_flags(th) & TH_SYN) { 7081 if (ntohl(th->th_seq) != state->src.seqlo) { 7082 REASON_SET(reason, PFRES_SYNPROXY); 7083 return (PF_DROP); 7084 } 7085 pf_send_tcp(state->rule, pd->af, pd->dst, 7086 pd->src, th->th_dport, th->th_sport, 7087 state->src.seqhi, ntohl(th->th_seq) + 1, 7088 TH_SYN|TH_ACK, 0, state->src.mss, 0, 7089 M_SKIP_FIREWALL, 0, 0, state->act.rtableid, 7090 reason); 7091 REASON_SET(reason, PFRES_SYNPROXY); 7092 return (PF_SYNPROXY_DROP); 7093 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 7094 (ntohl(th->th_ack) != state->src.seqhi + 1) || 7095 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 7096 REASON_SET(reason, PFRES_SYNPROXY); 7097 return (PF_DROP); 7098 } else if (state->sns[PF_SN_LIMIT] != NULL && 7099 pf_src_connlimit(state)) { 7100 REASON_SET(reason, PFRES_SRCLIMIT); 7101 return (PF_DROP); 7102 } else 7103 pf_set_protostate(state, PF_PEER_SRC, 7104 PF_TCPS_PROXY_DST); 7105 } 7106 if (state->src.state == PF_TCPS_PROXY_DST) { 7107 if (pd->dir == state->direction) { 7108 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 7109 (ntohl(th->th_ack) != state->src.seqhi + 1) || 7110 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 7111 REASON_SET(reason, PFRES_SYNPROXY); 7112 return (PF_DROP); 7113 } 7114 state->src.max_win = MAX(ntohs(th->th_win), 1); 7115 if (state->dst.seqhi == 1) 7116 state->dst.seqhi = arc4random(); 7117 pf_send_tcp(state->rule, pd->af, 7118 &sk->addr[pd->sidx], &sk->addr[pd->didx], 7119 sk->port[pd->sidx], sk->port[pd->didx], 7120 state->dst.seqhi, 0, TH_SYN, 0, 7121 state->src.mss, 0, 7122 state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 7123 state->tag, 0, state->act.rtableid, 7124 reason); 7125 REASON_SET(reason, PFRES_SYNPROXY); 7126 return (PF_SYNPROXY_DROP); 7127 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 7128 (TH_SYN|TH_ACK)) || 7129 (ntohl(th->th_ack) != state->dst.seqhi + 1)) { 7130 REASON_SET(reason, PFRES_SYNPROXY); 7131 return (PF_DROP); 7132 } else { 7133 state->dst.max_win = MAX(ntohs(th->th_win), 1); 7134 state->dst.seqlo = ntohl(th->th_seq); 7135 pf_send_tcp(state->rule, pd->af, pd->dst, 7136 pd->src, th->th_dport, th->th_sport, 7137 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 7138 TH_ACK, state->src.max_win, 0, 0, 0, 7139 state->tag, 0, state->act.rtableid, 7140 reason); 7141 pf_send_tcp(state->rule, pd->af, 7142 &sk->addr[pd->sidx], &sk->addr[pd->didx], 7143 sk->port[pd->sidx], sk->port[pd->didx], 7144 state->src.seqhi + 1, state->src.seqlo + 1, 7145 TH_ACK, state->dst.max_win, 0, 0, 7146 M_SKIP_FIREWALL, 0, 0, state->act.rtableid, 7147 reason); 7148 state->src.seqdiff = state->dst.seqhi - 7149 state->src.seqlo; 7150 state->dst.seqdiff = state->src.seqhi - 7151 state->dst.seqlo; 7152 state->src.seqhi = state->src.seqlo + 7153 state->dst.max_win; 7154 state->dst.seqhi = state->dst.seqlo + 7155 state->src.max_win; 7156 state->src.wscale = state->dst.wscale = 0; 7157 pf_set_protostate(state, PF_PEER_BOTH, 7158 TCPS_ESTABLISHED); 7159 REASON_SET(reason, PFRES_SYNPROXY); 7160 return (PF_SYNPROXY_DROP); 7161 } 7162 } 7163 7164 return (PF_PASS); 7165 } 7166 7167 static int 7168 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 7169 { 7170 struct pf_state_key_cmp key; 7171 int copyback = 0; 7172 struct pf_state_peer *src, *dst; 7173 uint8_t psrc, pdst; 7174 int action; 7175 7176 bzero(&key, sizeof(key)); 7177 key.af = pd->af; 7178 key.proto = pd->virtual_proto; 7179 pf_addrcpy(&key.addr[pd->sidx], pd->src, key.af); 7180 pf_addrcpy(&key.addr[pd->didx], pd->dst, key.af); 7181 key.port[pd->sidx] = pd->osport; 7182 key.port[pd->didx] = pd->odport; 7183 7184 action = pf_find_state(pd, &key, state); 7185 if (action != PF_MATCH) 7186 return (action); 7187 7188 action = PF_PASS; 7189 if (pd->dir == (*state)->direction) { 7190 if (PF_REVERSED_KEY(*state, pd->af)) { 7191 src = &(*state)->dst; 7192 dst = &(*state)->src; 7193 psrc = PF_PEER_DST; 7194 pdst = PF_PEER_SRC; 7195 } else { 7196 src = &(*state)->src; 7197 dst = &(*state)->dst; 7198 psrc = PF_PEER_SRC; 7199 pdst = PF_PEER_DST; 7200 } 7201 } else { 7202 if (PF_REVERSED_KEY(*state, pd->af)) { 7203 src = &(*state)->src; 7204 dst = &(*state)->dst; 7205 psrc = PF_PEER_SRC; 7206 pdst = PF_PEER_DST; 7207 } else { 7208 src = &(*state)->dst; 7209 dst = &(*state)->src; 7210 psrc = PF_PEER_DST; 7211 pdst = PF_PEER_SRC; 7212 } 7213 } 7214 7215 switch (pd->virtual_proto) { 7216 case IPPROTO_TCP: { 7217 struct tcphdr *th = &pd->hdr.tcp; 7218 7219 if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS) 7220 return (action); 7221 if (((tcp_get_flags(th) & (TH_SYN | TH_ACK)) == TH_SYN) || 7222 ((th->th_flags & (TH_SYN | TH_ACK | TH_RST)) == TH_ACK && 7223 pf_syncookie_check(pd) && pd->dir == PF_IN)) { 7224 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 7225 (*state)->dst.state >= TCPS_FIN_WAIT_2) { 7226 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7227 printf("pf: state reuse "); 7228 pf_print_state(*state); 7229 pf_print_flags(tcp_get_flags(th)); 7230 printf("\n"); 7231 } 7232 /* XXX make sure it's the same direction ?? */ 7233 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 7234 pf_remove_state(*state); 7235 *state = NULL; 7236 return (PF_DROP); 7237 } else if ((*state)->src.state >= TCPS_ESTABLISHED && 7238 (*state)->dst.state >= TCPS_ESTABLISHED) { 7239 /* 7240 * SYN matches existing state??? 7241 * Typically happens when sender boots up after 7242 * sudden panic. Certain protocols (NFSv3) are 7243 * always using same port numbers. Challenge 7244 * ACK enables all parties (firewall and peers) 7245 * to get in sync again. 7246 */ 7247 pf_send_challenge_ack(pd, *state, src, dst, reason); 7248 return (PF_DROP); 7249 } 7250 } 7251 if ((*state)->state_flags & PFSTATE_SLOPPY) { 7252 if (pf_tcp_track_sloppy(*state, pd, reason, src, dst, 7253 psrc, pdst) == PF_DROP) 7254 return (PF_DROP); 7255 } else { 7256 int ret; 7257 7258 ret = pf_tcp_track_full(*state, pd, reason, 7259 ©back, src, dst, psrc, pdst); 7260 if (ret == PF_DROP) 7261 return (PF_DROP); 7262 } 7263 break; 7264 } 7265 case IPPROTO_UDP: 7266 /* update states */ 7267 if (src->state < PFUDPS_SINGLE) 7268 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 7269 if (dst->state == PFUDPS_SINGLE) 7270 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 7271 7272 /* update expire time */ 7273 (*state)->expire = pf_get_uptime(); 7274 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 7275 (*state)->timeout = PFTM_UDP_MULTIPLE; 7276 else 7277 (*state)->timeout = PFTM_UDP_SINGLE; 7278 break; 7279 case IPPROTO_SCTP: 7280 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 7281 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 7282 pd->sctp_flags & PFDESC_SCTP_INIT) { 7283 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7284 pf_remove_state(*state); 7285 *state = NULL; 7286 return (PF_DROP); 7287 } 7288 7289 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7290 return (PF_DROP); 7291 7292 /* Track state. */ 7293 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7294 if (src->state < SCTP_COOKIE_WAIT) { 7295 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7296 (*state)->timeout = PFTM_SCTP_OPENING; 7297 } 7298 } 7299 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7300 MPASS(dst->scrub != NULL); 7301 if (dst->scrub->pfss_v_tag == 0) 7302 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7303 } 7304 7305 /* 7306 * Bind to the correct interface if we're if-bound. For multihomed 7307 * extra associations we don't know which interface that will be until 7308 * here, so we've inserted the state on V_pf_all. Fix that now. 7309 */ 7310 if ((*state)->kif == V_pfi_all && 7311 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7312 (*state)->kif = pd->kif; 7313 7314 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7315 if (src->state < SCTP_ESTABLISHED) { 7316 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7317 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7318 } 7319 } 7320 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7321 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7322 if (src->state < SCTP_SHUTDOWN_PENDING) { 7323 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7324 (*state)->timeout = PFTM_SCTP_CLOSING; 7325 } 7326 } 7327 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7328 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7329 (*state)->timeout = PFTM_SCTP_CLOSED; 7330 } 7331 7332 (*state)->expire = pf_get_uptime(); 7333 break; 7334 default: 7335 /* update states */ 7336 if (src->state < PFOTHERS_SINGLE) 7337 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7338 if (dst->state == PFOTHERS_SINGLE) 7339 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7340 7341 /* update expire time */ 7342 (*state)->expire = pf_get_uptime(); 7343 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7344 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7345 else 7346 (*state)->timeout = PFTM_OTHER_SINGLE; 7347 break; 7348 } 7349 7350 /* translate source/destination address, if necessary */ 7351 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7352 struct pf_state_key *nk; 7353 int afto, sidx, didx; 7354 7355 if (PF_REVERSED_KEY(*state, pd->af)) 7356 nk = (*state)->key[pd->sidx]; 7357 else 7358 nk = (*state)->key[pd->didx]; 7359 7360 afto = pd->af != nk->af; 7361 7362 if (afto && (*state)->direction == PF_IN) { 7363 sidx = pd->didx; 7364 didx = pd->sidx; 7365 } else { 7366 sidx = pd->sidx; 7367 didx = pd->didx; 7368 } 7369 7370 if (afto) { 7371 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 7372 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 7373 pd->naf = nk->af; 7374 action = PF_AFRT; 7375 } 7376 7377 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7378 nk->port[sidx] != pd->osport) 7379 pf_change_ap(pd, pd->src, pd->sport, 7380 &nk->addr[sidx], nk->port[sidx]); 7381 7382 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7383 nk->port[didx] != pd->odport) 7384 pf_change_ap(pd, pd->dst, pd->dport, 7385 &nk->addr[didx], nk->port[didx]); 7386 7387 copyback = 1; 7388 } 7389 7390 if (copyback && pd->hdrlen > 0) 7391 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 7392 7393 return (action); 7394 } 7395 7396 static int 7397 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7398 u_short *reason) 7399 { 7400 struct pf_state_peer *src; 7401 if (pd->dir == state->direction) { 7402 if (PF_REVERSED_KEY(state, pd->af)) 7403 src = &state->dst; 7404 else 7405 src = &state->src; 7406 } else { 7407 if (PF_REVERSED_KEY(state, pd->af)) 7408 src = &state->src; 7409 else 7410 src = &state->dst; 7411 } 7412 7413 if (src->scrub != NULL) { 7414 /* 7415 * Allow tags to be updated, in case of retransmission of 7416 * INIT/INIT_ACK chunks. 7417 **/ 7418 if (src->state <= SCTP_COOKIE_WAIT) 7419 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7420 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7421 return (PF_DROP); 7422 } 7423 7424 return (PF_PASS); 7425 } 7426 7427 static void 7428 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7429 { 7430 struct pf_sctp_endpoint key; 7431 struct pf_sctp_endpoint *ep; 7432 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7433 struct pf_sctp_source *i, *tmp; 7434 7435 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7436 return; 7437 7438 PF_SCTP_ENDPOINTS_LOCK(); 7439 7440 key.v_tag = s->dst.scrub->pfss_v_tag; 7441 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7442 if (ep != NULL) { 7443 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7444 if (pf_addr_cmp(&i->addr, 7445 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7446 s->key[PF_SK_WIRE]->af) == 0) { 7447 SDT_PROBE3(pf, sctp, multihome, remove, 7448 key.v_tag, s, i); 7449 TAILQ_REMOVE(&ep->sources, i, entry); 7450 free(i, M_PFTEMP); 7451 break; 7452 } 7453 } 7454 7455 if (TAILQ_EMPTY(&ep->sources)) { 7456 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7457 free(ep, M_PFTEMP); 7458 } 7459 } 7460 7461 /* Other direction. */ 7462 key.v_tag = s->src.scrub->pfss_v_tag; 7463 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7464 if (ep != NULL) { 7465 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7466 if (pf_addr_cmp(&i->addr, 7467 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7468 s->key[PF_SK_WIRE]->af) == 0) { 7469 SDT_PROBE3(pf, sctp, multihome, remove, 7470 key.v_tag, s, i); 7471 TAILQ_REMOVE(&ep->sources, i, entry); 7472 free(i, M_PFTEMP); 7473 break; 7474 } 7475 } 7476 7477 if (TAILQ_EMPTY(&ep->sources)) { 7478 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7479 free(ep, M_PFTEMP); 7480 } 7481 } 7482 7483 PF_SCTP_ENDPOINTS_UNLOCK(); 7484 } 7485 7486 static void 7487 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7488 { 7489 struct pf_sctp_endpoint key = { 7490 .v_tag = v_tag, 7491 }; 7492 struct pf_sctp_source *i; 7493 struct pf_sctp_endpoint *ep; 7494 int count; 7495 7496 PF_SCTP_ENDPOINTS_LOCK(); 7497 7498 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7499 if (ep == NULL) { 7500 ep = malloc(sizeof(struct pf_sctp_endpoint), 7501 M_PFTEMP, M_NOWAIT); 7502 if (ep == NULL) { 7503 PF_SCTP_ENDPOINTS_UNLOCK(); 7504 return; 7505 } 7506 7507 ep->v_tag = v_tag; 7508 TAILQ_INIT(&ep->sources); 7509 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7510 } 7511 7512 /* Avoid inserting duplicates. */ 7513 count = 0; 7514 TAILQ_FOREACH(i, &ep->sources, entry) { 7515 count++; 7516 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7517 PF_SCTP_ENDPOINTS_UNLOCK(); 7518 return; 7519 } 7520 } 7521 7522 /* Limit the number of addresses per endpoint. */ 7523 if (count >= PF_SCTP_MAX_ENDPOINTS) { 7524 PF_SCTP_ENDPOINTS_UNLOCK(); 7525 return; 7526 } 7527 7528 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7529 if (i == NULL) { 7530 PF_SCTP_ENDPOINTS_UNLOCK(); 7531 return; 7532 } 7533 7534 i->af = pd->af; 7535 memcpy(&i->addr, a, sizeof(*a)); 7536 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7537 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7538 7539 PF_SCTP_ENDPOINTS_UNLOCK(); 7540 } 7541 7542 static void 7543 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7544 struct pf_kstate *s, int action) 7545 { 7546 struct pf_krule_slist match_rules; 7547 struct pf_sctp_multihome_job *j, *tmp; 7548 struct pf_sctp_source *i; 7549 int ret; 7550 struct pf_kstate *sm = NULL; 7551 struct pf_krule *ra = NULL; 7552 struct pf_krule *r = &V_pf_default_rule; 7553 struct pf_kruleset *rs = NULL; 7554 u_short reason; 7555 bool do_extra = true; 7556 7557 PF_RULES_RLOCK_TRACKER; 7558 7559 again: 7560 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7561 if (s == NULL || action != PF_PASS) 7562 goto free; 7563 7564 /* Confirm we don't recurse here. */ 7565 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7566 7567 switch (j->op) { 7568 case SCTP_ADD_IP_ADDRESS: { 7569 uint32_t v_tag = pd->sctp_initiate_tag; 7570 7571 if (v_tag == 0) { 7572 if (s->direction == pd->dir) 7573 v_tag = s->src.scrub->pfss_v_tag; 7574 else 7575 v_tag = s->dst.scrub->pfss_v_tag; 7576 } 7577 7578 /* 7579 * Avoid duplicating states. We'll already have 7580 * created a state based on the source address of 7581 * the packet, but SCTP endpoints may also list this 7582 * address again in the INIT(_ACK) parameters. 7583 */ 7584 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7585 break; 7586 } 7587 7588 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7589 PF_RULES_RLOCK(); 7590 sm = NULL; 7591 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7592 j->pd.related_rule = s->rule; 7593 } 7594 SLIST_INIT(&match_rules); 7595 ret = pf_test_rule(&r, &sm, 7596 &j->pd, &ra, &rs, &reason, NULL, &match_rules); 7597 /* 7598 * Nothing to do about match rules, the processed 7599 * packet has already increased the counters. 