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 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 123 124 SDT_PROVIDER_DEFINE(pf); 125 SDT_PROBE_DEFINE2(pf, , test, reason_set, "int", "int"); 126 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 127 "struct pf_kstate *"); 128 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 129 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 130 "struct pf_kstate *"); 131 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 132 "struct pfi_kkif *"); 133 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 134 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 135 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 136 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 137 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 138 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 139 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 140 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 141 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 142 "struct pf_krule *", "struct mbuf *", "int"); 143 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 144 "struct pf_sctp_source *"); 145 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 146 "struct pf_kstate *", "struct pf_sctp_source *"); 147 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 148 "int", "struct pf_pdesc *", "int"); 149 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 150 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 151 "int"); 152 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 153 "int"); 154 155 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 156 "struct mbuf *"); 157 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 158 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 159 "int", "struct pf_keth_rule *", "char *"); 160 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 161 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 162 "int", "struct pf_keth_rule *"); 163 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 164 165 /* 166 * Global variables 167 */ 168 169 /* state tables */ 170 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 171 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 173 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 174 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 175 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 176 VNET_DEFINE(struct pf_kstatus, pf_status); 177 178 VNET_DEFINE(u_int32_t, ticket_altqs_active); 179 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 180 VNET_DEFINE(int, altqs_inactive_open); 181 VNET_DEFINE(u_int32_t, ticket_pabuf); 182 183 static const int PF_HDR_LIMIT = 20; /* arbitrary limit */ 184 185 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 186 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 187 VNET_DEFINE(u_char, pf_tcp_secret[16]); 188 #define V_pf_tcp_secret VNET(pf_tcp_secret) 189 VNET_DEFINE(int, pf_tcp_secret_init); 190 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 191 VNET_DEFINE(int, pf_tcp_iss_off); 192 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 193 VNET_DECLARE(int, pf_vnet_active); 194 #define V_pf_vnet_active VNET(pf_vnet_active) 195 196 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 197 #define V_pf_purge_idx VNET(pf_purge_idx) 198 199 #ifdef PF_WANT_32_TO_64_COUNTER 200 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 201 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 202 203 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 204 VNET_DEFINE(size_t, pf_allrulecount); 205 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 206 #endif 207 208 #define PF_SCTP_MAX_ENDPOINTS 8 209 210 struct pf_sctp_endpoint; 211 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 212 struct pf_sctp_source { 213 sa_family_t af; 214 struct pf_addr addr; 215 TAILQ_ENTRY(pf_sctp_source) entry; 216 }; 217 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 218 struct pf_sctp_endpoint 219 { 220 uint32_t v_tag; 221 struct pf_sctp_sources sources; 222 RB_ENTRY(pf_sctp_endpoint) entry; 223 }; 224 static int 225 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 226 { 227 return (a->v_tag - b->v_tag); 228 } 229 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 230 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 231 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 232 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 233 static struct mtx_padalign pf_sctp_endpoints_mtx; 234 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 235 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 236 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 237 238 /* 239 * Queue for pf_intr() sends. 240 */ 241 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 242 struct pf_send_entry { 243 STAILQ_ENTRY(pf_send_entry) pfse_next; 244 struct mbuf *pfse_m; 245 enum { 246 PFSE_IP, 247 PFSE_IP6, 248 PFSE_ICMP, 249 PFSE_ICMP6, 250 } pfse_type; 251 struct { 252 int type; 253 int code; 254 int mtu; 255 } icmpopts; 256 }; 257 258 STAILQ_HEAD(pf_send_head, pf_send_entry); 259 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 260 #define V_pf_sendqueue VNET(pf_sendqueue) 261 262 static struct mtx_padalign pf_sendqueue_mtx; 263 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 264 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 265 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 266 267 /* 268 * Queue for pf_overload_task() tasks. 269 */ 270 struct pf_overload_entry { 271 SLIST_ENTRY(pf_overload_entry) next; 272 struct pf_addr addr; 273 sa_family_t af; 274 uint8_t dir; 275 struct pf_krule *rule; 276 }; 277 278 SLIST_HEAD(pf_overload_head, pf_overload_entry); 279 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 280 #define V_pf_overloadqueue VNET(pf_overloadqueue) 281 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 282 #define V_pf_overloadtask VNET(pf_overloadtask) 283 284 static struct mtx_padalign pf_overloadqueue_mtx; 285 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 286 "pf overload/flush queue", MTX_DEF); 287 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 288 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 289 290 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 291 struct mtx_padalign pf_unlnkdrules_mtx; 292 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 293 MTX_DEF); 294 295 struct sx pf_config_lock; 296 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 297 298 struct mtx_padalign pf_table_stats_lock; 299 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 300 MTX_DEF); 301 302 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 303 #define V_pf_sources_z VNET(pf_sources_z) 304 uma_zone_t pf_mtag_z; 305 VNET_DEFINE(uma_zone_t, pf_state_z); 306 VNET_DEFINE(uma_zone_t, pf_state_key_z); 307 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 308 309 VNET_DEFINE(struct unrhdr64, pf_stateid); 310 311 static void pf_src_tree_remove_state(struct pf_kstate *); 312 static int pf_check_threshold(struct pf_kthreshold *); 313 314 static void pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *, 315 struct pf_addr *, u_int16_t); 316 static int pf_modulate_sack(struct pf_pdesc *, 317 struct tcphdr *, struct pf_state_peer *); 318 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 319 u_int16_t *, u_int16_t *); 320 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 321 struct pf_addr *, struct pf_addr *, u_int16_t, 322 u_int16_t *, u_int16_t *, u_int16_t *, 323 u_int16_t *, u_int8_t, sa_family_t); 324 int pf_change_icmp_af(struct mbuf *, int, 325 struct pf_pdesc *, struct pf_pdesc *, 326 struct pf_addr *, struct pf_addr *, sa_family_t, 327 sa_family_t); 328 int pf_translate_icmp_af(int, void *); 329 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 330 int, sa_family_t, struct pf_krule *, int); 331 static void pf_detach_state(struct pf_kstate *); 332 static int pf_state_key_attach(struct pf_state_key *, 333 struct pf_state_key *, struct pf_kstate *); 334 static void pf_state_key_detach(struct pf_kstate *, int); 335 static int pf_state_key_ctor(void *, int, void *, int); 336 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 337 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 338 struct pf_mtag *pf_mtag); 339 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 340 struct pf_krule *, struct mbuf **); 341 static int pf_dummynet_route(struct pf_pdesc *, 342 struct pf_kstate *, struct pf_krule *, 343 struct ifnet *, const struct sockaddr *, struct mbuf **); 344 static int pf_test_eth_rule(int, struct pfi_kkif *, 345 struct mbuf **); 346 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 347 struct pf_pdesc *, struct pf_krule **, 348 struct pf_kruleset **, u_short *, struct inpcb *); 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 static int pf_state_key_addr_setup(struct pf_pdesc *, 353 struct pf_state_key_cmp *, int); 354 static int pf_tcp_track_full(struct pf_kstate *, 355 struct pf_pdesc *, u_short *, int *, 356 struct pf_state_peer *, struct pf_state_peer *, 357 u_int8_t, u_int8_t); 358 static int pf_tcp_track_sloppy(struct pf_kstate *, 359 struct pf_pdesc *, u_short *, 360 struct pf_state_peer *, struct pf_state_peer *, 361 u_int8_t, u_int8_t); 362 static int pf_test_state(struct pf_kstate **, struct pf_pdesc *, 363 u_short *); 364 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 365 struct pf_pdesc *, struct pf_kstate **, 366 u_int16_t, u_int16_t, int, int *, int, int); 367 static int pf_test_state_icmp(struct pf_kstate **, 368 struct pf_pdesc *, u_short *); 369 static int pf_sctp_track(struct pf_kstate *, struct pf_pdesc *, 370 u_short *); 371 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 372 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 373 struct pfi_kkif *, struct pf_kstate *, int); 374 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 375 int, u_int16_t); 376 static int pf_check_proto_cksum(struct mbuf *, int, int, 377 u_int8_t, sa_family_t); 378 static int pf_walk_header(struct pf_pdesc *, struct ip *, u_short *); 379 #ifdef INET6 380 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 381 int, int, u_short *); 382 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 383 u_short *); 384 #endif 385 static void pf_print_state_parts(struct pf_kstate *, 386 struct pf_state_key *, struct pf_state_key *); 387 static int pf_patch_8(struct pf_pdesc *, u_int8_t *, u_int8_t, 388 bool); 389 static int pf_find_state(struct pf_pdesc *, 390 const struct pf_state_key_cmp *, struct pf_kstate **); 391 static bool pf_src_connlimit(struct pf_kstate *); 392 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 393 static void pf_counters_inc(int, struct pf_pdesc *, 394 struct pf_kstate *, struct pf_krule *, 395 struct pf_krule *); 396 static void pf_log_matches(struct pf_pdesc *, struct pf_krule *, 397 struct pf_krule *, struct pf_kruleset *, 398 struct pf_krule_slist *); 399 static void pf_overload_task(void *v, int pending); 400 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 401 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 402 struct pf_addr *, sa_family_t, struct pf_addr *, 403 struct pfi_kkif *, pf_sn_types_t); 404 static u_int pf_purge_expired_states(u_int, int); 405 static void pf_purge_unlinked_rules(void); 406 static int pf_mtag_uminit(void *, int, int); 407 static void pf_mtag_free(struct m_tag *); 408 static void pf_packet_rework_nat(struct pf_pdesc *, int, 409 struct pf_state_key *); 410 #ifdef INET 411 static void pf_route(struct pf_krule *, 412 struct ifnet *, struct pf_kstate *, 413 struct pf_pdesc *, struct inpcb *); 414 #endif /* INET */ 415 #ifdef INET6 416 static void pf_change_a6(struct pf_addr *, u_int16_t *, 417 struct pf_addr *, u_int8_t); 418 static void pf_route6(struct pf_krule *, 419 struct ifnet *, struct pf_kstate *, 420 struct pf_pdesc *, struct inpcb *); 421 #endif /* INET6 */ 422 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 423 424 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 425 426 extern int pf_end_threads; 427 extern struct proc *pf_purge_proc; 428 429 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 430 431 #define PACKET_UNDO_NAT(_pd, _off, _s) \ 432 do { \ 433 struct pf_state_key *nk; \ 434 if ((pd->dir) == PF_OUT) \ 435 nk = (_s)->key[PF_SK_STACK]; \ 436 else \ 437 nk = (_s)->key[PF_SK_WIRE]; \ 438 pf_packet_rework_nat(_pd, _off, nk); \ 439 } while (0) 440 441 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 442 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 443 444 static struct pfi_kkif * 445 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 446 { 447 struct pfi_kkif *k = pd->kif; 448 449 SDT_PROBE2(pf, ip, , bound_iface, st, k); 450 451 /* Floating unless otherwise specified. */ 452 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 453 return (V_pfi_all); 454 455 /* 456 * Initially set to all, because we don't know what interface we'll be 457 * sending this out when we create the state. 458 */ 459 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN)) 460 return (V_pfi_all); 461 462 /* 463 * If this state is created based on another state (e.g. SCTP 464 * multihome) always set it floating initially. We can't know for sure 465 * what interface the actual traffic for this state will come in on. 466 */ 467 if (pd->related_rule) 468 return (V_pfi_all); 469 470 /* Don't overrule the interface for states created on incoming packets. */ 471 if (st->direction == PF_IN) 472 return (k); 473 474 /* No route-to, so don't overrule. */ 475 if (st->act.rt != PF_ROUTETO) 476 return (k); 477 478 /* Bind to the route-to interface. */ 479 return (st->act.rt_kif); 480 } 481 482 #define STATE_INC_COUNTERS(s) \ 483 do { \ 484 struct pf_krule_item *mrm; \ 485 counter_u64_add(s->rule->states_cur, 1); \ 486 counter_u64_add(s->rule->states_tot, 1); \ 487 if (s->anchor != NULL) { \ 488 counter_u64_add(s->anchor->states_cur, 1); \ 489 counter_u64_add(s->anchor->states_tot, 1); \ 490 } \ 491 if (s->nat_rule != NULL) { \ 492 counter_u64_add(s->nat_rule->states_cur, 1);\ 493 counter_u64_add(s->nat_rule->states_tot, 1);\ 494 } \ 495 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 496 counter_u64_add(mrm->r->states_cur, 1); \ 497 counter_u64_add(mrm->r->states_tot, 1); \ 498 } \ 499 } while (0) 500 501 #define STATE_DEC_COUNTERS(s) \ 502 do { \ 503 struct pf_krule_item *mrm; \ 504 if (s->nat_rule != NULL) \ 505 counter_u64_add(s->nat_rule->states_cur, -1);\ 506 if (s->anchor != NULL) \ 507 counter_u64_add(s->anchor->states_cur, -1); \ 508 counter_u64_add(s->rule->states_cur, -1); \ 509 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 510 counter_u64_add(mrm->r->states_cur, -1); \ 511 } while (0) 512 513 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 514 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 515 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 516 VNET_DEFINE(struct pf_idhash *, pf_idhash); 517 VNET_DEFINE(struct pf_srchash *, pf_srchash); 518 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 519 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 520 521 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 522 "pf(4)"); 523 524 VNET_DEFINE(u_long, pf_hashmask); 525 VNET_DEFINE(u_long, pf_srchashmask); 526 VNET_DEFINE(u_long, pf_udpendpointhashmask); 527 VNET_DEFINE_STATIC(u_long, pf_hashsize); 528 #define V_pf_hashsize VNET(pf_hashsize) 529 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 530 #define V_pf_srchashsize VNET(pf_srchashsize) 531 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 532 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 533 u_long pf_ioctl_maxcount = 65535; 534 535 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 536 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 537 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 538 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 539 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 540 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 541 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 542 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 543 544 VNET_DEFINE(void *, pf_swi_cookie); 545 VNET_DEFINE(struct intr_event *, pf_swi_ie); 546 547 VNET_DEFINE(uint32_t, pf_hashseed); 548 #define V_pf_hashseed VNET(pf_hashseed) 549 550 static void 551 pf_sctp_checksum(struct mbuf *m, int off) 552 { 553 uint32_t sum = 0; 554 555 /* Zero out the checksum, to enable recalculation. */ 556 m_copyback(m, off + offsetof(struct sctphdr, checksum), 557 sizeof(sum), (caddr_t)&sum); 558 559 sum = sctp_calculate_cksum(m, off); 560 561 m_copyback(m, off + offsetof(struct sctphdr, checksum), 562 sizeof(sum), (caddr_t)&sum); 563 } 564 565 int 566 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 567 { 568 569 switch (af) { 570 #ifdef INET 571 case AF_INET: 572 if (a->addr32[0] > b->addr32[0]) 573 return (1); 574 if (a->addr32[0] < b->addr32[0]) 575 return (-1); 576 break; 577 #endif /* INET */ 578 #ifdef INET6 579 case AF_INET6: 580 if (a->addr32[3] > b->addr32[3]) 581 return (1); 582 if (a->addr32[3] < b->addr32[3]) 583 return (-1); 584 if (a->addr32[2] > b->addr32[2]) 585 return (1); 586 if (a->addr32[2] < b->addr32[2]) 587 return (-1); 588 if (a->addr32[1] > b->addr32[1]) 589 return (1); 590 if (a->addr32[1] < b->addr32[1]) 591 return (-1); 592 if (a->addr32[0] > b->addr32[0]) 593 return (1); 594 if (a->addr32[0] < b->addr32[0]) 595 return (-1); 596 break; 597 #endif /* INET6 */ 598 default: 599 unhandled_af(af); 600 } 601 return (0); 602 } 603 604 static bool 605 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 606 { 607 switch (af) { 608 #ifdef INET 609 case AF_INET: 610 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 611 #endif /* INET */ 612 case AF_INET6: 613 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 614 default: 615 unhandled_af(af); 616 } 617 } 618 619 static void 620 pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk) 621 { 622 623 switch (pd->proto) { 624 case IPPROTO_TCP: { 625 struct tcphdr *th = &pd->hdr.tcp; 626 627 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 628 pf_change_ap(pd, pd->src, &th->th_sport, 629 &nk->addr[pd->sidx], nk->port[pd->sidx]); 630 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 631 pf_change_ap(pd, pd->dst, &th->th_dport, 632 &nk->addr[pd->didx], nk->port[pd->didx]); 633 m_copyback(pd->m, off, sizeof(*th), (caddr_t)th); 634 break; 635 } 636 case IPPROTO_UDP: { 637 struct udphdr *uh = &pd->hdr.udp; 638 639 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 640 pf_change_ap(pd, pd->src, &uh->uh_sport, 641 &nk->addr[pd->sidx], nk->port[pd->sidx]); 642 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 643 pf_change_ap(pd, pd->dst, &uh->uh_dport, 644 &nk->addr[pd->didx], nk->port[pd->didx]); 645 m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh); 646 break; 647 } 648 case IPPROTO_SCTP: { 649 struct sctphdr *sh = &pd->hdr.sctp; 650 651 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 652 pf_change_ap(pd, pd->src, &sh->src_port, 653 &nk->addr[pd->sidx], nk->port[pd->sidx]); 654 } 655 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 656 pf_change_ap(pd, pd->dst, &sh->dest_port, 657 &nk->addr[pd->didx], nk->port[pd->didx]); 658 } 659 660 break; 661 } 662 case IPPROTO_ICMP: { 663 struct icmp *ih = &pd->hdr.icmp; 664 665 if (nk->port[pd->sidx] != ih->icmp_id) { 666 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 667 ih->icmp_cksum, ih->icmp_id, 668 nk->port[pd->sidx], 0); 669 ih->icmp_id = nk->port[pd->sidx]; 670 pd->sport = &ih->icmp_id; 671 672 m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih); 673 } 674 /* FALLTHROUGH */ 675 } 676 default: 677 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 678 switch (pd->af) { 679 case AF_INET: 680 pf_change_a(&pd->src->v4.s_addr, 681 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 682 0); 683 break; 684 case AF_INET6: 685 pf_addrcpy(pd->src, &nk->addr[pd->sidx], 686 pd->af); 687 break; 688 default: 689 unhandled_af(pd->af); 690 } 691 } 692 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 693 switch (pd->af) { 694 case AF_INET: 695 pf_change_a(&pd->dst->v4.s_addr, 696 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 697 0); 698 break; 699 case AF_INET6: 700 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 701 pd->af); 702 break; 703 default: 704 unhandled_af(pd->af); 705 } 706 } 707 break; 708 } 709 } 710 711 static __inline uint32_t 712 pf_hashkey(const struct pf_state_key *sk) 713 { 714 uint32_t h; 715 716 h = murmur3_32_hash32((const uint32_t *)sk, 717 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 718 V_pf_hashseed); 719 720 return (h & V_pf_hashmask); 721 } 722 723 __inline uint32_t 724 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 725 { 726 uint32_t h; 727 728 switch (af) { 729 case AF_INET: 730 h = murmur3_32_hash32((uint32_t *)&addr->v4, 731 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 732 break; 733 case AF_INET6: 734 h = murmur3_32_hash32((uint32_t *)&addr->v6, 735 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 736 break; 737 default: 738 unhandled_af(af); 739 } 740 741 return (h & V_pf_srchashmask); 742 } 743 744 static inline uint32_t 745 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 746 { 747 uint32_t h; 748 749 h = murmur3_32_hash32((uint32_t *)endpoint, 750 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 751 V_pf_hashseed); 752 return (h & V_pf_udpendpointhashmask); 753 } 754 755 #ifdef ALTQ 756 static int 757 pf_state_hash(struct pf_kstate *s) 758 { 759 u_int32_t hv = (intptr_t)s / sizeof(*s); 760 761 hv ^= crc32(&s->src, sizeof(s->src)); 762 hv ^= crc32(&s->dst, sizeof(s->dst)); 763 if (hv == 0) 764 hv = 1; 765 return (hv); 766 } 767 #endif /* ALTQ */ 768 769 static __inline void 770 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 771 { 772 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 773 s->dst.state = newstate; 774 if (which == PF_PEER_DST) 775 return; 776 if (s->src.state == newstate) 777 return; 778 if (s->creatorid == V_pf_status.hostid && 779 s->key[PF_SK_STACK] != NULL && 780 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 781 !(TCPS_HAVEESTABLISHED(s->src.state) || 782 s->src.state == TCPS_CLOSED) && 783 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 784 atomic_add_32(&V_pf_status.states_halfopen, -1); 785 786 s->src.state = newstate; 787 } 788 789 bool 790 pf_init_threshold(struct pf_kthreshold *threshold, 791 u_int32_t limit, u_int32_t seconds) 792 { 793 threshold->limit = limit; 794 threshold->seconds = seconds; 795 threshold->cr = counter_rate_alloc(M_NOWAIT, seconds); 796 797 return (threshold->cr != NULL); 798 } 799 800 static int 801 pf_check_threshold(struct pf_kthreshold *threshold) 802 { 803 return (counter_ratecheck(threshold->cr, threshold->limit) < 0); 804 } 805 806 static bool 807 pf_src_connlimit(struct pf_kstate *state) 808 { 809 struct pf_overload_entry *pfoe; 810 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 811 bool limited = false; 812 813 PF_STATE_LOCK_ASSERT(state); 814 PF_SRC_NODE_LOCK(src_node); 815 816 src_node->conn++; 817 state->src.tcp_est = 1; 818 819 if (state->rule->max_src_conn && 820 state->rule->max_src_conn < 821 src_node->conn) { 822 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 823 limited = true; 824 } 825 826 if (state->rule->max_src_conn_rate.limit && 827 pf_check_threshold(&src_node->conn_rate)) { 828 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 829 limited = true; 830 } 831 832 if (!limited) 833 goto done; 834 835 /* Kill this state. */ 836 state->timeout = PFTM_PURGE; 837 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 838 839 if (state->rule->overload_tbl == NULL) 840 goto done; 841 842 /* Schedule overloading and flushing task. */ 843 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 844 if (pfoe == NULL) 845 goto done; /* too bad :( */ 846 847 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 848 pfoe->af = state->key[PF_SK_WIRE]->af; 849 pfoe->rule = state->rule; 850 pfoe->dir = state->direction; 851 PF_OVERLOADQ_LOCK(); 852 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 853 PF_OVERLOADQ_UNLOCK(); 854 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 855 856 done: 857 PF_SRC_NODE_UNLOCK(src_node); 858 return (limited); 859 } 860 861 static void 862 pf_overload_task(void *v, int pending) 863 { 864 struct pf_overload_head queue; 865 struct pfr_addr p; 866 struct pf_overload_entry *pfoe, *pfoe1; 867 uint32_t killed = 0; 868 869 CURVNET_SET((struct vnet *)v); 870 871 PF_OVERLOADQ_LOCK(); 872 queue = V_pf_overloadqueue; 873 SLIST_INIT(&V_pf_overloadqueue); 874 PF_OVERLOADQ_UNLOCK(); 875 876 bzero(&p, sizeof(p)); 877 SLIST_FOREACH(pfoe, &queue, next) { 878 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 879 if (V_pf_status.debug >= PF_DEBUG_MISC) { 880 printf("%s: blocking address ", __func__); 881 pf_print_host(&pfoe->addr, 0, pfoe->af); 882 printf("\n"); 883 } 884 885 p.pfra_af = pfoe->af; 886 switch (pfoe->af) { 887 #ifdef INET 888 case AF_INET: 889 p.pfra_net = 32; 890 p.pfra_ip4addr = pfoe->addr.v4; 891 break; 892 #endif /* INET */ 893 #ifdef INET6 894 case AF_INET6: 895 p.pfra_net = 128; 896 p.pfra_ip6addr = pfoe->addr.v6; 897 break; 898 #endif /* INET6 */ 899 default: 900 unhandled_af(pfoe->af); 901 } 902 903 PF_RULES_WLOCK(); 904 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 905 PF_RULES_WUNLOCK(); 906 } 907 908 /* 909 * Remove those entries, that don't need flushing. 910 */ 911 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 912 if (pfoe->rule->flush == 0) { 913 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 914 free(pfoe, M_PFTEMP); 915 } else 916 counter_u64_add( 917 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 918 919 /* If nothing to flush, return. */ 920 if (SLIST_EMPTY(&queue)) { 921 CURVNET_RESTORE(); 922 return; 923 } 924 925 for (int i = 0; i <= V_pf_hashmask; i++) { 926 struct pf_idhash *ih = &V_pf_idhash[i]; 927 struct pf_state_key *sk; 928 struct pf_kstate *s; 929 930 PF_HASHROW_LOCK(ih); 931 LIST_FOREACH(s, &ih->states, entry) { 932 sk = s->key[PF_SK_WIRE]; 933 SLIST_FOREACH(pfoe, &queue, next) 934 if (sk->af == pfoe->af && 935 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 936 pfoe->rule == s->rule) && 937 ((pfoe->dir == PF_OUT && 938 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 939 (pfoe->dir == PF_IN && 940 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 941 s->timeout = PFTM_PURGE; 942 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 943 killed++; 944 } 945 } 946 PF_HASHROW_UNLOCK(ih); 947 } 948 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 949 free(pfoe, M_PFTEMP); 950 if (V_pf_status.debug >= PF_DEBUG_MISC) 951 printf("%s: %u states killed", __func__, killed); 952 953 CURVNET_RESTORE(); 954 } 955 956 /* 957 * On node found always returns locked. On not found its configurable. 958 */ 959 struct pf_ksrc_node * 960 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 961 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 962 { 963 struct pf_ksrc_node *n; 964 965 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 966 967 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 968 PF_HASHROW_LOCK(*sh); 969 LIST_FOREACH(n, &(*sh)->nodes, entry) 970 if (n->rule == rule && n->af == af && n->type == sn_type && 971 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 972 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 973 break; 974 975 if (n == NULL && !returnlocked) 976 PF_HASHROW_UNLOCK(*sh); 977 978 return (n); 979 } 980 981 bool 982 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 983 { 984 struct pf_ksrc_node *cur; 985 986 if ((*sn) == NULL) 987 return (false); 988 989 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 990 991 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 992 PF_HASHROW_LOCK(sh); 993 LIST_FOREACH(cur, &(sh->nodes), entry) { 994 if (cur == (*sn) && 995 cur->expire != 1) /* Ignore nodes being killed */ 996 return (true); 997 } 998 PF_HASHROW_UNLOCK(sh); 999 (*sn) = NULL; 1000 return (false); 1001 } 1002 1003 static void 1004 pf_free_src_node(struct pf_ksrc_node *sn) 1005 { 1006 1007 for (int i = 0; i < 2; i++) { 1008 counter_u64_free(sn->bytes[i]); 1009 counter_u64_free(sn->packets[i]); 1010 } 1011 counter_rate_free(sn->conn_rate.cr); 1012 uma_zfree(V_pf_sources_z, sn); 1013 } 1014 1015 static u_short 1016 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1017 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1018 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1019 struct pfi_kkif *rkif, pf_sn_types_t sn_type) 1020 { 1021 u_short reason = 0; 1022 struct pf_krule *r_track = rule; 1023 struct pf_ksrc_node **sn = &(sns[sn_type]); 1024 struct pf_srchash **sh = &(snhs[sn_type]); 1025 1026 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1027 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1028 1029 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1030 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1031 1032 /* 1033 * XXX: There could be a KASSERT for 1034 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1035 * but we'd need to pass pool *only* for this KASSERT. 1036 */ 1037 1038 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1039 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1040 r_track = &V_pf_default_rule; 1041 1042 /* 1043 * Request the sh to always be locked, as we might insert a new sn. 1044 */ 1045 if (*sn == NULL) 1046 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1047 1048 if (*sn == NULL) { 1049 PF_HASHROW_ASSERT(*sh); 1050 1051 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1052 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1053 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1054 reason = PFRES_SRCLIMIT; 1055 goto done; 1056 } 1057 1058 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1059 if ((*sn) == NULL) { 1060 reason = PFRES_MEMORY; 1061 goto done; 1062 } 1063 1064 for (int i = 0; i < 2; i++) { 1065 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1066 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1067 1068 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1069 pf_free_src_node(*sn); 1070 reason = PFRES_MEMORY; 1071 goto done; 1072 } 1073 } 1074 1075 if (sn_type == PF_SN_LIMIT) 1076 if (! pf_init_threshold(&(*sn)->conn_rate, 1077 rule->max_src_conn_rate.limit, 1078 rule->max_src_conn_rate.seconds)) { 1079 pf_free_src_node(*sn); 1080 reason = PFRES_MEMORY; 1081 goto done; 1082 } 1083 1084 MPASS((*sn)->lock == NULL); 1085 (*sn)->lock = &(*sh)->lock; 1086 1087 (*sn)->af = af; 1088 (*sn)->rule = r_track; 1089 pf_addrcpy(&(*sn)->addr, src, af); 1090 if (raddr != NULL) 1091 pf_addrcpy(&(*sn)->raddr, raddr, af); 1092 (*sn)->rkif = rkif; 1093 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1094 (*sn)->creation = time_uptime; 1095 (*sn)->ruletype = rule->action; 1096 (*sn)->type = sn_type; 1097 counter_u64_add(r_track->src_nodes[sn_type], 1); 1098 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1099 } else { 1100 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1101 (*sn)->states >= rule->max_src_states) { 1102 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1103 1); 1104 reason = PFRES_SRCLIMIT; 1105 goto done; 1106 } 1107 } 1108 done: 1109 if (reason == 0) 1110 (*sn)->states++; 1111 else 1112 (*sn) = NULL; 1113 1114 PF_HASHROW_UNLOCK(*sh); 1115 return (reason); 1116 } 1117 1118 void 1119 pf_unlink_src_node(struct pf_ksrc_node *src) 1120 { 1121 PF_SRC_NODE_LOCK_ASSERT(src); 1122 1123 LIST_REMOVE(src, entry); 1124 if (src->rule) 1125 counter_u64_add(src->rule->src_nodes[src->type], -1); 1126 } 1127 1128 u_int 1129 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1130 { 1131 struct pf_ksrc_node *sn, *tmp; 1132 u_int count = 0; 1133 1134 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1135 pf_free_src_node(sn); 1136 count++; 1137 } 1138 1139 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1140 1141 return (count); 1142 } 1143 1144 void 1145 pf_mtag_initialize(void) 1146 { 1147 1148 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1149 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1150 UMA_ALIGN_PTR, 0); 1151 } 1152 1153 /* Per-vnet data storage structures initialization. */ 1154 void 1155 pf_initialize(void) 1156 { 1157 struct pf_keyhash *kh; 1158 struct pf_idhash *ih; 1159 struct pf_srchash *sh; 1160 struct pf_udpendpointhash *uh; 1161 u_int i; 1162 1163 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1164 V_pf_hashsize = PF_HASHSIZ; 1165 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1166 V_pf_srchashsize = PF_SRCHASHSIZ; 1167 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1168 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1169 1170 V_pf_hashseed = arc4random(); 1171 1172 /* States and state keys storage. */ 1173 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1174 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1175 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1176 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1177 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1178 1179 V_pf_state_key_z = uma_zcreate("pf state keys", 1180 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1181 UMA_ALIGN_PTR, 0); 1182 1183 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1184 M_PFHASH, M_NOWAIT | M_ZERO); 1185 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1186 M_PFHASH, M_NOWAIT | M_ZERO); 1187 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1188 printf("pf: Unable to allocate memory for " 1189 "state_hashsize %lu.\n", V_pf_hashsize); 1190 1191 free(V_pf_keyhash, M_PFHASH); 1192 free(V_pf_idhash, M_PFHASH); 1193 1194 V_pf_hashsize = PF_HASHSIZ; 1195 V_pf_keyhash = mallocarray(V_pf_hashsize, 1196 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1197 V_pf_idhash = mallocarray(V_pf_hashsize, 1198 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1199 } 1200 1201 V_pf_hashmask = V_pf_hashsize - 1; 1202 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1203 i++, kh++, ih++) { 1204 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1205 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1206 } 1207 1208 /* Source nodes. */ 1209 V_pf_sources_z = uma_zcreate("pf source nodes", 1210 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1211 0); 1212 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1213 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1214 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1215 1216 V_pf_srchash = mallocarray(V_pf_srchashsize, 1217 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1218 if (V_pf_srchash == NULL) { 1219 printf("pf: Unable to allocate memory for " 1220 "source_hashsize %lu.\n", V_pf_srchashsize); 1221 1222 V_pf_srchashsize = PF_SRCHASHSIZ; 1223 V_pf_srchash = mallocarray(V_pf_srchashsize, 1224 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1225 } 1226 1227 V_pf_srchashmask = V_pf_srchashsize - 1; 1228 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1229 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1230 1231 1232 /* UDP endpoint mappings. */ 1233 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1234 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1235 UMA_ALIGN_PTR, 0); 1236 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1237 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1238 if (V_pf_udpendpointhash == NULL) { 1239 printf("pf: Unable to allocate memory for " 1240 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1241 1242 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1243 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1244 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1245 } 1246 1247 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1248 for (i = 0, uh = V_pf_udpendpointhash; 1249 i <= V_pf_udpendpointhashmask; 1250 i++, uh++) { 1251 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1252 MTX_DEF | MTX_DUPOK); 1253 } 1254 1255 /* ALTQ */ 1256 TAILQ_INIT(&V_pf_altqs[0]); 1257 TAILQ_INIT(&V_pf_altqs[1]); 1258 TAILQ_INIT(&V_pf_altqs[2]); 1259 TAILQ_INIT(&V_pf_altqs[3]); 1260 TAILQ_INIT(&V_pf_pabuf[0]); 1261 TAILQ_INIT(&V_pf_pabuf[1]); 1262 TAILQ_INIT(&V_pf_pabuf[2]); 1263 V_pf_altqs_active = &V_pf_altqs[0]; 1264 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1265 V_pf_altqs_inactive = &V_pf_altqs[2]; 1266 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1267 1268 /* Send & overload+flush queues. */ 1269 STAILQ_INIT(&V_pf_sendqueue); 1270 SLIST_INIT(&V_pf_overloadqueue); 1271 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1272 1273 /* Unlinked, but may be referenced rules. */ 1274 TAILQ_INIT(&V_pf_unlinked_rules); 1275 } 1276 1277 void 1278 pf_mtag_cleanup(void) 1279 { 1280 1281 uma_zdestroy(pf_mtag_z); 1282 } 1283 1284 void 1285 pf_cleanup(void) 1286 { 1287 struct pf_keyhash *kh; 1288 struct pf_idhash *ih; 1289 struct pf_srchash *sh; 1290 struct pf_udpendpointhash *uh; 1291 struct pf_send_entry *pfse, *next; 1292 u_int i; 1293 1294 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1295 i <= V_pf_hashmask; 1296 i++, kh++, ih++) { 1297 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1298 __func__)); 1299 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1300 __func__)); 1301 mtx_destroy(&kh->lock); 1302 mtx_destroy(&ih->lock); 1303 } 1304 free(V_pf_keyhash, M_PFHASH); 1305 free(V_pf_idhash, M_PFHASH); 1306 1307 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1308 KASSERT(LIST_EMPTY(&sh->nodes), 1309 ("%s: source node hash not empty", __func__)); 1310 mtx_destroy(&sh->lock); 1311 } 1312 free(V_pf_srchash, M_PFHASH); 1313 1314 for (i = 0, uh = V_pf_udpendpointhash; 1315 i <= V_pf_udpendpointhashmask; 1316 i++, uh++) { 1317 KASSERT(LIST_EMPTY(&uh->endpoints), 1318 ("%s: udp endpoint hash not empty", __func__)); 1319 mtx_destroy(&uh->lock); 1320 } 1321 free(V_pf_udpendpointhash, M_PFHASH); 1322 1323 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1324 m_freem(pfse->pfse_m); 1325 free(pfse, M_PFTEMP); 1326 } 1327 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1328 1329 uma_zdestroy(V_pf_sources_z); 1330 uma_zdestroy(V_pf_state_z); 1331 uma_zdestroy(V_pf_state_key_z); 1332 uma_zdestroy(V_pf_udp_mapping_z); 1333 } 1334 1335 static int 1336 pf_mtag_uminit(void *mem, int size, int how) 1337 { 1338 struct m_tag *t; 1339 1340 t = (struct m_tag *)mem; 1341 t->m_tag_cookie = MTAG_ABI_COMPAT; 1342 t->m_tag_id = PACKET_TAG_PF; 1343 t->m_tag_len = sizeof(struct pf_mtag); 1344 t->m_tag_free = pf_mtag_free; 1345 1346 return (0); 1347 } 1348 1349 static void 1350 pf_mtag_free(struct m_tag *t) 1351 { 1352 1353 uma_zfree(pf_mtag_z, t); 1354 } 1355 1356 struct pf_mtag * 1357 pf_get_mtag(struct mbuf *m) 1358 { 1359 struct m_tag *mtag; 1360 1361 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1362 return ((struct pf_mtag *)(mtag + 1)); 1363 1364 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1365 if (mtag == NULL) 1366 return (NULL); 1367 bzero(mtag + 1, sizeof(struct pf_mtag)); 1368 m_tag_prepend(m, mtag); 1369 1370 return ((struct pf_mtag *)(mtag + 1)); 1371 } 1372 1373 static int 1374 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1375 struct pf_kstate *s) 1376 { 1377 struct pf_keyhash *khs, *khw, *kh; 1378 struct pf_state_key *sk, *cur; 1379 struct pf_kstate *si, *olds = NULL; 1380 int idx; 1381 1382 NET_EPOCH_ASSERT(); 1383 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1384 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1385 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1386 1387 /* 1388 * We need to lock hash slots of both keys. To avoid deadlock 1389 * we always lock the slot with lower address first. Unlock order 1390 * isn't important. 