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_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 124 "struct pf_kstate *"); 125 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 126 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 127 "struct pf_kstate *"); 128 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 129 "struct pfi_kkif *"); 130 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 131 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 132 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 133 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 134 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 135 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 136 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 137 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 138 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 139 "struct pf_krule *", "struct mbuf *", "int"); 140 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 141 "struct pf_sctp_source *"); 142 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 143 "struct pf_kstate *", "struct pf_sctp_source *"); 144 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 145 "int", "struct pf_pdesc *", "int"); 146 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 147 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 148 "int"); 149 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 150 "int"); 151 152 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 153 "struct mbuf *"); 154 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 155 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 156 "int", "struct pf_keth_rule *", "char *"); 157 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 158 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 159 "int", "struct pf_keth_rule *"); 160 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 161 162 /* 163 * Global variables 164 */ 165 166 /* state tables */ 167 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 168 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 169 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 170 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 173 VNET_DEFINE(struct pf_kstatus, pf_status); 174 175 VNET_DEFINE(u_int32_t, ticket_altqs_active); 176 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 177 VNET_DEFINE(int, altqs_inactive_open); 178 VNET_DEFINE(u_int32_t, ticket_pabuf); 179 180 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 181 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 182 VNET_DEFINE(u_char, pf_tcp_secret[16]); 183 #define V_pf_tcp_secret VNET(pf_tcp_secret) 184 VNET_DEFINE(int, pf_tcp_secret_init); 185 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 186 VNET_DEFINE(int, pf_tcp_iss_off); 187 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 188 VNET_DECLARE(int, pf_vnet_active); 189 #define V_pf_vnet_active VNET(pf_vnet_active) 190 191 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 192 #define V_pf_purge_idx VNET(pf_purge_idx) 193 194 #ifdef PF_WANT_32_TO_64_COUNTER 195 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 196 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 197 198 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 199 VNET_DEFINE(size_t, pf_allrulecount); 200 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 201 #endif 202 203 struct pf_sctp_endpoint; 204 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 205 struct pf_sctp_source { 206 sa_family_t af; 207 struct pf_addr addr; 208 TAILQ_ENTRY(pf_sctp_source) entry; 209 }; 210 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 211 struct pf_sctp_endpoint 212 { 213 uint32_t v_tag; 214 struct pf_sctp_sources sources; 215 RB_ENTRY(pf_sctp_endpoint) entry; 216 }; 217 static int 218 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 219 { 220 return (a->v_tag - b->v_tag); 221 } 222 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 223 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 224 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 225 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 226 static struct mtx_padalign pf_sctp_endpoints_mtx; 227 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 228 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 229 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 230 231 /* 232 * Queue for pf_intr() sends. 