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 <machine/in_cksum.h> 118 #include <security/mac/mac_framework.h> 119 120 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 121 122 SDT_PROVIDER_DEFINE(pf); 123 SDT_PROBE_DEFINE2(pf, , test, reason_set, "int", "int"); 124 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 132 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 133 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 134 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 135 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 136 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 137 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 138 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 139 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 140 "struct pf_krule *", "struct mbuf *", "int"); 141 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 142 "struct pf_sctp_source *"); 143 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 144 "struct pf_kstate *", "struct pf_sctp_source *"); 145 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 146 "int", "struct pf_pdesc *", "int"); 147 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 148 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 149 "int"); 150 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 151 "int"); 152 153 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 154 "struct mbuf *"); 155 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 156 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 157 "int", "struct pf_keth_rule *", "char *"); 158 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 159 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 160 "int", "struct pf_keth_rule *"); 161 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 162 163 /* 164 * Global variables 165 */ 166 167 /* state tables */ 168 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 169 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 170 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 173 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 174 VNET_DEFINE(struct pf_kstatus, pf_status); 175 176 VNET_DEFINE(u_int32_t, ticket_altqs_active); 177 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 178 VNET_DEFINE(int, altqs_inactive_open); 179 VNET_DEFINE(u_int32_t, ticket_pabuf); 180 181 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 182 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 183 VNET_DEFINE(u_char, pf_tcp_secret[16]); 184 #define V_pf_tcp_secret VNET(pf_tcp_secret) 185 VNET_DEFINE(int, pf_tcp_secret_init); 186 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 187 VNET_DEFINE(int, pf_tcp_iss_off); 188 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 189 VNET_DECLARE(int, pf_vnet_active); 190 #define V_pf_vnet_active VNET(pf_vnet_active) 191 192 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 193 #define V_pf_purge_idx VNET(pf_purge_idx) 194 195 #ifdef PF_WANT_32_TO_64_COUNTER 196 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 197 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 198 199 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 200 VNET_DEFINE(size_t, pf_allrulecount); 201 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 202 #endif 203 204 struct pf_sctp_endpoint; 205 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 206 struct pf_sctp_source { 207 sa_family_t af; 208 struct pf_addr addr; 209 TAILQ_ENTRY(pf_sctp_source) entry; 210 }; 211 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 212 struct pf_sctp_endpoint 213 { 214 uint32_t v_tag; 215 struct pf_sctp_sources sources; 216 RB_ENTRY(pf_sctp_endpoint) entry; 217 }; 218 static int 219 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 220 { 221 return (a->v_tag - b->v_tag); 222 } 223 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 224 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 225 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 226 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 227 static struct mtx_padalign pf_sctp_endpoints_mtx; 228 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 229 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 230 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 231 232 /* 233 * Queue for pf_intr() sends. 234 */ 235 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 236 struct pf_send_entry { 237 STAILQ_ENTRY(pf_send_entry) pfse_next; 238 struct mbuf *pfse_m; 239 enum { 240 PFSE_IP, 241 PFSE_IP6, 242 PFSE_ICMP, 243 PFSE_ICMP6, 244 } pfse_type; 245 struct { 246 int type; 247 int code; 248 int mtu; 249 } icmpopts; 250 }; 251 252 STAILQ_HEAD(pf_send_head, pf_send_entry); 253 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 254 #define V_pf_sendqueue VNET(pf_sendqueue) 255 256 static struct mtx_padalign pf_sendqueue_mtx; 257 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 258 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 259 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 260 261 /* 262 * Queue for pf_overload_task() tasks. 263 */ 264 struct pf_overload_entry { 265 SLIST_ENTRY(pf_overload_entry) next; 266 struct pf_addr addr; 267 sa_family_t af; 268 uint8_t dir; 269 struct pf_krule *rule; 270 }; 271 272 SLIST_HEAD(pf_overload_head, pf_overload_entry); 273 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 274 #define V_pf_overloadqueue VNET(pf_overloadqueue) 275 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 276 #define V_pf_overloadtask VNET(pf_overloadtask) 277 278 static struct mtx_padalign pf_overloadqueue_mtx; 279 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 280 "pf overload/flush queue", MTX_DEF); 281 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 282 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 283 284 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 285 struct mtx_padalign pf_unlnkdrules_mtx; 286 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 287 MTX_DEF); 288 289 struct sx pf_config_lock; 290 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 291 292 struct mtx_padalign pf_table_stats_lock; 293 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 294 MTX_DEF); 295 296 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 297 #define V_pf_sources_z VNET(pf_sources_z) 298 uma_zone_t pf_mtag_z; 299 VNET_DEFINE(uma_zone_t, pf_state_z); 300 VNET_DEFINE(uma_zone_t, pf_state_key_z); 301 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 302 303 VNET_DEFINE(struct unrhdr64, pf_stateid); 304 305 static void pf_src_tree_remove_state(struct pf_kstate *); 306 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 307 u_int32_t); 308 static void pf_add_threshold(struct pf_threshold *); 309 static int pf_check_threshold(struct pf_threshold *); 310 311 static void pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *, 312 struct pf_addr *, u_int16_t); 313 static int pf_modulate_sack(struct pf_pdesc *, 314 struct tcphdr *, struct pf_state_peer *); 315 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 316 u_int16_t *, u_int16_t *); 317 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 318 struct pf_addr *, struct pf_addr *, u_int16_t, 319 u_int16_t *, u_int16_t *, u_int16_t *, 320 u_int16_t *, u_int8_t, sa_family_t); 321 int pf_change_icmp_af(struct mbuf *, int, 322 struct pf_pdesc *, struct pf_pdesc *, 323 struct pf_addr *, struct pf_addr *, sa_family_t, 324 sa_family_t); 325 int pf_translate_icmp_af(int, void *); 326 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 327 sa_family_t, struct pf_krule *, int); 328 static void pf_detach_state(struct pf_kstate *); 329 static int pf_state_key_attach(struct pf_state_key *, 330 struct pf_state_key *, struct pf_kstate *); 331 static void pf_state_key_detach(struct pf_kstate *, int); 332 static int pf_state_key_ctor(void *, int, void *, int); 333 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 334 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 335 struct pf_mtag *pf_mtag); 336 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 337 struct pf_krule *, struct mbuf **); 338 static int pf_dummynet_route(struct pf_pdesc *, 339 struct pf_kstate *, struct pf_krule *, 340 struct ifnet *, const struct sockaddr *, struct mbuf **); 341 static int pf_test_eth_rule(int, struct pfi_kkif *, 342 struct mbuf **); 343 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 344 struct pf_pdesc *, struct pf_krule **, 345 struct pf_kruleset **, u_short *, struct inpcb *); 346 static int pf_create_state(struct pf_krule *, struct pf_krule *, 347 struct pf_krule *, struct pf_pdesc *, 348 struct pf_state_key *, struct pf_state_key *, int *, 349 struct pf_kstate **, int, u_int16_t, u_int16_t, 350 struct pf_krule_slist *, struct pf_udp_mapping *, 351 u_short *); 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_option6(struct pf_pdesc *, struct ip6_hdr *, 379 int, int, u_short *); 380 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 381 u_short *); 382 static void pf_print_state_parts(struct pf_kstate *, 383 struct pf_state_key *, struct pf_state_key *); 384 static int pf_patch_8(struct pf_pdesc *, u_int8_t *, u_int8_t, 385 bool); 386 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 387 const struct pf_state_key_cmp *, u_int); 388 static bool pf_src_connlimit(struct pf_kstate *); 389 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 390 static void pf_counters_inc(int, struct pf_pdesc *, 391 struct pf_kstate *, struct pf_krule *, 392 struct pf_krule *); 393 static void pf_log_matches(struct pf_pdesc *, struct pf_krule *, 394 struct pf_krule *, struct pf_kruleset *, 395 struct pf_krule_slist *); 396 static void pf_overload_task(void *v, int pending); 397 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 398 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 399 struct pf_addr *, sa_family_t, struct pf_addr *, 400 struct pfi_kkif *, pf_sn_types_t); 401 static u_int pf_purge_expired_states(u_int, int); 402 static void pf_purge_unlinked_rules(void); 403 static int pf_mtag_uminit(void *, int, int); 404 static void pf_mtag_free(struct m_tag *); 405 static void pf_packet_rework_nat(struct pf_pdesc *, int, 406 struct pf_state_key *); 407 #ifdef INET 408 static void pf_route(struct mbuf **, struct pf_krule *, 409 struct ifnet *, struct pf_kstate *, 410 struct pf_pdesc *, struct inpcb *); 411 #endif /* INET */ 412 #ifdef INET6 413 static void pf_change_a6(struct pf_addr *, u_int16_t *, 414 struct pf_addr *, u_int8_t); 415 static void pf_route6(struct mbuf **, struct pf_krule *, 416 struct ifnet *, struct pf_kstate *, 417 struct pf_pdesc *, struct inpcb *); 418 #endif /* INET6 */ 419 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 420 421 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 422 423 extern int pf_end_threads; 424 extern struct proc *pf_purge_proc; 425 426 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 427 428 #define PACKET_UNDO_NAT(_pd, _off, _s) \ 429 do { \ 430 struct pf_state_key *nk; \ 431 if ((pd->dir) == PF_OUT) \ 432 nk = (_s)->key[PF_SK_STACK]; \ 433 else \ 434 nk = (_s)->key[PF_SK_WIRE]; \ 435 pf_packet_rework_nat(_pd, _off, nk); \ 436 } while (0) 437 438 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 439 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 440 441 #define STATE_LOOKUP(k, s, pd) \ 442 do { \ 443 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 444 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 445 if ((s) == NULL) \ 446 return (PF_DROP); \ 447 if (PACKET_LOOPED(pd)) \ 448 return (PF_PASS); \ 449 } while (0) 450 451 static struct pfi_kkif * 452 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 453 { 454 struct pfi_kkif *k = pd->kif; 455 456 SDT_PROBE2(pf, ip, , bound_iface, st, k); 457 458 /* Floating unless otherwise specified. */ 459 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 460 return (V_pfi_all); 461 462 /* 463 * Initially set to all, because we don't know what interface we'll be 464 * sending this out when we create the state. 465 */ 466 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN)) 467 return (V_pfi_all); 468 469 /* 470 * If this state is created based on another state (e.g. SCTP 471 * multihome) always set it floating initially. We can't know for sure 472 * what interface the actual traffic for this state will come in on. 473 */ 474 if (pd->related_rule) 475 return (V_pfi_all); 476 477 /* Don't overrule the interface for states created on incoming packets. */ 478 if (st->direction == PF_IN) 479 return (k); 480 481 /* No route-to, so don't overrule. */ 482 if (st->act.rt != PF_ROUTETO) 483 return (k); 484 485 /* Bind to the route-to interface. */ 486 return (st->act.rt_kif); 487 } 488 489 #define STATE_INC_COUNTERS(s) \ 490 do { \ 491 struct pf_krule_item *mrm; \ 492 counter_u64_add(s->rule->states_cur, 1); \ 493 counter_u64_add(s->rule->states_tot, 1); \ 494 if (s->anchor != NULL) { \ 495 counter_u64_add(s->anchor->states_cur, 1); \ 496 counter_u64_add(s->anchor->states_tot, 1); \ 497 } \ 498 if (s->nat_rule != NULL) { \ 499 counter_u64_add(s->nat_rule->states_cur, 1);\ 500 counter_u64_add(s->nat_rule->states_tot, 1);\ 501 } \ 502 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 503 counter_u64_add(mrm->r->states_cur, 1); \ 504 counter_u64_add(mrm->r->states_tot, 1); \ 505 } \ 506 } while (0) 507 508 #define STATE_DEC_COUNTERS(s) \ 509 do { \ 510 struct pf_krule_item *mrm; \ 511 if (s->nat_rule != NULL) \ 512 counter_u64_add(s->nat_rule->states_cur, -1);\ 513 if (s->anchor != NULL) \ 514 counter_u64_add(s->anchor->states_cur, -1); \ 515 counter_u64_add(s->rule->states_cur, -1); \ 516 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 517 counter_u64_add(mrm->r->states_cur, -1); \ 518 } while (0) 519 520 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 521 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 522 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 523 VNET_DEFINE(struct pf_idhash *, pf_idhash); 524 VNET_DEFINE(struct pf_srchash *, pf_srchash); 525 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 526 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 527 528 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 529 "pf(4)"); 530 531 VNET_DEFINE(u_long, pf_hashmask); 532 VNET_DEFINE(u_long, pf_srchashmask); 533 VNET_DEFINE(u_long, pf_udpendpointhashmask); 534 VNET_DEFINE_STATIC(u_long, pf_hashsize); 535 #define V_pf_hashsize VNET(pf_hashsize) 536 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 537 #define V_pf_srchashsize VNET(pf_srchashsize) 538 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 539 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 540 u_long pf_ioctl_maxcount = 65535; 541 542 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 543 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 544 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 545 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 546 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 547 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 548 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 549 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 550 551 VNET_DEFINE(void *, pf_swi_cookie); 552 VNET_DEFINE(struct intr_event *, pf_swi_ie); 553 554 VNET_DEFINE(uint32_t, pf_hashseed); 555 #define V_pf_hashseed VNET(pf_hashseed) 556 557 static void 558 pf_sctp_checksum(struct mbuf *m, int off) 559 { 560 uint32_t sum = 0; 561 562 /* Zero out the checksum, to enable recalculation. */ 563 m_copyback(m, off + offsetof(struct sctphdr, checksum), 564 sizeof(sum), (caddr_t)&sum); 565 566 sum = sctp_calculate_cksum(m, off); 567 568 m_copyback(m, off + offsetof(struct sctphdr, checksum), 569 sizeof(sum), (caddr_t)&sum); 570 } 571 572 int 573 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 574 { 575 576 switch (af) { 577 #ifdef INET 578 case AF_INET: 579 if (a->addr32[0] > b->addr32[0]) 580 return (1); 581 if (a->addr32[0] < b->addr32[0]) 582 return (-1); 583 break; 584 #endif /* INET */ 585 #ifdef INET6 586 case AF_INET6: 587 if (a->addr32[3] > b->addr32[3]) 588 return (1); 589 if (a->addr32[3] < b->addr32[3]) 590 return (-1); 591 if (a->addr32[2] > b->addr32[2]) 592 return (1); 593 if (a->addr32[2] < b->addr32[2]) 594 return (-1); 595 if (a->addr32[1] > b->addr32[1]) 596 return (1); 597 if (a->addr32[1] < b->addr32[1]) 598 return (-1); 599 if (a->addr32[0] > b->addr32[0]) 600 return (1); 601 if (a->addr32[0] < b->addr32[0]) 602 return (-1); 603 break; 604 #endif /* INET6 */ 605 default: 606 unhandled_af(af); 607 } 608 return (0); 609 } 610 611 static bool 612 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 613 { 614 switch (af) { 615 #ifdef INET 616 case AF_INET: 617 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 618 #endif /* INET */ 619 case AF_INET6: 620 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 621 default: 622 unhandled_af(af); 623 } 624 } 625 626 static void 627 pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk) 628 { 629 630 switch (pd->proto) { 631 case IPPROTO_TCP: { 632 struct tcphdr *th = &pd->hdr.tcp; 633 634 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 635 pf_change_ap(pd, pd->src, &th->th_sport, 636 &nk->addr[pd->sidx], nk->port[pd->sidx]); 637 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 638 pf_change_ap(pd, pd->dst, &th->th_dport, 639 &nk->addr[pd->didx], nk->port[pd->didx]); 640 m_copyback(pd->m, off, sizeof(*th), (caddr_t)th); 641 break; 642 } 643 case IPPROTO_UDP: { 644 struct udphdr *uh = &pd->hdr.udp; 645 646 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 647 pf_change_ap(pd, pd->src, &uh->uh_sport, 648 &nk->addr[pd->sidx], nk->port[pd->sidx]); 649 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 650 pf_change_ap(pd, pd->dst, &uh->uh_dport, 651 &nk->addr[pd->didx], nk->port[pd->didx]); 652 m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh); 653 break; 654 } 655 case IPPROTO_SCTP: { 656 struct sctphdr *sh = &pd->hdr.sctp; 657 658 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 659 pf_change_ap(pd, pd->src, &sh->src_port, 660 &nk->addr[pd->sidx], nk->port[pd->sidx]); 661 } 662 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 663 pf_change_ap(pd, pd->dst, &sh->dest_port, 664 &nk->addr[pd->didx], nk->port[pd->didx]); 665 } 666 667 break; 668 } 669 case IPPROTO_ICMP: { 670 struct icmp *ih = &pd->hdr.icmp; 671 672 if (nk->port[pd->sidx] != ih->icmp_id) { 673 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 674 ih->icmp_cksum, ih->icmp_id, 675 nk->port[pd->sidx], 0); 676 ih->icmp_id = nk->port[pd->sidx]; 677 pd->sport = &ih->icmp_id; 678 679 m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih); 680 } 681 /* FALLTHROUGH */ 682 } 683 default: 684 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 685 switch (pd->af) { 686 case AF_INET: 687 pf_change_a(&pd->src->v4.s_addr, 688 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 689 0); 690 break; 691 case AF_INET6: 692 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 693 break; 694 default: 695 unhandled_af(pd->af); 696 } 697 } 698 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 699 switch (pd->af) { 700 case AF_INET: 701 pf_change_a(&pd->dst->v4.s_addr, 702 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 703 0); 704 break; 705 case AF_INET6: 706 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 707 break; 708 default: 709 unhandled_af(pd->af); 710 } 711 } 712 break; 713 } 714 } 715 716 static __inline uint32_t 717 pf_hashkey(const struct pf_state_key *sk) 718 { 719 uint32_t h; 720 721 h = murmur3_32_hash32((const uint32_t *)sk, 722 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 723 V_pf_hashseed); 724 725 return (h & V_pf_hashmask); 726 } 727 728 __inline uint32_t 729 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 730 { 731 uint32_t h; 732 733 switch (af) { 734 case AF_INET: 735 h = murmur3_32_hash32((uint32_t *)&addr->v4, 736 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 737 break; 738 case AF_INET6: 739 h = murmur3_32_hash32((uint32_t *)&addr->v6, 740 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 741 break; 742 default: 743 unhandled_af(af); 744 } 745 746 return (h & V_pf_srchashmask); 747 } 748 749 static inline uint32_t 750 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 751 { 752 uint32_t h; 753 754 h = murmur3_32_hash32((uint32_t *)endpoint, 755 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 756 V_pf_hashseed); 757 return (h & V_pf_udpendpointhashmask); 758 } 759 760 #ifdef ALTQ 761 static int 762 pf_state_hash(struct pf_kstate *s) 763 { 764 u_int32_t hv = (intptr_t)s / sizeof(*s); 765 766 hv ^= crc32(&s->src, sizeof(s->src)); 767 hv ^= crc32(&s->dst, sizeof(s->dst)); 768 if (hv == 0) 769 hv = 1; 770 return (hv); 771 } 772 #endif /* ALTQ */ 773 774 static __inline void 775 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 776 { 777 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 778 s->dst.state = newstate; 779 if (which == PF_PEER_DST) 780 return; 781 if (s->src.state == newstate) 782 return; 783 if (s->creatorid == V_pf_status.hostid && 784 s->key[PF_SK_STACK] != NULL && 785 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 786 !(TCPS_HAVEESTABLISHED(s->src.state) || 787 s->src.state == TCPS_CLOSED) && 788 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 789 atomic_add_32(&V_pf_status.states_halfopen, -1); 790 791 s->src.state = newstate; 792 } 793 794 static void 795 pf_init_threshold(struct pf_threshold *threshold, 796 u_int32_t limit, u_int32_t seconds) 797 { 798 threshold->limit = limit * PF_THRESHOLD_MULT; 799 threshold->seconds = seconds; 800 threshold->count = 0; 801 threshold->last = time_uptime; 802 } 803 804 static void 805 pf_add_threshold(struct pf_threshold *threshold) 806 { 807 u_int32_t t = time_uptime, diff = t - threshold->last; 808 809 if (diff >= threshold->seconds) 810 threshold->count = 0; 811 else 812 threshold->count -= threshold->count * diff / 813 threshold->seconds; 814 threshold->count += PF_THRESHOLD_MULT; 815 threshold->last = t; 816 } 817 818 static int 819 pf_check_threshold(struct pf_threshold *threshold) 820 { 821 return (threshold->count > threshold->limit); 822 } 823 824 static bool 825 pf_src_connlimit(struct pf_kstate *state) 826 { 827 struct pf_overload_entry *pfoe; 828 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 829 bool limited = false; 830 831 PF_STATE_LOCK_ASSERT(state); 832 PF_SRC_NODE_LOCK(src_node); 833 834 src_node->conn++; 835 state->src.tcp_est = 1; 836 pf_add_threshold(&src_node->conn_rate); 837 838 if (state->rule->max_src_conn && 839 state->rule->max_src_conn < 840 src_node->conn) { 841 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 842 limited = true; 843 } 844 845 if (state->rule->max_src_conn_rate.limit && 846 pf_check_threshold(&src_node->conn_rate)) { 847 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 848 limited = true; 849 } 850 851 if (!limited) 852 goto done; 853 854 /* Kill this state. */ 855 state->timeout = PFTM_PURGE; 856 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 857 858 if (state->rule->overload_tbl == NULL) 859 goto done; 860 861 /* Schedule overloading and flushing task. */ 862 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 863 if (pfoe == NULL) 864 goto done; /* too bad :( */ 865 866 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 867 pfoe->af = state->key[PF_SK_WIRE]->af; 868 pfoe->rule = state->rule; 869 pfoe->dir = state->direction; 870 PF_OVERLOADQ_LOCK(); 871 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 872 PF_OVERLOADQ_UNLOCK(); 873 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 874 875 done: 876 PF_SRC_NODE_UNLOCK(src_node); 877 return (limited); 878 } 879 880 static void 881 pf_overload_task(void *v, int pending) 882 { 883 struct pf_overload_head queue; 884 struct pfr_addr p; 885 struct pf_overload_entry *pfoe, *pfoe1; 886 uint32_t killed = 0; 887 888 CURVNET_SET((struct vnet *)v); 889 890 PF_OVERLOADQ_LOCK(); 891 queue = V_pf_overloadqueue; 892 SLIST_INIT(&V_pf_overloadqueue); 893 PF_OVERLOADQ_UNLOCK(); 894 895 bzero(&p, sizeof(p)); 896 SLIST_FOREACH(pfoe, &queue, next) { 897 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 898 if (V_pf_status.debug >= PF_DEBUG_MISC) { 899 printf("%s: blocking address ", __func__); 900 pf_print_host(&pfoe->addr, 0, pfoe->af); 901 printf("\n"); 902 } 903 904 p.pfra_af = pfoe->af; 905 switch (pfoe->af) { 906 #ifdef INET 907 case AF_INET: 908 p.pfra_net = 32; 909 p.pfra_ip4addr = pfoe->addr.v4; 910 break; 911 #endif /* INET */ 912 #ifdef INET6 913 case AF_INET6: 914 p.pfra_net = 128; 915 p.pfra_ip6addr = pfoe->addr.v6; 916 break; 917 #endif /* INET6 */ 918 default: 919 unhandled_af(pfoe->af); 920 } 921 922 PF_RULES_WLOCK(); 923 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 924 PF_RULES_WUNLOCK(); 925 } 926 927 /* 928 * Remove those entries, that don't need flushing. 929 */ 930 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 931 if (pfoe->rule->flush == 0) { 932 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 933 free(pfoe, M_PFTEMP); 934 } else 935 counter_u64_add( 936 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 937 938 /* If nothing to flush, return. */ 939 if (SLIST_EMPTY(&queue)) { 940 CURVNET_RESTORE(); 941 return; 942 } 943 944 for (int i = 0; i <= V_pf_hashmask; i++) { 945 struct pf_idhash *ih = &V_pf_idhash[i]; 946 struct pf_state_key *sk; 947 struct pf_kstate *s; 948 949 PF_HASHROW_LOCK(ih); 950 LIST_FOREACH(s, &ih->states, entry) { 951 sk = s->key[PF_SK_WIRE]; 952 SLIST_FOREACH(pfoe, &queue, next) 953 if (sk->af == pfoe->af && 954 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 955 pfoe->rule == s->rule) && 956 ((pfoe->dir == PF_OUT && 957 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 958 (pfoe->dir == PF_IN && 959 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 960 s->timeout = PFTM_PURGE; 961 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 962 killed++; 963 } 964 } 965 PF_HASHROW_UNLOCK(ih); 966 } 967 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 968 free(pfoe, M_PFTEMP); 969 if (V_pf_status.debug >= PF_DEBUG_MISC) 970 printf("%s: %u states killed", __func__, killed); 971 972 CURVNET_RESTORE(); 973 } 974 975 /* 976 * On node found always returns locked. On not found its configurable. 977 */ 978 struct pf_ksrc_node * 979 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 980 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 981 { 982 struct pf_ksrc_node *n; 983 984 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 985 986 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 987 PF_HASHROW_LOCK(*sh); 988 LIST_FOREACH(n, &(*sh)->nodes, entry) 989 if (n->rule == rule && n->af == af && n->type == sn_type && 990 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 991 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 992 break; 993 994 if (n == NULL && !returnlocked) 995 PF_HASHROW_UNLOCK(*sh); 996 997 return (n); 998 } 999 1000 bool 1001 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 1002 { 1003 struct pf_ksrc_node *cur; 1004 1005 if ((*sn) == NULL) 1006 return (false); 1007 1008 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 1009 1010 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1011 PF_HASHROW_LOCK(sh); 1012 LIST_FOREACH(cur, &(sh->nodes), entry) { 1013 if (cur == (*sn) && 1014 cur->expire != 1) /* Ignore nodes being killed */ 1015 return (true); 1016 } 1017 PF_HASHROW_UNLOCK(sh); 1018 (*sn) = NULL; 1019 return (false); 1020 } 1021 1022 static void 1023 pf_free_src_node(struct pf_ksrc_node *sn) 1024 { 1025 1026 for (int i = 0; i < 2; i++) { 1027 counter_u64_free(sn->bytes[i]); 1028 counter_u64_free(sn->packets[i]); 1029 } 1030 uma_zfree(V_pf_sources_z, sn); 1031 } 1032 1033 static u_short 1034 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1035 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1036 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1037 struct pfi_kkif *rkif, pf_sn_types_t sn_type) 1038 { 1039 u_short reason = 0; 1040 struct pf_krule *r_track = rule; 1041 struct pf_ksrc_node **sn = &(sns[sn_type]); 1042 struct pf_srchash **sh = &(snhs[sn_type]); 1043 1044 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1045 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1046 1047 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1048 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1049 1050 /* 1051 * XXX: There could be a KASSERT for 1052 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1053 * but we'd need to pass pool *only* for this KASSERT. 1054 */ 1055 1056 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1057 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1058 r_track = &V_pf_default_rule; 1059 1060 /* 1061 * Request the sh to always be locked, as we might insert a new sn. 1062 */ 1063 if (*sn == NULL) 1064 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1065 1066 if (*sn == NULL) { 1067 PF_HASHROW_ASSERT(*sh); 1068 1069 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1070 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1071 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1072 reason = PFRES_SRCLIMIT; 1073 goto done; 1074 } 1075 1076 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1077 if ((*sn) == NULL) { 1078 reason = PFRES_MEMORY; 1079 goto done; 1080 } 1081 1082 for (int i = 0; i < 2; i++) { 1083 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1084 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1085 1086 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1087 pf_free_src_node(*sn); 1088 reason = PFRES_MEMORY; 1089 goto done; 1090 } 1091 } 1092 1093 if (sn_type == PF_SN_LIMIT) 1094 pf_init_threshold(&(*sn)->conn_rate, 1095 rule->max_src_conn_rate.limit, 1096 rule->max_src_conn_rate.seconds); 1097 1098 MPASS((*sn)->lock == NULL); 1099 (*sn)->lock = &(*sh)->lock; 1100 1101 (*sn)->af = af; 1102 (*sn)->rule = r_track; 1103 PF_ACPY(&(*sn)->addr, src, af); 1104 if (raddr != NULL) 1105 PF_ACPY(&(*sn)->raddr, raddr, af); 1106 (*sn)->rkif = rkif; 1107 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1108 (*sn)->creation = time_uptime; 1109 (*sn)->ruletype = rule->action; 1110 (*sn)->type = sn_type; 1111 counter_u64_add(r_track->src_nodes[sn_type], 1); 1112 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1113 } else { 1114 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1115 (*sn)->states >= rule->max_src_states) { 1116 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1117 1); 1118 reason = PFRES_SRCLIMIT; 1119 goto done; 1120 } 1121 } 1122 done: 1123 if (reason == 0) 1124 (*sn)->states++; 1125 else 1126 (*sn) = NULL; 1127 1128 PF_HASHROW_UNLOCK(*sh); 1129 return (reason); 1130 } 1131 1132 void 1133 pf_unlink_src_node(struct pf_ksrc_node *src) 1134 { 1135 PF_SRC_NODE_LOCK_ASSERT(src); 1136 1137 LIST_REMOVE(src, entry); 1138 if (src->rule) 1139 counter_u64_add(src->rule->src_nodes[src->type], -1); 1140 } 1141 1142 u_int 1143 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1144 { 1145 struct pf_ksrc_node *sn, *tmp; 1146 u_int count = 0; 1147 1148 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1149 pf_free_src_node(sn); 1150 count++; 1151 } 1152 1153 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1154 1155 return (count); 1156 } 1157 1158 void 1159 pf_mtag_initialize(void) 1160 { 1161 1162 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1163 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1164 UMA_ALIGN_PTR, 0); 1165 } 1166 1167 /* Per-vnet data storage structures initialization. */ 1168 void 1169 pf_initialize(void) 1170 { 1171 struct pf_keyhash *kh; 1172 struct pf_idhash *ih; 1173 struct pf_srchash *sh; 1174 struct pf_udpendpointhash *uh; 1175 u_int i; 1176 1177 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1178 V_pf_hashsize = PF_HASHSIZ; 1179 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1180 V_pf_srchashsize = PF_SRCHASHSIZ; 1181 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1182 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1183 1184 V_pf_hashseed = arc4random(); 1185 1186 /* States and state keys storage. */ 1187 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1189 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1190 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1191 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1192 1193 V_pf_state_key_z = uma_zcreate("pf state keys", 1194 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1195 UMA_ALIGN_PTR, 0); 1196 1197 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1198 M_PFHASH, M_NOWAIT | M_ZERO); 1199 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1200 M_PFHASH, M_NOWAIT | M_ZERO); 1201 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1202 printf("pf: Unable to allocate memory for " 1203 "state_hashsize %lu.\n", V_pf_hashsize); 1204 1205 free(V_pf_keyhash, M_PFHASH); 1206 free(V_pf_idhash, M_PFHASH); 1207 1208 V_pf_hashsize = PF_HASHSIZ; 1209 V_pf_keyhash = mallocarray(V_pf_hashsize, 1210 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1211 V_pf_idhash = mallocarray(V_pf_hashsize, 1212 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1213 } 1214 1215 V_pf_hashmask = V_pf_hashsize - 1; 1216 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1217 i++, kh++, ih++) { 1218 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1219 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1220 } 1221 1222 /* Source nodes. */ 1223 V_pf_sources_z = uma_zcreate("pf source nodes", 1224 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1225 0); 1226 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1227 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1228 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1229 1230 V_pf_srchash = mallocarray(V_pf_srchashsize, 1231 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1232 if (V_pf_srchash == NULL) { 1233 printf("pf: Unable to allocate memory for " 1234 "source_hashsize %lu.\n", V_pf_srchashsize); 1235 1236 V_pf_srchashsize = PF_SRCHASHSIZ; 1237 V_pf_srchash = mallocarray(V_pf_srchashsize, 1238 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1239 } 1240 1241 V_pf_srchashmask = V_pf_srchashsize - 1; 1242 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1243 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1244 1245 1246 /* UDP endpoint mappings. */ 1247 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1248 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1249 UMA_ALIGN_PTR, 0); 1250 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1251 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1252 if (V_pf_udpendpointhash == NULL) { 1253 printf("pf: Unable to allocate memory for " 1254 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1255 1256 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1257 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1258 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1259 } 1260 1261 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1262 for (i = 0, uh = V_pf_udpendpointhash; 1263 i <= V_pf_udpendpointhashmask; 1264 i++, uh++) { 1265 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1266 MTX_DEF | MTX_DUPOK); 1267 } 1268 1269 /* ALTQ */ 1270 TAILQ_INIT(&V_pf_altqs[0]); 1271 TAILQ_INIT(&V_pf_altqs[1]); 1272 TAILQ_INIT(&V_pf_altqs[2]); 1273 TAILQ_INIT(&V_pf_altqs[3]); 1274 TAILQ_INIT(&V_pf_pabuf[0]); 1275 TAILQ_INIT(&V_pf_pabuf[1]); 1276 TAILQ_INIT(&V_pf_pabuf[2]); 1277 V_pf_altqs_active = &V_pf_altqs[0]; 1278 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1279 V_pf_altqs_inactive = &V_pf_altqs[2]; 1280 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1281 1282 /* Send & overload+flush queues. */ 1283 STAILQ_INIT(&V_pf_sendqueue); 1284 SLIST_INIT(&V_pf_overloadqueue); 1285 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1286 1287 /* Unlinked, but may be referenced rules. */ 1288 TAILQ_INIT(&V_pf_unlinked_rules); 1289 } 1290 1291 void 1292 pf_mtag_cleanup(void) 1293 { 1294 1295 uma_zdestroy(pf_mtag_z); 1296 } 1297 1298 void 1299 pf_cleanup(void) 1300 { 1301 struct pf_keyhash *kh; 1302 struct pf_idhash *ih; 1303 struct pf_srchash *sh; 1304 struct pf_udpendpointhash *uh; 1305 struct pf_send_entry *pfse, *next; 1306 u_int i; 1307 1308 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1309 i <= V_pf_hashmask; 1310 i++, kh++, ih++) { 1311 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1312 __func__)); 1313 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1314 __func__)); 1315 mtx_destroy(&kh->lock); 1316 mtx_destroy(&ih->lock); 1317 } 1318 free(V_pf_keyhash, M_PFHASH); 1319 free(V_pf_idhash, M_PFHASH); 1320 1321 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1322 KASSERT(LIST_EMPTY(&sh->nodes), 1323 ("%s: source node hash not empty", __func__)); 1324 mtx_destroy(&sh->lock); 1325 } 1326 free(V_pf_srchash, M_PFHASH); 1327 1328 for (i = 0, uh = V_pf_udpendpointhash; 1329 i <= V_pf_udpendpointhashmask; 1330 i++, uh++) { 1331 KASSERT(LIST_EMPTY(&uh->endpoints), 1332 ("%s: udp endpoint hash not empty", __func__)); 1333 mtx_destroy(&uh->lock); 1334 } 1335 free(V_pf_udpendpointhash, M_PFHASH); 1336 1337 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1338 m_freem(pfse->pfse_m); 1339 free(pfse, M_PFTEMP); 1340 } 1341 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1342 1343 uma_zdestroy(V_pf_sources_z); 1344 uma_zdestroy(V_pf_state_z); 1345 uma_zdestroy(V_pf_state_key_z); 1346 uma_zdestroy(V_pf_udp_mapping_z); 1347 } 1348 1349 static int 1350 pf_mtag_uminit(void *mem, int size, int how) 1351 { 1352 struct m_tag *t; 1353 1354 t = (struct m_tag *)mem; 1355 t->m_tag_cookie = MTAG_ABI_COMPAT; 1356 t->m_tag_id = PACKET_TAG_PF; 1357 t->m_tag_len = sizeof(struct pf_mtag); 1358 t->m_tag_free = pf_mtag_free; 1359 1360 return (0); 1361 } 1362 1363 static void 1364 pf_mtag_free(struct m_tag *t) 1365 { 1366 1367 uma_zfree(pf_mtag_z, t); 1368 } 1369 1370 struct pf_mtag * 1371 pf_get_mtag(struct mbuf *m) 1372 { 1373 struct m_tag *mtag; 1374 1375 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1376 return ((struct pf_mtag *)(mtag + 1)); 1377 1378 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1379 if (mtag == NULL) 1380 return (NULL); 1381 bzero(mtag + 1, sizeof(struct pf_mtag)); 1382 m_tag_prepend(m, mtag); 1383 1384 return ((struct pf_mtag *)(mtag + 1)); 1385 } 1386 1387 static int 1388 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1389 struct pf_kstate *s) 1390 { 1391 struct pf_keyhash *khs, *khw, *kh; 1392 struct pf_state_key *sk, *cur; 1393 struct pf_kstate *si, *olds = NULL; 1394 int idx; 1395 1396 NET_EPOCH_ASSERT(); 1397 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1398 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1399 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1400 1401 /* 1402 * We need to lock hash slots of both keys. To avoid deadlock 1403 * we always lock the slot with lower address first. Unlock order 1404 * isn't important. 