1391 * 1392 * We also need to lock ID hash slot before dropping key 1393 * locks. On success we return with ID hash slot locked. 1394 */ 1395 1396 if (skw == sks) { 1397 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1398 PF_HASHROW_LOCK(khs); 1399 } else { 1400 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1401 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1402 if (khs == khw) { 1403 PF_HASHROW_LOCK(khs); 1404 } else if (khs < khw) { 1405 PF_HASHROW_LOCK(khs); 1406 PF_HASHROW_LOCK(khw); 1407 } else { 1408 PF_HASHROW_LOCK(khw); 1409 PF_HASHROW_LOCK(khs); 1410 } 1411 } 1412 1413 #define KEYS_UNLOCK() do { \ 1414 if (khs != khw) { \ 1415 PF_HASHROW_UNLOCK(khs); \ 1416 PF_HASHROW_UNLOCK(khw); \ 1417 } else \ 1418 PF_HASHROW_UNLOCK(khs); \ 1419 } while (0) 1420 1421 /* 1422 * First run: start with wire key. 1423 */ 1424 sk = skw; 1425 kh = khw; 1426 idx = PF_SK_WIRE; 1427 1428 MPASS(s->lock == NULL); 1429 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1430 1431 keyattach: 1432 LIST_FOREACH(cur, &kh->keys, entry) 1433 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1434 break; 1435 1436 if (cur != NULL) { 1437 /* Key exists. Check for same kif, if none, add to key. */ 1438 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1439 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1440 1441 PF_HASHROW_LOCK(ih); 1442 if (si->kif == s->kif && 1443 ((si->key[PF_SK_WIRE]->af == sk->af && 1444 si->direction == s->direction) || 1445 (si->key[PF_SK_WIRE]->af != 1446 si->key[PF_SK_STACK]->af && 1447 sk->af == si->key[PF_SK_STACK]->af && 1448 si->direction != s->direction))) { 1449 bool reuse = false; 1450 1451 if (sk->proto == IPPROTO_TCP && 1452 si->src.state >= TCPS_FIN_WAIT_2 && 1453 si->dst.state >= TCPS_FIN_WAIT_2) 1454 reuse = true; 1455 1456 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1457 printf("pf: %s key attach " 1458 "%s on %s: ", 1459 (idx == PF_SK_WIRE) ? 1460 "wire" : "stack", 1461 reuse ? "reuse" : "failed", 1462 s->kif->pfik_name); 1463 pf_print_state_parts(s, 1464 (idx == PF_SK_WIRE) ? 1465 sk : NULL, 1466 (idx == PF_SK_STACK) ? 1467 sk : NULL); 1468 printf(", existing: "); 1469 pf_print_state_parts(si, 1470 (idx == PF_SK_WIRE) ? 1471 sk : NULL, 1472 (idx == PF_SK_STACK) ? 1473 sk : NULL); 1474 printf("\n"); 1475 } 1476 1477 if (reuse) { 1478 /* 1479 * New state matches an old >FIN_WAIT_2 1480 * state. We can't drop key hash locks, 1481 * thus we can't unlink it properly. 1482 * 1483 * As a workaround we drop it into 1484 * TCPS_CLOSED state, schedule purge 1485 * ASAP and push it into the very end 1486 * of the slot TAILQ, so that it won't 1487 * conflict with our new state. 1488 */ 1489 pf_set_protostate(si, PF_PEER_BOTH, 1490 TCPS_CLOSED); 1491 si->timeout = PFTM_PURGE; 1492 olds = si; 1493 } else { 1494 s->timeout = PFTM_UNLINKED; 1495 if (idx == PF_SK_STACK) 1496 /* 1497 * Remove the wire key from 1498 * the hash. Other threads 1499 * can't be referencing it 1500 * because we still hold the 1501 * hash lock. 1502 */ 1503 pf_state_key_detach(s, 1504 PF_SK_WIRE); 1505 PF_HASHROW_UNLOCK(ih); 1506 KEYS_UNLOCK(); 1507 if (idx == PF_SK_WIRE) 1508 /* 1509 * We've not inserted either key. 1510 * Free both. 1511 */ 1512 uma_zfree(V_pf_state_key_z, skw); 1513 if (skw != sks) 1514 uma_zfree( 1515 V_pf_state_key_z, 1516 sks); 1517 return (EEXIST); /* collision! */ 1518 } 1519 } 1520 PF_HASHROW_UNLOCK(ih); 1521 } 1522 uma_zfree(V_pf_state_key_z, sk); 1523 s->key[idx] = cur; 1524 } else { 1525 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1526 s->key[idx] = sk; 1527 } 1528 1529 stateattach: 1530 /* List is sorted, if-bound states before floating. */ 1531 if (s->kif == V_pfi_all) 1532 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1533 else 1534 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1535 1536 if (olds) { 1537 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1538 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1539 key_list[idx]); 1540 olds = NULL; 1541 } 1542 1543 /* 1544 * Attach done. See how should we (or should not?) 1545 * attach a second key. 1546 */ 1547 if (sks == skw) { 1548 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1549 idx = PF_SK_STACK; 1550 sks = NULL; 1551 goto stateattach; 1552 } else if (sks != NULL) { 1553 /* 1554 * Continue attaching with stack key. 1555 */ 1556 sk = sks; 1557 kh = khs; 1558 idx = PF_SK_STACK; 1559 sks = NULL; 1560 goto keyattach; 1561 } 1562 1563 PF_STATE_LOCK(s); 1564 KEYS_UNLOCK(); 1565 1566 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1567 ("%s failure", __func__)); 1568 1569 return (0); 1570 #undef KEYS_UNLOCK 1571 } 1572 1573 static void 1574 pf_detach_state(struct pf_kstate *s) 1575 { 1576 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1577 struct pf_keyhash *kh; 1578 1579 NET_EPOCH_ASSERT(); 1580 MPASS(s->timeout >= PFTM_MAX); 1581 1582 pf_sctp_multihome_detach_addr(s); 1583 1584 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1585 V_pflow_export_state_ptr(s); 1586 1587 if (sks != NULL) { 1588 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1589 PF_HASHROW_LOCK(kh); 1590 if (s->key[PF_SK_STACK] != NULL) 1591 pf_state_key_detach(s, PF_SK_STACK); 1592 /* 1593 * If both point to same key, then we are done. 1594 */ 1595 if (sks == s->key[PF_SK_WIRE]) { 1596 pf_state_key_detach(s, PF_SK_WIRE); 1597 PF_HASHROW_UNLOCK(kh); 1598 return; 1599 } 1600 PF_HASHROW_UNLOCK(kh); 1601 } 1602 1603 if (s->key[PF_SK_WIRE] != NULL) { 1604 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1605 PF_HASHROW_LOCK(kh); 1606 if (s->key[PF_SK_WIRE] != NULL) 1607 pf_state_key_detach(s, PF_SK_WIRE); 1608 PF_HASHROW_UNLOCK(kh); 1609 } 1610 } 1611 1612 static void 1613 pf_state_key_detach(struct pf_kstate *s, int idx) 1614 { 1615 struct pf_state_key *sk = s->key[idx]; 1616 #ifdef INVARIANTS 1617 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1618 1619 PF_HASHROW_ASSERT(kh); 1620 #endif /* INVARIANTS */ 1621 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1622 s->key[idx] = NULL; 1623 1624 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1625 LIST_REMOVE(sk, entry); 1626 uma_zfree(V_pf_state_key_z, sk); 1627 } 1628 } 1629 1630 static int 1631 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1632 { 1633 struct pf_state_key *sk = mem; 1634 1635 bzero(sk, sizeof(struct pf_state_key_cmp)); 1636 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1637 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1638 1639 return (0); 1640 } 1641 1642 static int 1643 pf_state_key_addr_setup(struct pf_pdesc *pd, 1644 struct pf_state_key_cmp *key, int multi) 1645 { 1646 struct pf_addr *saddr = pd->src; 1647 struct pf_addr *daddr = pd->dst; 1648 #ifdef INET6 1649 struct nd_neighbor_solicit nd; 1650 struct pf_addr *target; 1651 u_short action, reason; 1652 1653 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1654 goto copy; 1655 1656 switch (pd->hdr.icmp6.icmp6_type) { 1657 case ND_NEIGHBOR_SOLICIT: 1658 if (multi) 1659 return (-1); 1660 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1661 return (-1); 1662 target = (struct pf_addr *)&nd.nd_ns_target; 1663 daddr = target; 1664 break; 1665 case ND_NEIGHBOR_ADVERT: 1666 if (multi) 1667 return (-1); 1668 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1669 return (-1); 1670 target = (struct pf_addr *)&nd.nd_ns_target; 1671 saddr = target; 1672 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1673 key->addr[pd->didx].addr32[0] = 0; 1674 key->addr[pd->didx].addr32[1] = 0; 1675 key->addr[pd->didx].addr32[2] = 0; 1676 key->addr[pd->didx].addr32[3] = 0; 1677 daddr = NULL; /* overwritten */ 1678 } 1679 break; 1680 default: 1681 if (multi) { 1682 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1683 key->addr[pd->sidx].addr32[1] = 0; 1684 key->addr[pd->sidx].addr32[2] = 0; 1685 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1686 saddr = NULL; /* overwritten */ 1687 } 1688 } 1689 copy: 1690 #endif /* INET6 */ 1691 if (saddr) 1692 pf_addrcpy(&key->addr[pd->sidx], saddr, pd->af); 1693 if (daddr) 1694 pf_addrcpy(&key->addr[pd->didx], daddr, pd->af); 1695 1696 return (0); 1697 } 1698 1699 int 1700 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1701 struct pf_state_key **sk, struct pf_state_key **nk) 1702 { 1703 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1704 if (*sk == NULL) 1705 return (ENOMEM); 1706 1707 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1708 0)) { 1709 uma_zfree(V_pf_state_key_z, *sk); 1710 *sk = NULL; 1711 return (ENOMEM); 1712 } 1713 1714 (*sk)->port[pd->sidx] = sport; 1715 (*sk)->port[pd->didx] = dport; 1716 (*sk)->proto = pd->proto; 1717 (*sk)->af = pd->af; 1718 1719 *nk = pf_state_key_clone(*sk); 1720 if (*nk == NULL) { 1721 uma_zfree(V_pf_state_key_z, *sk); 1722 *sk = NULL; 1723 return (ENOMEM); 1724 } 1725 1726 if (pd->af != pd->naf) { 1727 (*sk)->port[pd->sidx] = pd->osport; 1728 (*sk)->port[pd->didx] = pd->odport; 1729 1730 (*nk)->af = pd->naf; 1731 1732 /* 1733 * We're overwriting an address here, so potentially there's bits of an IPv6 1734 * address left in here. Clear that out first. 1735 */ 1736 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1737 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1738 if (pd->dir == PF_IN) { 1739 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->nsaddr, 1740 pd->naf); 1741 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->ndaddr, 1742 pd->naf); 1743 (*nk)->port[pd->didx] = pd->nsport; 1744 (*nk)->port[pd->sidx] = pd->ndport; 1745 } else { 1746 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->nsaddr, 1747 pd->naf); 1748 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->ndaddr, 1749 pd->naf); 1750 (*nk)->port[pd->sidx] = pd->nsport; 1751 (*nk)->port[pd->didx] = pd->ndport; 1752 } 1753 1754 switch (pd->proto) { 1755 case IPPROTO_ICMP: 1756 (*nk)->proto = IPPROTO_ICMPV6; 1757 break; 1758 case IPPROTO_ICMPV6: 1759 (*nk)->proto = IPPROTO_ICMP; 1760 break; 1761 default: 1762 (*nk)->proto = pd->proto; 1763 } 1764 } 1765 1766 return (0); 1767 } 1768 1769 struct pf_state_key * 1770 pf_state_key_clone(const struct pf_state_key *orig) 1771 { 1772 struct pf_state_key *sk; 1773 1774 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1775 if (sk == NULL) 1776 return (NULL); 1777 1778 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1779 1780 return (sk); 1781 } 1782 1783 int 1784 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1785 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1786 { 1787 struct pf_idhash *ih; 1788 struct pf_kstate *cur; 1789 int error; 1790 1791 NET_EPOCH_ASSERT(); 1792 1793 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1794 ("%s: sks not pristine", __func__)); 1795 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1796 ("%s: skw not pristine", __func__)); 1797 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1798 1799 s->kif = kif; 1800 s->orig_kif = orig_kif; 1801 1802 if (s->id == 0 && s->creatorid == 0) { 1803 s->id = alloc_unr64(&V_pf_stateid); 1804 s->id = htobe64(s->id); 1805 s->creatorid = V_pf_status.hostid; 1806 } 1807 1808 /* Returns with ID locked on success. */ 1809 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1810 return (error); 1811 skw = sks = NULL; 1812 1813 ih = &V_pf_idhash[PF_IDHASH(s)]; 1814 PF_HASHROW_ASSERT(ih); 1815 LIST_FOREACH(cur, &ih->states, entry) 1816 if (cur->id == s->id && cur->creatorid == s->creatorid) 1817 break; 1818 1819 if (cur != NULL) { 1820 s->timeout = PFTM_UNLINKED; 1821 PF_HASHROW_UNLOCK(ih); 1822 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1823 printf("pf: state ID collision: " 1824 "id: %016llx creatorid: %08x\n", 1825 (unsigned long long)be64toh(s->id), 1826 ntohl(s->creatorid)); 1827 } 1828 pf_detach_state(s); 1829 return (EEXIST); 1830 } 1831 LIST_INSERT_HEAD(&ih->states, s, entry); 1832 /* One for keys, one for ID hash. */ 1833 refcount_init(&s->refs, 2); 1834 1835 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1836 if (V_pfsync_insert_state_ptr != NULL) 1837 V_pfsync_insert_state_ptr(s); 1838 1839 /* Returns locked. */ 1840 return (0); 1841 } 1842 1843 /* 1844 * Find state by ID: returns with locked row on success. 1845 */ 1846 struct pf_kstate * 1847 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1848 { 1849 struct pf_idhash *ih; 1850 struct pf_kstate *s; 1851 1852 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1853 1854 ih = &V_pf_idhash[PF_IDHASHID(id)]; 1855 1856 PF_HASHROW_LOCK(ih); 1857 LIST_FOREACH(s, &ih->states, entry) 1858 if (s->id == id && s->creatorid == creatorid) 1859 break; 1860 1861 if (s == NULL) 1862 PF_HASHROW_UNLOCK(ih); 1863 1864 return (s); 1865 } 1866 1867 /* 1868 * Find state by key. 1869 * Returns with ID hash slot locked on success. 1870 */ 1871 static int 1872 pf_find_state(struct pf_pdesc *pd, const struct pf_state_key_cmp *key, 1873 struct pf_kstate **state) 1874 { 1875 struct pf_keyhash *kh; 1876 struct pf_state_key *sk; 1877 struct pf_kstate *s; 1878 int idx; 1879 1880 *state = NULL; 1881 1882 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1883 1884 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1885 1886 PF_HASHROW_LOCK(kh); 1887 LIST_FOREACH(sk, &kh->keys, entry) 1888 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1889 break; 1890 if (sk == NULL) { 1891 PF_HASHROW_UNLOCK(kh); 1892 return (PF_DROP); 1893 } 1894 1895 idx = (pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1896 1897 /* List is sorted, if-bound states before floating ones. */ 1898 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1899 if (s->kif == V_pfi_all || s->kif == pd->kif || 1900 s->orig_kif == pd->kif) { 1901 PF_STATE_LOCK(s); 1902 PF_HASHROW_UNLOCK(kh); 1903 if (__predict_false(s->timeout >= PFTM_MAX)) { 1904 /* 1905 * State is either being processed by 1906 * pf_remove_state() in an other thread, or 1907 * is scheduled for immediate expiry. 1908 */ 1909 PF_STATE_UNLOCK(s); 1910 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 1911 key, (pd->dir), pd, *state); 1912 return (PF_DROP); 1913 } 1914 goto out; 1915 } 1916 1917 /* Look through the other list, in case of AF-TO */ 1918 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1919 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1920 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1921 continue; 1922 if (s->kif == V_pfi_all || s->kif == pd->kif || 1923 s->orig_kif == pd->kif) { 1924 PF_STATE_LOCK(s); 1925 PF_HASHROW_UNLOCK(kh); 1926 if (__predict_false(s->timeout >= PFTM_MAX)) { 1927 /* 1928 * State is either being processed by 1929 * pf_remove_state() in an other thread, or 1930 * is scheduled for immediate expiry. 1931 */ 1932 PF_STATE_UNLOCK(s); 1933 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 1934 key, (pd->dir), pd, NULL); 1935 return (PF_DROP); 1936 } 1937 goto out; 1938 } 1939 } 1940 1941 PF_HASHROW_UNLOCK(kh); 1942 1943 out: 1944 SDT_PROBE5(pf, ip, state, lookup, pd->kif, key, (pd->dir), pd, *state); 1945 1946 if (s == NULL || s->timeout == PFTM_PURGE) { 1947 if (s) 1948 PF_STATE_UNLOCK(s); 1949 return (PF_DROP); 1950 } 1951 1952 if ((s)->rule->pktrate.limit && pd->dir == (s)->direction) { 1953 if (pf_check_threshold(&(s)->rule->pktrate)) { 1954 PF_STATE_UNLOCK(s); 1955 return (PF_DROP); 1956 } 1957 } 1958 if (PACKET_LOOPED(pd)) { 1959 PF_STATE_UNLOCK(s); 1960 return (PF_PASS); 1961 } 1962 1963 *state = s; 1964 1965 return (PF_MATCH); 1966 } 1967 1968 /* 1969 * Returns with ID hash slot locked on success. 1970 */ 1971 struct pf_kstate * 1972 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1973 { 1974 struct pf_keyhash *kh; 1975 struct pf_state_key *sk; 1976 struct pf_kstate *s, *ret = NULL; 1977 int idx, inout = 0; 1978 1979 if (more != NULL) 1980 *more = 0; 1981 1982 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1983 1984 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1985 1986 PF_HASHROW_LOCK(kh); 1987 LIST_FOREACH(sk, &kh->keys, entry) 1988 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1989 break; 1990 if (sk == NULL) { 1991 PF_HASHROW_UNLOCK(kh); 1992 return (NULL); 1993 } 1994 switch (dir) { 1995 case PF_IN: 1996 idx = PF_SK_WIRE; 1997 break; 1998 case PF_OUT: 1999 idx = PF_SK_STACK; 2000 break; 2001 case PF_INOUT: 2002 idx = PF_SK_WIRE; 2003 inout = 1; 2004 break; 2005 default: 2006 panic("%s: dir %u", __func__, dir); 2007 } 2008 second_run: 2009 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 2010 if (more == NULL) { 2011 PF_STATE_LOCK(s); 2012 PF_HASHROW_UNLOCK(kh); 2013 return (s); 2014 } 2015 2016 if (ret) 2017 (*more)++; 2018 else { 2019 ret = s; 2020 PF_STATE_LOCK(s); 2021 } 2022 } 2023 if (inout == 1) { 2024 inout = 0; 2025 idx = PF_SK_STACK; 2026 goto second_run; 2027 } 2028 PF_HASHROW_UNLOCK(kh); 2029 2030 return (ret); 2031 } 2032 2033 /* 2034 * FIXME 2035 * This routine is inefficient -- locks the state only to unlock immediately on 2036 * return. 2037 * It is racy -- after the state is unlocked nothing stops other threads from 2038 * removing it. 2039 */ 2040 bool 2041 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2042 { 2043 struct pf_kstate *s; 2044 2045 s = pf_find_state_all(key, dir, NULL); 2046 if (s != NULL) { 2047 PF_STATE_UNLOCK(s); 2048 return (true); 2049 } 2050 return (false); 2051 } 2052 2053 struct pf_udp_mapping * 2054 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2055 struct pf_addr *nat_addr, uint16_t nat_port) 2056 { 2057 struct pf_udp_mapping *mapping; 2058 2059 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2060 if (mapping == NULL) 2061 return (NULL); 2062 pf_addrcpy(&mapping->endpoints[0].addr, src_addr, af); 2063 mapping->endpoints[0].port = src_port; 2064 mapping->endpoints[0].af = af; 2065 mapping->endpoints[0].mapping = mapping; 2066 pf_addrcpy(&mapping->endpoints[1].addr, nat_addr, af); 2067 mapping->endpoints[1].port = nat_port; 2068 mapping->endpoints[1].af = af; 2069 mapping->endpoints[1].mapping = mapping; 2070 refcount_init(&mapping->refs, 1); 2071 return (mapping); 2072 } 2073 2074 int 2075 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2076 { 2077 struct pf_udpendpointhash *h0, *h1; 2078 struct pf_udp_endpoint *endpoint; 2079 int ret = EEXIST; 2080 2081 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2082 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2083 if (h0 == h1) { 2084 PF_HASHROW_LOCK(h0); 2085 } else if (h0 < h1) { 2086 PF_HASHROW_LOCK(h0); 2087 PF_HASHROW_LOCK(h1); 2088 } else { 2089 PF_HASHROW_LOCK(h1); 2090 PF_HASHROW_LOCK(h0); 2091 } 2092 2093 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2094 if (bcmp(endpoint, &mapping->endpoints[0], 2095 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2096 break; 2097 } 2098 if (endpoint != NULL) 2099 goto cleanup; 2100 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2101 if (bcmp(endpoint, &mapping->endpoints[1], 2102 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2103 break; 2104 } 2105 if (endpoint != NULL) 2106 goto cleanup; 2107 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2108 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2109 ret = 0; 2110 2111 cleanup: 2112 if (h0 != h1) { 2113 PF_HASHROW_UNLOCK(h0); 2114 PF_HASHROW_UNLOCK(h1); 2115 } else { 2116 PF_HASHROW_UNLOCK(h0); 2117 } 2118 return (ret); 2119 } 2120 2121 void 2122 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2123 { 2124 /* refcount is synchronized on the source endpoint's row lock */ 2125 struct pf_udpendpointhash *h0, *h1; 2126 2127 if (mapping == NULL) 2128 return; 2129 2130 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2131 PF_HASHROW_LOCK(h0); 2132 if (refcount_release(&mapping->refs)) { 2133 LIST_REMOVE(&mapping->endpoints[0], entry); 2134 PF_HASHROW_UNLOCK(h0); 2135 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2136 PF_HASHROW_LOCK(h1); 2137 LIST_REMOVE(&mapping->endpoints[1], entry); 2138 PF_HASHROW_UNLOCK(h1); 2139 2140 uma_zfree(V_pf_udp_mapping_z, mapping); 2141 } else { 2142 PF_HASHROW_UNLOCK(h0); 2143 } 2144 } 2145 2146 2147 struct pf_udp_mapping * 2148 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2149 { 2150 struct pf_udpendpointhash *uh; 2151 struct pf_udp_endpoint *endpoint; 2152 2153 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2154 2155 PF_HASHROW_LOCK(uh); 2156 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2157 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2158 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2159 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2160 break; 2161 } 2162 if (endpoint == NULL) { 2163 PF_HASHROW_UNLOCK(uh); 2164 return (NULL); 2165 } 2166 refcount_acquire(&endpoint->mapping->refs); 2167 PF_HASHROW_UNLOCK(uh); 2168 return (endpoint->mapping); 2169 } 2170 /* END state table stuff */ 2171 2172 static void 2173 pf_send(struct pf_send_entry *pfse) 2174 { 2175 2176 PF_SENDQ_LOCK(); 2177 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2178 PF_SENDQ_UNLOCK(); 2179 swi_sched(V_pf_swi_cookie, 0); 2180 } 2181 2182 static bool 2183 pf_isforlocal(struct mbuf *m, int af) 2184 { 2185 switch (af) { 2186 #ifdef INET 2187 case AF_INET: { 2188 struct ip *ip = mtod(m, struct ip *); 2189 2190 return (in_localip(ip->ip_dst)); 2191 } 2192 #endif /* INET */ 2193 #ifdef INET6 2194 case AF_INET6: { 2195 struct ip6_hdr *ip6; 2196 struct in6_ifaddr *ia; 2197 ip6 = mtod(m, struct ip6_hdr *); 2198 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2199 if (ia == NULL) 2200 return (false); 2201 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2202 } 2203 #endif /* INET6 */ 2204 default: 2205 unhandled_af(af); 2206 } 2207 2208 return (false); 2209 } 2210 2211 int 2212 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2213 int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type) 2214 { 2215 /* 2216 * ICMP types marked with PF_OUT are typically responses to 2217 * PF_IN, and will match states in the opposite direction. 2218 * PF_IN ICMP types need to match a state with that type. 2219 */ 2220 *icmp_dir = PF_OUT; 2221 2222 /* Queries (and responses) */ 2223 switch (pd->af) { 2224 #ifdef INET 2225 case AF_INET: 2226 switch (type) { 2227 case ICMP_ECHO: 2228 *icmp_dir = PF_IN; 2229 /* FALLTHROUGH */ 2230 case ICMP_ECHOREPLY: 2231 *virtual_type = ICMP_ECHO; 2232 *virtual_id = pd->hdr.icmp.icmp_id; 2233 break; 2234 2235 case ICMP_TSTAMP: 2236 *icmp_dir = PF_IN; 2237 /* FALLTHROUGH */ 2238 case ICMP_TSTAMPREPLY: 2239 *virtual_type = ICMP_TSTAMP; 2240 *virtual_id = pd->hdr.icmp.icmp_id; 2241 break; 2242 2243 case ICMP_IREQ: 2244 *icmp_dir = PF_IN; 2245 /* FALLTHROUGH */ 2246 case ICMP_IREQREPLY: 2247 *virtual_type = ICMP_IREQ; 2248 *virtual_id = pd->hdr.icmp.icmp_id; 2249 break; 2250 2251 case ICMP_MASKREQ: 2252 *icmp_dir = PF_IN; 2253 /* FALLTHROUGH */ 2254 case ICMP_MASKREPLY: 2255 *virtual_type = ICMP_MASKREQ; 2256 *virtual_id = pd->hdr.icmp.icmp_id; 2257 break; 2258 2259 case ICMP_IPV6_WHEREAREYOU: 2260 *icmp_dir = PF_IN; 2261 /* FALLTHROUGH */ 2262 case ICMP_IPV6_IAMHERE: 2263 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2264 *virtual_id = 0; /* Nothing sane to match on! */ 2265 break; 2266 2267 case ICMP_MOBILE_REGREQUEST: 2268 *icmp_dir = PF_IN; 2269 /* FALLTHROUGH */ 2270 case ICMP_MOBILE_REGREPLY: 2271 *virtual_type = ICMP_MOBILE_REGREQUEST; 2272 *virtual_id = 0; /* Nothing sane to match on! */ 2273 break; 2274 2275 case ICMP_ROUTERSOLICIT: 2276 *icmp_dir = PF_IN; 2277 /* FALLTHROUGH */ 2278 case ICMP_ROUTERADVERT: 2279 *virtual_type = ICMP_ROUTERSOLICIT; 2280 *virtual_id = 0; /* Nothing sane to match on! */ 2281 break; 2282 2283 /* These ICMP types map to other connections */ 2284 case ICMP_UNREACH: 2285 case ICMP_SOURCEQUENCH: 2286 case ICMP_REDIRECT: 2287 case ICMP_TIMXCEED: 2288 case ICMP_PARAMPROB: 2289 /* These will not be used, but set them anyway */ 2290 *icmp_dir = PF_IN; 2291 *virtual_type = type; 2292 *virtual_id = 0; 2293 *virtual_type = htons(*virtual_type); 2294 return (1); /* These types match to another state */ 2295 2296 /* 2297 * All remaining ICMP types get their own states, 2298 * and will only match in one direction. 2299 */ 2300 default: 2301 *icmp_dir = PF_IN; 2302 *virtual_type = type; 2303 *virtual_id = 0; 2304 break; 2305 } 2306 break; 2307 #endif /* INET */ 2308 #ifdef INET6 2309 case AF_INET6: 2310 switch (type) { 2311 case ICMP6_ECHO_REQUEST: 2312 *icmp_dir = PF_IN; 2313 /* FALLTHROUGH */ 2314 case ICMP6_ECHO_REPLY: 2315 *virtual_type = ICMP6_ECHO_REQUEST; 2316 *virtual_id = pd->hdr.icmp6.icmp6_id; 2317 break; 2318 2319 case MLD_LISTENER_QUERY: 2320 case MLD_LISTENER_REPORT: { 2321 /* 2322 * Listener Report can be sent by clients 2323 * without an associated Listener Query. 2324 * In addition to that, when Report is sent as a 2325 * reply to a Query its source and destination 2326 * address are different. 2327 */ 2328 *icmp_dir = PF_IN; 2329 *virtual_type = MLD_LISTENER_QUERY; 2330 *virtual_id = 0; 2331 break; 2332 } 2333 case MLD_MTRACE: 2334 *icmp_dir = PF_IN; 2335 /* FALLTHROUGH */ 2336 case MLD_MTRACE_RESP: 2337 *virtual_type = MLD_MTRACE; 2338 *virtual_id = 0; /* Nothing sane to match on! */ 2339 break; 2340 2341 case ND_NEIGHBOR_SOLICIT: 2342 *icmp_dir = PF_IN; 2343 /* FALLTHROUGH */ 2344 case ND_NEIGHBOR_ADVERT: { 2345 *virtual_type = ND_NEIGHBOR_SOLICIT; 2346 *virtual_id = 0; 2347 break; 2348 } 2349 2350 /* 2351 * These ICMP types map to other connections. 2352 * ND_REDIRECT can't be in this list because the triggering 2353 * packet header is optional. 2354 */ 2355 case ICMP6_DST_UNREACH: 2356 case ICMP6_PACKET_TOO_BIG: 2357 case ICMP6_TIME_EXCEEDED: 2358 case ICMP6_PARAM_PROB: 2359 /* These will not be used, but set them anyway */ 2360 *icmp_dir = PF_IN; 2361 *virtual_type = type; 2362 *virtual_id = 0; 2363 *virtual_type = htons(*virtual_type); 2364 return (1); /* These types match to another state */ 2365 /* 2366 * All remaining ICMP6 types get their own states, 2367 * and will only match in one direction. 2368 */ 2369 default: 2370 *icmp_dir = PF_IN; 2371 *virtual_type = type; 2372 *virtual_id = 0; 2373 break; 2374 } 2375 break; 2376 #endif /* INET6 */ 2377 default: 2378 unhandled_af(pd->af); 2379 } 2380 *virtual_type = htons(*virtual_type); 2381 return (0); /* These types match to their own state */ 2382 } 2383 2384 void 2385 pf_intr(void *v) 2386 { 2387 struct epoch_tracker et; 2388 struct pf_send_head queue; 2389 struct pf_send_entry *pfse, *next; 2390 2391 CURVNET_SET((struct vnet *)v); 2392 2393 PF_SENDQ_LOCK(); 2394 queue = V_pf_sendqueue; 2395 STAILQ_INIT(&V_pf_sendqueue); 2396 PF_SENDQ_UNLOCK(); 2397 2398 NET_EPOCH_ENTER(et); 2399 2400 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2401 switch (pfse->pfse_type) { 2402 #ifdef INET 2403 case PFSE_IP: { 2404 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2405 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2406 ("%s: rcvif != loif", __func__)); 2407 2408 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2409 pfse->pfse_m->m_pkthdr.csum_flags |= 2410 CSUM_IP_VALID | CSUM_IP_CHECKED | 2411 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2412 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2413 ip_input(pfse->pfse_m); 2414 } else { 2415 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2416 NULL); 2417 } 2418 break; 2419 } 2420 case PFSE_ICMP: 2421 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2422 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2423 break; 2424 #endif /* INET */ 2425 #ifdef INET6 2426 case PFSE_IP6: 2427 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2428 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2429 ("%s: rcvif != loif", __func__)); 2430 2431 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2432 M_LOOP; 2433 pfse->pfse_m->m_pkthdr.csum_flags |= 2434 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2435 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2436 ip6_input(pfse->pfse_m); 2437 } else { 2438 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2439 NULL, NULL); 2440 } 2441 break; 2442 case PFSE_ICMP6: 2443 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2444 pfse->icmpopts.code, pfse->icmpopts.mtu); 2445 break; 2446 #endif /* INET6 */ 2447 default: 2448 panic("%s: unknown type", __func__); 2449 } 2450 free(pfse, M_PFTEMP); 2451 } 2452 NET_EPOCH_EXIT(et); 2453 CURVNET_RESTORE(); 2454 } 2455 2456 #define pf_purge_thread_period (hz / 10) 2457 2458 #ifdef PF_WANT_32_TO_64_COUNTER 2459 static void 2460 pf_status_counter_u64_periodic(void) 2461 { 2462 2463 PF_RULES_RASSERT(); 2464 2465 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2466 return; 2467 } 2468 2469 for (int i = 0; i < FCNT_MAX; i++) { 2470 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2471 } 2472 } 2473 2474 static void 2475 pf_kif_counter_u64_periodic(void) 2476 { 2477 struct pfi_kkif *kif; 2478 size_t r, run; 2479 2480 PF_RULES_RASSERT(); 2481 2482 if (__predict_false(V_pf_allkifcount == 0)) { 2483 return; 2484 } 2485 2486 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2487 return; 2488 } 2489 2490 run = V_pf_allkifcount / 10; 2491 if (run < 5) 2492 run = 5; 2493 2494 for (r = 0; r < run; r++) { 2495 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2496 if (kif == NULL) { 2497 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2498 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2499 break; 2500 } 2501 2502 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2503 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2504 2505 for (int i = 0; i < 2; i++) { 2506 for (int j = 0; j < 2; j++) { 2507 for (int k = 0; k < 2; k++) { 2508 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2509 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2510 } 2511 } 2512 } 2513 } 2514 } 2515 2516 static void 2517 pf_rule_counter_u64_periodic(void) 2518 { 2519 struct pf_krule *rule; 2520 size_t r, run; 2521 2522 PF_RULES_RASSERT(); 2523 2524 if (__predict_false(V_pf_allrulecount == 0)) { 2525 return; 2526 } 2527 2528 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2529 return; 2530 } 2531 2532 run = V_pf_allrulecount / 10; 2533 if (run < 5) 2534 run = 5; 2535 2536 for (r = 0; r < run; r++) { 2537 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2538 if (rule == NULL) { 2539 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2540 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2541 break; 2542 } 2543 2544 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2545 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2546 2547 pf_counter_u64_periodic(&rule->evaluations); 2548 for (int i = 0; i < 2; i++) { 2549 pf_counter_u64_periodic(&rule->packets[i]); 2550 pf_counter_u64_periodic(&rule->bytes[i]); 2551 } 2552 } 2553 } 2554 2555 static void 2556 pf_counter_u64_periodic_main(void) 2557 { 2558 PF_RULES_RLOCK_TRACKER; 2559 2560 V_pf_counter_periodic_iter++; 2561 2562 PF_RULES_RLOCK(); 2563 pf_counter_u64_critical_enter(); 2564 pf_status_counter_u64_periodic(); 2565 pf_kif_counter_u64_periodic(); 2566 pf_rule_counter_u64_periodic(); 2567 pf_counter_u64_critical_exit(); 2568 PF_RULES_RUNLOCK(); 2569 } 2570 #else 2571 #define pf_counter_u64_periodic_main() do { } while (0) 2572 #endif 2573 2574 void 2575 pf_purge_thread(void *unused __unused) 2576 { 2577 struct epoch_tracker et; 2578 2579 VNET_ITERATOR_DECL(vnet_iter); 2580 2581 sx_xlock(&pf_end_lock); 2582 while (pf_end_threads == 0) { 2583 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2584 2585 VNET_LIST_RLOCK(); 2586 NET_EPOCH_ENTER(et); 2587 VNET_FOREACH(vnet_iter) { 2588 CURVNET_SET(vnet_iter); 2589 2590 /* Wait until V_pf_default_rule is initialized. */ 2591 if (V_pf_vnet_active == 0) { 2592 CURVNET_RESTORE(); 2593 continue; 2594 } 2595 2596 pf_counter_u64_periodic_main(); 2597 2598 /* 2599 * Process 1/interval fraction of the state 2600 * table every run. 2601 */ 2602 V_pf_purge_idx = 2603 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2604 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2605 2606 /* 2607 * Purge other expired types every 2608 * PFTM_INTERVAL seconds. 2609 */ 2610 if (V_pf_purge_idx == 0) { 2611 /* 2612 * Order is important: 2613 * - states and src nodes reference rules 2614 * - states and rules reference kifs 2615 */ 2616 pf_purge_expired_fragments(); 2617 pf_purge_expired_src_nodes(); 2618 pf_purge_unlinked_rules(); 2619 pfi_kkif_purge(); 2620 } 2621 CURVNET_RESTORE(); 2622 } 2623 NET_EPOCH_EXIT(et); 2624 VNET_LIST_RUNLOCK(); 2625 } 2626 2627 pf_end_threads++; 2628 sx_xunlock(&pf_end_lock); 2629 kproc_exit(0); 2630 } 2631 2632 void 2633 pf_unload_vnet_purge(void) 2634 { 2635 2636 /* 2637 * To cleanse up all kifs and rules we need 2638 * two runs: first one clears reference flags, 2639 * then pf_purge_expired_states() doesn't 2640 * raise them, and then second run frees. 2641 */ 2642 pf_purge_unlinked_rules(); 2643 pfi_kkif_purge(); 2644 2645 /* 2646 * Now purge everything. 2647 */ 2648 pf_purge_expired_states(0, V_pf_hashmask); 2649 pf_purge_fragments(UINT_MAX); 2650 pf_purge_expired_src_nodes(); 2651 2652 /* 2653 * Now all kifs & rules should be unreferenced, 2654 * thus should be successfully freed. 2655 */ 2656 pf_purge_unlinked_rules(); 2657 pfi_kkif_purge(); 2658 } 2659 2660 u_int32_t 2661 pf_state_expires(const struct pf_kstate *state) 2662 { 2663 u_int32_t timeout; 2664 u_int32_t start; 2665 u_int32_t end; 2666 u_int32_t states; 2667 2668 /* handle all PFTM_* > PFTM_MAX here */ 2669 if (state->timeout == PFTM_PURGE) 2670 return (time_uptime); 2671 KASSERT(state->timeout != PFTM_UNLINKED, 2672 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2673 KASSERT((state->timeout < PFTM_MAX), 2674 ("pf_state_expires: timeout > PFTM_MAX")); 2675 timeout = state->rule->timeout[state->timeout]; 2676 if (!timeout) 2677 timeout = V_pf_default_rule.timeout[state->timeout]; 2678 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2679 if (start && state->rule != &V_pf_default_rule) { 2680 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2681 states = counter_u64_fetch(state->rule->states_cur); 2682 } else { 2683 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2684 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2685 states = V_pf_status.states; 2686 } 2687 if (end && states > start && start < end) { 2688 if (states < end) { 2689 timeout = (u_int64_t)timeout * (end - states) / 2690 (end - start); 2691 return ((state->expire / 1000) + timeout); 2692 } 2693 else 2694 return (time_uptime); 2695 } 2696 return ((state->expire / 1000) + timeout); 2697 } 2698 2699 void 2700 pf_purge_expired_src_nodes(void) 2701 { 2702 struct pf_ksrc_node_list freelist; 2703 struct pf_srchash *sh; 2704 struct pf_ksrc_node *cur, *next; 2705 int i; 2706 2707 LIST_INIT(&freelist); 2708 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2709 PF_HASHROW_LOCK(sh); 2710 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2711 if (cur->states == 0 && cur->expire <= time_uptime) { 2712 pf_unlink_src_node(cur); 2713 LIST_INSERT_HEAD(&freelist, cur, entry); 2714 } else if (cur->rule != NULL) 2715 cur->rule->rule_ref |= PFRULE_REFS; 2716 PF_HASHROW_UNLOCK(sh); 2717 } 2718 2719 pf_free_src_nodes(&freelist); 2720 2721 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2722 } 2723 2724 static void 2725 pf_src_tree_remove_state(struct pf_kstate *s) 2726 { 2727 uint32_t timeout; 2728 2729 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2730 s->rule->timeout[PFTM_SRC_NODE] : 2731 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2732 2733 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 2734 if (s->sns[sn_type] == NULL) 2735 continue; 2736 PF_SRC_NODE_LOCK(s->sns[sn_type]); 2737 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 2738 --(s->sns[sn_type]->conn); 2739 if (--(s->sns[sn_type]->states) == 0) 2740 s->sns[sn_type]->expire = time_uptime + timeout; 2741 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 2742 s->sns[sn_type] = NULL; 2743 } 2744 2745 } 2746 2747 /* 2748 * Unlink and potentilly free a state. Function may be 2749 * called with ID hash row locked, but always returns 2750 * unlocked, since it needs to go through key hash locking. 2751 */ 2752 int 2753 pf_remove_state(struct pf_kstate *s) 2754 { 2755 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2756 2757 NET_EPOCH_ASSERT(); 2758 PF_HASHROW_ASSERT(ih); 2759 2760 if (s->timeout == PFTM_UNLINKED) { 2761 /* 2762 * State is being processed 2763 * by pf_remove_state() in 2764 * an other thread. 2765 */ 2766 PF_HASHROW_UNLOCK(ih); 2767 return (0); /* XXXGL: undefined actually */ 2768 } 2769 2770 if (s->src.state == PF_TCPS_PROXY_DST) { 2771 /* XXX wire key the right one? */ 2772 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2773 &s->key[PF_SK_WIRE]->addr[1], 2774 &s->key[PF_SK_WIRE]->addr[0], 2775 s->key[PF_SK_WIRE]->port[1], 2776 s->key[PF_SK_WIRE]->port[0], 2777 s->src.seqhi, s->src.seqlo + 1, 2778 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2779 s->act.rtableid); 2780 } 2781 2782 LIST_REMOVE(s, entry); 2783 pf_src_tree_remove_state(s); 2784 2785 if (V_pfsync_delete_state_ptr != NULL) 2786 V_pfsync_delete_state_ptr(s); 2787 2788 STATE_DEC_COUNTERS(s); 2789 2790 s->timeout = PFTM_UNLINKED; 2791 2792 /* Ensure we remove it from the list of halfopen states, if needed. */ 2793 if (s->key[PF_SK_STACK] != NULL && 2794 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2795 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2796 2797 PF_HASHROW_UNLOCK(ih); 2798 2799 pf_detach_state(s); 2800 2801 pf_udp_mapping_release(s->udp_mapping); 2802 2803 /* pf_state_insert() initialises refs to 2 */ 2804 return (pf_release_staten(s, 2)); 2805 } 2806 2807 struct pf_kstate * 2808 pf_alloc_state(int flags) 2809 { 2810 2811 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2812 } 2813 2814 void 2815 pf_free_state(struct pf_kstate *cur) 2816 { 2817 struct pf_krule_item *ri; 2818 2819 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2820 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2821 cur->timeout)); 2822 2823 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2824 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2825 free(ri, M_PF_RULE_ITEM); 2826 } 2827 2828 pf_normalize_tcp_cleanup(cur); 2829 uma_zfree(V_pf_state_z, cur); 2830 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2831 } 2832 2833 /* 2834 * Called only from pf_purge_thread(), thus serialized. 