233 */ 234 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 235 struct pf_send_entry { 236 STAILQ_ENTRY(pf_send_entry) pfse_next; 237 struct mbuf *pfse_m; 238 enum { 239 PFSE_IP, 240 PFSE_IP6, 241 PFSE_ICMP, 242 PFSE_ICMP6, 243 } pfse_type; 244 struct { 245 int type; 246 int code; 247 int mtu; 248 } icmpopts; 249 }; 250 251 STAILQ_HEAD(pf_send_head, pf_send_entry); 252 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 253 #define V_pf_sendqueue VNET(pf_sendqueue) 254 255 static struct mtx_padalign pf_sendqueue_mtx; 256 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 257 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 258 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 259 260 /* 261 * Queue for pf_overload_task() tasks. 262 */ 263 struct pf_overload_entry { 264 SLIST_ENTRY(pf_overload_entry) next; 265 struct pf_addr addr; 266 sa_family_t af; 267 uint8_t dir; 268 struct pf_krule *rule; 269 }; 270 271 SLIST_HEAD(pf_overload_head, pf_overload_entry); 272 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 273 #define V_pf_overloadqueue VNET(pf_overloadqueue) 274 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 275 #define V_pf_overloadtask VNET(pf_overloadtask) 276 277 static struct mtx_padalign pf_overloadqueue_mtx; 278 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 279 "pf overload/flush queue", MTX_DEF); 280 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 281 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 282 283 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 284 struct mtx_padalign pf_unlnkdrules_mtx; 285 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 286 MTX_DEF); 287 288 struct sx pf_config_lock; 289 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 290 291 struct mtx_padalign pf_table_stats_lock; 292 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 293 MTX_DEF); 294 295 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 296 #define V_pf_sources_z VNET(pf_sources_z) 297 uma_zone_t pf_mtag_z; 298 VNET_DEFINE(uma_zone_t, pf_state_z); 299 VNET_DEFINE(uma_zone_t, pf_state_key_z); 300 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 301 302 VNET_DEFINE(struct unrhdr64, pf_stateid); 303 304 static void pf_src_tree_remove_state(struct pf_kstate *); 305 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 306 u_int32_t); 307 static void pf_add_threshold(struct pf_threshold *); 308 static int pf_check_threshold(struct pf_threshold *); 309 310 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 311 u_int16_t *, u_int16_t *, struct pf_addr *, 312 u_int16_t, u_int8_t, sa_family_t, sa_family_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 int *, 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 *, 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 **, 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 static int pf_state_key_addr_setup(struct pf_pdesc *, 352 struct pf_state_key_cmp *, int); 353 static int pf_tcp_track_full(struct pf_kstate **, 354 struct pf_pdesc *, u_short *, int *); 355 static int pf_tcp_track_sloppy(struct pf_kstate **, 356 struct pf_pdesc *, u_short *); 357 static int pf_test_state(struct pf_kstate **, struct pf_pdesc *, 358 u_short *); 359 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 360 struct pf_pdesc *, struct pf_kstate **, 361 int, u_int16_t, u_int16_t, 362 int, int *, int, int); 363 static int pf_test_state_icmp(struct pf_kstate **, 364 struct pf_pdesc *, u_short *); 365 static int pf_sctp_track(struct pf_kstate *, struct pf_pdesc *, 366 u_short *); 367 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 368 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 369 struct pfi_kkif *, struct pf_kstate *, int); 370 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 371 int, u_int16_t); 372 static int pf_check_proto_cksum(struct mbuf *, int, int, 373 u_int8_t, sa_family_t); 374 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 375 int, int, u_short *); 376 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 377 u_short *); 378 static void