1405 * 1406 * We also need to lock ID hash slot before dropping key 1407 * locks. On success we return with ID hash slot locked. 1408 */ 1409 1410 if (skw == sks) { 1411 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1412 PF_HASHROW_LOCK(khs); 1413 } else { 1414 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1415 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1416 if (khs == khw) { 1417 PF_HASHROW_LOCK(khs); 1418 } else if (khs < khw) { 1419 PF_HASHROW_LOCK(khs); 1420 PF_HASHROW_LOCK(khw); 1421 } else { 1422 PF_HASHROW_LOCK(khw); 1423 PF_HASHROW_LOCK(khs); 1424 } 1425 } 1426 1427 #define KEYS_UNLOCK() do { \ 1428 if (khs != khw) { \ 1429 PF_HASHROW_UNLOCK(khs); \ 1430 PF_HASHROW_UNLOCK(khw); \ 1431 } else \ 1432 PF_HASHROW_UNLOCK(khs); \ 1433 } while (0) 1434 1435 /* 1436 * First run: start with wire key. 1437 */ 1438 sk = skw; 1439 kh = khw; 1440 idx = PF_SK_WIRE; 1441 1442 MPASS(s->lock == NULL); 1443 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1444 1445 keyattach: 1446 LIST_FOREACH(cur, &kh->keys, entry) 1447 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1448 break; 1449 1450 if (cur != NULL) { 1451 /* Key exists. Check for same kif, if none, add to key. */ 1452 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1453 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1454 1455 PF_HASHROW_LOCK(ih); 1456 if (si->kif == s->kif && 1457 ((si->key[PF_SK_WIRE]->af == sk->af && 1458 si->direction == s->direction) || 1459 (si->key[PF_SK_WIRE]->af != 1460 si->key[PF_SK_STACK]->af && 1461 sk->af == si->key[PF_SK_STACK]->af && 1462 si->direction != s->direction))) { 1463 bool reuse = false; 1464 1465 if (sk->proto == IPPROTO_TCP && 1466 si->src.state >= TCPS_FIN_WAIT_2 && 1467 si->dst.state >= TCPS_FIN_WAIT_2) 1468 reuse = true; 1469 1470 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1471 printf("pf: %s key attach " 1472 "%s on %s: ", 1473 (idx == PF_SK_WIRE) ? 1474 "wire" : "stack", 1475 reuse ? "reuse" : "failed", 1476 s->kif->pfik_name); 1477 pf_print_state_parts(s, 1478 (idx == PF_SK_WIRE) ? 1479 sk : NULL, 1480 (idx == PF_SK_STACK) ? 1481 sk : NULL); 1482 printf(", existing: "); 1483 pf_print_state_parts(si, 1484 (idx == PF_SK_WIRE) ? 1485 sk : NULL, 1486 (idx == PF_SK_STACK) ? 1487 sk : NULL); 1488 printf("\n"); 1489 } 1490 1491 if (reuse) { 1492 /* 1493 * New state matches an old >FIN_WAIT_2 1494 * state. We can't drop key hash locks, 1495 * thus we can't unlink it properly. 1496 * 1497 * As a workaround we drop it into 1498 * TCPS_CLOSED state, schedule purge 1499 * ASAP and push it into the very end 1500 * of the slot TAILQ, so that it won't 1501 * conflict with our new state. 1502 */ 1503 pf_set_protostate(si, PF_PEER_BOTH, 1504 TCPS_CLOSED); 1505 si->timeout = PFTM_PURGE; 1506 olds = si; 1507 } else { 1508 s->timeout = PFTM_UNLINKED; 1509 if (idx == PF_SK_STACK) 1510 /* 1511 * Remove the wire key from 1512 * the hash. Other threads 1513 * can't be referencing it 1514 * because we still hold the 1515 * hash lock. 1516 */ 1517 pf_state_key_detach(s, 1518 PF_SK_WIRE); 1519 PF_HASHROW_UNLOCK(ih); 1520 KEYS_UNLOCK(); 1521 if (idx == PF_SK_WIRE) 1522 /* 1523 * We've not inserted either key. 1524 * Free both. 1525 */ 1526 uma_zfree(V_pf_state_key_z, skw); 1527 if (skw != sks) 1528 uma_zfree( 1529 V_pf_state_key_z, 1530 sks); 1531 return (EEXIST); /* collision! */ 1532 } 1533 } 1534 PF_HASHROW_UNLOCK(ih); 1535 } 1536 uma_zfree(V_pf_state_key_z, sk); 1537 s->key[idx] = cur; 1538 } else { 1539 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1540 s->key[idx] = sk; 1541 } 1542 1543 stateattach: 1544 /* List is sorted, if-bound states before floating. */ 1545 if (s->kif == V_pfi_all) 1546 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1547 else 1548 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1549 1550 if (olds) { 1551 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1552 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1553 key_list[idx]); 1554 olds = NULL; 1555 } 1556 1557 /* 1558 * Attach done. See how should we (or should not?) 1559 * attach a second key. 1560 */ 1561 if (sks == skw) { 1562 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1563 idx = PF_SK_STACK; 1564 sks = NULL; 1565 goto stateattach; 1566 } else if (sks != NULL) { 1567 /* 1568 * Continue attaching with stack key. 1569 */ 1570 sk = sks; 1571 kh = khs; 1572 idx = PF_SK_STACK; 1573 sks = NULL; 1574 goto keyattach; 1575 } 1576 1577 PF_STATE_LOCK(s); 1578 KEYS_UNLOCK(); 1579 1580 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1581 ("%s failure", __func__)); 1582 1583 return (0); 1584 #undef KEYS_UNLOCK 1585 } 1586 1587 static void 1588 pf_detach_state(struct pf_kstate *s) 1589 { 1590 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1591 struct pf_keyhash *kh; 1592 1593 NET_EPOCH_ASSERT(); 1594 MPASS(s->timeout >= PFTM_MAX); 1595 1596 pf_sctp_multihome_detach_addr(s); 1597 1598 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1599 V_pflow_export_state_ptr(s); 1600 1601 if (sks != NULL) { 1602 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1603 PF_HASHROW_LOCK(kh); 1604 if (s->key[PF_SK_STACK] != NULL) 1605 pf_state_key_detach(s, PF_SK_STACK); 1606 /* 1607 * If both point to same key, then we are done. 1608 */ 1609 if (sks == s->key[PF_SK_WIRE]) { 1610 pf_state_key_detach(s, PF_SK_WIRE); 1611 PF_HASHROW_UNLOCK(kh); 1612 return; 1613 } 1614 PF_HASHROW_UNLOCK(kh); 1615 } 1616 1617 if (s->key[PF_SK_WIRE] != NULL) { 1618 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1619 PF_HASHROW_LOCK(kh); 1620 if (s->key[PF_SK_WIRE] != NULL) 1621 pf_state_key_detach(s, PF_SK_WIRE); 1622 PF_HASHROW_UNLOCK(kh); 1623 } 1624 } 1625 1626 static void 1627 pf_state_key_detach(struct pf_kstate *s, int idx) 1628 { 1629 struct pf_state_key *sk = s->key[idx]; 1630 #ifdef INVARIANTS 1631 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1632 1633 PF_HASHROW_ASSERT(kh); 1634 #endif /* INVARIANTS */ 1635 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1636 s->key[idx] = NULL; 1637 1638 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1639 LIST_REMOVE(sk, entry); 1640 uma_zfree(V_pf_state_key_z, sk); 1641 } 1642 } 1643 1644 static int 1645 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1646 { 1647 struct pf_state_key *sk = mem; 1648 1649 bzero(sk, sizeof(struct pf_state_key_cmp)); 1650 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1651 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1652 1653 return (0); 1654 } 1655 1656 static int 1657 pf_state_key_addr_setup(struct pf_pdesc *pd, 1658 struct pf_state_key_cmp *key, int multi) 1659 { 1660 struct pf_addr *saddr = pd->src; 1661 struct pf_addr *daddr = pd->dst; 1662 #ifdef INET6 1663 struct nd_neighbor_solicit nd; 1664 struct pf_addr *target; 1665 u_short action, reason; 1666 1667 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1668 goto copy; 1669 1670 switch (pd->hdr.icmp6.icmp6_type) { 1671 case ND_NEIGHBOR_SOLICIT: 1672 if (multi) 1673 return (-1); 1674 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1675 return (-1); 1676 target = (struct pf_addr *)&nd.nd_ns_target; 1677 daddr = target; 1678 break; 1679 case ND_NEIGHBOR_ADVERT: 1680 if (multi) 1681 return (-1); 1682 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1683 return (-1); 1684 target = (struct pf_addr *)&nd.nd_ns_target; 1685 saddr = target; 1686 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1687 key->addr[pd->didx].addr32[0] = 0; 1688 key->addr[pd->didx].addr32[1] = 0; 1689 key->addr[pd->didx].addr32[2] = 0; 1690 key->addr[pd->didx].addr32[3] = 0; 1691 daddr = NULL; /* overwritten */ 1692 } 1693 break; 1694 default: 1695 if (multi) { 1696 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1697 key->addr[pd->sidx].addr32[1] = 0; 1698 key->addr[pd->sidx].addr32[2] = 0; 1699 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1700 saddr = NULL; /* overwritten */ 1701 } 1702 } 1703 copy: 1704 #endif /* INET6 */ 1705 if (saddr) 1706 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1707 if (daddr) 1708 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1709 1710 return (0); 1711 } 1712 1713 int 1714 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1715 struct pf_state_key **sk, struct pf_state_key **nk) 1716 { 1717 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1718 if (*sk == NULL) 1719 return (ENOMEM); 1720 1721 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1722 0)) { 1723 uma_zfree(V_pf_state_key_z, *sk); 1724 *sk = NULL; 1725 return (ENOMEM); 1726 } 1727 1728 (*sk)->port[pd->sidx] = sport; 1729 (*sk)->port[pd->didx] = dport; 1730 (*sk)->proto = pd->proto; 1731 (*sk)->af = pd->af; 1732 1733 *nk = pf_state_key_clone(*sk); 1734 if (*nk == NULL) { 1735 uma_zfree(V_pf_state_key_z, *sk); 1736 *sk = NULL; 1737 return (ENOMEM); 1738 } 1739 1740 if (pd->af != pd->naf) { 1741 (*sk)->port[pd->sidx] = pd->osport; 1742 (*sk)->port[pd->didx] = pd->odport; 1743 1744 (*nk)->af = pd->naf; 1745 1746 /* 1747 * We're overwriting an address here, so potentially there's bits of an IPv6 1748 * address left in here. Clear that out first. 1749 */ 1750 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1751 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1752 if (pd->dir == PF_IN) { 1753 PF_ACPY(&(*nk)->addr[pd->didx], &pd->nsaddr, pd->naf); 1754 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->ndaddr, pd->naf); 1755 (*nk)->port[pd->didx] = pd->nsport; 1756 (*nk)->port[pd->sidx] = pd->ndport; 1757 } else { 1758 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->nsaddr, pd->naf); 1759 PF_ACPY(&(*nk)->addr[pd->didx], &pd->ndaddr, pd->naf); 1760 (*nk)->port[pd->sidx] = pd->nsport; 1761 (*nk)->port[pd->didx] = pd->ndport; 1762 } 1763 1764 switch (pd->proto) { 1765 case IPPROTO_ICMP: 1766 (*nk)->proto = IPPROTO_ICMPV6; 1767 break; 1768 case IPPROTO_ICMPV6: 1769 (*nk)->proto = IPPROTO_ICMP; 1770 break; 1771 default: 1772 (*nk)->proto = pd->proto; 1773 } 1774 } 1775 1776 return (0); 1777 } 1778 1779 struct pf_state_key * 1780 pf_state_key_clone(const struct pf_state_key *orig) 1781 { 1782 struct pf_state_key *sk; 1783 1784 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1785 if (sk == NULL) 1786 return (NULL); 1787 1788 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1789 1790 return (sk); 1791 } 1792 1793 int 1794 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1795 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1796 { 1797 struct pf_idhash *ih; 1798 struct pf_kstate *cur; 1799 int error; 1800 1801 NET_EPOCH_ASSERT(); 1802 1803 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1804 ("%s: sks not pristine", __func__)); 1805 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1806 ("%s: skw not pristine", __func__)); 1807 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1808 1809 s->kif = kif; 1810 s->orig_kif = orig_kif; 1811 1812 if (s->id == 0 && s->creatorid == 0) { 1813 s->id = alloc_unr64(&V_pf_stateid); 1814 s->id = htobe64(s->id); 1815 s->creatorid = V_pf_status.hostid; 1816 } 1817 1818 /* Returns with ID locked on success. */ 1819 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1820 return (error); 1821 skw = sks = NULL; 1822 1823 ih = &V_pf_idhash[PF_IDHASH(s)]; 1824 PF_HASHROW_ASSERT(ih); 1825 LIST_FOREACH(cur, &ih->states, entry) 1826 if (cur->id == s->id && cur->creatorid == s->creatorid) 1827 break; 1828 1829 if (cur != NULL) { 1830 s->timeout = PFTM_UNLINKED; 1831 PF_HASHROW_UNLOCK(ih); 1832 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1833 printf("pf: state ID collision: " 1834 "id: %016llx creatorid: %08x\n", 1835 (unsigned long long)be64toh(s->id), 1836 ntohl(s->creatorid)); 1837 } 1838 pf_detach_state(s); 1839 return (EEXIST); 1840 } 1841 LIST_INSERT_HEAD(&ih->states, s, entry); 1842 /* One for keys, one for ID hash. */ 1843 refcount_init(&s->refs, 2); 1844 1845 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1846 if (V_pfsync_insert_state_ptr != NULL) 1847 V_pfsync_insert_state_ptr(s); 1848 1849 /* Returns locked. */ 1850 return (0); 1851 } 1852 1853 /* 1854 * Find state by ID: returns with locked row on success. 1855 */ 1856 struct pf_kstate * 1857 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1858 { 1859 struct pf_idhash *ih; 1860 struct pf_kstate *s; 1861 1862 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1863 1864 ih = &V_pf_idhash[PF_IDHASHID(id)]; 1865 1866 PF_HASHROW_LOCK(ih); 1867 LIST_FOREACH(s, &ih->states, entry) 1868 if (s->id == id && s->creatorid == creatorid) 1869 break; 1870 1871 if (s == NULL) 1872 PF_HASHROW_UNLOCK(ih); 1873 1874 return (s); 1875 } 1876 1877 /* 1878 * Find state by key. 1879 * Returns with ID hash slot locked on success. 1880 */ 1881 static struct pf_kstate * 1882 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1883 u_int dir) 1884 { 1885 struct pf_keyhash *kh; 1886 struct pf_state_key *sk; 1887 struct pf_kstate *s; 1888 int idx; 1889 1890 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1891 1892 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1893 1894 PF_HASHROW_LOCK(kh); 1895 LIST_FOREACH(sk, &kh->keys, entry) 1896 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1897 break; 1898 if (sk == NULL) { 1899 PF_HASHROW_UNLOCK(kh); 1900 return (NULL); 1901 } 1902 1903 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1904 1905 /* List is sorted, if-bound states before floating ones. */ 1906 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1907 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1908 PF_STATE_LOCK(s); 1909 PF_HASHROW_UNLOCK(kh); 1910 if (__predict_false(s->timeout >= PFTM_MAX)) { 1911 /* 1912 * State is either being processed by 1913 * pf_remove_state() in an other thread, or 1914 * is scheduled for immediate expiry. 1915 */ 1916 PF_STATE_UNLOCK(s); 1917 return (NULL); 1918 } 1919 return (s); 1920 } 1921 1922 /* Look through the other list, in case of AF-TO */ 1923 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1924 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1925 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1926 continue; 1927 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1928 PF_STATE_LOCK(s); 1929 PF_HASHROW_UNLOCK(kh); 1930 if (__predict_false(s->timeout >= PFTM_MAX)) { 1931 /* 1932 * State is either being processed by 1933 * pf_remove_state() in an other thread, or 1934 * is scheduled for immediate expiry. 1935 */ 1936 PF_STATE_UNLOCK(s); 1937 return (NULL); 1938 } 1939 return (s); 1940 } 1941 } 1942 1943 PF_HASHROW_UNLOCK(kh); 1944 1945 return (NULL); 1946 } 1947 1948 /* 1949 * Returns with ID hash slot locked on success. 1950 */ 1951 struct pf_kstate * 1952 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1953 { 1954 struct pf_keyhash *kh; 1955 struct pf_state_key *sk; 1956 struct pf_kstate *s, *ret = NULL; 1957 int idx, inout = 0; 1958 1959 if (more != NULL) 1960 *more = 0; 1961 1962 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1963 1964 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1965 1966 PF_HASHROW_LOCK(kh); 1967 LIST_FOREACH(sk, &kh->keys, entry) 1968 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1969 break; 1970 if (sk == NULL) { 1971 PF_HASHROW_UNLOCK(kh); 1972 return (NULL); 1973 } 1974 switch (dir) { 1975 case PF_IN: 1976 idx = PF_SK_WIRE; 1977 break; 1978 case PF_OUT: 1979 idx = PF_SK_STACK; 1980 break; 1981 case PF_INOUT: 1982 idx = PF_SK_WIRE; 1983 inout = 1; 1984 break; 1985 default: 1986 panic("%s: dir %u", __func__, dir); 1987 } 1988 second_run: 1989 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1990 if (more == NULL) { 1991 PF_STATE_LOCK(s); 1992 PF_HASHROW_UNLOCK(kh); 1993 return (s); 1994 } 1995 1996 if (ret) 1997 (*more)++; 1998 else { 1999 ret = s; 2000 PF_STATE_LOCK(s); 2001 } 2002 } 2003 if (inout == 1) { 2004 inout = 0; 2005 idx = PF_SK_STACK; 2006 goto second_run; 2007 } 2008 PF_HASHROW_UNLOCK(kh); 2009 2010 return (ret); 2011 } 2012 2013 /* 2014 * FIXME 2015 * This routine is inefficient -- locks the state only to unlock immediately on 2016 * return. 2017 * It is racy -- after the state is unlocked nothing stops other threads from 2018 * removing it. 2019 */ 2020 bool 2021 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2022 { 2023 struct pf_kstate *s; 2024 2025 s = pf_find_state_all(key, dir, NULL); 2026 if (s != NULL) { 2027 PF_STATE_UNLOCK(s); 2028 return (true); 2029 } 2030 return (false); 2031 } 2032 2033 struct pf_udp_mapping * 2034 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2035 struct pf_addr *nat_addr, uint16_t nat_port) 2036 { 2037 struct pf_udp_mapping *mapping; 2038 2039 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2040 if (mapping == NULL) 2041 return (NULL); 2042 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 2043 mapping->endpoints[0].port = src_port; 2044 mapping->endpoints[0].af = af; 2045 mapping->endpoints[0].mapping = mapping; 2046 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 2047 mapping->endpoints[1].port = nat_port; 2048 mapping->endpoints[1].af = af; 2049 mapping->endpoints[1].mapping = mapping; 2050 refcount_init(&mapping->refs, 1); 2051 return (mapping); 2052 } 2053 2054 int 2055 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2056 { 2057 struct pf_udpendpointhash *h0, *h1; 2058 struct pf_udp_endpoint *endpoint; 2059 int ret = EEXIST; 2060 2061 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2062 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2063 if (h0 == h1) { 2064 PF_HASHROW_LOCK(h0); 2065 } else if (h0 < h1) { 2066 PF_HASHROW_LOCK(h0); 2067 PF_HASHROW_LOCK(h1); 2068 } else { 2069 PF_HASHROW_LOCK(h1); 2070 PF_HASHROW_LOCK(h0); 2071 } 2072 2073 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2074 if (bcmp(endpoint, &mapping->endpoints[0], 2075 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2076 break; 2077 } 2078 if (endpoint != NULL) 2079 goto cleanup; 2080 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2081 if (bcmp(endpoint, &mapping->endpoints[1], 2082 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2083 break; 2084 } 2085 if (endpoint != NULL) 2086 goto cleanup; 2087 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2088 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2089 ret = 0; 2090 2091 cleanup: 2092 if (h0 != h1) { 2093 PF_HASHROW_UNLOCK(h0); 2094 PF_HASHROW_UNLOCK(h1); 2095 } else { 2096 PF_HASHROW_UNLOCK(h0); 2097 } 2098 return (ret); 2099 } 2100 2101 void 2102 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2103 { 2104 /* refcount is synchronized on the source endpoint's row lock */ 2105 struct pf_udpendpointhash *h0, *h1; 2106 2107 if (mapping == NULL) 2108 return; 2109 2110 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2111 PF_HASHROW_LOCK(h0); 2112 if (refcount_release(&mapping->refs)) { 2113 LIST_REMOVE(&mapping->endpoints[0], entry); 2114 PF_HASHROW_UNLOCK(h0); 2115 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2116 PF_HASHROW_LOCK(h1); 2117 LIST_REMOVE(&mapping->endpoints[1], entry); 2118 PF_HASHROW_UNLOCK(h1); 2119 2120 uma_zfree(V_pf_udp_mapping_z, mapping); 2121 } else { 2122 PF_HASHROW_UNLOCK(h0); 2123 } 2124 } 2125 2126 2127 struct pf_udp_mapping * 2128 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2129 { 2130 struct pf_udpendpointhash *uh; 2131 struct pf_udp_endpoint *endpoint; 2132 2133 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2134 2135 PF_HASHROW_LOCK(uh); 2136 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2137 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2138 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2139 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2140 break; 2141 } 2142 if (endpoint == NULL) { 2143 PF_HASHROW_UNLOCK(uh); 2144 return (NULL); 2145 } 2146 refcount_acquire(&endpoint->mapping->refs); 2147 PF_HASHROW_UNLOCK(uh); 2148 return (endpoint->mapping); 2149 } 2150 /* END state table stuff */ 2151 2152 static void 2153 pf_send(struct pf_send_entry *pfse) 2154 { 2155 2156 PF_SENDQ_LOCK(); 2157 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2158 PF_SENDQ_UNLOCK(); 2159 swi_sched(V_pf_swi_cookie, 0); 2160 } 2161 2162 static bool 2163 pf_isforlocal(struct mbuf *m, int af) 2164 { 2165 switch (af) { 2166 #ifdef INET 2167 case AF_INET: { 2168 struct ip *ip = mtod(m, struct ip *); 2169 2170 return (in_localip(ip->ip_dst)); 2171 } 2172 #endif /* INET */ 2173 #ifdef INET6 2174 case AF_INET6: { 2175 struct ip6_hdr *ip6; 2176 struct in6_ifaddr *ia; 2177 ip6 = mtod(m, struct ip6_hdr *); 2178 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2179 if (ia == NULL) 2180 return (false); 2181 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2182 } 2183 #endif /* INET6 */ 2184 default: 2185 unhandled_af(af); 2186 } 2187 2188 return (false); 2189 } 2190 2191 int 2192 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2193 int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type) 2194 { 2195 /* 2196 * ICMP types marked with PF_OUT are typically responses to 2197 * PF_IN, and will match states in the opposite direction. 2198 * PF_IN ICMP types need to match a state with that type. 2199 */ 2200 *icmp_dir = PF_OUT; 2201 2202 /* Queries (and responses) */ 2203 switch (pd->af) { 2204 #ifdef INET 2205 case AF_INET: 2206 switch (type) { 2207 case ICMP_ECHO: 2208 *icmp_dir = PF_IN; 2209 /* FALLTHROUGH */ 2210 case ICMP_ECHOREPLY: 2211 *virtual_type = ICMP_ECHO; 2212 *virtual_id = pd->hdr.icmp.icmp_id; 2213 break; 2214 2215 case ICMP_TSTAMP: 2216 *icmp_dir = PF_IN; 2217 /* FALLTHROUGH */ 2218 case ICMP_TSTAMPREPLY: 2219 *virtual_type = ICMP_TSTAMP; 2220 *virtual_id = pd->hdr.icmp.icmp_id; 2221 break; 2222 2223 case ICMP_IREQ: 2224 *icmp_dir = PF_IN; 2225 /* FALLTHROUGH */ 2226 case ICMP_IREQREPLY: 2227 *virtual_type = ICMP_IREQ; 2228 *virtual_id = pd->hdr.icmp.icmp_id; 2229 break; 2230 2231 case ICMP_MASKREQ: 2232 *icmp_dir = PF_IN; 2233 /* FALLTHROUGH */ 2234 case ICMP_MASKREPLY: 2235 *virtual_type = ICMP_MASKREQ; 2236 *virtual_id = pd->hdr.icmp.icmp_id; 2237 break; 2238 2239 case ICMP_IPV6_WHEREAREYOU: 2240 *icmp_dir = PF_IN; 2241 /* FALLTHROUGH */ 2242 case ICMP_IPV6_IAMHERE: 2243 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2244 *virtual_id = 0; /* Nothing sane to match on! */ 2245 break; 2246 2247 case ICMP_MOBILE_REGREQUEST: 2248 *icmp_dir = PF_IN; 2249 /* FALLTHROUGH */ 2250 case ICMP_MOBILE_REGREPLY: 2251 *virtual_type = ICMP_MOBILE_REGREQUEST; 2252 *virtual_id = 0; /* Nothing sane to match on! */ 2253 break; 2254 2255 case ICMP_ROUTERSOLICIT: 2256 *icmp_dir = PF_IN; 2257 /* FALLTHROUGH */ 2258 case ICMP_ROUTERADVERT: 2259 *virtual_type = ICMP_ROUTERSOLICIT; 2260 *virtual_id = 0; /* Nothing sane to match on! */ 2261 break; 2262 2263 /* These ICMP types map to other connections */ 2264 case ICMP_UNREACH: 2265 case ICMP_SOURCEQUENCH: 2266 case ICMP_REDIRECT: 2267 case ICMP_TIMXCEED: 2268 case ICMP_PARAMPROB: 2269 /* These will not be used, but set them anyway */ 2270 *icmp_dir = PF_IN; 2271 *virtual_type = type; 2272 *virtual_id = 0; 2273 *virtual_type = htons(*virtual_type); 2274 return (1); /* These types match to another state */ 2275 2276 /* 2277 * All remaining ICMP types get their own states, 2278 * and will only match in one direction. 2279 */ 2280 default: 2281 *icmp_dir = PF_IN; 2282 *virtual_type = type; 2283 *virtual_id = 0; 2284 break; 2285 } 2286 break; 2287 #endif /* INET */ 2288 #ifdef INET6 2289 case AF_INET6: 2290 switch (type) { 2291 case ICMP6_ECHO_REQUEST: 2292 *icmp_dir = PF_IN; 2293 /* FALLTHROUGH */ 2294 case ICMP6_ECHO_REPLY: 2295 *virtual_type = ICMP6_ECHO_REQUEST; 2296 *virtual_id = pd->hdr.icmp6.icmp6_id; 2297 break; 2298 2299 case MLD_LISTENER_QUERY: 2300 case MLD_LISTENER_REPORT: { 2301 /* 2302 * Listener Report can be sent by clients 2303 * without an associated Listener Query. 2304 * In addition to that, when Report is sent as a 2305 * reply to a Query its source and destination 2306 * address are different. 2307 */ 2308 *icmp_dir = PF_IN; 2309 *virtual_type = MLD_LISTENER_QUERY; 2310 *virtual_id = 0; 2311 break; 2312 } 2313 case MLD_MTRACE: 2314 *icmp_dir = PF_IN; 2315 /* FALLTHROUGH */ 2316 case MLD_MTRACE_RESP: 2317 *virtual_type = MLD_MTRACE; 2318 *virtual_id = 0; /* Nothing sane to match on! */ 2319 break; 2320 2321 case ND_NEIGHBOR_SOLICIT: 2322 *icmp_dir = PF_IN; 2323 /* FALLTHROUGH */ 2324 case ND_NEIGHBOR_ADVERT: { 2325 *virtual_type = ND_NEIGHBOR_SOLICIT; 2326 *virtual_id = 0; 2327 break; 2328 } 2329 2330 /* 2331 * These ICMP types map to other connections. 2332 * ND_REDIRECT can't be in this list because the triggering 2333 * packet header is optional. 2334 */ 2335 case ICMP6_DST_UNREACH: 2336 case ICMP6_PACKET_TOO_BIG: 2337 case ICMP6_TIME_EXCEEDED: 2338 case ICMP6_PARAM_PROB: 2339 /* These will not be used, but set them anyway */ 2340 *icmp_dir = PF_IN; 2341 *virtual_type = type; 2342 *virtual_id = 0; 2343 *virtual_type = htons(*virtual_type); 2344 return (1); /* These types match to another state */ 2345 /* 2346 * All remaining ICMP6 types get their own states, 2347 * and will only match in one direction. 2348 */ 2349 default: 2350 *icmp_dir = PF_IN; 2351 *virtual_type = type; 2352 *virtual_id = 0; 2353 break; 2354 } 2355 break; 2356 #endif /* INET6 */ 2357 default: 2358 unhandled_af(pd->af); 2359 } 2360 *virtual_type = htons(*virtual_type); 2361 return (0); /* These types match to their own state */ 2362 } 2363 2364 void 2365 pf_intr(void *v) 2366 { 2367 struct epoch_tracker et; 2368 struct pf_send_head queue; 2369 struct pf_send_entry *pfse, *next; 2370 2371 CURVNET_SET((struct vnet *)v); 2372 2373 PF_SENDQ_LOCK(); 2374 queue = V_pf_sendqueue; 2375 STAILQ_INIT(&V_pf_sendqueue); 2376 PF_SENDQ_UNLOCK(); 2377 2378 NET_EPOCH_ENTER(et); 2379 2380 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2381 switch (pfse->pfse_type) { 2382 #ifdef INET 2383 case PFSE_IP: { 2384 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2385 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2386 ("%s: rcvif != loif", __func__)); 2387 2388 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2389 pfse->pfse_m->m_pkthdr.csum_flags |= 2390 CSUM_IP_VALID | CSUM_IP_CHECKED | 2391 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2392 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2393 ip_input(pfse->pfse_m); 2394 } else { 2395 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2396 NULL); 2397 } 2398 break; 2399 } 2400 case PFSE_ICMP: 2401 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2402 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2403 break; 2404 #endif /* INET */ 2405 #ifdef INET6 2406 case PFSE_IP6: 2407 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2408 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2409 ("%s: rcvif != loif", __func__)); 2410 2411 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2412 M_LOOP; 2413 pfse->pfse_m->m_pkthdr.csum_flags |= 2414 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2415 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2416 ip6_input(pfse->pfse_m); 2417 } else { 2418 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2419 NULL, NULL); 2420 } 2421 break; 2422 case PFSE_ICMP6: 2423 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2424 pfse->icmpopts.code, pfse->icmpopts.mtu); 2425 break; 2426 #endif /* INET6 */ 2427 default: 2428 panic("%s: unknown type", __func__); 2429 } 2430 free(pfse, M_PFTEMP); 2431 } 2432 NET_EPOCH_EXIT(et); 2433 CURVNET_RESTORE(); 2434 } 2435 2436 #define pf_purge_thread_period (hz / 10) 2437 2438 #ifdef PF_WANT_32_TO_64_COUNTER 2439 static void 2440 pf_status_counter_u64_periodic(void) 2441 { 2442 2443 PF_RULES_RASSERT(); 2444 2445 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2446 return; 2447 } 2448 2449 for (int i = 0; i < FCNT_MAX; i++) { 2450 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2451 } 2452 } 2453 2454 static void 2455 pf_kif_counter_u64_periodic(void) 2456 { 2457 struct pfi_kkif *kif; 2458 size_t r, run; 2459 2460 PF_RULES_RASSERT(); 2461 2462 if (__predict_false(V_pf_allkifcount == 0)) { 2463 return; 2464 } 2465 2466 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2467 return; 2468 } 2469 2470 run = V_pf_allkifcount / 10; 2471 if (run < 5) 2472 run = 5; 2473 2474 for (r = 0; r < run; r++) { 2475 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2476 if (kif == NULL) { 2477 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2478 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2479 break; 2480 } 2481 2482 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2483 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2484 2485 for (int i = 0; i < 2; i++) { 2486 for (int j = 0; j < 2; j++) { 2487 for (int k = 0; k < 2; k++) { 2488 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2489 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2490 } 2491 } 2492 } 2493 } 2494 } 2495 2496 static void 2497 pf_rule_counter_u64_periodic(void) 2498 { 2499 struct pf_krule *rule; 2500 size_t r, run; 2501 2502 PF_RULES_RASSERT(); 2503 2504 if (__predict_false(V_pf_allrulecount == 0)) { 2505 return; 2506 } 2507 2508 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2509 return; 2510 } 2511 2512 run = V_pf_allrulecount / 10; 2513 if (run < 5) 2514 run = 5; 2515 2516 for (r = 0; r < run; r++) { 2517 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2518 if (rule == NULL) { 2519 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2520 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2521 break; 2522 } 2523 2524 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2525 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2526 2527 pf_counter_u64_periodic(&rule->evaluations); 2528 for (int i = 0; i < 2; i++) { 2529 pf_counter_u64_periodic(&rule->packets[i]); 2530 pf_counter_u64_periodic(&rule->bytes[i]); 2531 } 2532 } 2533 } 2534 2535 static void 2536 pf_counter_u64_periodic_main(void) 2537 { 2538 PF_RULES_RLOCK_TRACKER; 2539 2540 V_pf_counter_periodic_iter++; 2541 2542 PF_RULES_RLOCK(); 2543 pf_counter_u64_critical_enter(); 2544 pf_status_counter_u64_periodic(); 2545 pf_kif_counter_u64_periodic(); 2546 pf_rule_counter_u64_periodic(); 2547 pf_counter_u64_critical_exit(); 2548 PF_RULES_RUNLOCK(); 2549 } 2550 #else 2551 #define pf_counter_u64_periodic_main() do { } while (0) 2552 #endif 2553 2554 void 2555 pf_purge_thread(void *unused __unused) 2556 { 2557 struct epoch_tracker et; 2558 2559 VNET_ITERATOR_DECL(vnet_iter); 2560 2561 sx_xlock(&pf_end_lock); 2562 while (pf_end_threads == 0) { 2563 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2564 2565 VNET_LIST_RLOCK(); 2566 NET_EPOCH_ENTER(et); 2567 VNET_FOREACH(vnet_iter) { 2568 CURVNET_SET(vnet_iter); 2569 2570 /* Wait until V_pf_default_rule is initialized. */ 2571 if (V_pf_vnet_active == 0) { 2572 CURVNET_RESTORE(); 2573 continue; 2574 } 2575 2576 pf_counter_u64_periodic_main(); 2577 2578 /* 2579 * Process 1/interval fraction of the state 2580 * table every run. 2581 */ 2582 V_pf_purge_idx = 2583 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2584 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2585 2586 /* 2587 * Purge other expired types every 2588 * PFTM_INTERVAL seconds. 2589 */ 2590 if (V_pf_purge_idx == 0) { 2591 /* 2592 * Order is important: 2593 * - states and src nodes reference rules 2594 * - states and rules reference kifs 2595 */ 2596 pf_purge_expired_fragments(); 2597 pf_purge_expired_src_nodes(); 2598 pf_purge_unlinked_rules(); 2599 pfi_kkif_purge(); 2600 } 2601 CURVNET_RESTORE(); 2602 } 2603 NET_EPOCH_EXIT(et); 2604 VNET_LIST_RUNLOCK(); 2605 } 2606 2607 pf_end_threads++; 2608 sx_xunlock(&pf_end_lock); 2609 kproc_exit(0); 2610 } 2611 2612 void 2613 pf_unload_vnet_purge(void) 2614 { 2615 2616 /* 2617 * To cleanse up all kifs and rules we need 2618 * two runs: first one clears reference flags, 2619 * then pf_purge_expired_states() doesn't 2620 * raise them, and then second run frees. 2621 */ 2622 pf_purge_unlinked_rules(); 2623 pfi_kkif_purge(); 2624 2625 /* 2626 * Now purge everything. 2627 */ 2628 pf_purge_expired_states(0, V_pf_hashmask); 2629 pf_purge_fragments(UINT_MAX); 2630 pf_purge_expired_src_nodes(); 2631 2632 /* 2633 * Now all kifs & rules should be unreferenced, 2634 * thus should be successfully freed. 2635 */ 2636 pf_purge_unlinked_rules(); 2637 pfi_kkif_purge(); 2638 } 2639 2640 u_int32_t 2641 pf_state_expires(const struct pf_kstate *state) 2642 { 2643 u_int32_t timeout; 2644 u_int32_t start; 2645 u_int32_t end; 2646 u_int32_t states; 2647 2648 /* handle all PFTM_* > PFTM_MAX here */ 2649 if (state->timeout == PFTM_PURGE) 2650 return (time_uptime); 2651 KASSERT(state->timeout != PFTM_UNLINKED, 2652 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2653 KASSERT((state->timeout < PFTM_MAX), 2654 ("pf_state_expires: timeout > PFTM_MAX")); 2655 timeout = state->rule->timeout[state->timeout]; 2656 if (!timeout) 2657 timeout = V_pf_default_rule.timeout[state->timeout]; 2658 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2659 if (start && state->rule != &V_pf_default_rule) { 2660 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2661 states = counter_u64_fetch(state->rule->states_cur); 2662 } else { 2663 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2664 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2665 states = V_pf_status.states; 2666 } 2667 if (end && states > start && start < end) { 2668 if (states < end) { 2669 timeout = (u_int64_t)timeout * (end - states) / 2670 (end - start); 2671 return ((state->expire / 1000) + timeout); 2672 } 2673 else 2674 return (time_uptime); 2675 } 2676 return ((state->expire / 1000) + timeout); 2677 } 2678 2679 void 2680 pf_purge_expired_src_nodes(void) 2681 { 2682 struct pf_ksrc_node_list freelist; 2683 struct pf_srchash *sh; 2684 struct pf_ksrc_node *cur, *next; 2685 int i; 2686 2687 LIST_INIT(&freelist); 2688 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2689 PF_HASHROW_LOCK(sh); 2690 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2691 if (cur->states == 0 && cur->expire <= time_uptime) { 2692 pf_unlink_src_node(cur); 2693 LIST_INSERT_HEAD(&freelist, cur, entry); 2694 } else if (cur->rule != NULL) 2695 cur->rule->rule_ref |= PFRULE_REFS; 2696 PF_HASHROW_UNLOCK(sh); 2697 } 2698 2699 pf_free_src_nodes(&freelist); 2700 2701 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2702 } 2703 2704 static void 2705 pf_src_tree_remove_state(struct pf_kstate *s) 2706 { 2707 uint32_t timeout; 2708 2709 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2710 s->rule->timeout[PFTM_SRC_NODE] : 2711 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2712 2713 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 2714 if (s->sns[sn_type] == NULL) 2715 continue; 2716 PF_SRC_NODE_LOCK(s->sns[sn_type]); 2717 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 2718 --(s->sns[sn_type]->conn); 2719 if (--(s->sns[sn_type]->states) == 0) 2720 s->sns[sn_type]->expire = time_uptime + timeout; 2721 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 2722 s->sns[sn_type] = NULL; 2723 } 2724 2725 } 2726 2727 /* 2728 * Unlink and potentilly free a state. Function may be 2729 * called with ID hash row locked, but always returns 2730 * unlocked, since it needs to go through key hash locking. 2731 */ 2732 int 2733 pf_remove_state(struct pf_kstate *s) 2734 { 2735 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2736 2737 NET_EPOCH_ASSERT(); 2738 PF_HASHROW_ASSERT(ih); 2739 2740 if (s->timeout == PFTM_UNLINKED) { 2741 /* 2742 * State is being processed 2743 * by pf_remove_state() in 2744 * an other thread. 