2835 */ 2836 static u_int 2837 pf_purge_expired_states(u_int i, int maxcheck) 2838 { 2839 struct pf_idhash *ih; 2840 struct pf_kstate *s; 2841 struct pf_krule_item *mrm; 2842 size_t count __unused; 2843 2844 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2845 2846 /* 2847 * Go through hash and unlink states that expire now. 2848 */ 2849 while (maxcheck > 0) { 2850 count = 0; 2851 ih = &V_pf_idhash[i]; 2852 2853 /* only take the lock if we expect to do work */ 2854 if (!LIST_EMPTY(&ih->states)) { 2855 relock: 2856 PF_HASHROW_LOCK(ih); 2857 LIST_FOREACH(s, &ih->states, entry) { 2858 if (pf_state_expires(s) <= time_uptime) { 2859 V_pf_status.states -= 2860 pf_remove_state(s); 2861 goto relock; 2862 } 2863 s->rule->rule_ref |= PFRULE_REFS; 2864 if (s->nat_rule != NULL) 2865 s->nat_rule->rule_ref |= PFRULE_REFS; 2866 if (s->anchor != NULL) 2867 s->anchor->rule_ref |= PFRULE_REFS; 2868 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2869 SLIST_FOREACH(mrm, &s->match_rules, entry) 2870 mrm->r->rule_ref |= PFRULE_REFS; 2871 if (s->act.rt_kif) 2872 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2873 count++; 2874 } 2875 PF_HASHROW_UNLOCK(ih); 2876 } 2877 2878 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2879 2880 /* Return when we hit end of hash. */ 2881 if (++i > V_pf_hashmask) { 2882 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2883 return (0); 2884 } 2885 2886 maxcheck--; 2887 } 2888 2889 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2890 2891 return (i); 2892 } 2893 2894 static void 2895 pf_purge_unlinked_rules(void) 2896 { 2897 struct pf_krulequeue tmpq; 2898 struct pf_krule *r, *r1; 2899 2900 /* 2901 * If we have overloading task pending, then we'd 2902 * better skip purging this time. There is a tiny 2903 * probability that overloading task references 2904 * an already unlinked rule. 2905 */ 2906 PF_OVERLOADQ_LOCK(); 2907 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2908 PF_OVERLOADQ_UNLOCK(); 2909 return; 2910 } 2911 PF_OVERLOADQ_UNLOCK(); 2912 2913 /* 2914 * Do naive mark-and-sweep garbage collecting of old rules. 2915 * Reference flag is raised by pf_purge_expired_states() 2916 * and pf_purge_expired_src_nodes(). 2917 * 2918 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2919 * use a temporary queue. 2920 */ 2921 TAILQ_INIT(&tmpq); 2922 PF_UNLNKDRULES_LOCK(); 2923 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2924 if (!(r->rule_ref & PFRULE_REFS)) { 2925 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2926 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2927 } else 2928 r->rule_ref &= ~PFRULE_REFS; 2929 } 2930 PF_UNLNKDRULES_UNLOCK(); 2931 2932 if (!TAILQ_EMPTY(&tmpq)) { 2933 PF_CONFIG_LOCK(); 2934 PF_RULES_WLOCK(); 2935 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2936 TAILQ_REMOVE(&tmpq, r, entries); 2937 pf_free_rule(r); 2938 } 2939 PF_RULES_WUNLOCK(); 2940 PF_CONFIG_UNLOCK(); 2941 } 2942 } 2943 2944 void 2945 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2946 { 2947 switch (af) { 2948 #ifdef INET 2949 case AF_INET: { 2950 u_int32_t a = ntohl(addr->addr32[0]); 2951 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2952 (a>>8)&255, a&255); 2953 if (p) { 2954 p = ntohs(p); 2955 printf(":%u", p); 2956 } 2957 break; 2958 } 2959 #endif /* INET */ 2960 #ifdef INET6 2961 case AF_INET6: { 2962 u_int16_t b; 2963 u_int8_t i, curstart, curend, maxstart, maxend; 2964 curstart = curend = maxstart = maxend = 255; 2965 for (i = 0; i < 8; i++) { 2966 if (!addr->addr16[i]) { 2967 if (curstart == 255) 2968 curstart = i; 2969 curend = i; 2970 } else { 2971 if ((curend - curstart) > 2972 (maxend - maxstart)) { 2973 maxstart = curstart; 2974 maxend = curend; 2975 } 2976 curstart = curend = 255; 2977 } 2978 } 2979 if ((curend - curstart) > 2980 (maxend - maxstart)) { 2981 maxstart = curstart; 2982 maxend = curend; 2983 } 2984 for (i = 0; i < 8; i++) { 2985 if (i >= maxstart && i <= maxend) { 2986 if (i == 0) 2987 printf(":"); 2988 if (i == maxend) 2989 printf(":"); 2990 } else { 2991 b = ntohs(addr->addr16[i]); 2992 printf("%x", b); 2993 if (i < 7) 2994 printf(":"); 2995 } 2996 } 2997 if (p) { 2998 p = ntohs(p); 2999 printf("[%u]", p); 3000 } 3001 break; 3002 } 3003 #endif /* INET6 */ 3004 default: 3005 unhandled_af(af); 3006 } 3007 } 3008 3009 void 3010 pf_print_state(struct pf_kstate *s) 3011 { 3012 pf_print_state_parts(s, NULL, NULL); 3013 } 3014 3015 static void 3016 pf_print_state_parts(struct pf_kstate *s, 3017 struct pf_state_key *skwp, struct pf_state_key *sksp) 3018 { 3019 struct pf_state_key *skw, *sks; 3020 u_int8_t proto, dir; 3021 3022 /* Do our best to fill these, but they're skipped if NULL */ 3023 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 3024 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 3025 proto = skw ? skw->proto : (sks ? sks->proto : 0); 3026 dir = s ? s->direction : 0; 3027 3028 switch (proto) { 3029 case IPPROTO_IPV4: 3030 printf("IPv4"); 3031 break; 3032 case IPPROTO_IPV6: 3033 printf("IPv6"); 3034 break; 3035 case IPPROTO_TCP: 3036 printf("TCP"); 3037 break; 3038 case IPPROTO_UDP: 3039 printf("UDP"); 3040 break; 3041 case IPPROTO_ICMP: 3042 printf("ICMP"); 3043 break; 3044 case IPPROTO_ICMPV6: 3045 printf("ICMPv6"); 3046 break; 3047 default: 3048 printf("%u", proto); 3049 break; 3050 } 3051 switch (dir) { 3052 case PF_IN: 3053 printf(" in"); 3054 break; 3055 case PF_OUT: 3056 printf(" out"); 3057 break; 3058 } 3059 if (skw) { 3060 printf(" wire: "); 3061 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3062 printf(" "); 3063 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3064 } 3065 if (sks) { 3066 printf(" stack: "); 3067 if (sks != skw) { 3068 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3069 printf(" "); 3070 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3071 } else 3072 printf("-"); 3073 } 3074 if (s) { 3075 if (proto == IPPROTO_TCP) { 3076 printf(" [lo=%u high=%u win=%u modulator=%u", 3077 s->src.seqlo, s->src.seqhi, 3078 s->src.max_win, s->src.seqdiff); 3079 if (s->src.wscale && s->dst.wscale) 3080 printf(" wscale=%u", 3081 s->src.wscale & PF_WSCALE_MASK); 3082 printf("]"); 3083 printf(" [lo=%u high=%u win=%u modulator=%u", 3084 s->dst.seqlo, s->dst.seqhi, 3085 s->dst.max_win, s->dst.seqdiff); 3086 if (s->src.wscale && s->dst.wscale) 3087 printf(" wscale=%u", 3088 s->dst.wscale & PF_WSCALE_MASK); 3089 printf("]"); 3090 } 3091 printf(" %u:%u", s->src.state, s->dst.state); 3092 if (s->rule) 3093 printf(" @%d", s->rule->nr); 3094 } 3095 } 3096 3097 void 3098 pf_print_flags(uint16_t f) 3099 { 3100 if (f) 3101 printf(" "); 3102 if (f & TH_FIN) 3103 printf("F"); 3104 if (f & TH_SYN) 3105 printf("S"); 3106 if (f & TH_RST) 3107 printf("R"); 3108 if (f & TH_PUSH) 3109 printf("P"); 3110 if (f & TH_ACK) 3111 printf("A"); 3112 if (f & TH_URG) 3113 printf("U"); 3114 if (f & TH_ECE) 3115 printf("E"); 3116 if (f & TH_CWR) 3117 printf("W"); 3118 if (f & TH_AE) 3119 printf("e"); 3120 } 3121 3122 #define PF_SET_SKIP_STEPS(i) \ 3123 do { \ 3124 while (head[i] != cur) { \ 3125 head[i]->skip[i] = cur; \ 3126 head[i] = TAILQ_NEXT(head[i], entries); \ 3127 } \ 3128 } while (0) 3129 3130 void 3131 pf_calc_skip_steps(struct pf_krulequeue *rules) 3132 { 3133 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3134 int i; 3135 3136 cur = TAILQ_FIRST(rules); 3137 prev = cur; 3138 for (i = 0; i < PF_SKIP_COUNT; ++i) 3139 head[i] = cur; 3140 while (cur != NULL) { 3141 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3142 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3143 if (cur->direction != prev->direction) 3144 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3145 if (cur->af != prev->af) 3146 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3147 if (cur->proto != prev->proto) 3148 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3149 if (cur->src.neg != prev->src.neg || 3150 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3151 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3152 if (cur->dst.neg != prev->dst.neg || 3153 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3154 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3155 if (cur->src.port[0] != prev->src.port[0] || 3156 cur->src.port[1] != prev->src.port[1] || 3157 cur->src.port_op != prev->src.port_op) 3158 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3159 if (cur->dst.port[0] != prev->dst.port[0] || 3160 cur->dst.port[1] != prev->dst.port[1] || 3161 cur->dst.port_op != prev->dst.port_op) 3162 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3163 3164 prev = cur; 3165 cur = TAILQ_NEXT(cur, entries); 3166 } 3167 for (i = 0; i < PF_SKIP_COUNT; ++i) 3168 PF_SET_SKIP_STEPS(i); 3169 } 3170 3171 int 3172 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3173 { 3174 if (aw1->type != aw2->type) 3175 return (1); 3176 switch (aw1->type) { 3177 case PF_ADDR_ADDRMASK: 3178 case PF_ADDR_RANGE: 3179 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3180 return (1); 3181 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3182 return (1); 3183 return (0); 3184 case PF_ADDR_DYNIFTL: 3185 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3186 case PF_ADDR_NONE: 3187 case PF_ADDR_NOROUTE: 3188 case PF_ADDR_URPFFAILED: 3189 return (0); 3190 case PF_ADDR_TABLE: 3191 return (aw1->p.tbl != aw2->p.tbl); 3192 default: 3193 printf("invalid address type: %d\n", aw1->type); 3194 return (1); 3195 } 3196 } 3197 3198 /** 3199 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3200 * header isn't always a full checksum. In some cases (i.e. output) it's a 3201 * pseudo-header checksum, which is a partial checksum over src/dst IP 3202 * addresses, protocol number and length. 3203 * 3204 * That means we have the following cases: 3205 * * Input or forwarding: we don't have TSO, the checksum fields are full 3206 * checksums, we need to update the checksum whenever we change anything. 3207 * * Output (i.e. the checksum is a pseudo-header checksum): 3208 * x The field being updated is src/dst address or affects the length of 3209 * the packet. We need to update the pseudo-header checksum (note that this 3210 * checksum is not ones' complement). 3211 * x Some other field is being modified (e.g. src/dst port numbers): We 3212 * don't have to update anything. 3213 **/ 3214 u_int16_t 3215 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3216 { 3217 u_int32_t x; 3218 3219 x = cksum + old - new; 3220 x = (x + (x >> 16)) & 0xffff; 3221 3222 /* optimise: eliminate a branch when not udp */ 3223 if (udp && cksum == 0x0000) 3224 return cksum; 3225 if (udp && x == 0x0000) 3226 x = 0xffff; 3227 3228 return (u_int16_t)(x); 3229 } 3230 3231 static int 3232 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 3233 { 3234 int rewrite = 0; 3235 3236 if (*f != v) { 3237 uint16_t old = htons(hi ? (*f << 8) : *f); 3238 uint16_t new = htons(hi ? ( v << 8) : v); 3239 3240 *f = v; 3241 3242 if (! (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3243 CSUM_DELAY_DATA_IPV6))) 3244 *pd->pcksum = pf_cksum_fixup(*pd->pcksum, old, new, 3245 pd->proto == IPPROTO_UDP); 3246 3247 rewrite = 1; 3248 } 3249 3250 return (rewrite); 3251 } 3252 3253 int 3254 pf_patch_16(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 3255 { 3256 int rewrite = 0; 3257 u_int8_t *fb = (u_int8_t *)f; 3258 u_int8_t *vb = (u_int8_t *)&v; 3259 3260 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3261 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3262 3263 return (rewrite); 3264 } 3265 3266 int 3267 pf_patch_32(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 3268 { 3269 int rewrite = 0; 3270 u_int8_t *fb = (u_int8_t *)f; 3271 u_int8_t *vb = (u_int8_t *)&v; 3272 3273 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3274 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3275 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3276 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3277 3278 return (rewrite); 3279 } 3280 3281 u_int16_t 3282 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3283 u_int16_t new, u_int8_t udp) 3284 { 3285 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3286 return (cksum); 3287 3288 return (pf_cksum_fixup(cksum, old, new, udp)); 3289 } 3290 3291 static void 3292 pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p, 3293 struct pf_addr *an, u_int16_t pn) 3294 { 3295 struct pf_addr ao; 3296 u_int16_t po; 3297 uint8_t u = pd->virtual_proto == IPPROTO_UDP; 3298 3299 MPASS(pd->pcksum); 3300 if (pd->af == AF_INET) { 3301 MPASS(pd->ip_sum); 3302 } 3303 3304 pf_addrcpy(&ao, a, pd->af); 3305 if (pd->af == pd->naf) 3306 pf_addrcpy(a, an, pd->af); 3307 3308 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3309 *pd->pcksum = ~*pd->pcksum; 3310 3311 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3312 return; 3313 po = *p; 3314 *p = pn; 3315 3316 switch (pd->af) { 3317 #ifdef INET 3318 case AF_INET: 3319 switch (pd->naf) { 3320 case AF_INET: 3321 *pd->ip_sum = pf_cksum_fixup(pf_cksum_fixup(*pd->ip_sum, 3322 ao.addr16[0], an->addr16[0], 0), 3323 ao.addr16[1], an->addr16[1], 0); 3324 *p = pn; 3325 3326 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3327 ao.addr16[0], an->addr16[0], u), 3328 ao.addr16[1], an->addr16[1], u); 3329 3330 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3331 break; 3332 #ifdef INET6 3333 case AF_INET6: 3334 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3335 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3336 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3337 ao.addr16[0], an->addr16[0], u), 3338 ao.addr16[1], an->addr16[1], u), 3339 0, an->addr16[2], u), 3340 0, an->addr16[3], u), 3341 0, an->addr16[4], u), 3342 0, an->addr16[5], u), 3343 0, an->addr16[6], u), 3344 0, an->addr16[7], u), 3345 po, pn, u); 3346 break; 3347 #endif /* INET6 */ 3348 default: 3349 unhandled_af(pd->naf); 3350 } 3351 break; 3352 #endif /* INET */ 3353 #ifdef INET6 3354 case AF_INET6: 3355 switch (pd->naf) { 3356 #ifdef INET 3357 case AF_INET: 3358 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3359 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3360 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3361 ao.addr16[0], an->addr16[0], u), 3362 ao.addr16[1], an->addr16[1], u), 3363 ao.addr16[2], 0, u), 3364 ao.addr16[3], 0, u), 3365 ao.addr16[4], 0, u), 3366 ao.addr16[5], 0, u), 3367 ao.addr16[6], 0, u), 3368 ao.addr16[7], 0, u), 3369 po, pn, u); 3370 break; 3371 #endif /* INET */ 3372 case AF_INET6: 3373 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3374 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3375 pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3376 ao.addr16[0], an->addr16[0], u), 3377 ao.addr16[1], an->addr16[1], u), 3378 ao.addr16[2], an->addr16[2], u), 3379 ao.addr16[3], an->addr16[3], u), 3380 ao.addr16[4], an->addr16[4], u), 3381 ao.addr16[5], an->addr16[5], u), 3382 ao.addr16[6], an->addr16[6], u), 3383 ao.addr16[7], an->addr16[7], u); 3384 3385 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3386 break; 3387 default: 3388 unhandled_af(pd->naf); 3389 } 3390 break; 3391 #endif /* INET6 */ 3392 default: 3393 unhandled_af(pd->af); 3394 } 3395 3396 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3397 CSUM_DELAY_DATA_IPV6)) { 3398 *pd->pcksum = ~*pd->pcksum; 3399 if (! *pd->pcksum) 3400 *pd->pcksum = 0xffff; 3401 } 3402 } 3403 3404 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3405 void 3406 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3407 { 3408 u_int32_t ao; 3409 3410 memcpy(&ao, a, sizeof(ao)); 3411 memcpy(a, &an, sizeof(u_int32_t)); 3412 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3413 ao % 65536, an % 65536, u); 3414 } 3415 3416 void 3417 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3418 { 3419 u_int32_t ao; 3420 3421 memcpy(&ao, a, sizeof(ao)); 3422 memcpy(a, &an, sizeof(u_int32_t)); 3423 3424 *c = pf_proto_cksum_fixup(m, 3425 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3426 ao % 65536, an % 65536, udp); 3427 } 3428 3429 #ifdef INET6 3430 static void 3431 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3432 { 3433 struct pf_addr ao; 3434 3435 pf_addrcpy(&ao, a, AF_INET6); 3436 pf_addrcpy(a, an, AF_INET6); 3437 3438 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3439 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3440 pf_cksum_fixup(pf_cksum_fixup(*c, 3441 ao.addr16[0], an->addr16[0], u), 3442 ao.addr16[1], an->addr16[1], u), 3443 ao.addr16[2], an->addr16[2], u), 3444 ao.addr16[3], an->addr16[3], u), 3445 ao.addr16[4], an->addr16[4], u), 3446 ao.addr16[5], an->addr16[5], u), 3447 ao.addr16[6], an->addr16[6], u), 3448 ao.addr16[7], an->addr16[7], u); 3449 } 3450 #endif /* INET6 */ 3451 3452 static void 3453 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3454 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3455 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3456 { 3457 struct pf_addr oia, ooa; 3458 3459 pf_addrcpy(&oia, ia, af); 3460 if (oa) 3461 pf_addrcpy(&ooa, oa, af); 3462 3463 /* Change inner protocol port, fix inner protocol checksum. */ 3464 if (ip != NULL) { 3465 u_int16_t oip = *ip; 3466 u_int32_t opc; 3467 3468 if (pc != NULL) 3469 opc = *pc; 3470 *ip = np; 3471 if (pc != NULL) 3472 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3473 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3474 if (pc != NULL) 3475 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3476 } 3477 /* Change inner ip address, fix inner ip and icmp checksums. */ 3478 pf_addrcpy(ia, na, af); 3479 switch (af) { 3480 #ifdef INET 3481 case AF_INET: { 3482 u_int32_t oh2c = *h2c; 3483 3484 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3485 oia.addr16[0], ia->addr16[0], 0), 3486 oia.addr16[1], ia->addr16[1], 0); 3487 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3488 oia.addr16[0], ia->addr16[0], 0), 3489 oia.addr16[1], ia->addr16[1], 0); 3490 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3491 break; 3492 } 3493 #endif /* INET */ 3494 #ifdef INET6 3495 case AF_INET6: 3496 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3497 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3498 pf_cksum_fixup(pf_cksum_fixup(*ic, 3499 oia.addr16[0], ia->addr16[0], u), 3500 oia.addr16[1], ia->addr16[1], u), 3501 oia.addr16[2], ia->addr16[2], u), 3502 oia.addr16[3], ia->addr16[3], u), 3503 oia.addr16[4], ia->addr16[4], u), 3504 oia.addr16[5], ia->addr16[5], u), 3505 oia.addr16[6], ia->addr16[6], u), 3506 oia.addr16[7], ia->addr16[7], u); 3507 break; 3508 #endif /* INET6 */ 3509 } 3510 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3511 if (oa) { 3512 pf_addrcpy(oa, na, af); 3513 switch (af) { 3514 #ifdef INET 3515 case AF_INET: 3516 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3517 ooa.addr16[0], oa->addr16[0], 0), 3518 ooa.addr16[1], oa->addr16[1], 0); 3519 break; 3520 #endif /* INET */ 3521 #ifdef INET6 3522 case AF_INET6: 3523 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3524 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3525 pf_cksum_fixup(pf_cksum_fixup(*ic, 3526 ooa.addr16[0], oa->addr16[0], u), 3527 ooa.addr16[1], oa->addr16[1], u), 3528 ooa.addr16[2], oa->addr16[2], u), 3529 ooa.addr16[3], oa->addr16[3], u), 3530 ooa.addr16[4], oa->addr16[4], u), 3531 ooa.addr16[5], oa->addr16[5], u), 3532 ooa.addr16[6], oa->addr16[6], u), 3533 ooa.addr16[7], oa->addr16[7], u); 3534 break; 3535 #endif /* INET6 */ 3536 } 3537 } 3538 } 3539 3540 int 3541 pf_translate_af(struct pf_pdesc *pd) 3542 { 3543 #if defined(INET) && defined(INET6) 3544 struct mbuf *mp; 3545 struct ip *ip4; 3546 struct ip6_hdr *ip6; 3547 struct icmp6_hdr *icmp; 3548 struct m_tag *mtag; 3549 struct pf_fragment_tag *ftag; 3550 int hlen; 3551 3552 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3553 3554 /* trim the old header */ 3555 m_adj(pd->m, pd->off); 3556 3557 /* prepend a new one */ 3558 M_PREPEND(pd->m, hlen, M_NOWAIT); 3559 if (pd->m == NULL) 3560 return (-1); 3561 3562 switch (pd->naf) { 3563 case AF_INET: 3564 ip4 = mtod(pd->m, struct ip *); 3565 bzero(ip4, hlen); 3566 ip4->ip_v = IPVERSION; 3567 ip4->ip_hl = hlen >> 2; 3568 ip4->ip_tos = pd->tos; 3569 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3570 ip_fillid(ip4, V_ip_random_id); 3571 ip4->ip_ttl = pd->ttl; 3572 ip4->ip_p = pd->proto; 3573 ip4->ip_src = pd->nsaddr.v4; 3574 ip4->ip_dst = pd->ndaddr.v4; 3575 pd->src = (struct pf_addr *)&ip4->ip_src; 3576 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3577 pd->off = sizeof(struct ip); 3578 break; 3579 case AF_INET6: 3580 ip6 = mtod(pd->m, struct ip6_hdr *); 3581 bzero(ip6, hlen); 3582 ip6->ip6_vfc = IPV6_VERSION; 3583 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3584 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3585 ip6->ip6_nxt = pd->proto; 3586 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3587 ip6->ip6_hlim = IPV6_DEFHLIM; 3588 else 3589 ip6->ip6_hlim = pd->ttl; 3590 ip6->ip6_src = pd->nsaddr.v6; 3591 ip6->ip6_dst = pd->ndaddr.v6; 3592 pd->src = (struct pf_addr *)&ip6->ip6_src; 3593 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3594 pd->off = sizeof(struct ip6_hdr); 3595 3596 /* 3597 * If we're dealing with a reassembled packet we need to adjust 3598 * the header length from the IPv4 header size to IPv6 header 3599 * size. 3600 */ 3601 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3602 if (mtag) { 3603 ftag = (struct pf_fragment_tag *)(mtag + 1); 3604 ftag->ft_hdrlen = sizeof(*ip6); 3605 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3606 sizeof(struct ip) + sizeof(struct ip6_frag); 3607 } 3608 break; 3609 default: 3610 return (-1); 3611 } 3612 3613 /* recalculate icmp/icmp6 checksums */ 3614 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3615 int off; 3616 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3617 NULL) { 3618 pd->m = NULL; 3619 return (-1); 3620 } 3621 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3622 icmp->icmp6_cksum = 0; 3623 icmp->icmp6_cksum = pd->naf == AF_INET ? 3624 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3625 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3626 ntohs(ip6->ip6_plen)); 3627 } 3628 #endif /* INET && INET6 */ 3629 3630 return (0); 3631 } 3632 3633 int 3634 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3635 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3636 sa_family_t af, sa_family_t naf) 3637 { 3638 #if defined(INET) && defined(INET6) 3639 struct mbuf *n = NULL; 3640 struct ip *ip4; 3641 struct ip6_hdr *ip6; 3642 int hlen, olen, mlen; 3643 3644 if (af == naf || (af != AF_INET && af != AF_INET6) || 3645 (naf != AF_INET && naf != AF_INET6)) 3646 return (-1); 3647 3648 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3649 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3650 return (-1); 3651 3652 /* old header */ 3653 olen = pd2->off - off; 3654 /* new header */ 3655 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3656 3657 /* trim old header */ 3658 m_adj(n, olen); 3659 3660 /* prepend a new one */ 3661 M_PREPEND(n, hlen, M_NOWAIT); 3662 if (n == NULL) 3663 return (-1); 3664 3665 /* translate inner ip/ip6 header */ 3666 switch (naf) { 3667 case AF_INET: 3668 ip4 = mtod(n, struct ip *); 3669 bzero(ip4, sizeof(*ip4)); 3670 ip4->ip_v = IPVERSION; 3671 ip4->ip_hl = sizeof(*ip4) >> 2; 3672 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 3673 ip_fillid(ip4, V_ip_random_id); 3674 ip4->ip_off = htons(IP_DF); 3675 ip4->ip_ttl = pd2->ttl; 3676 if (pd2->proto == IPPROTO_ICMPV6) 3677 ip4->ip_p = IPPROTO_ICMP; 3678 else 3679 ip4->ip_p = pd2->proto; 3680 ip4->ip_src = src->v4; 3681 ip4->ip_dst = dst->v4; 3682 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3683 break; 3684 case AF_INET6: 3685 ip6 = mtod(n, struct ip6_hdr *); 3686 bzero(ip6, sizeof(*ip6)); 3687 ip6->ip6_vfc = IPV6_VERSION; 3688 ip6->ip6_plen = htons(pd2->tot_len - olen); 3689 if (pd2->proto == IPPROTO_ICMP) 3690 ip6->ip6_nxt = IPPROTO_ICMPV6; 3691 else 3692 ip6->ip6_nxt = pd2->proto; 3693 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3694 ip6->ip6_hlim = IPV6_DEFHLIM; 3695 else 3696 ip6->ip6_hlim = pd2->ttl; 3697 ip6->ip6_src = src->v6; 3698 ip6->ip6_dst = dst->v6; 3699 break; 3700 default: 3701 unhandled_af(naf); 3702 } 3703 3704 /* adjust payload offset and total packet length */ 3705 pd2->off += hlen - olen; 3706 pd->tot_len += hlen - olen; 3707 3708 /* merge modified inner packet with the original header */ 3709 mlen = n->m_pkthdr.len; 3710 m_cat(m, n); 3711 m->m_pkthdr.len += mlen; 3712 #endif /* INET && INET6 */ 3713 3714 return (0); 3715 } 3716 3717 #define PTR_IP(field) (offsetof(struct ip, field)) 3718 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3719 3720 int 3721 pf_translate_icmp_af(int af, void *arg) 3722 { 3723 #if defined(INET) && defined(INET6) 3724 struct icmp *icmp4; 3725 struct icmp6_hdr *icmp6; 3726 u_int32_t mtu; 3727 int32_t ptr = -1; 3728 u_int8_t type; 3729 u_int8_t code; 3730 3731 switch (af) { 3732 case AF_INET: 3733 icmp6 = arg; 3734 type = icmp6->icmp6_type; 3735 code = icmp6->icmp6_code; 3736 mtu = ntohl(icmp6->icmp6_mtu); 3737 3738 switch (type) { 3739 case ICMP6_ECHO_REQUEST: 3740 type = ICMP_ECHO; 3741 break; 3742 case ICMP6_ECHO_REPLY: 3743 type = ICMP_ECHOREPLY; 3744 break; 3745 case ICMP6_DST_UNREACH: 3746 type = ICMP_UNREACH; 3747 switch (code) { 3748 case ICMP6_DST_UNREACH_NOROUTE: 3749 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3750 case ICMP6_DST_UNREACH_ADDR: 3751 code = ICMP_UNREACH_HOST; 3752 break; 3753 case ICMP6_DST_UNREACH_ADMIN: 3754 code = ICMP_UNREACH_HOST_PROHIB; 3755 break; 3756 case ICMP6_DST_UNREACH_NOPORT: 3757 code = ICMP_UNREACH_PORT; 3758 break; 3759 default: 3760 return (-1); 3761 } 3762 break; 3763 case ICMP6_PACKET_TOO_BIG: 3764 type = ICMP_UNREACH; 3765 code = ICMP_UNREACH_NEEDFRAG; 3766 mtu -= 20; 3767 break; 3768 case ICMP6_TIME_EXCEEDED: 3769 type = ICMP_TIMXCEED; 3770 break; 3771 case ICMP6_PARAM_PROB: 3772 switch (code) { 3773 case ICMP6_PARAMPROB_HEADER: 3774 type = ICMP_PARAMPROB; 3775 code = ICMP_PARAMPROB_ERRATPTR; 3776 ptr = ntohl(icmp6->icmp6_pptr); 3777 3778 if (ptr == PTR_IP6(ip6_vfc)) 3779 ; /* preserve */ 3780 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3781 ptr = PTR_IP(ip_tos); 3782 else if (ptr == PTR_IP6(ip6_plen) || 3783 ptr == PTR_IP6(ip6_plen) + 1) 3784 ptr = PTR_IP(ip_len); 3785 else if (ptr == PTR_IP6(ip6_nxt)) 3786 ptr = PTR_IP(ip_p); 3787 else if (ptr == PTR_IP6(ip6_hlim)) 3788 ptr = PTR_IP(ip_ttl); 3789 else if (ptr >= PTR_IP6(ip6_src) && 3790 ptr < PTR_IP6(ip6_dst)) 3791 ptr = PTR_IP(ip_src); 3792 else if (ptr >= PTR_IP6(ip6_dst) && 3793 ptr < sizeof(struct ip6_hdr)) 3794 ptr = PTR_IP(ip_dst); 3795 else { 3796 return (-1); 3797 } 3798 break; 3799 case ICMP6_PARAMPROB_NEXTHEADER: 3800 type = ICMP_UNREACH; 3801 code = ICMP_UNREACH_PROTOCOL; 3802 break; 3803 default: 3804 return (-1); 3805 } 3806 break; 3807 default: 3808 return (-1); 3809 } 3810 if (icmp6->icmp6_type != type) { 3811 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3812 icmp6->icmp6_type, type, 0); 3813 icmp6->icmp6_type = type; 3814 } 3815 if (icmp6->icmp6_code != code) { 3816 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3817 icmp6->icmp6_code, code, 0); 3818 icmp6->icmp6_code = code; 3819 } 3820 if (icmp6->icmp6_mtu != htonl(mtu)) { 3821 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3822 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3823 /* aligns well with a icmpv4 nextmtu */ 3824 icmp6->icmp6_mtu = htonl(mtu); 3825 } 3826 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3827 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3828 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3829 /* icmpv4 pptr is a one most significant byte */ 3830 icmp6->icmp6_pptr = htonl(ptr << 24); 3831 } 3832 break; 3833 case AF_INET6: 3834 icmp4 = arg; 3835 type = icmp4->icmp_type; 3836 code = icmp4->icmp_code; 3837 mtu = ntohs(icmp4->icmp_nextmtu); 3838 3839 switch (type) { 3840 case ICMP_ECHO: 3841 type = ICMP6_ECHO_REQUEST; 3842 break; 3843 case ICMP_ECHOREPLY: 3844 type = ICMP6_ECHO_REPLY; 3845 break; 3846 case ICMP_UNREACH: 3847 type = ICMP6_DST_UNREACH; 3848 switch (code) { 3849 case ICMP_UNREACH_NET: 3850 case ICMP_UNREACH_HOST: 3851 case ICMP_UNREACH_NET_UNKNOWN: 3852 case ICMP_UNREACH_HOST_UNKNOWN: 3853 case ICMP_UNREACH_ISOLATED: 3854 case ICMP_UNREACH_TOSNET: 3855 case ICMP_UNREACH_TOSHOST: 3856 code = ICMP6_DST_UNREACH_NOROUTE; 3857 break; 3858 case ICMP_UNREACH_PORT: 3859 code = ICMP6_DST_UNREACH_NOPORT; 3860 break; 3861 case ICMP_UNREACH_NET_PROHIB: 3862 case ICMP_UNREACH_HOST_PROHIB: 3863 case ICMP_UNREACH_FILTER_PROHIB: 3864 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3865 code = ICMP6_DST_UNREACH_ADMIN; 3866 break; 3867 case ICMP_UNREACH_PROTOCOL: 3868 type = ICMP6_PARAM_PROB; 3869 code = ICMP6_PARAMPROB_NEXTHEADER; 3870 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3871 break; 3872 case ICMP_UNREACH_NEEDFRAG: 3873 type = ICMP6_PACKET_TOO_BIG; 3874 code = 0; 3875 mtu += 20; 3876 break; 3877 default: 3878 return (-1); 3879 } 3880 break; 3881 case ICMP_TIMXCEED: 3882 type = ICMP6_TIME_EXCEEDED; 3883 break; 3884 case ICMP_PARAMPROB: 3885 type = ICMP6_PARAM_PROB; 3886 switch (code) { 3887 case ICMP_PARAMPROB_ERRATPTR: 3888 code = ICMP6_PARAMPROB_HEADER; 3889 break; 3890 case ICMP_PARAMPROB_LENGTH: 3891 code = ICMP6_PARAMPROB_HEADER; 3892 break; 3893 default: 3894 return (-1); 3895 } 3896 3897 ptr = icmp4->icmp_pptr; 3898 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3899 ; /* preserve */ 3900 else if (ptr == PTR_IP(ip_len) || 3901 ptr == PTR_IP(ip_len) + 1) 3902 ptr = PTR_IP6(ip6_plen); 3903 else if (ptr == PTR_IP(ip_ttl)) 3904 ptr = PTR_IP6(ip6_hlim); 3905 else if (ptr == PTR_IP(ip_p)) 3906 ptr = PTR_IP6(ip6_nxt); 3907 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 3908 ptr = PTR_IP6(ip6_src); 3909 else if (ptr >= PTR_IP(ip_dst) && 3910 ptr < sizeof(struct ip)) 3911 ptr = PTR_IP6(ip6_dst); 3912 else { 3913 return (-1); 3914 } 3915 break; 3916 default: 3917 return (-1); 3918 } 3919 if (icmp4->icmp_type != type) { 3920 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3921 icmp4->icmp_type, type, 0); 3922 icmp4->icmp_type = type; 3923 } 3924 if (icmp4->icmp_code != code) { 3925 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3926 icmp4->icmp_code, code, 0); 3927 icmp4->icmp_code = code; 3928 } 3929 if (icmp4->icmp_nextmtu != htons(mtu)) { 3930 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3931 icmp4->icmp_nextmtu, htons(mtu), 0); 3932 icmp4->icmp_nextmtu = htons(mtu); 3933 } 3934 if (ptr >= 0 && icmp4->icmp_void != ptr) { 3935 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3936 htons(icmp4->icmp_pptr), htons(ptr), 0); 3937 icmp4->icmp_void = htonl(ptr); 3938 } 3939 break; 3940 default: 3941 unhandled_af(af); 3942 } 3943 #endif /* INET && INET6 */ 3944 3945 return (0); 3946 } 3947 3948 /* 3949 * Need to modulate the sequence numbers in the TCP SACK option 3950 * (credits to Krzysztof Pfaff for report and patch) 3951 */ 3952 static int 3953 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3954 struct pf_state_peer *dst) 3955 { 3956 struct sackblk sack; 3957 int copyback = 0, i; 3958 int olen, optsoff; 3959 uint8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 3960 3961 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 3962 optsoff = pd->off + sizeof(struct tcphdr); 3963 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 3964 if (olen < TCPOLEN_MINSACK || 3965 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af)) 3966 return (0); 3967 3968 eoh = opts + olen; 3969 opt = opts; 3970 while ((opt = pf_find_tcpopt(opt, opts, olen, 3971 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 3972 { 3973 size_t safelen = MIN(opt[1], (eoh - opt)); 3974 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 3975 size_t startoff = (opt + i) - opts; 3976 memcpy(&sack, &opt[i], sizeof(sack)); 3977 pf_patch_32(pd, &sack.start, 3978 htonl(ntohl(sack.start) - dst->seqdiff), 3979 PF_ALGNMNT(startoff)); 3980 pf_patch_32(pd, &sack.end, 3981 htonl(ntohl(sack.end) - dst->seqdiff), 3982 PF_ALGNMNT(startoff + sizeof(sack.start))); 3983 memcpy(&opt[i], &sack, sizeof(sack)); 3984 } 3985 copyback = 1; 3986 opt += opt[1]; 3987 } 3988 3989 if (copyback) 3990 m_copyback(pd->m, optsoff, olen, (caddr_t)opts); 3991 3992 return (copyback); 3993 } 3994 3995 struct mbuf * 3996 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3997 const struct pf_addr *saddr, const struct pf_addr *daddr, 3998 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3999 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4000 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, u_int sack, 4001 int rtableid) 4002 { 4003 struct mbuf *m; 4004 int len, tlen; 4005 #ifdef INET 4006 struct ip *h = NULL; 4007 #endif /* INET */ 4008 #ifdef INET6 4009 struct ip6_hdr *h6 = NULL; 4010 #endif /* INET6 */ 4011 struct tcphdr *th; 4012 char *opt; 4013 struct pf_mtag *pf_mtag; 4014 4015 len = 0; 4016 th = NULL; 4017 4018 /* maximum segment size tcp option */ 4019 tlen = sizeof(struct tcphdr); 4020 if (mss) 4021 tlen += 4; 4022 if (sack) 4023 tlen += 2; 4024 4025 switch (af) { 4026 #ifdef INET 4027 case AF_INET: 4028 len = sizeof(struct ip) + tlen; 4029 break; 4030 #endif /* INET */ 4031 #ifdef INET6 4032 case AF_INET6: 4033 len = sizeof(struct ip6_hdr) + tlen; 4034 break; 4035 #endif /* INET6 */ 4036 default: 4037 unhandled_af(af); 4038 } 4039 4040 m = m_gethdr(M_NOWAIT, MT_DATA); 4041 if (m == NULL) 4042 return (NULL); 4043 4044 #ifdef MAC 4045 mac_netinet_firewall_send(m); 4046 #endif 4047 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 4048 m_freem(m); 4049 return (NULL); 4050 } 4051 m->m_flags |= mbuf_flags; 4052 pf_mtag->tag = mtag_tag; 4053 pf_mtag->flags = mtag_flags; 4054 4055 if (rtableid >= 0) 4056 M_SETFIB(m, rtableid); 4057 4058 #ifdef ALTQ 4059 if (r != NULL && r->qid) { 4060 pf_mtag->qid = r->qid; 4061 4062 /* add hints for ecn */ 4063 pf_mtag->hdr = mtod(m, struct ip *); 4064 } 4065 #endif /* ALTQ */ 4066 m->m_data += max_linkhdr; 4067 m->m_pkthdr.len = m->m_len = len; 4068 /* The rest of the stack assumes a rcvif, so provide one. 4069 * This is a locally generated packet, so .. close enough. */ 4070 m->m_pkthdr.rcvif = V_loif; 4071 bzero(m->m_data, len); 4072 switch (af) { 4073 #ifdef INET 4074 case AF_INET: 4075 m->m_pkthdr.csum_flags |= CSUM_TCP; 4076 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4077 4078 h = mtod(m, struct ip *); 4079 4080 h->ip_p = IPPROTO_TCP; 4081 h->ip_len = htons(tlen); 4082 h->ip_v = 4; 4083 h->ip_hl = sizeof(*h) >> 2; 4084 h->ip_tos = IPTOS_LOWDELAY; 4085 h->ip_len = htons(len); 4086 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4087 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4088 h->ip_sum = 0; 4089 h->ip_src.s_addr = saddr->v4.s_addr; 4090 h->ip_dst.s_addr = daddr->v4.s_addr; 4091 4092 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4093 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4094 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4095 break; 4096 #endif /* INET */ 4097 #ifdef INET6 4098 case AF_INET6: 4099 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4100 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4101 4102 h6 = mtod(m, struct ip6_hdr *); 4103 4104 /* IP header fields included in the TCP checksum */ 4105 h6->ip6_nxt = IPPROTO_TCP; 4106 h6->ip6_plen = htons(tlen); 4107 h6->ip6_vfc |= IPV6_VERSION; 4108 h6->ip6_hlim = V_ip6_defhlim; 4109 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4110 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4111 4112 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4113 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4114 IPPROTO_TCP, 0); 4115 break; 4116 #endif /* INET6 */ 4117 } 4118 4119 /* TCP header */ 4120 th->th_sport = sport; 4121 th->th_dport = dport; 4122 th->th_seq = htonl(seq); 4123 th->th_ack = htonl(ack); 4124 th->th_off = tlen >> 2; 4125 tcp_set_flags(th, tcp_flags); 4126 th->th_win = htons(win); 4127 4128 opt = (char *)(th + 1); 4129 if (mss) { 4130 opt = (char *)(th + 1); 4131 opt[0] = TCPOPT_MAXSEG; 4132 opt[1] = 4; 4133 mss = htons(mss); 4134 memcpy((opt + 2), &mss, 2); 4135 opt += 4; 4136 } 4137 if (sack) { 4138 opt[0] = TCPOPT_SACK_PERMITTED; 4139 opt[1] = 2; 4140 opt += 2; 4141 } 4142 4143 return (m); 4144 } 4145 4146 static void 4147 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4148 uint8_t ttl, int rtableid) 4149 { 4150 struct mbuf *m; 4151 #ifdef INET 4152 struct ip *h = NULL; 4153 #endif /* INET */ 4154 #ifdef INET6 4155 struct ip6_hdr *h6 = NULL; 4156 #endif /* INET6 */ 4157 struct sctphdr *hdr; 4158 struct sctp_chunkhdr *chunk; 4159 struct pf_send_entry *pfse; 4160 int off = 0; 4161 4162 MPASS(af == pd->af); 4163 4164 m = m_gethdr(M_NOWAIT, MT_DATA); 4165 if (m == NULL) 4166 return; 4167 4168 m->m_data += max_linkhdr; 4169 m->m_flags |= M_SKIP_FIREWALL; 4170 /* The rest of the stack assumes a rcvif, so provide one. 4171 * This is a locally generated packet, so .. close enough. */ 4172 m->m_pkthdr.rcvif = V_loif; 4173 4174 /* IPv4|6 header */ 4175 switch (af) { 4176 #ifdef INET 4177 case AF_INET: 4178 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4179 4180 h = mtod(m, struct ip *); 4181 4182 /* IP header fields included in the TCP checksum */ 4183 4184 h->ip_p = IPPROTO_SCTP; 4185 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4186 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4187 h->ip_src = pd->dst->v4; 4188 h->ip_dst = pd->src->v4; 4189 4190 off += sizeof(struct ip); 4191 break; 4192 #endif /* INET */ 4193 #ifdef INET6 4194 case AF_INET6: 4195 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4196 4197 h6 = mtod(m, struct ip6_hdr *); 4198 4199 /* IP header fields included in the TCP checksum */ 4200 h6->ip6_vfc |= IPV6_VERSION; 4201 h6->ip6_nxt = IPPROTO_SCTP; 4202 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4203 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4204 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4205 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4206 4207 off += sizeof(struct ip6_hdr); 4208 break; 4209 #endif /* INET6 */ 4210 default: 4211 unhandled_af(af); 4212 } 4213 4214 /* SCTP header */ 4215 hdr = mtodo(m, off); 4216 4217 hdr->src_port = pd->hdr.sctp.dest_port; 4218 hdr->dest_port = pd->hdr.sctp.src_port; 4219 hdr->v_tag = pd->sctp_initiate_tag; 4220 hdr->checksum = 0; 4221 4222 /* Abort chunk. */ 4223 off += sizeof(struct sctphdr); 4224 chunk = mtodo(m, off); 4225 4226 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4227 chunk->chunk_length = htons(sizeof(*chunk)); 4228 4229 /* SCTP checksum */ 4230 off += sizeof(*chunk); 4231 m->m_pkthdr.len = m->m_len = off; 4232 4233 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4234 4235 if (rtableid >= 0) 4236 M_SETFIB(m, rtableid); 4237 4238 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4239 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4240 if (pfse == NULL) { 4241 m_freem(m); 4242 return; 4243 } 4244 4245 switch (af) { 4246 #ifdef INET 4247 case AF_INET: 4248 pfse->pfse_type = PFSE_IP; 4249 break; 4250 #endif /* INET */ 4251 #ifdef INET6 4252 case AF_INET6: 4253 pfse->pfse_type = PFSE_IP6; 4254 break; 4255 #endif /* INET6 */ 4256 } 4257 4258 pfse->pfse_m = m; 4259 pf_send(pfse); 4260 } 4261 4262 void 4263 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4264 const struct pf_addr *saddr, const struct pf_addr *daddr, 4265 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4266 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4267 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 4268 { 4269 struct pf_send_entry *pfse; 4270 struct mbuf *m; 4271 4272 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4273 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, 0, rtableid); 4274 if (m == NULL) 4275 return; 4276 4277 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4278 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4279 if (pfse == NULL) { 4280 m_freem(m); 4281 return; 4282 } 4283 4284 switch (af) { 4285 #ifdef INET 4286 case AF_INET: 4287 pfse->pfse_type = PFSE_IP; 4288 break; 4289 #endif /* INET */ 4290 #ifdef INET6 4291 case AF_INET6: 4292 pfse->pfse_type = PFSE_IP6; 4293 break; 4294 #endif /* INET6 */ 4295 default: 4296 unhandled_af(af); 4297 } 4298 4299 pfse->pfse_m = m; 4300 pf_send(pfse); 4301 } 4302 4303 static void 4304 pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum) 4305 { 4306 /* undo NAT changes, if they have taken place */ 4307 if (nr != NULL) { 4308 pf_addrcpy(pd->src, &pd->osrc, pd->af); 4309 pf_addrcpy(pd->dst, &pd->odst, pd->af); 4310 if (pd->sport) 4311 *pd->sport = pd->osport; 4312 if (pd->dport) 4313 *pd->dport = pd->odport; 4314 if (pd->ip_sum) 4315 *pd->ip_sum = bip_sum; 4316 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4317 } 4318 } 4319 4320 static void 4321 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4322 struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum, 4323 u_short *reason, int rtableid) 4324 { 4325 pf_undo_nat(nr, pd, bip_sum); 4326 4327 if (pd->proto == IPPROTO_TCP && 4328 ((r->rule_flag & PFRULE_RETURNRST) || 4329 (r->rule_flag & PFRULE_RETURN)) && 4330 !