pf_print_state_parts(struct pf_kstate *, 379 struct pf_state_key *, struct pf_state_key *); 380 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 381 bool, u_int8_t); 382 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 383 const struct pf_state_key_cmp *, u_int); 384 static bool pf_src_connlimit(struct pf_kstate *); 385 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 386 static void pf_counters_inc(int, struct pf_pdesc *, 387 struct pf_kstate *, struct pf_krule *, 388 struct pf_krule *); 389 static void pf_overload_task(void *v, int pending); 390 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 391 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 392 struct pf_addr *, sa_family_t, struct pf_addr *, 393 struct pfi_kkif *, pf_sn_types_t); 394 static u_int pf_purge_expired_states(u_int, int); 395 static void pf_purge_unlinked_rules(void); 396 static int pf_mtag_uminit(void *, int, int); 397 static void pf_mtag_free(struct m_tag *); 398 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 399 int, struct pf_state_key *); 400 #ifdef INET 401 static void pf_route(struct mbuf **, struct pf_krule *, 402 struct ifnet *, struct pf_kstate *, 403 struct pf_pdesc *, struct inpcb *); 404 #endif /* INET */ 405 #ifdef INET6 406 static void pf_change_a6(struct pf_addr *, u_int16_t *, 407 struct pf_addr *, u_int8_t); 408 static void pf_route6(struct mbuf **, struct pf_krule *, 409 struct ifnet *, struct pf_kstate *, 410 struct pf_pdesc *, struct inpcb *); 411 #endif /* INET6 */ 412 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 413 414 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 415 416 extern int pf_end_threads; 417 extern struct proc *pf_purge_proc; 418 419 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 420 421 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 422 423 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 424 do { \ 425 struct pf_state_key *nk; \ 426 if ((pd->dir) == PF_OUT) \ 427 nk = (_s)->key[PF_SK_STACK]; \ 428 else \ 429 nk = (_s)->key[PF_SK_WIRE]; \ 430 pf_packet_rework_nat(_m, _pd, _off, nk); \ 431 } while (0) 432 433 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 434 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 435 436 #define STATE_LOOKUP(k, s, pd) \ 437 do { \ 438 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 439 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 440 if ((s) == NULL) \ 441 return (PF_DROP); \ 442 if (PACKET_LOOPED(pd)) \ 443 return (PF_PASS); \ 444 } while (0) 445 446 static struct pfi_kkif * 447 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 448 { 449 struct pfi_kkif *k = pd->kif; 450 451 SDT_PROBE2(pf, ip, , bound_iface, st, k); 452 453 /* Floating unless otherwise specified. */ 454 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 455 return (V_pfi_all); 456 457 /* 458 * Initially set to all, because we don't know what interface we'll be 459 * sending this out when we create the state. 460 */ 461 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf)) 462 return (V_pfi_all); 463 464 /* 465 * If this state is created based on another state (e.g. SCTP 466 * multihome) always set it floating initially. We can't know for sure 467 * what interface the actual traffic for this state will come in on. 468 */ 469 if (pd->related_rule) 470 return (V_pfi_all); 471 472 /* Don't overrule the interface for states created on incoming packets. */ 473 if (st->direction == PF_IN) 474 return (k); 475 476 /* No route-to, so don't overrule. */ 477 if (st->act.rt != PF_ROUTETO) 478 return (k); 479 480 /* Bind to the route-to interface. */ 481 return (st->act.rt_kif); 482 } 483 484 #define STATE_INC_COUNTERS(s) \ 485 do { \ 486 struct pf_krule_item *mrm; \ 487 counter_u64_add(s->rule->states_cur, 1); \ 488 counter_u64_add(s->rule->states_tot, 1); \ 489 if (s->anchor != NULL) { \ 490 counter_u64_add(s->anchor->states_cur, 1); \ 491 counter_u64_add(s->anchor->states_tot, 1); \ 492 } \ 493 if (s->nat_rule != NULL) { \ 494 counter_u64_add(s->nat_rule->states_cur, 1);\ 495 counter_u64_add(s->nat_rule->states_tot, 1);\ 496 } \ 497 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 498 counter_u64_add(mrm->r->states_cur, 1); \ 499 counter_u64_add(mrm->r->states_tot, 