2745 */ 2746 PF_HASHROW_UNLOCK(ih); 2747 return (0); /* XXXGL: undefined actually */ 2748 } 2749 2750 if (s->src.state == PF_TCPS_PROXY_DST) { 2751 /* XXX wire key the right one? */ 2752 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2753 &s->key[PF_SK_WIRE]->addr[1], 2754 &s->key[PF_SK_WIRE]->addr[0], 2755 s->key[PF_SK_WIRE]->port[1], 2756 s->key[PF_SK_WIRE]->port[0], 2757 s->src.seqhi, s->src.seqlo + 1, 2758 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2759 s->act.rtableid); 2760 } 2761 2762 LIST_REMOVE(s, entry); 2763 pf_src_tree_remove_state(s); 2764 2765 if (V_pfsync_delete_state_ptr != NULL) 2766 V_pfsync_delete_state_ptr(s); 2767 2768 STATE_DEC_COUNTERS(s); 2769 2770 s->timeout = PFTM_UNLINKED; 2771 2772 /* Ensure we remove it from the list of halfopen states, if needed. */ 2773 if (s->key[PF_SK_STACK] != NULL && 2774 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2775 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2776 2777 PF_HASHROW_UNLOCK(ih); 2778 2779 pf_detach_state(s); 2780 2781 pf_udp_mapping_release(s->udp_mapping); 2782 2783 /* pf_state_insert() initialises refs to 2 */ 2784 return (pf_release_staten(s, 2)); 2785 } 2786 2787 struct pf_kstate * 2788 pf_alloc_state(int flags) 2789 { 2790 2791 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2792 } 2793 2794 void 2795 pf_free_state(struct pf_kstate *cur) 2796 { 2797 struct pf_krule_item *ri; 2798 2799 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2800 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2801 cur->timeout)); 2802 2803 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2804 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2805 free(ri, M_PF_RULE_ITEM); 2806 } 2807 2808 pf_normalize_tcp_cleanup(cur); 2809 uma_zfree(V_pf_state_z, cur); 2810 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2811 } 2812 2813 /* 2814 * Called only from pf_purge_thread(), thus serialized. 2815 */ 2816 static u_int 2817 pf_purge_expired_states(u_int i, int maxcheck) 2818 { 2819 struct pf_idhash *ih; 2820 struct pf_kstate *s; 2821 struct pf_krule_item *mrm; 2822 size_t count __unused; 2823 2824 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2825 2826 /* 2827 * Go through hash and unlink states that expire now. 2828 */ 2829 while (maxcheck > 0) { 2830 count = 0; 2831 ih = &V_pf_idhash[i]; 2832 2833 /* only take the lock if we expect to do work */ 2834 if (!LIST_EMPTY(&ih->states)) { 2835 relock: 2836 PF_HASHROW_LOCK(ih); 2837 LIST_FOREACH(s, &ih->states, entry) { 2838 if (pf_state_expires(s) <= time_uptime) { 2839 V_pf_status.states -= 2840 pf_remove_state(s); 2841 goto relock; 2842 } 2843 s->rule->rule_ref |= PFRULE_REFS; 2844 if (s->nat_rule != NULL) 2845 s->nat_rule->rule_ref |= PFRULE_REFS; 2846 if (s->anchor != NULL) 2847 s->anchor->rule_ref |= PFRULE_REFS; 2848 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2849 SLIST_FOREACH(mrm, &s->match_rules, entry) 2850 mrm->r->rule_ref |= PFRULE_REFS; 2851 if (s->act.rt_kif) 2852 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2853 count++; 2854 } 2855 PF_HASHROW_UNLOCK(ih); 2856 } 2857 2858 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2859 2860 /* Return when we hit end of hash. */ 2861 if (++i > V_pf_hashmask) { 2862 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2863 return (0); 2864 } 2865 2866 maxcheck--; 2867 } 2868 2869 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2870 2871 return (i); 2872 } 2873 2874 static void 2875 pf_purge_unlinked_rules(void) 2876 { 2877 struct pf_krulequeue tmpq; 2878 struct pf_krule *r, *r1; 2879 2880 /* 2881 * If we have overloading task pending, then we'd 2882 * better skip purging this time. There is a tiny 2883 * probability that overloading task references 2884 * an already unlinked rule. 2885 */ 2886 PF_OVERLOADQ_LOCK(); 2887 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2888 PF_OVERLOADQ_UNLOCK(); 2889 return; 2890 } 2891 PF_OVERLOADQ_UNLOCK(); 2892 2893 /* 2894 * Do naive mark-and-sweep garbage collecting of old rules. 2895 * Reference flag is raised by pf_purge_expired_states() 2896 * and pf_purge_expired_src_nodes(). 2897 * 2898 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2899 * use a temporary queue. 2900 */ 2901 TAILQ_INIT(&tmpq); 2902 PF_UNLNKDRULES_LOCK(); 2903 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2904 if (!(r->rule_ref & PFRULE_REFS)) { 2905 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2906 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2907 } else 2908 r->rule_ref &= ~PFRULE_REFS; 2909 } 2910 PF_UNLNKDRULES_UNLOCK(); 2911 2912 if (!TAILQ_EMPTY(&tmpq)) { 2913 PF_CONFIG_LOCK(); 2914 PF_RULES_WLOCK(); 2915 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2916 TAILQ_REMOVE(&tmpq, r, entries); 2917 pf_free_rule(r); 2918 } 2919 PF_RULES_WUNLOCK(); 2920 PF_CONFIG_UNLOCK(); 2921 } 2922 } 2923 2924 void 2925 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2926 { 2927 switch (af) { 2928 #ifdef INET 2929 case AF_INET: { 2930 u_int32_t a = ntohl(addr->addr32[0]); 2931 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2932 (a>>8)&255, a&255); 2933 if (p) { 2934 p = ntohs(p); 2935 printf(":%u", p); 2936 } 2937 break; 2938 } 2939 #endif /* INET */ 2940 #ifdef INET6 2941 case AF_INET6: { 2942 u_int16_t b; 2943 u_int8_t i, curstart, curend, maxstart, maxend; 2944 curstart = curend = maxstart = maxend = 255; 2945 for (i = 0; i < 8; i++) { 2946 if (!addr->addr16[i]) { 2947 if (curstart == 255) 2948 curstart = i; 2949 curend = i; 2950 } else { 2951 if ((curend - curstart) > 2952 (maxend - maxstart)) { 2953 maxstart = curstart; 2954 maxend = curend; 2955 } 2956 curstart = curend = 255; 2957 } 2958 } 2959 if ((curend - curstart) > 2960 (maxend - maxstart)) { 2961 maxstart = curstart; 2962 maxend = curend; 2963 } 2964 for (i = 0; i < 8; i++) { 2965 if (i >= maxstart && i <= maxend) { 2966 if (i == 0) 2967 printf(":"); 2968 if (i == maxend) 2969 printf(":"); 2970 } else { 2971 b = ntohs(addr->addr16[i]); 2972 printf("%x", b); 2973 if (i < 7) 2974 printf(":"); 2975 } 2976 } 2977 if (p) { 2978 p = ntohs(p); 2979 printf("[%u]", p); 2980 } 2981 break; 2982 } 2983 #endif /* INET6 */ 2984 default: 2985 unhandled_af(af); 2986 } 2987 } 2988 2989 void 2990 pf_print_state(struct pf_kstate *s) 2991 { 2992 pf_print_state_parts(s, NULL, NULL); 2993 } 2994 2995 static void 2996 pf_print_state_parts(struct pf_kstate *s, 2997 struct pf_state_key *skwp, struct pf_state_key *sksp) 2998 { 2999 struct pf_state_key *skw, *sks; 3000 u_int8_t proto, dir; 3001 3002 /* Do our best to fill these, but they're skipped if NULL */ 3003 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 3004 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 3005 proto = skw ? skw->proto : (sks ? sks->proto : 0); 3006 dir = s ? s->direction : 0; 3007 3008 switch (proto) { 3009 case IPPROTO_IPV4: 3010 printf("IPv4"); 3011 break; 3012 case IPPROTO_IPV6: 3013 printf("IPv6"); 3014 break; 3015 case IPPROTO_TCP: 3016 printf("TCP"); 3017 break; 3018 case IPPROTO_UDP: 3019 printf("UDP"); 3020 break; 3021 case IPPROTO_ICMP: 3022 printf("ICMP"); 3023 break; 3024 case IPPROTO_ICMPV6: 3025 printf("ICMPv6"); 3026 break; 3027 default: 3028 printf("%u", proto); 3029 break; 3030 } 3031 switch (dir) { 3032 case PF_IN: 3033 printf(" in"); 3034 break; 3035 case PF_OUT: 3036 printf(" out"); 3037 break; 3038 } 3039 if (skw) { 3040 printf(" wire: "); 3041 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3042 printf(" "); 3043 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3044 } 3045 if (sks) { 3046 printf(" stack: "); 3047 if (sks != skw) { 3048 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3049 printf(" "); 3050 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3051 } else 3052 printf("-"); 3053 } 3054 if (s) { 3055 if (proto == IPPROTO_TCP) { 3056 printf(" [lo=%u high=%u win=%u modulator=%u", 3057 s->src.seqlo, s->src.seqhi, 3058 s->src.max_win, s->src.seqdiff); 3059 if (s->src.wscale && s->dst.wscale) 3060 printf(" wscale=%u", 3061 s->src.wscale & PF_WSCALE_MASK); 3062 printf("]"); 3063 printf(" [lo=%u high=%u win=%u modulator=%u", 3064 s->dst.seqlo, s->dst.seqhi, 3065 s->dst.max_win, s->dst.seqdiff); 3066 if (s->src.wscale && s->dst.wscale) 3067 printf(" wscale=%u", 3068 s->dst.wscale & PF_WSCALE_MASK); 3069 printf("]"); 3070 } 3071 printf(" %u:%u", s->src.state, s->dst.state); 3072 if (s->rule) 3073 printf(" @%d", s->rule->nr); 3074 } 3075 } 3076 3077 void 3078 pf_print_flags(uint16_t f) 3079 { 3080 if (f) 3081 printf(" "); 3082 if (f & TH_FIN) 3083 printf("F"); 3084 if (f & TH_SYN) 3085 printf("S"); 3086 if (f & TH_RST) 3087 printf("R"); 3088 if (f & TH_PUSH) 3089 printf("P"); 3090 if (f & TH_ACK) 3091 printf("A"); 3092 if (f & TH_URG) 3093 printf("U"); 3094 if (f & TH_ECE) 3095 printf("E"); 3096 if (f & TH_CWR) 3097 printf("W"); 3098 if (f & TH_AE) 3099 printf("e"); 3100 } 3101 3102 #define PF_SET_SKIP_STEPS(i) \ 3103 do { \ 3104 while (head[i] != cur) { \ 3105 head[i]->skip[i] = cur; \ 3106 head[i] = TAILQ_NEXT(head[i], entries); \ 3107 } \ 3108 } while (0) 3109 3110 void 3111 pf_calc_skip_steps(struct pf_krulequeue *rules) 3112 { 3113 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3114 int i; 3115 3116 cur = TAILQ_FIRST(rules); 3117 prev = cur; 3118 for (i = 0; i < PF_SKIP_COUNT; ++i) 3119 head[i] = cur; 3120 while (cur != NULL) { 3121 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3122 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3123 if (cur->direction != prev->direction) 3124 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3125 if (cur->af != prev->af) 3126 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3127 if (cur->proto != prev->proto) 3128 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3129 if (cur->src.neg != prev->src.neg || 3130 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3131 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3132 if (cur->dst.neg != prev->dst.neg || 3133 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3134 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3135 if (cur->src.port[0] != prev->src.port[0] || 3136 cur->src.port[1] != prev->src.port[1] || 3137 cur->src.port_op != prev->src.port_op) 3138 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3139 if (cur->dst.port[0] != prev->dst.port[0] || 3140 cur->dst.port[1] != prev->dst.port[1] || 3141 cur->dst.port_op != prev->dst.port_op) 3142 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3143 3144 prev = cur; 3145 cur = TAILQ_NEXT(cur, entries); 3146 } 3147 for (i = 0; i < PF_SKIP_COUNT; ++i) 3148 PF_SET_SKIP_STEPS(i); 3149 } 3150 3151 int 3152 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3153 { 3154 if (aw1->type != aw2->type) 3155 return (1); 3156 switch (aw1->type) { 3157 case PF_ADDR_ADDRMASK: 3158 case PF_ADDR_RANGE: 3159 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3160 return (1); 3161 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3162 return (1); 3163 return (0); 3164 case PF_ADDR_DYNIFTL: 3165 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3166 case PF_ADDR_NONE: 3167 case PF_ADDR_NOROUTE: 3168 case PF_ADDR_URPFFAILED: 3169 return (0); 3170 case PF_ADDR_TABLE: 3171 return (aw1->p.tbl != aw2->p.tbl); 3172 default: 3173 printf("invalid address type: %d\n", aw1->type); 3174 return (1); 3175 } 3176 } 3177 3178 /** 3179 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3180 * header isn't always a full checksum. In some cases (i.e. output) it's a 3181 * pseudo-header checksum, which is a partial checksum over src/dst IP 3182 * addresses, protocol number and length. 3183 * 3184 * That means we have the following cases: 3185 * * Input or forwarding: we don't have TSO, the checksum fields are full 3186 * checksums, we need to update the checksum whenever we change anything. 3187 * * Output (i.e. the checksum is a pseudo-header checksum): 3188 * x The field being updated is src/dst address or affects the length of 3189 * the packet. We need to update the pseudo-header checksum (note that this 3190 * checksum is not ones' complement). 3191 * x Some other field is being modified (e.g. src/dst port numbers): We 3192 * don't have to update anything. 3193 **/ 3194 u_int16_t 3195 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3196 { 3197 u_int32_t x; 3198 3199 x = cksum + old - new; 3200 x = (x + (x >> 16)) & 0xffff; 3201 3202 /* optimise: eliminate a branch when not udp */ 3203 if (udp && cksum == 0x0000) 3204 return cksum; 3205 if (udp && x == 0x0000) 3206 x = 0xffff; 3207 3208 return (u_int16_t)(x); 3209 } 3210 3211 static int 3212 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 3213 { 3214 int rewrite = 0; 3215 3216 if (*f != v) { 3217 uint16_t old = htons(hi ? (*f << 8) : *f); 3218 uint16_t new = htons(hi ? ( v << 8) : v); 3219 3220 *f = v; 3221 3222 if (! (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3223 CSUM_DELAY_DATA_IPV6))) 3224 *pd->pcksum = pf_cksum_fixup(*pd->pcksum, old, new, 3225 pd->proto == IPPROTO_UDP); 3226 3227 rewrite = 1; 3228 } 3229 3230 return (rewrite); 3231 } 3232 3233 int 3234 pf_patch_16(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 3235 { 3236 int rewrite = 0; 3237 u_int8_t *fb = (u_int8_t *)f; 3238 u_int8_t *vb = (u_int8_t *)&v; 3239 3240 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3241 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3242 3243 return (rewrite); 3244 } 3245 3246 int 3247 pf_patch_32(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 3248 { 3249 int rewrite = 0; 3250 u_int8_t *fb = (u_int8_t *)f; 3251 u_int8_t *vb = (u_int8_t *)&v; 3252 3253 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3254 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3255 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3256 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3257 3258 return (rewrite); 3259 } 3260 3261 u_int16_t 3262 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3263 u_int16_t new, u_int8_t udp) 3264 { 3265 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3266 return (cksum); 3267 3268 return (pf_cksum_fixup(cksum, old, new, udp)); 3269 } 3270 3271 static void 3272 pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p, 3273 struct pf_addr *an, u_int16_t pn) 3274 { 3275 struct pf_addr ao; 3276 u_int16_t po; 3277 uint8_t u = pd->virtual_proto == IPPROTO_UDP; 3278 3279 MPASS(pd->pcksum); 3280 if (pd->af == AF_INET) { 3281 MPASS(pd->ip_sum); 3282 } 3283 3284 PF_ACPY(&ao, a, pd->af); 3285 if (pd->af == pd->naf) 3286 PF_ACPY(a, an, pd->af); 3287 3288 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3289 *pd->pcksum = ~*pd->pcksum; 3290 3291 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3292 return; 3293 po = *p; 3294 *p = pn; 3295 3296 switch (pd->af) { 3297 #ifdef INET 3298 case AF_INET: 3299 switch (pd->naf) { 3300 case AF_INET: 3301 *pd->ip_sum = pf_cksum_fixup(pf_cksum_fixup(*pd->ip_sum, 3302 ao.addr16[0], an->addr16[0], 0), 3303 ao.addr16[1], an->addr16[1], 0); 3304 *p = pn; 3305 3306 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3307 ao.addr16[0], an->addr16[0], u), 3308 ao.addr16[1], an->addr16[1], u); 3309 3310 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3311 break; 3312 #ifdef INET6 3313 case AF_INET6: 3314 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3315 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3316 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3317 ao.addr16[0], an->addr16[0], u), 3318 ao.addr16[1], an->addr16[1], u), 3319 0, an->addr16[2], u), 3320 0, an->addr16[3], u), 3321 0, an->addr16[4], u), 3322 0, an->addr16[5], u), 3323 0, an->addr16[6], u), 3324 0, an->addr16[7], u), 3325 po, pn, u); 3326 break; 3327 #endif /* INET6 */ 3328 default: 3329 unhandled_af(pd->naf); 3330 } 3331 break; 3332 #endif /* INET */ 3333 #ifdef INET6 3334 case AF_INET6: 3335 switch (pd->naf) { 3336 #ifdef INET 3337 case AF_INET: 3338 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3339 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3340 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3341 ao.addr16[0], an->addr16[0], u), 3342 ao.addr16[1], an->addr16[1], u), 3343 ao.addr16[2], 0, u), 3344 ao.addr16[3], 0, u), 3345 ao.addr16[4], 0, u), 3346 ao.addr16[5], 0, u), 3347 ao.addr16[6], 0, u), 3348 ao.addr16[7], 0, u), 3349 po, pn, u); 3350 break; 3351 #endif /* INET */ 3352 case AF_INET6: 3353 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3354 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3355 pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3356 ao.addr16[0], an->addr16[0], u), 3357 ao.addr16[1], an->addr16[1], u), 3358 ao.addr16[2], an->addr16[2], u), 3359 ao.addr16[3], an->addr16[3], u), 3360 ao.addr16[4], an->addr16[4], u), 3361 ao.addr16[5], an->addr16[5], u), 3362 ao.addr16[6], an->addr16[6], u), 3363 ao.addr16[7], an->addr16[7], u); 3364 3365 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3366 break; 3367 default: 3368 unhandled_af(pd->naf); 3369 } 3370 break; 3371 #endif /* INET6 */ 3372 default: 3373 unhandled_af(pd->af); 3374 } 3375 3376 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3377 CSUM_DELAY_DATA_IPV6)) { 3378 *pd->pcksum = ~*pd->pcksum; 3379 if (! *pd->pcksum) 3380 *pd->pcksum = 0xffff; 3381 } 3382 } 3383 3384 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3385 void 3386 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3387 { 3388 u_int32_t ao; 3389 3390 memcpy(&ao, a, sizeof(ao)); 3391 memcpy(a, &an, sizeof(u_int32_t)); 3392 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3393 ao % 65536, an % 65536, u); 3394 } 3395 3396 void 3397 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3398 { 3399 u_int32_t ao; 3400 3401 memcpy(&ao, a, sizeof(ao)); 3402 memcpy(a, &an, sizeof(u_int32_t)); 3403 3404 *c = pf_proto_cksum_fixup(m, 3405 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3406 ao % 65536, an % 65536, udp); 3407 } 3408 3409 #ifdef INET6 3410 static void 3411 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3412 { 3413 struct pf_addr ao; 3414 3415 PF_ACPY(&ao, a, AF_INET6); 3416 PF_ACPY(a, an, AF_INET6); 3417 3418 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3419 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3420 pf_cksum_fixup(pf_cksum_fixup(*c, 3421 ao.addr16[0], an->addr16[0], u), 3422 ao.addr16[1], an->addr16[1], u), 3423 ao.addr16[2], an->addr16[2], u), 3424 ao.addr16[3], an->addr16[3], u), 3425 ao.addr16[4], an->addr16[4], u), 3426 ao.addr16[5], an->addr16[5], u), 3427 ao.addr16[6], an->addr16[6], u), 3428 ao.addr16[7], an->addr16[7], u); 3429 } 3430 #endif /* INET6 */ 3431 3432 static void 3433 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3434 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3435 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3436 { 3437 struct pf_addr oia, ooa; 3438 3439 PF_ACPY(&oia, ia, af); 3440 if (oa) 3441 PF_ACPY(&ooa, oa, af); 3442 3443 /* Change inner protocol port, fix inner protocol checksum. */ 3444 if (ip != NULL) { 3445 u_int16_t oip = *ip; 3446 u_int32_t opc; 3447 3448 if (pc != NULL) 3449 opc = *pc; 3450 *ip = np; 3451 if (pc != NULL) 3452 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3453 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3454 if (pc != NULL) 3455 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3456 } 3457 /* Change inner ip address, fix inner ip and icmp checksums. */ 3458 PF_ACPY(ia, na, af); 3459 switch (af) { 3460 #ifdef INET 3461 case AF_INET: { 3462 u_int32_t oh2c = *h2c; 3463 3464 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3465 oia.addr16[0], ia->addr16[0], 0), 3466 oia.addr16[1], ia->addr16[1], 0); 3467 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3468 oia.addr16[0], ia->addr16[0], 0), 3469 oia.addr16[1], ia->addr16[1], 0); 3470 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3471 break; 3472 } 3473 #endif /* INET */ 3474 #ifdef INET6 3475 case AF_INET6: 3476 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3477 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3478 pf_cksum_fixup(pf_cksum_fixup(*ic, 3479 oia.addr16[0], ia->addr16[0], u), 3480 oia.addr16[1], ia->addr16[1], u), 3481 oia.addr16[2], ia->addr16[2], u), 3482 oia.addr16[3], ia->addr16[3], u), 3483 oia.addr16[4], ia->addr16[4], u), 3484 oia.addr16[5], ia->addr16[5], u), 3485 oia.addr16[6], ia->addr16[6], u), 3486 oia.addr16[7], ia->addr16[7], u); 3487 break; 3488 #endif /* INET6 */ 3489 } 3490 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3491 if (oa) { 3492 PF_ACPY(oa, na, af); 3493 switch (af) { 3494 #ifdef INET 3495 case AF_INET: 3496 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3497 ooa.addr16[0], oa->addr16[0], 0), 3498 ooa.addr16[1], oa->addr16[1], 0); 3499 break; 3500 #endif /* INET */ 3501 #ifdef INET6 3502 case AF_INET6: 3503 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3504 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3505 pf_cksum_fixup(pf_cksum_fixup(*ic, 3506 ooa.addr16[0], oa->addr16[0], u), 3507 ooa.addr16[1], oa->addr16[1], u), 3508 ooa.addr16[2], oa->addr16[2], u), 3509 ooa.addr16[3], oa->addr16[3], u), 3510 ooa.addr16[4], oa->addr16[4], u), 3511 ooa.addr16[5], oa->addr16[5], u), 3512 ooa.addr16[6], oa->addr16[6], u), 3513 ooa.addr16[7], oa->addr16[7], u); 3514 break; 3515 #endif /* INET6 */ 3516 } 3517 } 3518 } 3519 3520 int 3521 pf_translate_af(struct pf_pdesc *pd) 3522 { 3523 #if defined(INET) && defined(INET6) 3524 struct mbuf *mp; 3525 struct ip *ip4; 3526 struct ip6_hdr *ip6; 3527 struct icmp6_hdr *icmp; 3528 struct m_tag *mtag; 3529 struct pf_fragment_tag *ftag; 3530 int hlen; 3531 3532 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3533 3534 /* trim the old header */ 3535 m_adj(pd->m, pd->off); 3536 3537 /* prepend a new one */ 3538 M_PREPEND(pd->m, hlen, M_NOWAIT); 3539 if (pd->m == NULL) 3540 return (-1); 3541 3542 switch (pd->naf) { 3543 case AF_INET: 3544 ip4 = mtod(pd->m, struct ip *); 3545 bzero(ip4, hlen); 3546 ip4->ip_v = IPVERSION; 3547 ip4->ip_hl = hlen >> 2; 3548 ip4->ip_tos = pd->tos; 3549 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3550 ip_fillid(ip4, V_ip_random_id); 3551 ip4->ip_ttl = pd->ttl; 3552 ip4->ip_p = pd->proto; 3553 ip4->ip_src = pd->nsaddr.v4; 3554 ip4->ip_dst = pd->ndaddr.v4; 3555 pd->src = (struct pf_addr *)&ip4->ip_src; 3556 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3557 pd->off = sizeof(struct ip); 3558 break; 3559 case AF_INET6: 3560 ip6 = mtod(pd->m, struct ip6_hdr *); 3561 bzero(ip6, hlen); 3562 ip6->ip6_vfc = IPV6_VERSION; 3563 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3564 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3565 ip6->ip6_nxt = pd->proto; 3566 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3567 ip6->ip6_hlim = IPV6_DEFHLIM; 3568 else 3569 ip6->ip6_hlim = pd->ttl; 3570 ip6->ip6_src = pd->nsaddr.v6; 3571 ip6->ip6_dst = pd->ndaddr.v6; 3572 pd->src = (struct pf_addr *)&ip6->ip6_src; 3573 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3574 pd->off = sizeof(struct ip6_hdr); 3575 3576 /* 3577 * If we're dealing with a reassembled packet we need to adjust 3578 * the header length from the IPv4 header size to IPv6 header 3579 * size. 3580 */ 3581 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3582 if (mtag) { 3583 ftag = (struct pf_fragment_tag *)(mtag + 1); 3584 ftag->ft_hdrlen = sizeof(*ip6); 3585 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3586 sizeof(struct ip) + sizeof(struct ip6_frag); 3587 } 3588 break; 3589 default: 3590 return (-1); 3591 } 3592 3593 /* recalculate icmp/icmp6 checksums */ 3594 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3595 int off; 3596 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3597 NULL) { 3598 pd->m = NULL; 3599 return (-1); 3600 } 3601 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3602 icmp->icmp6_cksum = 0; 3603 icmp->icmp6_cksum = pd->naf == AF_INET ? 3604 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3605 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3606 ntohs(ip6->ip6_plen)); 3607 } 3608 #endif /* INET && INET6 */ 3609 3610 return (0); 3611 } 3612 3613 int 3614 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3615 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3616 sa_family_t af, sa_family_t naf) 3617 { 3618 #if defined(INET) && defined(INET6) 3619 struct mbuf *n = NULL; 3620 struct ip *ip4; 3621 struct ip6_hdr *ip6; 3622 int hlen, olen, mlen; 3623 3624 if (af == naf || (af != AF_INET && af != AF_INET6) || 3625 (naf != AF_INET && naf != AF_INET6)) 3626 return (-1); 3627 3628 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3629 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3630 return (-1); 3631 3632 /* old header */ 3633 olen = pd2->off - off; 3634 /* new header */ 3635 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3636 3637 /* trim old header */ 3638 m_adj(n, olen); 3639 3640 /* prepend a new one */ 3641 M_PREPEND(n, hlen, M_NOWAIT); 3642 if (n == NULL) 3643 return (-1); 3644 3645 /* translate inner ip/ip6 header */ 3646 switch (naf) { 3647 case AF_INET: 3648 ip4 = mtod(n, struct ip *); 3649 bzero(ip4, sizeof(*ip4)); 3650 ip4->ip_v = IPVERSION; 3651 ip4->ip_hl = sizeof(*ip4) >> 2; 3652 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 3653 ip_fillid(ip4, V_ip_random_id); 3654 ip4->ip_off = htons(IP_DF); 3655 ip4->ip_ttl = pd2->ttl; 3656 if (pd2->proto == IPPROTO_ICMPV6) 3657 ip4->ip_p = IPPROTO_ICMP; 3658 else 3659 ip4->ip_p = pd2->proto; 3660 ip4->ip_src = src->v4; 3661 ip4->ip_dst = dst->v4; 3662 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3663 break; 3664 case AF_INET6: 3665 ip6 = mtod(n, struct ip6_hdr *); 3666 bzero(ip6, sizeof(*ip6)); 3667 ip6->ip6_vfc = IPV6_VERSION; 3668 ip6->ip6_plen = htons(pd2->tot_len - olen); 3669 if (pd2->proto == IPPROTO_ICMP) 3670 ip6->ip6_nxt = IPPROTO_ICMPV6; 3671 else 3672 ip6->ip6_nxt = pd2->proto; 3673 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3674 ip6->ip6_hlim = IPV6_DEFHLIM; 3675 else 3676 ip6->ip6_hlim = pd2->ttl; 3677 ip6->ip6_src = src->v6; 3678 ip6->ip6_dst = dst->v6; 3679 break; 3680 default: 3681 unhandled_af(naf); 3682 } 3683 3684 /* adjust payload offset and total packet length */ 3685 pd2->off += hlen - olen; 3686 pd->tot_len += hlen - olen; 3687 3688 /* merge modified inner packet with the original header */ 3689 mlen = n->m_pkthdr.len; 3690 m_cat(m, n); 3691 m->m_pkthdr.len += mlen; 3692 #endif /* INET && INET6 */ 3693 3694 return (0); 3695 } 3696 3697 #define PTR_IP(field) (offsetof(struct ip, field)) 3698 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3699 3700 int 3701 pf_translate_icmp_af(int af, void *arg) 3702 { 3703 #if defined(INET) && defined(INET6) 3704 struct icmp *icmp4; 3705 struct icmp6_hdr *icmp6; 3706 u_int32_t mtu; 3707 int32_t ptr = -1; 3708 u_int8_t type; 3709 u_int8_t code; 3710 3711 switch (af) { 3712 case AF_INET: 3713 icmp6 = arg; 3714 type = icmp6->icmp6_type; 3715 code = icmp6->icmp6_code; 3716 mtu = ntohl(icmp6->icmp6_mtu); 3717 3718 switch (type) { 3719 case ICMP6_ECHO_REQUEST: 3720 type = ICMP_ECHO; 3721 break; 3722 case ICMP6_ECHO_REPLY: 3723 type = ICMP_ECHOREPLY; 3724 break; 3725 case ICMP6_DST_UNREACH: 3726 type = ICMP_UNREACH; 3727 switch (code) { 3728 case ICMP6_DST_UNREACH_NOROUTE: 3729 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3730 case ICMP6_DST_UNREACH_ADDR: 3731 code = ICMP_UNREACH_HOST; 3732 break; 3733 case ICMP6_DST_UNREACH_ADMIN: 3734 code = ICMP_UNREACH_HOST_PROHIB; 3735 break; 3736 case ICMP6_DST_UNREACH_NOPORT: 3737 code = ICMP_UNREACH_PORT; 3738 break; 3739 default: 3740 return (-1); 3741 } 3742 break; 3743 case ICMP6_PACKET_TOO_BIG: 3744 type = ICMP_UNREACH; 3745 code = ICMP_UNREACH_NEEDFRAG; 3746 mtu -= 20; 3747 break; 3748 case ICMP6_TIME_EXCEEDED: 3749 type = ICMP_TIMXCEED; 3750 break; 3751 case ICMP6_PARAM_PROB: 3752 switch (code) { 3753 case ICMP6_PARAMPROB_HEADER: 3754 type = ICMP_PARAMPROB; 3755 code = ICMP_PARAMPROB_ERRATPTR; 3756 ptr = ntohl(icmp6->icmp6_pptr); 3757 3758 if (ptr == PTR_IP6(ip6_vfc)) 3759 ; /* preserve */ 3760 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3761 ptr = PTR_IP(ip_tos); 3762 else if (ptr == PTR_IP6(ip6_plen) || 3763 ptr == PTR_IP6(ip6_plen) + 1) 3764 ptr = PTR_IP(ip_len); 3765 else if (ptr == PTR_IP6(ip6_nxt)) 3766 ptr = PTR_IP(ip_p); 3767 else if (ptr == PTR_IP6(ip6_hlim)) 3768 ptr = PTR_IP(ip_ttl); 3769 else if (ptr >= PTR_IP6(ip6_src) && 3770 ptr < PTR_IP6(ip6_dst)) 3771 ptr = PTR_IP(ip_src); 3772 else if (ptr >= PTR_IP6(ip6_dst) && 3773 ptr < sizeof(struct ip6_hdr)) 3774 ptr = PTR_IP(ip_dst); 3775 else { 3776 return (-1); 3777 } 3778 break; 3779 case ICMP6_PARAMPROB_NEXTHEADER: 3780 type = ICMP_UNREACH; 3781 code = ICMP_UNREACH_PROTOCOL; 3782 break; 3783 default: 3784 return (-1); 3785 } 3786 break; 3787 default: 3788 return (-1); 3789 } 3790 if (icmp6->icmp6_type != type) { 3791 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3792 icmp6->icmp6_type, type, 0); 3793 icmp6->icmp6_type = type; 3794 } 3795 if (icmp6->icmp6_code != code) { 3796 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3797 icmp6->icmp6_code, code, 0); 3798 icmp6->icmp6_code = code; 3799 } 3800 if (icmp6->icmp6_mtu != htonl(mtu)) { 3801 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3802 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3803 /* aligns well with a icmpv4 nextmtu */ 3804 icmp6->icmp6_mtu = htonl(mtu); 3805 } 3806 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3807 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3808 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3809 /* icmpv4 pptr is a one most significant byte */ 3810 icmp6->icmp6_pptr = htonl(ptr << 24); 3811 } 3812 break; 3813 case AF_INET6: 3814 icmp4 = arg; 3815 type = icmp4->icmp_type; 3816 code = icmp4->icmp_code; 3817 mtu = ntohs(icmp4->icmp_nextmtu); 3818 3819 switch (type) { 3820 case ICMP_ECHO: 3821 type = ICMP6_ECHO_REQUEST; 3822 break; 3823 case ICMP_ECHOREPLY: 3824 type = ICMP6_ECHO_REPLY; 3825 break; 3826 case ICMP_UNREACH: 3827 type = ICMP6_DST_UNREACH; 3828 switch (code) { 3829 case ICMP_UNREACH_NET: 3830 case ICMP_UNREACH_HOST: 3831 case ICMP_UNREACH_NET_UNKNOWN: 3832 case ICMP_UNREACH_HOST_UNKNOWN: 3833 case ICMP_UNREACH_ISOLATED: 3834 case ICMP_UNREACH_TOSNET: 3835 case ICMP_UNREACH_TOSHOST: 3836 code = ICMP6_DST_UNREACH_NOROUTE; 3837 break; 3838 case ICMP_UNREACH_PORT: 3839 code = ICMP6_DST_UNREACH_NOPORT; 3840 break; 3841 case ICMP_UNREACH_NET_PROHIB: 3842 case ICMP_UNREACH_HOST_PROHIB: 3843 case ICMP_UNREACH_FILTER_PROHIB: 3844 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3845 code = ICMP6_DST_UNREACH_ADMIN; 3846 break; 3847 case ICMP_UNREACH_PROTOCOL: 3848 type = ICMP6_PARAM_PROB; 3849 code = ICMP6_PARAMPROB_NEXTHEADER; 3850 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3851 break; 3852 case ICMP_UNREACH_NEEDFRAG: 3853 type = ICMP6_PACKET_TOO_BIG; 3854 code = 0; 3855 mtu += 20; 3856 break; 3857 default: 3858 return (-1); 3859 } 3860 break; 3861 case ICMP_TIMXCEED: 3862 type = ICMP6_TIME_EXCEEDED; 3863 break; 3864 case ICMP_PARAMPROB: 3865 type = ICMP6_PARAM_PROB; 3866 switch (code) { 3867 case ICMP_PARAMPROB_ERRATPTR: 3868 code = ICMP6_PARAMPROB_HEADER; 3869 break; 3870 case ICMP_PARAMPROB_LENGTH: 3871 code = ICMP6_PARAMPROB_HEADER; 3872 break; 3873 default: 3874 return (-1); 3875 } 3876 3877 ptr = icmp4->icmp_pptr; 3878 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3879 ; /* preserve */ 3880 else if (ptr == PTR_IP(ip_len) || 3881 ptr == PTR_IP(ip_len) + 1) 3882 ptr = PTR_IP6(ip6_plen); 3883 else if (ptr == PTR_IP(ip_ttl)) 3884 ptr = PTR_IP6(ip6_hlim); 3885 else if (ptr == PTR_IP(ip_p)) 3886 ptr = PTR_IP6(ip6_nxt); 3887 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 3888 ptr = PTR_IP6(ip6_src); 3889 else if (ptr >= PTR_IP(ip_dst) && 3890 ptr < sizeof(struct ip)) 3891 ptr = PTR_IP6(ip6_dst); 3892 else { 3893 return (-1); 3894 } 3895 break; 3896 default: 3897 return (-1); 3898 } 3899 if (icmp4->icmp_type != type) { 3900 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3901 icmp4->icmp_type, type, 0); 3902 icmp4->icmp_type = type; 3903 } 3904 if (icmp4->icmp_code != code) { 3905 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3906 icmp4->icmp_code, code, 0); 3907 icmp4->icmp_code = code; 3908 } 3909 if (icmp4->icmp_nextmtu != htons(mtu)) { 3910 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3911 icmp4->icmp_nextmtu, htons(mtu), 0); 3912 icmp4->icmp_nextmtu = htons(mtu); 3913 } 3914 if (ptr >= 0 && icmp4->icmp_void != ptr) { 3915 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3916 htons(icmp4->icmp_pptr), htons(ptr), 0); 3917 icmp4->icmp_void = htonl(ptr); 3918 } 3919 break; 3920 default: 3921 unhandled_af(af); 3922 } 3923 #endif /* INET && INET6 */ 3924 3925 return (0); 3926 } 3927 3928 /* 3929 * Need to modulate the sequence numbers in the TCP SACK option 3930 * (credits to Krzysztof Pfaff for report and patch) 3931 */ 3932 static int 3933 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3934 struct pf_state_peer *dst) 3935 { 3936 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3937 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3938 int copyback = 0, i, olen; 3939 struct sackblk sack; 3940 3941 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3942 if (hlen < TCPOLEN_SACKLEN || hlen > MAX_TCPOPTLEN || 3943 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3944 return 0; 3945 3946 while (hlen >= TCPOLEN_SACKLEN) { 3947 size_t startoff = opt - opts; 3948 olen = opt[1]; 3949 switch (*opt) { 3950 case TCPOPT_EOL: /* FALLTHROUGH */ 3951 case TCPOPT_NOP: 3952 opt++; 3953 hlen--; 3954 break; 3955 case TCPOPT_SACK: 3956 if (olen > hlen) 3957 olen = hlen; 3958 if (olen >= TCPOLEN_SACKLEN) { 3959 for (i = 2; i + TCPOLEN_SACK <= olen; 3960 i += TCPOLEN_SACK) { 3961 memcpy(&sack, &opt[i], sizeof(sack)); 3962 pf_patch_32(pd, 3963 &sack.start, 3964 htonl(ntohl(sack.start) - dst->seqdiff), 3965 PF_ALGNMNT(startoff)); 3966 pf_patch_32(pd, 3967 &sack.end, 3968 htonl(ntohl(sack.end) - dst->seqdiff), 3969 PF_ALGNMNT(startoff)); 3970 memcpy(&opt[i], &sack, sizeof(sack)); 3971 copyback = 1; 3972 } 3973 } 3974 /* FALLTHROUGH */ 3975 default: 3976 if (olen < 2) 3977 olen = 2; 3978 hlen -= olen; 3979 opt += olen; 3980 } 3981 } 3982 3983 if (copyback) 3984 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3985 return (copyback); 3986 } 3987 3988 struct mbuf * 3989 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3990 const struct pf_addr *saddr, const struct pf_addr *daddr, 3991 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3992 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3993 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3994 { 3995 struct mbuf *m; 3996 int len, tlen; 3997 #ifdef INET 3998 struct ip *h = NULL; 3999 #endif /* INET */ 4000 #ifdef INET6 4001 struct ip6_hdr *h6 = NULL; 4002 #endif /* INET6 */ 4003 struct tcphdr *th; 4004 char *opt; 4005 struct pf_mtag *pf_mtag; 4006 4007 len = 0; 4008 th = NULL; 4009 4010 /* maximum segment size tcp option */ 4011 tlen = sizeof(struct tcphdr); 4012 if (mss) 4013 tlen += 4; 4014 4015 switch (af) { 4016 #ifdef INET 4017 case AF_INET: 4018 len = sizeof(struct ip) + tlen; 4019 break; 4020 #endif /* INET */ 4021 #ifdef INET6 4022 case AF_INET6: 4023 len = sizeof(struct ip6_hdr) + tlen; 4024 break; 4025 #endif /* INET6 */ 4026 default: 4027 unhandled_af(af); 4028 } 4029 4030 m = m_gethdr(M_NOWAIT, MT_DATA); 4031 if (m == NULL) 4032 return (NULL); 4033 4034 #ifdef MAC 4035 mac_netinet_firewall_send(m); 4036 #endif 4037 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 4038 m_freem(m); 4039 return (NULL); 4040 } 4041 m->m_flags |= mbuf_flags; 4042 pf_mtag->tag = mtag_tag; 4043 pf_mtag->flags = mtag_flags; 4044 4045 if (rtableid >= 0) 4046 M_SETFIB(m, rtableid); 4047 4048 #ifdef ALTQ 4049 if (r != NULL && r->qid) { 4050 pf_mtag->qid = r->qid; 4051 4052 /* add hints for ecn */ 4053 pf_mtag->hdr = mtod(m, struct ip *); 4054 } 4055 #endif /* ALTQ */ 4056 m->m_data += max_linkhdr; 4057 m->m_pkthdr.len = m->m_len = len; 4058 /* The rest of the stack assumes a rcvif, so provide one. 4059 * This is a locally generated packet, so .. close enough. */ 4060 m->m_pkthdr.rcvif = V_loif; 4061 bzero(m->m_data, len); 4062 switch (af) { 4063 #ifdef INET 4064 case AF_INET: 4065 m->m_pkthdr.csum_flags |= CSUM_TCP; 4066 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4067 4068 h = mtod(m, struct ip *); 4069 4070 h->ip_p = IPPROTO_TCP; 4071 h->ip_len = htons(tlen); 4072 h->ip_v = 4; 4073 h->ip_hl = sizeof(*h) >> 2; 4074 h->ip_tos = IPTOS_LOWDELAY; 4075 h->ip_len = htons(len); 4076 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4077 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4078 h->ip_sum = 0; 4079 h->ip_src.s_addr = saddr->v4.s_addr; 4080 h->ip_dst.s_addr = daddr->v4.s_addr; 4081 4082 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4083 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4084 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4085 break; 4086 #endif /* INET */ 4087 #ifdef INET6 4088 case AF_INET6: 4089 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4090 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4091 4092 h6 = mtod(m, struct ip6_hdr *); 4093 4094 /* IP header fields included in the TCP checksum */ 4095 h6->ip6_nxt = IPPROTO_TCP; 4096 h6->ip6_plen = htons(tlen); 4097 h6->ip6_vfc |= IPV6_VERSION; 4098 h6->ip6_hlim = V_ip6_defhlim; 4099 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4100 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4101 4102 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4103 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4104 IPPROTO_TCP, 0); 4105 break; 4106 #endif /* INET6 */ 4107 } 4108 4109 /* TCP header */ 4110 th->th_sport = sport; 4111 th->th_dport = dport; 4112 th->th_seq = htonl(seq); 4113 th->th_ack = htonl(ack); 4114 th->th_off = tlen >> 2; 4115 tcp_set_flags(th, tcp_flags); 4116 th->th_win = htons(win); 4117 4118 if (mss) { 4119 opt = (char *)(th + 1); 4120 opt[0] = TCPOPT_MAXSEG; 4121 opt[1] = 4; 4122 mss = htons(mss); 4123 memcpy((opt + 2), &mss, 2); 4124 } 4125 4126 return (m); 4127 } 4128 4129 static void 4130 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4131 uint8_t ttl, int rtableid) 4132 { 4133 struct mbuf *m; 4134 #ifdef INET 4135 struct ip *h = NULL; 4136 #endif /* INET */ 4137 #ifdef INET6 4138 struct ip6_hdr *h6 = NULL; 4139 #endif /* INET6 */ 4140 struct sctphdr *hdr; 4141 struct sctp_chunkhdr *chunk; 4142 struct pf_send_entry *pfse; 4143 int off = 0; 4144 4145 MPASS(af == pd->af); 4146 4147 m = m_gethdr(M_NOWAIT, MT_DATA); 4148 if (m == NULL) 4149 return; 4150 4151 m->m_data += max_linkhdr; 4152 m->m_flags |= M_SKIP_FIREWALL; 4153 /* The rest of the stack assumes a rcvif, so provide one. 4154 * This is a locally generated packet, so .. close enough. */ 4155 m->m_pkthdr.rcvif = V_loif; 4156 4157 /* IPv4|6 header */ 4158 switch (af) { 4159 #ifdef INET 4160 case AF_INET: 4161 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4162 4163 h = mtod(m, struct ip *); 4164 4165 /* IP header fields included in the TCP checksum */ 4166 4167 h->ip_p = IPPROTO_SCTP; 4168 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4169 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4170 h->ip_src = pd->dst->v4; 4171 h->ip_dst = pd->src->v4; 4172 4173 off += sizeof(struct ip); 4174 break; 4175 #endif /* INET */ 4176 #ifdef INET6 4177 case AF_INET6: 4178 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4179 4180 h6 = mtod(m, struct ip6_hdr *); 4181 4182 /* IP header fields included in the TCP checksum */ 4183 h6->ip6_vfc |= IPV6_VERSION; 4184 h6->ip6_nxt = IPPROTO_SCTP; 4185 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4186 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4187 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4188 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4189 4190 off += sizeof(struct ip6_hdr); 4191 break; 4192 #endif /* INET6 */ 4193 default: 4194 unhandled_af(af); 4195 } 4196 4197 /* SCTP header */ 4198 hdr = mtodo(m, off); 4199 4200 hdr->src_port = pd->hdr.sctp.dest_port; 4201 hdr->dest_port = pd->hdr.sctp.src_port; 4202 hdr->v_tag = pd->sctp_initiate_tag; 4203 hdr->checksum = 0; 4204 4205 /* Abort chunk. */ 4206 off += sizeof(struct sctphdr); 4207 chunk = mtodo(m, off); 4208 4209 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4210 chunk->chunk_length = htons(sizeof(*chunk)); 4211 4212 /* SCTP checksum */ 4213 off += sizeof(*chunk); 4214 m->m_pkthdr.len = m->m_len = off; 4215 4216 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4217 4218 if (rtableid >= 0) 4219 M_SETFIB(m, rtableid); 4220 4221 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4222 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4223 if (pfse == NULL) { 4224 m_freem(m); 4225 return; 4226 } 4227 4228 switch (af) { 4229 #ifdef INET 4230 case AF_INET: 4231 pfse->pfse_type = PFSE_IP; 4232 break; 4233 #endif /* INET */ 4234 #ifdef INET6 4235 case AF_INET6: 4236 pfse->pfse_type = PFSE_IP6; 4237 break; 4238 #endif /* INET6 */ 4239 } 4240 4241 pfse->pfse_m = m; 4242 pf_send(pfse); 4243 } 4244 4245 void 4246 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4247 const struct pf_addr *saddr, const struct pf_addr *daddr, 4248 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4249 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4250 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 4251 { 4252 struct pf_send_entry *pfse; 4253 struct mbuf *m; 4254 4255 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4256 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 4257 if (m == NULL) 4258 return; 4259 4260 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4261 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4262 if (pfse == NULL) { 4263 m_freem(m); 4264 return; 4265 } 4266 4267 switch (af) { 4268 #ifdef INET 4269 case AF_INET: 4270 pfse->pfse_type = PFSE_IP; 4271 break; 4272 #endif /* INET */ 4273 #ifdef INET6 4274 case AF_INET6: 4275 pfse->pfse_type = PFSE_IP6; 4276 break; 4277 #endif /* INET6 */ 4278 default: 4279 unhandled_af(af); 4280 } 4281 4282 pfse->pfse_m = m; 4283 pf_send(pfse); 4284 } 4285 4286 static void 4287 pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum) 4288 { 4289 /* undo NAT changes, if they have taken place */ 4290 if (nr != NULL) { 4291 PF_ACPY(pd->src, &pd->osrc, pd->af); 4292 PF_ACPY(pd->dst, &pd->odst, pd->af); 4293 if (pd->sport) 4294 *pd->sport = pd->osport; 4295 if (pd->dport) 4296 *pd->dport = pd->odport; 4297 if (pd->ip_sum) 4298 *pd->ip_sum = bip_sum; 4299 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4300 } 4301 } 4302 4303 static void 4304 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4305 struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum, 4306 u_short *reason, int rtableid) 4307 { 4308 pf_undo_nat(nr, pd, bip_sum); 4309 4310 if (pd->proto == IPPROTO_TCP && 4311 ((r->rule_flag & PFRULE_RETURNRST) || 4312 (r->rule_flag & PFRULE_RETURN)) && 4313 !