(tcp_get_flags(th) & TH_RST)) { 4331 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4332 4333 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4334 IPPROTO_TCP, pd->af)) 4335 REASON_SET(reason, PFRES_PROTCKSUM); 4336 else { 4337 if (tcp_get_flags(th) & TH_SYN) 4338 ack++; 4339 if (tcp_get_flags(th) & TH_FIN) 4340 ack++; 4341 pf_send_tcp(r, pd->af, pd->dst, 4342 pd->src, th->th_dport, th->th_sport, 4343 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4344 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 4345 } 4346 } else if (pd->proto == IPPROTO_SCTP && 4347 (r->rule_flag & PFRULE_RETURN)) { 4348 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4349 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4350 r->return_icmp) 4351 pf_send_icmp(pd->m, r->return_icmp >> 8, 4352 r->return_icmp & 255, 0, pd->af, r, rtableid); 4353 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4354 r->return_icmp6) 4355 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4356 r->return_icmp6 & 255, 0, pd->af, r, rtableid); 4357 } 4358 4359 static int 4360 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4361 { 4362 struct m_tag *mtag; 4363 u_int8_t mpcp; 4364 4365 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4366 if (mtag == NULL) 4367 return (0); 4368 4369 if (prio == PF_PRIO_ZERO) 4370 prio = 0; 4371 4372 mpcp = *(uint8_t *)(mtag + 1); 4373 4374 return (mpcp == prio); 4375 } 4376 4377 static int 4378 pf_icmp_to_bandlim(uint8_t type) 4379 { 4380 switch (type) { 4381 case ICMP_ECHO: 4382 case ICMP_ECHOREPLY: 4383 return (BANDLIM_ICMP_ECHO); 4384 case ICMP_TSTAMP: 4385 case ICMP_TSTAMPREPLY: 4386 return (BANDLIM_ICMP_TSTAMP); 4387 case ICMP_UNREACH: 4388 default: 4389 return (BANDLIM_ICMP_UNREACH); 4390 } 4391 } 4392 4393 static void 4394 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_kstate *s, 4395 struct pf_state_peer *src, struct pf_state_peer *dst) 4396 { 4397 /* 4398 * We are sending challenge ACK as a response to SYN packet, which 4399 * matches existing state (modulo TCP window check). Therefore packet 4400 * must be sent on behalf of destination. 4401 * 4402 * We expect sender to remain either silent, or send RST packet 4403 * so both, firewall and remote peer, can purge dead state from 4404 * memory. 4405 */ 4406 pf_send_tcp(s->rule, pd->af, pd->dst, pd->src, 4407 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 4408 src->seqlo, TH_ACK, 0, 0, s->rule->return_ttl, 0, 0, 0, 4409 s->rule->rtableid); 4410 } 4411 4412 static void 4413 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int mtu, 4414 sa_family_t af, struct pf_krule *r, int rtableid) 4415 { 4416 struct pf_send_entry *pfse; 4417 struct mbuf *m0; 4418 struct pf_mtag *pf_mtag; 4419 4420 /* ICMP packet rate limitation. */ 4421 switch (af) { 4422 #ifdef INET6 4423 case AF_INET6: 4424 if (icmp6_ratelimit(NULL, type, code)) 4425 return; 4426 break; 4427 #endif /* INET6 */ 4428 #ifdef INET 4429 case AF_INET: 4430 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4431 return; 4432 break; 4433 #endif /* INET */ 4434 } 4435 4436 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4437 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4438 if (pfse == NULL) 4439 return; 4440 4441 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4442 free(pfse, M_PFTEMP); 4443 return; 4444 } 4445 4446 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4447 free(pfse, M_PFTEMP); 4448 return; 4449 } 4450 /* XXX: revisit */ 4451 m0->m_flags |= M_SKIP_FIREWALL; 4452 4453 if (rtableid >= 0) 4454 M_SETFIB(m0, rtableid); 4455 4456 #ifdef ALTQ 4457 if (r->qid) { 4458 pf_mtag->qid = r->qid; 4459 /* add hints for ecn */ 4460 pf_mtag->hdr = mtod(m0, struct ip *); 4461 } 4462 #endif /* ALTQ */ 4463 4464 switch (af) { 4465 #ifdef INET 4466 case AF_INET: 4467 pfse->pfse_type = PFSE_ICMP; 4468 break; 4469 #endif /* INET */ 4470 #ifdef INET6 4471 case AF_INET6: 4472 pfse->pfse_type = PFSE_ICMP6; 4473 break; 4474 #endif /* INET6 */ 4475 } 4476 pfse->pfse_m = m0; 4477 pfse->icmpopts.type = type; 4478 pfse->icmpopts.code = code; 4479 pfse->icmpopts.mtu = mtu; 4480 pf_send(pfse); 4481 } 4482 4483 /* 4484 * Return ((n = 0) == (a = b [with mask m])) 4485 * Note: n != 0 => returns (a != b [with mask m]) 4486 */ 4487 int 4488 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4489 const struct pf_addr *b, sa_family_t af) 4490 { 4491 switch (af) { 4492 #ifdef INET 4493 case AF_INET: 4494 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4495 return (n == 0); 4496 break; 4497 #endif /* INET */ 4498 #ifdef INET6 4499 case AF_INET6: 4500 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4501 return (n == 0); 4502 break; 4503 #endif /* INET6 */ 4504 } 4505 4506 return (n != 0); 4507 } 4508 4509 /* 4510 * Return 1 if b <= a <= e, otherwise return 0. 4511 */ 4512 int 4513 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4514 const struct pf_addr *a, sa_family_t af) 4515 { 4516 switch (af) { 4517 #ifdef INET 4518 case AF_INET: 4519 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4520 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4521 return (0); 4522 break; 4523 #endif /* INET */ 4524 #ifdef INET6 4525 case AF_INET6: { 4526 int i; 4527 4528 /* check a >= b */ 4529 for (i = 0; i < 4; ++i) 4530 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4531 break; 4532 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4533 return (0); 4534 /* check a <= e */ 4535 for (i = 0; i < 4; ++i) 4536 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4537 break; 4538 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4539 return (0); 4540 break; 4541 } 4542 #endif /* INET6 */ 4543 } 4544 return (1); 4545 } 4546 4547 static int 4548 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4549 { 4550 switch (op) { 4551 case PF_OP_IRG: 4552 return ((p > a1) && (p < a2)); 4553 case PF_OP_XRG: 4554 return ((p < a1) || (p > a2)); 4555 case PF_OP_RRG: 4556 return ((p >= a1) && (p <= a2)); 4557 case PF_OP_EQ: 4558 return (p == a1); 4559 case PF_OP_NE: 4560 return (p != a1); 4561 case PF_OP_LT: 4562 return (p < a1); 4563 case PF_OP_LE: 4564 return (p <= a1); 4565 case PF_OP_GT: 4566 return (p > a1); 4567 case PF_OP_GE: 4568 return (p >= a1); 4569 } 4570 return (0); /* never reached */ 4571 } 4572 4573 int 4574 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4575 { 4576 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 4577 } 4578 4579 static int 4580 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4581 { 4582 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4583 return (0); 4584 return (pf_match(op, a1, a2, u)); 4585 } 4586 4587 static int 4588 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4589 { 4590 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4591 return (0); 4592 return (pf_match(op, a1, a2, g)); 4593 } 4594 4595 int 4596 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4597 { 4598 if (*tag == -1) 4599 *tag = mtag; 4600 4601 return ((!r->match_tag_not && r->match_tag == *tag) || 4602 (r->match_tag_not && r->match_tag != *tag)); 4603 } 4604 4605 static int 4606 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4607 { 4608 struct ifnet *ifp = m->m_pkthdr.rcvif; 4609 struct pfi_kkif *kif; 4610 4611 if (ifp == NULL) 4612 return (0); 4613 4614 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4615 4616 if (kif == NULL) { 4617 DPFPRINTF(PF_DEBUG_URGENT, 4618 ("%s: kif == NULL, @%d via %s\n", __func__, r->nr, 4619 r->rcv_ifname)); 4620 return (0); 4621 } 4622 4623 return (pfi_kkif_match(r->rcv_kif, kif)); 4624 } 4625 4626 int 4627 pf_tag_packet(struct pf_pdesc *pd, int tag) 4628 { 4629 4630 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4631 4632 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4633 return (ENOMEM); 4634 4635 pd->pf_mtag->tag = tag; 4636 4637 return (0); 4638 } 4639 4640 /* 4641 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4642 */ 4643 #define PF_ANCHORSTACK_MATCH 0x00000001 4644 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4645 4646 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4647 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4648 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4649 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4650 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4651 } while (0) 4652 4653 enum pf_test_status 4654 pf_step_into_anchor(struct pf_test_ctx *ctx, struct pf_krule *r) 4655 { 4656 enum pf_test_status rv; 4657 4658 PF_RULES_RASSERT(); 4659 4660 if (ctx->depth >= PF_ANCHOR_STACK_MAX) { 4661 printf("%s: anchor stack overflow on %s\n", 4662 __func__, r->anchor->name); 4663 return (PF_TEST_FAIL); 4664 } 4665 4666 ctx->depth++; 4667 4668 if (r->anchor_wildcard) { 4669 struct pf_kanchor *child; 4670 rv = PF_TEST_OK; 4671 RB_FOREACH(child, pf_kanchor_node, &r->anchor->children) { 4672 rv = pf_match_rule(ctx, &child->ruleset); 4673 if ((rv == PF_TEST_QUICK) || (rv == PF_TEST_FAIL)) { 4674 /* 4675 * we either hit a rule with quick action 4676 * (more likely), or hit some runtime 4677 * error (e.g. pool_get() failure). 4678 */ 4679 break; 4680 } 4681 } 4682 } else { 4683 rv = pf_match_rule(ctx, &r->anchor->ruleset); 4684 /* 4685 * Unless errors occured, stop iff any rule matched 4686 * within quick anchors. 4687 */ 4688 if (rv != PF_TEST_FAIL && r->quick == PF_TEST_QUICK && 4689 *ctx->am == r) 4690 rv = PF_TEST_QUICK; 4691 } 4692 4693 ctx->depth--; 4694 4695 return (rv); 4696 } 4697 4698 struct pf_keth_anchor_stackframe { 4699 struct pf_keth_ruleset *rs; 4700 struct pf_keth_rule *r; /* XXX: + match bit */ 4701 struct pf_keth_anchor *child; 4702 }; 4703 4704 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4705 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4706 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4707 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4708 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4709 } while (0) 4710 4711 void 4712 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4713 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4714 struct pf_keth_rule **a, int *match) 4715 { 4716 struct pf_keth_anchor_stackframe *f; 4717 4718 NET_EPOCH_ASSERT(); 4719 4720 if (match) 4721 *match = 0; 4722 if (*depth >= PF_ANCHOR_STACK_MAX) { 4723 printf("%s: anchor stack overflow on %s\n", 4724 __func__, (*r)->anchor->name); 4725 *r = TAILQ_NEXT(*r, entries); 4726 return; 4727 } else if (*depth == 0 && a != NULL) 4728 *a = *r; 4729 f = stack + (*depth)++; 4730 f->rs = *rs; 4731 f->r = *r; 4732 if ((*r)->anchor_wildcard) { 4733 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4734 4735 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4736 *r = NULL; 4737 return; 4738 } 4739 *rs = &f->child->ruleset; 4740 } else { 4741 f->child = NULL; 4742 *rs = &(*r)->anchor->ruleset; 4743 } 4744 *r = TAILQ_FIRST((*rs)->active.rules); 4745 } 4746 4747 int 4748 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4749 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4750 struct pf_keth_rule **a, int *match) 4751 { 4752 struct pf_keth_anchor_stackframe *f; 4753 struct pf_keth_rule *fr; 4754 int quick = 0; 4755 4756 NET_EPOCH_ASSERT(); 4757 4758 do { 4759 if (*depth <= 0) 4760 break; 4761 f = stack + *depth - 1; 4762 fr = PF_ETH_ANCHOR_RULE(f); 4763 if (f->child != NULL) { 4764 /* 4765 * This block traverses through 4766 * a wildcard anchor. 4767 */ 4768 if (match != NULL && *match) { 4769 /* 4770 * If any of "*" matched, then 4771 * "foo/ *" matched, mark frame 4772 * appropriately. 4773 */ 4774 PF_ETH_ANCHOR_SET_MATCH(f); 4775 *match = 0; 4776 } 4777 f->child = RB_NEXT(pf_keth_anchor_node, 4778 &fr->anchor->children, f->child); 4779 if (f->child != NULL) { 4780 *rs = &f->child->ruleset; 4781 *r = TAILQ_FIRST((*rs)->active.rules); 4782 if (*r == NULL) 4783 continue; 4784 else 4785 break; 4786 } 4787 } 4788 (*depth)--; 4789 if (*depth == 0 && a != NULL) 4790 *a = NULL; 4791 *rs = f->rs; 4792 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4793 quick = fr->quick; 4794 *r = TAILQ_NEXT(fr, entries); 4795 } while (*r == NULL); 4796 4797 return (quick); 4798 } 4799 4800 void 4801 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4802 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4803 { 4804 switch (af) { 4805 #ifdef INET 4806 case AF_INET: 4807 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4808 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4809 break; 4810 #endif /* INET */ 4811 #ifdef INET6 4812 case AF_INET6: 4813 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4814 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4815 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4816 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4817 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4818 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4819 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4820 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4821 break; 4822 #endif /* INET6 */ 4823 } 4824 } 4825 4826 void 4827 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4828 { 4829 switch (af) { 4830 #ifdef INET 4831 case AF_INET: 4832 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4833 break; 4834 #endif /* INET */ 4835 #ifdef INET6 4836 case AF_INET6: 4837 if (addr->addr32[3] == 0xffffffff) { 4838 addr->addr32[3] = 0; 4839 if (addr->addr32[2] == 0xffffffff) { 4840 addr->addr32[2] = 0; 4841 if (addr->addr32[1] == 0xffffffff) { 4842 addr->addr32[1] = 0; 4843 addr->addr32[0] = 4844 htonl(ntohl(addr->addr32[0]) + 1); 4845 } else 4846 addr->addr32[1] = 4847 htonl(ntohl(addr->addr32[1]) + 1); 4848 } else 4849 addr->addr32[2] = 4850 htonl(ntohl(addr->addr32[2]) + 1); 4851 } else 4852 addr->addr32[3] = 4853 htonl(ntohl(addr->addr32[3]) + 1); 4854 break; 4855 #endif /* INET6 */ 4856 } 4857 } 4858 4859 void 4860 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4861 { 4862 /* 4863 * Modern rules use the same flags in rules as they do in states. 4864 */ 4865 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4866 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4867 4868 /* 4869 * Old-style scrub rules have different flags which need to be translated. 4870 */ 4871 if (r->rule_flag & PFRULE_RANDOMID) 4872 a->flags |= PFSTATE_RANDOMID; 4873 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4874 a->flags |= PFSTATE_SETTOS; 4875 a->set_tos = r->set_tos; 4876 } 4877 4878 if (r->qid) 4879 a->qid = r->qid; 4880 if (r->pqid) 4881 a->pqid = r->pqid; 4882 if (r->rtableid >= 0) 4883 a->rtableid = r->rtableid; 4884 a->log |= r->log; 4885 if (r->min_ttl) 4886 a->min_ttl = r->min_ttl; 4887 if (r->max_mss) 4888 a->max_mss = r->max_mss; 4889 if (r->dnpipe) 4890 a->dnpipe = r->dnpipe; 4891 if (r->dnrpipe) 4892 a->dnrpipe = r->dnrpipe; 4893 if (r->dnpipe || r->dnrpipe) { 4894 if (r->free_flags & PFRULE_DN_IS_PIPE) 4895 a->flags |= PFSTATE_DN_IS_PIPE; 4896 else 4897 a->flags &= ~PFSTATE_DN_IS_PIPE; 4898 } 4899 if (r->scrub_flags & PFSTATE_SETPRIO) { 4900 a->set_prio[0] = r->set_prio[0]; 4901 a->set_prio[1] = r->set_prio[1]; 4902 } 4903 if (r->allow_opts) 4904 a->allow_opts = r->allow_opts; 4905 if (r->max_pkt_size) 4906 a->max_pkt_size = r->max_pkt_size; 4907 } 4908 4909 int 4910 pf_socket_lookup(struct pf_pdesc *pd) 4911 { 4912 struct pf_addr *saddr, *daddr; 4913 u_int16_t sport, dport; 4914 struct inpcbinfo *pi; 4915 struct inpcb *inp; 4916 4917 pd->lookup.uid = UID_MAX; 4918 pd->lookup.gid = GID_MAX; 4919 4920 switch (pd->proto) { 4921 case IPPROTO_TCP: 4922 sport = pd->hdr.tcp.th_sport; 4923 dport = pd->hdr.tcp.th_dport; 4924 pi = &V_tcbinfo; 4925 break; 4926 case IPPROTO_UDP: 4927 sport = pd->hdr.udp.uh_sport; 4928 dport = pd->hdr.udp.uh_dport; 4929 pi = &V_udbinfo; 4930 break; 4931 default: 4932 return (-1); 4933 } 4934 if (pd->dir == PF_IN) { 4935 saddr = pd->src; 4936 daddr = pd->dst; 4937 } else { 4938 u_int16_t p; 4939 4940 p = sport; 4941 sport = dport; 4942 dport = p; 4943 saddr = pd->dst; 4944 daddr = pd->src; 4945 } 4946 switch (pd->af) { 4947 #ifdef INET 4948 case AF_INET: 4949 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4950 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4951 if (inp == NULL) { 4952 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4953 daddr->v4, dport, INPLOOKUP_WILDCARD | 4954 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4955 if (inp == NULL) 4956 return (-1); 4957 } 4958 break; 4959 #endif /* INET */ 4960 #ifdef INET6 4961 case AF_INET6: 4962 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4963 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4964 if (inp == NULL) { 4965 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4966 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4967 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4968 if (inp == NULL) 4969 return (-1); 4970 } 4971 break; 4972 #endif /* INET6 */ 4973 default: 4974 unhandled_af(pd->af); 4975 } 4976 INP_RLOCK_ASSERT(inp); 4977 pd->lookup.uid = inp->inp_cred->cr_uid; 4978 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4979 INP_RUNLOCK(inp); 4980 4981 return (1); 4982 } 4983 4984 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 4985 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 4986 * 4987 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 4988 */ 4989 uint8_t* 4990 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 4991 u_int8_t min_typelen) 4992 { 4993 uint8_t *eoh = opts + hlen; 4994 4995 if (min_typelen < 2) 4996 return (NULL); 4997 4998 while ((eoh - opt) >= min_typelen) { 4999 switch (*opt) { 5000 case TCPOPT_EOL: 5001 /* FALLTHROUGH - Workaround the failure of some 5002 systems to NOP-pad their bzero'd option buffers, 5003 producing spurious EOLs */ 5004 case TCPOPT_NOP: 5005 opt++; 5006 continue; 5007 default: 5008 if (opt[0] == type && 5009 opt[1] >= min_typelen) 5010 return (opt); 5011 } 5012 5013 opt += MAX(opt[1], 2); /* evade infinite loops */ 5014 } 5015 5016 return (NULL); 5017 } 5018 5019 u_int8_t 5020 pf_get_wscale(struct pf_pdesc *pd) 5021 { 5022 int olen; 5023 uint8_t opts[MAX_TCPOPTLEN], *opt; 5024 uint8_t wscale = 0; 5025 5026 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5027 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 5028 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 5029 return (0); 5030 5031 opt = opts; 5032 while ((opt = pf_find_tcpopt(opt, opts, olen, 5033 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 5034 wscale = opt[2]; 5035 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 5036 wscale |= PF_WSCALE_FLAG; 5037 5038 opt += opt[1]; 5039 } 5040 5041 return (wscale); 5042 } 5043 5044 u_int16_t 5045 pf_get_mss(struct pf_pdesc *pd) 5046 { 5047 int olen; 5048 uint8_t opts[MAX_TCPOPTLEN], *opt; 5049 u_int16_t mss = V_tcp_mssdflt; 5050 5051 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5052 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 5053 pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af)) 5054 return (0); 5055 5056 opt = opts; 5057 while ((opt = pf_find_tcpopt(opt, opts, olen, 5058 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 5059 memcpy(&mss, (opt + 2), 2); 5060 mss = ntohs(mss); 5061 opt += opt[1]; 5062 } 5063 5064 return (mss); 5065 } 5066 5067 static u_int16_t 5068 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5069 { 5070 struct nhop_object *nh; 5071 #ifdef INET6 5072 struct in6_addr dst6; 5073 uint32_t scopeid; 5074 #endif /* INET6 */ 5075 int hlen = 0; 5076 uint16_t mss = 0; 5077 5078 NET_EPOCH_ASSERT(); 5079 5080 switch (af) { 5081 #ifdef INET 5082 case AF_INET: 5083 hlen = sizeof(struct ip); 5084 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5085 if (nh != NULL) 5086 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5087 break; 5088 #endif /* INET */ 5089 #ifdef INET6 5090 case AF_INET6: 5091 hlen = sizeof(struct ip6_hdr); 5092 in6_splitscope(&addr->v6, &dst6, &scopeid); 5093 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5094 if (nh != NULL) 5095 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5096 break; 5097 #endif /* INET6 */ 5098 } 5099 5100 mss = max(V_tcp_mssdflt, mss); 5101 mss = min(mss, offer); 5102 mss = max(mss, 64); /* sanity - at least max opt space */ 5103 return (mss); 5104 } 5105 5106 static u_int32_t 5107 pf_tcp_iss(struct pf_pdesc *pd) 5108 { 5109 SHA512_CTX ctx; 5110 union { 5111 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5112 uint32_t words[1]; 5113 } digest; 5114 5115 if (V_pf_tcp_secret_init == 0) { 5116 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5117 SHA512_Init(&V_pf_tcp_secret_ctx); 5118 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5119 sizeof(V_pf_tcp_secret)); 5120 V_pf_tcp_secret_init = 1; 5121 } 5122 5123 ctx = V_pf_tcp_secret_ctx; 5124 5125 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5126 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5127 switch (pd->af) { 5128 case AF_INET6: 5129 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5130 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5131 break; 5132 case AF_INET: 5133 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5134 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5135 break; 5136 } 5137 SHA512_Final(digest.bytes, &ctx); 5138 V_pf_tcp_iss_off += 4096; 5139 #define ISN_RANDOM_INCREMENT (4096 - 1) 5140 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5141 V_pf_tcp_iss_off); 5142 #undef ISN_RANDOM_INCREMENT 5143 } 5144 5145 static bool 5146 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5147 { 5148 bool match = true; 5149 5150 /* Always matches if not set */ 5151 if (! r->isset) 5152 return (!r->neg); 5153 5154 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5155 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5156 match = false; 5157 break; 5158 } 5159 } 5160 5161 return (match ^ r->neg); 5162 } 5163 5164 static int 5165 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5166 { 5167 if (*tag == -1) 5168 *tag = mtag; 5169 5170 return ((!r->match_tag_not && r->match_tag == *tag) || 5171 (r->match_tag_not && r->match_tag != *tag)); 5172 } 5173 5174 static void 5175 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5176 { 5177 /* If we don't have the interface drop the packet. */ 5178 if (ifp == NULL) { 5179 m_freem(m); 5180 return; 5181 } 5182 5183 switch (ifp->if_type) { 5184 case IFT_ETHER: 5185 case IFT_XETHER: 5186 case IFT_L2VLAN: 5187 case IFT_BRIDGE: 5188 case IFT_IEEE8023ADLAG: 5189 break; 5190 default: 5191 m_freem(m); 5192 return; 5193 } 5194 5195 ifp->if_transmit(ifp, m); 5196 } 5197 5198 static int 5199 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5200 { 5201 #ifdef INET 5202 struct ip ip; 5203 #endif /* INET */ 5204 #ifdef INET6 5205 struct ip6_hdr ip6; 5206 #endif /* INET6 */ 5207 struct mbuf *m = *m0; 5208 struct ether_header *e; 5209 struct pf_keth_rule *r, *rm, *a = NULL; 5210 struct pf_keth_ruleset *ruleset = NULL; 5211 struct pf_mtag *mtag; 5212 struct pf_keth_ruleq *rules; 5213 struct pf_addr *src = NULL, *dst = NULL; 5214 struct pfi_kkif *bridge_to; 5215 sa_family_t af = 0; 5216 uint16_t proto; 5217 int asd = 0, match = 0; 5218 int tag = -1; 5219 uint8_t action; 5220 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACK_MAX]; 5221 5222 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5223 NET_EPOCH_ASSERT(); 5224 5225 PF_RULES_RLOCK_TRACKER; 5226 5227 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5228 5229 mtag = pf_find_mtag(m); 5230 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5231 /* Dummynet re-injects packets after they've 5232 * completed their delay. We've already 5233 * processed them, so pass unconditionally. */ 5234 5235 /* But only once. We may see the packet multiple times (e.g. 5236 * PFIL_IN/PFIL_OUT). */ 5237 pf_dummynet_flag_remove(m, mtag); 5238 5239 return (PF_PASS); 5240 } 5241 5242 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5243 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5244 DPFPRINTF(PF_DEBUG_URGENT, 5245 ("%s: m_len < sizeof(struct ether_header)" 5246 ", pullup failed\n", __func__)); 5247 return (PF_DROP); 5248 } 5249 e = mtod(m, struct ether_header *); 5250 proto = ntohs(e->ether_type); 5251 5252 switch (proto) { 5253 #ifdef INET 5254 case ETHERTYPE_IP: { 5255 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5256 sizeof(ip))) 5257 return (PF_DROP); 5258 5259 af = AF_INET; 5260 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5261 (caddr_t)&ip); 5262 src = (struct pf_addr *)&ip.ip_src; 5263 dst = (struct pf_addr *)&ip.ip_dst; 5264 break; 5265 } 5266 #endif /* INET */ 5267 #ifdef INET6 5268 case ETHERTYPE_IPV6: { 5269 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5270 sizeof(ip6))) 5271 return (PF_DROP); 5272 5273 af = AF_INET6; 5274 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5275 (caddr_t)&ip6); 5276 src = (struct pf_addr *)&ip6.ip6_src; 5277 dst = (struct pf_addr *)&ip6.ip6_dst; 5278 break; 5279 } 5280 #endif /* INET6 */ 5281 } 5282 5283 PF_RULES_RLOCK(); 5284 5285 ruleset = V_pf_keth; 5286 rules = atomic_load_ptr(&ruleset->active.rules); 5287 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5288 counter_u64_add(r->evaluations, 1); 5289 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5290 5291 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5292 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5293 "kif"); 5294 r = r->skip[PFE_SKIP_IFP].ptr; 5295 } 5296 else if (r->direction && r->direction != dir) { 5297 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5298 "dir"); 5299 r = r->skip[PFE_SKIP_DIR].ptr; 5300 } 5301 else if (r->proto && r->proto != proto) { 5302 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5303 "proto"); 5304 r = r->skip[PFE_SKIP_PROTO].ptr; 5305 } 5306 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5307 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5308 "src"); 5309 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5310 } 5311 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5312 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5313 "dst"); 5314 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5315 } 5316 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5317 r->ipsrc.neg, kif, M_GETFIB(m))) { 5318 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5319 "ip_src"); 5320 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5321 } 5322 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5323 r->ipdst.neg, kif, M_GETFIB(m))) { 5324 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5325 "ip_dst"); 5326 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5327 } 5328 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5329 mtag ? mtag->tag : 0)) { 5330 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5331 "match_tag"); 5332 r = TAILQ_NEXT(r, entries); 5333 } 5334 else { 5335 if (r->tag) 5336 tag = r->tag; 5337 if (r->anchor == NULL) { 5338 /* Rule matches */ 5339 rm = r; 5340 5341 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5342 5343 if (r->quick) 5344 break; 5345 5346 r = TAILQ_NEXT(r, entries); 5347 } else { 5348 pf_step_into_keth_anchor(anchor_stack, &asd, 5349 &ruleset, &r, &a, &match); 5350 } 5351 } 5352 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5353 &ruleset, &r, &a, &match)) 5354 break; 5355 } 5356 5357 r = rm; 5358 5359 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5360 5361 /* Default to pass. */ 5362 if (r == NULL) { 5363 PF_RULES_RUNLOCK(); 5364 return (PF_PASS); 5365 } 5366 5367 /* Execute action. */ 5368 counter_u64_add(r->packets[dir == PF_OUT], 1); 5369 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5370 pf_update_timestamp(r); 5371 5372 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5373 if (r->action == PF_DROP) { 5374 PF_RULES_RUNLOCK(); 5375 return (PF_DROP); 5376 } 5377 5378 if (tag > 0) { 5379 if (mtag == NULL) 5380 mtag = pf_get_mtag(m); 5381 if (mtag == NULL) { 5382 PF_RULES_RUNLOCK(); 5383 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5384 return (PF_DROP); 5385 } 5386 mtag->tag = tag; 5387 } 5388 5389 if (r->qid != 0) { 5390 if (mtag == NULL) 5391 mtag = pf_get_mtag(m); 5392 if (mtag == NULL) { 5393 PF_RULES_RUNLOCK(); 5394 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5395 return (PF_DROP); 5396 } 5397 mtag->qid = r->qid; 5398 } 5399 5400 action = r->action; 5401 bridge_to = r->bridge_to; 5402 5403 /* Dummynet */ 5404 if (r->dnpipe) { 5405 struct ip_fw_args dnflow; 5406 5407 /* Drop packet if dummynet is not loaded. */ 5408 if (ip_dn_io_ptr == NULL) { 5409 PF_RULES_RUNLOCK(); 5410 m_freem(m); 5411 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5412 return (PF_DROP); 5413 } 5414 if (mtag == NULL) 5415 mtag = pf_get_mtag(m); 5416 if (mtag == NULL) { 5417 PF_RULES_RUNLOCK(); 5418 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5419 return (PF_DROP); 5420 } 5421 5422 bzero(&dnflow, sizeof(dnflow)); 5423 5424 /* We don't have port numbers here, so we set 0. That means 5425 * that we'll be somewhat limited in distinguishing flows (i.e. 5426 * only based on IP addresses, not based on port numbers), but 5427 * it's better than nothing. */ 5428 dnflow.f_id.dst_port = 0; 5429 dnflow.f_id.src_port = 0; 5430 dnflow.f_id.proto = 0; 5431 5432 dnflow.rule.info = r->dnpipe; 5433 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5434 if (r->dnflags & PFRULE_DN_IS_PIPE) 5435 dnflow.rule.info |= IPFW_IS_PIPE; 5436 5437 dnflow.f_id.extra = dnflow.rule.info; 5438 5439 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5440 dnflow.flags |= IPFW_ARGS_ETHER; 5441 dnflow.ifp = kif->pfik_ifp; 5442 5443 switch (af) { 5444 case AF_INET: 5445 dnflow.f_id.addr_type = 4; 5446 dnflow.f_id.src_ip = src->v4.s_addr; 5447 dnflow.f_id.dst_ip = dst->v4.s_addr; 5448 break; 5449 case AF_INET6: 5450 dnflow.flags |= IPFW_ARGS_IP6; 5451 dnflow.f_id.addr_type = 6; 5452 dnflow.f_id.src_ip6 = src->v6; 5453 dnflow.f_id.dst_ip6 = dst->v6; 5454 break; 5455 } 5456 5457 PF_RULES_RUNLOCK(); 5458 5459 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5460 ip_dn_io_ptr(m0, &dnflow); 5461 if (*m0 != NULL) 5462 pf_dummynet_flag_remove(m, mtag); 5463 } else { 5464 PF_RULES_RUNLOCK(); 5465 } 5466 5467 if (action == PF_PASS && bridge_to) { 5468 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5469 *m0 = NULL; /* We've eaten the packet. */ 5470 } 5471 5472 return (action); 5473 } 5474 5475 #define PF_TEST_ATTRIB(t, a) \ 5476 if (t) { \ 5477 r = a; \ 5478 continue; \ 5479 } else do { \ 5480 } while (0) 5481 5482 static __inline u_short 5483 pf_rule_apply_nat(struct pf_test_ctx *ctx, struct pf_krule *r) 5484 { 5485 struct pf_pdesc *pd = ctx->pd; 5486 u_short transerror; 5487 u_int8_t nat_action; 5488 5489 if (r->rule_flag & PFRULE_AFTO) { 5490 /* Don't translate if there was an old style NAT rule */ 5491 if (ctx->nr != NULL) 5492 return (PFRES_TRANSLATE); 5493 5494 /* pass af-to rules, unsupported on match rules */ 5495 KASSERT(r->action != PF_MATCH, ("%s: af-to on match rule", __func__)); 5496 /* XXX I can imagine scenarios where we have both NAT and RDR source tracking */ 5497 ctx->nat_pool = &(r->nat); 5498 ctx->nr = r; 5499 pd->naf = r->naf; 5500 if (pf_get_transaddr_af(ctx->nr, pd) == -1) { 5501 return (PFRES_TRANSLATE); 5502 } 5503 return (PFRES_MATCH); 5504 } else if (r->rdr.cur || r->nat.cur) { 5505 /* Don't translate if there was an old style NAT rule */ 5506 if (ctx->nr != NULL) 5507 return (PFRES_TRANSLATE); 5508 5509 /* match/pass nat-to/rdr-to rules */ 5510 ctx->nr = r; 5511 if (r->nat.cur) { 5512 nat_action = PF_NAT; 5513 ctx->nat_pool = &(r->nat); 5514 } else { 5515 nat_action = PF_RDR; 5516 ctx->nat_pool = &(r->rdr); 5517 } 5518 5519 transerror = pf_get_transaddr(ctx, ctx->nr, 5520 nat_action, ctx->nat_pool); 5521 if (transerror == PFRES_MATCH) { 5522 ctx->rewrite += pf_translate_compat(ctx); 5523 return(PFRES_MATCH); 5524 } 5525 return (transerror); 5526 } 5527 5528 return (PFRES_MAX); 5529 } 5530 5531 enum pf_test_status 5532 pf_match_rule(struct pf_test_ctx *ctx, struct pf_kruleset *ruleset) 5533 { 5534 struct pf_krule_item *ri; 5535 struct pf_krule *r; 5536 struct pf_krule *save_a; 5537 struct pf_kruleset *save_aruleset; 5538 struct pf_pdesc *pd = ctx->pd; 5539 u_short transerror; 5540 5541 r = TAILQ_FIRST(ruleset->rules[PF_RULESET_FILTER].active.ptr); 5542 while (r != NULL) { 5543 if (ctx->pd->related_rule) { 5544 *ctx->rm = ctx->pd->related_rule; 5545 break; 5546 } 5547 pf_counter_u64_add(&r->evaluations, 1); 5548 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5549 r->skip[PF_SKIP_IFP]); 5550 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5551 r->skip[PF_SKIP_DIR]); 5552 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5553 r->skip[PF_SKIP_AF]); 5554 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5555 r->skip[PF_SKIP_PROTO]); 5556 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5557 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5558 r->skip[PF_SKIP_SRC_ADDR]); 5559 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5560 r->dst.neg, NULL, M_GETFIB(pd->m)), 5561 r->skip[PF_SKIP_DST_ADDR]); 5562 switch (pd->virtual_proto) { 5563 case PF_VPROTO_FRAGMENT: 5564 /* tcp/udp only. port_op always 0 in other cases */ 5565 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5566 TAILQ_NEXT(r, entries)); 5567 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5568 TAILQ_NEXT(r, entries)); 5569 /* icmp only. type/code always 0 in other cases */ 5570 PF_TEST_ATTRIB((r->type || r->code), 5571 TAILQ_NEXT(r, entries)); 5572 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5573 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5574 TAILQ_NEXT(r, entries)); 5575 break; 5576 5577 case IPPROTO_TCP: 5578 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(ctx->th)) 5579 != r->flags, 5580 TAILQ_NEXT(r, entries)); 5581 /* FALLTHROUGH */ 5582 case IPPROTO_SCTP: 5583 case IPPROTO_UDP: 5584 /* tcp/udp only. port_op always 0 in other cases */ 5585 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5586 r->src.port[0], r->src.port[1], pd->nsport), 5587 r->skip[PF_SKIP_SRC_PORT]); 5588 /* tcp/udp only. port_op always 0 in other cases */ 5589 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5590 r->dst.port[0], r->dst.port[1], pd->ndport), 5591 r->skip[PF_SKIP_DST_PORT]); 5592 /* tcp/udp only. uid.op always 0 in other cases */ 5593 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5594 pf_socket_lookup(pd), 1)) && 5595 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5596 pd->lookup.uid), 5597 TAILQ_NEXT(r, entries)); 5598 /* tcp/udp only. gid.op always 0 in other cases */ 5599 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5600 pf_socket_lookup(pd), 1)) && 5601 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5602 pd->lookup.gid), 5603 TAILQ_NEXT(r, entries)); 5604 break; 5605 5606 case IPPROTO_ICMP: 5607 case IPPROTO_ICMPV6: 5608 /* icmp only. type always 0 in other cases */ 5609 PF_TEST_ATTRIB(r->type && r->type != ctx->icmptype + 1, 5610 TAILQ_NEXT(r, entries)); 5611 /* icmp only. type always 0 in other cases */ 5612 PF_TEST_ATTRIB(r->code && r->code != ctx->icmpcode + 1, 5613 TAILQ_NEXT(r, entries)); 5614 break; 5615 5616 default: 5617 break; 5618 } 5619 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5620 TAILQ_NEXT(r, entries)); 5621 PF_TEST_ATTRIB(r->prio && 5622 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5623 TAILQ_NEXT(r, entries)); 5624 PF_TEST_ATTRIB(r->prob && 5625 r->prob <= arc4random(), 5626 TAILQ_NEXT(r, entries)); 5627 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, 5628 &ctx->tag, pd->pf_mtag ? pd->pf_mtag->tag : 0), 5629 TAILQ_NEXT(r, entries)); 5630 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 5631 r->rcvifnot), 5632 TAILQ_NEXT(r, entries)); 5633 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5634 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5635 TAILQ_NEXT(r, entries)); 5636 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5637 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5638 pf_osfp_fingerprint(pd, ctx->th), 5639 r->os_fingerprint)), 5640 TAILQ_NEXT(r, entries)); 5641 /* must be last! */ 5642 if (r->pktrate.limit) { 5643 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 5644 TAILQ_NEXT(r, entries)); 5645 } 5646 /* FALLTHROUGH */ 5647 if (r->tag) 5648 ctx->tag = r->tag; 5649 if (r->anchor == NULL) { 5650 if (r->action == PF_MATCH) { 5651 /* 5652 * Apply translations before increasing counters, 5653 * in case it fails. 