7600 */ 7601 pf_free_match_rules(&match_rules); 7602 PF_RULES_RUNLOCK(); 7603 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7604 if (ret != PF_DROP && sm != NULL) { 7605 /* Inherit v_tag values. */ 7606 if (sm->direction == s->direction) { 7607 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7608 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7609 } else { 7610 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7611 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7612 } 7613 PF_STATE_UNLOCK(sm); 7614 } else { 7615 /* If we try duplicate inserts? */ 7616 break; 7617 } 7618 7619 /* Only add the address if we've actually allowed the state. */ 7620 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7621 7622 if (! do_extra) { 7623 break; 7624 } 7625 /* 7626 * We need to do this for each of our source addresses. 7627 * Find those based on the verification tag. 7628 */ 7629 struct pf_sctp_endpoint key = { 7630 .v_tag = pd->hdr.sctp.v_tag, 7631 }; 7632 struct pf_sctp_endpoint *ep; 7633 7634 PF_SCTP_ENDPOINTS_LOCK(); 7635 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7636 if (ep == NULL) { 7637 PF_SCTP_ENDPOINTS_UNLOCK(); 7638 break; 7639 } 7640 MPASS(ep != NULL); 7641 7642 TAILQ_FOREACH(i, &ep->sources, entry) { 7643 struct pf_sctp_multihome_job *nj; 7644 7645 /* SCTP can intermingle IPv4 and IPv6. */ 7646 if (i->af != pd->af) 7647 continue; 7648 7649 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7650 if (! nj) { 7651 continue; 7652 } 7653 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7654 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7655 nj->pd.src = &nj->src; 7656 // New destination address! 7657 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7658 nj->pd.dst = &nj->dst; 7659 nj->pd.m = j->pd.m; 7660 nj->op = j->op; 7661 7662 MPASS(nj->pd.pcksum); 7663 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7664 } 7665 PF_SCTP_ENDPOINTS_UNLOCK(); 7666 7667 break; 7668 } 7669 case SCTP_DEL_IP_ADDRESS: { 7670 struct pf_state_key_cmp key; 7671 uint8_t psrc; 7672 int action; 7673 7674 bzero(&key, sizeof(key)); 7675 key.af = j->pd.af; 7676 key.proto = IPPROTO_SCTP; 7677 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7678 pf_addrcpy(&key.addr[0], j->pd.src, key.af); 7679 pf_addrcpy(&key.addr[1], j->pd.dst, key.af); 7680 key.port[0] = j->pd.hdr.sctp.src_port; 7681 key.port[1] = j->pd.hdr.sctp.dest_port; 7682 } else { /* stack side, reverse */ 7683 pf_addrcpy(&key.addr[1], j->pd.src, key.af); 7684 pf_addrcpy(&key.addr[0], j->pd.dst, key.af); 7685 key.port[1] = j->pd.hdr.sctp.src_port; 7686 key.port[0] = j->pd.hdr.sctp.dest_port; 7687 } 7688 7689 action = pf_find_state(&j->pd, &key, &sm); 7690 if (action == PF_MATCH) { 7691 PF_STATE_LOCK_ASSERT(sm); 7692 if (j->pd.dir == sm->direction) { 7693 psrc = PF_PEER_SRC; 7694 } else { 7695 psrc = PF_PEER_DST; 7696 } 7697 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7698 sm->timeout = PFTM_SCTP_CLOSING; 7699 PF_STATE_UNLOCK(sm); 7700 } 7701 break; 7702 default: 7703 panic("Unknown op %#x", j->op); 7704 } 7705 } 7706 7707 free: 7708 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7709 free(j, M_PFTEMP); 7710 } 7711 7712 /* We may have inserted extra work while processing the list. */ 7713 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7714 do_extra = false; 7715 goto again; 7716 } 7717 } 7718 7719 static int 7720 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7721 { 7722 int off = 0; 7723 struct pf_sctp_multihome_job *job; 7724 7725 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7726 7727 while (off < len) { 7728 struct sctp_paramhdr h; 7729 7730 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, 7731 pd->af)) 7732 return (PF_DROP); 7733 7734 /* Parameters are at least 4 bytes. */ 7735 if (ntohs(h.param_length) < 4) 7736 return (PF_DROP); 7737 7738 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7739 ntohs(h.param_length)); 7740 7741 switch (ntohs(h.param_type)) { 7742 case SCTP_IPV4_ADDRESS: { 7743 struct in_addr t; 7744 7745 if (ntohs(h.param_length) != 7746 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7747 return (PF_DROP); 7748 7749 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7750 NULL, pd->af)) 7751 return (PF_DROP); 7752 7753 if (in_nullhost(t)) 7754 t.s_addr = pd->src->v4.s_addr; 7755 7756 /* 7757 * We hold the state lock (idhash) here, which means 7758 * that we can't acquire the keyhash, or we'll get a 7759 * LOR (and potentially double-lock things too). We also 7760 * can't release the state lock here, so instead we'll 7761 * enqueue this for async handling. 7762 * There's a relatively small race here, in that a 7763 * packet using the new addresses could arrive already, 7764 * but that's just though luck for it. 7765 */ 7766 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7767 if (! job) 7768 return (PF_DROP); 7769 7770 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7771 7772 memcpy(&job->pd, pd, sizeof(*pd)); 7773 7774 // New source address! 7775 memcpy(&job->src, &t, sizeof(t)); 7776 job->pd.src = &job->src; 7777 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7778 job->pd.dst = &job->dst; 7779 job->pd.m = pd->m; 7780 job->op = op; 7781 7782 MPASS(job->pd.pcksum); 7783 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7784 break; 7785 } 7786 #ifdef INET6 7787 case SCTP_IPV6_ADDRESS: { 7788 struct in6_addr t; 7789 7790 if (ntohs(h.param_length) != 7791 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7792 return (PF_DROP); 7793 7794 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7795 NULL, pd->af)) 7796 return (PF_DROP); 7797 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7798 break; 7799 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7800 memcpy(&t, &pd->src->v6, sizeof(t)); 7801 7802 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7803 if (! job) 7804 return (PF_DROP); 7805 7806 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7807 7808 memcpy(&job->pd, pd, sizeof(*pd)); 7809 memcpy(&job->src, &t, sizeof(t)); 7810 job->pd.src = &job->src; 7811 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7812 job->pd.dst = &job->dst; 7813 job->pd.m = pd->m; 7814 job->op = op; 7815 7816 MPASS(job->pd.pcksum); 7817 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7818 break; 7819 } 7820 #endif /* INET6 */ 7821 case SCTP_ADD_IP_ADDRESS: { 7822 int ret; 7823 struct sctp_asconf_paramhdr ah; 7824 7825 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7826 NULL, pd->af)) 7827 return (PF_DROP); 7828 7829 ret = pf_multihome_scan(start + off + sizeof(ah), 7830 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7831 SCTP_ADD_IP_ADDRESS); 7832 if (ret != PF_PASS) 7833 return (ret); 7834 break; 7835 } 7836 case SCTP_DEL_IP_ADDRESS: { 7837 int ret; 7838 struct sctp_asconf_paramhdr ah; 7839 7840 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7841 NULL, pd->af)) 7842 return (PF_DROP); 7843 ret = pf_multihome_scan(start + off + sizeof(ah), 7844 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7845 SCTP_DEL_IP_ADDRESS); 7846 if (ret != PF_PASS) 7847 return (ret); 7848 break; 7849 } 7850 default: 7851 break; 7852 } 7853 7854 off += roundup(ntohs(h.param_length), 4); 7855 } 7856 7857 return (PF_PASS); 7858 } 7859 7860 int 7861 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7862 { 7863 start += sizeof(struct sctp_init_chunk); 7864 len -= sizeof(struct sctp_init_chunk); 7865 7866 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7867 } 7868 7869 int 7870 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7871 { 7872 start += sizeof(struct sctp_asconf_chunk); 7873 len -= sizeof(struct sctp_asconf_chunk); 7874 7875 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7876 } 7877 7878 int 7879 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7880 struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir, 7881 int *iidx, int multi, int inner) 7882 { 7883 int action, direction = pd->dir; 7884 7885 key->af = pd->af; 7886 key->proto = pd->proto; 7887 if (icmp_dir == PF_IN) { 7888 *iidx = pd->sidx; 7889 key->port[pd->sidx] = icmpid; 7890 key->port[pd->didx] = type; 7891 } else { 7892 *iidx = pd->didx; 7893 key->port[pd->sidx] = type; 7894 key->port[pd->didx] = icmpid; 7895 } 7896 if (pf_state_key_addr_setup(pd, key, multi)) 7897 return (PF_DROP); 7898 7899 action = pf_find_state(pd, key, state); 7900 if (action != PF_MATCH) 7901 return (action); 7902 7903 if ((*state)->state_flags & PFSTATE_SLOPPY) 7904 return (-1); 7905 7906 /* Is this ICMP message flowing in right direction? */ 7907 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7908 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7909 PF_IN : PF_OUT; 7910 else 7911 direction = (*state)->direction; 7912 if ((*state)->rule->type && 7913 (((!inner && direction == pd->dir) || 7914 (inner && direction != pd->dir)) ? 7915 PF_IN : PF_OUT) != icmp_dir) { 7916 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7917 printf("pf: icmp type %d in wrong direction (%d): ", 7918 ntohs(type), icmp_dir); 7919 pf_print_state(*state); 7920 printf("\n"); 7921 } 7922 PF_STATE_UNLOCK(*state); 7923 *state = NULL; 7924 return (PF_DROP); 7925 } 7926 return (-1); 7927 } 7928 7929 static int 7930 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7931 u_short *reason) 7932 { 7933 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7934 u_int16_t *icmpsum, virtual_id, virtual_type; 7935 u_int8_t icmptype, icmpcode; 7936 int icmp_dir, iidx, ret; 7937 struct pf_state_key_cmp key; 7938 #ifdef INET 7939 u_int16_t icmpid; 7940 #endif /* INET*/ 7941 7942 MPASS(*state == NULL); 7943 7944 bzero(&key, sizeof(key)); 7945 switch (pd->proto) { 7946 #ifdef INET 7947 case IPPROTO_ICMP: 7948 icmptype = pd->hdr.icmp.icmp_type; 7949 icmpcode = pd->hdr.icmp.icmp_code; 7950 icmpid = pd->hdr.icmp.icmp_id; 7951 icmpsum = &pd->hdr.icmp.icmp_cksum; 7952 break; 7953 #endif /* INET */ 7954 #ifdef INET6 7955 case IPPROTO_ICMPV6: 7956 icmptype = pd->hdr.icmp6.icmp6_type; 7957 icmpcode = pd->hdr.icmp6.icmp6_code; 7958 #ifdef INET 7959 icmpid = pd->hdr.icmp6.icmp6_id; 7960 #endif /* INET */ 7961 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7962 break; 7963 #endif /* INET6 */ 7964 default: 7965 panic("unhandled proto %d", pd->proto); 7966 } 7967 7968 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 7969 &virtual_type) == 0) { 7970 /* 7971 * ICMP query/reply message not related to a TCP/UDP/SCTP 7972 * packet. Search for an ICMP state. 7973 */ 7974 ret = pf_icmp_state_lookup(&key, pd, state, virtual_id, 7975 virtual_type, icmp_dir, &iidx, 0, 0); 7976 /* IPv6? try matching a multicast address */ 7977 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) { 7978 MPASS(*state == NULL); 7979 ret = pf_icmp_state_lookup(&key, pd, state, 7980 virtual_id, virtual_type, 7981 icmp_dir, &iidx, 1, 0); 7982 } 7983 if (ret >= 0) { 7984 MPASS(*state == NULL); 7985 return (ret); 7986 } 7987 7988 (*state)->expire = pf_get_uptime(); 7989 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7990 7991 /* translate source/destination address, if necessary */ 7992 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7993 struct pf_state_key *nk; 7994 int afto, sidx, didx; 7995 7996 if (PF_REVERSED_KEY(*state, pd->af)) 7997 nk = (*state)->key[pd->sidx]; 7998 else 7999 nk = (*state)->key[pd->didx]; 8000 8001 afto = pd->af != nk->af; 8002 8003 if (afto && (*state)->direction == PF_IN) { 8004 sidx = pd->didx; 8005 didx = pd->sidx; 8006 iidx = !iidx; 8007 } else { 8008 sidx = pd->sidx; 8009 didx = pd->didx; 8010 } 8011 8012 switch (pd->af) { 8013 #ifdef INET 8014 case AF_INET: 8015 #ifdef INET6 8016 if (afto) { 8017 if (pf_translate_icmp_af(AF_INET6, 8018 &pd->hdr.icmp)) 8019 return (PF_DROP); 8020 pd->proto = IPPROTO_ICMPV6; 8021 } 8022 #endif /* INET6 */ 8023 if (!afto && 8024 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 8025 pf_change_a(&saddr->v4.s_addr, 8026 pd->ip_sum, 8027 nk->addr[sidx].v4.s_addr, 8028 0); 8029 8030 if (!afto && PF_ANEQ(pd->dst, 8031 &nk->addr[didx], AF_INET)) 8032 pf_change_a(&daddr->v4.s_addr, 8033 pd->ip_sum, 8034 nk->addr[didx].v4.s_addr, 0); 8035 8036 if (nk->port[iidx] != 8037 pd->hdr.icmp.icmp_id) { 8038 pd->hdr.icmp.icmp_cksum = 8039 pf_cksum_fixup( 8040 pd->hdr.icmp.icmp_cksum, icmpid, 8041 nk->port[iidx], 0); 8042 pd->hdr.icmp.icmp_id = 8043 nk->port[iidx]; 8044 } 8045 8046 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8047 (caddr_t )&pd->hdr.icmp); 8048 break; 8049 #endif /* INET */ 8050 #ifdef INET6 8051 case AF_INET6: 8052 #ifdef INET 8053 if (afto) { 8054 if (pf_translate_icmp_af(AF_INET, 8055 &pd->hdr.icmp6)) 8056 return (PF_DROP); 8057 pd->proto = IPPROTO_ICMP; 8058 } 8059 #endif /* INET */ 8060 if (!afto && 8061 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 8062 pf_change_a6(saddr, 8063 &pd->hdr.icmp6.icmp6_cksum, 8064 &nk->addr[sidx], 0); 8065 8066 if (!afto && PF_ANEQ(pd->dst, 8067 &nk->addr[didx], AF_INET6)) 8068 pf_change_a6(daddr, 8069 &pd->hdr.icmp6.icmp6_cksum, 8070 &nk->addr[didx], 0); 8071 8072 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 8073 pd->hdr.icmp6.icmp6_id = 8074 nk->port[iidx]; 8075 8076 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8077 (caddr_t )&pd->hdr.icmp6); 8078 break; 8079 #endif /* INET6 */ 8080 } 8081 if (afto) { 8082 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 8083 nk->af); 8084 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 8085 nk->af); 8086 pd->naf = nk->af; 8087 return (PF_AFRT); 8088 } 8089 } 8090 return (PF_PASS); 8091 8092 } else { 8093 /* 8094 * ICMP error message in response to a TCP/UDP packet. 8095 * Extract the inner TCP/UDP header and search for that state. 8096 */ 8097 8098 struct pf_pdesc pd2; 8099 bzero(&pd2, sizeof pd2); 8100 #ifdef INET 8101 struct ip h2; 8102 #endif /* INET */ 8103 #ifdef INET6 8104 struct ip6_hdr h2_6; 8105 #endif /* INET6 */ 8106 int ipoff2 = 0; 8107 8108 pd2.af = pd->af; 8109 pd2.dir = pd->dir; 8110 /* Payload packet is from the opposite direction. */ 8111 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 8112 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 8113 pd2.m = pd->m; 8114 pd2.pf_mtag = pd->pf_mtag; 8115 pd2.kif = pd->kif; 8116 switch (pd->af) { 8117 #ifdef INET 8118 case AF_INET: 8119 /* offset of h2 in mbuf chain */ 8120 ipoff2 = pd->off + ICMP_MINLEN; 8121 8122 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 8123 reason, pd2.af)) { 8124 DPFPRINTF(PF_DEBUG_MISC, 8125 "pf: ICMP error message too short " 8126 "(ip)"); 8127 return (PF_DROP); 8128 } 8129 /* 8130 * ICMP error messages don't refer to non-first 8131 * fragments 8132 */ 8133 if (h2.ip_off & htons(IP_OFFMASK)) { 8134 REASON_SET(reason, PFRES_FRAG); 8135 return (PF_DROP); 8136 } 8137 8138 /* offset of protocol header that follows h2 */ 8139 pd2.off = ipoff2; 8140 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 8141 return (PF_DROP); 8142 8143 pd2.tot_len = ntohs(h2.ip_len); 8144 pd2.ttl = h2.ip_ttl; 8145 pd2.src = (struct pf_addr *)&h2.ip_src; 8146 pd2.dst = (struct pf_addr *)&h2.ip_dst; 8147 pd2.ip_sum = &h2.ip_sum; 8148 break; 8149 #endif /* INET */ 8150 #ifdef INET6 8151 case AF_INET6: 8152 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 8153 8154 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 8155 reason, pd2.af)) { 8156 DPFPRINTF(PF_DEBUG_MISC, 8157 "pf: ICMP error message too short " 8158 "(ip6)"); 8159 return (PF_DROP); 8160 } 8161 pd2.off = ipoff2; 8162 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 8163 return (PF_DROP); 8164 8165 pd2.tot_len = ntohs(h2_6.ip6_plen) + 8166 sizeof(struct ip6_hdr); 8167 pd2.ttl = h2_6.ip6_hlim; 8168 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 8169 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 8170 pd2.ip_sum = NULL; 8171 break; 8172 #endif /* INET6 */ 8173 default: 8174 unhandled_af(pd->af); 8175 } 8176 8177 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 8178 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8179 printf("pf: BAD ICMP %d:%d outer dst: ", 8180 icmptype, icmpcode); 8181 pf_print_host(pd->src, 0, pd->af); 8182 printf(" -> "); 8183 pf_print_host(pd->dst, 0, pd->af); 8184 printf(" inner src: "); 8185 pf_print_host(pd2.src, 0, pd2.af); 8186 printf(" -> "); 8187 pf_print_host(pd2.dst, 0, pd2.af); 8188 printf("\n"); 8189 } 8190 REASON_SET(reason, PFRES_BADSTATE); 8191 return (PF_DROP); 8192 } 8193 8194 switch (pd2.proto) { 8195 case IPPROTO_TCP: { 8196 struct tcphdr *th = &pd2.hdr.tcp; 8197 u_int32_t seq; 8198 struct pf_state_peer *src, *dst; 8199 u_int8_t dws; 8200 int copyback = 0; 8201 int action; 8202 8203 /* 8204 * Only the first 8 bytes of the TCP header can be 8205 * expected. Don't access any TCP header fields after 8206 * th_seq, an ackskew test is not possible. 