(tcp_get_flags(th) & TH_RST)) { 4314 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4315 4316 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4317 IPPROTO_TCP, pd->af)) 4318 REASON_SET(reason, PFRES_PROTCKSUM); 4319 else { 4320 if (tcp_get_flags(th) & TH_SYN) 4321 ack++; 4322 if (tcp_get_flags(th) & TH_FIN) 4323 ack++; 4324 pf_send_tcp(r, pd->af, pd->dst, 4325 pd->src, th->th_dport, th->th_sport, 4326 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4327 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 4328 } 4329 } else if (pd->proto == IPPROTO_SCTP && 4330 (r->rule_flag & PFRULE_RETURN)) { 4331 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4332 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4333 r->return_icmp) 4334 pf_send_icmp(pd->m, r->return_icmp >> 8, 4335 r->return_icmp & 255, pd->af, r, rtableid); 4336 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4337 r->return_icmp6) 4338 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4339 r->return_icmp6 & 255, pd->af, r, rtableid); 4340 } 4341 4342 static int 4343 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4344 { 4345 struct m_tag *mtag; 4346 u_int8_t mpcp; 4347 4348 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4349 if (mtag == NULL) 4350 return (0); 4351 4352 if (prio == PF_PRIO_ZERO) 4353 prio = 0; 4354 4355 mpcp = *(uint8_t *)(mtag + 1); 4356 4357 return (mpcp == prio); 4358 } 4359 4360 static int 4361 pf_icmp_to_bandlim(uint8_t type) 4362 { 4363 switch (type) { 4364 case ICMP_ECHO: 4365 case ICMP_ECHOREPLY: 4366 return (BANDLIM_ICMP_ECHO); 4367 case ICMP_TSTAMP: 4368 case ICMP_TSTAMPREPLY: 4369 return (BANDLIM_ICMP_TSTAMP); 4370 case ICMP_UNREACH: 4371 default: 4372 return (BANDLIM_ICMP_UNREACH); 4373 } 4374 } 4375 4376 static void 4377 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 4378 struct pf_krule *r, int rtableid) 4379 { 4380 struct pf_send_entry *pfse; 4381 struct mbuf *m0; 4382 struct pf_mtag *pf_mtag; 4383 4384 /* ICMP packet rate limitation. */ 4385 switch (af) { 4386 #ifdef INET6 4387 case AF_INET6: 4388 if (icmp6_ratelimit(NULL, type, code)) 4389 return; 4390 break; 4391 #endif /* INET6 */ 4392 #ifdef INET 4393 case AF_INET: 4394 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4395 return; 4396 break; 4397 #endif /* INET */ 4398 } 4399 4400 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4401 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4402 if (pfse == NULL) 4403 return; 4404 4405 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4406 free(pfse, M_PFTEMP); 4407 return; 4408 } 4409 4410 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4411 free(pfse, M_PFTEMP); 4412 return; 4413 } 4414 /* XXX: revisit */ 4415 m0->m_flags |= M_SKIP_FIREWALL; 4416 4417 if (rtableid >= 0) 4418 M_SETFIB(m0, rtableid); 4419 4420 #ifdef ALTQ 4421 if (r->qid) { 4422 pf_mtag->qid = r->qid; 4423 /* add hints for ecn */ 4424 pf_mtag->hdr = mtod(m0, struct ip *); 4425 } 4426 #endif /* ALTQ */ 4427 4428 switch (af) { 4429 #ifdef INET 4430 case AF_INET: 4431 pfse->pfse_type = PFSE_ICMP; 4432 break; 4433 #endif /* INET */ 4434 #ifdef INET6 4435 case AF_INET6: 4436 pfse->pfse_type = PFSE_ICMP6; 4437 break; 4438 #endif /* INET6 */ 4439 } 4440 pfse->pfse_m = m0; 4441 pfse->icmpopts.type = type; 4442 pfse->icmpopts.code = code; 4443 pf_send(pfse); 4444 } 4445 4446 /* 4447 * Return ((n = 0) == (a = b [with mask m])) 4448 * Note: n != 0 => returns (a != b [with mask m]) 4449 */ 4450 int 4451 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4452 const struct pf_addr *b, sa_family_t af) 4453 { 4454 switch (af) { 4455 #ifdef INET 4456 case AF_INET: 4457 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4458 return (n == 0); 4459 break; 4460 #endif /* INET */ 4461 #ifdef INET6 4462 case AF_INET6: 4463 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4464 return (n == 0); 4465 break; 4466 #endif /* INET6 */ 4467 } 4468 4469 return (n != 0); 4470 } 4471 4472 /* 4473 * Return 1 if b <= a <= e, otherwise return 0. 4474 */ 4475 int 4476 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4477 const struct pf_addr *a, sa_family_t af) 4478 { 4479 switch (af) { 4480 #ifdef INET 4481 case AF_INET: 4482 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4483 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4484 return (0); 4485 break; 4486 #endif /* INET */ 4487 #ifdef INET6 4488 case AF_INET6: { 4489 int i; 4490 4491 /* check a >= b */ 4492 for (i = 0; i < 4; ++i) 4493 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4494 break; 4495 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4496 return (0); 4497 /* check a <= e */ 4498 for (i = 0; i < 4; ++i) 4499 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4500 break; 4501 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4502 return (0); 4503 break; 4504 } 4505 #endif /* INET6 */ 4506 } 4507 return (1); 4508 } 4509 4510 static int 4511 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4512 { 4513 switch (op) { 4514 case PF_OP_IRG: 4515 return ((p > a1) && (p < a2)); 4516 case PF_OP_XRG: 4517 return ((p < a1) || (p > a2)); 4518 case PF_OP_RRG: 4519 return ((p >= a1) && (p <= a2)); 4520 case PF_OP_EQ: 4521 return (p == a1); 4522 case PF_OP_NE: 4523 return (p != a1); 4524 case PF_OP_LT: 4525 return (p < a1); 4526 case PF_OP_LE: 4527 return (p <= a1); 4528 case PF_OP_GT: 4529 return (p > a1); 4530 case PF_OP_GE: 4531 return (p >= a1); 4532 } 4533 return (0); /* never reached */ 4534 } 4535 4536 int 4537 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4538 { 4539 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 4540 } 4541 4542 static int 4543 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4544 { 4545 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4546 return (0); 4547 return (pf_match(op, a1, a2, u)); 4548 } 4549 4550 static int 4551 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4552 { 4553 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4554 return (0); 4555 return (pf_match(op, a1, a2, g)); 4556 } 4557 4558 int 4559 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4560 { 4561 if (*tag == -1) 4562 *tag = mtag; 4563 4564 return ((!r->match_tag_not && r->match_tag == *tag) || 4565 (r->match_tag_not && r->match_tag != *tag)); 4566 } 4567 4568 static int 4569 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4570 { 4571 struct ifnet *ifp = m->m_pkthdr.rcvif; 4572 struct pfi_kkif *kif; 4573 4574 if (ifp == NULL) 4575 return (0); 4576 4577 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4578 4579 if (kif == NULL) { 4580 DPFPRINTF(PF_DEBUG_URGENT, 4581 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 4582 r->rcv_ifname)); 4583 return (0); 4584 } 4585 4586 return (pfi_kkif_match(r->rcv_kif, kif)); 4587 } 4588 4589 int 4590 pf_tag_packet(struct pf_pdesc *pd, int tag) 4591 { 4592 4593 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4594 4595 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4596 return (ENOMEM); 4597 4598 pd->pf_mtag->tag = tag; 4599 4600 return (0); 4601 } 4602 4603 #define PF_ANCHOR_STACKSIZE 32 4604 struct pf_kanchor_stackframe { 4605 struct pf_kruleset *rs; 4606 struct pf_krule *r; /* XXX: + match bit */ 4607 struct pf_kanchor *child; 4608 }; 4609 4610 /* 4611 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4612 */ 4613 #define PF_ANCHORSTACK_MATCH 0x00000001 4614 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4615 4616 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4617 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4618 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4619 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4620 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4621 } while (0) 4622 4623 void 4624 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4625 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a) 4626 { 4627 struct pf_kanchor_stackframe *f; 4628 4629 PF_RULES_RASSERT(); 4630 4631 if (*depth >= PF_ANCHOR_STACKSIZE) { 4632 printf("%s: anchor stack overflow on %s\n", 4633 __func__, (*r)->anchor->name); 4634 *r = TAILQ_NEXT(*r, entries); 4635 return; 4636 } else if (*depth == 0 && a != NULL) 4637 *a = *r; 4638 f = stack + (*depth)++; 4639 f->rs = *rs; 4640 f->r = *r; 4641 if ((*r)->anchor_wildcard) { 4642 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4643 4644 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4645 *r = NULL; 4646 return; 4647 } 4648 *rs = &f->child->ruleset; 4649 } else { 4650 f->child = NULL; 4651 *rs = &(*r)->anchor->ruleset; 4652 } 4653 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4654 } 4655 4656 int 4657 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4658 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4659 int *match) 4660 { 4661 struct pf_kanchor_stackframe *f; 4662 struct pf_krule *fr; 4663 int quick = 0; 4664 4665 PF_RULES_RASSERT(); 4666 4667 do { 4668 if (*depth <= 0) 4669 break; 4670 f = stack + *depth - 1; 4671 fr = PF_ANCHOR_RULE(f); 4672 if (f->child != NULL) { 4673 f->child = RB_NEXT(pf_kanchor_node, 4674 &fr->anchor->children, f->child); 4675 if (f->child != NULL) { 4676 *rs = &f->child->ruleset; 4677 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4678 if (*r == NULL) 4679 continue; 4680 else 4681 break; 4682 } 4683 } 4684 (*depth)--; 4685 if (*depth == 0 && a != NULL) 4686 *a = NULL; 4687 *rs = f->rs; 4688 if (match != NULL && *match > *depth) { 4689 *match = *depth; 4690 if (f->r->quick) 4691 quick = 1; 4692 } 4693 *r = TAILQ_NEXT(fr, entries); 4694 } while (*r == NULL); 4695 4696 return (quick); 4697 } 4698 4699 struct pf_keth_anchor_stackframe { 4700 struct pf_keth_ruleset *rs; 4701 struct pf_keth_rule *r; /* XXX: + match bit */ 4702 struct pf_keth_anchor *child; 4703 }; 4704 4705 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4706 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4707 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4708 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4709 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4710 } while (0) 4711 4712 void 4713 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4714 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4715 struct pf_keth_rule **a, int *match) 4716 { 4717 struct pf_keth_anchor_stackframe *f; 4718 4719 NET_EPOCH_ASSERT(); 4720 4721 if (match) 4722 *match = 0; 4723 if (*depth >= PF_ANCHOR_STACKSIZE) { 4724 printf("%s: anchor stack overflow on %s\n", 4725 __func__, (*r)->anchor->name); 4726 *r = TAILQ_NEXT(*r, entries); 4727 return; 4728 } else if (*depth == 0 && a != NULL) 4729 *a = *r; 4730 f = stack + (*depth)++; 4731 f->rs = *rs; 4732 f->r = *r; 4733 if ((*r)->anchor_wildcard) { 4734 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4735 4736 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4737 *r = NULL; 4738 return; 4739 } 4740 *rs = &f->child->ruleset; 4741 } else { 4742 f->child = NULL; 4743 *rs = &(*r)->anchor->ruleset; 4744 } 4745 *r = TAILQ_FIRST((*rs)->active.rules); 4746 } 4747 4748 int 4749 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4750 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4751 struct pf_keth_rule **a, int *match) 4752 { 4753 struct pf_keth_anchor_stackframe *f; 4754 struct pf_keth_rule *fr; 4755 int quick = 0; 4756 4757 NET_EPOCH_ASSERT(); 4758 4759 do { 4760 if (*depth <= 0) 4761 break; 4762 f = stack + *depth - 1; 4763 fr = PF_ETH_ANCHOR_RULE(f); 4764 if (f->child != NULL) { 4765 /* 4766 * This block traverses through 4767 * a wildcard anchor. 4768 */ 4769 if (match != NULL && *match) { 4770 /* 4771 * If any of "*" matched, then 4772 * "foo/ *" matched, mark frame 4773 * appropriately. 4774 */ 4775 PF_ETH_ANCHOR_SET_MATCH(f); 4776 *match = 0; 4777 } 4778 f->child = RB_NEXT(pf_keth_anchor_node, 4779 &fr->anchor->children, f->child); 4780 if (f->child != NULL) { 4781 *rs = &f->child->ruleset; 4782 *r = TAILQ_FIRST((*rs)->active.rules); 4783 if (*r == NULL) 4784 continue; 4785 else 4786 break; 4787 } 4788 } 4789 (*depth)--; 4790 if (*depth == 0 && a != NULL) 4791 *a = NULL; 4792 *rs = f->rs; 4793 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4794 quick = fr->quick; 4795 *r = TAILQ_NEXT(fr, entries); 4796 } while (*r == NULL); 4797 4798 return (quick); 4799 } 4800 4801 #ifdef INET6 4802 void 4803 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4804 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4805 { 4806 switch (af) { 4807 #ifdef INET 4808 case AF_INET: 4809 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4810 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4811 break; 4812 #endif /* INET */ 4813 case AF_INET6: 4814 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4815 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4816 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4817 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4818 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4819 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4820 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4821 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4822 break; 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 case AF_INET6: 4836 if (addr->addr32[3] == 0xffffffff) { 4837 addr->addr32[3] = 0; 4838 if (addr->addr32[2] == 0xffffffff) { 4839 addr->addr32[2] = 0; 4840 if (addr->addr32[1] == 0xffffffff) { 4841 addr->addr32[1] = 0; 4842 addr->addr32[0] = 4843 htonl(ntohl(addr->addr32[0]) + 1); 4844 } else 4845 addr->addr32[1] = 4846 htonl(ntohl(addr->addr32[1]) + 1); 4847 } else 4848 addr->addr32[2] = 4849 htonl(ntohl(addr->addr32[2]) + 1); 4850 } else 4851 addr->addr32[3] = 4852 htonl(ntohl(addr->addr32[3]) + 1); 4853 break; 4854 } 4855 } 4856 #endif /* INET6 */ 4857 4858 void 4859 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4860 { 4861 /* 4862 * Modern rules use the same flags in rules as they do in states. 4863 */ 4864 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4865 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4866 4867 /* 4868 * Old-style scrub rules have different flags which need to be translated. 4869 */ 4870 if (r->rule_flag & PFRULE_RANDOMID) 4871 a->flags |= PFSTATE_RANDOMID; 4872 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4873 a->flags |= PFSTATE_SETTOS; 4874 a->set_tos = r->set_tos; 4875 } 4876 4877 if (r->qid) 4878 a->qid = r->qid; 4879 if (r->pqid) 4880 a->pqid = r->pqid; 4881 if (r->rtableid >= 0) 4882 a->rtableid = r->rtableid; 4883 a->log |= r->log; 4884 if (r->min_ttl) 4885 a->min_ttl = r->min_ttl; 4886 if (r->max_mss) 4887 a->max_mss = r->max_mss; 4888 if (r->dnpipe) 4889 a->dnpipe = r->dnpipe; 4890 if (r->dnrpipe) 4891 a->dnrpipe = r->dnrpipe; 4892 if (r->dnpipe || r->dnrpipe) { 4893 if (r->free_flags & PFRULE_DN_IS_PIPE) 4894 a->flags |= PFSTATE_DN_IS_PIPE; 4895 else 4896 a->flags &= ~PFSTATE_DN_IS_PIPE; 4897 } 4898 if (r->scrub_flags & PFSTATE_SETPRIO) { 4899 a->set_prio[0] = r->set_prio[0]; 4900 a->set_prio[1] = r->set_prio[1]; 4901 } 4902 if (r->allow_opts) 4903 a->allow_opts = r->allow_opts; 4904 } 4905 4906 int 4907 pf_socket_lookup(struct pf_pdesc *pd) 4908 { 4909 struct pf_addr *saddr, *daddr; 4910 u_int16_t sport, dport; 4911 struct inpcbinfo *pi; 4912 struct inpcb *inp; 4913 4914 pd->lookup.uid = UID_MAX; 4915 pd->lookup.gid = GID_MAX; 4916 4917 switch (pd->proto) { 4918 case IPPROTO_TCP: 4919 sport = pd->hdr.tcp.th_sport; 4920 dport = pd->hdr.tcp.th_dport; 4921 pi = &V_tcbinfo; 4922 break; 4923 case IPPROTO_UDP: 4924 sport = pd->hdr.udp.uh_sport; 4925 dport = pd->hdr.udp.uh_dport; 4926 pi = &V_udbinfo; 4927 break; 4928 default: 4929 return (-1); 4930 } 4931 if (pd->dir == PF_IN) { 4932 saddr = pd->src; 4933 daddr = pd->dst; 4934 } else { 4935 u_int16_t p; 4936 4937 p = sport; 4938 sport = dport; 4939 dport = p; 4940 saddr = pd->dst; 4941 daddr = pd->src; 4942 } 4943 switch (pd->af) { 4944 #ifdef INET 4945 case AF_INET: 4946 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4947 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4948 if (inp == NULL) { 4949 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4950 daddr->v4, dport, INPLOOKUP_WILDCARD | 4951 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4952 if (inp == NULL) 4953 return (-1); 4954 } 4955 break; 4956 #endif /* INET */ 4957 #ifdef INET6 4958 case AF_INET6: 4959 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4960 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4961 if (inp == NULL) { 4962 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4963 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4964 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4965 if (inp == NULL) 4966 return (-1); 4967 } 4968 break; 4969 #endif /* INET6 */ 4970 default: 4971 unhandled_af(pd->af); 4972 } 4973 INP_RLOCK_ASSERT(inp); 4974 pd->lookup.uid = inp->inp_cred->cr_uid; 4975 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4976 INP_RUNLOCK(inp); 4977 4978 return (1); 4979 } 4980 4981 u_int8_t 4982 pf_get_wscale(struct pf_pdesc *pd) 4983 { 4984 struct tcphdr *th = &pd->hdr.tcp; 4985 int hlen; 4986 u_int8_t hdr[60]; 4987 u_int8_t *opt, optlen; 4988 u_int8_t wscale = 0; 4989 4990 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4991 if (hlen <= sizeof(struct tcphdr)) 4992 return (0); 4993 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4994 return (0); 4995 opt = hdr + sizeof(struct tcphdr); 4996 hlen -= sizeof(struct tcphdr); 4997 while (hlen >= 3) { 4998 switch (*opt) { 4999 case TCPOPT_EOL: 5000 case TCPOPT_NOP: 5001 ++opt; 5002 --hlen; 5003 break; 5004 case TCPOPT_WINDOW: 5005 wscale = opt[2]; 5006 if (wscale > TCP_MAX_WINSHIFT) 5007 wscale = TCP_MAX_WINSHIFT; 5008 wscale |= PF_WSCALE_FLAG; 5009 /* FALLTHROUGH */ 5010 default: 5011 optlen = opt[1]; 5012 if (optlen < 2) 5013 optlen = 2; 5014 hlen -= optlen; 5015 opt += optlen; 5016 break; 5017 } 5018 } 5019 return (wscale); 5020 } 5021 5022 u_int16_t 5023 pf_get_mss(struct pf_pdesc *pd) 5024 { 5025 struct tcphdr *th = &pd->hdr.tcp; 5026 int hlen; 5027 u_int8_t hdr[60]; 5028 u_int8_t *opt, optlen; 5029 u_int16_t mss = V_tcp_mssdflt; 5030 5031 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 5032 if (hlen <= sizeof(struct tcphdr)) 5033 return (0); 5034 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 5035 return (0); 5036 opt = hdr + sizeof(struct tcphdr); 5037 hlen -= sizeof(struct tcphdr); 5038 while (hlen >= TCPOLEN_MAXSEG) { 5039 switch (*opt) { 5040 case TCPOPT_EOL: 5041 case TCPOPT_NOP: 5042 ++opt; 5043 --hlen; 5044 break; 5045 case TCPOPT_MAXSEG: 5046 memcpy(&mss, (opt + 2), 2); 5047 mss = ntohs(mss); 5048 /* FALLTHROUGH */ 5049 default: 5050 optlen = opt[1]; 5051 if (optlen < 2) 5052 optlen = 2; 5053 hlen -= optlen; 5054 opt += optlen; 5055 break; 5056 } 5057 } 5058 return (mss); 5059 } 5060 5061 static u_int16_t 5062 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5063 { 5064 struct nhop_object *nh; 5065 #ifdef INET6 5066 struct in6_addr dst6; 5067 uint32_t scopeid; 5068 #endif /* INET6 */ 5069 int hlen = 0; 5070 uint16_t mss = 0; 5071 5072 NET_EPOCH_ASSERT(); 5073 5074 switch (af) { 5075 #ifdef INET 5076 case AF_INET: 5077 hlen = sizeof(struct ip); 5078 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5079 if (nh != NULL) 5080 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5081 break; 5082 #endif /* INET */ 5083 #ifdef INET6 5084 case AF_INET6: 5085 hlen = sizeof(struct ip6_hdr); 5086 in6_splitscope(&addr->v6, &dst6, &scopeid); 5087 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5088 if (nh != NULL) 5089 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5090 break; 5091 #endif /* INET6 */ 5092 } 5093 5094 mss = max(V_tcp_mssdflt, mss); 5095 mss = min(mss, offer); 5096 mss = max(mss, 64); /* sanity - at least max opt space */ 5097 return (mss); 5098 } 5099 5100 static u_int32_t 5101 pf_tcp_iss(struct pf_pdesc *pd) 5102 { 5103 SHA512_CTX ctx; 5104 union { 5105 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5106 uint32_t words[1]; 5107 } digest; 5108 5109 if (V_pf_tcp_secret_init == 0) { 5110 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5111 SHA512_Init(&V_pf_tcp_secret_ctx); 5112 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5113 sizeof(V_pf_tcp_secret)); 5114 V_pf_tcp_secret_init = 1; 5115 } 5116 5117 ctx = V_pf_tcp_secret_ctx; 5118 5119 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5120 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5121 switch (pd->af) { 5122 case AF_INET6: 5123 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5124 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5125 break; 5126 case AF_INET: 5127 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5128 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5129 break; 5130 } 5131 SHA512_Final(digest.bytes, &ctx); 5132 V_pf_tcp_iss_off += 4096; 5133 #define ISN_RANDOM_INCREMENT (4096 - 1) 5134 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5135 V_pf_tcp_iss_off); 5136 #undef ISN_RANDOM_INCREMENT 5137 } 5138 5139 static bool 5140 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5141 { 5142 bool match = true; 5143 5144 /* Always matches if not set */ 5145 if (! r->isset) 5146 return (!r->neg); 5147 5148 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5149 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5150 match = false; 5151 break; 5152 } 5153 } 5154 5155 return (match ^ r->neg); 5156 } 5157 5158 static int 5159 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5160 { 5161 if (*tag == -1) 5162 *tag = mtag; 5163 5164 return ((!r->match_tag_not && r->match_tag == *tag) || 5165 (r->match_tag_not && r->match_tag != *tag)); 5166 } 5167 5168 static void 5169 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5170 { 5171 /* If we don't have the interface drop the packet. */ 5172 if (ifp == NULL) { 5173 m_freem(m); 5174 return; 5175 } 5176 5177 switch (ifp->if_type) { 5178 case IFT_ETHER: 5179 case IFT_XETHER: 5180 case IFT_L2VLAN: 5181 case IFT_BRIDGE: 5182 case IFT_IEEE8023ADLAG: 5183 break; 5184 default: 5185 m_freem(m); 5186 return; 5187 } 5188 5189 ifp->if_transmit(ifp, m); 5190 } 5191 5192 static int 5193 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5194 { 5195 #ifdef INET 5196 struct ip ip; 5197 #endif /* INET */ 5198 #ifdef INET6 5199 struct ip6_hdr ip6; 5200 #endif /* INET6 */ 5201 struct mbuf *m = *m0; 5202 struct ether_header *e; 5203 struct pf_keth_rule *r, *rm, *a = NULL; 5204 struct pf_keth_ruleset *ruleset = NULL; 5205 struct pf_mtag *mtag; 5206 struct pf_keth_ruleq *rules; 5207 struct pf_addr *src = NULL, *dst = NULL; 5208 struct pfi_kkif *bridge_to; 5209 sa_family_t af = 0; 5210 uint16_t proto; 5211 int asd = 0, match = 0; 5212 int tag = -1; 5213 uint8_t action; 5214 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5215 5216 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5217 NET_EPOCH_ASSERT(); 5218 5219 PF_RULES_RLOCK_TRACKER; 5220 5221 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5222 5223 mtag = pf_find_mtag(m); 5224 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5225 /* Dummynet re-injects packets after they've 5226 * completed their delay. We've already 5227 * processed them, so pass unconditionally. */ 5228 5229 /* But only once. We may see the packet multiple times (e.g. 5230 * PFIL_IN/PFIL_OUT). */ 5231 pf_dummynet_flag_remove(m, mtag); 5232 5233 return (PF_PASS); 5234 } 5235 5236 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5237 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5238 DPFPRINTF(PF_DEBUG_URGENT, 5239 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 5240 ", pullup failed\n")); 5241 return (PF_DROP); 5242 } 5243 e = mtod(m, struct ether_header *); 5244 proto = ntohs(e->ether_type); 5245 5246 switch (proto) { 5247 #ifdef INET 5248 case ETHERTYPE_IP: { 5249 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5250 sizeof(ip))) 5251 return (PF_DROP); 5252 5253 af = AF_INET; 5254 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5255 (caddr_t)&ip); 5256 src = (struct pf_addr *)&ip.ip_src; 5257 dst = (struct pf_addr *)&ip.ip_dst; 5258 break; 5259 } 5260 #endif /* INET */ 5261 #ifdef INET6 5262 case ETHERTYPE_IPV6: { 5263 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5264 sizeof(ip6))) 5265 return (PF_DROP); 5266 5267 af = AF_INET6; 5268 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5269 (caddr_t)&ip6); 5270 src = (struct pf_addr *)&ip6.ip6_src; 5271 dst = (struct pf_addr *)&ip6.ip6_dst; 5272 break; 5273 } 5274 #endif /* INET6 */ 5275 } 5276 5277 PF_RULES_RLOCK(); 5278 5279 ruleset = V_pf_keth; 5280 rules = atomic_load_ptr(&ruleset->active.rules); 5281 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5282 counter_u64_add(r->evaluations, 1); 5283 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5284 5285 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5286 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5287 "kif"); 5288 r = r->skip[PFE_SKIP_IFP].ptr; 5289 } 5290 else if (r->direction && r->direction != dir) { 5291 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5292 "dir"); 5293 r = r->skip[PFE_SKIP_DIR].ptr; 5294 } 5295 else if (r->proto && r->proto != proto) { 5296 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5297 "proto"); 5298 r = r->skip[PFE_SKIP_PROTO].ptr; 5299 } 5300 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5301 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5302 "src"); 5303 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5304 } 5305 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5306 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5307 "dst"); 5308 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5309 } 5310 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5311 r->ipsrc.neg, kif, M_GETFIB(m))) { 5312 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5313 "ip_src"); 5314 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5315 } 5316 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5317 r->ipdst.neg, kif, M_GETFIB(m))) { 5318 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5319 "ip_dst"); 5320 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5321 } 5322 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5323 mtag ? mtag->tag : 0)) { 5324 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5325 "match_tag"); 5326 r = TAILQ_NEXT(r, entries); 5327 } 5328 else { 5329 if (r->tag) 5330 tag = r->tag; 5331 if (r->anchor == NULL) { 5332 /* Rule matches */ 5333 rm = r; 5334 5335 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5336 5337 if (r->quick) 5338 break; 5339 5340 r = TAILQ_NEXT(r, entries); 5341 } else { 5342 pf_step_into_keth_anchor(anchor_stack, &asd, 5343 &ruleset, &r, &a, &match); 5344 } 5345 } 5346 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5347 &ruleset, &r, &a, &match)) 5348 break; 5349 } 5350 5351 r = rm; 5352 5353 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5354 5355 /* Default to pass. */ 5356 if (r == NULL) { 5357 PF_RULES_RUNLOCK(); 5358 return (PF_PASS); 5359 } 5360 5361 /* Execute action. */ 5362 counter_u64_add(r->packets[dir == PF_OUT], 1); 5363 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5364 pf_update_timestamp(r); 5365 5366 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5367 if (r->action == PF_DROP) { 5368 PF_RULES_RUNLOCK(); 5369 return (PF_DROP); 5370 } 5371 5372 if (tag > 0) { 5373 if (mtag == NULL) 5374 mtag = pf_get_mtag(m); 5375 if (mtag == NULL) { 5376 PF_RULES_RUNLOCK(); 5377 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5378 return (PF_DROP); 5379 } 5380 mtag->tag = tag; 5381 } 5382 5383 if (r->qid != 0) { 5384 if (mtag == NULL) 5385 mtag = pf_get_mtag(m); 5386 if (mtag == NULL) { 5387 PF_RULES_RUNLOCK(); 5388 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5389 return (PF_DROP); 5390 } 5391 mtag->qid = r->qid; 5392 } 5393 5394 action = r->action; 5395 bridge_to = r->bridge_to; 5396 5397 /* Dummynet */ 5398 if (r->dnpipe) { 5399 struct ip_fw_args dnflow; 5400 5401 /* Drop packet if dummynet is not loaded. */ 5402 if (ip_dn_io_ptr == NULL) { 5403 PF_RULES_RUNLOCK(); 5404 m_freem(m); 5405 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5406 return (PF_DROP); 5407 } 5408 if (mtag == NULL) 5409 mtag = pf_get_mtag(m); 5410 if (mtag == NULL) { 5411 PF_RULES_RUNLOCK(); 5412 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5413 return (PF_DROP); 5414 } 5415 5416 bzero(&dnflow, sizeof(dnflow)); 5417 5418 /* We don't have port numbers here, so we set 0. That means 5419 * that we'll be somewhat limited in distinguishing flows (i.e. 5420 * only based on IP addresses, not based on port numbers), but 5421 * it's better than nothing. */ 5422 dnflow.f_id.dst_port = 0; 5423 dnflow.f_id.src_port = 0; 5424 dnflow.f_id.proto = 0; 5425 5426 dnflow.rule.info = r->dnpipe; 5427 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5428 if (r->dnflags & PFRULE_DN_IS_PIPE) 5429 dnflow.rule.info |= IPFW_IS_PIPE; 5430 5431 dnflow.f_id.extra = dnflow.rule.info; 5432 5433 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5434 dnflow.flags |= IPFW_ARGS_ETHER; 5435 dnflow.ifp = kif->pfik_ifp; 5436 5437 switch (af) { 5438 case AF_INET: 5439 dnflow.f_id.addr_type = 4; 5440 dnflow.f_id.src_ip = src->v4.s_addr; 5441 dnflow.f_id.dst_ip = dst->v4.s_addr; 5442 break; 5443 case AF_INET6: 5444 dnflow.flags |= IPFW_ARGS_IP6; 5445 dnflow.f_id.addr_type = 6; 5446 dnflow.f_id.src_ip6 = src->v6; 5447 dnflow.f_id.dst_ip6 = dst->v6; 5448 break; 5449 } 5450 5451 PF_RULES_RUNLOCK(); 5452 5453 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5454 ip_dn_io_ptr(m0, &dnflow); 5455 if (*m0 != NULL) 5456 pf_dummynet_flag_remove(m, mtag); 5457 } else { 5458 PF_RULES_RUNLOCK(); 5459 } 5460 5461 if (action == PF_PASS && bridge_to) { 5462 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5463 *m0 = NULL; /* We've eaten the packet. */ 5464 } 5465 5466 return (action); 5467 } 5468 5469 #define PF_TEST_ATTRIB(t, a)\ 5470 do { \ 5471 if (t) { \ 5472 r = a; \ 5473 goto nextrule; \ 5474 } \ 5475 } while (0) 5476 5477 static int 5478 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5479 struct pf_pdesc *pd, struct pf_krule **am, 5480 struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp) 5481 { 5482 struct pf_krule *nr = NULL; 5483 struct pf_krule *r, *a = NULL; 5484 struct pf_kruleset *ruleset = NULL; 5485 struct pf_krule_slist match_rules; 5486 struct pf_krule_item *ri; 5487 struct tcphdr *th = &pd->hdr.tcp; 5488 struct pf_state_key *sk = NULL, *nk = NULL; 5489 u_short transerror; 5490 int rewrite = 0; 5491 int tag = -1; 5492 int asd = 0; 5493 int match = 0; 5494 int state_icmp = 0, icmp_dir; 5495 int action = PF_PASS; 5496 u_int16_t virtual_type, virtual_id; 5497 u_int16_t bproto_sum = 0, bip_sum = 0; 5498 u_int8_t icmptype = 0, icmpcode = 0; 5499 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5500 struct pf_udp_mapping *udp_mapping = NULL; 5501 5502 PF_RULES_RASSERT(); 5503 5504 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5505 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5506 5507 SLIST_INIT(&match_rules); 5508 5509 if (inp != NULL) { 5510 INP_LOCK_ASSERT(inp); 5511 pd->lookup.uid = inp->inp_cred->cr_uid; 5512 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 5513 pd->lookup.done = 1; 5514 } 5515 5516 switch (pd->virtual_proto) { 5517 case IPPROTO_TCP: 5518 pd->nsport = th->th_sport; 5519 pd->ndport = th->th_dport; 5520 break; 5521 case IPPROTO_UDP: 5522 pd->nsport = pd->hdr.udp.uh_sport; 5523 pd->ndport = pd->hdr.udp.uh_dport; 5524 break; 5525 case IPPROTO_SCTP: 5526 pd->nsport = pd->hdr.sctp.src_port; 5527 pd->ndport = pd->hdr.sctp.dest_port; 5528 break; 5529 #ifdef INET 5530 case IPPROTO_ICMP: 5531 MPASS(pd->af == AF_INET); 5532 icmptype = pd->hdr.icmp.icmp_type; 5533 icmpcode = pd->hdr.icmp.icmp_code; 5534 state_icmp = pf_icmp_mapping(pd, icmptype, 5535 &icmp_dir, &virtual_id, &virtual_type); 5536 if (icmp_dir == PF_IN) { 5537 pd->nsport = virtual_id; 5538 pd->ndport = virtual_type; 5539 } else { 5540 pd->nsport = virtual_type; 5541 pd->ndport = virtual_id; 5542 } 5543 break; 5544 #endif /* INET */ 5545 #ifdef INET6 5546 case IPPROTO_ICMPV6: 5547 MPASS(pd->af == AF_INET6); 5548 icmptype = pd->hdr.icmp6.icmp6_type; 5549 icmpcode = pd->hdr.icmp6.icmp6_code; 5550 state_icmp = pf_icmp_mapping(pd, icmptype, 5551 &icmp_dir, &virtual_id, &virtual_type); 5552 if (icmp_dir == PF_IN) { 5553 pd->nsport = virtual_id; 5554 pd->ndport = virtual_type; 5555 } else { 5556 pd->nsport = virtual_type; 5557 pd->ndport = virtual_id; 5558 } 5559 5560 break; 5561 #endif /* INET6 */ 5562 default: 5563 pd->nsport = pd->ndport = 0; 5564 break; 5565 } 5566 pd->osport = pd->nsport; 5567 pd->odport = pd->ndport; 5568 5569 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5570 5571 /* check packet for BINAT/NAT/RDR */ 5572 transerror = pf_get_translation(pd, pd->off, &sk, &nk, anchor_stack, 5573 &nr, &udp_mapping); 5574 switch (transerror) { 5575 default: 5576 /* A translation error occurred. */ 5577 REASON_SET(reason, transerror); 5578 goto