5654 */ 5655 transerror = pf_rule_apply_nat(ctx, r); 5656 switch (transerror) { 5657 case PFRES_MATCH: 5658 /* Translation action found in rule and applied successfully */ 5659 case PFRES_MAX: 5660 /* No translation action found in rule */ 5661 break; 5662 default: 5663 /* Translation action found in rule but failed to apply */ 5664 REASON_SET(&ctx->reason, transerror); 5665 return (PF_TEST_FAIL); 5666 } 5667 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5668 if (ri == NULL) { 5669 REASON_SET(&ctx->reason, PFRES_MEMORY); 5670 return (PF_TEST_FAIL); 5671 } 5672 ri->r = r; 5673 SLIST_INSERT_HEAD(&ctx->rules, ri, entry); 5674 pf_counter_u64_critical_enter(); 5675 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5676 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5677 pf_counter_u64_critical_exit(); 5678 pf_rule_to_actions(r, &pd->act); 5679 if (r->log) 5680 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5681 ctx->a, ruleset, pd, 1, NULL); 5682 } else { 5683 /* 5684 * found matching r 5685 */ 5686 *ctx->rm = r; 5687 /* 5688 * anchor, with ruleset, where r belongs to 5689 */ 5690 *ctx->am = ctx->a; 5691 /* 5692 * ruleset where r belongs to 5693 */ 5694 *ctx->rsm = ruleset; 5695 /* 5696 * ruleset, where anchor belongs to. 5697 */ 5698 ctx->arsm = ctx->aruleset; 5699 } 5700 if (pd->act.log & PF_LOG_MATCHES) 5701 pf_log_matches(pd, r, ctx->a, ruleset, &ctx->rules); 5702 if (r->quick) { 5703 ctx->test_status = PF_TEST_QUICK; 5704 break; 5705 } 5706 } else { 5707 save_a = ctx->a; 5708 save_aruleset = ctx->aruleset; 5709 5710 ctx->a = r; /* remember anchor */ 5711 ctx->aruleset = ruleset; /* and its ruleset */ 5712 if (ctx->a->quick) 5713 ctx->test_status = PF_TEST_QUICK; 5714 /* 5715 * Note: we don't need to restore if we are not going 5716 * to continue with ruleset evaluation. 5717 */ 5718 if (pf_step_into_anchor(ctx, r) != PF_TEST_OK) { 5719 break; 5720 } 5721 ctx->a = save_a; 5722 ctx->aruleset = save_aruleset; 5723 } 5724 r = TAILQ_NEXT(r, entries); 5725 } 5726 5727 return (ctx->test_status); 5728 } 5729 5730 static int 5731 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5732 struct pf_pdesc *pd, struct pf_krule **am, 5733 struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp) 5734 { 5735 struct pf_krule *r = NULL; 5736 struct pf_kruleset *ruleset = NULL; 5737 struct pf_krule_item *ri; 5738 struct pf_test_ctx ctx; 5739 u_short transerror; 5740 int action = PF_PASS; 5741 u_int16_t bproto_sum = 0, bip_sum = 0; 5742 enum pf_test_status rv; 5743 5744 PF_RULES_RASSERT(); 5745 5746 bzero(&ctx, sizeof(ctx)); 5747 ctx.tag = -1; 5748 ctx.pd = pd; 5749 ctx.rm = rm; 5750 ctx.am = am; 5751 ctx.rsm = rsm; 5752 ctx.th = &pd->hdr.tcp; 5753 ctx.reason = *reason; 5754 SLIST_INIT(&ctx.rules); 5755 5756 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 5757 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 5758 5759 if (inp != NULL) { 5760 INP_LOCK_ASSERT(inp); 5761 pd->lookup.uid = inp->inp_cred->cr_uid; 5762 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 5763 pd->lookup.done = 1; 5764 } 5765 5766 if (pd->ip_sum) 5767 bip_sum = *pd->ip_sum; 5768 5769 switch (pd->virtual_proto) { 5770 case IPPROTO_TCP: 5771 bproto_sum = ctx.th->th_sum; 5772 pd->nsport = ctx.th->th_sport; 5773 pd->ndport = ctx.th->th_dport; 5774 break; 5775 case IPPROTO_UDP: 5776 bproto_sum = pd->hdr.udp.uh_sum; 5777 pd->nsport = pd->hdr.udp.uh_sport; 5778 pd->ndport = pd->hdr.udp.uh_dport; 5779 break; 5780 case IPPROTO_SCTP: 5781 pd->nsport = pd->hdr.sctp.src_port; 5782 pd->ndport = pd->hdr.sctp.dest_port; 5783 break; 5784 #ifdef INET 5785 case IPPROTO_ICMP: 5786 MPASS(pd->af == AF_INET); 5787 ctx.icmptype = pd->hdr.icmp.icmp_type; 5788 ctx.icmpcode = pd->hdr.icmp.icmp_code; 5789 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 5790 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 5791 if (ctx.icmp_dir == PF_IN) { 5792 pd->nsport = ctx.virtual_id; 5793 pd->ndport = ctx.virtual_type; 5794 } else { 5795 pd->nsport = ctx.virtual_type; 5796 pd->ndport = ctx.virtual_id; 5797 } 5798 break; 5799 #endif /* INET */ 5800 #ifdef INET6 5801 case IPPROTO_ICMPV6: 5802 MPASS(pd->af == AF_INET6); 5803 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 5804 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 5805 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 5806 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 5807 if (ctx.icmp_dir == PF_IN) { 5808 pd->nsport = ctx.virtual_id; 5809 pd->ndport = ctx.virtual_type; 5810 } else { 5811 pd->nsport = ctx.virtual_type; 5812 pd->ndport = ctx.virtual_id; 5813 } 5814 5815 break; 5816 #endif /* INET6 */ 5817 default: 5818 pd->nsport = pd->ndport = 0; 5819 break; 5820 } 5821 pd->osport = pd->nsport; 5822 pd->odport = pd->ndport; 5823 5824 /* check packet for BINAT/NAT/RDR */ 5825 transerror = pf_get_translation(&ctx); 5826 switch (transerror) { 5827 default: 5828 /* A translation error occurred. */ 5829 REASON_SET(&ctx.reason, transerror); 5830 goto cleanup; 5831 case PFRES_MAX: 5832 /* No match. */ 5833 break; 5834 case PFRES_MATCH: 5835 KASSERT(ctx.sk != NULL, ("%s: null sk", __func__)); 5836 KASSERT(ctx.nk != NULL, ("%s: null nk", __func__)); 5837 if (ctx.nr->log) { 5838 PFLOG_PACKET(ctx.nr->action, PFRES_MATCH, ctx.nr, ctx.a, 5839 ruleset, pd, 1, NULL); 5840 } 5841 5842 ctx.rewrite += pf_translate_compat(&ctx); 5843 ctx.nat_pool = &(ctx.nr->rdr); 5844 } 5845 5846 ruleset = &pf_main_ruleset; 5847 rv = pf_match_rule(&ctx, ruleset); 5848 if (rv == PF_TEST_FAIL) { 5849 /* 5850 * Reason has been set in pf_match_rule() already. 5851 */ 5852 goto cleanup; 5853 } 5854 5855 r = *ctx.rm; /* matching rule */ 5856 ctx.a = *ctx.am; /* rule that defines an anchor containing 'r' */ 5857 ruleset = *ctx.rsm; /* ruleset of the anchor defined by the rule 'a' */ 5858 ctx.aruleset = ctx.arsm; /* ruleset of the 'a' rule itself */ 5859 5860 REASON_SET(&ctx.reason, PFRES_MATCH); 5861 5862 /* apply actions for last matching pass/block rule */ 5863 pf_rule_to_actions(r, &pd->act); 5864 transerror = pf_rule_apply_nat(&ctx, r); 5865 switch (transerror) { 5866 case PFRES_MATCH: 5867 /* Translation action found in rule and applied successfully */ 5868 case PFRES_MAX: 5869 /* No translation action found in rule */ 5870 break; 5871 default: 5872 /* Translation action found in rule but failed to apply */ 5873 REASON_SET(&ctx.reason, transerror); 5874 goto cleanup; 5875 } 5876 5877 if (r->log) { 5878 if (ctx.rewrite) 5879 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5880 PFLOG_PACKET(r->action, ctx.reason, r, ctx.a, ruleset, pd, 1, NULL); 5881 } 5882 if (pd->act.log & PF_LOG_MATCHES) 5883 pf_log_matches(pd, r, ctx.a, ruleset, &ctx.rules); 5884 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5885 (r->action == PF_DROP) && 5886 ((r->rule_flag & PFRULE_RETURNRST) || 5887 (r->rule_flag & PFRULE_RETURNICMP) || 5888 (r->rule_flag & PFRULE_RETURN))) { 5889 pf_return(r, ctx.nr, pd, ctx.th, bproto_sum, 5890 bip_sum, &ctx.reason, r->rtableid); 5891 } 5892 5893 if (r->action == PF_DROP) 5894 goto cleanup; 5895 5896 if (ctx.tag > 0 && pf_tag_packet(pd, ctx.tag)) { 5897 REASON_SET(&ctx.reason, PFRES_MEMORY); 5898 goto cleanup; 5899 } 5900 if (pd->act.rtableid >= 0) 5901 M_SETFIB(pd->m, pd->act.rtableid); 5902 5903 if (r->rt) { 5904 struct pf_ksrc_node *sn = NULL; 5905 struct pf_srchash *snh = NULL; 5906 /* 5907 * Set act.rt here instead of in pf_rule_to_actions() because 5908 * it is applied only from the last pass rule. 5909 */ 5910 pd->act.rt = r->rt; 5911 /* Don't use REASON_SET, pf_map_addr increases the reason counters */ 5912 ctx.reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, 5913 &pd->act.rt_kif, NULL, &sn, &snh, &(r->route), PF_SN_ROUTE); 5914 if (ctx.reason != 0) 5915 goto cleanup; 5916 } 5917 5918 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5919 (!ctx.state_icmp && (r->keep_state || ctx.nr != NULL || 5920 (pd->flags & PFDESC_TCP_NORM)))) { 5921 bool nat64; 5922 5923 action = pf_create_state(r, &ctx, sm, bproto_sum, bip_sum); 5924 ctx.sk = ctx.nk = NULL; 5925 if (action != PF_PASS) { 5926 pf_udp_mapping_release(ctx.udp_mapping); 5927 if (r->log || (ctx.nr != NULL && ctx.nr->log) || 5928 ctx.reason == PFRES_MEMORY) 5929 pd->act.log |= PF_LOG_FORCE; 5930 if (action == PF_DROP && 5931 (r->rule_flag & PFRULE_RETURN)) 5932 pf_return(r, ctx.nr, pd, ctx.th, 5933 bproto_sum, bip_sum, &ctx.reason, 5934 pd->act.rtableid); 5935 *reason = ctx.reason; 5936 return (action); 5937 } 5938 5939 nat64 = pd->af != pd->naf; 5940 if (nat64) { 5941 int ret; 5942 5943 if (ctx.sk == NULL) 5944 ctx.sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 5945 if (ctx.nk == NULL) 5946 ctx.nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 5947 5948 if (pd->dir == PF_IN) { 5949 ret = pf_translate(pd, &ctx.sk->addr[pd->didx], 5950 ctx.sk->port[pd->didx], &ctx.sk->addr[pd->sidx], 5951 ctx.sk->port[pd->sidx], ctx.virtual_type, 5952 ctx.icmp_dir); 5953 } else { 5954 ret = pf_translate(pd, &ctx.sk->addr[pd->sidx], 5955 ctx.sk->port[pd->sidx], &ctx.sk->addr[pd->didx], 5956 ctx.sk->port[pd->didx], ctx.virtual_type, 5957 ctx.icmp_dir); 5958 } 5959 5960 if (ret < 0) 5961 goto cleanup; 5962 5963 ctx.rewrite += ret; 5964 5965 if (ctx.rewrite && ctx.sk->af != ctx.nk->af) 5966 action = PF_AFRT; 5967 } 5968 } else { 5969 while ((ri = SLIST_FIRST(&ctx.rules))) { 5970 SLIST_REMOVE_HEAD(&ctx.rules, entry); 5971 free(ri, M_PF_RULE_ITEM); 5972 } 5973 5974 uma_zfree(V_pf_state_key_z, ctx.sk); 5975 uma_zfree(V_pf_state_key_z, ctx.nk); 5976 ctx.sk = ctx.nk = NULL; 5977 pf_udp_mapping_release(ctx.udp_mapping); 5978 } 5979 5980 /* copy back packet headers if we performed NAT operations */ 5981 if (ctx.rewrite) 5982 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5983 5984 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5985 pd->dir == PF_OUT && 5986 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) { 5987 /* 5988 * We want the state created, but we dont 5989 * want to send this in case a partner 5990 * firewall has to know about it to allow 5991 * replies through it. 5992 */ 5993 *reason = ctx.reason; 5994 return (PF_DEFER); 5995 } 5996 5997 *reason = ctx.reason; 5998 return (action); 5999 6000 cleanup: 6001 while ((ri = SLIST_FIRST(&ctx.rules))) { 6002 SLIST_REMOVE_HEAD(&ctx.rules, entry); 6003 free(ri, M_PF_RULE_ITEM); 6004 } 6005 6006 uma_zfree(V_pf_state_key_z, ctx.sk); 6007 uma_zfree(V_pf_state_key_z, ctx.nk); 6008 pf_udp_mapping_release(ctx.udp_mapping); 6009 *reason = ctx.reason; 6010 6011 return (PF_DROP); 6012 } 6013 6014 static int 6015 pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx, 6016 struct pf_kstate **sm, u_int16_t bproto_sum, u_int16_t bip_sum) 6017 { 6018 struct pf_pdesc *pd = ctx->pd; 6019 struct pf_kstate *s = NULL; 6020 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6021 /* 6022 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6023 * but for PF_SN_NAT it is different. Don't try optimizing it, 6024 * just store all 3 hashes. 6025 */ 6026 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6027 struct tcphdr *th = &pd->hdr.tcp; 6028 u_int16_t mss = V_tcp_mssdflt; 6029 u_short sn_reason; 6030 struct pf_krule_item *ri; 6031 6032 /* check maximums */ 6033 if (r->max_states && 6034 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6035 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6036 REASON_SET(&ctx->reason, PFRES_MAXSTATES); 6037 goto csfailed; 6038 } 6039 /* src node for limits */ 6040 if ((r->rule_flag & PFRULE_SRCTRACK) && 6041 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6042 NULL, NULL, PF_SN_LIMIT)) != 0) { 6043 REASON_SET(&ctx->reason, sn_reason); 6044 goto csfailed; 6045 } 6046 /* src node for route-to rule */ 6047 if (r->rt) { 6048 if ((r->route.opts & PF_POOL_STICKYADDR) && 6049 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, 6050 pd->af, &pd->act.rt_addr, pd->act.rt_kif, 6051 PF_SN_ROUTE)) != 0) { 6052 REASON_SET(&ctx->reason, sn_reason); 6053 goto csfailed; 6054 } 6055 } 6056 /* src node for translation rule */ 6057 if (ctx->nr != NULL) { 6058 KASSERT(ctx->nat_pool != NULL, ("%s: nat_pool is NULL", __func__)); 6059 if ((ctx->nat_pool->opts & PF_POOL_STICKYADDR) && 6060 (sn_reason = pf_insert_src_node(sns, snhs, ctx->nr, 6061 &ctx->sk->addr[pd->sidx], pd->af, &ctx->nk->addr[1], NULL, 6062 PF_SN_NAT)) != 0 ) { 6063 REASON_SET(&ctx->reason, sn_reason); 6064 goto csfailed; 6065 } 6066 } 6067 s = pf_alloc_state(M_NOWAIT); 6068 if (s == NULL) { 6069 REASON_SET(&ctx->reason, PFRES_MEMORY); 6070 goto csfailed; 6071 } 6072 s->rule = r; 6073 s->nat_rule = ctx->nr; 6074 s->anchor = ctx->a; 6075 memcpy(&s->match_rules, &ctx->rules, sizeof(s->match_rules)); 6076 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6077 6078 if (pd->act.allow_opts) 6079 s->state_flags |= PFSTATE_ALLOWOPTS; 6080 if (r->rule_flag & PFRULE_STATESLOPPY) 6081 s->state_flags |= PFSTATE_SLOPPY; 6082 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6083 s->state_flags |= PFSTATE_SCRUB_TCP; 6084 if ((r->rule_flag & PFRULE_PFLOW) || 6085 (ctx->nr != NULL && ctx->nr->rule_flag & PFRULE_PFLOW)) 6086 s->state_flags |= PFSTATE_PFLOW; 6087 6088 s->act.log = pd->act.log & PF_LOG_ALL; 6089 s->sync_state = PFSYNC_S_NONE; 6090 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6091 6092 if (ctx->nr != NULL) 6093 s->act.log |= ctx->nr->log & PF_LOG_ALL; 6094 switch (pd->proto) { 6095 case IPPROTO_TCP: 6096 s->src.seqlo = ntohl(th->th_seq); 6097 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6098 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6099 r->keep_state == PF_STATE_MODULATE) { 6100 /* Generate sequence number modulator */ 6101 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6102 0) 6103 s->src.seqdiff = 1; 6104 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6105 htonl(s->src.seqlo + s->src.seqdiff), 0); 6106 ctx->rewrite = 1; 6107 } else 6108 s->src.seqdiff = 0; 6109 if (tcp_get_flags(th) & TH_SYN) { 6110 s->src.seqhi++; 6111 s->src.wscale = pf_get_wscale(pd); 6112 } 6113 s->src.max_win = MAX(ntohs(th->th_win), 1); 6114 if (s->src.wscale & PF_WSCALE_MASK) { 6115 /* Remove scale factor from initial window */ 6116 int win = s->src.max_win; 6117 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6118 s->src.max_win = (win - 1) >> 6119 (s->src.wscale & PF_WSCALE_MASK); 6120 } 6121 if (tcp_get_flags(th) & TH_FIN) 6122 s->src.seqhi++; 6123 s->dst.seqhi = 1; 6124 s->dst.max_win = 1; 6125 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6126 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6127 s->timeout = PFTM_TCP_FIRST_PACKET; 6128 atomic_add_32(&V_pf_status.states_halfopen, 1); 6129 break; 6130 case IPPROTO_UDP: 6131 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6132 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6133 s->timeout = PFTM_UDP_FIRST_PACKET; 6134 break; 6135 case IPPROTO_SCTP: 6136 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6137 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6138 s->timeout = PFTM_SCTP_FIRST_PACKET; 6139 break; 6140 case IPPROTO_ICMP: 6141 #ifdef INET6 6142 case IPPROTO_ICMPV6: 6143 #endif /* INET6 */ 6144 s->timeout = PFTM_ICMP_FIRST_PACKET; 6145 break; 6146 default: 6147 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6148 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6149 s->timeout = PFTM_OTHER_FIRST_PACKET; 6150 } 6151 6152 s->creation = s->expire = pf_get_uptime(); 6153 6154 if (pd->proto == IPPROTO_TCP) { 6155 if (s->state_flags & PFSTATE_SCRUB_TCP && 6156 pf_normalize_tcp_init(pd, th, &s->src)) { 6157 REASON_SET(&ctx->reason, PFRES_MEMORY); 6158 goto csfailed; 6159 } 6160 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6161 pf_normalize_tcp_stateful(pd, &ctx->reason, th, s, 6162 &s->src, &s->dst, &ctx->rewrite)) { 6163 /* This really shouldn't happen!!! */ 6164 DPFPRINTF(PF_DEBUG_URGENT, 6165 ("%s: tcp normalize failed on first " 6166 "pkt\n", __func__)); 6167 goto csfailed; 6168 } 6169 } else if (pd->proto == IPPROTO_SCTP) { 6170 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6171 goto csfailed; 6172 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6173 goto csfailed; 6174 } 6175 s->direction = pd->dir; 6176 6177 /* 6178 * sk/nk could already been setup by pf_get_translation(). 6179 */ 6180 if (ctx->sk == NULL && ctx->nk == NULL) { 6181 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6182 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6183 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, 6184 &ctx->sk, &ctx->nk)) { 6185 goto csfailed; 6186 } 6187 } else 6188 KASSERT((ctx->sk != NULL && ctx->nk != NULL), ("%s: nr %p sk %p, nk %p", 6189 __func__, ctx->nr, ctx->sk, ctx->nk)); 6190 6191 /* Swap sk/nk for PF_OUT. */ 6192 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6193 (pd->dir == PF_IN) ? ctx->sk : ctx->nk, 6194 (pd->dir == PF_IN) ? ctx->nk : ctx->sk, s)) { 6195 REASON_SET(&ctx->reason, PFRES_STATEINS); 6196 goto drop; 6197 } else 6198 *sm = s; 6199 ctx->sk = ctx->nk = NULL; 6200 6201 STATE_INC_COUNTERS(s); 6202 6203 /* 6204 * Lock order is important: first state, then source node. 6205 */ 6206 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6207 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6208 s->sns[sn_type] = sns[sn_type]; 6209 PF_HASHROW_UNLOCK(snhs[sn_type]); 6210 } 6211 } 6212 6213 if (ctx->tag > 0) 6214 s->tag = ctx->tag; 6215 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6216 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 6217 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6218 pf_undo_nat(ctx->nr, pd, bip_sum); 6219 s->src.seqhi = arc4random(); 6220 /* Find mss option */ 6221 int rtid = M_GETFIB(pd->m); 6222 mss = pf_get_mss(pd); 6223 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6224 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6225 s->src.mss = mss; 6226 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6227 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6228 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6229 pd->act.rtableid); 6230 REASON_SET(&ctx->reason, PFRES_SYNPROXY); 6231 return (PF_SYNPROXY_DROP); 6232 } 6233 6234 s->udp_mapping = ctx->udp_mapping; 6235 6236 return (PF_PASS); 6237 6238 csfailed: 6239 while ((ri = SLIST_FIRST(&ctx->rules))) { 6240 SLIST_REMOVE_HEAD(&ctx->rules, entry); 6241 free(ri, M_PF_RULE_ITEM); 6242 } 6243 6244 uma_zfree(V_pf_state_key_z, ctx->sk); 6245 uma_zfree(V_pf_state_key_z, ctx->nk); 6246 6247 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6248 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6249 if (--sns[sn_type]->states == 0 && 6250 sns[sn_type]->expire == 0) { 6251 pf_unlink_src_node(sns[sn_type]); 6252 pf_free_src_node(sns[sn_type]); 6253 counter_u64_add( 6254 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6255 } 6256 PF_HASHROW_UNLOCK(snhs[sn_type]); 6257 } 6258 } 6259 6260 drop: 6261 if (s != NULL) { 6262 pf_src_tree_remove_state(s); 6263 s->timeout = PFTM_UNLINKED; 6264 pf_free_state(s); 6265 } 6266 6267 return (PF_DROP); 6268 } 6269 6270 int 6271 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6272 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6273 int icmp_dir) 6274 { 6275 /* 6276 * pf_translate() implements OpenBSD's "new" NAT approach. 6277 * We don't follow it, because it involves a breaking syntax change 6278 * (removing nat/rdr rules, moving it into regular pf rules.) 6279 * It also moves NAT processing to be done after normal rules evaluation 6280 * whereas in FreeBSD that's done before rules processing. 6281 * 6282 * We adopt the function only for nat64, and keep other NAT processing 6283 * before rules processing. 6284 */ 6285 int rewrite = 0; 6286 int afto = pd->af != pd->naf; 6287 6288 MPASS(afto); 6289 6290 switch (pd->proto) { 6291 case IPPROTO_TCP: 6292 case IPPROTO_UDP: 6293 case IPPROTO_SCTP: 6294 if (afto || *pd->sport != sport) { 6295 pf_change_ap(pd, pd->src, pd->sport, 6296 saddr, sport); 6297 rewrite = 1; 6298 } 6299 if (afto || *pd->dport != dport) { 6300 pf_change_ap(pd, pd->dst, pd->dport, 6301 daddr, dport); 6302 rewrite = 1; 6303 } 6304 break; 6305 6306 #ifdef INET 6307 case IPPROTO_ICMP: 6308 /* pf_translate() is also used when logging invalid packets */ 6309 if (pd->af != AF_INET) 6310 return (0); 6311 6312 if (afto) { 6313 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6314 return (-1); 6315 pd->proto = IPPROTO_ICMPV6; 6316 rewrite = 1; 6317 } 6318 if (virtual_type == htons(ICMP_ECHO)) { 6319 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6320 6321 if (icmpid != pd->hdr.icmp.icmp_id) { 6322 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6323 pd->hdr.icmp.icmp_cksum, 6324 pd->hdr.icmp.icmp_id, icmpid, 0); 6325 pd->hdr.icmp.icmp_id = icmpid; 6326 /* XXX TODO copyback. */ 6327 rewrite = 1; 6328 } 6329 } 6330 break; 6331 #endif /* INET */ 6332 6333 #ifdef INET6 6334 case IPPROTO_ICMPV6: 6335 /* pf_translate() is also used when logging invalid packets */ 6336 if (pd->af != AF_INET6) 6337 return (0); 6338 6339 if (afto) { 6340 /* ip_sum will be recalculated in pf_translate_af */ 6341 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6342 return (0); 6343 pd->proto = IPPROTO_ICMP; 6344 rewrite = 1; 6345 } 6346 break; 6347 #endif /* INET6 */ 6348 6349 default: 6350 break; 6351 } 6352 6353 return (rewrite); 6354 } 6355 6356 int 6357 pf_translate_compat(struct pf_test_ctx *ctx) 6358 { 6359 struct pf_pdesc *pd = ctx->pd; 6360 struct pf_state_key *nk = ctx->nk; 6361 struct tcphdr *th = &pd->hdr.tcp; 6362 int rewrite = 0; 6363 6364 KASSERT(ctx->sk != NULL, ("%s: null sk", __func__)); 6365 KASSERT(ctx->nk != NULL, ("%s: null nk", __func__)); 6366 6367 switch (pd->proto) { 6368 case IPPROTO_TCP: 6369 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6370 nk->port[pd->sidx] != pd->nsport) { 6371 pf_change_ap(pd, pd->src, &th->th_sport, 6372 &nk->addr[pd->sidx], nk->port[pd->sidx]); 6373 pd->sport = &th->th_sport; 6374 pd->nsport = th->th_sport; 6375 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6376 } 6377 6378 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6379 nk->port[pd->didx] != pd->ndport) { 6380 pf_change_ap(pd, pd->dst, &th->th_dport, 6381 &nk->addr[pd->didx], nk->port[pd->didx]); 6382 pd->dport = &th->th_dport; 6383 pd->ndport = th->th_dport; 6384 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6385 } 6386 rewrite++; 6387 break; 6388 case IPPROTO_UDP: 6389 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6390 nk->port[pd->sidx] != pd->nsport) { 6391 pf_change_ap(pd, pd->src, 6392 &pd->hdr.udp.uh_sport, 6393 &nk->addr[pd->sidx], 6394 nk->port[pd->sidx]); 6395 pd->sport = &pd->hdr.udp.uh_sport; 6396 pd->nsport = pd->hdr.udp.uh_sport; 6397 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6398 } 6399 6400 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6401 nk->port[pd->didx] != pd->ndport) { 6402 pf_change_ap(pd, pd->dst, 6403 &pd->hdr.udp.uh_dport, 6404 &nk->addr[pd->didx], 6405 nk->port[pd->didx]); 6406 pd->dport = &pd->hdr.udp.uh_dport; 6407 pd->ndport = pd->hdr.udp.uh_dport; 6408 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6409 } 6410 rewrite++; 6411 break; 6412 case IPPROTO_SCTP: { 6413 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 6414 nk->port[pd->sidx] != pd->nsport) { 6415 pf_change_ap(pd, pd->src, 6416 &pd->hdr.sctp.src_port, 6417 &nk->addr[pd->sidx], 6418 nk->port[pd->sidx]); 6419 pd->sport = &pd->hdr.sctp.src_port; 6420 pd->nsport = pd->hdr.sctp.src_port; 6421 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6422 } 6423 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 6424 nk->port[pd->didx] != pd->ndport) { 6425 pf_change_ap(pd, pd->dst, 6426 &pd->hdr.sctp.dest_port, 6427 &nk->addr[pd->didx], 6428 nk->port[pd->didx]); 6429 pd->dport = &pd->hdr.sctp.dest_port; 6430 pd->ndport = pd->hdr.sctp.dest_port; 6431 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6432 } 6433 break; 6434 } 6435 #ifdef INET 6436 case IPPROTO_ICMP: 6437 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 6438 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 6439 nk->addr[pd->sidx].v4.s_addr, 0); 6440 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6441 } 6442 6443 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 6444 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 6445 nk->addr[pd->didx].v4.s_addr, 0); 6446 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6447 } 6448 6449 if (ctx->virtual_type == htons(ICMP_ECHO) && 6450 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 6451 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6452 pd->hdr.icmp.icmp_cksum, pd->nsport, 6453 nk->port[pd->sidx], 0); 6454 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 6455 pd->sport = &pd->hdr.icmp.icmp_id; 6456 } 6457 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6458 break; 6459 #endif /* INET */ 6460 #ifdef INET6 6461 case IPPROTO_ICMPV6: 6462 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 6463 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 6464 &nk->addr[pd->sidx], 0); 6465 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6466 } 6467 6468 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 6469 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 6470 &nk->addr[pd->didx], 0); 6471 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6472 } 6473 rewrite++; 6474 break; 6475 #endif /* INET */ 6476 default: 6477 switch (pd->af) { 6478 #ifdef INET 6479 case AF_INET: 6480 if (PF_ANEQ(&pd->nsaddr, 6481 &nk->addr[pd->sidx], AF_INET)) { 6482 pf_change_a(&pd->src->v4.s_addr, 6483 pd->ip_sum, 6484 nk->addr[pd->sidx].v4.s_addr, 0); 6485 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6486 } 6487 6488 if (PF_ANEQ(&pd->ndaddr, 6489 &nk->addr[pd->didx], AF_INET)) { 6490 pf_change_a(&pd->dst->v4.s_addr, 6491 pd->ip_sum, 6492 nk->addr[pd->didx].v4.s_addr, 0); 6493 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6494 } 6495 break; 6496 #endif /* INET */ 6497 #ifdef INET6 6498 case AF_INET6: 6499 if (PF_ANEQ(&pd->nsaddr, 6500 &nk->addr[pd->sidx], AF_INET6)) { 6501 pf_addrcpy(&pd->nsaddr, &nk->addr[pd->sidx], 6502 pd->af); 6503 pf_addrcpy(pd->src, &nk->addr[pd->sidx], pd->af); 6504 } 6505 6506 if (PF_ANEQ(&pd->ndaddr, 6507 &nk->addr[pd->didx], AF_INET6)) { 6508 pf_addrcpy(&pd->ndaddr, &nk->addr[pd->didx], 6509 pd->af); 6510 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 6511 pd->af); 6512 } 6513 break; 6514 #endif /* INET6 */ 6515 } 6516 break; 6517 } 6518 return (rewrite); 6519 } 6520 6521 static int 6522 pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd, 6523 u_short *reason, int *copyback, struct pf_state_peer *src, 6524 struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst) 6525 { 6526 struct tcphdr *th = &pd->hdr.tcp; 6527 u_int16_t win = ntohs(th->th_win); 6528 u_int32_t ack, end, data_end, seq, orig_seq; 6529 u_int8_t sws, dws; 6530 int ackskew; 6531 6532 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6533 sws = src->wscale & PF_WSCALE_MASK; 6534 dws = dst->wscale & PF_WSCALE_MASK; 6535 } else 6536 sws = dws = 0; 6537 6538 /* 6539 * Sequence tracking algorithm from Guido van Rooij's paper: 6540 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6541 * tcp_filtering.ps 6542 */ 6543 6544 orig_seq = seq = ntohl(th->th_seq); 6545 if (src->seqlo == 0) { 6546 /* First packet from this end. Set its state */ 6547 6548 if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6549 src->scrub == NULL) { 6550 if (pf_normalize_tcp_init(pd, th, src)) { 6551 REASON_SET(reason, PFRES_MEMORY); 6552 return (PF_DROP); 6553 } 6554 } 6555 6556 /* Deferred generation of sequence number modulator */ 6557 if (dst->seqdiff && !src->seqdiff) { 6558 /* use random iss for the TCP server */ 6559 while ((src->seqdiff = arc4random() - seq) == 0) 6560 ; 6561 ack = ntohl(th->th_ack) - dst->seqdiff; 6562 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6563 src->seqdiff), 0); 6564 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6565 *copyback = 1; 6566 } else { 6567 ack = ntohl(th->th_ack); 6568 } 6569 6570 end = seq + pd->p_len; 6571 if (tcp_get_flags(th) & TH_SYN) { 6572 end++; 6573 if (dst->wscale & PF_WSCALE_FLAG) { 6574 src->wscale = pf_get_wscale(pd); 6575 if (src->wscale & PF_WSCALE_FLAG) { 6576 /* Remove scale factor from initial 6577 * window */ 6578 sws = src->wscale & PF_WSCALE_MASK; 6579 win = ((u_int32_t)win + (1 << sws) - 1) 6580 >> sws; 6581 dws = dst->wscale & PF_WSCALE_MASK; 6582 } else { 6583 /* fixup other window */ 6584 dst->max_win = MIN(TCP_MAXWIN, 6585 (u_int32_t)dst->max_win << 6586 (dst->wscale & PF_WSCALE_MASK)); 6587 /* in case of a retrans SYN|ACK */ 6588 dst->wscale = 0; 6589 } 6590 } 6591 } 6592 data_end = end; 6593 if (tcp_get_flags(th) & TH_FIN) 6594 end++; 6595 6596 src->seqlo = seq; 6597 if (src->state < TCPS_SYN_SENT) 6598 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6599 6600 /* 6601 * May need to slide the window (seqhi may have been set by 6602 * the crappy stack check or if we picked up the connection 6603 * after establishment) 6604 */ 6605 if (src->seqhi == 1 || 6606 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6607 src->seqhi = end + MAX(1, dst->max_win << dws); 6608 if (win > src->max_win) 6609 src->max_win = win; 6610 6611 } else { 6612 ack = ntohl(th->th_ack) - dst->seqdiff; 6613 if (src->seqdiff) { 6614 /* Modulate sequence numbers */ 6615 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6616 src->seqdiff), 0); 6617 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6618 *copyback = 1; 6619 } 6620 end = seq + pd->p_len; 6621 if (tcp_get_flags(th) & TH_SYN) 6622 end++; 6623 data_end = end; 6624 if (tcp_get_flags(th) & TH_FIN) 6625 end++; 6626 } 6627 6628 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6629 /* Let it pass through the ack skew check */ 6630 ack = dst->seqlo; 6631 } else if ((ack == 0 && 6632 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6633 /* broken tcp stacks do not set ack */ 6634 (dst->state < TCPS_SYN_SENT)) { 6635 /* 6636 * Many stacks (ours included) will set the ACK number in an 6637 * FIN|ACK if the SYN times out -- no sequence to ACK. 6638 */ 6639 ack = dst->seqlo; 6640 } 6641 6642 if (seq == end) { 6643 /* Ease sequencing restrictions on no data packets */ 6644 seq = src->seqlo; 6645 data_end = end = seq; 6646 } 6647 6648 ackskew = dst->seqlo - ack; 6649 6650 /* 6651 * Need to demodulate the sequence numbers in any TCP SACK options 6652 * (Selective ACK). We could optionally validate the SACK values 6653 * against the current ACK window, either forwards or backwards, but 6654 * I'm not confident that SACK has been implemented properly 6655 * everywhere. It wouldn't surprise me if several stacks accidentally 6656 * SACK too far backwards of previously ACKed data. There really aren't 6657 * any security implications of bad SACKing unless the target stack 6658 * doesn't validate the option length correctly. Someone trying to 6659 * spoof into a TCP connection won't bother blindly sending SACK 6660 * options anyway. 6661 */ 6662 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6663 if (pf_modulate_sack(pd, th, dst)) 6664 *copyback = 1; 6665 } 6666 6667 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6668 if (SEQ_GEQ(src->seqhi, data_end) && 6669 /* Last octet inside other's window space */ 6670 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6671 /* Retrans: not more than one window back */ 6672 (ackskew >= -MAXACKWINDOW) && 6673 /* Acking not more than one reassembled fragment backwards */ 6674 (ackskew <= (MAXACKWINDOW << sws)) && 6675 /* Acking not more than one window forward */ 6676 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6677 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 6678 /* Require an exact/+1 sequence match on resets when possible */ 6679 6680 if (dst->scrub || src->scrub) { 6681 if (pf_normalize_tcp_stateful(pd, reason, th, 6682 state, src, dst, copyback)) 6683 return (PF_DROP); 6684 } 6685 6686 /* update max window */ 6687 if (src->max_win < win) 6688 src->max_win = win; 6689 /* synchronize sequencing */ 6690 if (SEQ_GT(end, src->seqlo)) 6691 src->seqlo = end; 6692 /* slide the window of what the other end can send */ 6693 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6694 dst->seqhi = ack + MAX((win << sws), 1); 6695 6696 /* update states */ 6697 if (tcp_get_flags(th) & TH_SYN) 6698 if (src->state < TCPS_SYN_SENT) 6699 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6700 if (tcp_get_flags(th) & TH_FIN) 6701 if (src->state < TCPS_CLOSING) 6702 pf_set_protostate(state, psrc, TCPS_CLOSING); 6703 if (tcp_get_flags(th) & TH_ACK) { 6704 if (dst->state == TCPS_SYN_SENT) { 6705 pf_set_protostate(state, pdst, 6706 TCPS_ESTABLISHED); 6707 if (src->state == TCPS_ESTABLISHED && 6708 state->sns[PF_SN_LIMIT] != NULL && 6709 pf_src_connlimit(state)) { 6710 REASON_SET(reason, PFRES_SRCLIMIT); 6711 return (PF_DROP); 6712 } 6713 } else if (dst->state == TCPS_CLOSING) 6714 pf_set_protostate(state, pdst, 6715 TCPS_FIN_WAIT_2); 6716 } 6717 if (tcp_get_flags(th) & TH_RST) 6718 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6719 6720 /* update expire time */ 6721 state->expire = pf_get_uptime(); 6722 if (src->state >= TCPS_FIN_WAIT_2 && 6723 dst->state >= TCPS_FIN_WAIT_2) 6724 state->timeout = PFTM_TCP_CLOSED; 6725 else if (src->state >= TCPS_CLOSING && 6726 dst->state >= TCPS_CLOSING) 6727 state->timeout = PFTM_TCP_FIN_WAIT; 6728 else if (src->state < TCPS_ESTABLISHED || 6729 dst->state < TCPS_ESTABLISHED) 6730 state->timeout = PFTM_TCP_OPENING; 6731 else if (src->state >= TCPS_CLOSING || 6732 dst->state >= TCPS_CLOSING) 6733 state->timeout = PFTM_TCP_CLOSING; 6734 else 6735 state->timeout = PFTM_TCP_ESTABLISHED; 6736 6737 /* Fall through to PASS packet */ 6738 6739 } else if ((dst->state < TCPS_SYN_SENT || 6740 dst->state >= TCPS_FIN_WAIT_2 || 6741 src->state >= TCPS_FIN_WAIT_2) && 6742 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6743 /* Within a window forward of the originating packet */ 6744 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6745 /* Within a window backward of the originating packet */ 6746 6747 /* 6748 * This currently handles three situations: 6749 * 1) Stupid stacks will shotgun SYNs before their peer 6750 * replies. 6751 * 2) When PF catches an already established stream (the 6752 * firewall rebooted, the state table was flushed, routes 6753 * changed...) 6754 * 3) Packets get funky immediately after the connection 6755 * closes (this should catch Solaris spurious ACK|FINs 6756 * that web servers like to spew after a close) 6757 * 6758 * This must be a little more careful than the above code 6759 * since packet floods will also be caught here. We don't 6760 * update the TTL here to mitigate the damage of a packet 6761 * flood and so the same code can handle awkward establishment 6762 * and a loosened connection close. 6763 * In the establishment case, a correct peer response will 6764 * validate the connection, go through the normal state code 6765 * and keep updating the state TTL. 6766 */ 6767 6768 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6769 printf("pf: loose state match: "); 6770 pf_print_state(state); 6771 pf_print_flags(tcp_get_flags(th)); 6772 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6773 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6774 pd->p_len, ackskew, (unsigned long long)state->packets[0], 6775 (unsigned long long)state->packets[1], 6776 pd->dir == PF_IN ? "in" : "out", 6777 pd->dir == state->direction ? "fwd" : "rev"); 6778 } 6779 6780 if (dst->scrub || src->scrub) { 6781 if (pf_normalize_tcp_stateful(pd, reason, th, 6782 state, src, dst, copyback)) 6783 return (PF_DROP); 6784 } 6785 6786 /* update max window */ 6787 if (src->max_win < win) 6788 src->max_win = win; 6789 /* synchronize sequencing */ 6790 if (SEQ_GT(end, src->seqlo)) 6791 src->seqlo = end; 6792 /* slide the window of what the other end can send */ 6793 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6794 dst->seqhi = ack + MAX((win << sws), 1); 6795 6796 /* 6797 * Cannot set dst->seqhi here since this could be a shotgunned 6798 * SYN and not an already established connection. 6799 */ 6800 6801 if (tcp_get_flags(th) & TH_FIN) 6802 if (src->state < TCPS_CLOSING) 6803 pf_set_protostate(state, psrc, TCPS_CLOSING); 6804 if (tcp_get_flags(th) & TH_RST) 6805 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6806 6807 /* Fall through to PASS packet */ 6808 6809 } else { 6810 if (state->dst.state == TCPS_SYN_SENT && 6811 state->src.state == TCPS_SYN_SENT) { 6812 /* Send RST for state mismatches during handshake */ 6813 if (!(tcp_get_flags(th) & TH_RST)) 6814 pf_send_tcp(state->rule, pd->af, 6815 pd->dst, pd->src, th->th_dport, 6816 th->th_sport, ntohl(th->th_ack), 0, 6817 TH_RST, 0, 0, 6818 state->rule->return_ttl, M_SKIP_FIREWALL, 6819 0, 0, state->act.rtableid); 6820 src->seqlo = 0; 6821 src->seqhi = 1; 6822 src->max_win = 1; 6823 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6824 printf("pf: BAD state: "); 6825 pf_print_state(state); 6826 pf_print_flags(tcp_get_flags(th)); 6827 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6828 "pkts=%llu:%llu dir=%s,%s\n", 6829 seq, orig_seq, ack, pd->p_len, ackskew, 6830 (unsigned long long)state->packets[0], 6831 (unsigned long long)state->packets[1], 6832 pd->dir == PF_IN ? "in" : "out", 6833 pd->dir == state->direction ? "fwd" : "rev"); 6834 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6835 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6836 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6837 ' ': '2', 6838 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6839 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6840 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6841 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6842 } 6843 REASON_SET(reason, PFRES_BADSTATE); 6844 return (PF_DROP); 6845 } 6846 6847 return (PF_PASS); 6848 } 6849 6850 static int 6851 pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd, 6852 u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst, 6853 u_int8_t psrc, u_int8_t pdst) 6854 { 6855 struct tcphdr *th = &pd->hdr.tcp; 6856 6857 if (tcp_get_flags(th) & TH_SYN) 6858 if (src->state < TCPS_SYN_SENT) 6859 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6860 if (tcp_get_flags(th) & TH_FIN) 6861 if (src->state < TCPS_CLOSING) 6862 pf_set_protostate(state, psrc, TCPS_CLOSING); 6863 if (tcp_get_flags(th) & TH_ACK) { 6864 if (dst->state == TCPS_SYN_SENT) { 6865 pf_set_protostate(state, pdst, TCPS_ESTABLISHED); 6866 if (src->state == TCPS_ESTABLISHED && 6867 state->sns[PF_SN_LIMIT] != NULL && 6868 pf_src_connlimit(state)) { 6869 REASON_SET(reason, PFRES_SRCLIMIT); 6870 return (PF_DROP); 6871 } 6872 } else if (dst->state == TCPS_CLOSING) { 6873 pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2); 6874 } else if (src->state == TCPS_SYN_SENT && 6875 dst->state < TCPS_SYN_SENT) { 6876 /* 6877 * Handle a special sloppy case where we only see one 6878 * half of the connection. If there is a ACK after 6879 * the initial SYN without ever seeing a packet from 6880 * the destination, set the connection to established. 