1); \ 500 } \ 501 } while (0) 502 503 #define STATE_DEC_COUNTERS(s) \ 504 do { \ 505 struct pf_krule_item *mrm; \ 506 if (s->nat_rule != NULL) \ 507 counter_u64_add(s->nat_rule->states_cur, -1);\ 508 if (s->anchor != NULL) \ 509 counter_u64_add(s->anchor->states_cur, -1); \ 510 counter_u64_add(s->rule->states_cur, -1); \ 511 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 512 counter_u64_add(mrm->r->states_cur, -1); \ 513 } while (0) 514 515 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 516 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 517 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 518 VNET_DEFINE(struct pf_idhash *, pf_idhash); 519 VNET_DEFINE(struct pf_srchash *, pf_srchash); 520 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 521 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 522 523 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 524 "pf(4)"); 525 526 VNET_DEFINE(u_long, pf_hashmask); 527 VNET_DEFINE(u_long, pf_srchashmask); 528 VNET_DEFINE(u_long, pf_udpendpointhashmask); 529 VNET_DEFINE_STATIC(u_long, pf_hashsize); 530 #define V_pf_hashsize VNET(pf_hashsize) 531 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 532 #define V_pf_srchashsize VNET(pf_srchashsize) 533 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 534 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 535 u_long pf_ioctl_maxcount = 65535; 536 537 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 538 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 539 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 540 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 541 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 542 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 543 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 544 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 545 546 VNET_DEFINE(void *, pf_swi_cookie); 547 VNET_DEFINE(struct intr_event *, pf_swi_ie); 548 549 VNET_DEFINE(uint32_t, pf_hashseed); 550 #define V_pf_hashseed VNET(pf_hashseed) 551 552 static void 553 pf_sctp_checksum(struct mbuf *m, int off) 554 { 555 uint32_t sum = 0; 556 557 /* Zero out the checksum, to enable recalculation. */ 558 m_copyback(m, off + offsetof(struct sctphdr, checksum), 559 sizeof(sum), (caddr_t)&sum); 560 561 sum = sctp_calculate_cksum(m, off); 562 563 m_copyback(m, off + offsetof(struct sctphdr, checksum), 564 sizeof(sum), (caddr_t)&sum); 565 } 566 567 int 568 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 569 { 570 571 switch (af) { 572 #ifdef INET 573 case AF_INET: 574 if (a->addr32[0] > b->addr32[0]) 575 return (1); 576 if (a->addr32[0] < b->addr32[0]) 577 return (-1); 578 break; 579 #endif /* INET */ 580 #ifdef INET6 581 case AF_INET6: 582 if (a->addr32[3] > b->addr32[3]) 583 return (1); 584 if (a->addr32[3] < b->addr32[3]) 585 return (-1); 586 if (a->addr32[2] > b->addr32[2]) 587 return (1); 588 if (a->addr32[2] < b->addr32[2]) 589 return (-1); 590 if (a->addr32[1] > b->addr32[1]) 591 return (1); 592 if (a->addr32[1] < b->addr32[1]) 593 return (-1); 594 if (a->addr32[0] > b->addr32[0]) 595 return (1); 596 if (a->addr32[0] < b->addr32[0]) 597 return (-1); 598 break; 599 #endif /* INET6 */ 600 } 601 return (0); 602 } 603 604 static bool 605 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 606 { 607 switch (af) { 608 #ifdef INET 609 case AF_INET: 610 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 611 #endif 612 case AF_INET6: 613 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 614 default: 615 panic("Unknown af %d", af); 616 } 617 } 618 619 static void 620 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 621 struct pf_state_key *nk) 622 { 623 624 switch (pd->proto) { 625 case IPPROTO_TCP: { 626 struct tcphdr *th = &pd->hdr.tcp; 627 628 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 629 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 630 &th->th_sum, &nk->addr[pd->sidx], 631 nk->port[pd->sidx], 0, pd->af, pd->naf); 632 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 633 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 634 &th->th_sum, &nk->addr[pd->didx], 635 nk->port[pd->didx], 0, pd->af, pd->naf); 636 m_copyback(m, off, sizeof(*th), (caddr_t)th); 