cleanup; 5579 case PFRES_MAX: 5580 /* No match. */ 5581 break; 5582 case PFRES_MATCH: 5583 KASSERT(sk != NULL, ("%s: null sk", __func__)); 5584 KASSERT(nk != NULL, ("%s: null nk", __func__)); 5585 5586 if (nr->log) { 5587 PFLOG_PACKET(nr->action, PFRES_MATCH, nr, a, 5588 ruleset, pd, 1, NULL); 5589 } 5590 5591 if (pd->ip_sum) 5592 bip_sum = *pd->ip_sum; 5593 5594 switch (pd->proto) { 5595 case IPPROTO_TCP: 5596 bproto_sum = th->th_sum; 5597 5598 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5599 nk->port[pd->sidx] != pd->nsport) { 5600 pf_change_ap(pd, pd->src, &th->th_sport, 5601 &nk->addr[pd->sidx], nk->port[pd->sidx]); 5602 pd->sport = &th->th_sport; 5603 pd->nsport = th->th_sport; 5604 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5605 } 5606 5607 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5608 nk->port[pd->didx] != pd->ndport) { 5609 pf_change_ap(pd, pd->dst, &th->th_dport, 5610 &nk->addr[pd->didx], nk->port[pd->didx]); 5611 pd->dport = &th->th_dport; 5612 pd->ndport = th->th_dport; 5613 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5614 } 5615 rewrite++; 5616 break; 5617 case IPPROTO_UDP: 5618 bproto_sum = pd->hdr.udp.uh_sum; 5619 5620 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5621 nk->port[pd->sidx] != pd->nsport) { 5622 pf_change_ap(pd, pd->src, 5623 &pd->hdr.udp.uh_sport, 5624 &nk->addr[pd->sidx], 5625 nk->port[pd->sidx]); 5626 pd->sport = &pd->hdr.udp.uh_sport; 5627 pd->nsport = pd->hdr.udp.uh_sport; 5628 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5629 } 5630 5631 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5632 nk->port[pd->didx] != pd->ndport) { 5633 pf_change_ap(pd, pd->dst, 5634 &pd->hdr.udp.uh_dport, 5635 &nk->addr[pd->didx], 5636 nk->port[pd->didx]); 5637 pd->dport = &pd->hdr.udp.uh_dport; 5638 pd->ndport = pd->hdr.udp.uh_dport; 5639 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5640 } 5641 rewrite++; 5642 break; 5643 case IPPROTO_SCTP: { 5644 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5645 nk->port[pd->sidx] != pd->nsport) { 5646 pf_change_ap(pd, pd->src, 5647 &pd->hdr.sctp.src_port, 5648 &nk->addr[pd->sidx], 5649 nk->port[pd->sidx]); 5650 pd->sport = &pd->hdr.sctp.src_port; 5651 pd->nsport = pd->hdr.sctp.src_port; 5652 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5653 } 5654 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5655 nk->port[pd->didx] != pd->ndport) { 5656 pf_change_ap(pd, pd->dst, 5657 &pd->hdr.sctp.dest_port, 5658 &nk->addr[pd->didx], 5659 nk->port[pd->didx]); 5660 pd->dport = &pd->hdr.sctp.dest_port; 5661 pd->ndport = pd->hdr.sctp.dest_port; 5662 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5663 } 5664 break; 5665 } 5666 #ifdef INET 5667 case IPPROTO_ICMP: 5668 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 5669 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 5670 nk->addr[pd->sidx].v4.s_addr, 0); 5671 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5672 } 5673 5674 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 5675 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 5676 nk->addr[pd->didx].v4.s_addr, 0); 5677 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5678 } 5679 5680 if (virtual_type == htons(ICMP_ECHO) && 5681 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5682 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5683 pd->hdr.icmp.icmp_cksum, pd->nsport, 5684 nk->port[pd->sidx], 0); 5685 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5686 pd->sport = &pd->hdr.icmp.icmp_id; 5687 } 5688 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5689 break; 5690 #endif /* INET */ 5691 #ifdef INET6 5692 case IPPROTO_ICMPV6: 5693 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 5694 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 5695 &nk->addr[pd->sidx], 0); 5696 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5697 } 5698 5699 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 5700 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 5701 &nk->addr[pd->didx], 0); 5702 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5703 } 5704 rewrite++; 5705 break; 5706 #endif /* INET */ 5707 default: 5708 switch (pd->af) { 5709 #ifdef INET 5710 case AF_INET: 5711 if (PF_ANEQ(&pd->nsaddr, 5712 &nk->addr[pd->sidx], AF_INET)) { 5713 pf_change_a(&pd->src->v4.s_addr, 5714 pd->ip_sum, 5715 nk->addr[pd->sidx].v4.s_addr, 0); 5716 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5717 } 5718 5719 if (PF_ANEQ(&pd->ndaddr, 5720 &nk->addr[pd->didx], AF_INET)) { 5721 pf_change_a(&pd->dst->v4.s_addr, 5722 pd->ip_sum, 5723 nk->addr[pd->didx].v4.s_addr, 0); 5724 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5725 } 5726 break; 5727 #endif /* INET */ 5728 #ifdef INET6 5729 case AF_INET6: 5730 if (PF_ANEQ(&pd->nsaddr, 5731 &nk->addr[pd->sidx], AF_INET6)) { 5732 PF_ACPY(&pd->nsaddr, &nk->addr[pd->sidx], pd->af); 5733 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5734 } 5735 5736 if (PF_ANEQ(&pd->ndaddr, 5737 &nk->addr[pd->didx], AF_INET6)) { 5738 PF_ACPY(&pd->ndaddr, &nk->addr[pd->didx], pd->af); 5739 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5740 } 5741 break; 5742 #endif /* INET6 */ 5743 } 5744 break; 5745 } 5746 if (nr->natpass) 5747 r = NULL; 5748 } 5749 5750 while (r != NULL) { 5751 if (pd->related_rule) { 5752 *rm = pd->related_rule; 5753 break; 5754 } 5755 pf_counter_u64_add(&r->evaluations, 1); 5756 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5757 r->skip[PF_SKIP_IFP]); 5758 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5759 r->skip[PF_SKIP_DIR]); 5760 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5761 r->skip[PF_SKIP_AF]); 5762 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5763 r->skip[PF_SKIP_PROTO]); 5764 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5765 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5766 r->skip[PF_SKIP_SRC_ADDR]); 5767 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5768 r->dst.neg, NULL, M_GETFIB(pd->m)), 5769 r->skip[PF_SKIP_DST_ADDR]); 5770 switch (pd->virtual_proto) { 5771 case PF_VPROTO_FRAGMENT: 5772 /* tcp/udp only. port_op always 0 in other cases */ 5773 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5774 TAILQ_NEXT(r, entries)); 5775 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5776 TAILQ_NEXT(r, entries)); 5777 /* icmp only. type/code always 0 in other cases */ 5778 PF_TEST_ATTRIB((r->type || r->code), 5779 TAILQ_NEXT(r, entries)); 5780 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5781 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5782 TAILQ_NEXT(r, entries)); 5783 break; 5784 5785 case IPPROTO_TCP: 5786 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(th)) != r->flags, 5787 TAILQ_NEXT(r, entries)); 5788 /* FALLTHROUGH */ 5789 case IPPROTO_SCTP: 5790 case IPPROTO_UDP: 5791 /* tcp/udp only. port_op always 0 in other cases */ 5792 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5793 r->src.port[0], r->src.port[1], pd->nsport), 5794 r->skip[PF_SKIP_SRC_PORT]); 5795 /* tcp/udp only. port_op always 0 in other cases */ 5796 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5797 r->dst.port[0], r->dst.port[1], pd->ndport), 5798 r->skip[PF_SKIP_DST_PORT]); 5799 /* tcp/udp only. uid.op always 0 in other cases */ 5800 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5801 pf_socket_lookup(pd), 1)) && 5802 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5803 pd->lookup.uid), 5804 TAILQ_NEXT(r, entries)); 5805 /* tcp/udp only. gid.op always 0 in other cases */ 5806 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5807 pf_socket_lookup(pd), 1)) && 5808 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5809 pd->lookup.gid), 5810 TAILQ_NEXT(r, entries)); 5811 break; 5812 5813 case IPPROTO_ICMP: 5814 case IPPROTO_ICMPV6: 5815 /* icmp only. type always 0 in other cases */ 5816 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5817 TAILQ_NEXT(r, entries)); 5818 /* icmp only. type always 0 in other cases */ 5819 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5820 TAILQ_NEXT(r, entries)); 5821 break; 5822 5823 default: 5824 break; 5825 } 5826 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5827 TAILQ_NEXT(r, entries)); 5828 PF_TEST_ATTRIB(r->prio && 5829 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5830 TAILQ_NEXT(r, entries)); 5831 PF_TEST_ATTRIB(r->prob && 5832 r->prob <= arc4random(), 5833 TAILQ_NEXT(r, entries)); 5834 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5835 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5836 TAILQ_NEXT(r, entries)); 5837 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 5838 r->rcvifnot), 5839 TAILQ_NEXT(r, entries)); 5840 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5841 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5842 TAILQ_NEXT(r, entries)); 5843 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5844 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5845 pf_osfp_fingerprint(pd, th), 5846 r->os_fingerprint)), 5847 TAILQ_NEXT(r, entries)); 5848 /* FALLTHROUGH */ 5849 if (r->tag) 5850 tag = r->tag; 5851 if (r->anchor == NULL) { 5852 if (r->action == PF_MATCH) { 5853 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5854 if (ri == NULL) { 5855 REASON_SET(reason, PFRES_MEMORY); 5856 goto cleanup; 5857 } 5858 ri->r = r; 5859 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5860 pf_counter_u64_critical_enter(); 5861 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5862 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5863 pf_counter_u64_critical_exit(); 5864 pf_rule_to_actions(r, &pd->act); 5865 if (r->rule_flag & PFRULE_AFTO) 5866 pd->naf = r->naf; 5867 if (pd->af != pd->naf) { 5868 if (pf_get_transaddr_af(r, pd) == -1) { 5869 REASON_SET(reason, PFRES_TRANSLATE); 5870 goto cleanup; 5871 } 5872 } 5873 if (r->log) 5874 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5875 a, ruleset, pd, 1, NULL); 5876 } else { 5877 match = asd; 5878 *rm = r; 5879 *am = a; 5880 *rsm = ruleset; 5881 } 5882 if (pd->act.log & PF_LOG_MATCHES) 5883 pf_log_matches(pd, r, a, ruleset, &match_rules); 5884 if (r->quick) 5885 break; 5886 r = TAILQ_NEXT(r, entries); 5887 } else 5888 pf_step_into_anchor(anchor_stack, &asd, 5889 &ruleset, PF_RULESET_FILTER, &r, &a); 5890 nextrule: 5891 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5892 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5893 break; 5894 } 5895 r = *rm; 5896 a = *am; 5897 ruleset = *rsm; 5898 5899 REASON_SET(reason, PFRES_MATCH); 5900 5901 /* apply actions for last matching pass/block rule */ 5902 pf_rule_to_actions(r, &pd->act); 5903 if (r->rule_flag & PFRULE_AFTO) 5904 pd->naf = r->naf; 5905 if (pd->af != pd->naf) { 5906 if (pf_get_transaddr_af(r, pd) == -1) { 5907 REASON_SET(reason, PFRES_TRANSLATE); 5908 goto cleanup; 5909 } 5910 } 5911 5912 if (r->log) { 5913 if (rewrite) 5914 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5915 PFLOG_PACKET(r->action, *reason, r, a, ruleset, pd, 1, NULL); 5916 } 5917 if (pd->act.log & PF_LOG_MATCHES) 5918 pf_log_matches(pd, r, a, ruleset, &match_rules); 5919 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5920 (r->action == PF_DROP) && 5921 ((r->rule_flag & PFRULE_RETURNRST) || 5922 (r->rule_flag & PFRULE_RETURNICMP) || 5923 (r->rule_flag & PFRULE_RETURN))) { 5924 pf_return(r, nr, pd, th, bproto_sum, 5925 bip_sum, reason, r->rtableid); 5926 } 5927 5928 if (r->action == PF_DROP) 5929 goto cleanup; 5930 5931 if (tag > 0 && pf_tag_packet(pd, tag)) { 5932 REASON_SET(reason, PFRES_MEMORY); 5933 goto cleanup; 5934 } 5935 if (pd->act.rtableid >= 0) 5936 M_SETFIB(pd->m, pd->act.rtableid); 5937 5938 if (r->rt) { 5939 struct pf_ksrc_node *sn = NULL; 5940 struct pf_srchash *snh = NULL; 5941 struct pf_kpool *pool = &r->route; 5942 5943 /* Backwards compatibility. */ 5944 if (TAILQ_EMPTY(&pool->list)) 5945 pool = &r->rdr; 5946 5947 /* 5948 * Set act.rt here instead of in pf_rule_to_actions() because 5949 * it is applied only from the last pass rule. 5950 */ 5951 pd->act.rt = r->rt; 5952 /* Don't use REASON_SET, pf_map_addr increases the reason counters */ 5953 *reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, 5954 &pd->act.rt_kif, NULL, &sn, &snh, pool, PF_SN_ROUTE); 5955 if (*reason != 0) 5956 goto cleanup; 5957 } 5958 5959 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5960 (!state_icmp && (r->keep_state || nr != NULL || 5961 (pd->flags & PFDESC_TCP_NORM)))) { 5962 bool nat64; 5963 5964 action = pf_create_state(r, nr, a, pd, nk, sk, 5965 &rewrite, sm, tag, bproto_sum, bip_sum, 5966 &match_rules, udp_mapping, reason); 5967 sk = nk = NULL; 5968 if (action != PF_PASS) { 5969 pf_udp_mapping_release(udp_mapping); 5970 if (r->log || (nr != NULL && nr->log) || 5971 *reason == PFRES_MEMORY) 5972 pd->act.log |= PF_LOG_FORCE; 5973 if (action == PF_DROP && 5974 (r->rule_flag & PFRULE_RETURN)) 5975 pf_return(r, nr, pd, th, 5976 bproto_sum, bip_sum, reason, 5977 pd->act.rtableid); 5978 return (action); 5979 } 5980 5981 nat64 = pd->af != pd->naf; 5982 if (nat64) { 5983 int ret; 5984 5985 if (sk == NULL) 5986 sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 5987 if (nk == NULL) 5988 nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 5989 5990 if (pd->dir == PF_IN) { 5991 ret = pf_translate(pd, &sk->addr[pd->didx], 5992 sk->port[pd->didx], &sk->addr[pd->sidx], 5993 sk->port[pd->sidx], virtual_type, 5994 icmp_dir); 5995 } else { 5996 ret = pf_translate(pd, &sk->addr[pd->sidx], 5997 sk->port[pd->sidx], &sk->addr[pd->didx], 5998 sk->port[pd->didx], virtual_type, 5999 icmp_dir); 6000 } 6001 6002 if (ret < 0) 6003 goto cleanup; 6004 6005 rewrite += ret; 6006 6007 if (rewrite && sk->af != nk->af) 6008 action = PF_AFRT; 6009 } 6010 } else { 6011 while ((ri = SLIST_FIRST(&match_rules))) { 6012 SLIST_REMOVE_HEAD(&match_rules, entry); 6013 free(ri, M_PF_RULE_ITEM); 6014 } 6015 6016 uma_zfree(V_pf_state_key_z, sk); 6017 uma_zfree(V_pf_state_key_z, nk); 6018 sk = nk = NULL; 6019 pf_udp_mapping_release(udp_mapping); 6020 } 6021 6022 /* copy back packet headers if we performed NAT operations */ 6023 if (rewrite) 6024 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6025 6026 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 6027 pd->dir == PF_OUT && 6028 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 6029 /* 6030 * We want the state created, but we dont 6031 * want to send this in case a partner 6032 * firewall has to know about it to allow 6033 * replies through it. 6034 */ 6035 return (PF_DEFER); 6036 6037 return (action); 6038 6039 cleanup: 6040 while ((ri = SLIST_FIRST(&match_rules))) { 6041 SLIST_REMOVE_HEAD(&match_rules, entry); 6042 free(ri, M_PF_RULE_ITEM); 6043 } 6044 6045 uma_zfree(V_pf_state_key_z, sk); 6046 uma_zfree(V_pf_state_key_z, nk); 6047 pf_udp_mapping_release(udp_mapping); 6048 6049 return (PF_DROP); 6050 } 6051 6052 static int 6053 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 6054 struct pf_pdesc *pd, struct pf_state_key *nk, struct pf_state_key *sk, 6055 int *rewrite, struct pf_kstate **sm, int tag, u_int16_t bproto_sum, 6056 u_int16_t bip_sum, struct pf_krule_slist *match_rules, 6057 struct pf_udp_mapping *udp_mapping, u_short *reason) 6058 { 6059 struct pf_kstate *s = NULL; 6060 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6061 /* 6062 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6063 * but for PF_SN_NAT it is different. Don't try optimizing it, 6064 * just store all 3 hashes. 6065 */ 6066 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6067 struct tcphdr *th = &pd->hdr.tcp; 6068 u_int16_t mss = V_tcp_mssdflt; 6069 u_short sn_reason; 6070 struct pf_krule_item *ri; 6071 struct pf_kpool *pool_route = &r->route; 6072 6073 /* check maximums */ 6074 if (r->max_states && 6075 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6076 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6077 REASON_SET(reason, PFRES_MAXSTATES); 6078 goto csfailed; 6079 } 6080 /* src node for limits */ 6081 if ((r->rule_flag & PFRULE_SRCTRACK) && 6082 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6083 NULL, NULL, PF_SN_LIMIT)) != 0) { 6084 REASON_SET(reason, sn_reason); 6085 goto csfailed; 6086 } 6087 /* src node for route-to rule */ 6088 if (TAILQ_EMPTY(&pool_route->list)) /* Backwards compatibility. */ 6089 pool_route = &r->rdr; 6090 if ((pool_route->opts & PF_POOL_STICKYADDR) && 6091 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6092 &pd->act.rt_addr, pd->act.rt_kif, PF_SN_ROUTE)) != 0) { 6093 REASON_SET(reason, sn_reason); 6094 goto csfailed; 6095 } 6096 /* src node for translation rule */ 6097 if (nr != NULL && (nr->rdr.opts & PF_POOL_STICKYADDR) && 6098 (sn_reason = pf_insert_src_node(sns, snhs, nr, &sk->addr[pd->sidx], 6099 pd->af, &nk->addr[1], NULL, PF_SN_NAT)) != 0 ) { 6100 REASON_SET(reason, sn_reason); 6101 goto csfailed; 6102 } 6103 s = pf_alloc_state(M_NOWAIT); 6104 if (s == NULL) { 6105 REASON_SET(reason, PFRES_MEMORY); 6106 goto csfailed; 6107 } 6108 s->rule = r; 6109 s->nat_rule = nr; 6110 s->anchor = a; 6111 memcpy(&s->match_rules, match_rules, sizeof(s->match_rules)); 6112 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6113 6114 if (pd->act.allow_opts) 6115 s->state_flags |= PFSTATE_ALLOWOPTS; 6116 if (r->rule_flag & PFRULE_STATESLOPPY) 6117 s->state_flags |= PFSTATE_SLOPPY; 6118 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6119 s->state_flags |= PFSTATE_SCRUB_TCP; 6120 if ((r->rule_flag & PFRULE_PFLOW) || 6121 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 6122 s->state_flags |= PFSTATE_PFLOW; 6123 6124 s->act.log = pd->act.log & PF_LOG_ALL; 6125 s->sync_state = PFSYNC_S_NONE; 6126 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6127 6128 if (nr != NULL) 6129 s->act.log |= nr->log & PF_LOG_ALL; 6130 switch (pd->proto) { 6131 case IPPROTO_TCP: 6132 s->src.seqlo = ntohl(th->th_seq); 6133 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6134 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6135 r->keep_state == PF_STATE_MODULATE) { 6136 /* Generate sequence number modulator */ 6137 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6138 0) 6139 s->src.seqdiff = 1; 6140 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6141 htonl(s->src.seqlo + s->src.seqdiff), 0); 6142 *rewrite = 1; 6143 } else 6144 s->src.seqdiff = 0; 6145 if (tcp_get_flags(th) & TH_SYN) { 6146 s->src.seqhi++; 6147 s->src.wscale = pf_get_wscale(pd); 6148 } 6149 s->src.max_win = MAX(ntohs(th->th_win), 1); 6150 if (s->src.wscale & PF_WSCALE_MASK) { 6151 /* Remove scale factor from initial window */ 6152 int win = s->src.max_win; 6153 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6154 s->src.max_win = (win - 1) >> 6155 (s->src.wscale & PF_WSCALE_MASK); 6156 } 6157 if (tcp_get_flags(th) & TH_FIN) 6158 s->src.seqhi++; 6159 s->dst.seqhi = 1; 6160 s->dst.max_win = 1; 6161 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6162 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6163 s->timeout = PFTM_TCP_FIRST_PACKET; 6164 atomic_add_32(&V_pf_status.states_halfopen, 1); 6165 break; 6166 case IPPROTO_UDP: 6167 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6168 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6169 s->timeout = PFTM_UDP_FIRST_PACKET; 6170 break; 6171 case IPPROTO_SCTP: 6172 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6173 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6174 s->timeout = PFTM_SCTP_FIRST_PACKET; 6175 break; 6176 case IPPROTO_ICMP: 6177 #ifdef INET6 6178 case IPPROTO_ICMPV6: 6179 #endif /* INET6 */ 6180 s->timeout = PFTM_ICMP_FIRST_PACKET; 6181 break; 6182 default: 6183 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6184 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6185 s->timeout = PFTM_OTHER_FIRST_PACKET; 6186 } 6187 6188 s->creation = s->expire = pf_get_uptime(); 6189 6190 if (pd->proto == IPPROTO_TCP) { 6191 if (s->state_flags & PFSTATE_SCRUB_TCP && 6192 pf_normalize_tcp_init(pd, th, &s->src)) { 6193 REASON_SET(reason, PFRES_MEMORY); 6194 goto csfailed; 6195 } 6196 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6197 pf_normalize_tcp_stateful(pd, reason, th, s, 6198 &s->src, &s->dst, rewrite)) { 6199 /* This really shouldn't happen!!! */ 6200 DPFPRINTF(PF_DEBUG_URGENT, 6201 ("pf_normalize_tcp_stateful failed on first " 6202 "pkt\n")); 6203 goto csfailed; 6204 } 6205 } else if (pd->proto == IPPROTO_SCTP) { 6206 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6207 goto csfailed; 6208 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6209 goto csfailed; 6210 } 6211 s->direction = pd->dir; 6212 6213 /* 6214 * sk/nk could already been setup by pf_get_translation(). 6215 */ 6216 if (nr == NULL) { 6217 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 6218 __func__, nr, sk, nk)); 6219 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6220 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6221 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, &sk, &nk)) { 6222 goto csfailed; 6223 } 6224 } else 6225 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 6226 __func__, nr, sk, nk)); 6227 6228 /* Swap sk/nk for PF_OUT. */ 6229 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6230 (pd->dir == PF_IN) ? sk : nk, 6231 (pd->dir == PF_IN) ? nk : sk, s)) { 6232 REASON_SET(reason, PFRES_STATEINS); 6233 goto drop; 6234 } else 6235 *sm = s; 6236 sk = nk = NULL; 6237 6238 STATE_INC_COUNTERS(s); 6239 6240 /* 6241 * Lock order is important: first state, then source node. 6242 */ 6243 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6244 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6245 s->sns[sn_type] = sns[sn_type]; 6246 PF_HASHROW_UNLOCK(snhs[sn_type]); 6247 } 6248 } 6249 6250 if (tag > 0) 6251 s->tag = tag; 6252 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6253 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 6254 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6255 pf_undo_nat(nr, pd, bip_sum); 6256 s->src.seqhi = arc4random(); 6257 /* Find mss option */ 6258 int rtid = M_GETFIB(pd->m); 6259 mss = pf_get_mss(pd); 6260 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6261 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6262 s->src.mss = mss; 6263 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6264 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6265 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6266 pd->act.rtableid); 6267 REASON_SET(reason, PFRES_SYNPROXY); 6268 return (PF_SYNPROXY_DROP); 6269 } 6270 6271 s->udp_mapping = udp_mapping; 6272 6273 return (PF_PASS); 6274 6275 csfailed: 6276 while ((ri = SLIST_FIRST(match_rules))) { 6277 SLIST_REMOVE_HEAD(match_rules, entry); 6278 free(ri, M_PF_RULE_ITEM); 6279 } 6280 6281 uma_zfree(V_pf_state_key_z, sk); 6282 uma_zfree(V_pf_state_key_z, nk); 6283 6284 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6285 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6286 if (--sns[sn_type]->states == 0 && 6287 sns[sn_type]->expire == 0) { 6288 pf_unlink_src_node(sns[sn_type]); 6289 pf_free_src_node(sns[sn_type]); 6290 counter_u64_add( 6291 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6292 } 6293 PF_HASHROW_UNLOCK(snhs[sn_type]); 6294 } 6295 } 6296 6297 drop: 6298 if (s != NULL) { 6299 pf_src_tree_remove_state(s); 6300 s->timeout = PFTM_UNLINKED; 6301 pf_free_state(s); 6302 } 6303 6304 return (PF_DROP); 6305 } 6306 6307 int 6308 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6309 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6310 int icmp_dir) 6311 { 6312 /* 6313 * pf_translate() implements OpenBSD's "new" NAT approach. 6314 * We don't follow it, because it involves a breaking syntax change 6315 * (removing nat/rdr rules, moving it into regular pf rules.) 6316 * It also moves NAT processing to be done after normal rules evaluation 6317 * whereas in FreeBSD that's done before rules processing. 6318 * 6319 * We adopt the function only for nat64, and keep other NAT processing 6320 * before rules processing. 6321 */ 6322 int rewrite = 0; 6323 int afto = pd->af != pd->naf; 6324 6325 MPASS(afto); 6326 6327 switch (pd->proto) { 6328 case IPPROTO_TCP: 6329 case IPPROTO_UDP: 6330 case IPPROTO_SCTP: 6331 if (afto || *pd->sport != sport) { 6332 pf_change_ap(pd, pd->src, pd->sport, 6333 saddr, sport); 6334 rewrite = 1; 6335 } 6336 if (afto || *pd->dport != dport) { 6337 pf_change_ap(pd, pd->dst, pd->dport, 6338 daddr, dport); 6339 rewrite = 1; 6340 } 6341 break; 6342 6343 #ifdef INET 6344 case IPPROTO_ICMP: 6345 /* pf_translate() is also used when logging invalid packets */ 6346 if (pd->af != AF_INET) 6347 return (0); 6348 6349 if (afto) { 6350 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6351 return (-1); 6352 pd->proto = IPPROTO_ICMPV6; 6353 rewrite = 1; 6354 } 6355 if (virtual_type == htons(ICMP_ECHO)) { 6356 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6357 6358 if (icmpid != pd->hdr.icmp.icmp_id) { 6359 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6360 pd->hdr.icmp.icmp_cksum, 6361 pd->hdr.icmp.icmp_id, icmpid, 0); 6362 pd->hdr.icmp.icmp_id = icmpid; 6363 /* XXX TODO copyback. */ 6364 rewrite = 1; 6365 } 6366 } 6367 break; 6368 #endif /* INET */ 6369 6370 #ifdef INET6 6371 case IPPROTO_ICMPV6: 6372 /* pf_translate() is also used when logging invalid packets */ 6373 if (pd->af != AF_INET6) 6374 return (0); 6375 6376 if (afto) { 6377 /* ip_sum will be recalculated in pf_translate_af */ 6378 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6379 return (0); 6380 pd->proto = IPPROTO_ICMP; 6381 rewrite = 1; 6382 } 6383 break; 6384 #endif /* INET6 */ 6385 6386 default: 6387 break; 6388 } 6389 6390 return (rewrite); 6391 } 6392 6393 static int 6394 pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd, 6395 u_short *reason, int *copyback, struct pf_state_peer *src, 6396 struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst) 6397 { 6398 struct tcphdr *th = &pd->hdr.tcp; 6399 u_int16_t win = ntohs(th->th_win); 6400 u_int32_t ack, end, data_end, seq, orig_seq; 6401 u_int8_t sws, dws; 6402 int ackskew; 6403 6404 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6405 sws = src->wscale & PF_WSCALE_MASK; 6406 dws = dst->wscale & PF_WSCALE_MASK; 6407 } else 6408 sws = dws = 0; 6409 6410 /* 6411 * Sequence tracking algorithm from Guido van Rooij's paper: 6412 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6413 * tcp_filtering.ps 6414 */ 6415 6416 orig_seq = seq = ntohl(th->th_seq); 6417 if (src->seqlo == 0) { 6418 /* First packet from this end. Set its state */ 6419 6420 if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6421 src->scrub == NULL) { 6422 if (pf_normalize_tcp_init(pd, th, src)) { 6423 REASON_SET(reason, PFRES_MEMORY); 6424 return (PF_DROP); 6425 } 6426 } 6427 6428 /* Deferred generation of sequence number modulator */ 6429 if (dst->seqdiff && !src->seqdiff) { 6430 /* use random iss for the TCP server */ 6431 while ((src->seqdiff = arc4random() - seq) == 0) 6432 ; 6433 ack = ntohl(th->th_ack) - dst->seqdiff; 6434 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6435 src->seqdiff), 0); 6436 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6437 *copyback = 1; 6438 } else { 6439 ack = ntohl(th->th_ack); 6440 } 6441 6442 end = seq + pd->p_len; 6443 if (tcp_get_flags(th) & TH_SYN) { 6444 end++; 6445 if (dst->wscale & PF_WSCALE_FLAG) { 6446 src->wscale = pf_get_wscale(pd); 6447 if (src->wscale & PF_WSCALE_FLAG) { 6448 /* Remove scale factor from initial 6449 * window */ 6450 sws = src->wscale & PF_WSCALE_MASK; 6451 win = ((u_int32_t)win + (1 << sws) - 1) 6452 >> sws; 6453 dws = dst->wscale & PF_WSCALE_MASK; 6454 } else { 6455 /* fixup other window */ 6456 dst->max_win = MIN(TCP_MAXWIN, 6457 (u_int32_t)dst->max_win << 6458 (dst->wscale & PF_WSCALE_MASK)); 6459 /* in case of a retrans SYN|ACK */ 6460 dst->wscale = 0; 6461 } 6462 } 6463 } 6464 data_end = end; 6465 if (tcp_get_flags(th) & TH_FIN) 6466 end++; 6467 6468 src->seqlo = seq; 6469 if (src->state < TCPS_SYN_SENT) 6470 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6471 6472 /* 6473 * May need to slide the window (seqhi may have been set by 6474 * the crappy stack check or if we picked up the connection 6475 * after establishment) 6476 */ 6477 if (src->seqhi == 1 || 6478 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6479 src->seqhi = end + MAX(1, dst->max_win << dws); 6480 if (win > src->max_win) 6481 src->max_win = win; 6482 6483 } else { 6484 ack = ntohl(th->th_ack) - dst->seqdiff; 6485 if (src->seqdiff) { 6486 /* Modulate sequence numbers */ 6487 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6488 src->seqdiff), 0); 6489 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6490 *copyback = 1; 6491 } 6492 end = seq + pd->p_len; 6493 if (tcp_get_flags(th) & TH_SYN) 6494 end++; 6495 data_end = end; 6496 if (tcp_get_flags(th) & TH_FIN) 6497 end++; 6498 } 6499 6500 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6501 /* Let it pass through the ack skew check */ 6502 ack = dst->seqlo; 6503 } else if ((ack == 0 && 6504 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6505 /* broken tcp stacks do not set ack */ 6506 (dst->state < TCPS_SYN_SENT)) { 6507 /* 6508 * Many stacks (ours included) will set the ACK number in an 6509 * FIN|ACK if the SYN times out -- no sequence to ACK. 6510 */ 6511 ack = dst->seqlo; 6512 } 6513 6514 if (seq == end) { 6515 /* Ease sequencing restrictions on no data packets */ 6516 seq = src->seqlo; 6517 data_end = end = seq; 6518 } 6519 6520 ackskew = dst->seqlo - ack; 6521 6522 /* 6523 * Need to demodulate the sequence numbers in any TCP SACK options 6524 * (Selective ACK). We could optionally validate the SACK values 6525 * against the current ACK window, either forwards or backwards, but 6526 * I'm not confident that SACK has been implemented properly 6527 * everywhere. It wouldn't surprise me if several stacks accidentally 6528 * SACK too far backwards of previously ACKed data. There really aren't 6529 * any security implications of bad SACKing unless the target stack 6530 * doesn't validate the option length correctly. Someone trying to 6531 * spoof into a TCP connection won't bother blindly sending SACK 6532 * options anyway. 6533 */ 6534 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6535 if (pf_modulate_sack(pd, th, dst)) 6536 *copyback = 1; 6537 } 6538 6539 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6540 if (SEQ_GEQ(src->seqhi, data_end) && 6541 /* Last octet inside other's window space */ 6542 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6543 /* Retrans: not more than one window back */ 6544 (ackskew >= -MAXACKWINDOW) && 6545 /* Acking not more than one reassembled fragment backwards */ 6546 (ackskew <= (MAXACKWINDOW << sws)) && 6547 /* Acking not more than one window forward */ 6548 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6549 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 6550 /* Require an exact/+1 sequence match on resets when possible */ 6551 6552 if (dst->scrub || src->scrub) { 6553 if (pf_normalize_tcp_stateful(pd, reason, th, 6554 state, src, dst, copyback)) 6555 return (PF_DROP); 6556 } 6557 6558 /* update max window */ 6559 if (src->max_win < win) 6560 src->max_win = win; 6561 /* synchronize sequencing */ 6562 if (SEQ_GT(end, src->seqlo)) 6563 src->seqlo = end; 6564 /* slide the window of what the other end can send */ 6565 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6566 dst->seqhi = ack + MAX((win << sws), 1); 6567 6568 /* update states */ 6569 if (tcp_get_flags(th) & TH_SYN) 6570 if (src->state < TCPS_SYN_SENT) 6571 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6572 if (tcp_get_flags(th) & TH_FIN) 6573 if (src->state < TCPS_CLOSING) 6574 pf_set_protostate(state, psrc, TCPS_CLOSING); 6575 if (tcp_get_flags(th) & TH_ACK) { 6576 if (dst->state == TCPS_SYN_SENT) { 6577 pf_set_protostate(state, pdst, 6578 TCPS_ESTABLISHED); 6579 if (src->state == TCPS_ESTABLISHED && 6580 state->sns[PF_SN_LIMIT] != NULL && 6581 pf_src_connlimit(state)) { 6582 REASON_SET(reason, PFRES_SRCLIMIT); 6583 return (PF_DROP); 6584 } 6585 } else if (dst->state == TCPS_CLOSING) 6586 pf_set_protostate(state, pdst, 6587 TCPS_FIN_WAIT_2); 6588 } 6589 if (tcp_get_flags(th) & TH_RST) 6590 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6591 6592 /* update expire time */ 6593 state->expire = pf_get_uptime(); 6594 if (src->state >= TCPS_FIN_WAIT_2 && 6595 dst->state >= TCPS_FIN_WAIT_2) 6596 state->timeout = PFTM_TCP_CLOSED; 6597 else if (src->state >= TCPS_CLOSING && 6598 dst->state >= TCPS_CLOSING) 6599 state->timeout = PFTM_TCP_FIN_WAIT; 6600 else if (src->state < TCPS_ESTABLISHED || 6601 dst->state < TCPS_ESTABLISHED) 6602 state->timeout = PFTM_TCP_OPENING; 6603 else if (src->state >= TCPS_CLOSING || 6604 dst->state >= TCPS_CLOSING) 6605 state->timeout = PFTM_TCP_CLOSING; 6606 else 6607 state->timeout = PFTM_TCP_ESTABLISHED; 6608 6609 /* Fall through to PASS packet */ 6610 6611 } else if ((dst->state < TCPS_SYN_SENT || 6612 dst->state >= TCPS_FIN_WAIT_2 || 6613 src->state >= TCPS_FIN_WAIT_2) && 6614 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6615 /* Within a window forward of the originating packet */ 6616 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6617 /* Within a window backward of the originating packet */ 6618 6619 /* 6620 * This currently handles three situations: 6621 * 1) Stupid stacks will shotgun SYNs before their peer 6622 * replies. 6623 * 2) When PF catches an already established stream (the 6624 * firewall rebooted, the state table was flushed, routes 6625 * changed...) 6626 * 3) Packets get funky immediately after the connection 6627 * closes (this should catch Solaris spurious ACK|FINs 6628 * that web servers like to spew after a close) 6629 * 6630 * This must be a little more careful than the above code 6631 * since packet floods will also be caught here. We don't 6632 * update the TTL here to mitigate the damage of a packet 6633 * flood and so the same code can handle awkward establishment 6634 * and a loosened connection close. 6635 * In the establishment case, a correct peer response will 6636 * validate the connection, go through the normal state code 6637 * and keep updating the state TTL. 6638 */ 6639 6640 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6641 printf("pf: loose state match: "); 6642 pf_print_state(state); 6643 pf_print_flags(tcp_get_flags(th)); 6644 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6645 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6646 pd->p_len, ackskew, (unsigned long long)state->packets[0], 6647 (unsigned long long)state->packets[1], 6648 pd->dir == PF_IN ? "in" : "out", 6649 pd->dir == state->direction ? "fwd" : "rev"); 6650 } 6651 6652 if (dst->scrub || src->scrub) { 6653 if (pf_normalize_tcp_stateful(pd, reason, th, 6654 state, src, dst, copyback)) 6655 return (PF_DROP); 6656 } 6657 6658 /* update max window */ 6659 if (src->max_win < win) 6660 src->max_win = win; 6661 /* synchronize sequencing */ 6662 if (SEQ_GT(end, src->seqlo)) 6663 src->seqlo = end; 6664 /* slide the window of what the other end can send */ 6665 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6666 dst->seqhi = ack + MAX((win << sws), 1); 6667 6668 /* 6669 * Cannot set dst->seqhi here since this could be a shotgunned 6670 * SYN and not an already established connection. 6671 */ 6672 6673 if (tcp_get_flags(th) & TH_FIN) 6674 if (src->state < TCPS_CLOSING) 6675 pf_set_protostate(state, psrc, TCPS_CLOSING); 6676 if (tcp_get_flags(th) & TH_RST) 6677 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6678 6679 /* Fall through to PASS packet */ 6680 6681 } else { 6682 if (state->dst.state == TCPS_SYN_SENT && 6683 state->src.state == TCPS_SYN_SENT) { 6684 /* Send RST for state mismatches during handshake */ 6685 if (!(tcp_get_flags(th) & TH_RST)) 6686 pf_send_tcp(state->rule, pd->af, 6687 pd->dst, pd->src, th->th_dport, 6688 th->th_sport, ntohl(th->th_ack), 0, 6689 TH_RST, 0, 0, 6690 state->rule->return_ttl, M_SKIP_FIREWALL, 6691 0, 0, state->act.rtableid); 6692 src->seqlo = 0; 6693 src->seqhi = 1; 6694 src->max_win = 1; 6695 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6696 printf("pf: BAD state: "); 6697 pf_print_state(state); 6698 pf_print_flags(tcp_get_flags(th)); 6699 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6700 "pkts=%llu:%llu dir=%s,%s\n", 6701 seq, orig_seq, ack, pd->p_len, ackskew, 6702 (unsigned long long)state->packets[0], 6703 (unsigned long long)state->packets[1], 6704 pd->dir == PF_IN ? "in" : "out", 6705 pd->dir == state->direction ? "fwd" : "rev"); 6706 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6707 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6708 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6709 ' ': '2', 6710 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6711 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6712 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6713 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6714 } 6715 REASON_SET(reason, PFRES_BADSTATE); 6716 return (PF_DROP); 6717 } 6718 6719 return (PF_PASS); 6720 } 6721 6722 static int 6723 pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd, 6724 u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst, 6725 u_int8_t psrc, u_int8_t pdst) 6726 { 6727 struct tcphdr *th = &pd->hdr.tcp; 6728 6729 if (tcp_get_flags(th) & TH_SYN) 6730 if (src->state < TCPS_SYN_SENT) 6731 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6732 if (tcp_get_flags(th) & TH_FIN) 6733 if (src->state < TCPS_CLOSING) 6734 pf_set_protostate(state, psrc, TCPS_CLOSING); 6735 if (tcp_get_flags(th) & TH_ACK) { 6736 if (dst->state == TCPS_SYN_SENT) { 6737 pf_set_protostate(state, pdst, TCPS_ESTABLISHED); 6738 if (src->state == TCPS_ESTABLISHED && 6739 state->sns[PF_SN_LIMIT] != NULL && 6740 pf_src_connlimit(state)) { 6741 REASON_SET(reason, PFRES_SRCLIMIT); 6742 return (PF_DROP); 6743 } 6744 } else if (dst->state == TCPS_CLOSING) { 6745 pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2); 6746 } else if (src->state == TCPS_SYN_SENT && 6747 dst->state < TCPS_SYN_SENT) { 6748 /* 6749 * Handle a special sloppy case where we only see one 6750 * half of the connection. If there is a ACK after 6751 * the initial SYN without ever seeing a packet from 6752 * the destination, set the connection to established. 