8207 */ 8208 if (!pf_pull_hdr(pd->m, pd2.off, th, 8, reason, 8209 pd2.af)) { 8210 DPFPRINTF(PF_DEBUG_MISC, 8211 "pf: ICMP error message too short " 8212 "(tcp)"); 8213 return (PF_DROP); 8214 } 8215 pd2.pcksum = &pd2.hdr.tcp.th_sum; 8216 8217 key.af = pd2.af; 8218 key.proto = IPPROTO_TCP; 8219 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8220 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8221 key.port[pd2.sidx] = th->th_sport; 8222 key.port[pd2.didx] = th->th_dport; 8223 8224 action = pf_find_state(&pd2, &key, state); 8225 if (action != PF_MATCH) 8226 return (action); 8227 8228 if (pd->dir == (*state)->direction) { 8229 if (PF_REVERSED_KEY(*state, pd->af)) { 8230 src = &(*state)->src; 8231 dst = &(*state)->dst; 8232 } else { 8233 src = &(*state)->dst; 8234 dst = &(*state)->src; 8235 } 8236 } else { 8237 if (PF_REVERSED_KEY(*state, pd->af)) { 8238 src = &(*state)->dst; 8239 dst = &(*state)->src; 8240 } else { 8241 src = &(*state)->src; 8242 dst = &(*state)->dst; 8243 } 8244 } 8245 8246 if (src->wscale && dst->wscale) 8247 dws = dst->wscale & PF_WSCALE_MASK; 8248 else 8249 dws = 0; 8250 8251 /* Demodulate sequence number */ 8252 seq = ntohl(th->th_seq) - src->seqdiff; 8253 if (src->seqdiff) { 8254 pf_change_a(&th->th_seq, icmpsum, 8255 htonl(seq), 0); 8256 copyback = 1; 8257 } 8258 8259 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 8260 (!SEQ_GEQ(src->seqhi, seq) || 8261 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 8262 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8263 printf("pf: BAD ICMP %d:%d ", 8264 icmptype, icmpcode); 8265 pf_print_host(pd->src, 0, pd->af); 8266 printf(" -> "); 8267 pf_print_host(pd->dst, 0, pd->af); 8268 printf(" state: "); 8269 pf_print_state(*state); 8270 printf(" seq=%u\n", seq); 8271 } 8272 REASON_SET(reason, PFRES_BADSTATE); 8273 return (PF_DROP); 8274 } else { 8275 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8276 printf("pf: OK ICMP %d:%d ", 8277 icmptype, icmpcode); 8278 pf_print_host(pd->src, 0, pd->af); 8279 printf(" -> "); 8280 pf_print_host(pd->dst, 0, pd->af); 8281 printf(" state: "); 8282 pf_print_state(*state); 8283 printf(" seq=%u\n", seq); 8284 } 8285 } 8286 8287 /* translate source/destination address, if necessary */ 8288 if ((*state)->key[PF_SK_WIRE] != 8289 (*state)->key[PF_SK_STACK]) { 8290 8291 struct pf_state_key *nk; 8292 8293 if (PF_REVERSED_KEY(*state, pd->af)) 8294 nk = (*state)->key[pd->sidx]; 8295 else 8296 nk = (*state)->key[pd->didx]; 8297 8298 #if defined(INET) && defined(INET6) 8299 int afto, sidx, didx; 8300 8301 afto = pd->af != nk->af; 8302 8303 if (afto && (*state)->direction == PF_IN) { 8304 sidx = pd2.didx; 8305 didx = pd2.sidx; 8306 } else { 8307 sidx = pd2.sidx; 8308 didx = pd2.didx; 8309 } 8310 8311 if (afto) { 8312 if (pf_translate_icmp_af(nk->af, 8313 &pd->hdr.icmp)) 8314 return (PF_DROP); 8315 m_copyback(pd->m, pd->off, 8316 sizeof(struct icmp6_hdr), 8317 (c_caddr_t)&pd->hdr.icmp6); 8318 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8319 &pd2, &nk->addr[sidx], 8320 &nk->addr[didx], pd->af, 8321 nk->af)) 8322 return (PF_DROP); 8323 pf_addrcpy(&pd->nsaddr, 8324 &nk->addr[pd2.sidx], nk->af); 8325 pf_addrcpy(&pd->ndaddr, 8326 &nk->addr[pd2.didx], nk->af); 8327 if (nk->af == AF_INET) { 8328 pd->proto = IPPROTO_ICMP; 8329 } else { 8330 pd->proto = IPPROTO_ICMPV6; 8331 /* 8332 * IPv4 becomes IPv6 so we must 8333 * copy IPv4 src addr to least 8334 * 32bits in IPv6 address to 8335 * keep traceroute/icmp 8336 * working. 8337 */ 8338 pd->nsaddr.addr32[3] = 8339 pd->src->addr32[0]; 8340 } 8341 pd->naf = pd2.naf = nk->af; 8342 pf_change_ap(&pd2, pd2.src, &th->th_sport, 8343 &nk->addr[pd2.sidx], nk->port[sidx]); 8344 pf_change_ap(&pd2, pd2.dst, &th->th_dport, 8345 &nk->addr[pd2.didx], nk->port[didx]); 8346 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)th); 8347 return (PF_AFRT); 8348 } 8349 #endif /* INET && INET6 */ 8350 8351 if (PF_ANEQ(pd2.src, 8352 &nk->addr[pd2.sidx], pd2.af) || 8353 nk->port[pd2.sidx] != th->th_sport) 8354 pf_change_icmp(pd2.src, &th->th_sport, 8355 daddr, &nk->addr[pd2.sidx], 8356 nk->port[pd2.sidx], NULL, 8357 pd2.ip_sum, icmpsum, 8358 pd->ip_sum, 0, pd2.af); 8359 8360 if (PF_ANEQ(pd2.dst, 8361 &nk->addr[pd2.didx], pd2.af) || 8362 nk->port[pd2.didx] != th->th_dport) 8363 pf_change_icmp(pd2.dst, &th->th_dport, 8364 saddr, &nk->addr[pd2.didx], 8365 nk->port[pd2.didx], NULL, 8366 pd2.ip_sum, icmpsum, 8367 pd->ip_sum, 0, pd2.af); 8368 copyback = 1; 8369 } 8370 8371 if (copyback) { 8372 switch (pd2.af) { 8373 #ifdef INET 8374 case AF_INET: 8375 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8376 (caddr_t )&pd->hdr.icmp); 8377 m_copyback(pd->m, ipoff2, sizeof(h2), 8378 (caddr_t )&h2); 8379 break; 8380 #endif /* INET */ 8381 #ifdef INET6 8382 case AF_INET6: 8383 m_copyback(pd->m, pd->off, 8384 sizeof(struct icmp6_hdr), 8385 (caddr_t )&pd->hdr.icmp6); 8386 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8387 (caddr_t )&h2_6); 8388 break; 8389 #endif /* INET6 */ 8390 default: 8391 unhandled_af(pd->af); 8392 } 8393 m_copyback(pd->m, pd2.off, 8, (caddr_t)th); 8394 } 8395 8396 return (PF_PASS); 8397 break; 8398 } 8399 case IPPROTO_UDP: { 8400 struct udphdr *uh = &pd2.hdr.udp; 8401 int action; 8402 8403 if (!pf_pull_hdr(pd->m, pd2.off, uh, sizeof(*uh), 8404 reason, pd2.af)) { 8405 DPFPRINTF(PF_DEBUG_MISC, 8406 "pf: ICMP error message too short " 8407 "(udp)"); 8408 return (PF_DROP); 8409 } 8410 pd2.pcksum = &pd2.hdr.udp.uh_sum; 8411 8412 key.af = pd2.af; 8413 key.proto = IPPROTO_UDP; 8414 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8415 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8416 key.port[pd2.sidx] = uh->uh_sport; 8417 key.port[pd2.didx] = uh->uh_dport; 8418 8419 action = pf_find_state(&pd2, &key, state); 8420 if (action != PF_MATCH) 8421 return (action); 8422 8423 /* translate source/destination address, if necessary */ 8424 if ((*state)->key[PF_SK_WIRE] != 8425 (*state)->key[PF_SK_STACK]) { 8426 struct pf_state_key *nk; 8427 8428 if (PF_REVERSED_KEY(*state, pd->af)) 8429 nk = (*state)->key[pd->sidx]; 8430 else 8431 nk = (*state)->key[pd->didx]; 8432 8433 #if defined(INET) && defined(INET6) 8434 int afto, sidx, didx; 8435 8436 afto = pd->af != nk->af; 8437 8438 if (afto && (*state)->direction == PF_IN) { 8439 sidx = pd2.didx; 8440 didx = pd2.sidx; 8441 } else { 8442 sidx = pd2.sidx; 8443 didx = pd2.didx; 8444 } 8445 8446 if (afto) { 8447 if (pf_translate_icmp_af(nk->af, 8448 &pd->hdr.icmp)) 8449 return (PF_DROP); 8450 m_copyback(pd->m, pd->off, 8451 sizeof(struct icmp6_hdr), 8452 (c_caddr_t)&pd->hdr.icmp6); 8453 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8454 &pd2, &nk->addr[sidx], 8455 &nk->addr[didx], pd->af, 8456 nk->af)) 8457 return (PF_DROP); 8458 pf_addrcpy(&pd->nsaddr, 8459 &nk->addr[pd2.sidx], nk->af); 8460 pf_addrcpy(&pd->ndaddr, 8461 &nk->addr[pd2.didx], nk->af); 8462 if (nk->af == AF_INET) { 8463 pd->proto = IPPROTO_ICMP; 8464 } else { 8465 pd->proto = IPPROTO_ICMPV6; 8466 /* 8467 * IPv4 becomes IPv6 so we must 8468 * copy IPv4 src addr to least 8469 * 32bits in IPv6 address to 8470 * keep traceroute/icmp 8471 * working. 8472 */ 8473 pd->nsaddr.addr32[3] = 8474 pd->src->addr32[0]; 8475 } 8476 pd->naf = pd2.naf = nk->af; 8477 pf_change_ap(&pd2, pd2.src, &uh->uh_sport, 8478 &nk->addr[pd2.sidx], nk->port[sidx]); 8479 pf_change_ap(&pd2, pd2.dst, &uh->uh_dport, 8480 &nk->addr[pd2.didx], nk->port[didx]); 8481 m_copyback(pd2.m, pd2.off, sizeof(*uh), 8482 (c_caddr_t)uh); 8483 return (PF_AFRT); 8484 } 8485 #endif /* INET && INET6 */ 8486 8487 if (PF_ANEQ(pd2.src, 8488 &nk->addr[pd2.sidx], pd2.af) || 8489 nk->port[pd2.sidx] != uh->uh_sport) 8490 pf_change_icmp(pd2.src, &uh->uh_sport, 8491 daddr, &nk->addr[pd2.sidx], 8492 nk->port[pd2.sidx], &uh->uh_sum, 8493 pd2.ip_sum, icmpsum, 8494 pd->ip_sum, 1, pd2.af); 8495 8496 if (PF_ANEQ(pd2.dst, 8497 &nk->addr[pd2.didx], pd2.af) || 8498 nk->port[pd2.didx] != uh->uh_dport) 8499 pf_change_icmp(pd2.dst, &uh->uh_dport, 8500 saddr, &nk->addr[pd2.didx], 8501 nk->port[pd2.didx], &uh->uh_sum, 8502 pd2.ip_sum, icmpsum, 8503 pd->ip_sum, 1, pd2.af); 8504 8505 switch (pd2.af) { 8506 #ifdef INET 8507 case AF_INET: 8508 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8509 (caddr_t )&pd->hdr.icmp); 8510 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8511 break; 8512 #endif /* INET */ 8513 #ifdef INET6 8514 case AF_INET6: 8515 m_copyback(pd->m, pd->off, 8516 sizeof(struct icmp6_hdr), 8517 (caddr_t )&pd->hdr.icmp6); 8518 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8519 (caddr_t )&h2_6); 8520 break; 8521 #endif /* INET6 */ 8522 } 8523 m_copyback(pd->m, pd2.off, sizeof(*uh), (caddr_t)uh); 8524 } 8525 return (PF_PASS); 8526 break; 8527 } 8528 #ifdef INET 8529 case IPPROTO_SCTP: { 8530 struct sctphdr *sh = &pd2.hdr.sctp; 8531 struct pf_state_peer *src; 8532 int copyback = 0; 8533 int action; 8534 8535 if (! pf_pull_hdr(pd->m, pd2.off, sh, sizeof(*sh), reason, 8536 pd2.af)) { 8537 DPFPRINTF(PF_DEBUG_MISC, 8538 "pf: ICMP error message too short " 8539 "(sctp)"); 8540 return (PF_DROP); 8541 } 8542 pd2.pcksum = &pd2.sctp_dummy_sum; 8543 8544 key.af = pd2.af; 8545 key.proto = IPPROTO_SCTP; 8546 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8547 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8548 key.port[pd2.sidx] = sh->src_port; 8549 key.port[pd2.didx] = sh->dest_port; 8550 8551 action = pf_find_state(&pd2, &key, state); 8552 if (action != PF_MATCH) 8553 return (action); 8554 8555 if (pd->dir == (*state)->direction) { 8556 if (PF_REVERSED_KEY(*state, pd->af)) 8557 src = &(*state)->src; 8558 else 8559 src = &(*state)->dst; 8560 } else { 8561 if (PF_REVERSED_KEY(*state, pd->af)) 8562 src = &(*state)->dst; 8563 else 8564 src = &(*state)->src; 8565 } 8566 8567 if (src->scrub->pfss_v_tag != sh->v_tag) { 8568 DPFPRINTF(PF_DEBUG_MISC, 8569 "pf: ICMP error message has incorrect " 8570 "SCTP v_tag"); 8571 return (PF_DROP); 8572 } 8573 8574 /* translate source/destination address, if necessary */ 8575 if ((*state)->key[PF_SK_WIRE] != 8576 (*state)->key[PF_SK_STACK]) { 8577 8578 struct pf_state_key *nk; 8579 8580 if (PF_REVERSED_KEY(*state, pd->af)) 8581 nk = (*state)->key[pd->sidx]; 8582 else 8583 nk = (*state)->key[pd->didx]; 8584 8585 #if defined(INET) && defined(INET6) 8586 int afto, sidx, didx; 8587 8588 afto = pd->af != nk->af; 8589 8590 if (afto && (*state)->direction == PF_IN) { 8591 sidx = pd2.didx; 8592 didx = pd2.sidx; 8593 } else { 8594 sidx = pd2.sidx; 8595 didx = pd2.didx; 8596 } 8597 8598 if (afto) { 8599 if (pf_translate_icmp_af(nk->af, 8600 &pd->hdr.icmp)) 8601 return (PF_DROP); 8602 m_copyback(pd->m, pd->off, 8603 sizeof(struct icmp6_hdr), 8604 (c_caddr_t)&pd->hdr.icmp6); 8605 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8606 &pd2, &nk->addr[sidx], 8607 &nk->addr[didx], pd->af, 8608 nk->af)) 8609 return (PF_DROP); 8610 sh->src_port = nk->port[sidx]; 8611 sh->dest_port = nk->port[didx]; 8612 m_copyback(pd2.m, pd2.off, sizeof(*sh), (c_caddr_t)sh); 8613 pf_addrcpy(&pd->nsaddr, 8614 &nk->addr[pd2.sidx], nk->af); 8615 pf_addrcpy(&pd->ndaddr, 8616 &nk->addr[pd2.didx], nk->af); 8617 if (nk->af == AF_INET) { 8618 pd->proto = IPPROTO_ICMP; 8619 } else { 8620 pd->proto = IPPROTO_ICMPV6; 8621 /* 8622 * IPv4 becomes IPv6 so we must 8623 * copy IPv4 src addr to least 8624 * 32bits in IPv6 address to 8625 * keep traceroute/icmp 8626 * working. 8627 */ 8628 pd->nsaddr.addr32[3] = 8629 pd->src->addr32[0]; 8630 } 8631 pd->naf = nk->af; 8632 return (PF_AFRT); 8633 } 8634 #endif /* INET && INET6 */ 8635 8636 if (PF_ANEQ(pd2.src, 8637 &nk->addr[pd2.sidx], pd2.af) || 8638 nk->port[pd2.sidx] != sh->src_port) 8639 pf_change_icmp(pd2.src, &sh->src_port, 8640 daddr, &nk->addr[pd2.sidx], 8641 nk->port[pd2.sidx], NULL, 8642 pd2.ip_sum, icmpsum, 8643 pd->ip_sum, 0, pd2.af); 8644 8645 if (PF_ANEQ(pd2.dst, 8646 &nk->addr[pd2.didx], pd2.af) || 8647 nk->port[pd2.didx] != sh->dest_port) 8648 pf_change_icmp(pd2.dst, &sh->dest_port, 8649 saddr, &nk->addr[pd2.didx], 8650 nk->port[pd2.didx], NULL, 8651 pd2.ip_sum, icmpsum, 8652 pd->ip_sum, 0, pd2.af); 8653 copyback = 1; 8654 } 8655 8656 if (copyback) { 8657 switch (pd2.af) { 8658 #ifdef INET 8659 case AF_INET: 8660 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8661 (caddr_t )&pd->hdr.icmp); 8662 m_copyback(pd->m, ipoff2, sizeof(h2), 8663 (caddr_t )&h2); 8664 break; 8665 #endif /* INET */ 8666 #ifdef INET6 8667 case AF_INET6: 8668 m_copyback(pd->m, pd->off, 8669 sizeof(struct icmp6_hdr), 8670 (caddr_t )&pd->hdr.icmp6); 8671 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8672 (caddr_t )&h2_6); 8673 break; 8674 #endif /* INET6 */ 8675 } 8676 m_copyback(pd->m, pd2.off, sizeof(*sh), (caddr_t)sh); 8677 } 8678 8679 return (PF_PASS); 8680 break; 8681 } 8682 case IPPROTO_ICMP: { 8683 struct icmp *iih = &pd2.hdr.icmp; 8684 8685 if (pd2.af != AF_INET) { 8686 REASON_SET(reason, PFRES_NORM); 8687 return (PF_DROP); 8688 } 8689 8690 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8691 reason, pd2.af)) { 8692 DPFPRINTF(PF_DEBUG_MISC, 8693 "pf: ICMP error message too short i" 8694 "(icmp)"); 8695 return (PF_DROP); 8696 } 8697 pd2.pcksum = &pd2.hdr.icmp.icmp_cksum; 8698 8699 icmpid = iih->icmp_id; 8700 pf_icmp_mapping(&pd2, iih->icmp_type, 8701 &icmp_dir, &virtual_id, &virtual_type); 8702 8703 ret = pf_icmp_state_lookup(&key, &pd2, state, 8704 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8705 if (ret >= 0) { 8706 MPASS(*state == NULL); 8707 return (ret); 8708 } 8709 8710 /* translate source/destination address, if necessary */ 8711 if ((*state)->key[PF_SK_WIRE] != 8712 (*state)->key[PF_SK_STACK]) { 8713 struct pf_state_key *nk; 8714 8715 if (PF_REVERSED_KEY(*state, pd->af)) 8716 nk = (*state)->key[pd->sidx]; 8717 else 8718 nk = (*state)->key[pd->didx]; 8719 8720 #if defined(INET) && defined(INET6) 8721 int afto, sidx, didx; 8722 8723 afto = pd->af != nk->af; 8724 8725 if (afto && (*state)->direction == PF_IN) { 8726 sidx = pd2.didx; 8727 didx = pd2.sidx; 8728 iidx = !iidx; 8729 } else { 8730 sidx = pd2.sidx; 8731 didx = pd2.didx; 8732 } 8733 8734 if (afto) { 8735 if (nk->af != AF_INET6) 8736 return (PF_DROP); 8737 if (pf_translate_icmp_af(nk->af, 8738 &pd->hdr.icmp)) 8739 return (PF_DROP); 8740 m_copyback(pd->m, pd->off, 8741 sizeof(struct icmp6_hdr), 8742 (c_caddr_t)&pd->hdr.icmp6); 8743 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8744 &pd2, &nk->addr[sidx], 8745 &nk->addr[didx], pd->af, 8746 nk->af)) 8747 return (PF_DROP); 8748 pd->proto = IPPROTO_ICMPV6; 8749 if (pf_translate_icmp_af(nk->af, iih)) 8750 return (PF_DROP); 8751 if (virtual_type == htons(ICMP_ECHO) && 8752 nk->port[iidx] != iih->icmp_id) 8753 iih->icmp_id = nk->port[iidx]; 8754 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8755 (c_caddr_t)iih); 8756 pf_addrcpy(&pd->nsaddr, 8757 &nk->addr[pd2.sidx], nk->af); 8758 pf_addrcpy(&pd->ndaddr, 8759 &nk->addr[pd2.didx], nk->af); 8760 /* 8761 * IPv4 becomes IPv6 so we must copy 8762 * IPv4 src addr to least 32bits in 8763 * IPv6 address to keep traceroute 8764 * working. 