637 break; 638 } 639 case IPPROTO_UDP: { 640 struct udphdr *uh = &pd->hdr.udp; 641 642 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 643 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 644 &uh->uh_sum, &nk->addr[pd->sidx], 645 nk->port[pd->sidx], 1, pd->af, pd->naf); 646 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 647 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 648 &uh->uh_sum, &nk->addr[pd->didx], 649 nk->port[pd->didx], 1, pd->af, pd->naf); 650 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 651 break; 652 } 653 case IPPROTO_SCTP: { 654 struct sctphdr *sh = &pd->hdr.sctp; 655 uint16_t checksum = 0; 656 657 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 658 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 659 &checksum, &nk->addr[pd->sidx], 660 nk->port[pd->sidx], 1, pd->af, pd->naf); 661 } 662 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 663 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 664 &checksum, &nk->addr[pd->didx], 665 nk->port[pd->didx], 1, pd->af, pd->naf); 666 } 667 668 break; 669 } 670 case IPPROTO_ICMP: { 671 struct icmp *ih = &pd->hdr.icmp; 672 673 if (nk->port[pd->sidx] != ih->icmp_id) { 674 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 675 ih->icmp_cksum, ih->icmp_id, 676 nk->port[pd->sidx], 0); 677 ih->icmp_id = nk->port[pd->sidx]; 678 pd->sport = &ih->icmp_id; 679 680 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 681 } 682 /* FALLTHROUGH */ 683 } 684 default: 685 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 686 switch (pd->af) { 687 case AF_INET: 688 pf_change_a(&pd->src->v4.s_addr, 689 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 690 0); 691 break; 692 case AF_INET6: 693 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 694 break; 695 } 696 } 697 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 698 switch (pd->af) { 699 case AF_INET: 700 pf_change_a(&pd->dst->v4.s_addr, 701 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 702 0); 703 break; 704 case AF_INET6: 705 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 706 break; 707 } 708 } 709 break; 710 } 711 } 712 713 static __inline uint32_t 714 pf_hashkey(const struct pf_state_key *sk) 715 { 716 uint32_t h; 717 718 h = murmur3_32_hash32((const uint32_t *)sk, 719 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 720 V_pf_hashseed); 721 722 return (h & V_pf_hashmask); 723 } 724 725 __inline uint32_t 726 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 727 { 728 uint32_t h; 729 730 switch (af) { 731 case AF_INET: 732 h = murmur3_32_hash32((uint32_t *)&addr->v4, 733 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 734 break; 735 case AF_INET6: 736 h = murmur3_32_hash32((uint32_t *)&addr->v6, 737 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 738 break; 739 } 740 741 return (h & V_pf_srchashmask); 742 } 743 744 static inline uint32_t 745 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 746 { 747 uint32_t h; 748 749 h = murmur3_32_hash32((uint32_t *)endpoint, 750 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 751 V_pf_hashseed); 752 return (h & V_pf_udpendpointhashmask); 753 } 754 755 #ifdef ALTQ 756 static int 757 pf_state_hash(struct pf_kstate *s) 758 { 759 u_int32_t hv = (intptr_t)s / sizeof(*s); 760 761 hv ^= crc32(&s->src, sizeof(s->src)); 762 hv ^= crc32(&s->dst, sizeof(s->dst)); 763 if (hv == 0) 764 hv = 1; 765 return (hv); 766 } 767 #endif 768 769 static __inline void 770 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 771 { 772 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 773 s->dst.state = newstate; 774 if (which == PF_PEER_DST) 775 return; 776 if (s->src.state == newstate) 777 return; 778 if (s->creatorid == V_pf_status.hostid && 779 s->key[PF_SK_STACK] != NULL && 780 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 781 !