6753 */ 6754 pf_set_protostate(state, PF_PEER_BOTH, 6755 TCPS_ESTABLISHED); 6756 dst->state = src->state = TCPS_ESTABLISHED; 6757 if (state->sns[PF_SN_LIMIT] != NULL && 6758 pf_src_connlimit(state)) { 6759 REASON_SET(reason, PFRES_SRCLIMIT); 6760 return (PF_DROP); 6761 } 6762 } else if (src->state == TCPS_CLOSING && 6763 dst->state == TCPS_ESTABLISHED && 6764 dst->seqlo == 0) { 6765 /* 6766 * Handle the closing of half connections where we 6767 * don't see the full bidirectional FIN/ACK+ACK 6768 * handshake. 6769 */ 6770 pf_set_protostate(state, pdst, TCPS_CLOSING); 6771 } 6772 } 6773 if (tcp_get_flags(th) & TH_RST) 6774 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6775 6776 /* update expire time */ 6777 state->expire = pf_get_uptime(); 6778 if (src->state >= TCPS_FIN_WAIT_2 && 6779 dst->state >= TCPS_FIN_WAIT_2) 6780 state->timeout = PFTM_TCP_CLOSED; 6781 else if (src->state >= TCPS_CLOSING && 6782 dst->state >= TCPS_CLOSING) 6783 state->timeout = PFTM_TCP_FIN_WAIT; 6784 else if (src->state < TCPS_ESTABLISHED || 6785 dst->state < TCPS_ESTABLISHED) 6786 state->timeout = PFTM_TCP_OPENING; 6787 else if (src->state >= TCPS_CLOSING || 6788 dst->state >= TCPS_CLOSING) 6789 state->timeout = PFTM_TCP_CLOSING; 6790 else 6791 state->timeout = PFTM_TCP_ESTABLISHED; 6792 6793 return (PF_PASS); 6794 } 6795 6796 static int 6797 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason) 6798 { 6799 struct pf_state_key *sk = state->key[pd->didx]; 6800 struct tcphdr *th = &pd->hdr.tcp; 6801 6802 if (state->src.state == PF_TCPS_PROXY_SRC) { 6803 if (pd->dir != state->direction) { 6804 REASON_SET(reason, PFRES_SYNPROXY); 6805 return (PF_SYNPROXY_DROP); 6806 } 6807 if (tcp_get_flags(th) & TH_SYN) { 6808 if (ntohl(th->th_seq) != state->src.seqlo) { 6809 REASON_SET(reason, PFRES_SYNPROXY); 6810 return (PF_DROP); 6811 } 6812 pf_send_tcp(state->rule, pd->af, pd->dst, 6813 pd->src, th->th_dport, th->th_sport, 6814 state->src.seqhi, ntohl(th->th_seq) + 1, 6815 TH_SYN|TH_ACK, 0, state->src.mss, 0, 6816 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6817 REASON_SET(reason, PFRES_SYNPROXY); 6818 return (PF_SYNPROXY_DROP); 6819 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6820 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6821 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6822 REASON_SET(reason, PFRES_SYNPROXY); 6823 return (PF_DROP); 6824 } else if (state->sns[PF_SN_LIMIT] != NULL && 6825 pf_src_connlimit(state)) { 6826 REASON_SET(reason, PFRES_SRCLIMIT); 6827 return (PF_DROP); 6828 } else 6829 pf_set_protostate(state, PF_PEER_SRC, 6830 PF_TCPS_PROXY_DST); 6831 } 6832 if (state->src.state == PF_TCPS_PROXY_DST) { 6833 if (pd->dir == state->direction) { 6834 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 6835 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6836 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6837 REASON_SET(reason, PFRES_SYNPROXY); 6838 return (PF_DROP); 6839 } 6840 state->src.max_win = MAX(ntohs(th->th_win), 1); 6841 if (state->dst.seqhi == 1) 6842 state->dst.seqhi = arc4random(); 6843 pf_send_tcp(state->rule, pd->af, 6844 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6845 sk->port[pd->sidx], sk->port[pd->didx], 6846 state->dst.seqhi, 0, TH_SYN, 0, 6847 state->src.mss, 0, 6848 state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6849 state->tag, 0, state->act.rtableid); 6850 REASON_SET(reason, PFRES_SYNPROXY); 6851 return (PF_SYNPROXY_DROP); 6852 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 6853 (TH_SYN|TH_ACK)) || 6854 (ntohl(th->th_ack) != state->dst.seqhi + 1)) { 6855 REASON_SET(reason, PFRES_SYNPROXY); 6856 return (PF_DROP); 6857 } else { 6858 state->dst.max_win = MAX(ntohs(th->th_win), 1); 6859 state->dst.seqlo = ntohl(th->th_seq); 6860 pf_send_tcp(state->rule, pd->af, pd->dst, 6861 pd->src, th->th_dport, th->th_sport, 6862 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6863 TH_ACK, state->src.max_win, 0, 0, 0, 6864 state->tag, 0, state->act.rtableid); 6865 pf_send_tcp(state->rule, pd->af, 6866 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6867 sk->port[pd->sidx], sk->port[pd->didx], 6868 state->src.seqhi + 1, state->src.seqlo + 1, 6869 TH_ACK, state->dst.max_win, 0, 0, 6870 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6871 state->src.seqdiff = state->dst.seqhi - 6872 state->src.seqlo; 6873 state->dst.seqdiff = state->src.seqhi - 6874 state->dst.seqlo; 6875 state->src.seqhi = state->src.seqlo + 6876 state->dst.max_win; 6877 state->dst.seqhi = state->dst.seqlo + 6878 state->src.max_win; 6879 state->src.wscale = state->dst.wscale = 0; 6880 pf_set_protostate(state, PF_PEER_BOTH, 6881 TCPS_ESTABLISHED); 6882 REASON_SET(reason, PFRES_SYNPROXY); 6883 return (PF_SYNPROXY_DROP); 6884 } 6885 } 6886 6887 return (PF_PASS); 6888 } 6889 6890 static int 6891 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6892 { 6893 struct pf_state_key_cmp key; 6894 int copyback = 0; 6895 struct pf_state_peer *src, *dst; 6896 uint8_t psrc, pdst; 6897 int action = PF_PASS; 6898 6899 bzero(&key, sizeof(key)); 6900 key.af = pd->af; 6901 key.proto = pd->virtual_proto; 6902 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 6903 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 6904 key.port[pd->sidx] = pd->osport; 6905 key.port[pd->didx] = pd->odport; 6906 6907 STATE_LOOKUP(&key, *state, pd); 6908 6909 if (pd->dir == (*state)->direction) { 6910 if (PF_REVERSED_KEY(*state, pd->af)) { 6911 src = &(*state)->dst; 6912 dst = &(*state)->src; 6913 psrc = PF_PEER_DST; 6914 pdst = PF_PEER_SRC; 6915 } else { 6916 src = &(*state)->src; 6917 dst = &(*state)->dst; 6918 psrc = PF_PEER_SRC; 6919 pdst = PF_PEER_DST; 6920 } 6921 } else { 6922 if (PF_REVERSED_KEY(*state, pd->af)) { 6923 src = &(*state)->src; 6924 dst = &(*state)->dst; 6925 psrc = PF_PEER_SRC; 6926 pdst = PF_PEER_DST; 6927 } else { 6928 src = &(*state)->dst; 6929 dst = &(*state)->src; 6930 psrc = PF_PEER_DST; 6931 pdst = PF_PEER_SRC; 6932 } 6933 } 6934 6935 switch (pd->virtual_proto) { 6936 case IPPROTO_TCP: { 6937 struct tcphdr *th = &pd->hdr.tcp; 6938 6939 if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS) 6940 return (action); 6941 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 6942 (*state)->dst.state >= TCPS_FIN_WAIT_2 && 6943 (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN) || 6944 ((tcp_get_flags(th) & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6945 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6946 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6947 printf("pf: state reuse "); 6948 pf_print_state(*state); 6949 pf_print_flags(tcp_get_flags(th)); 6950 printf("\n"); 6951 } 6952 /* XXX make sure it's the same direction ?? */ 6953 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6954 pf_remove_state(*state); 6955 *state = NULL; 6956 return (PF_DROP); 6957 } 6958 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6959 if (pf_tcp_track_sloppy(*state, pd, reason, src, dst, 6960 psrc, pdst) == PF_DROP) 6961 return (PF_DROP); 6962 } else { 6963 int ret; 6964 6965 ret = pf_tcp_track_full(*state, pd, reason, 6966 ©back, src, dst, psrc, pdst); 6967 if (ret == PF_DROP) 6968 return (PF_DROP); 6969 } 6970 break; 6971 } 6972 case IPPROTO_UDP: 6973 /* update states */ 6974 if (src->state < PFUDPS_SINGLE) 6975 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6976 if (dst->state == PFUDPS_SINGLE) 6977 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6978 6979 /* update expire time */ 6980 (*state)->expire = pf_get_uptime(); 6981 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6982 (*state)->timeout = PFTM_UDP_MULTIPLE; 6983 else 6984 (*state)->timeout = PFTM_UDP_SINGLE; 6985 break; 6986 case IPPROTO_SCTP: 6987 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6988 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6989 pd->sctp_flags & PFDESC_SCTP_INIT) { 6990 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6991 pf_remove_state(*state); 6992 *state = NULL; 6993 return (PF_DROP); 6994 } 6995 6996 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 6997 return (PF_DROP); 6998 6999 /* Track state. */ 7000 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7001 if (src->state < SCTP_COOKIE_WAIT) { 7002 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7003 (*state)->timeout = PFTM_SCTP_OPENING; 7004 } 7005 } 7006 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7007 MPASS(dst->scrub != NULL); 7008 if (dst->scrub->pfss_v_tag == 0) 7009 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7010 } 7011 7012 /* 7013 * Bind to the correct interface if we're if-bound. For multihomed 7014 * extra associations we don't know which interface that will be until 7015 * here, so we've inserted the state on V_pf_all. Fix that now. 7016 */ 7017 if ((*state)->kif == V_pfi_all && 7018 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7019 (*state)->kif = pd->kif; 7020 7021 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7022 if (src->state < SCTP_ESTABLISHED) { 7023 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7024 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7025 } 7026 } 7027 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7028 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7029 if (src->state < SCTP_SHUTDOWN_PENDING) { 7030 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7031 (*state)->timeout = PFTM_SCTP_CLOSING; 7032 } 7033 } 7034 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7035 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7036 (*state)->timeout = PFTM_SCTP_CLOSED; 7037 } 7038 7039 (*state)->expire = pf_get_uptime(); 7040 break; 7041 default: 7042 /* update states */ 7043 if (src->state < PFOTHERS_SINGLE) 7044 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7045 if (dst->state == PFOTHERS_SINGLE) 7046 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7047 7048 /* update expire time */ 7049 (*state)->expire = pf_get_uptime(); 7050 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7051 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7052 else 7053 (*state)->timeout = PFTM_OTHER_SINGLE; 7054 break; 7055 } 7056 7057 /* translate source/destination address, if necessary */ 7058 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7059 struct pf_state_key *nk; 7060 int afto, sidx, didx; 7061 7062 if (PF_REVERSED_KEY(*state, pd->af)) 7063 nk = (*state)->key[pd->sidx]; 7064 else 7065 nk = (*state)->key[pd->didx]; 7066 7067 afto = pd->af != nk->af; 7068 7069 if (afto && (*state)->direction == PF_IN) { 7070 sidx = pd->didx; 7071 didx = pd->sidx; 7072 } else { 7073 sidx = pd->sidx; 7074 didx = pd->didx; 7075 } 7076 7077 if (afto) { 7078 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7079 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7080 pd->naf = nk->af; 7081 action = PF_AFRT; 7082 } 7083 7084 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7085 nk->port[sidx] != pd->osport) 7086 pf_change_ap(pd, pd->src, pd->sport, 7087 &nk->addr[sidx], nk->port[sidx]); 7088 7089 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7090 nk->port[didx] != pd->odport) 7091 pf_change_ap(pd, pd->dst, pd->dport, 7092 &nk->addr[didx], nk->port[didx]); 7093 7094 copyback = 1; 7095 } 7096 7097 if (copyback && pd->hdrlen > 0) 7098 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 7099 7100 return (action); 7101 } 7102 7103 static int 7104 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7105 u_short *reason) 7106 { 7107 struct pf_state_peer *src; 7108 if (pd->dir == state->direction) { 7109 if (PF_REVERSED_KEY(state, pd->af)) 7110 src = &state->dst; 7111 else 7112 src = &state->src; 7113 } else { 7114 if (PF_REVERSED_KEY(state, pd->af)) 7115 src = &state->src; 7116 else 7117 src = &state->dst; 7118 } 7119 7120 if (src->scrub != NULL) { 7121 if (src->scrub->pfss_v_tag == 0) 7122 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7123 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7124 return (PF_DROP); 7125 } 7126 7127 return (PF_PASS); 7128 } 7129 7130 static void 7131 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7132 { 7133 struct pf_sctp_endpoint key; 7134 struct pf_sctp_endpoint *ep; 7135 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7136 struct pf_sctp_source *i, *tmp; 7137 7138 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7139 return; 7140 7141 PF_SCTP_ENDPOINTS_LOCK(); 7142 7143 key.v_tag = s->dst.scrub->pfss_v_tag; 7144 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7145 if (ep != NULL) { 7146 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7147 if (pf_addr_cmp(&i->addr, 7148 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7149 s->key[PF_SK_WIRE]->af) == 0) { 7150 SDT_PROBE3(pf, sctp, multihome, remove, 7151 key.v_tag, s, i); 7152 TAILQ_REMOVE(&ep->sources, i, entry); 7153 free(i, M_PFTEMP); 7154 break; 7155 } 7156 } 7157 7158 if (TAILQ_EMPTY(&ep->sources)) { 7159 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7160 free(ep, M_PFTEMP); 7161 } 7162 } 7163 7164 /* Other direction. */ 7165 key.v_tag = s->src.scrub->pfss_v_tag; 7166 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7167 if (ep != NULL) { 7168 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7169 if (pf_addr_cmp(&i->addr, 7170 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7171 s->key[PF_SK_WIRE]->af) == 0) { 7172 SDT_PROBE3(pf, sctp, multihome, remove, 7173 key.v_tag, s, i); 7174 TAILQ_REMOVE(&ep->sources, i, entry); 7175 free(i, M_PFTEMP); 7176 break; 7177 } 7178 } 7179 7180 if (TAILQ_EMPTY(&ep->sources)) { 7181 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7182 free(ep, M_PFTEMP); 7183 } 7184 } 7185 7186 PF_SCTP_ENDPOINTS_UNLOCK(); 7187 } 7188 7189 static void 7190 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7191 { 7192 struct pf_sctp_endpoint key = { 7193 .v_tag = v_tag, 7194 }; 7195 struct pf_sctp_source *i; 7196 struct pf_sctp_endpoint *ep; 7197 7198 PF_SCTP_ENDPOINTS_LOCK(); 7199 7200 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7201 if (ep == NULL) { 7202 ep = malloc(sizeof(struct pf_sctp_endpoint), 7203 M_PFTEMP, M_NOWAIT); 7204 if (ep == NULL) { 7205 PF_SCTP_ENDPOINTS_UNLOCK(); 7206 return; 7207 } 7208 7209 ep->v_tag = v_tag; 7210 TAILQ_INIT(&ep->sources); 7211 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7212 } 7213 7214 /* Avoid inserting duplicates. */ 7215 TAILQ_FOREACH(i, &ep->sources, entry) { 7216 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7217 PF_SCTP_ENDPOINTS_UNLOCK(); 7218 return; 7219 } 7220 } 7221 7222 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7223 if (i == NULL) { 7224 PF_SCTP_ENDPOINTS_UNLOCK(); 7225 return; 7226 } 7227 7228 i->af = pd->af; 7229 memcpy(&i->addr, a, sizeof(*a)); 7230 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7231 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7232 7233 PF_SCTP_ENDPOINTS_UNLOCK(); 7234 } 7235 7236 static void 7237 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7238 struct pf_kstate *s, int action) 7239 { 7240 struct pf_sctp_multihome_job *j, *tmp; 7241 struct pf_sctp_source *i; 7242 int ret __unused; 7243 struct pf_kstate *sm = NULL; 7244 struct pf_krule *ra = NULL; 7245 struct pf_krule *r = &V_pf_default_rule; 7246 struct pf_kruleset *rs = NULL; 7247 u_short reason; 7248 bool do_extra = true; 7249 7250 PF_RULES_RLOCK_TRACKER; 7251 7252 again: 7253 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7254 if (s == NULL || action != PF_PASS) 7255 goto free; 7256 7257 /* Confirm we don't recurse here. */ 7258 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7259 7260 switch (j->op) { 7261 case SCTP_ADD_IP_ADDRESS: { 7262 uint32_t v_tag = pd->sctp_initiate_tag; 7263 7264 if (v_tag == 0) { 7265 if (s->direction == pd->dir) 7266 v_tag = s->src.scrub->pfss_v_tag; 7267 else 7268 v_tag = s->dst.scrub->pfss_v_tag; 7269 } 7270 7271 /* 7272 * Avoid duplicating states. We'll already have 7273 * created a state based on the source address of 7274 * the packet, but SCTP endpoints may also list this 7275 * address again in the INIT(_ACK) parameters. 7276 */ 7277 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7278 break; 7279 } 7280 7281 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7282 PF_RULES_RLOCK(); 7283 sm = NULL; 7284 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7285 j->pd.related_rule = s->rule; 7286 } 7287 ret = pf_test_rule(&r, &sm, 7288 &j->pd, &ra, &rs, &reason, NULL); 7289 PF_RULES_RUNLOCK(); 7290 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7291 if (ret != PF_DROP && sm != NULL) { 7292 /* Inherit v_tag values. */ 7293 if (sm->direction == s->direction) { 7294 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7295 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7296 } else { 7297 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7298 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7299 } 7300 PF_STATE_UNLOCK(sm); 7301 } else { 7302 /* If we try duplicate inserts? */ 7303 break; 7304 } 7305 7306 /* Only add the address if we've actually allowed the state. */ 7307 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7308 7309 if (! do_extra) { 7310 break; 7311 } 7312 /* 7313 * We need to do this for each of our source addresses. 7314 * Find those based on the verification tag. 7315 */ 7316 struct pf_sctp_endpoint key = { 7317 .v_tag = pd->hdr.sctp.v_tag, 7318 }; 7319 struct pf_sctp_endpoint *ep; 7320 7321 PF_SCTP_ENDPOINTS_LOCK(); 7322 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7323 if (ep == NULL) { 7324 PF_SCTP_ENDPOINTS_UNLOCK(); 7325 break; 7326 } 7327 MPASS(ep != NULL); 7328 7329 TAILQ_FOREACH(i, &ep->sources, entry) { 7330 struct pf_sctp_multihome_job *nj; 7331 7332 /* SCTP can intermingle IPv4 and IPv6. */ 7333 if (i->af != pd->af) 7334 continue; 7335 7336 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7337 if (! nj) { 7338 continue; 7339 } 7340 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7341 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7342 nj->pd.src = &nj->src; 7343 // New destination address! 7344 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7345 nj->pd.dst = &nj->dst; 7346 nj->pd.m = j->pd.m; 7347 nj->op = j->op; 7348 7349 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7350 } 7351 PF_SCTP_ENDPOINTS_UNLOCK(); 7352 7353 break; 7354 } 7355 case SCTP_DEL_IP_ADDRESS: { 7356 struct pf_state_key_cmp key; 7357 uint8_t psrc; 7358 7359 bzero(&key, sizeof(key)); 7360 key.af = j->pd.af; 7361 key.proto = IPPROTO_SCTP; 7362 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7363 PF_ACPY(&key.addr[0], j->pd.src, key.af); 7364 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 7365 key.port[0] = j->pd.hdr.sctp.src_port; 7366 key.port[1] = j->pd.hdr.sctp.dest_port; 7367 } else { /* stack side, reverse */ 7368 PF_ACPY(&key.addr[1], j->pd.src, key.af); 7369 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 7370 key.port[1] = j->pd.hdr.sctp.src_port; 7371 key.port[0] = j->pd.hdr.sctp.dest_port; 7372 } 7373 7374 sm = pf_find_state(kif, &key, j->pd.dir); 7375 if (sm != NULL) { 7376 PF_STATE_LOCK_ASSERT(sm); 7377 if (j->pd.dir == sm->direction) { 7378 psrc = PF_PEER_SRC; 7379 } else { 7380 psrc = PF_PEER_DST; 7381 } 7382 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7383 sm->timeout = PFTM_SCTP_CLOSING; 7384 PF_STATE_UNLOCK(sm); 7385 } 7386 break; 7387 default: 7388 panic("Unknown op %#x", j->op); 7389 } 7390 } 7391 7392 free: 7393 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7394 free(j, M_PFTEMP); 7395 } 7396 7397 /* We may have inserted extra work while processing the list. */ 7398 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7399 do_extra = false; 7400 goto again; 7401 } 7402 } 7403 7404 static int 7405 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7406 { 7407 int off = 0; 7408 struct pf_sctp_multihome_job *job; 7409 7410 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7411 7412 while (off < len) { 7413 struct sctp_paramhdr h; 7414 7415 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 7416 pd->af)) 7417 return (PF_DROP); 7418 7419 /* Parameters are at least 4 bytes. */ 7420 if (ntohs(h.param_length) < 4) 7421 return (PF_DROP); 7422 7423 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7424 ntohs(h.param_length)); 7425 7426 switch (ntohs(h.param_type)) { 7427 case SCTP_IPV4_ADDRESS: { 7428 struct in_addr t; 7429 7430 if (ntohs(h.param_length) != 7431 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7432 return (PF_DROP); 7433 7434 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7435 NULL, NULL, pd->af)) 7436 return (PF_DROP); 7437 7438 if (in_nullhost(t)) 7439 t.s_addr = pd->src->v4.s_addr; 7440 7441 /* 7442 * We hold the state lock (idhash) here, which means 7443 * that we can't acquire the keyhash, or we'll get a 7444 * LOR (and potentially double-lock things too). We also 7445 * can't release the state lock here, so instead we'll 7446 * enqueue this for async handling. 7447 * There's a relatively small race here, in that a 7448 * packet using the new addresses could arrive already, 7449 * but that's just though luck for it. 7450 */ 7451 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7452 if (! job) 7453 return (PF_DROP); 7454 7455 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7456 7457 memcpy(&job->pd, pd, sizeof(*pd)); 7458 7459 // New source address! 7460 memcpy(&job->src, &t, sizeof(t)); 7461 job->pd.src = &job->src; 7462 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7463 job->pd.dst = &job->dst; 7464 job->pd.m = pd->m; 7465 job->op = op; 7466 7467 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7468 break; 7469 } 7470 #ifdef INET6 7471 case SCTP_IPV6_ADDRESS: { 7472 struct in6_addr t; 7473 7474 if (ntohs(h.param_length) != 7475 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7476 return (PF_DROP); 7477 7478 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7479 NULL, NULL, pd->af)) 7480 return (PF_DROP); 7481 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7482 break; 7483 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7484 memcpy(&t, &pd->src->v6, sizeof(t)); 7485 7486 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7487 if (! job) 7488 return (PF_DROP); 7489 7490 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7491 7492 memcpy(&job->pd, pd, sizeof(*pd)); 7493 memcpy(&job->src, &t, sizeof(t)); 7494 job->pd.src = &job->src; 7495 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7496 job->pd.dst = &job->dst; 7497 job->pd.m = pd->m; 7498 job->op = op; 7499 7500 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7501 break; 7502 } 7503 #endif /* INET6 */ 7504 case SCTP_ADD_IP_ADDRESS: { 7505 int ret; 7506 struct sctp_asconf_paramhdr ah; 7507 7508 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7509 NULL, NULL, pd->af)) 7510 return (PF_DROP); 7511 7512 ret = pf_multihome_scan(start + off + sizeof(ah), 7513 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7514 SCTP_ADD_IP_ADDRESS); 7515 if (ret != PF_PASS) 7516 return (ret); 7517 break; 7518 } 7519 case SCTP_DEL_IP_ADDRESS: { 7520 int ret; 7521 struct sctp_asconf_paramhdr ah; 7522 7523 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7524 NULL, NULL, pd->af)) 7525 return (PF_DROP); 7526 ret = pf_multihome_scan(start + off + sizeof(ah), 7527 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7528 SCTP_DEL_IP_ADDRESS); 7529 if (ret != PF_PASS) 7530 return (ret); 7531 break; 7532 } 7533 default: 7534 break; 7535 } 7536 7537 off += roundup(ntohs(h.param_length), 4); 7538 } 7539 7540 return (PF_PASS); 7541 } 7542 7543 int 7544 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7545 { 7546 start += sizeof(struct sctp_init_chunk); 7547 len -= sizeof(struct sctp_init_chunk); 7548 7549 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7550 } 7551 7552 int 7553 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7554 { 7555 start += sizeof(struct sctp_asconf_chunk); 7556 len -= sizeof(struct sctp_asconf_chunk); 7557 7558 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7559 } 7560 7561 int 7562 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7563 struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir, 7564 int *iidx, int multi, int inner) 7565 { 7566 int direction = pd->dir; 7567 7568 key->af = pd->af; 7569 key->proto = pd->proto; 7570 if (icmp_dir == PF_IN) { 7571 *iidx = pd->sidx; 7572 key->port[pd->sidx] = icmpid; 7573 key->port[pd->didx] = type; 7574 } else { 7575 *iidx = pd->didx; 7576 key->port[pd->sidx] = type; 7577 key->port[pd->didx] = icmpid; 7578 } 7579 if (pf_state_key_addr_setup(pd, key, multi)) 7580 return (PF_DROP); 7581 7582 STATE_LOOKUP(key, *state, pd); 7583 7584 if ((*state)->state_flags & PFSTATE_SLOPPY) 7585 return (-1); 7586 7587 /* Is this ICMP message flowing in right direction? */ 7588 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7589 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7590 PF_IN : PF_OUT; 7591 else 7592 direction = (*state)->direction; 7593 if ((*state)->rule->type && 7594 (((!inner && direction == pd->dir) || 7595 (inner && direction != pd->dir)) ? 7596 PF_IN : PF_OUT) != icmp_dir) { 7597 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7598 printf("pf: icmp type %d in wrong direction (%d): ", 7599 ntohs(type), icmp_dir); 7600 pf_print_state(*state); 7601 printf("\n"); 7602 } 7603 PF_STATE_UNLOCK(*state); 7604 *state = NULL; 7605 return (PF_DROP); 7606 } 7607 return (-1); 7608 } 7609 7610 static int 7611 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7612 u_short *reason) 7613 { 7614 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7615 u_int16_t *icmpsum, virtual_id, virtual_type; 7616 u_int8_t icmptype, icmpcode; 7617 int icmp_dir, iidx, ret; 7618 struct pf_state_key_cmp key; 7619 #ifdef INET 7620 u_int16_t icmpid; 7621 #endif /* INET*/ 7622 7623 MPASS(*state == NULL); 7624 7625 bzero(&key, sizeof(key)); 7626 switch (pd->proto) { 7627 #ifdef INET 7628 case IPPROTO_ICMP: 7629 icmptype = pd->hdr.icmp.icmp_type; 7630 icmpcode = pd->hdr.icmp.icmp_code; 7631 icmpid = pd->hdr.icmp.icmp_id; 7632 icmpsum = &pd->hdr.icmp.icmp_cksum; 7633 break; 7634 #endif /* INET */ 7635 #ifdef INET6 7636 case IPPROTO_ICMPV6: 7637 icmptype = pd->hdr.icmp6.icmp6_type; 7638 icmpcode = pd->hdr.icmp6.icmp6_code; 7639 #ifdef INET 7640 icmpid = pd->hdr.icmp6.icmp6_id; 7641 #endif /* INET */ 7642 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7643 break; 7644 #endif /* INET6 */ 7645 default: 7646 panic("unhandled proto %d", pd->proto); 7647 } 7648 7649 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 7650 &virtual_type) == 0) { 7651 /* 7652 * ICMP query/reply message not related to a TCP/UDP/SCTP 7653 * packet. Search for an ICMP state. 7654 */ 7655 ret = pf_icmp_state_lookup(&key, pd, state, virtual_id, 7656 virtual_type, icmp_dir, &iidx, 0, 0); 7657 /* IPv6? try matching a multicast address */ 7658 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) { 7659 MPASS(*state == NULL); 7660 ret = pf_icmp_state_lookup(&key, pd, state, 7661 virtual_id, virtual_type, 7662 icmp_dir, &iidx, 1, 0); 7663 } 7664 if (ret >= 0) { 7665 MPASS(*state == NULL); 7666 return (ret); 7667 } 7668 7669 (*state)->expire = pf_get_uptime(); 7670 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7671 7672 /* translate source/destination address, if necessary */ 7673 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7674 struct pf_state_key *nk; 7675 int afto, sidx, didx; 7676 7677 if (PF_REVERSED_KEY(*state, pd->af)) 7678 nk = (*state)->key[pd->sidx]; 7679 else 7680 nk = (*state)->key[pd->didx]; 7681 7682 afto = pd->af != nk->af; 7683 7684 if (afto && (*state)->direction == PF_IN) { 7685 sidx = pd->didx; 7686 didx = pd->sidx; 7687 iidx = !iidx; 7688 } else { 7689 sidx = pd->sidx; 7690 didx = pd->didx; 7691 } 7692 7693 switch (pd->af) { 7694 #ifdef INET 7695 case AF_INET: 7696 #ifdef INET6 7697 if (afto) { 7698 if (pf_translate_icmp_af(AF_INET6, 7699 &pd->hdr.icmp)) 7700 return (PF_DROP); 7701 pd->proto = IPPROTO_ICMPV6; 7702 } 7703 #endif /* INET6 */ 7704 if (!afto && 7705 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 7706 pf_change_a(&saddr->v4.s_addr, 7707 pd->ip_sum, 7708 nk->addr[sidx].v4.s_addr, 7709 0); 7710 7711 if (!afto && PF_ANEQ(pd->dst, 7712 &nk->addr[didx], AF_INET)) 7713 pf_change_a(&daddr->v4.s_addr, 7714 pd->ip_sum, 7715 nk->addr[didx].v4.s_addr, 0); 7716 7717 if (nk->port[iidx] != 7718 pd->hdr.icmp.icmp_id) { 7719 pd->hdr.icmp.icmp_cksum = 7720 pf_cksum_fixup( 7721 pd->hdr.icmp.icmp_cksum, icmpid, 7722 nk->port[iidx], 0); 7723 pd->hdr.icmp.icmp_id = 7724 nk->port[iidx]; 7725 } 7726 7727 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7728 (caddr_t )&pd->hdr.icmp); 7729 break; 7730 #endif /* INET */ 7731 #ifdef INET6 7732 case AF_INET6: 7733 #ifdef INET 7734 if (afto) { 7735 if (pf_translate_icmp_af(AF_INET, 7736 &pd->hdr.icmp6)) 7737 return (PF_DROP); 7738 pd->proto = IPPROTO_ICMP; 7739 } 7740 #endif /* INET */ 7741 if (!afto && 7742 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 7743 pf_change_a6(saddr, 7744 &pd->hdr.icmp6.icmp6_cksum, 7745 &nk->addr[sidx], 0); 7746 7747 if (!afto && PF_ANEQ(pd->dst, 7748 &nk->addr[didx], AF_INET6)) 7749 pf_change_a6(daddr, 7750 &pd->hdr.icmp6.icmp6_cksum, 7751 &nk->addr[didx], 0); 7752 7753 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 7754 pd->hdr.icmp6.icmp6_id = 7755 nk->port[iidx]; 7756 7757 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7758 (caddr_t )&pd->hdr.icmp6); 7759 break; 7760 #endif /* INET6 */ 7761 } 7762 if (afto) { 7763 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7764 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7765 pd->naf = nk->af; 7766 return (PF_AFRT); 7767 } 7768 } 7769 return (PF_PASS); 7770 7771 } else { 7772 /* 7773 * ICMP error message in response to a TCP/UDP packet. 7774 * Extract the inner TCP/UDP header and search for that state. 7775 */ 7776 7777 struct pf_pdesc pd2; 7778 bzero(&pd2, sizeof pd2); 7779 #ifdef INET 7780 struct ip h2; 7781 #endif /* INET */ 7782 #ifdef INET6 7783 struct ip6_hdr h2_6; 7784 #endif /* INET6 */ 7785 int ipoff2 = 0; 7786 7787 pd2.af = pd->af; 7788 pd2.dir = pd->dir; 7789 /* Payload packet is from the opposite direction. */ 7790 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7791 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7792 pd2.m = pd->m; 7793 pd2.kif = pd->kif; 7794 switch (pd->af) { 7795 #ifdef INET 7796 case AF_INET: 7797 /* offset of h2 in mbuf chain */ 7798 ipoff2 = pd->off + ICMP_MINLEN; 7799 7800 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7801 NULL, reason, pd2.af)) { 7802 DPFPRINTF(PF_DEBUG_MISC, 7803 ("pf: ICMP error message too short " 7804 "(ip)\n")); 7805 return (PF_DROP); 7806 } 7807 /* 7808 * ICMP error messages don't refer to non-first 7809 * fragments 7810 */ 7811 if (h2.ip_off & htons(IP_OFFMASK)) { 7812 REASON_SET(reason, PFRES_FRAG); 7813 return (PF_DROP); 7814 } 7815 7816 /* offset of protocol header that follows h2 */ 7817 pd2.off = ipoff2 + (h2.ip_hl << 2); 7818 7819 pd2.proto = h2.ip_p; 7820 pd2.tot_len = ntohs(h2.ip_len); 7821 pd2.src = (struct pf_addr *)&h2.ip_src; 7822 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7823 pd2.ip_sum = &h2.ip_sum; 7824 break; 7825 #endif /* INET */ 7826 #ifdef INET6 7827 case AF_INET6: 7828 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7829 7830 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7831 NULL, reason, pd2.af)) { 7832 DPFPRINTF(PF_DEBUG_MISC, 7833 ("pf: ICMP error message too short " 7834 "(ip6)\n")); 7835 return (PF_DROP); 7836 } 7837 pd2.off = ipoff2; 7838 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 7839 return (PF_DROP); 7840 7841 pd2.tot_len = ntohs(h2_6.ip6_plen) + 7842 sizeof(struct ip6_hdr); 7843 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7844 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7845 pd2.ip_sum = NULL; 7846 break; 7847 #endif /* INET6 */ 7848 default: 7849 unhandled_af(pd->af); 7850 } 7851 7852 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7853 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7854 printf("pf: BAD ICMP %d:%d outer dst: ", 7855 icmptype, icmpcode); 7856 pf_print_host(pd->src, 0, pd->af); 7857 printf(" -> "); 7858 pf_print_host(pd->dst, 0, pd->af); 7859 printf(" inner src: "); 7860 pf_print_host(pd2.src, 0, pd2.af); 7861 printf(" -> "); 7862 pf_print_host(pd2.dst, 0, pd2.af); 7863 printf("\n"); 7864 } 7865 REASON_SET(reason, PFRES_BADSTATE); 7866 return (PF_DROP); 7867 } 7868 7869 switch (pd2.proto) { 7870 case IPPROTO_TCP: { 7871 struct tcphdr *th = &pd2.hdr.tcp; 7872 u_int32_t seq; 7873 struct pf_state_peer *src, *dst; 7874 u_int8_t dws; 7875 int copyback = 0; 7876 7877 /* 7878 * Only the first 8 bytes of the TCP header can be 7879 * expected. Don't access any TCP header fields after 7880 * th_seq, an ackskew test is not possible. 7881 */ 7882 if (!pf_pull_hdr(pd->m, pd2.off, th, 8, NULL, reason, 7883 pd2.af)) { 7884 DPFPRINTF(PF_DEBUG_MISC, 7885 ("pf: ICMP error message too short " 7886 "(tcp)\n")); 7887 return (PF_DROP); 7888 } 7889 pd2.pcksum = &pd2.hdr.tcp.th_sum; 7890 7891 key.af = pd2.af; 7892 key.proto = IPPROTO_TCP; 7893 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7894 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7895 key.port[pd2.sidx] = th->th_sport; 7896 key.port[pd2.didx] = th->th_dport; 7897 7898 STATE_LOOKUP(&key, *state, pd); 7899 7900 if (pd->dir == (*state)->direction) { 7901 if (PF_REVERSED_KEY(*state, pd->af)) { 7902 src = &(*state)->src; 7903 dst = &(*state)->dst; 7904 } else { 7905 src = &(*state)->dst; 7906 dst = &(*state)->src; 7907 } 7908 } else { 7909 if (PF_REVERSED_KEY(*state, pd->af)) { 7910 src = &(*state)->dst; 7911 dst = &(*state)->src; 7912 } else { 7913 src = &(*state)->src; 7914 dst = &(*state)->dst; 7915 } 7916 } 7917 7918 if (src->wscale && dst->wscale) 7919 dws = dst->wscale & PF_WSCALE_MASK; 7920 else 7921 dws = 0; 7922 7923 /* Demodulate sequence number */ 7924 seq = ntohl(th->th_seq) - src->seqdiff; 7925 if (src->seqdiff) { 7926 pf_change_a(&th->th_seq, icmpsum, 7927 htonl(seq), 0); 7928 copyback = 1; 7929 } 7930 7931 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7932 (!SEQ_GEQ(src->seqhi, seq) || 7933 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7934 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7935 printf("pf: BAD ICMP %d:%d ", 7936 icmptype, icmpcode); 7937 pf_print_host(pd->src, 0, pd->af); 7938 printf(" -> "); 7939 pf_print_host(pd->dst, 0, pd->af); 7940 printf(" state: "); 7941 pf_print_state(*state); 7942 printf(" seq=%u\n", seq); 7943 } 7944 REASON_SET(reason, PFRES_BADSTATE); 7945 return (PF_DROP); 7946 } else { 7947 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7948 printf("pf: OK ICMP %d:%d ", 7949 icmptype, icmpcode); 7950 pf_print_host(pd->src, 0, pd->af); 7951 printf(" -> "); 7952 pf_print_host(pd->dst, 0, pd->af); 7953 printf(" state: "); 7954 pf_print_state(*state); 7955 printf(" seq=%u\n", seq); 7956 } 7957 } 7958 7959 /* translate source/destination address, if necessary */ 7960 if ((*state)->key[PF_SK_WIRE] != 7961 (*state)->key[PF_SK_STACK]) { 7962 7963 struct pf_state_key *nk; 7964 7965 if (PF_REVERSED_KEY(*state, pd->af)) 7966 nk = (*state)->key[pd->sidx]; 7967 else 7968 nk = (*state)->key[pd->didx]; 7969 7970 #if defined(INET) && defined(INET6) 7971 int afto, sidx, didx; 7972 7973 afto = pd->af != nk->af; 7974 7975 if (afto && (*state)->direction == PF_IN) { 7976 sidx = pd2.didx; 7977 didx = pd2.sidx; 7978 } else { 7979 sidx = pd2.sidx; 7980 didx = pd2.didx; 7981 } 7982 7983 if (afto) { 7984 if (pf_translate_icmp_af(nk->af, 7985 &pd->hdr.icmp)) 7986 return (PF_DROP); 7987 m_copyback(pd->m, pd->off, 7988 sizeof(struct icmp6_hdr), 7989 (c_caddr_t)&pd->hdr.icmp6); 7990 if (pf_change_icmp_af(pd->m, ipoff2, pd, 7991 &pd2, &nk->addr[sidx], 7992 &nk->addr[didx], pd->af, 7993 nk->af)) 7994 return (PF_DROP); 7995 PF_ACPY(&pd->nsaddr, &nk->addr[pd2.sidx], 7996 nk->af); 7997 PF_ACPY(&pd->ndaddr, 7998 &nk->addr[pd2.didx], nk->af); 7999 if (nk->af == AF_INET) { 8000 pd->proto = IPPROTO_ICMP; 8001 } else { 8002 pd->proto = IPPROTO_ICMPV6; 8003 /* 8004 * IPv4 becomes IPv6 so we must 8005 * copy IPv4 src addr to least 8006 * 32bits in IPv6 address to 8007 * keep traceroute/icmp 8008 * working. 