8765 */ 8766 pd->nsaddr.addr32[3] = 8767 pd->src->addr32[0]; 8768 pd->naf = nk->af; 8769 return (PF_AFRT); 8770 } 8771 #endif /* INET && INET6 */ 8772 8773 if (PF_ANEQ(pd2.src, 8774 &nk->addr[pd2.sidx], pd2.af) || 8775 (virtual_type == htons(ICMP_ECHO) && 8776 nk->port[iidx] != iih->icmp_id)) 8777 pf_change_icmp(pd2.src, 8778 (virtual_type == htons(ICMP_ECHO)) ? 8779 &iih->icmp_id : NULL, 8780 daddr, &nk->addr[pd2.sidx], 8781 (virtual_type == htons(ICMP_ECHO)) ? 8782 nk->port[iidx] : 0, NULL, 8783 pd2.ip_sum, icmpsum, 8784 pd->ip_sum, 0, AF_INET); 8785 8786 if (PF_ANEQ(pd2.dst, 8787 &nk->addr[pd2.didx], pd2.af)) 8788 pf_change_icmp(pd2.dst, NULL, NULL, 8789 &nk->addr[pd2.didx], 0, NULL, 8790 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8791 AF_INET); 8792 8793 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8794 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8795 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8796 } 8797 return (PF_PASS); 8798 break; 8799 } 8800 #endif /* INET */ 8801 #ifdef INET6 8802 case IPPROTO_ICMPV6: { 8803 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8804 8805 if (pd2.af != AF_INET6) { 8806 REASON_SET(reason, PFRES_NORM); 8807 return (PF_DROP); 8808 } 8809 8810 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8811 sizeof(struct icmp6_hdr), reason, pd2.af)) { 8812 DPFPRINTF(PF_DEBUG_MISC, 8813 "pf: ICMP error message too short " 8814 "(icmp6)"); 8815 return (PF_DROP); 8816 } 8817 pd2.pcksum = &pd2.hdr.icmp6.icmp6_cksum; 8818 8819 pf_icmp_mapping(&pd2, iih->icmp6_type, 8820 &icmp_dir, &virtual_id, &virtual_type); 8821 8822 ret = pf_icmp_state_lookup(&key, &pd2, state, 8823 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8824 /* IPv6? try matching a multicast address */ 8825 if (ret == PF_DROP && pd2.af == AF_INET6 && 8826 icmp_dir == PF_OUT) { 8827 MPASS(*state == NULL); 8828 ret = pf_icmp_state_lookup(&key, &pd2, 8829 state, virtual_id, virtual_type, 8830 icmp_dir, &iidx, 1, 1); 8831 } 8832 if (ret >= 0) { 8833 MPASS(*state == NULL); 8834 return (ret); 8835 } 8836 8837 /* translate source/destination address, if necessary */ 8838 if ((*state)->key[PF_SK_WIRE] != 8839 (*state)->key[PF_SK_STACK]) { 8840 struct pf_state_key *nk; 8841 8842 if (PF_REVERSED_KEY(*state, pd->af)) 8843 nk = (*state)->key[pd->sidx]; 8844 else 8845 nk = (*state)->key[pd->didx]; 8846 8847 #if defined(INET) && defined(INET6) 8848 int afto, sidx, didx; 8849 8850 afto = pd->af != nk->af; 8851 8852 if (afto && (*state)->direction == PF_IN) { 8853 sidx = pd2.didx; 8854 didx = pd2.sidx; 8855 iidx = !iidx; 8856 } else { 8857 sidx = pd2.sidx; 8858 didx = pd2.didx; 8859 } 8860 8861 if (afto) { 8862 if (nk->af != AF_INET) 8863 return (PF_DROP); 8864 if (pf_translate_icmp_af(nk->af, 8865 &pd->hdr.icmp)) 8866 return (PF_DROP); 8867 m_copyback(pd->m, pd->off, 8868 sizeof(struct icmp6_hdr), 8869 (c_caddr_t)&pd->hdr.icmp6); 8870 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8871 &pd2, &nk->addr[sidx], 8872 &nk->addr[didx], pd->af, 8873 nk->af)) 8874 return (PF_DROP); 8875 pd->proto = IPPROTO_ICMP; 8876 if (pf_translate_icmp_af(nk->af, iih)) 8877 return (PF_DROP); 8878 if (virtual_type == 8879 htons(ICMP6_ECHO_REQUEST) && 8880 nk->port[iidx] != iih->icmp6_id) 8881 iih->icmp6_id = nk->port[iidx]; 8882 m_copyback(pd2.m, pd2.off, 8883 sizeof(struct icmp6_hdr), (c_caddr_t)iih); 8884 pf_addrcpy(&pd->nsaddr, 8885 &nk->addr[pd2.sidx], nk->af); 8886 pf_addrcpy(&pd->ndaddr, 8887 &nk->addr[pd2.didx], nk->af); 8888 pd->naf = nk->af; 8889 return (PF_AFRT); 8890 } 8891 #endif /* INET && INET6 */ 8892 8893 if (PF_ANEQ(pd2.src, 8894 &nk->addr[pd2.sidx], pd2.af) || 8895 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8896 nk->port[pd2.sidx] != iih->icmp6_id)) 8897 pf_change_icmp(pd2.src, 8898 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8899 ? &iih->icmp6_id : NULL, 8900 daddr, &nk->addr[pd2.sidx], 8901 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8902 ? nk->port[iidx] : 0, NULL, 8903 pd2.ip_sum, icmpsum, 8904 pd->ip_sum, 0, AF_INET6); 8905 8906 if (PF_ANEQ(pd2.dst, 8907 &nk->addr[pd2.didx], pd2.af)) 8908 pf_change_icmp(pd2.dst, NULL, NULL, 8909 &nk->addr[pd2.didx], 0, NULL, 8910 pd2.ip_sum, icmpsum, 8911 pd->ip_sum, 0, AF_INET6); 8912 8913 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8914 (caddr_t)&pd->hdr.icmp6); 8915 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8916 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8917 (caddr_t)iih); 8918 } 8919 return (PF_PASS); 8920 break; 8921 } 8922 #endif /* INET6 */ 8923 default: { 8924 int action; 8925 8926 /* 8927 * Placeholder value, so future calls to pf_change_ap() 8928 * don't try to update a NULL checksum pointer. 8929 */ 8930 pd->pcksum = &pd->sctp_dummy_sum; 8931 key.af = pd2.af; 8932 key.proto = pd2.proto; 8933 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8934 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8935 key.port[0] = key.port[1] = 0; 8936 8937 action = pf_find_state(&pd2, &key, state); 8938 if (action != PF_MATCH) 8939 return (action); 8940 8941 /* translate source/destination address, if necessary */ 8942 if ((*state)->key[PF_SK_WIRE] != 8943 (*state)->key[PF_SK_STACK]) { 8944 struct pf_state_key *nk = 8945 (*state)->key[pd->didx]; 8946 8947 if (PF_ANEQ(pd2.src, 8948 &nk->addr[pd2.sidx], pd2.af)) 8949 pf_change_icmp(pd2.src, NULL, daddr, 8950 &nk->addr[pd2.sidx], 0, NULL, 8951 pd2.ip_sum, icmpsum, 8952 pd->ip_sum, 0, pd2.af); 8953 8954 if (PF_ANEQ(pd2.dst, 8955 &nk->addr[pd2.didx], pd2.af)) 8956 pf_change_icmp(pd2.dst, NULL, saddr, 8957 &nk->addr[pd2.didx], 0, NULL, 8958 pd2.ip_sum, icmpsum, 8959 pd->ip_sum, 0, pd2.af); 8960 8961 switch (pd2.af) { 8962 #ifdef INET 8963 case AF_INET: 8964 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8965 (caddr_t)&pd->hdr.icmp); 8966 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8967 break; 8968 #endif /* INET */ 8969 #ifdef INET6 8970 case AF_INET6: 8971 m_copyback(pd->m, pd->off, 8972 sizeof(struct icmp6_hdr), 8973 (caddr_t )&pd->hdr.icmp6); 8974 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8975 (caddr_t )&h2_6); 8976 break; 8977 #endif /* INET6 */ 8978 } 8979 } 8980 return (PF_PASS); 8981 break; 8982 } 8983 } 8984 } 8985 } 8986 8987 /* 8988 * ipoff and off are measured from the start of the mbuf chain. 8989 * h must be at "ipoff" on the mbuf chain. 8990 */ 8991 void * 8992 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8993 u_short *reasonp, sa_family_t af) 8994 { 8995 int iplen = 0; 8996 switch (af) { 8997 #ifdef INET 8998 case AF_INET: { 8999 const struct ip *h = mtod(m, struct ip *); 9000 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 9001 9002 if (fragoff) { 9003 REASON_SET(reasonp, PFRES_FRAG); 9004 return (NULL); 9005 } 9006 iplen = ntohs(h->ip_len); 9007 break; 9008 } 9009 #endif /* INET */ 9010 #ifdef INET6 9011 case AF_INET6: { 9012 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 9013 9014 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 9015 break; 9016 } 9017 #endif /* INET6 */ 9018 } 9019 if (m->m_pkthdr.len < off + len || iplen < off + len) { 9020 REASON_SET(reasonp, PFRES_SHORT); 9021 return (NULL); 9022 } 9023 m_copydata(m, off, len, p); 9024 return (p); 9025 } 9026 9027 int 9028 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 9029 int rtableid) 9030 { 9031 struct ifnet *ifp; 9032 9033 /* 9034 * Skip check for addresses with embedded interface scope, 9035 * as they would always match anyway. 9036 */ 9037 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 9038 return (1); 9039 9040 if (af != AF_INET && af != AF_INET6) 9041 return (0); 9042 9043 if (kif == V_pfi_all) 9044 return (1); 9045 9046 /* Skip checks for ipsec interfaces */ 9047 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 9048 return (1); 9049 9050 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 9051 9052 switch (af) { 9053 #ifdef INET6 9054 case AF_INET6: 9055 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 9056 ifp)); 9057 #endif /* INET6 */ 9058 #ifdef INET 9059 case AF_INET: 9060 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 9061 ifp)); 9062 #endif /* INET */ 9063 } 9064 9065 return (0); 9066 } 9067 9068 #ifdef INET 9069 static int 9070 pf_route(struct pf_krule *r, struct ifnet *oifp, 9071 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9072 { 9073 struct mbuf *m0, *m1, *md; 9074 struct route_in6 ro; 9075 union sockaddr_union rt_gw; 9076 const union sockaddr_union *gw = (const union sockaddr_union *)&ro.ro_dst; 9077 union sockaddr_union *dst; 9078 struct ip *ip; 9079 struct ifnet *ifp = NULL; 9080 int error = 0; 9081 uint16_t ip_len, ip_off; 9082 uint16_t tmp; 9083 int r_dir; 9084 bool skip_test = false; 9085 int action = PF_PASS; 9086 9087 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 9088 9089 SDT_PROBE4(pf, ip, route_to, entry, pd->m, pd, s, oifp); 9090 9091 if (s) { 9092 r_dir = s->direction; 9093 } else { 9094 r_dir = r->direction; 9095 } 9096 9097 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9098 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9099 __func__)); 9100 9101 if ((pd->pf_mtag == NULL && 9102 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 9103 pd->pf_mtag->routed++ > 3) { 9104 m0 = pd->m; 9105 pd->m = NULL; 9106 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9107 action = PF_DROP; 9108 goto bad_locked; 9109 } 9110 9111 if (pd->act.rt_kif != NULL) 9112 ifp = pd->act.rt_kif->pfik_ifp; 9113 9114 if (pd->act.rt == PF_DUPTO) { 9115 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9116 if (s != NULL) { 9117 PF_STATE_UNLOCK(s); 9118 } 9119 if (ifp == oifp) { 9120 /* When the 2nd interface is not skipped */ 9121 return (action); 9122 } else { 9123 m0 = pd->m; 9124 pd->m = NULL; 9125 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9126 action = PF_DROP; 9127 goto bad; 9128 } 9129 } else { 9130 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9131 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 9132 if (s) 9133 PF_STATE_UNLOCK(s); 9134 return (action); 9135 } 9136 } 9137 } else { 9138 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9139 if (pd->af == pd->naf) { 9140 pf_dummynet(pd, s, r, &pd->m); 9141 if (s) 9142 PF_STATE_UNLOCK(s); 9143 return (action); 9144 } else { 9145 if (r_dir == PF_IN) { 9146 skip_test = true; 9147 } 9148 } 9149 } 9150 9151 /* 9152 * If we're actually doing route-to and af-to and are in the 9153 * reply direction. 9154 */ 9155 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9156 pd->af != pd->naf) { 9157 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 9158 /* Un-set ifp so we do a plain route lookup. */ 9159 ifp = NULL; 9160 } 9161 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 9162 /* Un-set ifp so we do a plain route lookup. */ 9163 ifp = NULL; 9164 } 9165 } 9166 m0 = pd->m; 9167 } 9168 9169 ip = mtod(m0, struct ip *); 9170 9171 bzero(&ro, sizeof(ro)); 9172 dst = (union sockaddr_union *)&ro.ro_dst; 9173 dst->sin.sin_family = AF_INET; 9174 dst->sin.sin_len = sizeof(struct sockaddr_in); 9175 dst->sin.sin_addr = ip->ip_dst; 9176 if (ifp) { /* Only needed in forward direction and route-to */ 9177 bzero(&rt_gw, sizeof(rt_gw)); 9178 ro.ro_flags |= RT_HAS_GW; 9179 gw = &rt_gw; 9180 switch (pd->act.rt_af) { 9181 #ifdef INET 9182 case AF_INET: 9183 rt_gw.sin.sin_family = AF_INET; 9184 rt_gw.sin.sin_len = sizeof(struct sockaddr_in); 9185 rt_gw.sin.sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 9186 break; 9187 #endif /* INET */ 9188 #ifdef INET6 9189 case AF_INET6: 9190 rt_gw.sin6.sin6_family = AF_INET6; 9191 rt_gw.sin6.sin6_len = sizeof(struct sockaddr_in6); 9192 pf_addrcpy((struct pf_addr *)&rt_gw.sin6.sin6_addr, 9193 &pd->act.rt_addr, AF_INET6); 9194 break; 9195 #endif /* INET6 */ 9196 default: 9197 /* Normal af-to without route-to */ 9198 break; 9199 } 9200 } 9201 9202 if (pd->dir == PF_IN) { 9203 if (ip->ip_ttl <= IPTTLDEC) { 9204 if (r->rt != PF_DUPTO && pd->naf == pd->af) 9205 pf_send_icmp(m0, ICMP_TIMXCEED, 9206 ICMP_TIMXCEED_INTRANS, 0, pd->af, r, 9207 pd->act.rtableid); 9208 action = PF_DROP; 9209 goto bad_locked; 9210 } 9211 ip->ip_ttl -= IPTTLDEC; 9212 } 9213 9214 if (s != NULL) { 9215 if (ifp == NULL && (pd->af != pd->naf)) { 9216 /* We're in the AFTO case. Do a route lookup. */ 9217 const struct nhop_object *nh; 9218 nh = fib4_lookup(M_GETFIB(m0), ip->ip_dst, 0, NHR_NONE, 0); 9219 if (nh) { 9220 ifp = nh->nh_ifp; 9221 9222 /* Use the gateway if needed. */ 9223 if (nh->nh_flags & NHF_GATEWAY) { 9224 gw = (const union sockaddr_union *)&nh->gw_sa; 9225 ro.ro_flags |= RT_HAS_GW; 9226 } else { 9227 dst->sin.sin_addr = ip->ip_dst; 9228 } 9229 } 9230 } 9231 PF_STATE_UNLOCK(s); 9232 } 9233 9234 /* It must have been either set from rt_af or from fib4_lookup */ 9235 KASSERT(gw->sin.sin_family != 0, ("%s: gw address family undetermined", __func__)); 9236 9237 if (ifp == NULL) { 9238 m0 = pd->m; 9239 pd->m = NULL; 9240 action = PF_DROP; 9241 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9242 goto bad; 9243 } 9244 9245 /* 9246 * Bind to the correct interface if we're if-bound. We don't know which 9247 * interface that will be until here, so we've inserted the state 9248 * on V_pf_all. Fix that now. 9249 */ 9250 if (s != NULL && s->kif == V_pfi_all && r->rule_flag & PFRULE_IFBOUND) { 9251 /* Verify that we're here because of BOUND_IFACE */ 9252 MPASS(r->rt == PF_REPLYTO || (pd->af != pd->naf && s->direction == PF_IN)); 9253 s->kif = ifp->if_pf_kif; 9254 if (pd->act.rt == PF_REPLYTO) { 9255 s->orig_kif = oifp->if_pf_kif; 9256 } 9257 } 9258 9259 if (r->rt == PF_DUPTO || (pd->af != pd->naf && s->direction == PF_IN)) 9260 skip_test = true; 9261 9262 if (pd->dir == PF_IN) { 9263 if (skip_test) { 9264 struct pfi_kkif *out_kif = (struct pfi_kkif *)ifp->if_pf_kif; 9265 MPASS(s != NULL); 9266 pf_counter_u64_critical_enter(); 9267 pf_counter_u64_add_protected( 9268 &out_kif->pfik_bytes[pd->naf == AF_INET6][1] 9269 [action != PF_PASS && action != PF_AFRT], pd->tot_len); 9270 pf_counter_u64_add_protected( 9271 &out_kif->pfik_packets[pd->naf == AF_INET6][1] 9272 [action != PF_PASS && action != PF_AFRT], 1); 9273 pf_counter_u64_critical_exit(); 9274 } else { 9275 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 9276 &pd->act) != PF_PASS) { 9277 action = PF_DROP; 9278 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9279 goto bad; 9280 } else if (m0 == NULL) { 9281 action = PF_DROP; 9282 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9283 goto done; 9284 } 9285 if (m0->m_len < sizeof(struct ip)) { 9286 DPFPRINTF(PF_DEBUG_URGENT, 9287 "%s: m0->m_len < sizeof(struct ip)", __func__); 9288 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9289 action = PF_DROP; 9290 goto bad; 9291 } 9292 ip = mtod(m0, struct ip *); 9293 } 9294 } 9295 9296 if (ifp->if_flags & IFF_LOOPBACK) 9297 m0->m_flags |= M_SKIP_FIREWALL; 9298 9299 ip_len = ntohs(ip->ip_len); 9300 ip_off = ntohs(ip->ip_off); 9301 9302 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 9303 m0->m_pkthdr.csum_flags |= CSUM_IP; 9304 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 9305 in_delayed_cksum(m0); 9306 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 9307 } 9308 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 9309 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 9310 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 9311 } 9312 9313 if (pd->dir == PF_IN) { 9314 /* 9315 * Make sure dummynet gets the correct direction, in case it needs to 9316 * re-inject later. 