(TCPS_HAVEESTABLISHED(s->src.state) || 782 s->src.state == TCPS_CLOSED) && 783 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 784 atomic_add_32(&V_pf_status.states_halfopen, -1); 785 786 s->src.state = newstate; 787 } 788 789 #ifdef INET6 790 void 791 pf_addrcpy(struct pf_addr *dst, const struct pf_addr *src, sa_family_t af) 792 { 793 switch (af) { 794 #ifdef INET 795 case AF_INET: 796 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 797 break; 798 #endif /* INET */ 799 case AF_INET6: 800 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 801 break; 802 } 803 } 804 #endif /* INET6 */ 805 806 static void 807 pf_init_threshold(struct pf_threshold *threshold, 808 u_int32_t limit, u_int32_t seconds) 809 { 810 threshold->limit = limit * PF_THRESHOLD_MULT; 811 threshold->seconds = seconds; 812 threshold->count = 0; 813 threshold->last = time_uptime; 814 } 815 816 static void 817 pf_add_threshold(struct pf_threshold *threshold) 818 { 819 u_int32_t t = time_uptime, diff = t - threshold->last; 820 821 if (diff >= threshold->seconds) 822 threshold->count = 0; 823 else 824 threshold->count -= threshold->count * diff / 825 threshold->seconds; 826 threshold->count += PF_THRESHOLD_MULT; 827 threshold->last = t; 828 } 829 830 static int 831 pf_check_threshold(struct pf_threshold *threshold) 832 { 833 return (threshold->count > threshold->limit); 834 } 835 836 static bool 837 pf_src_connlimit(struct pf_kstate *state) 838 { 839 struct pf_overload_entry *pfoe; 840 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 841 bool limited = false; 842 843 PF_STATE_LOCK_ASSERT(state); 844 PF_SRC_NODE_LOCK(src_node); 845 846 src_node->conn++; 847 state->src.tcp_est = 1; 848 pf_add_threshold(&src_node->conn_rate); 849 850 if (state->rule->max_src_conn && 851 state->rule->max_src_conn < 852 src_node->conn) { 853 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 854 limited = true; 855 } 856 857 if (state->rule->max_src_conn_rate.limit && 858 pf_check_threshold(&src_node->conn_rate)) { 859 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 860 limited = true; 861 } 862 863 if (!limited) 864 goto done; 865 866 /* Kill this state. */ 867 state->timeout = PFTM_PURGE; 868 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 869 870 if (state->rule->overload_tbl == NULL) 871 goto done; 872 873 /* Schedule overloading and flushing task. */ 874 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 875 if (pfoe == NULL) 876 goto done; /* too bad :( */ 877 878 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 879 pfoe->af = state->key[PF_SK_WIRE]->af; 880 pfoe->rule = state->rule; 881 pfoe->dir = state->direction; 882 PF_OVERLOADQ_LOCK(); 883 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 884 PF_OVERLOADQ_UNLOCK(); 885 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 886 887 done: 888 PF_SRC_NODE_UNLOCK(src_node); 889 return (limited); 890 } 891 892 static void 893 pf_overload_task(void *v, int pending) 894 { 895 struct pf_overload_head queue; 896 struct pfr_addr p; 897 struct pf_overload_entry *pfoe, *pfoe1; 898 uint32_t killed = 0; 899 900 CURVNET_SET((struct vnet *)v); 901 902 PF_OVERLOADQ_LOCK(); 903 queue = V_pf_overloadqueue; 904 SLIST_INIT(&V_pf_overloadqueue); 905 PF_OVERLOADQ_UNLOCK(); 906 907 bzero(&p, sizeof(p)); 908 SLIST_FOREACH(pfoe, &queue, next) { 909 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 910 if (V_pf_status.debug >= PF_DEBUG_MISC) { 911 printf("%s: blocking address ", __func__); 912 pf_print_host(&pfoe->addr, 0, pfoe->af); 913 printf("\n"); 914 } 915 916 p.pfra_af = pfoe->af; 917 switch (pfoe->af) { 918 #ifdef INET 919 case AF_INET: 920 p.pfra_net = 32; 921 p.pfra_ip4addr = pfoe->addr.v4; 922 break; 923 #endif 924 #ifdef INET6 925 case AF_INET6: 926 p.pfra_net = 128; 927 p.pfra_ip6addr = pfoe->addr.v6; 928 break; 929 #endif 930 } 931 932 PF_RULES_WLOCK(); 933 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 934 PF_RULES_WUNLOCK(); 935 } 936 937 /* 938 * Remove those entries, that don't need flushing. 