8009 */ 8010 pd->nsaddr.addr32[3] = 8011 pd->src->addr32[0]; 8012 } 8013 pd->naf = pd2.naf = nk->af; 8014 pf_change_ap(&pd2, pd2.src, &th->th_sport, 8015 &nk->addr[pd2.sidx], nk->port[sidx]); 8016 pf_change_ap(&pd2, pd2.dst, &th->th_dport, 8017 &nk->addr[pd2.didx], nk->port[didx]); 8018 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)th); 8019 return (PF_AFRT); 8020 } 8021 #endif /* INET && INET6 */ 8022 8023 if (PF_ANEQ(pd2.src, 8024 &nk->addr[pd2.sidx], pd2.af) || 8025 nk->port[pd2.sidx] != th->th_sport) 8026 pf_change_icmp(pd2.src, &th->th_sport, 8027 daddr, &nk->addr[pd2.sidx], 8028 nk->port[pd2.sidx], NULL, 8029 pd2.ip_sum, icmpsum, 8030 pd->ip_sum, 0, pd2.af); 8031 8032 if (PF_ANEQ(pd2.dst, 8033 &nk->addr[pd2.didx], pd2.af) || 8034 nk->port[pd2.didx] != th->th_dport) 8035 pf_change_icmp(pd2.dst, &th->th_dport, 8036 saddr, &nk->addr[pd2.didx], 8037 nk->port[pd2.didx], NULL, 8038 pd2.ip_sum, icmpsum, 8039 pd->ip_sum, 0, pd2.af); 8040 copyback = 1; 8041 } 8042 8043 if (copyback) { 8044 switch (pd2.af) { 8045 #ifdef INET 8046 case AF_INET: 8047 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8048 (caddr_t )&pd->hdr.icmp); 8049 m_copyback(pd->m, ipoff2, sizeof(h2), 8050 (caddr_t )&h2); 8051 break; 8052 #endif /* INET */ 8053 #ifdef INET6 8054 case AF_INET6: 8055 m_copyback(pd->m, pd->off, 8056 sizeof(struct icmp6_hdr), 8057 (caddr_t )&pd->hdr.icmp6); 8058 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8059 (caddr_t )&h2_6); 8060 break; 8061 #endif /* INET6 */ 8062 default: 8063 unhandled_af(pd->af); 8064 } 8065 m_copyback(pd->m, pd2.off, 8, (caddr_t)th); 8066 } 8067 8068 return (PF_PASS); 8069 break; 8070 } 8071 case IPPROTO_UDP: { 8072 struct udphdr *uh = &pd2.hdr.udp; 8073 8074 if (!pf_pull_hdr(pd->m, pd2.off, uh, sizeof(*uh), 8075 NULL, reason, pd2.af)) { 8076 DPFPRINTF(PF_DEBUG_MISC, 8077 ("pf: ICMP error message too short " 8078 "(udp)\n")); 8079 return (PF_DROP); 8080 } 8081 pd2.pcksum = &pd2.hdr.udp.uh_sum; 8082 8083 key.af = pd2.af; 8084 key.proto = IPPROTO_UDP; 8085 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8086 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8087 key.port[pd2.sidx] = uh->uh_sport; 8088 key.port[pd2.didx] = uh->uh_dport; 8089 8090 STATE_LOOKUP(&key, *state, pd); 8091 8092 /* translate source/destination address, if necessary */ 8093 if ((*state)->key[PF_SK_WIRE] != 8094 (*state)->key[PF_SK_STACK]) { 8095 struct pf_state_key *nk; 8096 8097 if (PF_REVERSED_KEY(*state, pd->af)) 8098 nk = (*state)->key[pd->sidx]; 8099 else 8100 nk = (*state)->key[pd->didx]; 8101 8102 #if defined(INET) && defined(INET6) 8103 int afto, sidx, didx; 8104 8105 afto = pd->af != nk->af; 8106 8107 if (afto && (*state)->direction == PF_IN) { 8108 sidx = pd2.didx; 8109 didx = pd2.sidx; 8110 } else { 8111 sidx = pd2.sidx; 8112 didx = pd2.didx; 8113 } 8114 8115 if (afto) { 8116 if (pf_translate_icmp_af(nk->af, 8117 &pd->hdr.icmp)) 8118 return (PF_DROP); 8119 m_copyback(pd->m, pd->off, 8120 sizeof(struct icmp6_hdr), 8121 (c_caddr_t)&pd->hdr.icmp6); 8122 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8123 &pd2, &nk->addr[sidx], 8124 &nk->addr[didx], pd->af, 8125 nk->af)) 8126 return (PF_DROP); 8127 PF_ACPY(&pd->nsaddr, 8128 &nk->addr[pd2.sidx], nk->af); 8129 PF_ACPY(&pd->ndaddr, 8130 &nk->addr[pd2.didx], nk->af); 8131 if (nk->af == AF_INET) { 8132 pd->proto = IPPROTO_ICMP; 8133 } else { 8134 pd->proto = IPPROTO_ICMPV6; 8135 /* 8136 * IPv4 becomes IPv6 so we must 8137 * copy IPv4 src addr to least 8138 * 32bits in IPv6 address to 8139 * keep traceroute/icmp 8140 * working. 8141 */ 8142 pd->nsaddr.addr32[3] = 8143 pd->src->addr32[0]; 8144 } 8145 pd->naf = pd2.naf = nk->af; 8146 pf_change_ap(&pd2, pd2.src, &uh->uh_sport, 8147 &nk->addr[pd2.sidx], nk->port[sidx]); 8148 pf_change_ap(&pd2, pd2.dst, &uh->uh_dport, 8149 &nk->addr[pd2.didx], nk->port[didx]); 8150 m_copyback(pd2.m, pd2.off, sizeof(*uh), 8151 (c_caddr_t)uh); 8152 return (PF_AFRT); 8153 } 8154 #endif /* INET && INET6 */ 8155 8156 if (PF_ANEQ(pd2.src, 8157 &nk->addr[pd2.sidx], pd2.af) || 8158 nk->port[pd2.sidx] != uh->uh_sport) 8159 pf_change_icmp(pd2.src, &uh->uh_sport, 8160 daddr, &nk->addr[pd2.sidx], 8161 nk->port[pd2.sidx], &uh->uh_sum, 8162 pd2.ip_sum, icmpsum, 8163 pd->ip_sum, 1, pd2.af); 8164 8165 if (PF_ANEQ(pd2.dst, 8166 &nk->addr[pd2.didx], pd2.af) || 8167 nk->port[pd2.didx] != uh->uh_dport) 8168 pf_change_icmp(pd2.dst, &uh->uh_dport, 8169 saddr, &nk->addr[pd2.didx], 8170 nk->port[pd2.didx], &uh->uh_sum, 8171 pd2.ip_sum, icmpsum, 8172 pd->ip_sum, 1, pd2.af); 8173 8174 switch (pd2.af) { 8175 #ifdef INET 8176 case AF_INET: 8177 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8178 (caddr_t )&pd->hdr.icmp); 8179 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8180 break; 8181 #endif /* INET */ 8182 #ifdef INET6 8183 case AF_INET6: 8184 m_copyback(pd->m, pd->off, 8185 sizeof(struct icmp6_hdr), 8186 (caddr_t )&pd->hdr.icmp6); 8187 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8188 (caddr_t )&h2_6); 8189 break; 8190 #endif /* INET6 */ 8191 } 8192 m_copyback(pd->m, pd2.off, sizeof(*uh), (caddr_t)uh); 8193 } 8194 return (PF_PASS); 8195 break; 8196 } 8197 #ifdef INET 8198 case IPPROTO_SCTP: { 8199 struct sctphdr *sh = &pd2.hdr.sctp; 8200 struct pf_state_peer *src; 8201 int copyback = 0; 8202 8203 if (! pf_pull_hdr(pd->m, pd2.off, sh, sizeof(*sh), NULL, reason, 8204 pd2.af)) { 8205 DPFPRINTF(PF_DEBUG_MISC, 8206 ("pf: ICMP error message too short " 8207 "(sctp)\n")); 8208 return (PF_DROP); 8209 } 8210 pd2.pcksum = &pd2.sctp_dummy_sum; 8211 8212 key.af = pd2.af; 8213 key.proto = IPPROTO_SCTP; 8214 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8215 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8216 key.port[pd2.sidx] = sh->src_port; 8217 key.port[pd2.didx] = sh->dest_port; 8218 8219 STATE_LOOKUP(&key, *state, pd); 8220 8221 if (pd->dir == (*state)->direction) { 8222 if (PF_REVERSED_KEY(*state, pd->af)) 8223 src = &(*state)->src; 8224 else 8225 src = &(*state)->dst; 8226 } else { 8227 if (PF_REVERSED_KEY(*state, pd->af)) 8228 src = &(*state)->dst; 8229 else 8230 src = &(*state)->src; 8231 } 8232 8233 if (src->scrub->pfss_v_tag != sh->v_tag) { 8234 DPFPRINTF(PF_DEBUG_MISC, 8235 ("pf: ICMP error message has incorrect " 8236 "SCTP v_tag\n")); 8237 return (PF_DROP); 8238 } 8239 8240 /* translate source/destination address, if necessary */ 8241 if ((*state)->key[PF_SK_WIRE] != 8242 (*state)->key[PF_SK_STACK]) { 8243 8244 struct pf_state_key *nk; 8245 8246 if (PF_REVERSED_KEY(*state, pd->af)) 8247 nk = (*state)->key[pd->sidx]; 8248 else 8249 nk = (*state)->key[pd->didx]; 8250 8251 #if defined(INET) && defined(INET6) 8252 int afto, sidx, didx; 8253 8254 afto = pd->af != nk->af; 8255 8256 if (afto && (*state)->direction == PF_IN) { 8257 sidx = pd2.didx; 8258 didx = pd2.sidx; 8259 } else { 8260 sidx = pd2.sidx; 8261 didx = pd2.didx; 8262 } 8263 8264 if (afto) { 8265 if (pf_translate_icmp_af(nk->af, 8266 &pd->hdr.icmp)) 8267 return (PF_DROP); 8268 m_copyback(pd->m, pd->off, 8269 sizeof(struct icmp6_hdr), 8270 (c_caddr_t)&pd->hdr.icmp6); 8271 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8272 &pd2, &nk->addr[sidx], 8273 &nk->addr[didx], pd->af, 8274 nk->af)) 8275 return (PF_DROP); 8276 sh->src_port = nk->port[sidx]; 8277 sh->dest_port = nk->port[didx]; 8278 m_copyback(pd2.m, pd2.off, sizeof(*sh), (c_caddr_t)sh); 8279 PF_ACPY(&pd->nsaddr, 8280 &nk->addr[pd2.sidx], nk->af); 8281 PF_ACPY(&pd->ndaddr, 8282 &nk->addr[pd2.didx], nk->af); 8283 if (nk->af == AF_INET) { 8284 pd->proto = IPPROTO_ICMP; 8285 } else { 8286 pd->proto = IPPROTO_ICMPV6; 8287 /* 8288 * IPv4 becomes IPv6 so we must 8289 * copy IPv4 src addr to least 8290 * 32bits in IPv6 address to 8291 * keep traceroute/icmp 8292 * working. 8293 */ 8294 pd->nsaddr.addr32[3] = 8295 pd->src->addr32[0]; 8296 } 8297 pd->naf = nk->af; 8298 return (PF_AFRT); 8299 } 8300 #endif /* INET && INET6 */ 8301 8302 if (PF_ANEQ(pd2.src, 8303 &nk->addr[pd2.sidx], pd2.af) || 8304 nk->port[pd2.sidx] != sh->src_port) 8305 pf_change_icmp(pd2.src, &sh->src_port, 8306 daddr, &nk->addr[pd2.sidx], 8307 nk->port[pd2.sidx], NULL, 8308 pd2.ip_sum, icmpsum, 8309 pd->ip_sum, 0, pd2.af); 8310 8311 if (PF_ANEQ(pd2.dst, 8312 &nk->addr[pd2.didx], pd2.af) || 8313 nk->port[pd2.didx] != sh->dest_port) 8314 pf_change_icmp(pd2.dst, &sh->dest_port, 8315 saddr, &nk->addr[pd2.didx], 8316 nk->port[pd2.didx], NULL, 8317 pd2.ip_sum, icmpsum, 8318 pd->ip_sum, 0, pd2.af); 8319 copyback = 1; 8320 } 8321 8322 if (copyback) { 8323 switch (pd2.af) { 8324 #ifdef INET 8325 case AF_INET: 8326 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8327 (caddr_t )&pd->hdr.icmp); 8328 m_copyback(pd->m, ipoff2, sizeof(h2), 8329 (caddr_t )&h2); 8330 break; 8331 #endif /* INET */ 8332 #ifdef INET6 8333 case AF_INET6: 8334 m_copyback(pd->m, pd->off, 8335 sizeof(struct icmp6_hdr), 8336 (caddr_t )&pd->hdr.icmp6); 8337 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8338 (caddr_t )&h2_6); 8339 break; 8340 #endif /* INET6 */ 8341 } 8342 m_copyback(pd->m, pd2.off, sizeof(*sh), (caddr_t)sh); 8343 } 8344 8345 return (PF_PASS); 8346 break; 8347 } 8348 case IPPROTO_ICMP: { 8349 struct icmp *iih = &pd2.hdr.icmp; 8350 8351 if (pd2.af != AF_INET) { 8352 REASON_SET(reason, PFRES_NORM); 8353 return (PF_DROP); 8354 } 8355 8356 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8357 NULL, reason, pd2.af)) { 8358 DPFPRINTF(PF_DEBUG_MISC, 8359 ("pf: ICMP error message too short i" 8360 "(icmp)\n")); 8361 return (PF_DROP); 8362 } 8363 pd2.pcksum = &pd2.hdr.icmp.icmp_cksum; 8364 8365 icmpid = iih->icmp_id; 8366 pf_icmp_mapping(&pd2, iih->icmp_type, 8367 &icmp_dir, &virtual_id, &virtual_type); 8368 8369 ret = pf_icmp_state_lookup(&key, &pd2, state, 8370 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8371 if (ret >= 0) { 8372 MPASS(*state == NULL); 8373 return (ret); 8374 } 8375 8376 /* translate source/destination address, if necessary */ 8377 if ((*state)->key[PF_SK_WIRE] != 8378 (*state)->key[PF_SK_STACK]) { 8379 struct pf_state_key *nk; 8380 8381 if (PF_REVERSED_KEY(*state, pd->af)) 8382 nk = (*state)->key[pd->sidx]; 8383 else 8384 nk = (*state)->key[pd->didx]; 8385 8386 #if defined(INET) && defined(INET6) 8387 int afto, sidx, didx; 8388 8389 afto = pd->af != nk->af; 8390 8391 if (afto && (*state)->direction == PF_IN) { 8392 sidx = pd2.didx; 8393 didx = pd2.sidx; 8394 iidx = !iidx; 8395 } else { 8396 sidx = pd2.sidx; 8397 didx = pd2.didx; 8398 } 8399 8400 if (afto) { 8401 if (nk->af != AF_INET6) 8402 return (PF_DROP); 8403 if (pf_translate_icmp_af(nk->af, 8404 &pd->hdr.icmp)) 8405 return (PF_DROP); 8406 m_copyback(pd->m, pd->off, 8407 sizeof(struct icmp6_hdr), 8408 (c_caddr_t)&pd->hdr.icmp6); 8409 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8410 &pd2, &nk->addr[sidx], 8411 &nk->addr[didx], pd->af, 8412 nk->af)) 8413 return (PF_DROP); 8414 pd->proto = IPPROTO_ICMPV6; 8415 if (pf_translate_icmp_af(nk->af, iih)) 8416 return (PF_DROP); 8417 if (virtual_type == htons(ICMP_ECHO) && 8418 nk->port[iidx] != iih->icmp_id) 8419 iih->icmp_id = nk->port[iidx]; 8420 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8421 (c_caddr_t)iih); 8422 PF_ACPY(&pd->nsaddr, 8423 &nk->addr[pd2.sidx], nk->af); 8424 PF_ACPY(&pd->ndaddr, 8425 &nk->addr[pd2.didx], nk->af); 8426 /* 8427 * IPv4 becomes IPv6 so we must copy 8428 * IPv4 src addr to least 32bits in 8429 * IPv6 address to keep traceroute 8430 * working. 8431 */ 8432 pd->nsaddr.addr32[3] = 8433 pd->src->addr32[0]; 8434 pd->naf = nk->af; 8435 return (PF_AFRT); 8436 } 8437 #endif /* INET && INET6 */ 8438 8439 if (PF_ANEQ(pd2.src, 8440 &nk->addr[pd2.sidx], pd2.af) || 8441 (virtual_type == htons(ICMP_ECHO) && 8442 nk->port[iidx] != iih->icmp_id)) 8443 pf_change_icmp(pd2.src, 8444 (virtual_type == htons(ICMP_ECHO)) ? 8445 &iih->icmp_id : NULL, 8446 daddr, &nk->addr[pd2.sidx], 8447 (virtual_type == htons(ICMP_ECHO)) ? 8448 nk->port[iidx] : 0, NULL, 8449 pd2.ip_sum, icmpsum, 8450 pd->ip_sum, 0, AF_INET); 8451 8452 if (PF_ANEQ(pd2.dst, 8453 &nk->addr[pd2.didx], pd2.af)) 8454 pf_change_icmp(pd2.dst, NULL, NULL, 8455 &nk->addr[pd2.didx], 0, NULL, 8456 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8457 AF_INET); 8458 8459 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8460 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8461 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8462 } 8463 return (PF_PASS); 8464 break; 8465 } 8466 #endif /* INET */ 8467 #ifdef INET6 8468 case IPPROTO_ICMPV6: { 8469 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8470 8471 if (pd2.af != AF_INET6) { 8472 REASON_SET(reason, PFRES_NORM); 8473 return (PF_DROP); 8474 } 8475 8476 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8477 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 8478 DPFPRINTF(PF_DEBUG_MISC, 8479 ("pf: ICMP error message too short " 8480 "(icmp6)\n")); 8481 return (PF_DROP); 8482 } 8483 pd2.pcksum = &pd2.hdr.icmp6.icmp6_cksum; 8484 8485 pf_icmp_mapping(&pd2, iih->icmp6_type, 8486 &icmp_dir, &virtual_id, &virtual_type); 8487 8488 ret = pf_icmp_state_lookup(&key, &pd2, state, 8489 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8490 /* IPv6? try matching a multicast address */ 8491 if (ret == PF_DROP && pd2.af == AF_INET6 && 8492 icmp_dir == PF_OUT) { 8493 MPASS(*state == NULL); 8494 ret = pf_icmp_state_lookup(&key, &pd2, 8495 state, virtual_id, virtual_type, 8496 icmp_dir, &iidx, 1, 1); 8497 } 8498 if (ret >= 0) { 8499 MPASS(*state == NULL); 8500 return (ret); 8501 } 8502 8503 /* translate source/destination address, if necessary */ 8504 if ((*state)->key[PF_SK_WIRE] != 8505 (*state)->key[PF_SK_STACK]) { 8506 struct pf_state_key *nk; 8507 8508 if (PF_REVERSED_KEY(*state, pd->af)) 8509 nk = (*state)->key[pd->sidx]; 8510 else 8511 nk = (*state)->key[pd->didx]; 8512 8513 #if defined(INET) && defined(INET6) 8514 int afto, sidx, didx; 8515 8516 afto = pd->af != nk->af; 8517 8518 if (afto && (*state)->direction == PF_IN) { 8519 sidx = pd2.didx; 8520 didx = pd2.sidx; 8521 iidx = !iidx; 8522 } else { 8523 sidx = pd2.sidx; 8524 didx = pd2.didx; 8525 } 8526 8527 if (afto) { 8528 if (nk->af != AF_INET) 8529 return (PF_DROP); 8530 if (pf_translate_icmp_af(nk->af, 8531 &pd->hdr.icmp)) 8532 return (PF_DROP); 8533 m_copyback(pd->m, pd->off, 8534 sizeof(struct icmp6_hdr), 8535 (c_caddr_t)&pd->hdr.icmp6); 8536 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8537 &pd2, &nk->addr[sidx], 8538 &nk->addr[didx], pd->af, 8539 nk->af)) 8540 return (PF_DROP); 8541 pd->proto = IPPROTO_ICMP; 8542 if (pf_translate_icmp_af(nk->af, iih)) 8543 return (PF_DROP); 8544 if (virtual_type == 8545 htons(ICMP6_ECHO_REQUEST) && 8546 nk->port[iidx] != iih->icmp6_id) 8547 iih->icmp6_id = nk->port[iidx]; 8548 m_copyback(pd2.m, pd2.off, 8549 sizeof(struct icmp6_hdr), (c_caddr_t)iih); 8550 PF_ACPY(&pd->nsaddr, 8551 &nk->addr[pd2.sidx], nk->af); 8552 PF_ACPY(&pd->ndaddr, 8553 &nk->addr[pd2.didx], nk->af); 8554 pd->naf = nk->af; 8555 return (PF_AFRT); 8556 } 8557 #endif /* INET && INET6 */ 8558 8559 if (PF_ANEQ(pd2.src, 8560 &nk->addr[pd2.sidx], pd2.af) || 8561 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8562 nk->port[pd2.sidx] != iih->icmp6_id)) 8563 pf_change_icmp(pd2.src, 8564 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8565 ? &iih->icmp6_id : NULL, 8566 daddr, &nk->addr[pd2.sidx], 8567 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8568 ? nk->port[iidx] : 0, NULL, 8569 pd2.ip_sum, icmpsum, 8570 pd->ip_sum, 0, AF_INET6); 8571 8572 if (PF_ANEQ(pd2.dst, 8573 &nk->addr[pd2.didx], pd2.af)) 8574 pf_change_icmp(pd2.dst, NULL, NULL, 8575 &nk->addr[pd2.didx], 0, NULL, 8576 pd2.ip_sum, icmpsum, 8577 pd->ip_sum, 0, AF_INET6); 8578 8579 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8580 (caddr_t)&pd->hdr.icmp6); 8581 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8582 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8583 (caddr_t)iih); 8584 } 8585 return (PF_PASS); 8586 break; 8587 } 8588 #endif /* INET6 */ 8589 default: { 8590 key.af = pd2.af; 8591 key.proto = pd2.proto; 8592 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8593 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8594 key.port[0] = key.port[1] = 0; 8595 8596 STATE_LOOKUP(&key, *state, pd); 8597 8598 /* translate source/destination address, if necessary */ 8599 if ((*state)->key[PF_SK_WIRE] != 8600 (*state)->key[PF_SK_STACK]) { 8601 struct pf_state_key *nk = 8602 (*state)->key[pd->didx]; 8603 8604 if (PF_ANEQ(pd2.src, 8605 &nk->addr[pd2.sidx], pd2.af)) 8606 pf_change_icmp(pd2.src, NULL, daddr, 8607 &nk->addr[pd2.sidx], 0, NULL, 8608 pd2.ip_sum, icmpsum, 8609 pd->ip_sum, 0, pd2.af); 8610 8611 if (PF_ANEQ(pd2.dst, 8612 &nk->addr[pd2.didx], pd2.af)) 8613 pf_change_icmp(pd2.dst, NULL, saddr, 8614 &nk->addr[pd2.didx], 0, NULL, 8615 pd2.ip_sum, icmpsum, 8616 pd->ip_sum, 0, pd2.af); 8617 8618 switch (pd2.af) { 8619 #ifdef INET 8620 case AF_INET: 8621 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8622 (caddr_t)&pd->hdr.icmp); 8623 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8624 break; 8625 #endif /* INET */ 8626 #ifdef INET6 8627 case AF_INET6: 8628 m_copyback(pd->m, pd->off, 8629 sizeof(struct icmp6_hdr), 8630 (caddr_t )&pd->hdr.icmp6); 8631 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8632 (caddr_t )&h2_6); 8633 break; 8634 #endif /* INET6 */ 8635 } 8636 } 8637 return (PF_PASS); 8638 break; 8639 } 8640 } 8641 } 8642 } 8643 8644 /* 8645 * ipoff and off are measured from the start of the mbuf chain. 8646 * h must be at "ipoff" on the mbuf chain. 8647 */ 8648 void * 8649 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8650 u_short *actionp, u_short *reasonp, sa_family_t af) 8651 { 8652 switch (af) { 8653 #ifdef INET 8654 case AF_INET: { 8655 const struct ip *h = mtod(m, struct ip *); 8656 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 8657 8658 if (fragoff) { 8659 if (fragoff >= len) 8660 ACTION_SET(actionp, PF_PASS); 8661 else { 8662 ACTION_SET(actionp, PF_DROP); 8663 REASON_SET(reasonp, PFRES_FRAG); 8664 } 8665 return (NULL); 8666 } 8667 if (m->m_pkthdr.len < off + len || 8668 ntohs(h->ip_len) < off + len) { 8669 ACTION_SET(actionp, PF_DROP); 8670 REASON_SET(reasonp, PFRES_SHORT); 8671 return (NULL); 8672 } 8673 break; 8674 } 8675 #endif /* INET */ 8676 #ifdef INET6 8677 case AF_INET6: { 8678 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8679 8680 if (m->m_pkthdr.len < off + len || 8681 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 8682 (unsigned)(off + len)) { 8683 ACTION_SET(actionp, PF_DROP); 8684 REASON_SET(reasonp, PFRES_SHORT); 8685 return (NULL); 8686 } 8687 break; 8688 } 8689 #endif /* INET6 */ 8690 } 8691 m_copydata(m, off, len, p); 8692 return (p); 8693 } 8694 8695 int 8696 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 8697 int rtableid) 8698 { 8699 struct ifnet *ifp; 8700 8701 /* 8702 * Skip check for addresses with embedded interface scope, 8703 * as they would always match anyway. 8704 */ 8705 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 8706 return (1); 8707 8708 if (af != AF_INET && af != AF_INET6) 8709 return (0); 8710 8711 if (kif == V_pfi_all) 8712 return (1); 8713 8714 /* Skip checks for ipsec interfaces */ 8715 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 8716 return (1); 8717 8718 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 8719 8720 switch (af) { 8721 #ifdef INET6 8722 case AF_INET6: 8723 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 8724 ifp)); 8725 #endif /* INET6 */ 8726 #ifdef INET 8727 case AF_INET: 8728 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 8729 ifp)); 8730 #endif /* INET */ 8731 } 8732 8733 return (0); 8734 } 8735 8736 #ifdef INET 8737 static void 8738 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8739 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8740 { 8741 struct mbuf *m0, *m1, *md; 8742 struct route ro; 8743 const struct sockaddr *gw = &ro.ro_dst; 8744 struct sockaddr_in *dst; 8745 struct ip *ip; 8746 struct ifnet *ifp = NULL; 8747 int error = 0; 8748 uint16_t ip_len, ip_off; 8749 uint16_t tmp; 8750 int r_dir; 8751 bool skip_test = false; 8752 8753 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8754 8755 SDT_PROBE4(pf, ip, route_to, entry, *m, pd, s, oifp); 8756 8757 if (s) { 8758 r_dir = s->direction; 8759 } else { 8760 r_dir = r->direction; 8761 } 8762 8763 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8764 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8765 __func__)); 8766 8767 if ((pd->pf_mtag == NULL && 8768 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8769 pd->pf_mtag->routed++ > 3) { 8770 m0 = *m; 8771 *m = NULL; 8772 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8773 goto bad_locked; 8774 } 8775 8776 if (pd->act.rt_kif != NULL) 8777 ifp = pd->act.rt_kif->pfik_ifp; 8778 8779 if (pd->act.rt == PF_DUPTO) { 8780 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8781 if (s != NULL) { 8782 PF_STATE_UNLOCK(s); 8783 } 8784 if (ifp == oifp) { 8785 /* When the 2nd interface is not skipped */ 8786 return; 8787 } else { 8788 m0 = *m; 8789 *m = NULL; 8790 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8791 goto bad; 8792 } 8793 } else { 8794 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8795 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8796 if (s) 8797 PF_STATE_UNLOCK(s); 8798 return; 8799 } 8800 } 8801 } else { 8802 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8803 if (pd->af == pd->naf) { 8804 pf_dummynet(pd, s, r, m); 8805 if (s) 8806 PF_STATE_UNLOCK(s); 8807 return; 8808 } else { 8809 if (r_dir == PF_IN) { 8810 skip_test = true; 8811 } 8812 } 8813 } 8814 8815 /* 8816 * If we're actually doing route-to and af-to and are in the 8817 * reply direction. 8818 */ 8819 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 8820 pd->af != pd->naf) { 8821 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 8822 /* Un-set ifp so we do a plain route lookup. */ 8823 ifp = NULL; 8824 } 8825 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 8826 /* Un-set ifp so we do a plain route lookup. */ 8827 ifp = NULL; 8828 } 8829 } 8830 m0 = *m; 8831 } 8832 8833 ip = mtod(m0, struct ip *); 8834 8835 bzero(&ro, sizeof(ro)); 8836 dst = (struct sockaddr_in *)&ro.ro_dst; 8837 dst->sin_family = AF_INET; 8838 dst->sin_len = sizeof(struct sockaddr_in); 8839 dst->sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 8840 8841 if (s != NULL){ 8842 if (ifp == NULL && (pd->af != pd->naf)) { 8843 /* We're in the AFTO case. Do a route lookup. */ 8844 const struct nhop_object *nh; 8845 nh = fib4_lookup(M_GETFIB(*m), ip->ip_dst, 0, NHR_NONE, 0); 8846 if (nh) { 8847 ifp = nh->nh_ifp; 8848 8849 /* Use the gateway if needed. */ 8850 if (nh->nh_flags & NHF_GATEWAY) { 8851 gw = &nh->gw_sa; 8852 ro.ro_flags |= RT_HAS_GW; 8853 } else { 8854 dst->sin_addr = ip->ip_dst; 8855 } 8856 8857 /* 8858 * Bind to the correct interface if we're 8859 * if-bound. We don't know which interface 8860 * that will be until here, so we've inserted 8861 * the state on V_pf_all. Fix that now. 8862 */ 8863 if (s->kif == V_pfi_all && ifp != NULL && 8864 r->rule_flag & PFRULE_IFBOUND) 8865 s->kif = ifp->if_pf_kif; 8866 } 8867 } 8868 8869 if (r->rule_flag & PFRULE_IFBOUND && 8870 pd->act.rt == PF_REPLYTO && 8871 s->kif == V_pfi_all) { 8872 s->kif = pd->act.rt_kif; 8873 s->orig_kif = oifp->if_pf_kif; 8874 } 8875 8876 PF_STATE_UNLOCK(s); 8877 } 8878 8879 if (ifp == NULL) { 8880 m0 = *m; 8881 *m = NULL; 8882 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8883 goto bad; 8884 } 8885 8886 if (pd->dir == PF_IN && !skip_test) { 8887 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8888 &pd->act) != PF_PASS) { 8889 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8890 goto bad; 8891 } else if (m0 == NULL) { 8892 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8893 goto done; 8894 } 8895 if (m0->m_len < sizeof(struct ip)) { 8896 DPFPRINTF(PF_DEBUG_URGENT, 8897 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 8898 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8899 goto bad; 8900 } 8901 ip = mtod(m0, struct ip *); 8902 } 8903 8904 if (ifp->if_flags & IFF_LOOPBACK) 8905 m0->m_flags |= M_SKIP_FIREWALL; 8906 8907 ip_len = ntohs(ip->ip_len); 8908 ip_off = ntohs(ip->ip_off); 8909 8910 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 8911 m0->m_pkthdr.csum_flags |= CSUM_IP; 8912 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 8913 in_delayed_cksum(m0); 8914 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 8915 } 8916 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 8917 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 8918 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 8919 } 8920 8921 if (pd->dir == PF_IN) { 8922 /* 8923 * Make sure dummynet gets the correct direction, in case it needs to 8924 * re-inject later. 8925 */ 8926 pd->dir = PF_OUT; 8927 8928 /* 8929 * The following processing is actually the rest of the inbound processing, even 8930 * though we've marked it as outbound (so we don't look through dummynet) and it 8931 * happens after the outbound processing (pf_test(PF_OUT) above). 8932 * Swap the dummynet pipe numbers, because it's going to come to the wrong 8933 * conclusion about what direction it's processing, and we can't fix it or it 8934 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 8935 * decision will pick the right pipe, and everything will mostly work as expected. 8936 */ 8937 tmp = pd->act.dnrpipe; 8938 pd->act.dnrpipe = pd->act.dnpipe; 8939 pd->act.dnpipe = tmp; 8940 } 8941 8942 /* 8943 * If small enough for interface, or the interface will take 8944 * care of the fragmentation for us, we can just send directly. 8945 */ 8946 if (ip_len <= ifp->if_mtu || 8947 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 8948 ip->ip_sum = 0; 8949 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 8950 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 8951 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 8952 } 8953 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 8954 8955 md = m0; 8956 error = pf_dummynet_route(pd, s, r, ifp, gw, &md); 8957 if (md != NULL) { 8958 error = (*ifp->if_output)(ifp, md, gw, &ro); 8959 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 8960 } 8961 goto done; 8962 } 8963 8964 /* Balk when DF bit is set or the interface didn't support TSO. */ 8965 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 8966 error = EMSGSIZE; 8967 KMOD_IPSTAT_INC(ips_cantfrag); 8968 if (pd->act.rt != PF_DUPTO) { 8969 if (s && s->nat_rule != NULL) 8970 MPASS(m0 == pd->m); 8971 PACKET_UNDO_NAT(pd, 8972 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 8973 s); 8974 8975 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 8976 ifp->if_mtu); 8977 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8978 goto done; 8979 } else { 8980 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8981 goto bad; 8982 } 8983 } 8984 8985 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 8986 if (error) { 8987 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8988 goto bad; 8989 } 8990 8991 for (; m0; m0 = m1) { 8992 m1 = m0->m_nextpkt; 8993 m0->m_nextpkt = NULL; 8994 if (error == 0) { 8995 m_clrprotoflags(m0); 8996 md = m0; 8997 pd->pf_mtag = pf_find_mtag(md); 8998 error = pf_dummynet_route(pd, s, r, ifp, 8999 gw, &md); 9000 if (md != NULL) { 9001 error = (*ifp->if_output)(ifp, md, gw, &ro); 9002 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9003 } 9004 } else 9005 m_freem(m0); 9006 } 9007 9008 if (error == 0) 9009 KMOD_IPSTAT_INC(ips_fragmented); 9010 9011 done: 9012 if (pd->act.rt != PF_DUPTO) 9013 *m = NULL; 9014 return; 9015 9016 bad_locked: 9017 if (s) 9018 PF_STATE_UNLOCK(s); 9019 bad: 9020 m_freem(m0); 9021 goto done; 9022 } 9023 #endif /* INET */ 9024 9025 #ifdef INET6 9026 static void 9027 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 9028 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9029 { 9030 struct mbuf *m0, *md; 9031 struct m_tag *mtag; 9032 struct sockaddr_in6 dst; 9033 struct ip6_hdr *ip6; 9034 struct ifnet *ifp = NULL; 9035 int r_dir; 9036 bool skip_test = false; 9037 9038 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 9039 9040 SDT_PROBE4(pf, ip6, route_to, entry, *m, pd, s, oifp); 9041 9042 if (s) { 9043 r_dir = s->direction; 9044 } else { 9045 r_dir = r->direction; 9046 } 9047 9048 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9049 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9050 __func__)); 9051 9052 if ((pd->pf_mtag == NULL && 9053 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 9054 pd->pf_mtag->routed++ > 3) { 9055 m0 = *m; 9056 *m = NULL; 9057 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9058 goto bad_locked; 9059 } 9060 9061 if (pd->act.rt_kif != NULL) 9062 ifp = pd->act.rt_kif->pfik_ifp; 9063 9064 if (pd->act.rt == PF_DUPTO) { 9065 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9066 if (s != NULL) { 9067 PF_STATE_UNLOCK(s); 9068 } 9069 if (ifp == oifp) { 9070 /* When the 2nd interface is not skipped */ 9071 return; 9072 } else { 9073 m0 = *m; 9074 *m = NULL; 9075 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9076 goto bad; 9077 } 9078 } else { 9079 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9080 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 9081 if (s) 9082 PF_STATE_UNLOCK(s); 9083 return; 9084 } 9085 } 9086 } else { 9087 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9088 if (pd->af == pd->naf) { 9089 pf_dummynet(pd, s, r, m); 9090 if (s) 9091 PF_STATE_UNLOCK(s); 9092 return; 9093 } else { 9094 if (r_dir == PF_IN) { 9095 skip_test = true; 9096 } 9097 } 9098 } 9099 9100 /* 9101 * If we're actually doing route-to and af-to and are in the 9102 * reply direction. 9103 */ 9104 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9105 pd->af != pd->naf) { 9106 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9107 /* Un-set ifp so we do a plain route lookup. */ 9108 ifp = NULL; 9109 } 9110 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9111 /* Un-set ifp so we do a plain route lookup. */ 9112 ifp = NULL; 9113 } 9114 } 9115 m0 = *m; 9116 } 9117 9118 ip6 = mtod(m0, struct ip6_hdr *); 9119 9120 bzero(&dst, sizeof(dst)); 9121 dst.sin6_family = AF_INET6; 9122 dst.sin6_len = sizeof(dst); 9123 PF_ACPY((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, AF_INET6); 9124 9125 if (s != NULL) { 9126 if (ifp == NULL && (pd->af != pd->naf)) { 9127 const struct nhop_object *nh; 9128 nh = fib6_lookup(M_GETFIB(*m), &ip6->ip6_dst, 0, NHR_NONE, 0); 9129 if (nh) { 9130 ifp = nh->nh_ifp; 9131 9132 /* Use the gateway if needed. */ 9133 if (nh->nh_flags & NHF_GATEWAY) 9134 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9135 sizeof(dst.sin6_addr)); 9136 else 9137 dst.sin6_addr = ip6->ip6_dst; 9138 9139 /* 9140 * Bind to the correct interface if we're 9141 * if-bound. We don't know which interface 9142 * that will be until here, so we've inserted 9143 * the state on V_pf_all. Fix that now. 9144 */ 9145 if (s->kif == V_pfi_all && ifp != NULL && 9146 r->rule_flag & PFRULE_IFBOUND) 9147 s->kif = ifp->if_pf_kif; 9148 } 9149 } 9150 9151 if (r->rule_flag & PFRULE_IFBOUND && 9152 pd->act.rt == PF_REPLYTO && 9153 s->kif == V_pfi_all) { 9154 s->kif = pd->act.rt_kif; 9155 s->orig_kif = oifp->if_pf_kif; 9156 } 9157 9158 PF_STATE_UNLOCK(s); 9159 } 9160 9161 if (pd->af != pd->naf) { 9162 struct udphdr *uh = &pd->hdr.udp; 9163 9164 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9165 uh->uh_sum = in6_cksum_pseudo(ip6, 9166 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9167 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9168 } 9169 } 9170 9171 if (ifp == NULL) { 9172 m0 = *m; 9173 *m = NULL; 9174 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9175 goto bad; 9176 } 9177 9178 if (pd->dir == PF_IN && !skip_test) { 9179 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9180 ifp, &m0, inp, &pd->act) != PF_PASS) { 9181 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9182 goto bad; 9183 } else if (m0 == NULL) { 9184 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9185 goto done; 9186 } 9187 if (m0->m_len < sizeof(struct ip6_hdr)) { 9188 DPFPRINTF(PF_DEBUG_URGENT, 9189 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 9190 __func__)); 9191 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9192 goto bad; 9193 } 9194 ip6 = mtod(m0, struct ip6_hdr *); 9195 } 9196 9197 if (ifp->if_flags & IFF_LOOPBACK) 9198 m0->m_flags |= M_SKIP_FIREWALL; 9199 9200 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9201 ~ifp->if_hwassist) { 9202 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9203 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9204 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9205 } 9206 9207 if (pd->dir == PF_IN) { 9208 uint16_t tmp; 9209 /* 9210 * Make sure dummynet gets the correct direction, in case it needs to 9211 * re-inject later. 9212 */ 9213 pd->dir = PF_OUT; 9214 9215 /* 9216 * The following processing is actually the rest of the inbound processing, even 9217 * though we've marked it as outbound (so we don't look through dummynet) and it 9218 * happens after the outbound processing (pf_test(PF_OUT) above). 9219 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9220 * conclusion about what direction it's processing, and we can't fix it or it 9221 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9222 * decision will pick the right pipe, and everything will mostly work as expected. 9223 */ 9224 tmp = pd->act.dnrpipe; 9225 pd->act.dnrpipe = pd->act.dnpipe; 9226 pd->act.dnpipe = tmp; 9227 } 9228 9229 /* 9230 * If the packet is too large for the outgoing interface, 9231 * send back an icmp6 error. 9232 */ 9233 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9234 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9235 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9236 if (mtag != NULL) { 9237 int ret __sdt_used; 9238 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9239 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9240 goto done; 9241 } 9242 9243 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9244 md = m0; 9245 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9246 if (md != NULL) { 9247 int ret __sdt_used; 9248 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9249 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9250 } 9251 } 9252 else { 9253 in6_ifstat_inc(ifp, ifs6_in_toobig); 9254 if (pd->act.rt != PF_DUPTO) { 9255 if (s && s->nat_rule != NULL) 9256 MPASS(m0 == pd->m); 9257 PACKET_UNDO_NAT(pd, 9258 ((caddr_t)ip6 - m0->m_data) + 9259 sizeof(struct ip6_hdr), s); 9260 9261 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 9262 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9263 } else { 9264 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9265 goto bad; 9266 } 9267 } 9268 9269 done: 9270 if (pd->act.rt != PF_DUPTO) 9271 *m = NULL; 9272 return; 9273 9274 bad_locked: 9275 if (s) 9276 PF_STATE_UNLOCK(s); 9277 bad: 9278 m_freem(m0); 9279 goto done; 9280 } 9281 #endif /* INET6 */ 9282 9283 /* 9284 * FreeBSD supports cksum offloads for the following drivers. 9285 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9286 * 9287 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9288 * network driver performed cksum including pseudo header, need to verify 9289 * csum_data 9290 * CSUM_DATA_VALID : 9291 * network driver performed cksum, needs to additional pseudo header 9292 * cksum computation with partial csum_data(i.e. lack of H/W support for 9293 * pseudo header, for instance sk(4) and possibly gem(4)) 9294 * 9295 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9296 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9297 * TCP/UDP layer. 9298 * Also, set csum_data to 0xffff to force cksum validation. 