9317 */ 9318 pd->dir = PF_OUT; 9319 9320 /* 9321 * The following processing is actually the rest of the inbound processing, even 9322 * though we've marked it as outbound (so we don't look through dummynet) and it 9323 * happens after the outbound processing (pf_test(PF_OUT) above). 9324 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9325 * conclusion about what direction it's processing, and we can't fix it or it 9326 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9327 * decision will pick the right pipe, and everything will mostly work as expected. 9328 */ 9329 tmp = pd->act.dnrpipe; 9330 pd->act.dnrpipe = pd->act.dnpipe; 9331 pd->act.dnpipe = tmp; 9332 } 9333 9334 /* 9335 * If small enough for interface, or the interface will take 9336 * care of the fragmentation for us, we can just send directly. 9337 */ 9338 if (ip_len <= ifp->if_mtu || 9339 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 9340 ip->ip_sum = 0; 9341 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 9342 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 9343 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 9344 } 9345 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 9346 9347 md = m0; 9348 error = pf_dummynet_route(pd, s, r, ifp, 9349 (const struct sockaddr *)gw, &md); 9350 if (md != NULL) { 9351 error = (*ifp->if_output)(ifp, md, 9352 (const struct sockaddr *)gw, (struct route *)&ro); 9353 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9354 } 9355 goto done; 9356 } 9357 9358 /* Balk when DF bit is set or the interface didn't support TSO. */ 9359 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 9360 error = EMSGSIZE; 9361 KMOD_IPSTAT_INC(ips_cantfrag); 9362 if (pd->act.rt != PF_DUPTO) { 9363 if (s && s->nat_rule != NULL) { 9364 MPASS(m0 == pd->m); 9365 PACKET_UNDO_NAT(pd, 9366 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 9367 s); 9368 } 9369 9370 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 9371 ifp->if_mtu, pd->af, r, pd->act.rtableid); 9372 } 9373 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9374 action = PF_DROP; 9375 goto bad; 9376 } 9377 9378 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 9379 if (error) { 9380 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9381 action = PF_DROP; 9382 goto bad; 9383 } 9384 9385 for (; m0; m0 = m1) { 9386 m1 = m0->m_nextpkt; 9387 m0->m_nextpkt = NULL; 9388 if (error == 0) { 9389 m_clrprotoflags(m0); 9390 md = m0; 9391 pd->pf_mtag = pf_find_mtag(md); 9392 error = pf_dummynet_route(pd, s, r, ifp, 9393 (const struct sockaddr *)gw, &md); 9394 if (md != NULL) { 9395 error = (*ifp->if_output)(ifp, md, 9396 (const struct sockaddr *)gw, 9397 (struct route *)&ro); 9398 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9399 } 9400 } else 9401 m_freem(m0); 9402 } 9403 9404 if (error == 0) 9405 KMOD_IPSTAT_INC(ips_fragmented); 9406 9407 done: 9408 if (pd->act.rt != PF_DUPTO) 9409 pd->m = NULL; 9410 else 9411 action = PF_PASS; 9412 return (action); 9413 9414 bad_locked: 9415 if (s) 9416 PF_STATE_UNLOCK(s); 9417 bad: 9418 m_freem(m0); 9419 goto done; 9420 } 9421 #endif /* INET */ 9422 9423 #ifdef INET6 9424 static int 9425 pf_route6(struct pf_krule *r, struct ifnet *oifp, 9426 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9427 { 9428 struct mbuf *m0, *md; 9429 struct m_tag *mtag; 9430 struct sockaddr_in6 dst; 9431 struct ip6_hdr *ip6; 9432 struct ifnet *ifp = NULL; 9433 int r_dir; 9434 bool skip_test = false; 9435 int action = PF_PASS; 9436 9437 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 9438 9439 SDT_PROBE4(pf, ip6, route_to, entry, pd->m, pd, s, oifp); 9440 9441 if (s) { 9442 r_dir = s->direction; 9443 } else { 9444 r_dir = r->direction; 9445 } 9446 9447 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9448 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9449 __func__)); 9450 9451 if ((pd->pf_mtag == NULL && 9452 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 9453 pd->pf_mtag->routed++ > 3) { 9454 m0 = pd->m; 9455 pd->m = NULL; 9456 action = PF_DROP; 9457 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9458 goto bad_locked; 9459 } 9460 9461 if (pd->act.rt_kif != NULL) 9462 ifp = pd->act.rt_kif->pfik_ifp; 9463 9464 if (pd->act.rt == PF_DUPTO) { 9465 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9466 if (s != NULL) { 9467 PF_STATE_UNLOCK(s); 9468 } 9469 if (ifp == oifp) { 9470 /* When the 2nd interface is not skipped */ 9471 return (action); 9472 } else { 9473 m0 = pd->m; 9474 pd->m = NULL; 9475 action = PF_DROP; 9476 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9477 goto bad; 9478 } 9479 } else { 9480 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9481 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 9482 if (s) 9483 PF_STATE_UNLOCK(s); 9484 return (action); 9485 } 9486 } 9487 } else { 9488 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9489 if (pd->af == pd->naf) { 9490 pf_dummynet(pd, s, r, &pd->m); 9491 if (s) 9492 PF_STATE_UNLOCK(s); 9493 return (action); 9494 } else { 9495 if (r_dir == PF_IN) { 9496 skip_test = true; 9497 } 9498 } 9499 } 9500 9501 /* 9502 * If we're actually doing route-to and af-to and are in the 9503 * reply direction. 9504 */ 9505 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9506 pd->af != pd->naf) { 9507 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9508 /* Un-set ifp so we do a plain route lookup. */ 9509 ifp = NULL; 9510 } 9511 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9512 /* Un-set ifp so we do a plain route lookup. */ 9513 ifp = NULL; 9514 } 9515 } 9516 m0 = pd->m; 9517 } 9518 9519 ip6 = mtod(m0, struct ip6_hdr *); 9520 9521 bzero(&dst, sizeof(dst)); 9522 dst.sin6_family = AF_INET6; 9523 dst.sin6_len = sizeof(dst); 9524 pf_addrcpy((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, 9525 AF_INET6); 9526 9527 if (pd->dir == PF_IN) { 9528 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 9529 if (r->rt != PF_DUPTO && pd->naf == pd->af) 9530 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 9531 ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r, 9532 pd->act.rtableid); 9533 action = PF_DROP; 9534 goto bad_locked; 9535 } 9536 ip6->ip6_hlim -= IPV6_HLIMDEC; 9537 } 9538 9539 if (s != NULL) { 9540 if (ifp == NULL && (pd->af != pd->naf)) { 9541 const struct nhop_object *nh; 9542 nh = fib6_lookup(M_GETFIB(m0), &ip6->ip6_dst, 0, NHR_NONE, 0); 9543 if (nh) { 9544 ifp = nh->nh_ifp; 9545 9546 /* Use the gateway if needed. */ 9547 if (nh->nh_flags & NHF_GATEWAY) 9548 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9549 sizeof(dst.sin6_addr)); 9550 else 9551 dst.sin6_addr = ip6->ip6_dst; 9552 } 9553 } 9554 PF_STATE_UNLOCK(s); 9555 } 9556 9557 if (pd->af != pd->naf) { 9558 struct udphdr *uh = &pd->hdr.udp; 9559 9560 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9561 uh->uh_sum = in6_cksum_pseudo(ip6, 9562 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9563 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9564 } 9565 } 9566 9567 if (ifp == NULL) { 9568 m0 = pd->m; 9569 pd->m = NULL; 9570 action = PF_DROP; 9571 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9572 goto bad; 9573 } 9574 9575 /* 9576 * Bind to the correct interface if we're if-bound. We don't know which 9577 * interface that will be until here, so we've inserted the state 9578 * on V_pf_all. Fix that now. 9579 */ 9580 if (s != NULL && s->kif == V_pfi_all && r->rule_flag & PFRULE_IFBOUND) { 9581 /* Verify that we're here because of BOUND_IFACE */ 9582 MPASS(r->rt == PF_REPLYTO || (pd->af != pd->naf && s->direction == PF_IN)); 9583 s->kif = ifp->if_pf_kif; 9584 if (pd->act.rt == PF_REPLYTO) { 9585 s->orig_kif = oifp->if_pf_kif; 9586 } 9587 } 9588 9589 if (r->rt == PF_DUPTO || (pd->af != pd->naf && s->direction == PF_IN)) 9590 skip_test = true; 9591 9592 if (pd->dir == PF_IN) { 9593 if (skip_test) { 9594 struct pfi_kkif *out_kif = (struct pfi_kkif *)ifp->if_pf_kif; 9595 MPASS(s != NULL); 9596 pf_counter_u64_critical_enter(); 9597 pf_counter_u64_add_protected( 9598 &out_kif->pfik_bytes[pd->naf == AF_INET6][1] 9599 [action != PF_PASS && action != PF_AFRT], pd->tot_len); 9600 pf_counter_u64_add_protected( 9601 &out_kif->pfik_packets[pd->naf == AF_INET6][1] 9602 [action != PF_PASS && action != PF_AFRT], 1); 9603 pf_counter_u64_critical_exit(); 9604 } else { 9605 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9606 ifp, &m0, inp, &pd->act) != PF_PASS) { 9607 action = PF_DROP; 9608 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9609 goto bad; 9610 } else if (m0 == NULL) { 9611 action = PF_DROP; 9612 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9613 goto done; 9614 } 9615 if (m0->m_len < sizeof(struct ip6_hdr)) { 9616 DPFPRINTF(PF_DEBUG_URGENT, 9617 "%s: m0->m_len < sizeof(struct ip6_hdr)", 9618 __func__); 9619 action = PF_DROP; 9620 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9621 goto bad; 9622 } 9623 ip6 = mtod(m0, struct ip6_hdr *); 9624 } 9625 } 9626 9627 if (ifp->if_flags & IFF_LOOPBACK) 9628 m0->m_flags |= M_SKIP_FIREWALL; 9629 9630 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9631 ~ifp->if_hwassist) { 9632 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9633 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9634 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9635 } 9636 9637 if (pd->dir == PF_IN) { 9638 uint16_t tmp; 9639 /* 9640 * Make sure dummynet gets the correct direction, in case it needs to 9641 * re-inject later. 9642 */ 9643 pd->dir = PF_OUT; 9644 9645 /* 9646 * The following processing is actually the rest of the inbound processing, even 9647 * though we've marked it as outbound (so we don't look through dummynet) and it 9648 * happens after the outbound processing (pf_test(PF_OUT) above). 9649 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9650 * conclusion about what direction it's processing, and we can't fix it or it 9651 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9652 * decision will pick the right pipe, and everything will mostly work as expected. 9653 */ 9654 tmp = pd->act.dnrpipe; 9655 pd->act.dnrpipe = pd->act.dnpipe; 9656 pd->act.dnpipe = tmp; 9657 } 9658 9659 /* 9660 * If the packet is too large for the outgoing interface, 9661 * send back an icmp6 error. 9662 */ 9663 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9664 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9665 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9666 if (mtag != NULL) { 9667 int ret __sdt_used; 9668 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9669 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9670 goto done; 9671 } 9672 9673 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9674 md = m0; 9675 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9676 if (md != NULL) { 9677 int ret __sdt_used; 9678 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9679 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9680 } 9681 } 9682 else { 9683 in6_ifstat_inc(ifp, ifs6_in_toobig); 9684 if (pd->act.rt != PF_DUPTO) { 9685 if (s && s->nat_rule != NULL) { 9686 MPASS(m0 == pd->m); 9687 PACKET_UNDO_NAT(pd, 9688 ((caddr_t)ip6 - m0->m_data) + 9689 sizeof(struct ip6_hdr), s); 9690 } 9691 9692 if (r->rt != PF_DUPTO) 9693 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 9694 ifp->if_mtu, pd->af, r, pd->act.rtableid); 9695 } 9696 action = PF_DROP; 9697 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9698 goto bad; 9699 } 9700 9701 done: 9702 if (pd->act.rt != PF_DUPTO) 9703 pd->m = NULL; 9704 else 9705 action = PF_PASS; 9706 return (action); 9707 9708 bad_locked: 9709 if (s) 9710 PF_STATE_UNLOCK(s); 9711 bad: 9712 m_freem(m0); 9713 goto done; 9714 } 9715 #endif /* INET6 */ 9716 9717 /* 9718 * FreeBSD supports cksum offloads for the following drivers. 9719 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9720 * 9721 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9722 * network driver performed cksum including pseudo header, need to verify 9723 * csum_data 9724 * CSUM_DATA_VALID : 9725 * network driver performed cksum, needs to additional pseudo header 9726 * cksum computation with partial csum_data(i.e. lack of H/W support for 9727 * pseudo header, for instance sk(4) and possibly gem(4)) 9728 * 9729 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9730 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9731 * TCP/UDP layer. 9732 * Also, set csum_data to 0xffff to force cksum validation. 9733 */ 9734 static int 9735 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9736 { 9737 u_int16_t sum = 0; 9738 int hw_assist = 0; 9739 struct ip *ip; 9740 9741 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9742 return (1); 9743 if (m->m_pkthdr.len < off + len) 9744 return (1); 9745 9746 switch (p) { 9747 case IPPROTO_TCP: 9748 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9749 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9750 sum = m->m_pkthdr.csum_data; 9751 } else { 9752 ip = mtod(m, struct ip *); 9753 sum = in_pseudo(ip->ip_src.s_addr, 9754 ip->ip_dst.s_addr, htonl((u_short)len + 9755 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9756 } 9757 sum ^= 0xffff; 9758 ++hw_assist; 9759 } 9760 break; 9761 case IPPROTO_UDP: 9762 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9763 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9764 sum = m->m_pkthdr.csum_data; 9765 } else { 9766 ip = mtod(m, struct ip *); 9767 sum = in_pseudo(ip->ip_src.s_addr, 9768 ip->ip_dst.s_addr, htonl((u_short)len + 9769 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9770 } 9771 sum ^= 0xffff; 9772 ++hw_assist; 9773 } 9774 break; 9775 case IPPROTO_ICMP: 9776 #ifdef INET6 9777 case IPPROTO_ICMPV6: 9778 #endif /* INET6 */ 9779 break; 9780 default: 9781 return (1); 9782 } 9783 9784 if (!hw_assist) { 9785 switch (af) { 9786 case AF_INET: 9787 if (m->m_len < sizeof(struct ip)) 9788 return (1); 9789 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 9790 break; 9791 #ifdef INET6 9792 case AF_INET6: 9793 if (m->m_len < sizeof(struct ip6_hdr)) 9794 return (1); 9795 sum = in6_cksum(m, p, off, len); 9796 break; 9797 #endif /* INET6 */ 9798 } 9799 } 9800 if (sum) { 9801 switch (p) { 9802 case IPPROTO_TCP: 9803 { 9804 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9805 break; 9806 } 9807 case IPPROTO_UDP: 9808 { 9809 KMOD_UDPSTAT_INC(udps_badsum); 9810 break; 9811 } 9812 #ifdef INET 9813 case IPPROTO_ICMP: 9814 { 9815 KMOD_ICMPSTAT_INC(icps_checksum); 9816 break; 9817 } 9818 #endif 9819 #ifdef INET6 9820 case IPPROTO_ICMPV6: 9821 { 9822 KMOD_ICMP6STAT_INC(icp6s_checksum); 9823 break; 9824 } 9825 #endif /* INET6 */ 9826 } 9827 return (1); 9828 } else { 9829 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9830 m->m_pkthdr.csum_flags |= 9831 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9832 m->m_pkthdr.csum_data = 0xffff; 9833 } 9834 } 9835 return (0); 9836 } 9837 9838 static bool 9839 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9840 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9841 { 9842 int dndir = r->direction; 9843 sa_family_t af = pd->naf; 9844 9845 if (s && dndir == PF_INOUT) { 9846 dndir = s->direction; 9847 } else if (dndir == PF_INOUT) { 9848 /* Assume primary direction. Happens when we've set dnpipe in 9849 * the ethernet level code. */ 9850 dndir = pd->dir; 9851 } 9852 9853 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9854 return (false); 9855 9856 memset(dnflow, 0, sizeof(*dnflow)); 9857 9858 if (pd->dport != NULL) 9859 dnflow->f_id.dst_port = ntohs(*pd->dport); 9860 if (pd->sport != NULL) 9861 dnflow->f_id.src_port = ntohs(*pd->sport); 9862 9863 if (pd->dir == PF_IN) 9864 dnflow->flags |= IPFW_ARGS_IN; 9865 else 9866 dnflow->flags |= IPFW_ARGS_OUT; 9867 9868 if (pd->dir != dndir && pd->act.dnrpipe) { 9869 dnflow->rule.info = pd->act.dnrpipe; 9870 } 9871 else if (pd->dir == dndir && pd->act.dnpipe) { 9872 dnflow->rule.info = pd->act.dnpipe; 9873 } 9874 else { 9875 return (false); 9876 } 9877 9878 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9879 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9880 dnflow->rule.info |= IPFW_IS_PIPE; 9881 9882 dnflow->f_id.proto = pd->proto; 9883 dnflow->f_id.extra = dnflow->rule.info; 9884 if (s) 9885 af = s->key[PF_SK_STACK]->af; 9886 9887 switch (af) { 9888 case AF_INET: 9889 dnflow->f_id.addr_type = 4; 9890 if (s) { 9891 dnflow->f_id.src_ip = htonl( 9892 s->key[PF_SK_STACK]->addr[pd->sidx].v4.s_addr); 9893 dnflow->f_id.dst_ip = htonl( 9894 s->key[PF_SK_STACK]->addr[pd->didx].v4.s_addr); 9895 } else { 9896 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9897 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9898 } 9899 break; 9900 case AF_INET6: 9901 dnflow->f_id.addr_type = 6; 9902 9903 if (s) { 9904 dnflow->f_id.src_ip6 = 9905 s->key[PF_SK_STACK]->addr[pd->sidx].v6; 9906 dnflow->f_id.dst_ip6 = 9907 s->key[PF_SK_STACK]->addr[pd->didx].v6; 9908 } else { 9909 dnflow->f_id.src_ip6 = pd->src->v6; 9910 dnflow->f_id.dst_ip6 = pd->dst->v6; 9911 } 9912 break; 9913 } 9914 9915 /* 9916 * Separate this out, because while we pass the pre-NAT addresses to 9917 * dummynet we want the post-nat address family in case of nat64. 