939 */ 940 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 941 if (pfoe->rule->flush == 0) { 942 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 943 free(pfoe, M_PFTEMP); 944 } else 945 counter_u64_add( 946 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 947 948 /* If nothing to flush, return. */ 949 if (SLIST_EMPTY(&queue)) { 950 CURVNET_RESTORE(); 951 return; 952 } 953 954 for (int i = 0; i <= V_pf_hashmask; i++) { 955 struct pf_idhash *ih = &V_pf_idhash[i]; 956 struct pf_state_key *sk; 957 struct pf_kstate *s; 958 959 PF_HASHROW_LOCK(ih); 960 LIST_FOREACH(s, &ih->states, entry) { 961 sk = s->key[PF_SK_WIRE]; 962 SLIST_FOREACH(pfoe, &queue, next) 963 if (sk->af == pfoe->af && 964 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 965 pfoe->rule == s->rule) && 966 ((pfoe->dir == PF_OUT && 967 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 968 (pfoe->dir == PF_IN && 969 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 970 s->timeout = PFTM_PURGE; 971 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 972 killed++; 973 } 974 } 975 PF_HASHROW_UNLOCK(ih); 976 } 977 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 978 free(pfoe, M_PFTEMP); 979 if (V_pf_status.debug >= PF_DEBUG_MISC) 980 printf("%s: %u states killed", __func__, killed); 981 982 CURVNET_RESTORE(); 983 } 984 985 /* 986 * On node found always returns locked. On not found its configurable. 987 */ 988 struct pf_ksrc_node * 989 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 990 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 991 { 992 struct pf_ksrc_node *n; 993 994 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 995 996 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 997 PF_HASHROW_LOCK(*sh); 998 LIST_FOREACH(n, &(*sh)->nodes, entry) 999 if (n->rule == rule && n->af == af && n->type == sn_type && 1000 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 1001 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 1002 break; 1003 1004 if (n == NULL && !returnlocked) 1005 PF_HASHROW_UNLOCK(*sh); 1006 1007 return (n); 1008 } 1009 1010 bool 1011 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 1012 { 1013 struct pf_ksrc_node *cur; 1014 1015 if ((*sn) == NULL) 1016 return (false); 1017 1018 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 1019 1020 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1021 PF_HASHROW_LOCK(sh); 1022 LIST_FOREACH(cur, &(sh->nodes), entry) { 1023 if (cur == (*sn) && 1024 cur->expire != 1) /* Ignore nodes being killed */ 1025 return (true); 1026 } 1027 PF_HASHROW_UNLOCK(sh); 1028 (*sn) = NULL; 1029 return (false); 1030 } 1031 1032 static void 1033 pf_free_src_node(struct pf_ksrc_node *sn) 1034 { 1035 1036 for (int i = 0; i < 2; i++) { 1037 counter_u64_free(sn->bytes[i]); 1038 counter_u64_free(sn->packets[i]); 1039 } 1040 uma_zfree(V_pf_sources_z, sn); 1041 } 1042 1043 static u_short 1044 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1045 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1046 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1047 struct pfi_kkif *rkif, pf_sn_types_t sn_type) 1048 { 1049 u_short reason = 0; 1050 struct pf_krule *r_track = rule; 1051 struct pf_ksrc_node **sn = &(sns[sn_type]); 1052 struct pf_srchash **sh = &(snhs[sn_type]); 1053 1054 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1055 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1056 1057 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1058 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1059 1060 /* 1061 * XXX: There could be a KASSERT for 1062 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1063 * but we'd need to pass pool *only* for this KASSERT. 1064 */ 1065 1066 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1067 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1068 r_track = &V_pf_default_rule; 1069 1070 /* 1071 * Request the sh to always be locked, as we might insert a new sn. 1072 */ 1073 if (*sn == NULL) 1074 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1075 1076 if (*sn == NULL) { 1077 PF_HASHROW_ASSERT(*sh); 1078 1079 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1080 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1081 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1082 reason = PFRES_SRCLIMIT; 1083 goto done; 1084 } 1085 1086 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1087 if ((*sn) == NULL) { 1088 reason = PFRES_MEMORY; 1089 goto done; 1090 } 1091 1092 for (int i = 0; i < 2; i++) { 1093 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1094 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1095 1096 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1097 pf_free_src_node(*sn); 1098 reason = PFRES_MEMORY; 1099 goto done; 1100 } 1101 } 1102 1103 if (sn_type == PF_SN_LIMIT) 1104 