9299 */ 9300 static int 9301 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9302 { 9303 u_int16_t sum = 0; 9304 int hw_assist = 0; 9305 struct ip *ip; 9306 9307 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9308 return (1); 9309 if (m->m_pkthdr.len < off + len) 9310 return (1); 9311 9312 switch (p) { 9313 case IPPROTO_TCP: 9314 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9315 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9316 sum = m->m_pkthdr.csum_data; 9317 } else { 9318 ip = mtod(m, struct ip *); 9319 sum = in_pseudo(ip->ip_src.s_addr, 9320 ip->ip_dst.s_addr, htonl((u_short)len + 9321 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9322 } 9323 sum ^= 0xffff; 9324 ++hw_assist; 9325 } 9326 break; 9327 case IPPROTO_UDP: 9328 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9329 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9330 sum = m->m_pkthdr.csum_data; 9331 } else { 9332 ip = mtod(m, struct ip *); 9333 sum = in_pseudo(ip->ip_src.s_addr, 9334 ip->ip_dst.s_addr, htonl((u_short)len + 9335 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9336 } 9337 sum ^= 0xffff; 9338 ++hw_assist; 9339 } 9340 break; 9341 case IPPROTO_ICMP: 9342 #ifdef INET6 9343 case IPPROTO_ICMPV6: 9344 #endif /* INET6 */ 9345 break; 9346 default: 9347 return (1); 9348 } 9349 9350 if (!hw_assist) { 9351 switch (af) { 9352 case AF_INET: 9353 if (m->m_len < sizeof(struct ip)) 9354 return (1); 9355 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 9356 break; 9357 #ifdef INET6 9358 case AF_INET6: 9359 if (m->m_len < sizeof(struct ip6_hdr)) 9360 return (1); 9361 sum = in6_cksum(m, p, off, len); 9362 break; 9363 #endif /* INET6 */ 9364 } 9365 } 9366 if (sum) { 9367 switch (p) { 9368 case IPPROTO_TCP: 9369 { 9370 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9371 break; 9372 } 9373 case IPPROTO_UDP: 9374 { 9375 KMOD_UDPSTAT_INC(udps_badsum); 9376 break; 9377 } 9378 #ifdef INET 9379 case IPPROTO_ICMP: 9380 { 9381 KMOD_ICMPSTAT_INC(icps_checksum); 9382 break; 9383 } 9384 #endif 9385 #ifdef INET6 9386 case IPPROTO_ICMPV6: 9387 { 9388 KMOD_ICMP6STAT_INC(icp6s_checksum); 9389 break; 9390 } 9391 #endif /* INET6 */ 9392 } 9393 return (1); 9394 } else { 9395 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9396 m->m_pkthdr.csum_flags |= 9397 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9398 m->m_pkthdr.csum_data = 0xffff; 9399 } 9400 } 9401 return (0); 9402 } 9403 9404 static bool 9405 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9406 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9407 { 9408 int dndir = r->direction; 9409 9410 if (s && dndir == PF_INOUT) { 9411 dndir = s->direction; 9412 } else if (dndir == PF_INOUT) { 9413 /* Assume primary direction. Happens when we've set dnpipe in 9414 * the ethernet level code. */ 9415 dndir = pd->dir; 9416 } 9417 9418 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9419 return (false); 9420 9421 memset(dnflow, 0, sizeof(*dnflow)); 9422 9423 if (pd->dport != NULL) 9424 dnflow->f_id.dst_port = ntohs(*pd->dport); 9425 if (pd->sport != NULL) 9426 dnflow->f_id.src_port = ntohs(*pd->sport); 9427 9428 if (pd->dir == PF_IN) 9429 dnflow->flags |= IPFW_ARGS_IN; 9430 else 9431 dnflow->flags |= IPFW_ARGS_OUT; 9432 9433 if (pd->dir != dndir && pd->act.dnrpipe) { 9434 dnflow->rule.info = pd->act.dnrpipe; 9435 } 9436 else if (pd->dir == dndir && pd->act.dnpipe) { 9437 dnflow->rule.info = pd->act.dnpipe; 9438 } 9439 else { 9440 return (false); 9441 } 9442 9443 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9444 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9445 dnflow->rule.info |= IPFW_IS_PIPE; 9446 9447 dnflow->f_id.proto = pd->proto; 9448 dnflow->f_id.extra = dnflow->rule.info; 9449 switch (pd->naf) { 9450 case AF_INET: 9451 dnflow->f_id.addr_type = 4; 9452 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9453 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9454 break; 9455 case AF_INET6: 9456 dnflow->flags |= IPFW_ARGS_IP6; 9457 dnflow->f_id.addr_type = 6; 9458 dnflow->f_id.src_ip6 = pd->src->v6; 9459 dnflow->f_id.dst_ip6 = pd->dst->v6; 9460 break; 9461 } 9462 9463 return (true); 9464 } 9465 9466 int 9467 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9468 struct inpcb *inp) 9469 { 9470 struct pfi_kkif *kif; 9471 struct mbuf *m = *m0; 9472 9473 M_ASSERTPKTHDR(m); 9474 MPASS(ifp->if_vnet == curvnet); 9475 NET_EPOCH_ASSERT(); 9476 9477 if (!V_pf_status.running) 9478 return (PF_PASS); 9479 9480 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9481 9482 if (kif == NULL) { 9483 DPFPRINTF(PF_DEBUG_URGENT, 9484 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 9485 return (PF_DROP); 9486 } 9487 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9488 return (PF_PASS); 9489 9490 if (m->m_flags & M_SKIP_FIREWALL) 9491 return (PF_PASS); 9492 9493 if (__predict_false(! M_WRITABLE(*m0))) { 9494 m = *m0 = m_unshare(*m0, M_NOWAIT); 9495 if (*m0 == NULL) 9496 return (PF_DROP); 9497 } 9498 9499 /* Stateless! */ 9500 return (pf_test_eth_rule(dir, kif, m0)); 9501 } 9502 9503 static __inline void 9504 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9505 { 9506 struct m_tag *mtag; 9507 9508 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9509 9510 /* dummynet adds this tag, but pf does not need it, 9511 * and keeping it creates unexpected behavior, 9512 * e.g. in case of divert(4) usage right after dummynet. */ 9513 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9514 if (mtag != NULL) 9515 m_tag_delete(m, mtag); 9516 } 9517 9518 static int 9519 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9520 struct pf_krule *r, struct mbuf **m0) 9521 { 9522 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9523 } 9524 9525 static int 9526 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9527 struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa, 9528 struct mbuf **m0) 9529 { 9530 struct ip_fw_args dnflow; 9531 9532 NET_EPOCH_ASSERT(); 9533 9534 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9535 return (0); 9536 9537 if (ip_dn_io_ptr == NULL) { 9538 m_freem(*m0); 9539 *m0 = NULL; 9540 return (ENOMEM); 9541 } 9542 9543 if (pd->pf_mtag == NULL && 9544 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 9545 m_freem(*m0); 9546 *m0 = NULL; 9547 return (ENOMEM); 9548 } 9549 9550 if (ifp != NULL) { 9551 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 9552 9553 pd->pf_mtag->if_index = ifp->if_index; 9554 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 9555 9556 MPASS(sa != NULL); 9557 9558 switch (sa->sa_family) { 9559 case AF_INET: 9560 memcpy(&pd->pf_mtag->dst, sa, 9561 sizeof(struct sockaddr_in)); 9562 break; 9563 case AF_INET6: 9564 memcpy(&pd->pf_mtag->dst, sa, 9565 sizeof(struct sockaddr_in6)); 9566 break; 9567 } 9568 } 9569 9570 if (s != NULL && s->nat_rule != NULL && 9571 s->nat_rule->action == PF_RDR && 9572 ( 9573 #ifdef INET 9574 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 9575 #endif /* INET */ 9576 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 9577 /* 9578 * If we're redirecting to loopback mark this packet 9579 * as being local. Otherwise it might get dropped 9580 * if dummynet re-injects. 9581 */ 9582 (*m0)->m_pkthdr.rcvif = V_loif; 9583 } 9584 9585 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 9586 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 9587 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 9588 ip_dn_io_ptr(m0, &dnflow); 9589 if (*m0 != NULL) { 9590 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9591 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 9592 } 9593 } 9594 9595 return (0); 9596 } 9597 9598 #ifdef INET6 9599 static int 9600 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 9601 u_short *reason) 9602 { 9603 struct ip6_opt opt; 9604 struct ip6_opt_jumbo jumbo; 9605 9606 while (off < end) { 9607 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 9608 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 9609 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 9610 return (PF_DROP); 9611 } 9612 if (opt.ip6o_type == IP6OPT_PAD1) { 9613 off++; 9614 continue; 9615 } 9616 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL, 9617 reason, AF_INET6)) { 9618 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 9619 return (PF_DROP); 9620 } 9621 if (off + sizeof(opt) + opt.ip6o_len > end) { 9622 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 9623 REASON_SET(reason, PFRES_IPOPTIONS); 9624 return (PF_DROP); 9625 } 9626 switch (opt.ip6o_type) { 9627 case IP6OPT_JUMBO: 9628 if (pd->jumbolen != 0) { 9629 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 9630 REASON_SET(reason, PFRES_IPOPTIONS); 9631 return (PF_DROP); 9632 } 9633 if (ntohs(h->ip6_plen) != 0) { 9634 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 9635 REASON_SET(reason, PFRES_IPOPTIONS); 9636 return (PF_DROP); 9637 } 9638 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL, 9639 reason, AF_INET6)) { 9640 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 9641 return (PF_DROP); 9642 } 9643 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 9644 sizeof(pd->jumbolen)); 9645 pd->jumbolen = ntohl(pd->jumbolen); 9646 if (pd->jumbolen < IPV6_MAXPACKET) { 9647 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 9648 REASON_SET(reason, PFRES_IPOPTIONS); 9649 return (PF_DROP); 9650 } 9651 break; 9652 default: 9653 break; 9654 } 9655 off += sizeof(opt) + opt.ip6o_len; 9656 } 9657 9658 return (PF_PASS); 9659 } 9660 9661 int 9662 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 9663 { 9664 struct ip6_frag frag; 9665 struct ip6_ext ext; 9666 struct ip6_rthdr rthdr; 9667 uint32_t end; 9668 int fraghdr_cnt = 0, rthdr_cnt = 0; 9669 9670 pd->off += sizeof(struct ip6_hdr); 9671 end = pd->off + ntohs(h->ip6_plen); 9672 pd->fragoff = pd->extoff = pd->jumbolen = 0; 9673 pd->proto = h->ip6_nxt; 9674 for (;;) { 9675 switch (pd->proto) { 9676 case IPPROTO_FRAGMENT: 9677 if (fraghdr_cnt++) { 9678 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 9679 REASON_SET(reason, PFRES_FRAG); 9680 return (PF_DROP); 9681 } 9682 /* jumbo payload packets cannot be fragmented */ 9683 if (pd->jumbolen != 0) { 9684 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 9685 REASON_SET(reason, PFRES_FRAG); 9686 return (PF_DROP); 9687 } 9688 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 9689 NULL, reason, AF_INET6)) { 9690 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 9691 return (PF_DROP); 9692 } 9693 /* stop walking over non initial fragments */ 9694 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 9695 pd->fragoff = pd->off; 9696 return (PF_PASS); 9697 } 9698 /* RFC6946: reassemble only non atomic fragments */ 9699 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 9700 pd->fragoff = pd->off; 9701 pd->off += sizeof(frag); 9702 pd->proto = frag.ip6f_nxt; 9703 break; 9704 case IPPROTO_ROUTING: 9705 if (rthdr_cnt++) { 9706 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 9707 REASON_SET(reason, PFRES_IPOPTIONS); 9708 return (PF_DROP); 9709 } 9710 /* fragments may be short */ 9711 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 9712 pd->off = pd->fragoff; 9713 pd->proto = IPPROTO_FRAGMENT; 9714 return (PF_PASS); 9715 } 9716 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 9717 NULL, reason, AF_INET6)) { 9718 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 9719 return (PF_DROP); 9720 } 9721 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9722 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 9723 REASON_SET(reason, PFRES_IPOPTIONS); 9724 return (PF_DROP); 9725 } 9726 /* FALLTHROUGH */ 9727 case IPPROTO_AH: 9728 case IPPROTO_HOPOPTS: 9729 case IPPROTO_DSTOPTS: 9730 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9731 NULL, reason, AF_INET6)) { 9732 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 9733 return (PF_DROP); 9734 } 9735 /* fragments may be short */ 9736 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 9737 pd->off = pd->fragoff; 9738 pd->proto = IPPROTO_FRAGMENT; 9739 return (PF_PASS); 9740 } 9741 /* reassembly needs the ext header before the frag */ 9742 if (pd->fragoff == 0) 9743 pd->extoff = pd->off; 9744 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0) { 9745 if (pf_walk_option6(pd, h, 9746 pd->off + sizeof(ext), 9747 pd->off + (ext.ip6e_len + 1) * 8, reason) 9748 != PF_PASS) 9749 return (PF_DROP); 9750 if (ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 9751 DPFPRINTF(PF_DEBUG_MISC, 9752 ("IPv6 missing jumbo")); 9753 REASON_SET(reason, PFRES_IPOPTIONS); 9754 return (PF_DROP); 9755 } 9756 } 9757 if (pd->proto == IPPROTO_AH) 9758 pd->off += (ext.ip6e_len + 2) * 4; 9759 else 9760 pd->off += (ext.ip6e_len + 1) * 8; 9761 pd->proto = ext.ip6e_nxt; 9762 break; 9763 case IPPROTO_TCP: 9764 case IPPROTO_UDP: 9765 case IPPROTO_SCTP: 9766 case IPPROTO_ICMPV6: 9767 /* fragments may be short, ignore inner header then */ 9768 if (pd->fragoff != 0 && end < pd->off + 9769 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 9770 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 9771 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 9772 sizeof(struct icmp6_hdr))) { 9773 pd->off = pd->fragoff; 9774 pd->proto = IPPROTO_FRAGMENT; 9775 } 9776 /* FALLTHROUGH */ 9777 default: 9778 return (PF_PASS); 9779 } 9780 } 9781 } 9782 #endif /* INET6 */ 9783 9784 static void 9785 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 9786 { 9787 memset(pd, 0, sizeof(*pd)); 9788 pd->pf_mtag = pf_find_mtag(m); 9789 pd->m = m; 9790 } 9791 9792 static int 9793 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 9794 u_short *action, u_short *reason, struct pfi_kkif *kif, 9795 struct pf_rule_actions *default_actions) 9796 { 9797 pd->dir = dir; 9798 pd->kif = kif; 9799 pd->m = *m0; 9800 pd->sidx = (dir == PF_IN) ? 0 : 1; 9801 pd->didx = (dir == PF_IN) ? 1 : 0; 9802 pd->af = pd->naf = af; 9803 9804 TAILQ_INIT(&pd->sctp_multihome_jobs); 9805 if (default_actions != NULL) 9806 memcpy(&pd->act, default_actions, sizeof(pd->act)); 9807 9808 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 9809 pd->act.dnpipe = pd->pf_mtag->dnpipe; 9810 pd->act.flags = pd->pf_mtag->dnflags; 9811 } 9812 9813 switch (af) { 9814 #ifdef INET 9815 case AF_INET: { 9816 struct ip *h; 9817 9818 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 9819 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 9820 DPFPRINTF(PF_DEBUG_URGENT, 9821 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 9822 *action = PF_DROP; 9823 REASON_SET(reason, PFRES_SHORT); 9824 return (-1); 9825 } 9826 9827 if (pf_normalize_ip(reason, pd) != PF_PASS) { 9828 /* We do IP header normalization and packet reassembly here */ 9829 *m0 = pd->m; 9830 *action = PF_DROP; 9831 return (-1); 9832 } 9833 *m0 = pd->m; 9834 9835 h = mtod(pd->m, struct ip *); 9836 pd->off = h->ip_hl << 2; 9837 if (pd->off < (int)sizeof(*h)) { 9838 *action = PF_DROP; 9839 REASON_SET(reason, PFRES_SHORT); 9840 return (-1); 9841 } 9842 pd->src = (struct pf_addr *)&h->ip_src; 9843 pd->dst = (struct pf_addr *)&h->ip_dst; 9844 PF_ACPY(&pd->osrc, pd->src, af); 9845 PF_ACPY(&pd->odst, pd->dst, af); 9846 pd->ip_sum = &h->ip_sum; 9847 pd->virtual_proto = pd->proto = h->ip_p; 9848 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 9849 pd->ttl = h->ip_ttl; 9850 pd->tot_len = ntohs(h->ip_len); 9851 pd->act.rtableid = -1; 9852 pd->df = h->ip_off & htons(IP_DF); 9853 9854 if (h->ip_hl > 5) /* has options */ 9855 pd->badopts++; 9856 9857 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 9858 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9859 9860 break; 9861 } 9862 #endif /* INET */ 9863 #ifdef INET6 9864 case AF_INET6: { 9865 struct ip6_hdr *h; 9866 9867 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 9868 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 9869 DPFPRINTF(PF_DEBUG_URGENT, 9870 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 9871 ", pullup failed\n")); 9872 *action = PF_DROP; 9873 REASON_SET(reason, PFRES_SHORT); 9874 return (-1); 9875 } 9876 9877 h = mtod(pd->m, struct ip6_hdr *); 9878 pd->off = 0; 9879 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9880 *action = PF_DROP; 9881 return (-1); 9882 } 9883 9884 h = mtod(pd->m, struct ip6_hdr *); 9885 pd->src = (struct pf_addr *)&h->ip6_src; 9886 pd->dst = (struct pf_addr *)&h->ip6_dst; 9887 PF_ACPY(&pd->osrc, pd->src, af); 9888 PF_ACPY(&pd->odst, pd->dst, af); 9889 pd->ip_sum = NULL; 9890 pd->tos = IPV6_DSCP(h); 9891 pd->ttl = h->ip6_hlim; 9892 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 9893 pd->virtual_proto = pd->proto = h->ip6_nxt; 9894 pd->act.rtableid = -1; 9895 9896 if (pd->fragoff != 0) 9897 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9898 9899 /* 9900 * we do not support jumbogram. if we keep going, zero ip6_plen 9901 * will do something bad, so drop the packet for now. 9902 */ 9903 if (htons(h->ip6_plen) == 0) { 9904 *action = PF_DROP; 9905 return (-1); 9906 } 9907 9908 /* We do IP header normalization and packet reassembly here */ 9909 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 9910 PF_PASS) { 9911 *m0 = pd->m; 9912 *action = PF_DROP; 9913 return (-1); 9914 } 9915 *m0 = pd->m; 9916 if (pd->m == NULL) { 9917 /* packet sits in reassembly queue, no error */ 9918 *action = PF_PASS; 9919 return (-1); 9920 } 9921 9922 /* Update pointers into the packet. */ 9923 h = mtod(pd->m, struct ip6_hdr *); 9924 pd->src = (struct pf_addr *)&h->ip6_src; 9925 pd->dst = (struct pf_addr *)&h->ip6_dst; 9926 9927 pd->off = 0; 9928 9929 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9930 *action = PF_DROP; 9931 return (-1); 9932 } 9933 9934 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 9935 /* 9936 * Reassembly may have changed the next protocol from 9937 * fragment to something else, so update. 9938 */ 9939 pd->virtual_proto = pd->proto; 9940 MPASS(pd->fragoff == 0); 9941 } 9942 9943 if (pd->fragoff != 0) 9944 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9945 9946 break; 9947 } 9948 #endif /* INET6 */ 9949 default: 9950 panic("pf_setup_pdesc called with illegal af %u", af); 9951 } 9952 9953 switch (pd->virtual_proto) { 9954 case IPPROTO_TCP: { 9955 struct tcphdr *th = &pd->hdr.tcp; 9956 9957 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 9958 reason, af)) { 9959 *action = PF_DROP; 9960 REASON_SET(reason, PFRES_SHORT); 9961 return (-1); 9962 } 9963 pd->hdrlen = sizeof(*th); 9964 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 9965 pd->sport = &th->th_sport; 9966 pd->dport = &th->th_dport; 9967 pd->pcksum = &th->th_sum; 9968 break; 9969 } 9970 case IPPROTO_UDP: { 9971 struct udphdr *uh = &pd->hdr.udp; 9972 9973 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 9974 reason, af)) { 9975 *action = PF_DROP; 9976 REASON_SET(reason, PFRES_SHORT); 9977 return (-1); 9978 } 9979 pd->hdrlen = sizeof(*uh); 9980 if (uh->uh_dport == 0 || 9981 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 9982 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 9983 *action = PF_DROP; 9984 REASON_SET(reason, PFRES_SHORT); 9985 return (-1); 9986 } 9987 pd->sport = &uh->uh_sport; 9988 pd->dport = &uh->uh_dport; 9989 pd->pcksum = &uh->uh_sum; 9990 break; 9991 } 9992 case IPPROTO_SCTP: { 9993 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 9994 action, reason, af)) { 9995 *action = PF_DROP; 9996 REASON_SET(reason, PFRES_SHORT); 9997 return (-1); 9998 } 9999 pd->hdrlen = sizeof(pd->hdr.sctp); 10000 pd->p_len = pd->tot_len - pd->off; 10001 10002 pd->sport = &pd->hdr.sctp.src_port; 10003 pd->dport = &pd->hdr.sctp.dest_port; 10004 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 10005 *action = PF_DROP; 10006 REASON_SET(reason, PFRES_SHORT); 10007 return (-1); 10008 } 10009 if (pf_scan_sctp(pd) != PF_PASS) { 10010 *action = PF_DROP; 10011 REASON_SET(reason, PFRES_SHORT); 10012 return (-1); 10013 } 10014 /* 10015 * Placeholder. The SCTP checksum is 32-bits, but 10016 * pf_test_state() expects to update a 16-bit checksum. 10017 * Provide a dummy value which we'll subsequently ignore. 10018 */ 10019 pd->pcksum = &pd->sctp_dummy_sum; 10020 break; 10021 } 10022 case IPPROTO_ICMP: { 10023 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 10024 action, reason, af)) { 10025 *action = PF_DROP; 10026 REASON_SET(reason, PFRES_SHORT); 10027 return (-1); 10028 } 10029 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 10030 pd->hdrlen = ICMP_MINLEN; 10031 break; 10032 } 10033 #ifdef INET6 10034 case IPPROTO_ICMPV6: { 10035 size_t icmp_hlen = sizeof(struct icmp6_hdr); 10036 10037 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10038 action, reason, af)) { 10039 *action = PF_DROP; 10040 REASON_SET(reason, PFRES_SHORT); 10041 return (-1); 10042 } 10043 /* ICMP headers we look further into to match state */ 10044 switch (pd->hdr.icmp6.icmp6_type) { 10045 case MLD_LISTENER_QUERY: 10046 case MLD_LISTENER_REPORT: 10047 icmp_hlen = sizeof(struct mld_hdr); 10048 break; 10049 case ND_NEIGHBOR_SOLICIT: 10050 case ND_NEIGHBOR_ADVERT: 10051 icmp_hlen = sizeof(struct nd_neighbor_solicit); 10052 break; 10053 } 10054 if (icmp_hlen > sizeof(struct icmp6_hdr) && 10055 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10056 action, reason, af)) { 10057 *action = PF_DROP; 10058 REASON_SET(reason, PFRES_SHORT); 10059 return (-1); 10060 } 10061 pd->hdrlen = icmp_hlen; 10062 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 10063 break; 10064 } 10065 #endif /* INET6 */ 10066 } 10067 10068 if (pd->sport) 10069 pd->osport = pd->nsport = *pd->sport; 10070 if (pd->dport) 10071 pd->odport = pd->ndport = *pd->dport; 10072 10073 return (0); 10074 } 10075 10076 static void 10077 pf_counters_inc(int action, struct pf_pdesc *pd, 10078 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 10079 { 10080 struct pf_krule *tr; 10081 int dir = pd->dir; 10082 int dirndx; 10083 10084 pf_counter_u64_critical_enter(); 10085 pf_counter_u64_add_protected( 10086 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10087 pd->tot_len); 10088 pf_counter_u64_add_protected( 10089 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10090 1); 10091 10092 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 10093 dirndx = (dir == PF_OUT); 10094 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 10095 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 10096 pf_update_timestamp(r); 10097 10098 if (a != NULL) { 10099 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 10100 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 10101 } 10102 if (s != NULL) { 10103 struct pf_krule_item *ri; 10104 10105 if (s->nat_rule != NULL) { 10106 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 10107 1); 10108 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 10109 pd->tot_len); 10110 } 10111 /* 10112 * Source nodes are accessed unlocked here. 10113 * But since we are operating with stateful tracking 10114 * and the state is locked, those SNs could not have 10115 * been freed. 10116 */ 10117 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 10118 if (s->sns[sn_type] != NULL) { 10119 counter_u64_add( 10120 s->sns[sn_type]->packets[dirndx], 10121 1); 10122 counter_u64_add( 10123 s->sns[sn_type]->bytes[dirndx], 10124 pd->tot_len); 10125 } 10126 } 10127 dirndx = (dir == s->direction) ? 0 : 1; 10128 s->packets[dirndx]++; 10129 s->bytes[dirndx] += pd->tot_len; 10130 10131 SLIST_FOREACH(ri, &s->match_rules, entry) { 10132 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 10133 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 10134 10135 if (ri->r->src.addr.type == PF_ADDR_TABLE) 10136 pfr_update_stats(ri->r->src.addr.p.tbl, 10137 (s == NULL) ? pd->src : 10138 &s->key[(s->direction == PF_IN)]-> 10139 addr[(s->direction == PF_OUT)], 10140 pd->af, pd->tot_len, dir == PF_OUT, 10141 r->action == PF_PASS, ri->r->src.neg); 10142 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 10143 pfr_update_stats(ri->r->dst.addr.p.tbl, 10144 (s == NULL) ? pd->dst : 10145 &s->key[(s->direction == PF_IN)]-> 10146 addr[(s->direction == PF_IN)], 10147 pd->af, pd->tot_len, dir == PF_OUT, 10148 r->action == PF_PASS, ri->r->dst.neg); 10149 } 10150 } 10151 10152 tr = r; 10153 if (s != NULL && s->nat_rule != NULL && 10154 r == &V_pf_default_rule) 10155 tr = s->nat_rule; 10156 10157 if (tr->src.addr.type == PF_ADDR_TABLE) 10158 pfr_update_stats(tr->src.addr.p.tbl, 10159 (s == NULL) ? pd->src : 10160 &s->key[(s->direction == PF_IN)]-> 10161 addr[(s->direction == PF_OUT)], 10162 pd->af, pd->tot_len, dir == PF_OUT, 10163 r->action == PF_PASS, tr->src.neg); 10164 if (tr->dst.addr.type == PF_ADDR_TABLE) 10165 pfr_update_stats(tr->dst.addr.p.tbl, 10166 (s == NULL) ? pd->dst : 10167 &s->key[(s->direction == PF_IN)]-> 10168 addr[(s->direction == PF_IN)], 10169 pd->af, pd->tot_len, dir == PF_OUT, 10170 r->action == PF_PASS, tr->dst.neg); 10171 } 10172 pf_counter_u64_critical_exit(); 10173 } 10174 static void 10175 pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm, 10176 struct pf_krule *am, struct pf_kruleset *ruleset, 10177 struct pf_krule_slist *matchrules) 10178 { 10179 struct pf_krule_item *ri; 10180 10181 /* if this is the log(matches) rule, packet has been logged already */ 10182 if (rm->log & PF_LOG_MATCHES) 10183 return; 10184 10185 SLIST_FOREACH(ri, matchrules, entry) 10186 if (ri->r->log & PF_LOG_MATCHES) 10187 PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am, 10188 ruleset, pd, 1, ri->r); 10189 } 10190 10191 #if defined(INET) || defined(INET6) 10192 int 10193 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10194 struct inpcb *inp, struct pf_rule_actions *default_actions) 10195 { 10196 struct pfi_kkif *kif; 10197 u_short action, reason = 0; 10198 struct m_tag *mtag; 10199 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10200 struct pf_kstate *s = NULL; 10201 struct pf_kruleset *ruleset = NULL; 10202 struct pf_pdesc pd; 10203 int use_2nd_queue = 0; 10204 uint16_t tag; 10205 10206 PF_RULES_RLOCK_TRACKER; 10207 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10208 M_ASSERTPKTHDR(*m0); 10209 10210 if (!V_pf_status.running) 10211 return (PF_PASS); 10212 10213 PF_RULES_RLOCK(); 10214 10215 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10216 10217 if (__predict_false(kif == NULL)) { 10218 DPFPRINTF(PF_DEBUG_URGENT, 10219 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 10220 PF_RULES_RUNLOCK(); 10221 return (PF_DROP); 10222 } 10223 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10224 PF_RULES_RUNLOCK(); 10225 return (PF_PASS); 10226 } 10227 10228 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10229 PF_RULES_RUNLOCK(); 10230 return (PF_PASS); 10231 } 10232 10233 if (__predict_false(! M_WRITABLE(*m0))) { 10234 *m0 = m_unshare(*m0, M_NOWAIT); 10235 if (*m0 == NULL) { 10236 PF_RULES_RUNLOCK(); 10237 return (PF_DROP); 10238 } 10239 } 10240 10241 pf_init_pdesc(&pd, *m0); 10242 10243 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10244 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10245 10246 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10247 pd.pf_mtag->if_idxgen); 10248 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10249 PF_RULES_RUNLOCK(); 10250 m_freem(*m0); 10251 *m0 = NULL; 10252 return (PF_PASS); 10253 } 10254 PF_RULES_RUNLOCK(); 10255 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10256 *m0 = NULL; 10257 return (PF_PASS); 10258 } 10259 10260 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10261 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10262 /* Dummynet re-injects packets after they've 10263 * completed their delay. We've already 10264 * processed them, so pass unconditionally. */ 10265 10266 /* But only once. We may see the packet multiple times (e.g. 10267 * PFIL_IN/PFIL_OUT). */ 10268 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 10269 PF_RULES_RUNLOCK(); 10270 10271 return (PF_PASS); 10272 } 10273 10274 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 10275 kif, default_actions) == -1) { 10276 if (action != PF_PASS) 10277 pd.act.log |= PF_LOG_FORCE; 10278 goto done; 10279 } 10280 10281 #ifdef INET 10282 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 10283 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 10284 PF_RULES_RUNLOCK(); 10285 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 10286 0, ifp->if_mtu); 10287 *m0 = NULL; 10288 return (PF_DROP); 10289 } 10290 #endif /* INET */ 10291 #ifdef INET6 10292 /* 10293 * If we end up changing IP addresses (e.g. binat) the stack may get 10294 * confused and fail to send the icmp6 packet too big error. Just send 10295 * it here, before we do any NAT. 10296 */ 10297 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 10298 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 10299 PF_RULES_RUNLOCK(); 10300 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 10301 *m0 = NULL; 10302 return (PF_DROP); 10303 } 10304 #endif /* INET6 */ 10305 10306 if (__predict_false(ip_divert_ptr != NULL) && 10307 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 10308 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 10309 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 10310 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 10311 if (pd.pf_mtag == NULL && 10312 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10313 action = PF_DROP; 10314 goto done; 10315 } 10316 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 10317 } 10318 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 10319 pd.m->m_flags |= M_FASTFWD_OURS; 10320 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10321 } 10322 m_tag_delete(pd.m, mtag); 10323 10324 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 10325 if (mtag != NULL) 10326 m_tag_delete(pd.m, mtag); 10327 } 10328 10329 switch (pd.virtual_proto) { 10330 case PF_VPROTO_FRAGMENT: 10331 /* 10332 * handle fragments that aren't reassembled by 10333 * normalization 10334 */ 10335 if (kif == NULL || r == NULL) /* pflog */ 10336 action = PF_DROP; 10337 else 10338 action = pf_test_rule(&r, &s, &pd, &a, 10339 &ruleset, &reason, inp); 10340 if (action != PF_PASS) 10341 REASON_SET(&reason, PFRES_FRAG); 10342 break; 10343 10344 case IPPROTO_TCP: { 10345 /* Respond to SYN with a syncookie. */ 10346 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 10347 pd.dir == PF_IN && pf_synflood_check(&pd)) { 10348 pf_syncookie_send(&pd); 10349 action = PF_DROP; 10350 break; 10351 } 10352 10353 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 10354 use_2nd_queue = 1; 10355 action = pf_normalize_tcp(&pd); 10356 if (action == PF_DROP) 10357 goto done; 10358 action = pf_test_state(&s, &pd, &reason); 10359 if (action == PF_PASS || action == PF_AFRT) { 10360 if (V_pfsync_update_state_ptr != NULL) 10361 V_pfsync_update_state_ptr(s); 10362 r = s->rule; 10363 a = s->anchor; 10364 } else if (s == NULL) { 10365 /* Validate remote SYN|ACK, re-create original SYN if 10366 * valid. */ 10367 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 10368 TH_ACK && pf_syncookie_validate(&pd) && 10369 pd.dir == PF_IN) { 10370 struct mbuf *msyn; 10371 10372 msyn = pf_syncookie_recreate_syn(&pd); 10373 if (msyn == NULL) { 10374 action = PF_DROP; 10375 break; 10376 } 10377 10378 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 10379 &pd.act); 10380 m_freem(msyn); 10381 if (action != PF_PASS) 10382 break; 10383 10384 action = pf_test_state(&s, &pd, &reason); 10385 if (action != PF_PASS || s == NULL) { 10386 action = PF_DROP; 10387 break; 10388 } 10389 10390 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 10391 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 10392 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 10393 action = pf_synproxy(&pd, s, &reason); 10394 break; 10395 } else { 10396 action = pf_test_rule(&r, &s, &pd, 10397 &a, &ruleset, &reason, inp); 10398 } 10399 } 10400 break; 10401 } 10402 10403 case IPPROTO_SCTP: 10404 action = pf_normalize_sctp(&pd); 10405 if (action == PF_DROP) 10406 goto done; 10407 /* fallthrough */ 10408 case IPPROTO_UDP: 10409 default: 10410 action = pf_test_state(&s, &pd, &reason); 10411 if (action == PF_PASS || action == PF_AFRT) { 10412 if (V_pfsync_update_state_ptr != NULL) 10413 V_pfsync_update_state_ptr(s); 10414 r = s->rule; 10415 a = s->anchor; 10416 } else if (s == NULL) { 10417 action = pf_test_rule(&r, &s, 10418 &pd, &a, &ruleset, &reason, inp); 10419 } 10420 break; 10421 10422 case IPPROTO_ICMP: 10423 case IPPROTO_ICMPV6: { 10424 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 10425 action = PF_DROP; 10426 REASON_SET(&reason, PFRES_NORM); 10427 DPFPRINTF(PF_DEBUG_MISC, 10428 ("dropping IPv6 packet with ICMPv4 payload")); 10429 goto done; 10430 } 10431 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 10432 action = PF_DROP; 10433 REASON_SET(&reason, PFRES_NORM); 10434 DPFPRINTF(PF_DEBUG_MISC, 10435 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 10436 goto done; 10437 } 10438 action = pf_test_state_icmp(&s, &pd, &reason); 10439 if (action == PF_PASS || action == PF_AFRT) { 10440 if (V_pfsync_update_state_ptr != NULL) 10441 V_pfsync_update_state_ptr(s); 10442 r = s->rule; 10443 a = s->anchor; 10444 } else if (s == NULL) 10445 action = pf_test_rule(&r, &s, &pd, 10446 &a, &ruleset, &reason, inp); 10447 break; 10448 } 10449 10450 } 10451 10452 done: 10453 PF_RULES_RUNLOCK(); 10454 10455 if (pd.m == NULL) 10456 goto eat_pkt; 10457 10458 if (action == PF_PASS && pd.badopts && 10459 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || pd.act.allow_opts)) { 10460 action = PF_DROP; 10461 REASON_SET(&reason, PFRES_IPOPTIONS); 10462 pd.act.log = PF_LOG_FORCE; 10463 DPFPRINTF(PF_DEBUG_MISC, 10464 ("pf: dropping packet with dangerous headers\n")); 10465 } 10466 10467 if (s) { 10468 uint8_t log = pd.act.log; 10469 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 10470 pd.act.log |= log; 10471 tag = s->tag; 10472 } else { 10473 tag = r->tag; 10474 } 10475 10476 if (tag > 0 && pf_tag_packet(&pd, tag)) { 10477 action = PF_DROP; 10478 REASON_SET(&reason, PFRES_MEMORY); 10479 } 10480 10481 pf_scrub(&pd); 10482 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 10483 pf_normalize_mss(&pd); 10484 10485 if (pd.act.rtableid >= 0) 10486 M_SETFIB(pd.m, pd.act.rtableid); 10487 10488 if (pd.act.flags & PFSTATE_SETPRIO) { 10489 if (pd.tos & IPTOS_LOWDELAY) 10490 use_2nd_queue = 1; 10491 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 10492 action = PF_DROP; 10493 REASON_SET(&reason, PFRES_MEMORY); 10494 pd.act.log = PF_LOG_FORCE; 10495 DPFPRINTF(PF_DEBUG_MISC, 10496 ("pf: failed to allocate 802.1q mtag\n")); 10497 } 10498 } 10499 10500 #ifdef ALTQ 10501 if (action == PF_PASS && pd.act.qid) { 10502 if (pd.pf_mtag == NULL && 10503 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10504 action = PF_DROP; 10505 REASON_SET(&reason, PFRES_MEMORY); 10506 } else { 10507 if (s != NULL) 10508 pd.pf_mtag->qid_hash = pf_state_hash(s); 10509 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 10510 pd.pf_mtag->qid = pd.act.pqid; 10511 else 10512 pd.pf_mtag->qid = pd.act.qid; 10513 /* Add hints for ecn. */ 10514 pd.pf_mtag->hdr = mtod(pd.m, void *); 10515 } 10516 } 10517 #endif /* ALTQ */ 10518 10519 /* 10520 * connections redirected to loopback should not match sockets 10521 * bound specifically to loopback due to security implications, 10522 * see tcp_input() and in_pcblookup_listen(). 10523 */ 10524 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 10525 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 10526 (s->nat_rule->action == PF_RDR || 10527 s->nat_rule->action == PF_BINAT) && 10528 pf_is_loopback(af, pd.dst)) 10529 pd.m->m_flags |= M_SKIP_FIREWALL; 10530 10531 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 10532 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 10533 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 10534 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 10535 if (mtag != NULL) { 10536 ((struct pf_divert_mtag *)(mtag+1))->port = 10537 ntohs(r->divert.port); 10538 ((struct pf_divert_mtag *)(mtag+1))->idir = 10539 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 10540 PF_DIVERT_MTAG_DIR_OUT; 10541 10542 if (s) 10543 PF_STATE_UNLOCK(s); 10544 10545 m_tag_prepend(pd.m, mtag); 10546 if (pd.m->m_flags & M_FASTFWD_OURS) { 10547 if (pd.pf_mtag == NULL && 10548 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10549 action = PF_DROP; 10550 REASON_SET(&reason, PFRES_MEMORY); 10551 pd.act.log = PF_LOG_FORCE; 10552 DPFPRINTF(PF_DEBUG_MISC, 10553 ("pf: failed to allocate tag\n")); 10554 } else { 10555 pd.pf_mtag->flags |= 10556 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10557 pd.m->m_flags &= ~M_FASTFWD_OURS; 10558 } 10559 } 10560 ip_divert_ptr(*m0, dir == PF_IN); 10561 *m0 = NULL; 10562 10563 return (action); 10564 } else { 10565 /* XXX: ipfw has the same behaviour! */ 10566 action = PF_DROP; 10567 REASON_SET(&reason, PFRES_MEMORY); 10568 pd.act.log = PF_LOG_FORCE; 10569 DPFPRINTF(PF_DEBUG_MISC, 10570 ("pf: failed to allocate divert tag\n")); 10571 } 10572 } 10573 /* XXX: Anybody working on it?! */ 10574 if (af == AF_INET6 && r->divert.port) 10575 printf("pf: divert(9) is not supported for IPv6\n"); 10576 10577 /* this flag will need revising if the pkt is forwarded */ 10578 if (pd.pf_mtag) 10579 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 10580 10581 if (pd.act.log) { 10582 struct pf_krule *lr; 10583 struct pf_krule_item *ri; 10584 10585 if (s != NULL && s->nat_rule != NULL && 10586 s->nat_rule->log & PF_LOG_ALL) 10587 lr = s->nat_rule; 10588 else 10589 lr = r; 10590 10591 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 10592 PFLOG_PACKET(action, reason, lr, a, 10593 ruleset, &pd, (s == NULL), NULL); 10594 if (s) { 10595 SLIST_FOREACH(ri, &s->match_rules, entry) 10596 if (ri->r->log & PF_LOG_ALL) 10597 PFLOG_PACKET(action, 10598 reason, ri->r, a, ruleset, &pd, 0, NULL); 10599 } 10600 } 10601 10602 pf_counters_inc(action, &pd, s, r, a); 10603 10604 switch (action) { 10605 case PF_SYNPROXY_DROP: 10606 m_freem(*m0); 10607 case PF_DEFER: 10608 *m0 = NULL; 10609 action = PF_PASS; 10610 break; 10611 case PF_DROP: 10612 m_freem(*m0); 10613 *m0 = NULL; 10614 break; 10615 case PF_AFRT: 10616 if (pf_translate_af(&pd)) { 10617 if (!pd.m) 10618 *m0 = NULL; 10619 action = PF_DROP; 10620 break; 10621 } 10622 *m0 = pd.m; /* pf_translate_af may change pd.m */ 10623 #ifdef INET 10624 if (pd.naf == AF_INET) 10625 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10626 #endif /* INET */ 10627 #ifdef INET6 10628 if (pd.naf == AF_INET6) 10629 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10630 #endif /* INET6 */ 10631 *m0 = NULL; 10632 action = PF_PASS; 10633 goto out; 10634 break; 10635 default: 10636 if (pd.act.rt) { 10637 switch (af) { 10638 #ifdef INET 10639 case AF_INET: 10640 /* pf_route() returns unlocked. */ 10641 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10642 break; 10643 #endif /* INET */ 10644 #ifdef INET6 10645 case AF_INET6: 10646 /* pf_route6() returns unlocked. */ 10647 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10648 break; 10649 #endif /* INET6 */ 10650 } 10651 goto out; 10652 } 10653 if (pf_dummynet(&pd, s, r, m0) != 0) { 10654 action = PF_DROP; 10655 REASON_SET(&reason, PFRES_MEMORY); 10656 } 10657 break; 10658 } 10659 10660 eat_pkt: 10661 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 10662 10663 if (s && action != PF_DROP) { 10664 if (!s->if_index_in && dir == PF_IN) 10665 s->if_index_in = ifp->if_index; 10666 else if (!s->if_index_out && dir == PF_OUT) 10667 s->if_index_out = ifp->if_index; 10668 } 10669 10670 if (s) 10671 PF_STATE_UNLOCK(s); 10672 10673 out: 10674 #ifdef INET6 10675 /* If reassembled packet passed, create new fragments. */ 10676 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 10677 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 10678 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 10679 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 10680 #endif /* INET6 */ 10681 10682 pf_sctp_multihome_delayed(&pd, kif, s, action); 10683 10684 return (action); 10685 } 10686 #endif /* INET || INET6 */ 10687