6881 */ 6882 pf_set_protostate(state, PF_PEER_BOTH, 6883 TCPS_ESTABLISHED); 6884 dst->state = src->state = TCPS_ESTABLISHED; 6885 if (state->sns[PF_SN_LIMIT] != NULL && 6886 pf_src_connlimit(state)) { 6887 REASON_SET(reason, PFRES_SRCLIMIT); 6888 return (PF_DROP); 6889 } 6890 } else if (src->state == TCPS_CLOSING && 6891 dst->state == TCPS_ESTABLISHED && 6892 dst->seqlo == 0) { 6893 /* 6894 * Handle the closing of half connections where we 6895 * don't see the full bidirectional FIN/ACK+ACK 6896 * handshake. 6897 */ 6898 pf_set_protostate(state, pdst, TCPS_CLOSING); 6899 } 6900 } 6901 if (tcp_get_flags(th) & TH_RST) 6902 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6903 6904 /* update expire time */ 6905 state->expire = pf_get_uptime(); 6906 if (src->state >= TCPS_FIN_WAIT_2 && 6907 dst->state >= TCPS_FIN_WAIT_2) 6908 state->timeout = PFTM_TCP_CLOSED; 6909 else if (src->state >= TCPS_CLOSING && 6910 dst->state >= TCPS_CLOSING) 6911 state->timeout = PFTM_TCP_FIN_WAIT; 6912 else if (src->state < TCPS_ESTABLISHED || 6913 dst->state < TCPS_ESTABLISHED) 6914 state->timeout = PFTM_TCP_OPENING; 6915 else if (src->state >= TCPS_CLOSING || 6916 dst->state >= TCPS_CLOSING) 6917 state->timeout = PFTM_TCP_CLOSING; 6918 else 6919 state->timeout = PFTM_TCP_ESTABLISHED; 6920 6921 return (PF_PASS); 6922 } 6923 6924 static int 6925 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason) 6926 { 6927 struct pf_state_key *sk = state->key[pd->didx]; 6928 struct tcphdr *th = &pd->hdr.tcp; 6929 6930 if (state->src.state == PF_TCPS_PROXY_SRC) { 6931 if (pd->dir != state->direction) { 6932 REASON_SET(reason, PFRES_SYNPROXY); 6933 return (PF_SYNPROXY_DROP); 6934 } 6935 if (tcp_get_flags(th) & TH_SYN) { 6936 if (ntohl(th->th_seq) != state->src.seqlo) { 6937 REASON_SET(reason, PFRES_SYNPROXY); 6938 return (PF_DROP); 6939 } 6940 pf_send_tcp(state->rule, pd->af, pd->dst, 6941 pd->src, th->th_dport, th->th_sport, 6942 state->src.seqhi, ntohl(th->th_seq) + 1, 6943 TH_SYN|TH_ACK, 0, state->src.mss, 0, 6944 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6945 REASON_SET(reason, PFRES_SYNPROXY); 6946 return (PF_SYNPROXY_DROP); 6947 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6948 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6949 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6950 REASON_SET(reason, PFRES_SYNPROXY); 6951 return (PF_DROP); 6952 } else if (state->sns[PF_SN_LIMIT] != NULL && 6953 pf_src_connlimit(state)) { 6954 REASON_SET(reason, PFRES_SRCLIMIT); 6955 return (PF_DROP); 6956 } else 6957 pf_set_protostate(state, PF_PEER_SRC, 6958 PF_TCPS_PROXY_DST); 6959 } 6960 if (state->src.state == PF_TCPS_PROXY_DST) { 6961 if (pd->dir == state->direction) { 6962 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 6963 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6964 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6965 REASON_SET(reason, PFRES_SYNPROXY); 6966 return (PF_DROP); 6967 } 6968 state->src.max_win = MAX(ntohs(th->th_win), 1); 6969 if (state->dst.seqhi == 1) 6970 state->dst.seqhi = arc4random(); 6971 pf_send_tcp(state->rule, pd->af, 6972 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6973 sk->port[pd->sidx], sk->port[pd->didx], 6974 state->dst.seqhi, 0, TH_SYN, 0, 6975 state->src.mss, 0, 6976 state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6977 state->tag, 0, state->act.rtableid); 6978 REASON_SET(reason, PFRES_SYNPROXY); 6979 return (PF_SYNPROXY_DROP); 6980 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 6981 (TH_SYN|TH_ACK)) || 6982 (ntohl(th->th_ack) != state->dst.seqhi + 1)) { 6983 REASON_SET(reason, PFRES_SYNPROXY); 6984 return (PF_DROP); 6985 } else { 6986 state->dst.max_win = MAX(ntohs(th->th_win), 1); 6987 state->dst.seqlo = ntohl(th->th_seq); 6988 pf_send_tcp(state->rule, pd->af, pd->dst, 6989 pd->src, th->th_dport, th->th_sport, 6990 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6991 TH_ACK, state->src.max_win, 0, 0, 0, 6992 state->tag, 0, state->act.rtableid); 6993 pf_send_tcp(state->rule, pd->af, 6994 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6995 sk->port[pd->sidx], sk->port[pd->didx], 6996 state->src.seqhi + 1, state->src.seqlo + 1, 6997 TH_ACK, state->dst.max_win, 0, 0, 6998 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6999 state->src.seqdiff = state->dst.seqhi - 7000 state->src.seqlo; 7001 state->dst.seqdiff = state->src.seqhi - 7002 state->dst.seqlo; 7003 state->src.seqhi = state->src.seqlo + 7004 state->dst.max_win; 7005 state->dst.seqhi = state->dst.seqlo + 7006 state->src.max_win; 7007 state->src.wscale = state->dst.wscale = 0; 7008 pf_set_protostate(state, PF_PEER_BOTH, 7009 TCPS_ESTABLISHED); 7010 REASON_SET(reason, PFRES_SYNPROXY); 7011 return (PF_SYNPROXY_DROP); 7012 } 7013 } 7014 7015 return (PF_PASS); 7016 } 7017 7018 static int 7019 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 7020 { 7021 struct pf_state_key_cmp key; 7022 int copyback = 0; 7023 struct pf_state_peer *src, *dst; 7024 uint8_t psrc, pdst; 7025 int action; 7026 7027 bzero(&key, sizeof(key)); 7028 key.af = pd->af; 7029 key.proto = pd->virtual_proto; 7030 pf_addrcpy(&key.addr[pd->sidx], pd->src, key.af); 7031 pf_addrcpy(&key.addr[pd->didx], pd->dst, key.af); 7032 key.port[pd->sidx] = pd->osport; 7033 key.port[pd->didx] = pd->odport; 7034 7035 action = pf_find_state(pd, &key, state); 7036 if (action != PF_MATCH) 7037 return (action); 7038 7039 action = PF_PASS; 7040 if (pd->dir == (*state)->direction) { 7041 if (PF_REVERSED_KEY(*state, pd->af)) { 7042 src = &(*state)->dst; 7043 dst = &(*state)->src; 7044 psrc = PF_PEER_DST; 7045 pdst = PF_PEER_SRC; 7046 } else { 7047 src = &(*state)->src; 7048 dst = &(*state)->dst; 7049 psrc = PF_PEER_SRC; 7050 pdst = PF_PEER_DST; 7051 } 7052 } else { 7053 if (PF_REVERSED_KEY(*state, pd->af)) { 7054 src = &(*state)->src; 7055 dst = &(*state)->dst; 7056 psrc = PF_PEER_SRC; 7057 pdst = PF_PEER_DST; 7058 } else { 7059 src = &(*state)->dst; 7060 dst = &(*state)->src; 7061 psrc = PF_PEER_DST; 7062 pdst = PF_PEER_SRC; 7063 } 7064 } 7065 7066 switch (pd->virtual_proto) { 7067 case IPPROTO_TCP: { 7068 struct tcphdr *th = &pd->hdr.tcp; 7069 7070 if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS) 7071 return (action); 7072 if (((tcp_get_flags(th) & (TH_SYN | TH_ACK)) == TH_SYN) || 7073 ((th->th_flags & (TH_SYN | TH_ACK | TH_RST)) == TH_ACK && 7074 pf_syncookie_check(pd) && pd->dir == PF_IN)) { 7075 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 7076 (*state)->dst.state >= TCPS_FIN_WAIT_2) { 7077 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7078 printf("pf: state reuse "); 7079 pf_print_state(*state); 7080 pf_print_flags(tcp_get_flags(th)); 7081 printf("\n"); 7082 } 7083 /* XXX make sure it's the same direction ?? */ 7084 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 7085 pf_remove_state(*state); 7086 *state = NULL; 7087 return (PF_DROP); 7088 } else if ((*state)->src.state >= TCPS_ESTABLISHED && 7089 (*state)->dst.state >= TCPS_ESTABLISHED) { 7090 /* 7091 * SYN matches existing state??? 7092 * Typically happens when sender boots up after 7093 * sudden panic. Certain protocols (NFSv3) are 7094 * always using same port numbers. Challenge 7095 * ACK enables all parties (firewall and peers) 7096 * to get in sync again. 7097 */ 7098 pf_send_challenge_ack(pd, *state, src, dst); 7099 return (PF_DROP); 7100 } 7101 } 7102 if ((*state)->state_flags & PFSTATE_SLOPPY) { 7103 if (pf_tcp_track_sloppy(*state, pd, reason, src, dst, 7104 psrc, pdst) == PF_DROP) 7105 return (PF_DROP); 7106 } else { 7107 int ret; 7108 7109 ret = pf_tcp_track_full(*state, pd, reason, 7110 ©back, src, dst, psrc, pdst); 7111 if (ret == PF_DROP) 7112 return (PF_DROP); 7113 } 7114 break; 7115 } 7116 case IPPROTO_UDP: 7117 /* update states */ 7118 if (src->state < PFUDPS_SINGLE) 7119 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 7120 if (dst->state == PFUDPS_SINGLE) 7121 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 7122 7123 /* update expire time */ 7124 (*state)->expire = pf_get_uptime(); 7125 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 7126 (*state)->timeout = PFTM_UDP_MULTIPLE; 7127 else 7128 (*state)->timeout = PFTM_UDP_SINGLE; 7129 break; 7130 case IPPROTO_SCTP: 7131 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 7132 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 7133 pd->sctp_flags & PFDESC_SCTP_INIT) { 7134 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7135 pf_remove_state(*state); 7136 *state = NULL; 7137 return (PF_DROP); 7138 } 7139 7140 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7141 return (PF_DROP); 7142 7143 /* Track state. */ 7144 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7145 if (src->state < SCTP_COOKIE_WAIT) { 7146 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7147 (*state)->timeout = PFTM_SCTP_OPENING; 7148 } 7149 } 7150 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7151 MPASS(dst->scrub != NULL); 7152 if (dst->scrub->pfss_v_tag == 0) 7153 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7154 } 7155 7156 /* 7157 * Bind to the correct interface if we're if-bound. For multihomed 7158 * extra associations we don't know which interface that will be until 7159 * here, so we've inserted the state on V_pf_all. Fix that now. 7160 */ 7161 if ((*state)->kif == V_pfi_all && 7162 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7163 (*state)->kif = pd->kif; 7164 7165 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7166 if (src->state < SCTP_ESTABLISHED) { 7167 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7168 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7169 } 7170 } 7171 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7172 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7173 if (src->state < SCTP_SHUTDOWN_PENDING) { 7174 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7175 (*state)->timeout = PFTM_SCTP_CLOSING; 7176 } 7177 } 7178 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7179 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7180 (*state)->timeout = PFTM_SCTP_CLOSED; 7181 } 7182 7183 (*state)->expire = pf_get_uptime(); 7184 break; 7185 default: 7186 /* update states */ 7187 if (src->state < PFOTHERS_SINGLE) 7188 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7189 if (dst->state == PFOTHERS_SINGLE) 7190 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7191 7192 /* update expire time */ 7193 (*state)->expire = pf_get_uptime(); 7194 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7195 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7196 else 7197 (*state)->timeout = PFTM_OTHER_SINGLE; 7198 break; 7199 } 7200 7201 /* translate source/destination address, if necessary */ 7202 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7203 struct pf_state_key *nk; 7204 int afto, sidx, didx; 7205 7206 if (PF_REVERSED_KEY(*state, pd->af)) 7207 nk = (*state)->key[pd->sidx]; 7208 else 7209 nk = (*state)->key[pd->didx]; 7210 7211 afto = pd->af != nk->af; 7212 7213 if (afto && (*state)->direction == PF_IN) { 7214 sidx = pd->didx; 7215 didx = pd->sidx; 7216 } else { 7217 sidx = pd->sidx; 7218 didx = pd->didx; 7219 } 7220 7221 if (afto) { 7222 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 7223 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 7224 pd->naf = nk->af; 7225 action = PF_AFRT; 7226 } 7227 7228 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7229 nk->port[sidx] != pd->osport) 7230 pf_change_ap(pd, pd->src, pd->sport, 7231 &nk->addr[sidx], nk->port[sidx]); 7232 7233 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7234 nk->port[didx] != pd->odport) 7235 pf_change_ap(pd, pd->dst, pd->dport, 7236 &nk->addr[didx], nk->port[didx]); 7237 7238 copyback = 1; 7239 } 7240 7241 if (copyback && pd->hdrlen > 0) 7242 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 7243 7244 return (action); 7245 } 7246 7247 static int 7248 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7249 u_short *reason) 7250 { 7251 struct pf_state_peer *src; 7252 if (pd->dir == state->direction) { 7253 if (PF_REVERSED_KEY(state, pd->af)) 7254 src = &state->dst; 7255 else 7256 src = &state->src; 7257 } else { 7258 if (PF_REVERSED_KEY(state, pd->af)) 7259 src = &state->src; 7260 else 7261 src = &state->dst; 7262 } 7263 7264 if (src->scrub != NULL) { 7265 if (src->scrub->pfss_v_tag == 0) 7266 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7267 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7268 return (PF_DROP); 7269 } 7270 7271 return (PF_PASS); 7272 } 7273 7274 static void 7275 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7276 { 7277 struct pf_sctp_endpoint key; 7278 struct pf_sctp_endpoint *ep; 7279 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7280 struct pf_sctp_source *i, *tmp; 7281 7282 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7283 return; 7284 7285 PF_SCTP_ENDPOINTS_LOCK(); 7286 7287 key.v_tag = s->dst.scrub->pfss_v_tag; 7288 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7289 if (ep != NULL) { 7290 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7291 if (pf_addr_cmp(&i->addr, 7292 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7293 s->key[PF_SK_WIRE]->af) == 0) { 7294 SDT_PROBE3(pf, sctp, multihome, remove, 7295 key.v_tag, s, i); 7296 TAILQ_REMOVE(&ep->sources, i, entry); 7297 free(i, M_PFTEMP); 7298 break; 7299 } 7300 } 7301 7302 if (TAILQ_EMPTY(&ep->sources)) { 7303 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7304 free(ep, M_PFTEMP); 7305 } 7306 } 7307 7308 /* Other direction. */ 7309 key.v_tag = s->src.scrub->pfss_v_tag; 7310 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7311 if (ep != NULL) { 7312 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7313 if (pf_addr_cmp(&i->addr, 7314 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7315 s->key[PF_SK_WIRE]->af) == 0) { 7316 SDT_PROBE3(pf, sctp, multihome, remove, 7317 key.v_tag, s, i); 7318 TAILQ_REMOVE(&ep->sources, i, entry); 7319 free(i, M_PFTEMP); 7320 break; 7321 } 7322 } 7323 7324 if (TAILQ_EMPTY(&ep->sources)) { 7325 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7326 free(ep, M_PFTEMP); 7327 } 7328 } 7329 7330 PF_SCTP_ENDPOINTS_UNLOCK(); 7331 } 7332 7333 static void 7334 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7335 { 7336 struct pf_sctp_endpoint key = { 7337 .v_tag = v_tag, 7338 }; 7339 struct pf_sctp_source *i; 7340 struct pf_sctp_endpoint *ep; 7341 int count; 7342 7343 PF_SCTP_ENDPOINTS_LOCK(); 7344 7345 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7346 if (ep == NULL) { 7347 ep = malloc(sizeof(struct pf_sctp_endpoint), 7348 M_PFTEMP, M_NOWAIT); 7349 if (ep == NULL) { 7350 PF_SCTP_ENDPOINTS_UNLOCK(); 7351 return; 7352 } 7353 7354 ep->v_tag = v_tag; 7355 TAILQ_INIT(&ep->sources); 7356 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7357 } 7358 7359 /* Avoid inserting duplicates. */ 7360 count = 0; 7361 TAILQ_FOREACH(i, &ep->sources, entry) { 7362 count++; 7363 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7364 PF_SCTP_ENDPOINTS_UNLOCK(); 7365 return; 7366 } 7367 } 7368 7369 /* Limit the number of addresses per endpoint. */ 7370 if (count >= PF_SCTP_MAX_ENDPOINTS) { 7371 PF_SCTP_ENDPOINTS_UNLOCK(); 7372 return; 7373 } 7374 7375 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7376 if (i == NULL) { 7377 PF_SCTP_ENDPOINTS_UNLOCK(); 7378 return; 7379 } 7380 7381 i->af = pd->af; 7382 memcpy(&i->addr, a, sizeof(*a)); 7383 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7384 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7385 7386 PF_SCTP_ENDPOINTS_UNLOCK(); 7387 } 7388 7389 static void 7390 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7391 struct pf_kstate *s, int action) 7392 { 7393 struct pf_sctp_multihome_job *j, *tmp; 7394 struct pf_sctp_source *i; 7395 int ret __unused; 7396 struct pf_kstate *sm = NULL; 7397 struct pf_krule *ra = NULL; 7398 struct pf_krule *r = &V_pf_default_rule; 7399 struct pf_kruleset *rs = NULL; 7400 u_short reason; 7401 bool do_extra = true; 7402 7403 PF_RULES_RLOCK_TRACKER; 7404 7405 again: 7406 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7407 if (s == NULL || action != PF_PASS) 7408 goto free; 7409 7410 /* Confirm we don't recurse here. */ 7411 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7412 7413 switch (j->op) { 7414 case SCTP_ADD_IP_ADDRESS: { 7415 uint32_t v_tag = pd->sctp_initiate_tag; 7416 7417 if (v_tag == 0) { 7418 if (s->direction == pd->dir) 7419 v_tag = s->src.scrub->pfss_v_tag; 7420 else 7421 v_tag = s->dst.scrub->pfss_v_tag; 7422 } 7423 7424 /* 7425 * Avoid duplicating states. We'll already have 7426 * created a state based on the source address of 7427 * the packet, but SCTP endpoints may also list this 7428 * address again in the INIT(_ACK) parameters. 7429 */ 7430 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7431 break; 7432 } 7433 7434 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7435 PF_RULES_RLOCK(); 7436 sm = NULL; 7437 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7438 j->pd.related_rule = s->rule; 7439 } 7440 ret = pf_test_rule(&r, &sm, 7441 &j->pd, &ra, &rs, &reason, NULL); 7442 PF_RULES_RUNLOCK(); 7443 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7444 if (ret != PF_DROP && sm != NULL) { 7445 /* Inherit v_tag values. */ 7446 if (sm->direction == s->direction) { 7447 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7448 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7449 } else { 7450 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7451 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7452 } 7453 PF_STATE_UNLOCK(sm); 7454 } else { 7455 /* If we try duplicate inserts? */ 7456 break; 7457 } 7458 7459 /* Only add the address if we've actually allowed the state. */ 7460 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7461 7462 if (! do_extra) { 7463 break; 7464 } 7465 /* 7466 * We need to do this for each of our source addresses. 7467 * Find those based on the verification tag. 7468 */ 7469 struct pf_sctp_endpoint key = { 7470 .v_tag = pd->hdr.sctp.v_tag, 7471 }; 7472 struct pf_sctp_endpoint *ep; 7473 7474 PF_SCTP_ENDPOINTS_LOCK(); 7475 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7476 if (ep == NULL) { 7477 PF_SCTP_ENDPOINTS_UNLOCK(); 7478 break; 7479 } 7480 MPASS(ep != NULL); 7481 7482 TAILQ_FOREACH(i, &ep->sources, entry) { 7483 struct pf_sctp_multihome_job *nj; 7484 7485 /* SCTP can intermingle IPv4 and IPv6. */ 7486 if (i->af != pd->af) 7487 continue; 7488 7489 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7490 if (! nj) { 7491 continue; 7492 } 7493 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7494 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7495 nj->pd.src = &nj->src; 7496 // New destination address! 7497 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7498 nj->pd.dst = &nj->dst; 7499 nj->pd.m = j->pd.m; 7500 nj->op = j->op; 7501 7502 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7503 } 7504 PF_SCTP_ENDPOINTS_UNLOCK(); 7505 7506 break; 7507 } 7508 case SCTP_DEL_IP_ADDRESS: { 7509 struct pf_state_key_cmp key; 7510 uint8_t psrc; 7511 int action; 7512 7513 bzero(&key, sizeof(key)); 7514 key.af = j->pd.af; 7515 key.proto = IPPROTO_SCTP; 7516 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7517 pf_addrcpy(&key.addr[0], j->pd.src, key.af); 7518 pf_addrcpy(&key.addr[1], j->pd.dst, key.af); 7519 key.port[0] = j->pd.hdr.sctp.src_port; 7520 key.port[1] = j->pd.hdr.sctp.dest_port; 7521 } else { /* stack side, reverse */ 7522 pf_addrcpy(&key.addr[1], j->pd.src, key.af); 7523 pf_addrcpy(&key.addr[0], j->pd.dst, key.af); 7524 key.port[1] = j->pd.hdr.sctp.src_port; 7525 key.port[0] = j->pd.hdr.sctp.dest_port; 7526 } 7527 7528 action = pf_find_state(&j->pd, &key, &sm); 7529 if (action == PF_MATCH) { 7530 PF_STATE_LOCK_ASSERT(sm); 7531 if (j->pd.dir == sm->direction) { 7532 psrc = PF_PEER_SRC; 7533 } else { 7534 psrc = PF_PEER_DST; 7535 } 7536 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7537 sm->timeout = PFTM_SCTP_CLOSING; 7538 PF_STATE_UNLOCK(sm); 7539 } 7540 break; 7541 default: 7542 panic("Unknown op %#x", j->op); 7543 } 7544 } 7545 7546 free: 7547 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7548 free(j, M_PFTEMP); 7549 } 7550 7551 /* We may have inserted extra work while processing the list. */ 7552 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7553 do_extra = false; 7554 goto again; 7555 } 7556 } 7557 7558 static int 7559 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7560 { 7561 int off = 0; 7562 struct pf_sctp_multihome_job *job; 7563 7564 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7565 7566 while (off < len) { 7567 struct sctp_paramhdr h; 7568 7569 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 7570 pd->af)) 7571 return (PF_DROP); 7572 7573 /* Parameters are at least 4 bytes. */ 7574 if (ntohs(h.param_length) < 4) 7575 return (PF_DROP); 7576 7577 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7578 ntohs(h.param_length)); 7579 7580 switch (ntohs(h.param_type)) { 7581 case SCTP_IPV4_ADDRESS: { 7582 struct in_addr t; 7583 7584 if (ntohs(h.param_length) != 7585 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7586 return (PF_DROP); 7587 7588 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7589 NULL, NULL, pd->af)) 7590 return (PF_DROP); 7591 7592 if (in_nullhost(t)) 7593 t.s_addr = pd->src->v4.s_addr; 7594 7595 /* 7596 * We hold the state lock (idhash) here, which means 7597 * that we can't acquire the keyhash, or we'll get a 7598 * LOR (and potentially double-lock things too). We also 7599 * can't release the state lock here, so instead we'll 7600 * enqueue this for async handling. 7601 * There's a relatively small race here, in that a 7602 * packet using the new addresses could arrive already, 7603 * but that's just though luck for it. 7604 */ 7605 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7606 if (! job) 7607 return (PF_DROP); 7608 7609 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7610 7611 memcpy(&job->pd, pd, sizeof(*pd)); 7612 7613 // New source address! 7614 memcpy(&job->src, &t, sizeof(t)); 7615 job->pd.src = &job->src; 7616 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7617 job->pd.dst = &job->dst; 7618 job->pd.m = pd->m; 7619 job->op = op; 7620 7621 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7622 break; 7623 } 7624 #ifdef INET6 7625 case SCTP_IPV6_ADDRESS: { 7626 struct in6_addr t; 7627 7628 if (ntohs(h.param_length) != 7629 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7630 return (PF_DROP); 7631 7632 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7633 NULL, NULL, pd->af)) 7634 return (PF_DROP); 7635 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7636 break; 7637 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7638 memcpy(&t, &pd->src->v6, sizeof(t)); 7639 7640 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7641 if (! job) 7642 return (PF_DROP); 7643 7644 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7645 7646 memcpy(&job->pd, pd, sizeof(*pd)); 7647 memcpy(&job->src, &t, sizeof(t)); 7648 job->pd.src = &job->src; 7649 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7650 job->pd.dst = &job->dst; 7651 job->pd.m = pd->m; 7652 job->op = op; 7653 7654 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7655 break; 7656 } 7657 #endif /* INET6 */ 7658 case SCTP_ADD_IP_ADDRESS: { 7659 int ret; 7660 struct sctp_asconf_paramhdr ah; 7661 7662 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7663 NULL, NULL, pd->af)) 7664 return (PF_DROP); 7665 7666 ret = pf_multihome_scan(start + off + sizeof(ah), 7667 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7668 SCTP_ADD_IP_ADDRESS); 7669 if (ret != PF_PASS) 7670 return (ret); 7671 break; 7672 } 7673 case SCTP_DEL_IP_ADDRESS: { 7674 int ret; 7675 struct sctp_asconf_paramhdr ah; 7676 7677 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7678 NULL, NULL, pd->af)) 7679 return (PF_DROP); 7680 ret = pf_multihome_scan(start + off + sizeof(ah), 7681 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7682 SCTP_DEL_IP_ADDRESS); 7683 if (ret != PF_PASS) 7684 return (ret); 7685 break; 7686 } 7687 default: 7688 break; 7689 } 7690 7691 off += roundup(ntohs(h.param_length), 4); 7692 } 7693 7694 return (PF_PASS); 7695 } 7696 7697 int 7698 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7699 { 7700 start += sizeof(struct sctp_init_chunk); 7701 len -= sizeof(struct sctp_init_chunk); 7702 7703 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7704 } 7705 7706 int 7707 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7708 { 7709 start += sizeof(struct sctp_asconf_chunk); 7710 len -= sizeof(struct sctp_asconf_chunk); 7711 7712 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7713 } 7714 7715 int 7716 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7717 struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir, 7718 int *iidx, int multi, int inner) 7719 { 7720 int action, direction = pd->dir; 7721 7722 key->af = pd->af; 7723 key->proto = pd->proto; 7724 if (icmp_dir == PF_IN) { 7725 *iidx = pd->sidx; 7726 key->port[pd->sidx] = icmpid; 7727 key->port[pd->didx] = type; 7728 } else { 7729 *iidx = pd->didx; 7730 key->port[pd->sidx] = type; 7731 key->port[pd->didx] = icmpid; 7732 } 7733 if (pf_state_key_addr_setup(pd, key, multi)) 7734 return (PF_DROP); 7735 7736 action = pf_find_state(pd, key, state); 7737 if (action != PF_MATCH) 7738 return (action); 7739 7740 if ((*state)->state_flags & PFSTATE_SLOPPY) 7741 return (-1); 7742 7743 /* Is this ICMP message flowing in right direction? */ 7744 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7745 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7746 PF_IN : PF_OUT; 7747 else 7748 direction = (*state)->direction; 7749 if ((*state)->rule->type && 7750 (((!inner && direction == pd->dir) || 7751 (inner && direction != pd->dir)) ? 7752 PF_IN : PF_OUT) != icmp_dir) { 7753 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7754 printf("pf: icmp type %d in wrong direction (%d): ", 7755 ntohs(type), icmp_dir); 7756 pf_print_state(*state); 7757 printf("\n"); 7758 } 7759 PF_STATE_UNLOCK(*state); 7760 *state = NULL; 7761 return (PF_DROP); 7762 } 7763 return (-1); 7764 } 7765 7766 static int 7767 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7768 u_short *reason) 7769 { 7770 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7771 u_int16_t *icmpsum, virtual_id, virtual_type; 7772 u_int8_t icmptype, icmpcode; 7773 int icmp_dir, iidx, ret; 7774 struct pf_state_key_cmp key; 7775 #ifdef INET 7776 u_int16_t icmpid; 7777 #endif /* INET*/ 7778 7779 MPASS(*state == NULL); 7780 7781 bzero(&key, sizeof(key)); 7782 switch (pd->proto) { 7783 #ifdef INET 7784 case IPPROTO_ICMP: 7785 icmptype = pd->hdr.icmp.icmp_type; 7786 icmpcode = pd->hdr.icmp.icmp_code; 7787 icmpid = pd->hdr.icmp.icmp_id; 7788 icmpsum = &pd->hdr.icmp.icmp_cksum; 7789 break; 7790 #endif /* INET */ 7791 #ifdef INET6 7792 case IPPROTO_ICMPV6: 7793 icmptype = pd->hdr.icmp6.icmp6_type; 7794 icmpcode = pd->hdr.icmp6.icmp6_code; 7795 #ifdef INET 7796 icmpid = pd->hdr.icmp6.icmp6_id; 7797 #endif /* INET */ 7798 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7799 break; 7800 #endif /* INET6 */ 7801 default: 7802 panic("unhandled proto %d", pd->proto); 7803 } 7804 7805 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 7806 &virtual_type) == 0) { 7807 /* 7808 * ICMP query/reply message not related to a TCP/UDP/SCTP 7809 * packet. Search for an ICMP state. 7810 */ 7811 ret = pf_icmp_state_lookup(&key, pd, state, virtual_id, 7812 virtual_type, icmp_dir, &iidx, 0, 0); 7813 /* IPv6? try matching a multicast address */ 7814 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) { 7815 MPASS(*state == NULL); 7816 ret = pf_icmp_state_lookup(&key, pd, state, 7817 virtual_id, virtual_type, 7818 icmp_dir, &iidx, 1, 0); 7819 } 7820 if (ret >= 0) { 7821 MPASS(*state == NULL); 7822 return (ret); 7823 } 7824 7825 (*state)->expire = pf_get_uptime(); 7826 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7827 7828 /* translate source/destination address, if necessary */ 7829 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7830 struct pf_state_key *nk; 7831 int afto, sidx, didx; 7832 7833 if (PF_REVERSED_KEY(*state, pd->af)) 7834 nk = (*state)->key[pd->sidx]; 7835 else 7836 nk = (*state)->key[pd->didx]; 7837 7838 afto = pd->af != nk->af; 7839 7840 if (afto && (*state)->direction == PF_IN) { 7841 sidx = pd->didx; 7842 didx = pd->sidx; 7843 iidx = !iidx; 7844 } else { 7845 sidx = pd->sidx; 7846 didx = pd->didx; 7847 } 7848 7849 switch (pd->af) { 7850 #ifdef INET 7851 case AF_INET: 7852 #ifdef INET6 7853 if (afto) { 7854 if (pf_translate_icmp_af(AF_INET6, 7855 &pd->hdr.icmp)) 7856 return (PF_DROP); 7857 pd->proto = IPPROTO_ICMPV6; 7858 } 7859 #endif /* INET6 */ 7860 if (!afto && 7861 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 7862 pf_change_a(&saddr->v4.s_addr, 7863 pd->ip_sum, 7864 nk->addr[sidx].v4.s_addr, 7865 0); 7866 7867 if (!afto && PF_ANEQ(pd->dst, 7868 &nk->addr[didx], AF_INET)) 7869 pf_change_a(&daddr->v4.s_addr, 7870 pd->ip_sum, 7871 nk->addr[didx].v4.s_addr, 0); 7872 7873 if (nk->port[iidx] != 7874 pd->hdr.icmp.icmp_id) { 7875 pd->hdr.icmp.icmp_cksum = 7876 pf_cksum_fixup( 7877 pd->hdr.icmp.icmp_cksum, icmpid, 7878 nk->port[iidx], 0); 7879 pd->hdr.icmp.icmp_id = 7880 nk->port[iidx]; 7881 } 7882 7883 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7884 (caddr_t )&pd->hdr.icmp); 7885 break; 7886 #endif /* INET */ 7887 #ifdef INET6 7888 case AF_INET6: 7889 #ifdef INET 7890 if (afto) { 7891 if (pf_translate_icmp_af(AF_INET, 7892 &pd->hdr.icmp6)) 7893 return (PF_DROP); 7894 pd->proto = IPPROTO_ICMP; 7895 } 7896 #endif /* INET */ 7897 if (!afto && 7898 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 7899 pf_change_a6(saddr, 7900 &pd->hdr.icmp6.icmp6_cksum, 7901 &nk->addr[sidx], 0); 7902 7903 if (!afto && PF_ANEQ(pd->dst, 7904 &nk->addr[didx], AF_INET6)) 7905 pf_change_a6(daddr, 7906 &pd->hdr.icmp6.icmp6_cksum, 7907 &nk->addr[didx], 0); 7908 7909 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 7910 pd->hdr.icmp6.icmp6_id = 7911 nk->port[iidx]; 7912 7913 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7914 (caddr_t )&pd->hdr.icmp6); 7915 break; 7916 #endif /* INET6 */ 7917 } 7918 if (afto) { 7919 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 7920 nk->af); 7921 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 7922 nk->af); 7923 pd->naf = nk->af; 7924 return (PF_AFRT); 7925 } 7926 } 7927 return (PF_PASS); 7928 7929 } else { 7930 /* 7931 * ICMP error message in response to a TCP/UDP packet. 7932 * Extract the inner TCP/UDP header and search for that state. 7933 */ 7934 7935 struct pf_pdesc pd2; 7936 bzero(&pd2, sizeof pd2); 7937 #ifdef INET 7938 struct ip h2; 7939 #endif /* INET */ 7940 #ifdef INET6 7941 struct ip6_hdr h2_6; 7942 #endif /* INET6 */ 7943 int ipoff2 = 0; 7944 7945 pd2.af = pd->af; 7946 pd2.dir = pd->dir; 7947 /* Payload packet is from the opposite direction. */ 7948 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7949 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7950 pd2.m = pd->m; 7951 pd2.pf_mtag = pd->pf_mtag; 7952 pd2.kif = pd->kif; 7953 switch (pd->af) { 7954 #ifdef INET 7955 case AF_INET: 7956 /* offset of h2 in mbuf chain */ 7957 ipoff2 = pd->off + ICMP_MINLEN; 7958 7959 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7960 NULL, reason, pd2.af)) { 7961 DPFPRINTF(PF_DEBUG_MISC, 7962 ("pf: ICMP error message too short " 7963 "(ip)\n")); 7964 return (PF_DROP); 7965 } 7966 /* 7967 * ICMP error messages don't refer to non-first 7968 * fragments 7969 */ 7970 if (h2.ip_off & htons(IP_OFFMASK)) { 7971 REASON_SET(reason, PFRES_FRAG); 7972 return (PF_DROP); 7973 } 7974 7975 /* offset of protocol header that follows h2 */ 7976 pd2.off = ipoff2; 7977 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 7978 return (PF_DROP); 7979 7980 pd2.tot_len = ntohs(h2.ip_len); 7981 pd2.src = (struct pf_addr *)&h2.ip_src; 7982 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7983 pd2.ip_sum = &h2.ip_sum; 7984 break; 7985 #endif /* INET */ 7986 #ifdef INET6 7987 case AF_INET6: 7988 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7989 7990 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7991 NULL, reason, pd2.af)) { 7992 DPFPRINTF(PF_DEBUG_MISC, 7993 ("pf: ICMP error message too short " 7994 "(ip6)\n")); 7995 return (PF_DROP); 7996 } 7997 pd2.off = ipoff2; 7998 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 7999 return (PF_DROP); 8000 8001 pd2.tot_len = ntohs(h2_6.ip6_plen) + 8002 sizeof(struct ip6_hdr); 8003 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 8004 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 8005 pd2.ip_sum = NULL; 8006 break; 8007 #endif /* INET6 */ 8008 default: 8009 unhandled_af(pd->af); 8010 } 8011 8012 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 8013 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8014 printf("pf: BAD ICMP %d:%d outer dst: ", 8015 icmptype, icmpcode); 8016 pf_print_host(pd->src, 0, pd->af); 8017 printf(" -> "); 8018 pf_print_host(pd->dst, 0, pd->af); 8019 printf(" inner src: "); 8020 pf_print_host(pd2.src, 0, pd2.af); 8021 printf(" -> "); 8022 pf_print_host(pd2.dst, 0, pd2.af); 8023 printf("\n"); 8024 } 8025 REASON_SET(reason, PFRES_BADSTATE); 8026 return (PF_DROP); 8027 } 8028 8029 switch (pd2.proto) { 8030 case IPPROTO_TCP: { 8031 struct tcphdr *th = &pd2.hdr.tcp; 8032 u_int32_t seq; 8033 struct pf_state_peer *src, *dst; 8034 u_int8_t dws; 8035 int copyback = 0; 8036 int action; 8037 8038 /* 8039 * Only the first 8 bytes of the TCP header can be 8040 * expected. Don't access any TCP header fields after 8041 * th_seq, an ackskew test is not possible. 8042 */ 8043 if (!pf_pull_hdr(pd->m, pd2.off, th, 8, NULL, reason, 8044 pd2.af)) { 8045 DPFPRINTF(PF_DEBUG_MISC, 8046 ("pf: ICMP error message too short " 8047 "(tcp)\n")); 8048 return (PF_DROP); 8049 } 8050 pd2.pcksum = &pd2.hdr.tcp.th_sum; 8051 8052 key.af = pd2.af; 8053 key.proto = IPPROTO_TCP; 8054 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8055 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8056 key.port[pd2.sidx] = th->th_sport; 8057 key.port[pd2.didx] = th->th_dport; 8058 8059 action = pf_find_state(&pd2, &key, state); 8060 if (action != PF_MATCH) 8061 return (action); 8062 8063 if (pd->dir == (*state)->direction) { 8064 if (PF_REVERSED_KEY(*state, pd->af)) { 8065 src = &(*state)->src; 8066 dst = &(*state)->dst; 8067 } else { 8068 src = &(*state)->dst; 8069 dst = &(*state)->src; 8070 } 8071 } else { 8072 if (PF_REVERSED_KEY(*state, pd->af)) { 8073 src = &(*state)->dst; 8074 dst = &(*state)->src; 8075 } else { 8076 src = &(*state)->src; 8077 dst = &(*state)->dst; 8078 } 8079 } 8080 8081 if (src->wscale && dst->wscale) 8082 dws = dst->wscale & PF_WSCALE_MASK; 8083 else 8084 dws = 0; 8085 8086 /* Demodulate sequence number */ 8087 seq = ntohl(th->th_seq) - src->seqdiff; 8088 if (src->seqdiff) { 8089 pf_change_a(&th->th_seq, icmpsum, 8090 htonl(seq), 0); 8091 copyback = 1; 8092 } 8093 8094 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 8095 (!SEQ_GEQ(src->seqhi, seq) || 8096 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 8097 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8098 printf("pf: BAD ICMP %d:%d ", 8099 icmptype, icmpcode); 8100 pf_print_host(pd->src, 0, pd->af); 8101 printf(" -> "); 8102 pf_print_host(pd->dst, 0, pd->af); 8103 printf(" state: "); 8104 pf_print_state(*state); 8105 printf(" seq=%u\n", seq); 8106 } 8107 REASON_SET(reason, PFRES_BADSTATE); 8108 return (PF_DROP); 8109 } else { 8110 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8111 printf("pf: OK ICMP %d:%d ", 8112 icmptype, icmpcode); 8113 pf_print_host(pd->src, 0, pd->af); 8114 printf(" -> "); 8115 pf_print_host(pd->dst, 0, pd->af); 8116 printf(" state: "); 8117 pf_print_state(*state); 8118 printf(" seq=%u\n", seq); 8119 } 8120 } 8121 8122 /* translate source/destination address, if necessary */ 8123 if ((*state)->key[PF_SK_WIRE] != 8124 (*state)->key[PF_SK_STACK]) { 8125 8126 struct pf_state_key *nk; 8127 8128 if (PF_REVERSED_KEY(*state, pd->af)) 8129 nk = (*state)->key[pd->sidx]; 8130 else 8131 nk = (*state)->key[pd->didx]; 8132 8133 #if defined(INET) && defined(INET6) 8134 int afto, sidx, didx; 8135 8136 afto = pd->af != nk->af; 8137 8138 if (afto && (*state)->direction == PF_IN) { 8139 sidx = pd2.didx; 8140 didx = pd2.sidx; 8141 } else { 8142 sidx = pd2.sidx; 8143 didx = pd2.didx; 8144 } 8145 8146 if (afto) { 8147 if (pf_translate_icmp_af(nk->af, 8148 &pd->hdr.icmp)) 8149 return (PF_DROP); 8150 m_copyback(pd->m, pd->off, 8151 sizeof(struct icmp6_hdr), 8152 (c_caddr_t)&pd->hdr.icmp6); 8153 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8154 &pd2, &nk->addr[sidx], 8155 &nk->addr[didx], pd->af, 8156 nk->af)) 8157 return (PF_DROP); 8158 pf_addrcpy(&pd->nsaddr, 8159 &nk->addr[pd2.sidx], nk->af); 8160 pf_addrcpy(&pd->ndaddr, 8161 &nk->addr[pd2.didx], nk->af); 8162 if (nk->af == AF_INET) { 8163 pd->proto = IPPROTO_ICMP; 8164 } else { 8165 pd->proto = IPPROTO_ICMPV6; 8166 /* 8167 * IPv4 becomes IPv6 so we must 8168 * copy IPv4 src addr to least 8169 * 32bits in IPv6 address to 8170 * keep traceroute/icmp 8171 * working. 