9918 * Dummynet may call ip_output/ip6_output itself, and we need it to 9919 * call the correct one. 9920 */ 9921 if (pd->naf == AF_INET6) 9922 dnflow->flags |= IPFW_ARGS_IP6; 9923 9924 return (true); 9925 } 9926 9927 int 9928 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9929 struct inpcb *inp) 9930 { 9931 struct pfi_kkif *kif; 9932 struct mbuf *m = *m0; 9933 9934 M_ASSERTPKTHDR(m); 9935 MPASS(ifp->if_vnet == curvnet); 9936 NET_EPOCH_ASSERT(); 9937 9938 if (!V_pf_status.running) 9939 return (PF_PASS); 9940 9941 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9942 9943 if (kif == NULL) { 9944 DPFPRINTF(PF_DEBUG_URGENT, 9945 "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname); 9946 return (PF_DROP); 9947 } 9948 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9949 return (PF_PASS); 9950 9951 if (m->m_flags & M_SKIP_FIREWALL) 9952 return (PF_PASS); 9953 9954 if (__predict_false(! M_WRITABLE(*m0))) { 9955 m = *m0 = m_unshare(*m0, M_NOWAIT); 9956 if (*m0 == NULL) 9957 return (PF_DROP); 9958 } 9959 9960 /* Stateless! */ 9961 return (pf_test_eth_rule(dir, kif, m0)); 9962 } 9963 9964 static __inline void 9965 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9966 { 9967 struct m_tag *mtag; 9968 9969 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9970 9971 /* dummynet adds this tag, but pf does not need it, 9972 * and keeping it creates unexpected behavior, 9973 * e.g. in case of divert(4) usage right after dummynet. */ 9974 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9975 if (mtag != NULL) 9976 m_tag_delete(m, mtag); 9977 } 9978 9979 static int 9980 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9981 struct pf_krule *r, struct mbuf **m0) 9982 { 9983 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9984 } 9985 9986 static int 9987 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9988 struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa, 9989 struct mbuf **m0) 9990 { 9991 struct ip_fw_args dnflow; 9992 9993 NET_EPOCH_ASSERT(); 9994 9995 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9996 return (0); 9997 9998 if (ip_dn_io_ptr == NULL) { 9999 m_freem(*m0); 10000 *m0 = NULL; 10001 return (ENOMEM); 10002 } 10003 10004 if (pd->pf_mtag == NULL && 10005 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 10006 m_freem(*m0); 10007 *m0 = NULL; 10008 return (ENOMEM); 10009 } 10010 10011 if (ifp != NULL) { 10012 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 10013 10014 pd->pf_mtag->if_index = ifp->if_index; 10015 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 10016 10017 MPASS(sa != NULL); 10018 10019 switch (sa->sa_family) { 10020 case AF_INET: 10021 memcpy(&pd->pf_mtag->dst, sa, 10022 sizeof(struct sockaddr_in)); 10023 break; 10024 case AF_INET6: 10025 memcpy(&pd->pf_mtag->dst, sa, 10026 sizeof(struct sockaddr_in6)); 10027 break; 10028 } 10029 } 10030 10031 if (s != NULL && s->nat_rule != NULL && 10032 s->nat_rule->action == PF_RDR && 10033 ( 10034 #ifdef INET 10035 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 10036 #endif /* INET */ 10037 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 10038 /* 10039 * If we're redirecting to loopback mark this packet 10040 * as being local. Otherwise it might get dropped 10041 * if dummynet re-injects. 10042 */ 10043 (*m0)->m_pkthdr.rcvif = V_loif; 10044 } 10045 10046 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 10047 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 10048 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 10049 ip_dn_io_ptr(m0, &dnflow); 10050 if (*m0 != NULL) { 10051 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10052 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 10053 } 10054 } 10055 10056 return (0); 10057 } 10058 10059 static int 10060 pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end, 10061 u_short *reason) 10062 { 10063 uint8_t type, length, opts[15 * 4 - sizeof(struct ip)]; 10064 10065 /* IP header in payload of ICMP packet may be too short */ 10066 if (pd->m->m_pkthdr.len < end) { 10067 DPFPRINTF(PF_DEBUG_MISC, "IP option too short"); 10068 REASON_SET(reason, PFRES_SHORT); 10069 return (PF_DROP); 10070 } 10071 10072 MPASS(end - off <= sizeof(opts)); 10073 m_copydata(pd->m, off, end - off, opts); 10074 end -= off; 10075 off = 0; 10076 10077 while (off < end) { 10078 type = opts[off]; 10079 if (type == IPOPT_EOL) 10080 break; 10081 if (type == IPOPT_NOP) { 10082 off++; 10083 continue; 10084 } 10085 if (off + 2 > end) { 10086 DPFPRINTF(PF_DEBUG_MISC, "IP length opt"); 10087 REASON_SET(reason, PFRES_IPOPTIONS); 10088 return (PF_DROP); 10089 } 10090 length = opts[off + 1]; 10091 if (length < 2) { 10092 DPFPRINTF(PF_DEBUG_MISC, "IP short opt"); 10093 REASON_SET(reason, PFRES_IPOPTIONS); 10094 return (PF_DROP); 10095 } 10096 if (off + length > end) { 10097 DPFPRINTF(PF_DEBUG_MISC, "IP long opt"); 10098 REASON_SET(reason, PFRES_IPOPTIONS); 10099 return (PF_DROP); 10100 } 10101 switch (type) { 10102 case IPOPT_RA: 10103 pd->badopts |= PF_OPT_ROUTER_ALERT; 10104 break; 10105 default: 10106 pd->badopts |= PF_OPT_OTHER; 10107 break; 10108 } 10109 off += length; 10110 } 10111 10112 return (PF_PASS); 10113 } 10114 10115 static int 10116 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 10117 { 10118 struct ah ext; 10119 u_int32_t hlen, end; 10120 int hdr_cnt; 10121 10122 hlen = h->ip_hl << 2; 10123 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 10124 REASON_SET(reason, PFRES_SHORT); 10125 return (PF_DROP); 10126 } 10127 if (hlen != sizeof(struct ip)) { 10128 if (pf_walk_option(pd, h, pd->off + sizeof(struct ip), 10129 pd->off + hlen, reason) != PF_PASS) 10130 return (PF_DROP); 10131 /* header options which contain only padding is fishy */ 10132 if (pd->badopts == 0) 10133 pd->badopts |= PF_OPT_OTHER; 10134 } 10135 end = pd->off + ntohs(h->ip_len); 10136 pd->off += hlen; 10137 pd->proto = h->ip_p; 10138 /* IGMP packets have router alert options, allow them */ 10139 if (pd->proto == IPPROTO_IGMP) { 10140 /* 10141 * According to RFC 1112 ttl must be set to 1 in all IGMP 10142 * packets sent to 224.0.0.1 10143 */ 10144 if ((h->ip_ttl != 1) && 10145 (h->ip_dst.s_addr == INADDR_ALLHOSTS_GROUP)) { 10146 DPFPRINTF(PF_DEBUG_MISC, "Invalid IGMP"); 10147 REASON_SET(reason, PFRES_IPOPTIONS); 10148 return (PF_DROP); 10149 } 10150 pd->badopts &= ~PF_OPT_ROUTER_ALERT; 10151 } 10152 /* stop walking over non initial fragments */ 10153 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 10154 return (PF_PASS); 10155 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 10156 switch (pd->proto) { 10157 case IPPROTO_AH: 10158 /* fragments may be short */ 10159 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 10160 end < pd->off + sizeof(ext)) 10161 return (PF_PASS); 10162 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10163 reason, AF_INET)) { 10164 DPFPRINTF(PF_DEBUG_MISC, "IP short exthdr"); 10165 return (PF_DROP); 10166 } 10167 pd->off += (ext.ah_len + 2) * 4; 10168 pd->proto = ext.ah_nxt; 10169 break; 10170 default: 10171 return (PF_PASS); 10172 } 10173 } 10174 DPFPRINTF(PF_DEBUG_MISC, "IPv4 nested authentication header limit"); 10175 REASON_SET(reason, PFRES_IPOPTIONS); 10176 return (PF_DROP); 10177 } 10178 10179 #ifdef INET6 10180 static int 10181 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 10182 u_short *reason) 10183 { 10184 struct ip6_opt opt; 10185 struct ip6_opt_jumbo jumbo; 10186 10187 while (off < end) { 10188 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 10189 sizeof(opt.ip6o_type), reason, AF_INET6)) { 10190 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt type"); 10191 return (PF_DROP); 10192 } 10193 if (opt.ip6o_type == IP6OPT_PAD1) { 10194 off++; 10195 continue; 10196 } 10197 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), 10198 reason, AF_INET6)) { 10199 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt"); 10200 return (PF_DROP); 10201 } 10202 if (off + sizeof(opt) + opt.ip6o_len > end) { 10203 DPFPRINTF(PF_DEBUG_MISC, "IPv6 long opt"); 10204 REASON_SET(reason, PFRES_IPOPTIONS); 10205 return (PF_DROP); 10206 } 10207 switch (opt.ip6o_type) { 10208 case IP6OPT_PADN: 10209 break; 10210 case IP6OPT_JUMBO: 10211 pd->badopts |= PF_OPT_JUMBO; 10212 if (pd->jumbolen != 0) { 10213 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple jumbo"); 10214 REASON_SET(reason, PFRES_IPOPTIONS); 10215 return (PF_DROP); 10216 } 10217 if (ntohs(h->ip6_plen) != 0) { 10218 DPFPRINTF(PF_DEBUG_MISC, "IPv6 bad jumbo plen"); 10219 REASON_SET(reason, PFRES_IPOPTIONS); 10220 return (PF_DROP); 10221 } 10222 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), 10223 reason, AF_INET6)) { 10224 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbo"); 10225 return (PF_DROP); 10226 } 10227 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 10228 sizeof(pd->jumbolen)); 10229 pd->jumbolen = ntohl(pd->jumbolen); 10230 if (pd->jumbolen < IPV6_MAXPACKET) { 10231 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbolen"); 10232 REASON_SET(reason, PFRES_IPOPTIONS); 10233 return (PF_DROP); 10234 } 10235 break; 10236 case IP6OPT_ROUTER_ALERT: 10237 pd->badopts |= PF_OPT_ROUTER_ALERT; 10238 break; 10239 default: 10240 pd->badopts |= PF_OPT_OTHER; 10241 break; 10242 } 10243 off += sizeof(opt) + opt.ip6o_len; 10244 } 10245 10246 return (PF_PASS); 10247 } 10248 10249 int 10250 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 10251 { 10252 struct ip6_frag frag; 10253 struct ip6_ext ext; 10254 struct icmp6_hdr icmp6; 10255 struct ip6_rthdr rthdr; 10256 uint32_t end; 10257 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 10258 10259 pd->off += sizeof(struct ip6_hdr); 10260 end = pd->off + ntohs(h->ip6_plen); 10261 pd->fragoff = pd->extoff = pd->jumbolen = 0; 10262 pd->proto = h->ip6_nxt; 10263 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 10264 switch (pd->proto) { 10265 case IPPROTO_ROUTING: 10266 case IPPROTO_DSTOPTS: 10267 pd->badopts |= PF_OPT_OTHER; 10268 break; 10269 case IPPROTO_HOPOPTS: 10270 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10271 reason, AF_INET6)) { 10272 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr"); 10273 return (PF_DROP); 10274 } 10275 if (pf_walk_option6(pd, h, pd->off + sizeof(ext), 10276 pd->off + (ext.ip6e_len + 1) * 8, 10277 reason) != PF_PASS) 10278 return (PF_DROP); 10279 /* option header which contains only padding is fishy */ 10280 if (pd->badopts == 0) 10281 pd->badopts |= PF_OPT_OTHER; 10282 break; 10283 } 10284 switch (pd->proto) { 10285 case IPPROTO_FRAGMENT: 10286 if (fraghdr_cnt++) { 10287 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple fragment"); 10288 REASON_SET(reason, PFRES_FRAG); 10289 return (PF_DROP); 10290 } 10291 /* jumbo payload packets cannot be fragmented */ 10292 if (pd->jumbolen != 0) { 10293 DPFPRINTF(PF_DEBUG_MISC, "IPv6 fragmented jumbo"); 10294 REASON_SET(reason, PFRES_FRAG); 10295 return (PF_DROP); 10296 } 10297 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 10298 reason, AF_INET6)) { 10299 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short fragment"); 10300 return (PF_DROP); 10301 } 10302 /* stop walking over non initial fragments */ 10303 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 10304 pd->fragoff = pd->off; 10305 return (PF_PASS); 10306 } 10307 /* RFC6946: reassemble only non atomic fragments */ 10308 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 10309 pd->fragoff = pd->off; 10310 pd->off += sizeof(frag); 10311 pd->proto = frag.ip6f_nxt; 10312 break; 10313 case IPPROTO_ROUTING: 10314 if (rthdr_cnt++) { 10315 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple rthdr"); 10316 REASON_SET(reason, PFRES_IPOPTIONS); 10317 return (PF_DROP); 10318 } 10319 /* fragments may be short */ 10320 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 10321 pd->off = pd->fragoff; 10322 pd->proto = IPPROTO_FRAGMENT; 10323 return (PF_PASS); 10324 } 10325 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 10326 reason, AF_INET6)) { 10327 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short rthdr"); 10328 return (PF_DROP); 10329 } 10330 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 10331 DPFPRINTF(PF_DEBUG_MISC, "IPv6 rthdr0"); 10332 REASON_SET(reason, PFRES_IPOPTIONS); 10333 return (PF_DROP); 10334 } 10335 /* FALLTHROUGH */ 10336 case IPPROTO_HOPOPTS: 10337 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 10338 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 10339 DPFPRINTF(PF_DEBUG_MISC, "IPv6 hopopts not first"); 10340 REASON_SET(reason, PFRES_IPOPTIONS); 10341 return (PF_DROP); 10342 } 10343 /* FALLTHROUGH */ 10344 case IPPROTO_AH: 10345 case IPPROTO_DSTOPTS: 10346 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10347 reason, AF_INET6)) { 10348 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr"); 10349 return (PF_DROP); 10350 } 10351 /* fragments may be short */ 10352 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 10353 pd->off = pd->fragoff; 10354 pd->proto = IPPROTO_FRAGMENT; 10355 return (PF_PASS); 10356 } 10357 /* reassembly needs the ext header before the frag */ 10358 if (pd->fragoff == 0) 10359 pd->extoff = pd->off; 10360 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 && 10361 ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 10362 DPFPRINTF(PF_DEBUG_MISC, "IPv6 missing jumbo"); 10363 REASON_SET(reason, PFRES_IPOPTIONS); 10364 return (PF_DROP); 10365 } 10366 if (pd->proto == IPPROTO_AH) 10367 pd->off += (ext.ip6e_len + 2) * 4; 10368 else 10369 pd->off += (ext.ip6e_len + 1) * 8; 10370 pd->proto = ext.ip6e_nxt; 10371 break; 10372 case IPPROTO_ICMPV6: 10373 /* fragments may be short, ignore inner header then */ 10374 if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) { 10375 pd->off = pd->fragoff; 10376 pd->proto = IPPROTO_FRAGMENT; 10377 return (PF_PASS); 10378 } 10379 if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6), 10380 reason, AF_INET6)) { 10381 DPFPRINTF(PF_DEBUG_MISC, 10382 "IPv6 short icmp6hdr"); 10383 return (PF_DROP); 10384 } 10385 /* ICMP multicast packets have router alert options */ 10386 switch (icmp6.icmp6_type) { 10387 case MLD_LISTENER_QUERY: 10388 case MLD_LISTENER_REPORT: 10389 case MLD_LISTENER_DONE: 10390 case MLDV2_LISTENER_REPORT: 10391 /* 10392 * According to RFC 2710 all MLD messages are 10393 * sent with hop-limit (ttl) set to 1, and link 10394 * local source address. If either one is 10395 * missing then MLD message is invalid and 10396 * should be discarded. 