8172 */ 8173 pd->nsaddr.addr32[3] = 8174 pd->src->addr32[0]; 8175 } 8176 pd->naf = pd2.naf = nk->af; 8177 pf_change_ap(&pd2, pd2.src, &th->th_sport, 8178 &nk->addr[pd2.sidx], nk->port[sidx]); 8179 pf_change_ap(&pd2, pd2.dst, &th->th_dport, 8180 &nk->addr[pd2.didx], nk->port[didx]); 8181 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)th); 8182 return (PF_AFRT); 8183 } 8184 #endif /* INET && INET6 */ 8185 8186 if (PF_ANEQ(pd2.src, 8187 &nk->addr[pd2.sidx], pd2.af) || 8188 nk->port[pd2.sidx] != th->th_sport) 8189 pf_change_icmp(pd2.src, &th->th_sport, 8190 daddr, &nk->addr[pd2.sidx], 8191 nk->port[pd2.sidx], NULL, 8192 pd2.ip_sum, icmpsum, 8193 pd->ip_sum, 0, pd2.af); 8194 8195 if (PF_ANEQ(pd2.dst, 8196 &nk->addr[pd2.didx], pd2.af) || 8197 nk->port[pd2.didx] != th->th_dport) 8198 pf_change_icmp(pd2.dst, &th->th_dport, 8199 saddr, &nk->addr[pd2.didx], 8200 nk->port[pd2.didx], NULL, 8201 pd2.ip_sum, icmpsum, 8202 pd->ip_sum, 0, pd2.af); 8203 copyback = 1; 8204 } 8205 8206 if (copyback) { 8207 switch (pd2.af) { 8208 #ifdef INET 8209 case AF_INET: 8210 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8211 (caddr_t )&pd->hdr.icmp); 8212 m_copyback(pd->m, ipoff2, sizeof(h2), 8213 (caddr_t )&h2); 8214 break; 8215 #endif /* INET */ 8216 #ifdef INET6 8217 case AF_INET6: 8218 m_copyback(pd->m, pd->off, 8219 sizeof(struct icmp6_hdr), 8220 (caddr_t )&pd->hdr.icmp6); 8221 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8222 (caddr_t )&h2_6); 8223 break; 8224 #endif /* INET6 */ 8225 default: 8226 unhandled_af(pd->af); 8227 } 8228 m_copyback(pd->m, pd2.off, 8, (caddr_t)th); 8229 } 8230 8231 return (PF_PASS); 8232 break; 8233 } 8234 case IPPROTO_UDP: { 8235 struct udphdr *uh = &pd2.hdr.udp; 8236 int action; 8237 8238 if (!pf_pull_hdr(pd->m, pd2.off, uh, sizeof(*uh), 8239 NULL, reason, pd2.af)) { 8240 DPFPRINTF(PF_DEBUG_MISC, 8241 ("pf: ICMP error message too short " 8242 "(udp)\n")); 8243 return (PF_DROP); 8244 } 8245 pd2.pcksum = &pd2.hdr.udp.uh_sum; 8246 8247 key.af = pd2.af; 8248 key.proto = IPPROTO_UDP; 8249 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8250 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8251 key.port[pd2.sidx] = uh->uh_sport; 8252 key.port[pd2.didx] = uh->uh_dport; 8253 8254 action = pf_find_state(&pd2, &key, state); 8255 if (action != PF_MATCH) 8256 return (action); 8257 8258 /* translate source/destination address, if necessary */ 8259 if ((*state)->key[PF_SK_WIRE] != 8260 (*state)->key[PF_SK_STACK]) { 8261 struct pf_state_key *nk; 8262 8263 if (PF_REVERSED_KEY(*state, pd->af)) 8264 nk = (*state)->key[pd->sidx]; 8265 else 8266 nk = (*state)->key[pd->didx]; 8267 8268 #if defined(INET) && defined(INET6) 8269 int afto, sidx, didx; 8270 8271 afto = pd->af != nk->af; 8272 8273 if (afto && (*state)->direction == PF_IN) { 8274 sidx = pd2.didx; 8275 didx = pd2.sidx; 8276 } else { 8277 sidx = pd2.sidx; 8278 didx = pd2.didx; 8279 } 8280 8281 if (afto) { 8282 if (pf_translate_icmp_af(nk->af, 8283 &pd->hdr.icmp)) 8284 return (PF_DROP); 8285 m_copyback(pd->m, pd->off, 8286 sizeof(struct icmp6_hdr), 8287 (c_caddr_t)&pd->hdr.icmp6); 8288 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8289 &pd2, &nk->addr[sidx], 8290 &nk->addr[didx], pd->af, 8291 nk->af)) 8292 return (PF_DROP); 8293 pf_addrcpy(&pd->nsaddr, 8294 &nk->addr[pd2.sidx], nk->af); 8295 pf_addrcpy(&pd->ndaddr, 8296 &nk->addr[pd2.didx], nk->af); 8297 if (nk->af == AF_INET) { 8298 pd->proto = IPPROTO_ICMP; 8299 } else { 8300 pd->proto = IPPROTO_ICMPV6; 8301 /* 8302 * IPv4 becomes IPv6 so we must 8303 * copy IPv4 src addr to least 8304 * 32bits in IPv6 address to 8305 * keep traceroute/icmp 8306 * working. 8307 */ 8308 pd->nsaddr.addr32[3] = 8309 pd->src->addr32[0]; 8310 } 8311 pd->naf = pd2.naf = nk->af; 8312 pf_change_ap(&pd2, pd2.src, &uh->uh_sport, 8313 &nk->addr[pd2.sidx], nk->port[sidx]); 8314 pf_change_ap(&pd2, pd2.dst, &uh->uh_dport, 8315 &nk->addr[pd2.didx], nk->port[didx]); 8316 m_copyback(pd2.m, pd2.off, sizeof(*uh), 8317 (c_caddr_t)uh); 8318 return (PF_AFRT); 8319 } 8320 #endif /* INET && INET6 */ 8321 8322 if (PF_ANEQ(pd2.src, 8323 &nk->addr[pd2.sidx], pd2.af) || 8324 nk->port[pd2.sidx] != uh->uh_sport) 8325 pf_change_icmp(pd2.src, &uh->uh_sport, 8326 daddr, &nk->addr[pd2.sidx], 8327 nk->port[pd2.sidx], &uh->uh_sum, 8328 pd2.ip_sum, icmpsum, 8329 pd->ip_sum, 1, pd2.af); 8330 8331 if (PF_ANEQ(pd2.dst, 8332 &nk->addr[pd2.didx], pd2.af) || 8333 nk->port[pd2.didx] != uh->uh_dport) 8334 pf_change_icmp(pd2.dst, &uh->uh_dport, 8335 saddr, &nk->addr[pd2.didx], 8336 nk->port[pd2.didx], &uh->uh_sum, 8337 pd2.ip_sum, icmpsum, 8338 pd->ip_sum, 1, pd2.af); 8339 8340 switch (pd2.af) { 8341 #ifdef INET 8342 case AF_INET: 8343 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8344 (caddr_t )&pd->hdr.icmp); 8345 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8346 break; 8347 #endif /* INET */ 8348 #ifdef INET6 8349 case AF_INET6: 8350 m_copyback(pd->m, pd->off, 8351 sizeof(struct icmp6_hdr), 8352 (caddr_t )&pd->hdr.icmp6); 8353 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8354 (caddr_t )&h2_6); 8355 break; 8356 #endif /* INET6 */ 8357 } 8358 m_copyback(pd->m, pd2.off, sizeof(*uh), (caddr_t)uh); 8359 } 8360 return (PF_PASS); 8361 break; 8362 } 8363 #ifdef INET 8364 case IPPROTO_SCTP: { 8365 struct sctphdr *sh = &pd2.hdr.sctp; 8366 struct pf_state_peer *src; 8367 int copyback = 0; 8368 int action; 8369 8370 if (! pf_pull_hdr(pd->m, pd2.off, sh, sizeof(*sh), NULL, reason, 8371 pd2.af)) { 8372 DPFPRINTF(PF_DEBUG_MISC, 8373 ("pf: ICMP error message too short " 8374 "(sctp)\n")); 8375 return (PF_DROP); 8376 } 8377 pd2.pcksum = &pd2.sctp_dummy_sum; 8378 8379 key.af = pd2.af; 8380 key.proto = IPPROTO_SCTP; 8381 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8382 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8383 key.port[pd2.sidx] = sh->src_port; 8384 key.port[pd2.didx] = sh->dest_port; 8385 8386 action = pf_find_state(&pd2, &key, state); 8387 if (action != PF_MATCH) 8388 return (action); 8389 8390 if (pd->dir == (*state)->direction) { 8391 if (PF_REVERSED_KEY(*state, pd->af)) 8392 src = &(*state)->src; 8393 else 8394 src = &(*state)->dst; 8395 } else { 8396 if (PF_REVERSED_KEY(*state, pd->af)) 8397 src = &(*state)->dst; 8398 else 8399 src = &(*state)->src; 8400 } 8401 8402 if (src->scrub->pfss_v_tag != sh->v_tag) { 8403 DPFPRINTF(PF_DEBUG_MISC, 8404 ("pf: ICMP error message has incorrect " 8405 "SCTP v_tag\n")); 8406 return (PF_DROP); 8407 } 8408 8409 /* translate source/destination address, if necessary */ 8410 if ((*state)->key[PF_SK_WIRE] != 8411 (*state)->key[PF_SK_STACK]) { 8412 8413 struct pf_state_key *nk; 8414 8415 if (PF_REVERSED_KEY(*state, pd->af)) 8416 nk = (*state)->key[pd->sidx]; 8417 else 8418 nk = (*state)->key[pd->didx]; 8419 8420 #if defined(INET) && defined(INET6) 8421 int afto, sidx, didx; 8422 8423 afto = pd->af != nk->af; 8424 8425 if (afto && (*state)->direction == PF_IN) { 8426 sidx = pd2.didx; 8427 didx = pd2.sidx; 8428 } else { 8429 sidx = pd2.sidx; 8430 didx = pd2.didx; 8431 } 8432 8433 if (afto) { 8434 if (pf_translate_icmp_af(nk->af, 8435 &pd->hdr.icmp)) 8436 return (PF_DROP); 8437 m_copyback(pd->m, pd->off, 8438 sizeof(struct icmp6_hdr), 8439 (c_caddr_t)&pd->hdr.icmp6); 8440 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8441 &pd2, &nk->addr[sidx], 8442 &nk->addr[didx], pd->af, 8443 nk->af)) 8444 return (PF_DROP); 8445 sh->src_port = nk->port[sidx]; 8446 sh->dest_port = nk->port[didx]; 8447 m_copyback(pd2.m, pd2.off, sizeof(*sh), (c_caddr_t)sh); 8448 pf_addrcpy(&pd->nsaddr, 8449 &nk->addr[pd2.sidx], nk->af); 8450 pf_addrcpy(&pd->ndaddr, 8451 &nk->addr[pd2.didx], nk->af); 8452 if (nk->af == AF_INET) { 8453 pd->proto = IPPROTO_ICMP; 8454 } else { 8455 pd->proto = IPPROTO_ICMPV6; 8456 /* 8457 * IPv4 becomes IPv6 so we must 8458 * copy IPv4 src addr to least 8459 * 32bits in IPv6 address to 8460 * keep traceroute/icmp 8461 * working. 8462 */ 8463 pd->nsaddr.addr32[3] = 8464 pd->src->addr32[0]; 8465 } 8466 pd->naf = nk->af; 8467 return (PF_AFRT); 8468 } 8469 #endif /* INET && INET6 */ 8470 8471 if (PF_ANEQ(pd2.src, 8472 &nk->addr[pd2.sidx], pd2.af) || 8473 nk->port[pd2.sidx] != sh->src_port) 8474 pf_change_icmp(pd2.src, &sh->src_port, 8475 daddr, &nk->addr[pd2.sidx], 8476 nk->port[pd2.sidx], NULL, 8477 pd2.ip_sum, icmpsum, 8478 pd->ip_sum, 0, pd2.af); 8479 8480 if (PF_ANEQ(pd2.dst, 8481 &nk->addr[pd2.didx], pd2.af) || 8482 nk->port[pd2.didx] != sh->dest_port) 8483 pf_change_icmp(pd2.dst, &sh->dest_port, 8484 saddr, &nk->addr[pd2.didx], 8485 nk->port[pd2.didx], NULL, 8486 pd2.ip_sum, icmpsum, 8487 pd->ip_sum, 0, pd2.af); 8488 copyback = 1; 8489 } 8490 8491 if (copyback) { 8492 switch (pd2.af) { 8493 #ifdef INET 8494 case AF_INET: 8495 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8496 (caddr_t )&pd->hdr.icmp); 8497 m_copyback(pd->m, ipoff2, sizeof(h2), 8498 (caddr_t )&h2); 8499 break; 8500 #endif /* INET */ 8501 #ifdef INET6 8502 case AF_INET6: 8503 m_copyback(pd->m, pd->off, 8504 sizeof(struct icmp6_hdr), 8505 (caddr_t )&pd->hdr.icmp6); 8506 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8507 (caddr_t )&h2_6); 8508 break; 8509 #endif /* INET6 */ 8510 } 8511 m_copyback(pd->m, pd2.off, sizeof(*sh), (caddr_t)sh); 8512 } 8513 8514 return (PF_PASS); 8515 break; 8516 } 8517 case IPPROTO_ICMP: { 8518 struct icmp *iih = &pd2.hdr.icmp; 8519 8520 if (pd2.af != AF_INET) { 8521 REASON_SET(reason, PFRES_NORM); 8522 return (PF_DROP); 8523 } 8524 8525 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8526 NULL, reason, pd2.af)) { 8527 DPFPRINTF(PF_DEBUG_MISC, 8528 ("pf: ICMP error message too short i" 8529 "(icmp)\n")); 8530 return (PF_DROP); 8531 } 8532 pd2.pcksum = &pd2.hdr.icmp.icmp_cksum; 8533 8534 icmpid = iih->icmp_id; 8535 pf_icmp_mapping(&pd2, iih->icmp_type, 8536 &icmp_dir, &virtual_id, &virtual_type); 8537 8538 ret = pf_icmp_state_lookup(&key, &pd2, state, 8539 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8540 if (ret >= 0) { 8541 MPASS(*state == NULL); 8542 return (ret); 8543 } 8544 8545 /* translate source/destination address, if necessary */ 8546 if ((*state)->key[PF_SK_WIRE] != 8547 (*state)->key[PF_SK_STACK]) { 8548 struct pf_state_key *nk; 8549 8550 if (PF_REVERSED_KEY(*state, pd->af)) 8551 nk = (*state)->key[pd->sidx]; 8552 else 8553 nk = (*state)->key[pd->didx]; 8554 8555 #if defined(INET) && defined(INET6) 8556 int afto, sidx, didx; 8557 8558 afto = pd->af != nk->af; 8559 8560 if (afto && (*state)->direction == PF_IN) { 8561 sidx = pd2.didx; 8562 didx = pd2.sidx; 8563 iidx = !iidx; 8564 } else { 8565 sidx = pd2.sidx; 8566 didx = pd2.didx; 8567 } 8568 8569 if (afto) { 8570 if (nk->af != AF_INET6) 8571 return (PF_DROP); 8572 if (pf_translate_icmp_af(nk->af, 8573 &pd->hdr.icmp)) 8574 return (PF_DROP); 8575 m_copyback(pd->m, pd->off, 8576 sizeof(struct icmp6_hdr), 8577 (c_caddr_t)&pd->hdr.icmp6); 8578 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8579 &pd2, &nk->addr[sidx], 8580 &nk->addr[didx], pd->af, 8581 nk->af)) 8582 return (PF_DROP); 8583 pd->proto = IPPROTO_ICMPV6; 8584 if (pf_translate_icmp_af(nk->af, iih)) 8585 return (PF_DROP); 8586 if (virtual_type == htons(ICMP_ECHO) && 8587 nk->port[iidx] != iih->icmp_id) 8588 iih->icmp_id = nk->port[iidx]; 8589 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8590 (c_caddr_t)iih); 8591 pf_addrcpy(&pd->nsaddr, 8592 &nk->addr[pd2.sidx], nk->af); 8593 pf_addrcpy(&pd->ndaddr, 8594 &nk->addr[pd2.didx], nk->af); 8595 /* 8596 * IPv4 becomes IPv6 so we must copy 8597 * IPv4 src addr to least 32bits in 8598 * IPv6 address to keep traceroute 8599 * working. 8600 */ 8601 pd->nsaddr.addr32[3] = 8602 pd->src->addr32[0]; 8603 pd->naf = nk->af; 8604 return (PF_AFRT); 8605 } 8606 #endif /* INET && INET6 */ 8607 8608 if (PF_ANEQ(pd2.src, 8609 &nk->addr[pd2.sidx], pd2.af) || 8610 (virtual_type == htons(ICMP_ECHO) && 8611 nk->port[iidx] != iih->icmp_id)) 8612 pf_change_icmp(pd2.src, 8613 (virtual_type == htons(ICMP_ECHO)) ? 8614 &iih->icmp_id : NULL, 8615 daddr, &nk->addr[pd2.sidx], 8616 (virtual_type == htons(ICMP_ECHO)) ? 8617 nk->port[iidx] : 0, NULL, 8618 pd2.ip_sum, icmpsum, 8619 pd->ip_sum, 0, AF_INET); 8620 8621 if (PF_ANEQ(pd2.dst, 8622 &nk->addr[pd2.didx], pd2.af)) 8623 pf_change_icmp(pd2.dst, NULL, NULL, 8624 &nk->addr[pd2.didx], 0, NULL, 8625 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8626 AF_INET); 8627 8628 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8629 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8630 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8631 } 8632 return (PF_PASS); 8633 break; 8634 } 8635 #endif /* INET */ 8636 #ifdef INET6 8637 case IPPROTO_ICMPV6: { 8638 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8639 8640 if (pd2.af != AF_INET6) { 8641 REASON_SET(reason, PFRES_NORM); 8642 return (PF_DROP); 8643 } 8644 8645 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8646 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 8647 DPFPRINTF(PF_DEBUG_MISC, 8648 ("pf: ICMP error message too short " 8649 "(icmp6)\n")); 8650 return (PF_DROP); 8651 } 8652 pd2.pcksum = &pd2.hdr.icmp6.icmp6_cksum; 8653 8654 pf_icmp_mapping(&pd2, iih->icmp6_type, 8655 &icmp_dir, &virtual_id, &virtual_type); 8656 8657 ret = pf_icmp_state_lookup(&key, &pd2, state, 8658 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8659 /* IPv6? try matching a multicast address */ 8660 if (ret == PF_DROP && pd2.af == AF_INET6 && 8661 icmp_dir == PF_OUT) { 8662 MPASS(*state == NULL); 8663 ret = pf_icmp_state_lookup(&key, &pd2, 8664 state, virtual_id, virtual_type, 8665 icmp_dir, &iidx, 1, 1); 8666 } 8667 if (ret >= 0) { 8668 MPASS(*state == NULL); 8669 return (ret); 8670 } 8671 8672 /* translate source/destination address, if necessary */ 8673 if ((*state)->key[PF_SK_WIRE] != 8674 (*state)->key[PF_SK_STACK]) { 8675 struct pf_state_key *nk; 8676 8677 if (PF_REVERSED_KEY(*state, pd->af)) 8678 nk = (*state)->key[pd->sidx]; 8679 else 8680 nk = (*state)->key[pd->didx]; 8681 8682 #if defined(INET) && defined(INET6) 8683 int afto, sidx, didx; 8684 8685 afto = pd->af != nk->af; 8686 8687 if (afto && (*state)->direction == PF_IN) { 8688 sidx = pd2.didx; 8689 didx = pd2.sidx; 8690 iidx = !iidx; 8691 } else { 8692 sidx = pd2.sidx; 8693 didx = pd2.didx; 8694 } 8695 8696 if (afto) { 8697 if (nk->af != AF_INET) 8698 return (PF_DROP); 8699 if (pf_translate_icmp_af(nk->af, 8700 &pd->hdr.icmp)) 8701 return (PF_DROP); 8702 m_copyback(pd->m, pd->off, 8703 sizeof(struct icmp6_hdr), 8704 (c_caddr_t)&pd->hdr.icmp6); 8705 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8706 &pd2, &nk->addr[sidx], 8707 &nk->addr[didx], pd->af, 8708 nk->af)) 8709 return (PF_DROP); 8710 pd->proto = IPPROTO_ICMP; 8711 if (pf_translate_icmp_af(nk->af, iih)) 8712 return (PF_DROP); 8713 if (virtual_type == 8714 htons(ICMP6_ECHO_REQUEST) && 8715 nk->port[iidx] != iih->icmp6_id) 8716 iih->icmp6_id = nk->port[iidx]; 8717 m_copyback(pd2.m, pd2.off, 8718 sizeof(struct icmp6_hdr), (c_caddr_t)iih); 8719 pf_addrcpy(&pd->nsaddr, 8720 &nk->addr[pd2.sidx], nk->af); 8721 pf_addrcpy(&pd->ndaddr, 8722 &nk->addr[pd2.didx], nk->af); 8723 pd->naf = nk->af; 8724 return (PF_AFRT); 8725 } 8726 #endif /* INET && INET6 */ 8727 8728 if (PF_ANEQ(pd2.src, 8729 &nk->addr[pd2.sidx], pd2.af) || 8730 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8731 nk->port[pd2.sidx] != iih->icmp6_id)) 8732 pf_change_icmp(pd2.src, 8733 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8734 ? &iih->icmp6_id : NULL, 8735 daddr, &nk->addr[pd2.sidx], 8736 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8737 ? nk->port[iidx] : 0, NULL, 8738 pd2.ip_sum, icmpsum, 8739 pd->ip_sum, 0, AF_INET6); 8740 8741 if (PF_ANEQ(pd2.dst, 8742 &nk->addr[pd2.didx], pd2.af)) 8743 pf_change_icmp(pd2.dst, NULL, NULL, 8744 &nk->addr[pd2.didx], 0, NULL, 8745 pd2.ip_sum, icmpsum, 8746 pd->ip_sum, 0, AF_INET6); 8747 8748 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8749 (caddr_t)&pd->hdr.icmp6); 8750 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8751 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8752 (caddr_t)iih); 8753 } 8754 return (PF_PASS); 8755 break; 8756 } 8757 #endif /* INET6 */ 8758 default: { 8759 int action; 8760 8761 key.af = pd2.af; 8762 key.proto = pd2.proto; 8763 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8764 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8765 key.port[0] = key.port[1] = 0; 8766 8767 action = pf_find_state(&pd2, &key, state); 8768 if (action != PF_MATCH) 8769 return (action); 8770 8771 /* translate source/destination address, if necessary */ 8772 if ((*state)->key[PF_SK_WIRE] != 8773 (*state)->key[PF_SK_STACK]) { 8774 struct pf_state_key *nk = 8775 (*state)->key[pd->didx]; 8776 8777 if (PF_ANEQ(pd2.src, 8778 &nk->addr[pd2.sidx], pd2.af)) 8779 pf_change_icmp(pd2.src, NULL, daddr, 8780 &nk->addr[pd2.sidx], 0, NULL, 8781 pd2.ip_sum, icmpsum, 8782 pd->ip_sum, 0, pd2.af); 8783 8784 if (PF_ANEQ(pd2.dst, 8785 &nk->addr[pd2.didx], pd2.af)) 8786 pf_change_icmp(pd2.dst, NULL, saddr, 8787 &nk->addr[pd2.didx], 0, NULL, 8788 pd2.ip_sum, icmpsum, 8789 pd->ip_sum, 0, pd2.af); 8790 8791 switch (pd2.af) { 8792 #ifdef INET 8793 case AF_INET: 8794 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8795 (caddr_t)&pd->hdr.icmp); 8796 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8797 break; 8798 #endif /* INET */ 8799 #ifdef INET6 8800 case AF_INET6: 8801 m_copyback(pd->m, pd->off, 8802 sizeof(struct icmp6_hdr), 8803 (caddr_t )&pd->hdr.icmp6); 8804 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8805 (caddr_t )&h2_6); 8806 break; 8807 #endif /* INET6 */ 8808 } 8809 } 8810 return (PF_PASS); 8811 break; 8812 } 8813 } 8814 } 8815 } 8816 8817 /* 8818 * ipoff and off are measured from the start of the mbuf chain. 8819 * h must be at "ipoff" on the mbuf chain. 8820 */ 8821 void * 8822 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8823 u_short *actionp, u_short *reasonp, sa_family_t af) 8824 { 8825 int iplen = 0; 8826 switch (af) { 8827 #ifdef INET 8828 case AF_INET: { 8829 const struct ip *h = mtod(m, struct ip *); 8830 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 8831 8832 if (fragoff) { 8833 if (fragoff >= len) 8834 ACTION_SET(actionp, PF_PASS); 8835 else { 8836 ACTION_SET(actionp, PF_DROP); 8837 REASON_SET(reasonp, PFRES_FRAG); 8838 } 8839 return (NULL); 8840 } 8841 iplen = ntohs(h->ip_len); 8842 break; 8843 } 8844 #endif /* INET */ 8845 #ifdef INET6 8846 case AF_INET6: { 8847 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8848 8849 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8850 break; 8851 } 8852 #endif /* INET6 */ 8853 } 8854 if (m->m_pkthdr.len < off + len || iplen < off + len) { 8855 ACTION_SET(actionp, PF_DROP); 8856 REASON_SET(reasonp, PFRES_SHORT); 8857 return (NULL); 8858 } 8859 m_copydata(m, off, len, p); 8860 return (p); 8861 } 8862 8863 int 8864 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 8865 int rtableid) 8866 { 8867 struct ifnet *ifp; 8868 8869 /* 8870 * Skip check for addresses with embedded interface scope, 8871 * as they would always match anyway. 8872 */ 8873 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 8874 return (1); 8875 8876 if (af != AF_INET && af != AF_INET6) 8877 return (0); 8878 8879 if (kif == V_pfi_all) 8880 return (1); 8881 8882 /* Skip checks for ipsec interfaces */ 8883 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 8884 return (1); 8885 8886 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 8887 8888 switch (af) { 8889 #ifdef INET6 8890 case AF_INET6: 8891 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 8892 ifp)); 8893 #endif /* INET6 */ 8894 #ifdef INET 8895 case AF_INET: 8896 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 8897 ifp)); 8898 #endif /* INET */ 8899 } 8900 8901 return (0); 8902 } 8903 8904 #ifdef INET 8905 static void 8906 pf_route(struct pf_krule *r, struct ifnet *oifp, 8907 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8908 { 8909 struct mbuf *m0, *m1, *md; 8910 struct route ro; 8911 const struct sockaddr *gw = &ro.ro_dst; 8912 struct sockaddr_in *dst; 8913 struct ip *ip; 8914 struct ifnet *ifp = NULL; 8915 int error = 0; 8916 uint16_t ip_len, ip_off; 8917 uint16_t tmp; 8918 int r_dir; 8919 bool skip_test = false; 8920 8921 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 8922 8923 SDT_PROBE4(pf, ip, route_to, entry, pd->m, pd, s, oifp); 8924 8925 if (s) { 8926 r_dir = s->direction; 8927 } else { 8928 r_dir = r->direction; 8929 } 8930 8931 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8932 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8933 __func__)); 8934 8935 if ((pd->pf_mtag == NULL && 8936 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 8937 pd->pf_mtag->routed++ > 3) { 8938 m0 = pd->m; 8939 pd->m = NULL; 8940 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8941 goto bad_locked; 8942 } 8943 8944 if (pd->act.rt_kif != NULL) 8945 ifp = pd->act.rt_kif->pfik_ifp; 8946 8947 if (pd->act.rt == PF_DUPTO) { 8948 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8949 if (s != NULL) { 8950 PF_STATE_UNLOCK(s); 8951 } 8952 if (ifp == oifp) { 8953 /* When the 2nd interface is not skipped */ 8954 return; 8955 } else { 8956 m0 = pd->m; 8957 pd->m = NULL; 8958 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8959 goto bad; 8960 } 8961 } else { 8962 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8963 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 8964 if (s) 8965 PF_STATE_UNLOCK(s); 8966 return; 8967 } 8968 } 8969 } else { 8970 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8971 if (pd->af == pd->naf) { 8972 pf_dummynet(pd, s, r, &pd->m); 8973 if (s) 8974 PF_STATE_UNLOCK(s); 8975 return; 8976 } else { 8977 if (r_dir == PF_IN) { 8978 skip_test = true; 8979 } 8980 } 8981 } 8982 8983 /* 8984 * If we're actually doing route-to and af-to and are in the 8985 * reply direction. 8986 */ 8987 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 8988 pd->af != pd->naf) { 8989 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 8990 /* Un-set ifp so we do a plain route lookup. */ 8991 ifp = NULL; 8992 } 8993 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 8994 /* Un-set ifp so we do a plain route lookup. */ 8995 ifp = NULL; 8996 } 8997 } 8998 m0 = pd->m; 8999 } 9000 9001 ip = mtod(m0, struct ip *); 9002 9003 bzero(&ro, sizeof(ro)); 9004 dst = (struct sockaddr_in *)&ro.ro_dst; 9005 dst->sin_family = AF_INET; 9006 dst->sin_len = sizeof(struct sockaddr_in); 9007 dst->sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 9008 9009 if (pd->dir == PF_IN) { 9010 if (ip->ip_ttl <= IPTTLDEC) { 9011 if (r->rt != PF_DUPTO) 9012 pf_send_icmp(m0, ICMP_TIMXCEED, 9013 ICMP_TIMXCEED_INTRANS, 0, pd->af, r, 9014 pd->act.rtableid); 9015 goto bad_locked; 9016 } 9017 ip->ip_ttl -= IPTTLDEC; 9018 } 9019 9020 if (s != NULL) { 9021 if (ifp == NULL && (pd->af != pd->naf)) { 9022 /* We're in the AFTO case. Do a route lookup. */ 9023 const struct nhop_object *nh; 9024 nh = fib4_lookup(M_GETFIB(m0), ip->ip_dst, 0, NHR_NONE, 0); 9025 if (nh) { 9026 ifp = nh->nh_ifp; 9027 9028 /* Use the gateway if needed. */ 9029 if (nh->nh_flags & NHF_GATEWAY) { 9030 gw = &nh->gw_sa; 9031 ro.ro_flags |= RT_HAS_GW; 9032 } else { 9033 dst->sin_addr = ip->ip_dst; 9034 } 9035 9036 /* 9037 * Bind to the correct interface if we're 9038 * if-bound. We don't know which interface 9039 * that will be until here, so we've inserted 9040 * the state on V_pf_all. Fix that now. 9041 */ 9042 if (s->kif == V_pfi_all && ifp != NULL && 9043 r->rule_flag & PFRULE_IFBOUND) 9044 s->kif = ifp->if_pf_kif; 9045 } 9046 } 9047 9048 if (r->rule_flag & PFRULE_IFBOUND && 9049 pd->act.rt == PF_REPLYTO && 9050 s->kif == V_pfi_all) { 9051 s->kif = pd->act.rt_kif; 9052 s->orig_kif = oifp->if_pf_kif; 9053 } 9054 9055 PF_STATE_UNLOCK(s); 9056 } 9057 9058 if (ifp == NULL) { 9059 m0 = pd->m; 9060 pd->m = NULL; 9061 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9062 goto bad; 9063 } 9064 9065 if (pd->dir == PF_IN && !skip_test) { 9066 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 9067 &pd->act) != PF_PASS) { 9068 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9069 goto bad; 9070 } else if (m0 == NULL) { 9071 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9072 goto done; 9073 } 9074 if (m0->m_len < sizeof(struct ip)) { 9075 DPFPRINTF(PF_DEBUG_URGENT, 9076 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 9077 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9078 goto bad; 9079 } 9080 ip = mtod(m0, struct ip *); 9081 } 9082 9083 if (ifp->if_flags & IFF_LOOPBACK) 9084 m0->m_flags |= M_SKIP_FIREWALL; 9085 9086 ip_len = ntohs(ip->ip_len); 9087 ip_off = ntohs(ip->ip_off); 9088 9089 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 9090 m0->m_pkthdr.csum_flags |= CSUM_IP; 9091 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 9092 in_delayed_cksum(m0); 9093 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 9094 } 9095 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 9096 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 9097 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 9098 } 9099 9100 if (pd->dir == PF_IN) { 9101 /* 9102 * Make sure dummynet gets the correct direction, in case it needs to 9103 * re-inject later. 9104 */ 9105 pd->dir = PF_OUT; 9106 9107 /* 9108 * The following processing is actually the rest of the inbound processing, even 9109 * though we've marked it as outbound (so we don't look through dummynet) and it 9110 * happens after the outbound processing (pf_test(PF_OUT) above). 9111 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9112 * conclusion about what direction it's processing, and we can't fix it or it 9113 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9114 * decision will pick the right pipe, and everything will mostly work as expected. 9115 */ 9116 tmp = pd->act.dnrpipe; 9117 pd->act.dnrpipe = pd->act.dnpipe; 9118 pd->act.dnpipe = tmp; 9119 } 9120 9121 /* 9122 * If small enough for interface, or the interface will take 9123 * care of the fragmentation for us, we can just send directly. 9124 */ 9125 if (ip_len <= ifp->if_mtu || 9126 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 9127 ip->ip_sum = 0; 9128 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 9129 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 9130 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 9131 } 9132 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 9133 9134 md = m0; 9135 error = pf_dummynet_route(pd, s, r, ifp, gw, &md); 9136 if (md != NULL) { 9137 error = (*ifp->if_output)(ifp, md, gw, &ro); 9138 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9139 } 9140 goto done; 9141 } 9142 9143 /* Balk when DF bit is set or the interface didn't support TSO. */ 9144 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 9145 error = EMSGSIZE; 9146 KMOD_IPSTAT_INC(ips_cantfrag); 9147 if (pd->act.rt != PF_DUPTO) { 9148 if (s && s->nat_rule != NULL) { 9149 MPASS(m0 == pd->m); 9150 PACKET_UNDO_NAT(pd, 9151 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 9152 s); 9153 } 9154 9155 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 9156 ifp->if_mtu, pd->af, r, pd->act.rtableid); 9157 } 9158 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9159 goto bad; 9160 } 9161 9162 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 9163 if (error) { 9164 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9165 goto bad; 9166 } 9167 9168 for (; m0; m0 = m1) { 9169 m1 = m0->m_nextpkt; 9170 m0->m_nextpkt = NULL; 9171 if (error == 0) { 9172 m_clrprotoflags(m0); 9173 md = m0; 9174 pd->pf_mtag = pf_find_mtag(md); 9175 error = pf_dummynet_route(pd, s, r, ifp, 9176 gw, &md); 9177 if (md != NULL) { 9178 error = (*ifp->if_output)(ifp, md, gw, &ro); 9179 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9180 } 9181 } else 9182 m_freem(m0); 9183 } 9184 9185 if (error == 0) 9186 KMOD_IPSTAT_INC(ips_fragmented); 9187 9188 done: 9189 if (pd->act.rt != PF_DUPTO) 9190 pd->m = NULL; 9191 return; 9192 9193 bad_locked: 9194 if (s) 9195 PF_STATE_UNLOCK(s); 9196 bad: 9197 m_freem(m0); 9198 goto done; 9199 } 9200 #endif /* INET */ 9201 9202 #ifdef INET6 9203 static void 9204 pf_route6(struct pf_krule *r, struct ifnet *oifp, 9205 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9206 { 9207 struct mbuf *m0, *md; 9208 struct m_tag *mtag; 9209 struct sockaddr_in6 dst; 9210 struct ip6_hdr *ip6; 9211 struct ifnet *ifp = NULL; 9212 int r_dir; 9213 bool skip_test = false; 9214 9215 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 9216 9217 SDT_PROBE4(pf, ip6, route_to, entry, pd->m, pd, s, oifp); 9218 9219 if (s) { 9220 r_dir = s->direction; 9221 } else { 9222 r_dir = r->direction; 9223 } 9224 9225 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9226 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9227 __func__)); 9228 9229 if ((pd->pf_mtag == NULL && 9230 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 9231 pd->pf_mtag->routed++ > 3) { 9232 m0 = pd->m; 9233 pd->m = NULL; 9234 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9235 goto bad_locked; 9236 } 9237 9238 if (pd->act.rt_kif != NULL) 9239 ifp = pd->act.rt_kif->pfik_ifp; 9240 9241 if (pd->act.rt == PF_DUPTO) { 9242 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9243 if (s != NULL) { 9244 PF_STATE_UNLOCK(s); 9245 } 9246 if (ifp == oifp) { 9247 /* When the 2nd interface is not skipped */ 9248 return; 9249 } else { 9250 m0 = pd->m; 9251 pd->m = NULL; 9252 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9253 goto bad; 9254 } 9255 } else { 9256 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9257 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 9258 if (s) 9259 PF_STATE_UNLOCK(s); 9260 return; 9261 } 9262 } 9263 } else { 9264 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9265 if (pd->af == pd->naf) { 9266 pf_dummynet(pd, s, r, &pd->m); 9267 if (s) 9268 PF_STATE_UNLOCK(s); 9269 return; 9270 } else { 9271 if (r_dir == PF_IN) { 9272 skip_test = true; 9273 } 9274 } 9275 } 9276 9277 /* 9278 * If we're actually doing route-to and af-to and are in the 9279 * reply direction. 9280 */ 9281 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9282 pd->af != pd->naf) { 9283 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9284 /* Un-set ifp so we do a plain route lookup. */ 9285 ifp = NULL; 9286 } 9287 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9288 /* Un-set ifp so we do a plain route lookup. */ 9289 ifp = NULL; 9290 } 9291 } 9292 m0 = pd->m; 9293 } 9294 9295 ip6 = mtod(m0, struct ip6_hdr *); 9296 9297 bzero(&dst, sizeof(dst)); 9298 dst.sin6_family = AF_INET6; 9299 dst.sin6_len = sizeof(dst); 9300 pf_addrcpy((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, 9301 AF_INET6); 9302 9303 if (pd->dir == PF_IN) { 9304 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 9305 if (r->rt != PF_DUPTO) 9306 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 9307 ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r, 9308 pd->act.rtableid); 9309 goto bad_locked; 9310 } 9311 ip6->ip6_hlim -= IPV6_HLIMDEC; 9312 } 9313 9314 if (s != NULL) { 9315 if (ifp == NULL && (pd->af != pd->naf)) { 9316 const struct nhop_object *nh; 9317 nh = fib6_lookup(M_GETFIB(m0), &ip6->ip6_dst, 0, NHR_NONE, 0); 9318 if (nh) { 9319 ifp = nh->nh_ifp; 9320 9321 /* Use the gateway if needed. */ 9322 if (nh->nh_flags & NHF_GATEWAY) 9323 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9324 sizeof(dst.sin6_addr)); 9325 else 9326 dst.sin6_addr = ip6->ip6_dst; 9327 9328 /* 9329 * Bind to the correct interface if we're 9330 * if-bound. We don't know which interface 9331 * that will be until here, so we've inserted 9332 * the state on V_pf_all. Fix that now. 9333 */ 9334 if (s->kif == V_pfi_all && ifp != NULL && 9335 r->rule_flag & PFRULE_IFBOUND) 9336 s->kif = ifp->if_pf_kif; 9337 } 9338 } 9339 9340 if (r->rule_flag & PFRULE_IFBOUND && 9341 pd->act.rt == PF_REPLYTO && 9342 s->kif == V_pfi_all) { 9343 s->kif = pd->act.rt_kif; 9344 s->orig_kif = oifp->if_pf_kif; 9345 } 9346 9347 PF_STATE_UNLOCK(s); 9348 } 9349 9350 if (pd->af != pd->naf) { 9351 struct udphdr *uh = &pd->hdr.udp; 9352 9353 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9354 uh->uh_sum = in6_cksum_pseudo(ip6, 9355 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9356 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9357 } 9358 } 9359 9360 if (ifp == NULL) { 9361 m0 = pd->m; 9362 pd->m = NULL; 9363 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9364 goto bad; 9365 } 9366 9367 if (pd->dir == PF_IN && !skip_test) { 9368 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9369 ifp, &m0, inp, &pd->act) != PF_PASS) { 9370 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9371 goto bad; 9372 } else if (m0 == NULL) { 9373 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9374 goto done; 9375 } 9376 if (m0->m_len < sizeof(struct ip6_hdr)) { 9377 DPFPRINTF(PF_DEBUG_URGENT, 9378 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 9379 __func__)); 9380 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9381 goto bad; 9382 } 9383 ip6 = mtod(m0, struct ip6_hdr *); 9384 } 9385 9386 if (ifp->if_flags & IFF_LOOPBACK) 9387 m0->m_flags |= M_SKIP_FIREWALL; 9388 9389 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9390 ~ifp->if_hwassist) { 9391 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9392 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9393 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9394 } 9395 9396 if (pd->dir == PF_IN) { 9397 uint16_t tmp; 9398 /* 9399 * Make sure dummynet gets the correct direction, in case it needs to 9400 * re-inject later. 9401 */ 9402 pd->dir = PF_OUT; 9403 9404 /* 9405 * The following processing is actually the rest of the inbound processing, even 9406 * though we've marked it as outbound (so we don't look through dummynet) and it 9407 * happens after the outbound processing (pf_test(PF_OUT) above). 9408 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9409 * conclusion about what direction it's processing, and we can't fix it or it 9410 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9411 * decision will pick the right pipe, and everything will mostly work as expected. 9412 */ 9413 tmp = pd->act.dnrpipe; 9414 pd->act.dnrpipe = pd->act.dnpipe; 9415 pd->act.dnpipe = tmp; 9416 } 9417 9418 /* 9419 * If the packet is too large for the outgoing interface, 9420 * send back an icmp6 error. 9421 */ 9422 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9423 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9424 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9425 if (mtag != NULL) { 9426 int ret __sdt_used; 9427 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9428 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9429 goto done; 9430 } 9431 9432 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9433 md = m0; 9434 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9435 if (md != NULL) { 9436 int ret __sdt_used; 9437 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9438 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9439 } 9440 } 9441 else { 9442 in6_ifstat_inc(ifp, ifs6_in_toobig); 9443 if (pd->act.rt != PF_DUPTO) { 9444 if (s && s->nat_rule != NULL) { 9445 MPASS(m0 == pd->m); 9446 PACKET_UNDO_NAT(pd, 9447 ((caddr_t)ip6 - m0->m_data) + 9448 sizeof(struct ip6_hdr), s); 9449 } 9450 9451 if (r->rt != PF_DUPTO) 9452 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 9453 ifp->if_mtu, pd->af, r, pd->act.rtableid); 9454 } 9455 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9456 goto bad; 9457 } 9458 9459 done: 9460 if (pd->act.rt != PF_DUPTO) 9461 pd->m = NULL; 9462 return; 9463 9464 bad_locked: 9465 if (s) 9466 PF_STATE_UNLOCK(s); 9467 bad: 9468 m_freem(m0); 9469 goto done; 9470 } 9471 #endif /* INET6 */ 9472 9473 /* 9474 * FreeBSD supports cksum offloads for the following drivers. 9475 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9476 * 9477 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9478 * network driver performed cksum including pseudo header, need to verify 9479 * csum_data 9480 * CSUM_DATA_VALID : 9481 * network driver performed cksum, needs to additional pseudo header 9482 * cksum computation with partial csum_data(i.e. lack of H/W support for 9483 * pseudo header, for instance sk(4) and possibly gem(4)) 9484 * 9485 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9486 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9487 * TCP/UDP layer. 9488 * Also, set csum_data to 0xffff to force cksum validation. 9489 */ 9490 static int 9491 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9492 { 9493 u_int16_t sum = 0; 9494 int hw_assist = 0; 9495 struct ip *ip; 9496 9497 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9498 return (1); 9499 if (m->m_pkthdr.len < off + len) 9500 return (1); 9501 9502 switch (p) { 9503 case IPPROTO_TCP: 9504 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9505 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9506 sum = m->m_pkthdr.csum_data; 9507 } else { 9508 ip = mtod(m, struct ip *); 9509 sum = in_pseudo(ip->ip_src.s_addr, 9510 ip->ip_dst.s_addr, htonl((u_short)len + 9511 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9512 } 9513 sum ^= 0xffff; 9514 ++hw_assist; 9515 } 9516 break; 9517 case IPPROTO_UDP: 9518 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9519 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9520 sum = m->m_pkthdr.csum_data; 9521 } else { 9522 ip = mtod(m, struct ip *); 9523 sum = in_pseudo(ip->ip_src.s_addr, 9524 ip->ip_dst.s_addr, htonl((u_short)len + 9525 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9526 } 9527 sum ^= 0xffff; 9528 ++hw_assist; 9529 } 9530 break; 9531 case IPPROTO_ICMP: 9532 #ifdef INET6 9533 case IPPROTO_ICMPV6: 9534 #endif /* INET6 */ 9535 break; 9536 default: 9537 return (1); 9538 } 9539 9540 if (!hw_assist) { 9541 switch (af) { 9542 case AF_INET: 9543 if (m->m_len < sizeof(struct ip)) 9544 return (1); 9545 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 9546 break; 9547 #ifdef INET6 9548 case AF_INET6: 9549 if (m->m_len < sizeof(struct ip6_hdr)) 9550 return (1); 9551 sum = in6_cksum(m, p, off, len); 9552 break; 9553 #endif /* INET6 */ 9554 } 9555 } 9556 if (sum) { 9557 switch (p) { 9558 case IPPROTO_TCP: 9559 { 9560 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9561 break; 9562 } 9563 case IPPROTO_UDP: 9564 { 9565 KMOD_UDPSTAT_INC(udps_badsum); 9566 break; 9567 } 9568 #ifdef INET 9569 case IPPROTO_ICMP: 9570 { 9571 KMOD_ICMPSTAT_INC(icps_checksum); 9572 break; 9573 } 9574 #endif 9575 #ifdef INET6 9576 case IPPROTO_ICMPV6: 9577 { 9578 KMOD_ICMP6STAT_INC(icp6s_checksum); 9579 break; 9580 } 9581 #endif /* INET6 */ 9582 } 9583 return (1); 9584 } else { 9585 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9586 m->m_pkthdr.csum_flags |= 9587 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9588 m->m_pkthdr.csum_data = 0xffff; 9589 } 9590 } 9591 return (0); 9592 } 9593 9594 static bool 9595 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9596 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9597 { 9598 int dndir = r->direction; 9599 9600 if (s && dndir == PF_INOUT) { 9601 dndir = s->direction; 9602 } else if (dndir == PF_INOUT) { 9603 /* Assume primary direction. Happens when we've set dnpipe in 9604 * the ethernet level code. */ 9605 dndir = pd->dir; 9606 } 9607 9608 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9609 return (false); 9610 9611 memset(dnflow, 0, sizeof(*dnflow)); 9612 9613 if (pd->dport != NULL) 9614 dnflow->f_id.dst_port = ntohs(*pd->dport); 9615 if (pd->sport != NULL) 9616 dnflow->f_id.src_port = ntohs(*pd->sport); 9617 9618 if (pd->dir == PF_IN) 9619 dnflow->flags |= IPFW_ARGS_IN; 9620 else 9621 dnflow->flags |= IPFW_ARGS_OUT; 9622 9623 if (pd->dir != dndir && pd->act.dnrpipe) { 9624 dnflow->rule.info = pd->act.dnrpipe; 9625 } 9626 else if (pd->dir == dndir && pd->act.dnpipe) { 9627 dnflow->rule.info = pd->act.dnpipe; 9628 } 9629 else { 9630 return (false); 9631 } 9632 9633 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9634 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9635 dnflow->rule.info |= IPFW_IS_PIPE; 9636 9637 dnflow->f_id.proto = pd->proto; 9638 dnflow->f_id.extra = dnflow->rule.info; 9639 switch (pd->naf) { 9640 case AF_INET: 9641 dnflow->f_id.addr_type = 4; 9642 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9643 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9644 break; 9645 case AF_INET6: 9646 dnflow->flags |= IPFW_ARGS_IP6; 9647 dnflow->f_id.addr_type = 6; 9648 dnflow->f_id.src_ip6 = pd->src->v6; 9649 dnflow->f_id.dst_ip6 = pd->dst->v6; 9650 break; 9651 } 9652 9653 return (true); 9654 } 9655 9656 int 9657 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9658 struct inpcb *inp) 9659 { 9660 struct pfi_kkif *kif; 9661 struct mbuf *m = *m0; 9662 9663 M_ASSERTPKTHDR(m); 9664 MPASS(ifp->if_vnet == curvnet); 9665 NET_EPOCH_ASSERT(); 9666 9667 if (!V_pf_status.running) 9668 return (PF_PASS); 9669 9670 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9671 9672 if (kif == NULL) { 9673 DPFPRINTF(PF_DEBUG_URGENT, 9674 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 9675 return (PF_DROP); 9676 } 9677 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9678 return (PF_PASS); 9679 9680 if (m->m_flags & M_SKIP_FIREWALL) 9681 return (PF_PASS); 9682 9683 if (__predict_false(! M_WRITABLE(*m0))) { 9684 m = *m0 = m_unshare(*m0, M_NOWAIT); 9685 if (*m0 == NULL) 9686 return (PF_DROP); 9687 } 9688 9689 /* Stateless! */ 9690 return (pf_test_eth_rule(dir, kif, m0)); 9691 } 9692 9693 static __inline void 9694 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9695 { 9696 struct m_tag *mtag; 9697 9698 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9699 9700 /* dummynet adds this tag, but pf does not need it, 9701 * and keeping it creates unexpected behavior, 9702 * e.g. in case of divert(4) usage right after dummynet. */ 9703 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9704 if (mtag != NULL) 9705 m_tag_delete(m, mtag); 9706 } 9707 9708 static int 9709 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9710 struct pf_krule *r, struct mbuf **m0) 9711 { 9712 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9713 } 9714 9715 static int 9716 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9717 struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa, 9718 struct mbuf **m0) 9719 { 9720 struct ip_fw_args dnflow; 9721 9722 NET_EPOCH_ASSERT(); 9723 9724 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9725 return (0); 9726 9727 if (ip_dn_io_ptr == NULL) { 9728 m_freem(*m0); 9729 *m0 = NULL; 9730 return (ENOMEM); 9731 } 9732 9733 if (pd->pf_mtag == NULL && 9734 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 9735 m_freem(*m0); 9736 *m0 = NULL; 9737 return (ENOMEM); 9738 } 9739 9740 if (ifp != NULL) { 9741 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 9742 9743 pd->pf_mtag->if_index = ifp->if_index; 9744 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 9745 9746 MPASS(sa != NULL); 9747 9748 switch (sa->sa_family) { 9749 case AF_INET: 9750 memcpy(&pd->pf_mtag->dst, sa, 9751 sizeof(struct sockaddr_in)); 9752 break; 9753 case AF_INET6: 9754 memcpy(&pd->pf_mtag->dst, sa, 9755 sizeof(struct sockaddr_in6)); 9756 break; 9757 } 9758 } 9759 9760 if (s != NULL && s->nat_rule != NULL && 9761 s->nat_rule->action == PF_RDR && 9762 ( 9763 #ifdef INET 9764 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 9765 #endif /* INET */ 9766 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 9767 /* 9768 * If we're redirecting to loopback mark this packet 9769 * as being local. Otherwise it might get dropped 9770 * if dummynet re-injects. 9771 */ 9772 (*m0)->m_pkthdr.rcvif = V_loif; 9773 } 9774 9775 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 9776 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 9777 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 9778 ip_dn_io_ptr(m0, &dnflow); 9779 if (*m0 != NULL) { 9780 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9781 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 9782 } 9783 } 9784 9785 return (0); 9786 } 9787 9788 static int 9789 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 9790 { 9791 struct ah ext; 9792 u_int32_t hlen, end; 9793 int hdr_cnt; 9794 9795 hlen = h->ip_hl << 2; 9796 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 9797 REASON_SET(reason, PFRES_SHORT); 9798 return (PF_DROP); 9799 } 9800 if (hlen != sizeof(struct ip)) 9801 pd->badopts++; 9802 end = pd->off + ntohs(h->ip_len); 9803 pd->off += hlen; 9804 pd->proto = h->ip_p; 9805 /* stop walking over non initial fragments */ 9806 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 9807 return (PF_PASS); 9808 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 9809 switch (pd->proto) { 9810 case IPPROTO_AH: 9811 /* fragments may be short */ 9812 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 9813 end < pd->off + sizeof(ext)) 9814 return (PF_PASS); 9815 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9816 NULL, reason, AF_INET)) { 9817 DPFPRINTF(PF_DEBUG_MISC, ("IP short exthdr")); 9818 return (PF_DROP); 9819 } 9820 pd->off += (ext.ah_len + 2) * 4; 9821 pd->proto = ext.ah_nxt; 9822 break; 9823 default: 9824 return (PF_PASS); 9825 } 9826 } 9827 DPFPRINTF(PF_DEBUG_MISC, ("IPv4 nested authentication header limit")); 9828 REASON_SET(reason, PFRES_IPOPTIONS); 9829 return (PF_DROP); 9830 } 9831 9832 #ifdef INET6 9833 static int 9834 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 9835 u_short *reason) 9836 { 9837 struct ip6_opt opt; 9838 struct ip6_opt_jumbo jumbo; 9839 9840 while (off < end) { 9841 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 9842 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 9843 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 9844 return (PF_DROP); 9845 } 9846 if (opt.ip6o_type == IP6OPT_PAD1) { 9847 off++; 9848 continue; 9849 } 9850 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL, 9851 reason, AF_INET6)) { 9852 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 9853 return (PF_DROP); 9854 } 9855 if (off + sizeof(opt) + opt.ip6o_len > end) { 9856 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 9857 REASON_SET(reason, PFRES_IPOPTIONS); 9858 return (PF_DROP); 9859 } 9860 switch (opt.ip6o_type) { 9861 case IP6OPT_JUMBO: 9862 if (pd->jumbolen != 0) { 9863 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 9864 REASON_SET(reason, PFRES_IPOPTIONS); 9865 return (PF_DROP); 9866 } 9867 if (ntohs(h->ip6_plen) != 0) { 9868 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 9869 REASON_SET(reason, PFRES_IPOPTIONS); 9870 return (PF_DROP); 9871 } 9872 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL, 9873 reason, AF_INET6)) { 9874 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 9875 return (PF_DROP); 9876 } 9877 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 9878 sizeof(pd->jumbolen)); 9879 pd->jumbolen = ntohl(pd->jumbolen); 9880 if (pd->jumbolen < IPV6_MAXPACKET) { 9881 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 9882 REASON_SET(reason, PFRES_IPOPTIONS); 9883 return (PF_DROP); 9884 } 9885 break; 9886 default: 9887 break; 9888 } 9889 off += sizeof(opt) + opt.ip6o_len; 9890 } 9891 9892 return (PF_PASS); 9893 } 9894 9895 int 9896 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 9897 { 9898 struct ip6_frag frag; 9899 struct ip6_ext ext; 9900 struct ip6_rthdr rthdr; 9901 uint32_t end; 9902 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 9903 9904 pd->off += sizeof(struct ip6_hdr); 9905 end = pd->off + ntohs(h->ip6_plen); 9906 pd->fragoff = pd->extoff = pd->jumbolen = 0; 9907 pd->proto = h->ip6_nxt; 9908 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 9909 switch (pd->proto) { 9910 case IPPROTO_ROUTING: 9911 case IPPROTO_HOPOPTS: 9912 case IPPROTO_DSTOPTS: 9913 pd->badopts++; 9914 break; 9915 } 9916 switch (pd->proto) { 9917 case IPPROTO_FRAGMENT: 9918 if (fraghdr_cnt++) { 9919 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 9920 REASON_SET(reason, PFRES_FRAG); 9921 return (PF_DROP); 9922 } 9923 /* jumbo payload packets cannot be fragmented */ 9924 if (pd->jumbolen != 0) { 9925 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 9926 REASON_SET(reason, PFRES_FRAG); 9927 return (PF_DROP); 9928 } 9929 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 9930 NULL, reason, AF_INET6)) { 9931 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 9932 return (PF_DROP); 9933 } 9934 /* stop walking over non initial fragments */ 9935 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 9936 pd->fragoff = pd->off; 9937 return (PF_PASS); 9938 } 9939 /* RFC6946: reassemble only non atomic fragments */ 9940 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 9941 pd->fragoff = pd->off; 9942 pd->off += sizeof(frag); 9943 pd->proto = frag.ip6f_nxt; 9944 break; 9945 case IPPROTO_ROUTING: 9946 if (rthdr_cnt++) { 9947 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 9948 REASON_SET(reason, PFRES_IPOPTIONS); 9949 return (PF_DROP); 9950 } 9951 /* fragments may be short */ 9952 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 9953 pd->off = pd->fragoff; 9954 pd->proto = IPPROTO_FRAGMENT; 9955 return (PF_PASS); 9956 } 9957 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 9958 NULL, reason, AF_INET6)) { 9959 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 9960 return (PF_DROP); 9961 } 9962 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9963 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 9964 REASON_SET(reason, PFRES_IPOPTIONS); 9965 return (PF_DROP); 9966 } 9967 /* FALLTHROUGH */ 9968 case IPPROTO_HOPOPTS: 9969 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 9970 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 9971 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 hopopts not first")); 9972 REASON_SET(reason, PFRES_IPOPTIONS); 9973 return (PF_DROP); 9974 } 9975 /* FALLTHROUGH */ 9976 case IPPROTO_AH: 9977 case IPPROTO_DSTOPTS: 9978 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9979 NULL, reason, AF_INET6)) { 9980 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 9981 return (PF_DROP); 9982 } 9983 /* fragments may be short */ 9984 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 9985 pd->off = pd->fragoff; 9986 pd->proto = IPPROTO_FRAGMENT; 9987 return (PF_PASS); 9988 } 9989 /* reassembly needs the ext header before the frag */ 9990 if (pd->fragoff == 0) 9991 pd->extoff = pd->off; 9992 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0) { 9993 if (pf_walk_option6(pd, h, 9994 pd->off + sizeof(ext), 9995 pd->off + (ext.ip6e_len + 1) * 8, reason) 9996 != PF_PASS) 9997 return (PF_DROP); 9998 if (ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 9999 DPFPRINTF(PF_DEBUG_MISC, 10000 ("IPv6 missing jumbo")); 10001 REASON_SET(reason, PFRES_IPOPTIONS); 10002 return (PF_DROP); 10003 } 10004 } 10005 if (pd->proto == IPPROTO_AH) 10006 pd->off += (ext.ip6e_len + 2) * 4; 10007 else 10008 pd->off += (ext.ip6e_len + 1) * 8; 10009 pd->proto = ext.ip6e_nxt; 10010 break; 10011 case IPPROTO_TCP: 10012 case IPPROTO_UDP: 10013 case IPPROTO_SCTP: 10014 case IPPROTO_ICMPV6: 10015 /* fragments may be short, ignore inner header then */ 10016 if (pd->fragoff != 0 && end < pd->off + 10017 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 10018 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 10019 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 10020 sizeof(struct icmp6_hdr))) { 10021 pd->off = pd->fragoff; 10022 pd->proto = IPPROTO_FRAGMENT; 10023 } 10024 /* FALLTHROUGH */ 10025 default: 10026 return (PF_PASS); 10027 } 10028 } 10029 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 nested extension header limit")); 10030 REASON_SET(reason, PFRES_IPOPTIONS); 10031 return (PF_DROP); 10032 } 10033 #endif /* INET6 */ 10034 10035 static void 10036 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 10037 { 10038 memset(pd, 0, sizeof(*pd)); 10039 pd->pf_mtag = pf_find_mtag(m); 10040 pd->m = m; 10041 } 10042 10043 static int 10044 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 10045 u_short *action, u_short *reason, struct pfi_kkif *kif, 10046 struct pf_rule_actions *default_actions) 10047 { 10048 pd->dir = dir; 10049 pd->kif = kif; 10050 pd->m = *m0; 10051 pd->sidx = (dir == PF_IN) ? 0 : 1; 10052 pd->didx = (dir == PF_IN) ? 1 : 0; 10053 pd->af = pd->naf = af; 10054 10055 TAILQ_INIT(&pd->sctp_multihome_jobs); 10056 if (default_actions != NULL) 10057 memcpy(&pd->act, default_actions, sizeof(pd->act)); 10058 10059 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 10060 pd->act.dnpipe = pd->pf_mtag->dnpipe; 10061 pd->act.flags = pd->pf_mtag->dnflags; 10062 } 10063 10064 switch (af) { 10065 #ifdef INET 10066 case AF_INET: { 10067 struct ip *h; 10068 10069 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 10070 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 10071 DPFPRINTF(PF_DEBUG_URGENT, 10072 ("%s: m_len < sizeof(struct ip), pullup failed\n", 10073 __func__)); 10074 *action = PF_DROP; 10075 REASON_SET(reason, PFRES_SHORT); 10076 return (-1); 10077 } 10078 10079 if (pf_normalize_ip(reason, pd) != PF_PASS) { 10080 /* We do IP header normalization and packet reassembly here */ 10081 *m0 = pd->m; 10082 *action = PF_DROP; 10083 return (-1); 10084 } 10085 *m0 = pd->m; 10086 10087 h = mtod(pd->m, struct ip *); 10088 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 10089 *action = PF_DROP; 10090 REASON_SET(reason, PFRES_SHORT); 10091 return (-1); 10092 } 10093 10094 if (pf_walk_header(pd, h, reason) != PF_PASS) { 10095 *action = PF_DROP; 10096 return (-1); 10097 } 10098 10099 pd->src = (struct pf_addr *)&h->ip_src; 10100 pd->dst = (struct pf_addr *)&h->ip_dst; 10101 pf_addrcpy(&pd->osrc, pd->src, af); 10102 pf_addrcpy(&pd->odst, pd->dst, af); 10103 pd->ip_sum = &h->ip_sum; 10104 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 10105 pd->ttl = h->ip_ttl; 10106 pd->tot_len = ntohs(h->ip_len); 10107 pd->act.rtableid = -1; 10108 pd->df = h->ip_off & htons(IP_DF); 10109 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 10110 PF_VPROTO_FRAGMENT : pd->proto; 10111 10112 break; 10113 } 10114 #endif /* INET */ 10115 #ifdef INET6 10116 case AF_INET6: { 10117 struct ip6_hdr *h; 10118 10119 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 10120 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 10121 DPFPRINTF(PF_DEBUG_URGENT, 10122 ("%s: m_len < sizeof(struct ip6_hdr)" 10123 ", pullup failed\n", __func__)); 10124 *action = PF_DROP; 10125 REASON_SET(reason, PFRES_SHORT); 10126 return (-1); 10127 } 10128 10129 h = mtod(pd->m, struct ip6_hdr *); 10130 10131 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10132 *action = PF_DROP; 10133 return (-1); 10134 } 10135 10136 h = mtod(pd->m, struct ip6_hdr *); 10137 pd->src = (struct pf_addr *)&h->ip6_src; 10138 pd->dst = (struct pf_addr *)&h->ip6_dst; 10139 pf_addrcpy(&pd->osrc, pd->src, af); 10140 pf_addrcpy(&pd->odst, pd->dst, af); 10141 pd->ip_sum = NULL; 10142 pd->tos = IPV6_DSCP(h); 10143 pd->ttl = h->ip6_hlim; 10144 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 10145 pd->act.rtableid = -1; 10146 10147 pd->virtual_proto = (pd->fragoff != 0) ? 10148 PF_VPROTO_FRAGMENT : pd->proto; 10149 10150 /* 10151 * we do not support jumbogram. if we keep going, zero ip6_plen 10152 * will do something bad, so drop the packet for now. 10153 */ 10154 if (htons(h->ip6_plen) == 0) { 10155 *action = PF_DROP; 10156 return (-1); 10157 } 10158 10159 /* We do IP header normalization and packet reassembly here */ 10160 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 10161 PF_PASS) { 10162 *m0 = pd->m; 10163 *action = PF_DROP; 10164 return (-1); 10165 } 10166 *m0 = pd->m; 10167 if (pd->m == NULL) { 10168 /* packet sits in reassembly queue, no error */ 10169 *action = PF_PASS; 10170 return (-1); 10171 } 10172 10173 /* Update pointers into the packet. */ 10174 h = mtod(pd->m, struct ip6_hdr *); 10175 pd->src = (struct pf_addr *)&h->ip6_src; 10176 pd->dst = (struct pf_addr *)&h->ip6_dst; 10177 10178 pd->off = 0; 10179 10180 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10181 *action = PF_DROP; 10182 return (-1); 10183 } 10184 10185 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 10186 /* 10187 * Reassembly may have changed the next protocol from 10188 * fragment to something else, so update. 10189 */ 10190 pd->virtual_proto = pd->proto; 10191 MPASS(pd->fragoff == 0); 10192 } 10193 10194 if (pd->fragoff != 0) 10195 pd->virtual_proto = PF_VPROTO_FRAGMENT; 10196 10197 break; 10198 } 10199 #endif /* INET6 */ 10200 default: 10201 panic("pf_setup_pdesc called with illegal af %u", af); 10202 } 10203 10204 switch (pd->virtual_proto) { 10205 case IPPROTO_TCP: { 10206 struct tcphdr *th = &pd->hdr.tcp; 10207 10208 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 10209 reason, af)) { 10210 *action = PF_DROP; 10211 REASON_SET(reason, PFRES_SHORT); 10212 return (-1); 10213 } 10214 pd->hdrlen = sizeof(*th); 10215 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 10216 pd->sport = &th->th_sport; 10217 pd->dport = &th->th_dport; 10218 pd->pcksum = &th->th_sum; 10219 break; 10220 } 10221 case IPPROTO_UDP: { 10222 struct udphdr *uh = &pd->hdr.udp; 10223 10224 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 10225 reason, af)) { 10226 *action = PF_DROP; 10227 REASON_SET(reason, PFRES_SHORT); 10228 return (-1); 10229 } 10230 pd->hdrlen = sizeof(*uh); 10231 if (uh->uh_dport == 0 || 10232 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 10233 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 10234 *action = PF_DROP; 10235 REASON_SET(reason, PFRES_SHORT); 10236 return (-1); 10237 } 10238 pd->sport = &uh->uh_sport; 10239 pd->dport = &uh->uh_dport; 10240 pd->pcksum = &uh->uh_sum; 10241 break; 10242 } 10243 case IPPROTO_SCTP: { 10244 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 10245 action, reason, af)) { 10246 *action = PF_DROP; 10247 REASON_SET(reason, PFRES_SHORT); 10248 return (-1); 10249 } 10250 pd->hdrlen = sizeof(pd->hdr.sctp); 10251 pd->p_len = pd->tot_len - pd->off; 10252 10253 pd->sport = &pd->hdr.sctp.src_port; 10254 pd->dport = &pd->hdr.sctp.dest_port; 10255 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 10256 *action = PF_DROP; 10257 REASON_SET(reason, PFRES_SHORT); 10258 return (-1); 10259 } 10260 if (pf_scan_sctp(pd) != PF_PASS) { 10261 *action = PF_DROP; 10262 REASON_SET(reason, PFRES_SHORT); 10263 return (-1); 10264 } 10265 /* 10266 * Placeholder. The SCTP checksum is 32-bits, but 10267 * pf_test_state() expects to update a 16-bit checksum. 10268 * Provide a dummy value which we'll subsequently ignore. 10269 */ 10270 pd->pcksum = &pd->sctp_dummy_sum; 10271 break; 10272 } 10273 case IPPROTO_ICMP: { 10274 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 10275 action, reason, af)) { 10276 *action = PF_DROP; 10277 REASON_SET(reason, PFRES_SHORT); 10278 return (-1); 10279 } 10280 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 10281 pd->hdrlen = ICMP_MINLEN; 10282 break; 10283 } 10284 #ifdef INET6 10285 case IPPROTO_ICMPV6: { 10286 size_t icmp_hlen = sizeof(struct icmp6_hdr); 10287 10288 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10289 action, reason, af)) { 10290 *action = PF_DROP; 10291 REASON_SET(reason, PFRES_SHORT); 10292 return (-1); 10293 } 10294 /* ICMP headers we look further into to match state */ 10295 switch (pd->hdr.icmp6.icmp6_type) { 10296 case MLD_LISTENER_QUERY: 10297 case MLD_LISTENER_REPORT: 10298 icmp_hlen = sizeof(struct mld_hdr); 10299 break; 10300 case ND_NEIGHBOR_SOLICIT: 10301 case ND_NEIGHBOR_ADVERT: 10302 icmp_hlen = sizeof(struct nd_neighbor_solicit); 10303 /* FALLTHROUGH */ 10304 case ND_ROUTER_SOLICIT: 10305 case ND_ROUTER_ADVERT: 10306 case ND_REDIRECT: 10307 if (pd->ttl != 255) { 10308 REASON_SET(reason, PFRES_NORM); 10309 return (PF_DROP); 10310 } 10311 break; 10312 } 10313 if (icmp_hlen > sizeof(struct icmp6_hdr) && 10314 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10315 action, reason, af)) { 10316 *action = PF_DROP; 10317 REASON_SET(reason, PFRES_SHORT); 10318 return (-1); 10319 } 10320 pd->hdrlen = icmp_hlen; 10321 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 10322 break; 10323 } 10324 #endif /* INET6 */ 10325 } 10326 10327 if (pd->sport) 10328 pd->osport = pd->nsport = *pd->sport; 10329 if (pd->dport) 10330 pd->odport = pd->ndport = *pd->dport; 10331 10332 return (0); 10333 } 10334 10335 static void 10336 pf_counters_inc(int action, struct pf_pdesc *pd, 10337 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 10338 { 10339 struct pf_krule *tr; 10340 int dir = pd->dir; 10341 int dirndx; 10342 10343 pf_counter_u64_critical_enter(); 10344 pf_counter_u64_add_protected( 10345 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10346 pd->tot_len); 10347 pf_counter_u64_add_protected( 10348 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10349 1); 10350 10351 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 10352 dirndx = (dir == PF_OUT); 10353 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 10354 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 10355 pf_update_timestamp(r); 10356 10357 if (a != NULL) { 10358 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 10359 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 10360 } 10361 if (s != NULL) { 10362 struct pf_krule_item *ri; 10363 10364 if (s->nat_rule != NULL) { 10365 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 10366 1); 10367 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 10368 pd->tot_len); 10369 } 10370 /* 10371 * Source nodes are accessed unlocked here. 10372 * But since we are operating with stateful tracking 10373 * and the state is locked, those SNs could not have 10374 * been freed. 10375 */ 10376 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 10377 if (s->sns[sn_type] != NULL) { 10378 counter_u64_add( 10379 s->sns[sn_type]->packets[dirndx], 10380 1); 10381 counter_u64_add( 10382 s->sns[sn_type]->bytes[dirndx], 10383 pd->tot_len); 10384 } 10385 } 10386 dirndx = (dir == s->direction) ? 0 : 1; 10387 s->packets[dirndx]++; 10388 s->bytes[dirndx] += pd->tot_len; 10389 10390 SLIST_FOREACH(ri, &s->match_rules, entry) { 10391 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 10392 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 10393 10394 if (ri->r->src.addr.type == PF_ADDR_TABLE) 10395 pfr_update_stats(ri->r->src.addr.p.tbl, 10396 (s == NULL) ? pd->src : 10397 &s->key[(s->direction == PF_IN)]-> 10398 addr[(s->direction == PF_OUT)], 10399 pd->af, pd->tot_len, dir == PF_OUT, 10400 r->action == PF_PASS, ri->r->src.neg); 10401 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 10402 pfr_update_stats(ri->r->dst.addr.p.tbl, 10403 (s == NULL) ? pd->dst : 10404 &s->key[(s->direction == PF_IN)]-> 10405 addr[(s->direction == PF_IN)], 10406 pd->af, pd->tot_len, dir == PF_OUT, 10407 r->action == PF_PASS, ri->r->dst.neg); 10408 } 10409 } 10410 10411 tr = r; 10412 if (s != NULL && s->nat_rule != NULL && 10413 r == &V_pf_default_rule) 10414 tr = s->nat_rule; 10415 10416 if (tr->src.addr.type == PF_ADDR_TABLE) 10417 pfr_update_stats(tr->src.addr.p.tbl, 10418 (s == NULL) ? pd->src : 10419 &s->key[(s->direction == PF_IN)]-> 10420 addr[(s->direction == PF_OUT)], 10421 pd->af, pd->tot_len, dir == PF_OUT, 10422 r->action == PF_PASS, tr->src.neg); 10423 if (tr->dst.addr.type == PF_ADDR_TABLE) 10424 pfr_update_stats(tr->dst.addr.p.tbl, 10425 (s == NULL) ? pd->dst : 10426 &s->key[(s->direction == PF_IN)]-> 10427 addr[(s->direction == PF_IN)], 10428 pd->af, pd->tot_len, dir == PF_OUT, 10429 r->action == PF_PASS, tr->dst.neg); 10430 } 10431 pf_counter_u64_critical_exit(); 10432 } 10433 static void 10434 pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm, 10435 struct pf_krule *am, struct pf_kruleset *ruleset, 10436 struct pf_krule_slist *matchrules) 10437 { 10438 struct pf_krule_item *ri; 10439 10440 /* if this is the log(matches) rule, packet has been logged already */ 10441 if (rm->log & PF_LOG_MATCHES) 10442 return; 10443 10444 SLIST_FOREACH(ri, matchrules, entry) 10445 if (ri->r->log & PF_LOG_MATCHES) 10446 PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am, 10447 ruleset, pd, 1, ri->r); 10448 } 10449 10450 #if defined(INET) || defined(INET6) 10451 int 10452 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10453 struct inpcb *inp, struct pf_rule_actions *default_actions) 10454 { 10455 struct pfi_kkif *kif; 10456 u_short action, reason = 0; 10457 struct m_tag *mtag; 10458 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10459 struct pf_kstate *s = NULL; 10460 struct pf_kruleset *ruleset = NULL; 10461 struct pf_pdesc pd; 10462 int use_2nd_queue = 0; 10463 uint16_t tag; 10464 10465 PF_RULES_RLOCK_TRACKER; 10466 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10467 M_ASSERTPKTHDR(*m0); 10468 10469 if (!V_pf_status.running) 10470 return (PF_PASS); 10471 10472 PF_RULES_RLOCK(); 10473 10474 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10475 10476 if (__predict_false(kif == NULL)) { 10477 DPFPRINTF(PF_DEBUG_URGENT, 10478 ("%s: kif == NULL, if_xname %s\n", 10479 __func__, ifp->if_xname)); 10480 PF_RULES_RUNLOCK(); 10481 return (PF_DROP); 10482 } 10483 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10484 PF_RULES_RUNLOCK(); 10485 return (PF_PASS); 10486 } 10487 10488 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10489 PF_RULES_RUNLOCK(); 10490 return (PF_PASS); 10491 } 10492 10493 if (__predict_false(! M_WRITABLE(*m0))) { 10494 *m0 = m_unshare(*m0, M_NOWAIT); 10495 if (*m0 == NULL) { 10496 PF_RULES_RUNLOCK(); 10497 return (PF_DROP); 10498 } 10499 } 10500 10501 pf_init_pdesc(&pd, *m0); 10502 10503 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10504 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10505 10506 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10507 pd.pf_mtag->if_idxgen); 10508 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10509 PF_RULES_RUNLOCK(); 10510 m_freem(*m0); 10511 *m0 = NULL; 10512 return (PF_PASS); 10513 } 10514 PF_RULES_RUNLOCK(); 10515 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10516 *m0 = NULL; 10517 return (PF_PASS); 10518 } 10519 10520 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10521 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10522 /* Dummynet re-injects packets after they've 10523 * completed their delay. We've already 10524 * processed them, so pass unconditionally. */ 10525 10526 /* But only once. We may see the packet multiple times (e.g. 10527 * PFIL_IN/PFIL_OUT). */ 10528 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 10529 PF_RULES_RUNLOCK(); 10530 10531 return (PF_PASS); 10532 } 10533 10534 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 10535 kif, default_actions) == -1) { 10536 if (action != PF_PASS) 10537 pd.act.log |= PF_LOG_FORCE; 10538 goto done; 10539 } 10540 10541 #ifdef INET 10542 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 10543 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 10544 PF_RULES_RUNLOCK(); 10545 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 10546 0, ifp->if_mtu); 10547 *m0 = NULL; 10548 return (PF_DROP); 10549 } 10550 #endif /* INET */ 10551 #ifdef INET6 10552 /* 10553 * If we end up changing IP addresses (e.g. binat) the stack may get 10554 * confused and fail to send the icmp6 packet too big error. Just send 10555 * it here, before we do any NAT. 10556 */ 10557 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 10558 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 10559 PF_RULES_RUNLOCK(); 10560 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 10561 *m0 = NULL; 10562 return (PF_DROP); 10563 } 10564 #endif /* INET6 */ 10565 10566 if (__predict_false(ip_divert_ptr != NULL) && 10567 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 10568 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 10569 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 10570 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 10571 if (pd.pf_mtag == NULL && 10572 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10573 action = PF_DROP; 10574 goto done; 10575 } 10576 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 10577 } 10578 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 10579 pd.m->m_flags |= M_FASTFWD_OURS; 10580 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10581 } 10582 m_tag_delete(pd.m, mtag); 10583 10584 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 10585 if (mtag != NULL) 10586 m_tag_delete(pd.m, mtag); 10587 } 10588 10589 switch (pd.virtual_proto) { 10590 case PF_VPROTO_FRAGMENT: 10591 /* 10592 * handle fragments that aren't reassembled by 10593 * normalization 10594 */ 10595 if (kif == NULL || r == NULL) /* pflog */ 10596 action = PF_DROP; 10597 else 10598 action = pf_test_rule(&r, &s, &pd, &a, 10599 &ruleset, &reason, inp); 10600 if (action != PF_PASS) 10601 REASON_SET(&reason, PFRES_FRAG); 10602 break; 10603 10604 case IPPROTO_TCP: { 10605 /* Respond to SYN with a syncookie. */ 10606 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 10607 pd.dir == PF_IN && pf_synflood_check(&pd)) { 10608 pf_syncookie_send(&pd); 10609 action = PF_DROP; 10610 break; 10611 } 10612 10613 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 10614 use_2nd_queue = 1; 10615 action = pf_normalize_tcp(&pd); 10616 if (action == PF_DROP) 10617 break; 10618 action = pf_test_state(&s, &pd, &reason); 10619 if (action == PF_PASS || action == PF_AFRT) { 10620 if (V_pfsync_update_state_ptr != NULL) 10621 V_pfsync_update_state_ptr(s); 10622 r = s->rule; 10623 a = s->anchor; 10624 } else if (s == NULL) { 10625 /* Validate remote SYN|ACK, re-create original SYN if 10626 * valid. */ 10627 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 10628 TH_ACK && pf_syncookie_validate(&pd) && 10629 pd.dir == PF_IN) { 10630 struct mbuf *msyn; 10631 10632 msyn = pf_syncookie_recreate_syn(&pd); 10633 if (msyn == NULL) { 10634 action = PF_DROP; 10635 break; 10636 } 10637 10638 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 10639 &pd.act); 10640 m_freem(msyn); 10641 if (action != PF_PASS) 10642 break; 10643 10644 action = pf_test_state(&s, &pd, &reason); 10645 if (action != PF_PASS || s == NULL) { 10646 action = PF_DROP; 10647 break; 10648 } 10649 10650 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 10651 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 10652 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 10653 action = pf_synproxy(&pd, s, &reason); 10654 break; 10655 } else { 10656 action = pf_test_rule(&r, &s, &pd, 10657 &a, &ruleset, &reason, inp); 10658 } 10659 } 10660 break; 10661 } 10662 10663 case IPPROTO_SCTP: 10664 action = pf_normalize_sctp(&pd); 10665 if (action == PF_DROP) 10666 break; 10667 /* fallthrough */ 10668 case IPPROTO_UDP: 10669 default: 10670 action = pf_test_state(&s, &pd, &reason); 10671 if (action == PF_PASS || action == PF_AFRT) { 10672 if (V_pfsync_update_state_ptr != NULL) 10673 V_pfsync_update_state_ptr(s); 10674 r = s->rule; 10675 a = s->anchor; 10676 } else if (s == NULL) { 10677 action = pf_test_rule(&r, &s, 10678 &pd, &a, &ruleset, &reason, inp); 10679 } 10680 break; 10681 10682 case IPPROTO_ICMP: 10683 case IPPROTO_ICMPV6: { 10684 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 10685 action = PF_DROP; 10686 REASON_SET(&reason, PFRES_NORM); 10687 DPFPRINTF(PF_DEBUG_MISC, 10688 ("dropping IPv6 packet with ICMPv4 payload")); 10689 break; 10690 } 10691 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 10692 action = PF_DROP; 10693 REASON_SET(&reason, PFRES_NORM); 10694 DPFPRINTF(PF_DEBUG_MISC, 10695 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 10696 break; 10697 } 10698 action = pf_test_state_icmp(&s, &pd, &reason); 10699 if (action == PF_PASS || action == PF_AFRT) { 10700 if (V_pfsync_update_state_ptr != NULL) 10701 V_pfsync_update_state_ptr(s); 10702 r = s->rule; 10703 a = s->anchor; 10704 } else if (s == NULL) 10705 action = pf_test_rule(&r, &s, &pd, 10706 &a, &ruleset, &reason, inp); 10707 break; 10708 } 10709 10710 } 10711 10712 done: 10713 PF_RULES_RUNLOCK(); 10714 10715 if (pd.m == NULL) 10716 goto eat_pkt; 10717 10718 if (s) 10719 memcpy(&pd.act, &s->act, sizeof(s->act)); 10720 10721 if (action == PF_PASS && pd.badopts && !pd.act.allow_opts) { 10722 action = PF_DROP; 10723 REASON_SET(&reason, PFRES_IPOPTIONS); 10724 pd.act.log = PF_LOG_FORCE; 10725 DPFPRINTF(PF_DEBUG_MISC, 10726 ("pf: dropping packet with dangerous headers\n")); 10727 } 10728 10729 if (pd.act.max_pkt_size && pd.act.max_pkt_size && 10730 pd.tot_len > pd.act.max_pkt_size) { 10731 action = PF_DROP; 10732 REASON_SET(&reason, PFRES_NORM); 10733 pd.act.log = PF_LOG_FORCE; 10734 DPFPRINTF(PF_DEBUG_MISC, 10735 ("pf: dropping overly long packet\n")); 10736 } 10737 10738 if (s) { 10739 uint8_t log = pd.act.log; 10740 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 10741 pd.act.log |= log; 10742 tag = s->tag; 10743 } else { 10744 tag = r->tag; 10745 } 10746 10747 if (tag > 0 && pf_tag_packet(&pd, tag)) { 10748 action = PF_DROP; 10749 REASON_SET(&reason, PFRES_MEMORY); 10750 } 10751 10752 pf_scrub(&pd); 10753 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 10754 pf_normalize_mss(&pd); 10755 10756 if (pd.act.rtableid >= 0) 10757 M_SETFIB(pd.m, pd.act.rtableid); 10758 10759 if (pd.act.flags & PFSTATE_SETPRIO) { 10760 if (pd.tos & IPTOS_LOWDELAY) 10761 use_2nd_queue = 1; 10762 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 10763 action = PF_DROP; 10764 REASON_SET(&reason, PFRES_MEMORY); 10765 pd.act.log = PF_LOG_FORCE; 10766 DPFPRINTF(PF_DEBUG_MISC, 10767 ("pf: failed to allocate 802.1q mtag\n")); 10768 } 10769 } 10770 10771 #ifdef ALTQ 10772 if (action == PF_PASS && pd.act.qid) { 10773 if (pd.pf_mtag == NULL && 10774 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10775 action = PF_DROP; 10776 REASON_SET(&reason, PFRES_MEMORY); 10777 } else { 10778 if (s != NULL) 10779 pd.pf_mtag->qid_hash = pf_state_hash(s); 10780 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 10781 pd.pf_mtag->qid = pd.act.pqid; 10782 else 10783 pd.pf_mtag->qid = pd.act.qid; 10784 /* Add hints for ecn. */ 10785 pd.pf_mtag->hdr = mtod(pd.m, void *); 10786 } 10787 } 10788 #endif /* ALTQ */ 10789 10790 /* 10791 * connections redirected to loopback should not match sockets 10792 * bound specifically to loopback due to security implications, 10793 * see tcp_input() and in_pcblookup_listen(). 10794 */ 10795 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 10796 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 10797 (s->nat_rule->action == PF_RDR || 10798 s->nat_rule->action == PF_BINAT) && 10799 pf_is_loopback(af, pd.dst)) 10800 pd.m->m_flags |= M_SKIP_FIREWALL; 10801 10802 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 10803 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 10804 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 10805 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 10806 if (mtag != NULL) { 10807 ((struct pf_divert_mtag *)(mtag+1))->port = 10808 ntohs(r->divert.port); 10809 ((struct pf_divert_mtag *)(mtag+1))->idir = 10810 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 10811 PF_DIVERT_MTAG_DIR_OUT; 10812 10813 if (s) 10814 PF_STATE_UNLOCK(s); 10815 10816 m_tag_prepend(pd.m, mtag); 10817 if (pd.m->m_flags & M_FASTFWD_OURS) { 10818 if (pd.pf_mtag == NULL && 10819 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10820 action = PF_DROP; 10821 REASON_SET(&reason, PFRES_MEMORY); 10822 pd.act.log = PF_LOG_FORCE; 10823 DPFPRINTF(PF_DEBUG_MISC, 10824 ("pf: failed to allocate tag\n")); 10825 } else { 10826 pd.pf_mtag->flags |= 10827 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10828 pd.m->m_flags &= ~M_FASTFWD_OURS; 10829 } 10830 } 10831 ip_divert_ptr(*m0, dir == PF_IN); 10832 *m0 = NULL; 10833 10834 return (action); 10835 } else { 10836 /* XXX: ipfw has the same behaviour! */ 10837 action = PF_DROP; 10838 REASON_SET(&reason, PFRES_MEMORY); 10839 pd.act.log = PF_LOG_FORCE; 10840 DPFPRINTF(PF_DEBUG_MISC, 10841 ("pf: failed to allocate divert tag\n")); 10842 } 10843 } 10844 /* XXX: Anybody working on it?! */ 10845 if (af == AF_INET6 && r->divert.port) 10846 printf("pf: divert(9) is not supported for IPv6\n"); 10847 10848 /* this flag will need revising if the pkt is forwarded */ 10849 if (pd.pf_mtag) 10850 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 10851 10852 if (pd.act.log) { 10853 struct pf_krule *lr; 10854 struct pf_krule_item *ri; 10855 10856 if (s != NULL && s->nat_rule != NULL && 10857 s->nat_rule->log & PF_LOG_ALL) 10858 lr = s->nat_rule; 10859 else 10860 lr = r; 10861 10862 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 10863 PFLOG_PACKET(action, reason, lr, a, 10864 ruleset, &pd, (s == NULL), NULL); 10865 if (s) { 10866 SLIST_FOREACH(ri, &s->match_rules, entry) 10867 if (ri->r->log & PF_LOG_ALL) 10868 PFLOG_PACKET(action, 10869 reason, ri->r, a, ruleset, &pd, 0, NULL); 10870 } 10871 } 10872 10873 pf_counters_inc(action, &pd, s, r, a); 10874 10875 switch (action) { 10876 case PF_SYNPROXY_DROP: 10877 m_freem(*m0); 10878 case PF_DEFER: 10879 *m0 = NULL; 10880 action = PF_PASS; 10881 break; 10882 case PF_DROP: 10883 m_freem(*m0); 10884 *m0 = NULL; 10885 break; 10886 case PF_AFRT: 10887 if (pf_translate_af(&pd)) { 10888 *m0 = pd.m; 10889 action = PF_DROP; 10890 break; 10891 } 10892 #ifdef INET 10893 if (pd.naf == AF_INET) 10894 pf_route(r, kif->pfik_ifp, s, &pd, inp); 10895 #endif /* INET */ 10896 #ifdef INET6 10897 if (pd.naf == AF_INET6) 10898 pf_route6(r, kif->pfik_ifp, s, &pd, inp); 10899 #endif /* INET6 */ 10900 *m0 = pd.m; 10901 action = PF_PASS; 10902 goto out; 10903 break; 10904 default: 10905 if (pd.act.rt) { 10906 switch (af) { 10907 #ifdef INET 10908 case AF_INET: 10909 /* pf_route() returns unlocked. */ 10910 pf_route(r, kif->pfik_ifp, s, &pd, inp); 10911 break; 10912 #endif /* INET */ 10913 #ifdef INET6 10914 case AF_INET6: 10915 /* pf_route6() returns unlocked. */ 10916 pf_route6(r, kif->pfik_ifp, s, &pd, inp); 10917 break; 10918 #endif /* INET6 */ 10919 } 10920 *m0 = pd.m; 10921 goto out; 10922 } 10923 if (pf_dummynet(&pd, s, r, m0) != 0) { 10924 action = PF_DROP; 10925 REASON_SET(&reason, PFRES_MEMORY); 10926 } 10927 break; 10928 } 10929 10930 eat_pkt: 10931 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 10932 10933 if (s && action != PF_DROP) { 10934 if (!s->if_index_in && dir == PF_IN) 10935 s->if_index_in = ifp->if_index; 10936 else if (!s->if_index_out && dir == PF_OUT) 10937 s->if_index_out = ifp->if_index; 10938 } 10939 10940 if (s) 10941 PF_STATE_UNLOCK(s); 10942 10943 out: 10944 #ifdef INET6 10945 /* If reassembled packet passed, create new fragments. */ 10946 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 10947 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 10948 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 10949 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 10950 #endif /* INET6 */ 10951 10952 pf_sctp_multihome_delayed(&pd, kif, s, action); 10953 10954 return (action); 10955 } 10956 #endif /* INET || INET6 */ 10957