10397 */ 10398 if ((h->ip6_hlim != 1) || 10399 !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) { 10400 DPFPRINTF(PF_DEBUG_MISC, "Invalid MLD"); 10401 REASON_SET(reason, PFRES_IPOPTIONS); 10402 return (PF_DROP); 10403 } 10404 pd->badopts &= ~PF_OPT_ROUTER_ALERT; 10405 break; 10406 } 10407 return (PF_PASS); 10408 case IPPROTO_TCP: 10409 case IPPROTO_UDP: 10410 case IPPROTO_SCTP: 10411 /* fragments may be short, ignore inner header then */ 10412 if (pd->fragoff != 0 && end < pd->off + 10413 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 10414 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 10415 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 10416 sizeof(struct icmp6_hdr))) { 10417 pd->off = pd->fragoff; 10418 pd->proto = IPPROTO_FRAGMENT; 10419 } 10420 /* FALLTHROUGH */ 10421 default: 10422 return (PF_PASS); 10423 } 10424 } 10425 DPFPRINTF(PF_DEBUG_MISC, "IPv6 nested extension header limit"); 10426 REASON_SET(reason, PFRES_IPOPTIONS); 10427 return (PF_DROP); 10428 } 10429 #endif /* INET6 */ 10430 10431 static void 10432 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 10433 { 10434 memset(pd, 0, sizeof(*pd)); 10435 pd->pf_mtag = pf_find_mtag(m); 10436 pd->m = m; 10437 } 10438 10439 static int 10440 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 10441 u_short *action, u_short *reason, struct pfi_kkif *kif, 10442 struct pf_rule_actions *default_actions) 10443 { 10444 pd->dir = dir; 10445 pd->kif = kif; 10446 pd->m = *m0; 10447 pd->sidx = (dir == PF_IN) ? 0 : 1; 10448 pd->didx = (dir == PF_IN) ? 1 : 0; 10449 pd->af = pd->naf = af; 10450 10451 PF_RULES_ASSERT(); 10452 10453 TAILQ_INIT(&pd->sctp_multihome_jobs); 10454 if (default_actions != NULL) 10455 memcpy(&pd->act, default_actions, sizeof(pd->act)); 10456 10457 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 10458 pd->act.dnpipe = pd->pf_mtag->dnpipe; 10459 pd->act.flags = pd->pf_mtag->dnflags; 10460 } 10461 10462 switch (af) { 10463 #ifdef INET 10464 case AF_INET: { 10465 struct ip *h; 10466 10467 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 10468 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 10469 DPFPRINTF(PF_DEBUG_URGENT, 10470 "%s: m_len < sizeof(struct ip), pullup failed", 10471 __func__); 10472 *action = PF_DROP; 10473 REASON_SET(reason, PFRES_SHORT); 10474 return (PF_DROP); 10475 } 10476 10477 h = mtod(pd->m, struct ip *); 10478 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 10479 *action = PF_DROP; 10480 REASON_SET(reason, PFRES_SHORT); 10481 return (PF_DROP); 10482 } 10483 10484 if (pf_normalize_ip(reason, pd) != PF_PASS) { 10485 /* We do IP header normalization and packet reassembly here */ 10486 *m0 = pd->m; 10487 *action = PF_DROP; 10488 return (PF_DROP); 10489 } 10490 *m0 = pd->m; 10491 h = mtod(pd->m, struct ip *); 10492 10493 if (pf_walk_header(pd, h, reason) != PF_PASS) { 10494 *action = PF_DROP; 10495 return (PF_DROP); 10496 } 10497 10498 pd->src = (struct pf_addr *)&h->ip_src; 10499 pd->dst = (struct pf_addr *)&h->ip_dst; 10500 pf_addrcpy(&pd->osrc, pd->src, af); 10501 pf_addrcpy(&pd->odst, pd->dst, af); 10502 pd->ip_sum = &h->ip_sum; 10503 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 10504 pd->ttl = h->ip_ttl; 10505 pd->tot_len = ntohs(h->ip_len); 10506 pd->act.rtableid = -1; 10507 pd->df = h->ip_off & htons(IP_DF); 10508 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 10509 PF_VPROTO_FRAGMENT : pd->proto; 10510 10511 break; 10512 } 10513 #endif /* INET */ 10514 #ifdef INET6 10515 case AF_INET6: { 10516 struct ip6_hdr *h; 10517 10518 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 10519 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 10520 DPFPRINTF(PF_DEBUG_URGENT, 10521 "%s: m_len < sizeof(struct ip6_hdr)" 10522 ", pullup failed", __func__); 10523 *action = PF_DROP; 10524 REASON_SET(reason, PFRES_SHORT); 10525 return (PF_DROP); 10526 } 10527 10528 h = mtod(pd->m, struct ip6_hdr *); 10529 if (pd->m->m_pkthdr.len < 10530 sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) { 10531 *action = PF_DROP; 10532 REASON_SET(reason, PFRES_SHORT); 10533 return (PF_DROP); 10534 } 10535 10536 /* 10537 * we do not support jumbogram. if we keep going, zero ip6_plen 10538 * will do something bad, so drop the packet for now. 10539 */ 10540 if (htons(h->ip6_plen) == 0) { 10541 *action = PF_DROP; 10542 return (PF_DROP); 10543 } 10544 10545 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10546 *action = PF_DROP; 10547 return (PF_DROP); 10548 } 10549 10550 h = mtod(pd->m, struct ip6_hdr *); 10551 pd->src = (struct pf_addr *)&h->ip6_src; 10552 pd->dst = (struct pf_addr *)&h->ip6_dst; 10553 pf_addrcpy(&pd->osrc, pd->src, af); 10554 pf_addrcpy(&pd->odst, pd->dst, af); 10555 pd->ip_sum = NULL; 10556 pd->tos = IPV6_DSCP(h); 10557 pd->ttl = h->ip6_hlim; 10558 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 10559 pd->act.rtableid = -1; 10560 10561 pd->virtual_proto = (pd->fragoff != 0) ? 10562 PF_VPROTO_FRAGMENT : pd->proto; 10563 10564 /* We do IP header normalization and packet reassembly here */ 10565 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 10566 PF_PASS) { 10567 *m0 = pd->m; 10568 *action = PF_DROP; 10569 return (PF_DROP); 10570 } 10571 *m0 = pd->m; 10572 if (pd->m == NULL) { 10573 /* packet sits in reassembly queue, no error */ 10574 *action = PF_PASS; 10575 return (PF_DROP); 10576 } 10577 10578 /* Update pointers into the packet. */ 10579 h = mtod(pd->m, struct ip6_hdr *); 10580 pd->src = (struct pf_addr *)&h->ip6_src; 10581 pd->dst = (struct pf_addr *)&h->ip6_dst; 10582 10583 pd->off = 0; 10584 10585 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10586 *action = PF_DROP; 10587 return (PF_DROP); 10588 } 10589 10590 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 10591 /* 10592 * Reassembly may have changed the next protocol from 10593 * fragment to something else, so update. 10594 */ 10595 pd->virtual_proto = pd->proto; 10596 MPASS(pd->fragoff == 0); 10597 } 10598 10599 if (pd->fragoff != 0) 10600 pd->virtual_proto = PF_VPROTO_FRAGMENT; 10601 10602 break; 10603 } 10604 #endif /* INET6 */ 10605 default: 10606 panic("pf_setup_pdesc called with illegal af %u", af); 10607 } 10608 10609 switch (pd->virtual_proto) { 10610 case IPPROTO_TCP: { 10611 struct tcphdr *th = &pd->hdr.tcp; 10612 10613 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), 10614 reason, af)) { 10615 *action = PF_DROP; 10616 REASON_SET(reason, PFRES_SHORT); 10617 return (PF_DROP); 10618 } 10619 pd->hdrlen = sizeof(*th); 10620 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 10621 pd->sport = &th->th_sport; 10622 pd->dport = &th->th_dport; 10623 pd->pcksum = &th->th_sum; 10624 break; 10625 } 10626 case IPPROTO_UDP: { 10627 struct udphdr *uh = &pd->hdr.udp; 10628 10629 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), 10630 reason, af)) { 10631 *action = PF_DROP; 10632 REASON_SET(reason, PFRES_SHORT); 10633 return (PF_DROP); 10634 } 10635 pd->hdrlen = sizeof(*uh); 10636 if (uh->uh_dport == 0 || 10637 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 10638 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 10639 *action = PF_DROP; 10640 REASON_SET(reason, PFRES_SHORT); 10641 return (PF_DROP); 10642 } 10643 pd->sport = &uh->uh_sport; 10644 pd->dport = &uh->uh_dport; 10645 pd->pcksum = &uh->uh_sum; 10646 break; 10647 } 10648 case IPPROTO_SCTP: { 10649 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 10650 reason, af)) { 10651 *action = PF_DROP; 10652 REASON_SET(reason, PFRES_SHORT); 10653 return (PF_DROP); 10654 } 10655 pd->hdrlen = sizeof(pd->hdr.sctp); 10656 pd->p_len = pd->tot_len - pd->off; 10657 10658 pd->sport = &pd->hdr.sctp.src_port; 10659 pd->dport = &pd->hdr.sctp.dest_port; 10660 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 10661 *action = PF_DROP; 10662 REASON_SET(reason, PFRES_SHORT); 10663 return (PF_DROP); 10664 } 10665 10666 /* 10667 * Placeholder. The SCTP checksum is 32-bits, but 10668 * pf_test_state() expects to update a 16-bit checksum. 10669 * Provide a dummy value which we'll subsequently ignore. 10670 * Do this before pf_scan_sctp() so any jobs we enqueue 10671 * have a pcksum set. 10672 */ 10673 pd->pcksum = &pd->sctp_dummy_sum; 10674 10675 if (pf_scan_sctp(pd) != PF_PASS) { 10676 *action = PF_DROP; 10677 REASON_SET(reason, PFRES_SHORT); 10678 return (PF_DROP); 10679 } 10680 break; 10681 } 10682 case IPPROTO_ICMP: { 10683 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 10684 reason, af)) { 10685 *action = PF_DROP; 10686 REASON_SET(reason, PFRES_SHORT); 10687 return (PF_DROP); 10688 } 10689 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 10690 pd->hdrlen = ICMP_MINLEN; 10691 break; 10692 } 10693 #ifdef INET6 10694 case IPPROTO_ICMPV6: { 10695 size_t icmp_hlen = sizeof(struct icmp6_hdr); 10696 10697 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10698 reason, af)) { 10699 *action = PF_DROP; 10700 REASON_SET(reason, PFRES_SHORT); 10701 return (PF_DROP); 10702 } 10703 /* ICMP headers we look further into to match state */ 10704 switch (pd->hdr.icmp6.icmp6_type) { 10705 case MLD_LISTENER_QUERY: 10706 case MLD_LISTENER_REPORT: 10707 icmp_hlen = sizeof(struct mld_hdr); 10708 break; 10709 case ND_NEIGHBOR_SOLICIT: 10710 case ND_NEIGHBOR_ADVERT: 10711 icmp_hlen = sizeof(struct nd_neighbor_solicit); 10712 /* FALLTHROUGH */ 10713 case ND_ROUTER_SOLICIT: 10714 case ND_ROUTER_ADVERT: 10715 case ND_REDIRECT: 10716 if (pd->ttl != 255) { 10717 REASON_SET(reason, PFRES_NORM); 10718 return (PF_DROP); 10719 } 10720 break; 10721 } 10722 if (icmp_hlen > sizeof(struct icmp6_hdr) && 10723 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10724 reason, af)) { 10725 *action = PF_DROP; 10726 REASON_SET(reason, PFRES_SHORT); 10727 return (PF_DROP); 10728 } 10729 pd->hdrlen = icmp_hlen; 10730 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 10731 break; 10732 } 10733 #endif /* INET6 */ 10734 default: 10735 /* 10736 * Placeholder value, so future calls to pf_change_ap() don't 10737 * try to update a NULL checksum pointer. 10738 */ 10739 pd->pcksum = &pd->sctp_dummy_sum; 10740 break; 10741 } 10742 10743 if (pd->sport) 10744 pd->osport = pd->nsport = *pd->sport; 10745 if (pd->dport) 10746 pd->odport = pd->ndport = *pd->dport; 10747 10748 MPASS(pd->pcksum != NULL); 10749 10750 return (PF_PASS); 10751 } 10752 10753 static __inline void 10754 pf_rule_counters_inc(struct pf_pdesc *pd, struct pf_krule *r, int dir_out, 10755 int op_pass, sa_family_t af, struct pf_addr *src_host, 10756 struct pf_addr *dst_host) 10757 { 10758 pf_counter_u64_add_protected(&(r->packets[dir_out]), 1); 10759 pf_counter_u64_add_protected(&(r->bytes[dir_out]), pd->tot_len); 10760 pf_update_timestamp(r); 10761 10762 if (r->src.addr.type == PF_ADDR_TABLE) 10763 pfr_update_stats(r->src.addr.p.tbl, src_host, af, 10764 pd->tot_len, dir_out, op_pass, r->src.neg); 10765 if (r->dst.addr.type == PF_ADDR_TABLE) 10766 pfr_update_stats(r->dst.addr.p.tbl, dst_host, af, 10767 pd->tot_len, dir_out, op_pass, r->dst.neg); 10768 } 10769 10770 static void 10771 pf_counters_inc(int action, struct pf_pdesc *pd, struct pf_kstate *s, 10772 struct pf_krule *r, struct pf_krule *a, struct pf_krule_slist *match_rules) 10773 { 10774 struct pf_krule_slist *mr = match_rules; 10775 struct pf_krule_item *ri; 10776 struct pf_krule *nr = NULL; 10777 struct pf_addr *src_host = pd->src; 10778 struct pf_addr *dst_host = pd->dst; 10779 struct pf_state_key *key; 10780 int dir_out = (pd->dir == PF_OUT); 10781 int op_r_pass = (r->action == PF_PASS); 10782 int op_pass = (action == PF_PASS || action == PF_AFRT); 10783 int s_dir_in, s_dir_out, s_dir_rev; 10784 sa_family_t af = pd->af; 10785 10786 pf_counter_u64_critical_enter(); 10787 10788 /* 10789 * Set AF for interface counters, it will be later overwritten for 10790 * rule and state counters with value from proper state key. 10791 */ 10792 if (action == PF_AFRT) { 10793 MPASS(s != NULL); 10794 if (s->direction == PF_OUT && dir_out) 10795 af = pd->naf; 10796 } 10797 10798 pf_counter_u64_add_protected( 10799 &pd->kif->pfik_bytes[af == AF_INET6][dir_out][!op_pass], 10800 pd->tot_len); 10801 pf_counter_u64_add_protected( 10802 &pd->kif->pfik_packets[af == AF_INET6][dir_out][!op_pass], 10803 1); 10804 10805 /* If the rule has failed to apply, don't increase its counters */ 10806 if (!(op_pass || r->action == PF_DROP)) { 10807 pf_counter_u64_critical_exit(); 10808 return; 10809 } 10810 10811 if (s != NULL) { 10812 PF_STATE_LOCK_ASSERT(s); 10813 mr = &(s->match_rules); 10814 10815 /* 10816 * For af-to on the inbound direction we can determine 10817 * the direction of passing packet only by checking direction 10818 * of AF translation. The af-to in "in" direction covers both 10819 * the inbound and the outbound side of state tracking, 10820 * so pd->dir is always PF_IN. We set dir_out and s_dir_rev 10821 * in a way to count packets as if the state was outbound, 10822 * because pfctl -ss shows the state with "->", as if it was 10823 * oubound. 10824 */ 10825 if (action == PF_AFRT && s->direction == PF_IN) { 10826 dir_out = (pd->naf == s->rule->naf); 10827 s_dir_in = 1; 10828 s_dir_out = 0; 10829 s_dir_rev = (pd->naf == s->rule->af); 10830 } else { 10831 dir_out = (pd->dir == PF_OUT); 10832 s_dir_in = (s->direction == PF_IN); 10833 s_dir_out = (s->direction == PF_OUT); 10834 s_dir_rev = (pd->dir != s->direction); 10835 } 10836 10837 /* pd->tot_len is a problematic with af-to rules. Sure, we can 10838 * agree that it's the post-af-to packet length that was 10839 * forwarded through a state, but what about tables which match 10840 * on pre-af-to addresses? We don't have access the the original 10841 * packet length anymore. 10842 */ 10843 s->packets[s_dir_rev]++; 10844 s->bytes[s_dir_rev] += pd->tot_len; 10845 10846 /* 10847 * Source nodes are accessed unlocked here. But since we are 10848 * operating with stateful tracking and the state is locked, 10849 * those SNs could not have been freed. 10850 */ 10851 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 10852 if (s->sns[sn_type] != NULL) { 10853 counter_u64_add( 10854 s->sns[sn_type]->packets[dir_out], 10855 1); 10856 counter_u64_add( 10857 s->sns[sn_type]->bytes[dir_out], 10858 pd->tot_len); 10859 } 10860 } 10861 10862 /* Start with pre-NAT addresses */ 10863 key = s->key[(s->direction == PF_OUT)]; 10864 src_host = &(key->addr[s_dir_out]); 10865 dst_host = &(key->addr[s_dir_in]); 10866 af = key->af; 10867 if (s->nat_rule) { 10868 /* Old-style NAT rules */ 10869 if (s->nat_rule->action == PF_NAT || 10870 s->nat_rule->action == PF_RDR || 10871 s->nat_rule->action == PF_BINAT) { 10872 nr = s->nat_rule; 10873 pf_rule_counters_inc(pd, s->nat_rule, dir_out, 10874 op_r_pass, af, src_host, dst_host); 10875 /* Use post-NAT addresses from now on */ 10876 key = s->key[s_dir_in]; 10877 src_host = &(key->addr[s_dir_out]); 10878 dst_host = &(key->addr[s_dir_in]); 10879 af = key->af; 10880 } 10881 } 10882 } 10883 10884 SLIST_FOREACH(ri, mr, entry) { 10885 pf_rule_counters_inc(pd, ri->r, dir_out, op_r_pass, af, 10886 src_host, dst_host); 10887 if (s && s->nat_rule == ri->r) { 10888 /* Use post-NAT addresses after a match NAT rule */ 10889 key = s->key[s_dir_in]; 10890 src_host = &(key->addr[s_dir_out]); 10891 dst_host = &(key->addr[s_dir_in]); 10892 af = key->af; 10893 } 10894 } 10895 10896 if (s == NULL) { 10897 pf_free_match_rules(mr); 10898 } 10899 10900 if (a != NULL) { 10901 pf_rule_counters_inc(pd, a, dir_out, op_r_pass, af, 10902 src_host, dst_host); 10903 } 10904 10905 if (r != nr) { 10906 pf_rule_counters_inc(pd, r, dir_out, op_r_pass, af, 10907 src_host, dst_host); 10908 } 10909 10910 pf_counter_u64_critical_exit(); 10911 } 10912 10913 static void 10914 pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm, 10915 struct pf_krule *am, struct pf_kruleset *ruleset, 10916 struct pf_krule_slist *match_rules) 10917 { 10918 struct pf_krule_item *ri; 10919 10920 /* if this is the log(matches) rule, packet has been logged already */ 10921 if (rm->log & PF_LOG_MATCHES) 10922 return; 10923 10924 SLIST_FOREACH(ri, match_rules, entry) 10925 if (ri->r->log & PF_LOG_MATCHES) 10926 PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am, 10927 ruleset, pd, 1, ri->r); 10928 } 10929 10930 #if defined(INET) || defined(INET6) 10931 int 10932 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10933 struct inpcb *inp, struct pf_rule_actions *default_actions) 10934 { 10935 struct pfi_kkif *kif; 10936 u_short action, reason = 0; 10937 struct m_tag *mtag; 10938 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10939 struct pf_kstate *s = NULL; 10940 struct pf_kruleset *ruleset = NULL; 10941 struct pf_krule_item *ri; 10942 struct pf_krule_slist match_rules; 10943 struct pf_pdesc pd; 10944 int use_2nd_queue = 0; 10945 uint16_t tag; 10946 10947 PF_RULES_RLOCK_TRACKER; 10948 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10949 M_ASSERTPKTHDR(*m0); 10950 NET_EPOCH_ASSERT(); 10951 10952 if (!V_pf_status.running) 10953 return (PF_PASS); 10954 10955 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10956 10957 if (__predict_false(kif == NULL)) { 10958 DPFPRINTF(PF_DEBUG_URGENT, 10959 "%s: kif == NULL, if_xname %s", 10960 __func__, ifp->if_xname); 10961 return (PF_DROP); 10962 } 10963 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10964 return (PF_PASS); 10965 } 10966 10967 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10968 return (PF_PASS); 10969 } 10970 10971 if (__predict_false(! M_WRITABLE(*m0))) { 10972 *m0 = m_unshare(*m0, M_NOWAIT); 10973 if (*m0 == NULL) { 10974 return (PF_DROP); 10975 } 10976 } 10977 10978 pf_init_pdesc(&pd, *m0); 10979 SLIST_INIT(&match_rules); 10980 10981 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10982 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10983 10984 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10985 pd.pf_mtag->if_idxgen); 10986 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10987 m_freem(*m0); 10988 *m0 = NULL; 10989 return (PF_PASS); 10990 } 10991 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10992 *m0 = NULL; 10993 return (PF_PASS); 10994 } 10995 10996 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10997 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10998 /* Dummynet re-injects packets after they've 10999 * completed their delay. We've already 11000 * processed them, so pass unconditionally. */ 11001 11002 /* But only once. We may see the packet multiple times (e.g. 11003 * PFIL_IN/PFIL_OUT). */ 11004 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 11005 11006 return (PF_PASS); 11007 } 11008 11009 PF_RULES_RLOCK(); 11010 11011 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 11012 kif, default_actions) != PF_PASS) { 11013 if (action != PF_PASS) 11014 pd.act.log |= PF_LOG_FORCE; 11015 goto done; 11016 } 11017 11018 #ifdef INET 11019 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 11020 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 11021 PF_RULES_RUNLOCK(); 11022 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 11023 0, ifp->if_mtu); 11024 *m0 = NULL; 11025 return (PF_DROP); 11026 } 11027 #endif /* INET */ 11028 #ifdef INET6 11029 /* 11030 * If we end up changing IP addresses (e.g. binat) the stack may get 11031 * confused and fail to send the icmp6 packet too big error. Just send 11032 * it here, before we do any NAT. 11033 */ 11034 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 11035 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 11036 PF_RULES_RUNLOCK(); 11037 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 11038 *m0 = NULL; 11039 return (PF_DROP); 11040 } 11041 #endif /* INET6 */ 11042 11043 if (__predict_false(ip_divert_ptr != NULL) && 11044 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 11045 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 11046 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 11047 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 11048 if (pd.pf_mtag == NULL && 11049 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11050 action = PF_DROP; 11051 goto done; 11052 } 11053 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 11054 } 11055 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 11056 pd.m->m_flags |= M_FASTFWD_OURS; 11057 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 11058 } 11059 m_tag_delete(pd.m, mtag); 11060 11061 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 11062 if (mtag != NULL) 11063 m_tag_delete(pd.m, mtag); 11064 } 11065 11066 switch (pd.virtual_proto) { 11067 case PF_VPROTO_FRAGMENT: 11068 /* 11069 * handle fragments that aren't reassembled by 11070 * normalization 11071 */ 11072 if (kif == NULL || r == NULL) /* pflog */ 11073 action = PF_DROP; 11074 else 11075 action = pf_test_rule(&r, &s, &pd, &a, 11076 &ruleset, &reason, inp, &match_rules); 11077 if (action != PF_PASS) 11078 REASON_SET(&reason, PFRES_FRAG); 11079 break; 11080 11081 case IPPROTO_TCP: { 11082 /* Respond to SYN with a syncookie. */ 11083 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 11084 pd.dir == PF_IN && pf_synflood_check(&pd)) { 11085 pf_syncookie_send(&pd, &reason); 11086 action = PF_DROP; 11087 break; 11088 } 11089 11090 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 11091 use_2nd_queue = 1; 11092 action = pf_normalize_tcp(&pd); 11093 if (action == PF_DROP) 11094 break; 11095 action = pf_test_state(&s, &pd, &reason); 11096 if (action == PF_PASS || action == PF_AFRT) { 11097 if (s != NULL) { 11098 if (V_pfsync_update_state_ptr != NULL) 11099 V_pfsync_update_state_ptr(s); 11100 r = s->rule; 11101 a = s->anchor; 11102 } 11103 } else if (s == NULL) { 11104 /* Validate remote SYN|ACK, re-create original SYN if 11105 * valid. */ 11106 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 11107 TH_ACK && pf_syncookie_validate(&pd) && 11108 pd.dir == PF_IN) { 11109 struct mbuf *msyn; 11110 11111 msyn = pf_syncookie_recreate_syn(&pd, &reason); 11112 if (msyn == NULL) { 11113 action = PF_DROP; 11114 break; 11115 } 11116 11117 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 11118 &pd.act); 11119 m_freem(msyn); 11120 if (action != PF_PASS) 11121 break; 11122 11123 action = pf_test_state(&s, &pd, &reason); 11124 if (action != PF_PASS || s == NULL) { 11125 action = PF_DROP; 11126 break; 11127 } 11128 11129 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 11130 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 11131 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 11132 action = pf_synproxy(&pd, s, &reason); 11133 break; 11134 } else { 11135 action = pf_test_rule(&r, &s, &pd, 11136 &a, &ruleset, &reason, inp, &match_rules); 11137 } 11138 } 11139 break; 11140 } 11141 11142 case IPPROTO_SCTP: 11143 action = pf_normalize_sctp(&pd); 11144 if (action == PF_DROP) 11145 break; 11146 /* fallthrough */ 11147 case IPPROTO_UDP: 11148 default: 11149 action = pf_test_state(&s, &pd, &reason); 11150 if (action == PF_PASS || action == PF_AFRT) { 11151 if (s != NULL) { 11152 if (V_pfsync_update_state_ptr != NULL) 11153 V_pfsync_update_state_ptr(s); 11154 r = s->rule; 11155 a = s->anchor; 11156 } 11157 } else if (s == NULL) { 11158 action = pf_test_rule(&r, &s, 11159 &pd, &a, &ruleset, &reason, inp, &match_rules); 11160 } 11161 break; 11162 11163 case IPPROTO_ICMP: 11164 case IPPROTO_ICMPV6: { 11165 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 11166 action = PF_DROP; 11167 REASON_SET(&reason, PFRES_NORM); 11168 DPFPRINTF(PF_DEBUG_MISC, 11169 "dropping IPv6 packet with ICMPv4 payload"); 11170 break; 11171 } 11172 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 11173 action = PF_DROP; 11174 REASON_SET(&reason, PFRES_NORM); 11175 DPFPRINTF(PF_DEBUG_MISC, 11176 "pf: dropping IPv4 packet with ICMPv6 payload"); 11177 break; 11178 } 11179 action = pf_test_state_icmp(&s, &pd, &reason); 11180 if (action == PF_PASS || action == PF_AFRT) { 11181 if (V_pfsync_update_state_ptr != NULL) 11182 V_pfsync_update_state_ptr(s); 11183 r = s->rule; 11184 a = s->anchor; 11185 } else if (s == NULL) 11186 action = pf_test_rule(&r, &s, &pd, 11187 &a, &ruleset, &reason, inp, &match_rules); 11188 break; 11189 } 11190 11191 } 11192 11193 done: 11194 PF_RULES_RUNLOCK(); 11195 11196 /* if packet sits in reassembly queue, return without error */ 11197 if (pd.m == NULL) { 11198 pf_free_match_rules(&match_rules); 11199 goto eat_pkt; 11200 } 11201 11202 if (s) 11203 memcpy(&pd.act, &s->act, sizeof(s->act)); 11204 11205 if (action == PF_PASS && pd.badopts != 0 && !pd.act.allow_opts) { 11206 action = PF_DROP; 11207 REASON_SET(&reason, PFRES_IPOPTIONS); 11208 pd.act.log = PF_LOG_FORCE; 11209 DPFPRINTF(PF_DEBUG_MISC, 11210 "pf: dropping packet with dangerous headers"); 11211 } 11212 11213 if (pd.act.max_pkt_size && pd.act.max_pkt_size && 11214 pd.tot_len > pd.act.max_pkt_size) { 11215 action = PF_DROP; 11216 REASON_SET(&reason, PFRES_NORM); 11217 pd.act.log = PF_LOG_FORCE; 11218 DPFPRINTF(PF_DEBUG_MISC, 11219 "pf: dropping overly long packet"); 11220 } 11221 11222 if (s) { 11223 uint8_t log = pd.act.log; 11224 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 11225 pd.act.log |= log; 11226 tag = s->tag; 11227 } else { 11228 tag = r->tag; 11229 } 11230 11231 if (tag > 0 && pf_tag_packet(&pd, tag)) { 11232 action = PF_DROP; 11233 REASON_SET(&reason, PFRES_MEMORY); 11234 } 11235 11236 pf_scrub(&pd); 11237 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 11238 pf_normalize_mss(&pd); 11239 11240 if (pd.act.rtableid >= 0) 11241 M_SETFIB(pd.m, pd.act.rtableid); 11242 11243 if (pd.act.flags & PFSTATE_SETPRIO) { 11244 if (pd.tos & IPTOS_LOWDELAY) 11245 use_2nd_queue = 1; 11246 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 11247 action = PF_DROP; 11248 REASON_SET(&reason, PFRES_MEMORY); 11249 pd.act.log = PF_LOG_FORCE; 11250 DPFPRINTF(PF_DEBUG_MISC, 11251 "pf: failed to allocate 802.1q mtag"); 11252 } 11253 } 11254 11255 #ifdef ALTQ 11256 if (action == PF_PASS && pd.act.qid) { 11257 if (pd.pf_mtag == NULL && 11258 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11259 action = PF_DROP; 11260 REASON_SET(&reason, PFRES_MEMORY); 11261 } else { 11262 if (s != NULL) 11263 pd.pf_mtag->qid_hash = pf_state_hash(s); 11264 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 11265 pd.pf_mtag->qid = pd.act.pqid; 11266 else 11267 pd.pf_mtag->qid = pd.act.qid; 11268 /* Add hints for ecn. */ 11269 pd.pf_mtag->hdr = mtod(pd.m, void *); 11270 } 11271 } 11272 #endif /* ALTQ */ 11273 11274 /* 11275 * connections redirected to loopback should not match sockets 11276 * bound specifically to loopback due to security implications, 11277 * see tcp_input() and in_pcblookup_listen(). 11278 */ 11279 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 11280 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 11281 (s->nat_rule->action == PF_RDR || 11282 s->nat_rule->action == PF_BINAT) && 11283 pf_is_loopback(af, pd.dst)) 11284 pd.m->m_flags |= M_SKIP_FIREWALL; 11285 11286 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 11287 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 11288 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 11289 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 11290 if (mtag != NULL) { 11291 ((struct pf_divert_mtag *)(mtag+1))->port = 11292 ntohs(r->divert.port); 11293 ((struct pf_divert_mtag *)(mtag+1))->idir = 11294 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 11295 PF_DIVERT_MTAG_DIR_OUT; 11296 11297 pf_counters_inc(action, &pd, s, r, a, &match_rules); 11298 11299 if (s) 11300 PF_STATE_UNLOCK(s); 11301 11302 m_tag_prepend(pd.m, mtag); 11303 if (pd.m->m_flags & M_FASTFWD_OURS) { 11304 if (pd.pf_mtag == NULL && 11305 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11306 action = PF_DROP; 11307 REASON_SET(&reason, PFRES_MEMORY); 11308 pd.act.log = PF_LOG_FORCE; 11309 DPFPRINTF(PF_DEBUG_MISC, 11310 "pf: failed to allocate tag"); 11311 } else { 11312 pd.pf_mtag->flags |= 11313 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 11314 pd.m->m_flags &= ~M_FASTFWD_OURS; 11315 } 11316 } 11317 ip_divert_ptr(*m0, dir == PF_IN); 11318 *m0 = NULL; 11319 11320 return (action); 11321 } else { 11322 /* XXX: ipfw has the same behaviour! */ 11323 action = PF_DROP; 11324 REASON_SET(&reason, PFRES_MEMORY); 11325 pd.act.log = PF_LOG_FORCE; 11326 DPFPRINTF(PF_DEBUG_MISC, 11327 "pf: failed to allocate divert tag"); 11328 } 11329 } 11330 /* XXX: Anybody working on it?! */ 11331 if (af == AF_INET6 && r->divert.port) 11332 printf("pf: divert(9) is not supported for IPv6\n"); 11333 11334 /* this flag will need revising if the pkt is forwarded */ 11335 if (pd.pf_mtag) 11336 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 11337 11338 if (pd.act.log) { 11339 struct pf_krule *lr; 11340 11341 if (s != NULL && s->nat_rule != NULL && 11342 s->nat_rule->log & PF_LOG_ALL) 11343 lr = s->nat_rule; 11344 else 11345 lr = r; 11346 11347 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 11348 PFLOG_PACKET(action, reason, lr, a, 11349 ruleset, &pd, (s == NULL), NULL); 11350 if (s) { 11351 SLIST_FOREACH(ri, &s->match_rules, entry) 11352 if (ri->r->log & PF_LOG_ALL) 11353 PFLOG_PACKET(action, 11354 reason, ri->r, a, ruleset, &pd, 0, NULL); 11355 } 11356 } 11357 11358 pf_counters_inc(action, &pd, s, r, a, &match_rules); 11359 11360 switch (action) { 11361 case PF_SYNPROXY_DROP: 11362 m_freem(*m0); 11363 case PF_DEFER: 11364 *m0 = NULL; 11365 action = PF_PASS; 11366 break; 11367 case PF_DROP: 11368 m_freem(*m0); 11369 *m0 = NULL; 11370 break; 11371 case PF_AFRT: 11372 if (pf_translate_af(&pd, r)) { 11373 *m0 = pd.m; 11374 action = PF_DROP; 11375 break; 11376 } 11377 #ifdef INET 11378 if (pd.naf == AF_INET) { 11379 action = pf_route(r, kif->pfik_ifp, s, &pd, 11380 inp); 11381 } 11382 #endif /* INET */ 11383 #ifdef INET6 11384 if (pd.naf == AF_INET6) { 11385 action = pf_route6(r, kif->pfik_ifp, s, &pd, 11386 inp); 11387 } 11388 #endif /* INET6 */ 11389 *m0 = pd.m; 11390 goto out; 11391 break; 11392 default: 11393 if (pd.act.rt) { 11394 switch (af) { 11395 #ifdef INET 11396 case AF_INET: 11397 /* pf_route() returns unlocked. */ 11398 action = pf_route(r, kif->pfik_ifp, s, &pd, 11399 inp); 11400 break; 11401 #endif /* INET */ 11402 #ifdef INET6 11403 case AF_INET6: 11404 /* pf_route6() returns unlocked. */ 11405 action = pf_route6(r, kif->pfik_ifp, s, &pd, 11406 inp); 11407 break; 11408 #endif /* INET6 */ 11409 } 11410 *m0 = pd.m; 11411 goto out; 11412 } 11413 if (pf_dummynet(&pd, s, r, m0) != 0) { 11414 action = PF_DROP; 11415 REASON_SET(&reason, PFRES_MEMORY); 11416 } 11417 break; 11418 } 11419 11420 eat_pkt: 11421 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 11422 11423 if (s && action != PF_DROP) { 11424 if (!s->if_index_in && dir == PF_IN) 11425 s->if_index_in = ifp->if_index; 11426 else if (!s->if_index_out && dir == PF_OUT) 11427 s->if_index_out = ifp->if_index; 11428 } 11429 11430 if (s) 11431 PF_STATE_UNLOCK(s); 11432 11433 out: 11434 #ifdef INET6 11435 /* If reassembled packet passed, create new fragments. */ 11436 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 11437 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 11438 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 11439 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 11440 #endif /* INET6 */ 11441 11442 pf_sctp_multihome_delayed(&pd, kif, s, action); 11443 11444 return (action); 11445 } 11446 #endif /* INET || INET6 */ 11447