pf_init_threshold(&(*sn)->conn_rate, 1105 rule->max_src_conn_rate.limit, 1106 rule->max_src_conn_rate.seconds); 1107 1108 MPASS((*sn)->lock == NULL); 1109 (*sn)->lock = &(*sh)->lock; 1110 1111 (*sn)->af = af; 1112 (*sn)->rule = r_track; 1113 PF_ACPY(&(*sn)->addr, src, af); 1114 if (raddr != NULL) 1115 PF_ACPY(&(*sn)->raddr, raddr, af); 1116 (*sn)->rkif = rkif; 1117 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1118 (*sn)->creation = time_uptime; 1119 (*sn)->ruletype = rule->action; 1120 (*sn)->type = sn_type; 1121 counter_u64_add(r_track->src_nodes[sn_type], 1); 1122 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1123 } else { 1124 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1125 (*sn)->states >= rule->max_src_states) { 1126 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1127 1); 1128 reason = PFRES_SRCLIMIT; 1129 goto done; 1130 } 1131 } 1132 done: 1133 if (reason == 0) 1134 (*sn)->states++; 1135 else 1136 (*sn) = NULL; 1137 1138 PF_HASHROW_UNLOCK(*sh); 1139 return (reason); 1140 } 1141 1142 void 1143 pf_unlink_src_node(struct pf_ksrc_node *src) 1144 { 1145 PF_SRC_NODE_LOCK_ASSERT(src); 1146 1147 LIST_REMOVE(src, entry); 1148 if (src->rule) 1149 counter_u64_add(src->rule->src_nodes[src->type], -1); 1150 } 1151 1152 u_int 1153 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1154 { 1155 struct pf_ksrc_node *sn, *tmp; 1156 u_int count = 0; 1157 1158 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1159 pf_free_src_node(sn); 1160 count++; 1161 } 1162 1163 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1164 1165 return (count); 1166 } 1167 1168 void 1169 pf_mtag_initialize(void) 1170 { 1171 1172 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1173 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1174 UMA_ALIGN_PTR, 0); 1175 } 1176 1177 /* Per-vnet data storage structures initialization. */ 1178 void 1179 pf_initialize(void) 1180 { 1181 struct pf_keyhash *kh; 1182 struct pf_idhash *ih; 1183 struct pf_srchash *sh; 1184 struct pf_udpendpointhash *uh; 1185 u_int i; 1186 1187 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1188 V_pf_hashsize = PF_HASHSIZ; 1189 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1190 V_pf_srchashsize = PF_SRCHASHSIZ; 1191 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1192 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1193 1194 V_pf_hashseed = arc4random(); 1195 1196 /* States and state keys storage. */ 1197 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1198 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1199 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1200 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1201 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1202 1203 V_pf_state_key_z = uma_zcreate("pf state keys", 1204 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1205 UMA_ALIGN_PTR, 0); 1206 1207 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1208 M_PFHASH, M_NOWAIT | M_ZERO); 1209 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1210 M_PFHASH, M_NOWAIT | M_ZERO); 1211 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1212 printf("pf: Unable to allocate memory for " 1213 "state_hashsize %lu.\n", V_pf_hashsize); 1214 1215 free(V_pf_keyhash, M_PFHASH); 1216 free(V_pf_idhash, M_PFHASH); 1217 1218 V_pf_hashsize = PF_HASHSIZ; 1219 V_pf_keyhash = mallocarray(V_pf_hashsize, 1220 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1221 V_pf_idhash = mallocarray(V_pf_hashsize, 1222 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1223 } 1224 1225 V_pf_hashmask = V_pf_hashsize - 1; 1226 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1227 i++, kh++, ih++) { 1228 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1229 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1230 } 1231 1232 /* Source nodes. */ 1233 V_pf_sources_z = uma_zcreate("pf source nodes", 1234 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1235 0); 1236 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1237 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1238 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1239 1240 V_pf_srchash = mallocarray(V_pf_srchashsize, 1241 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1242 if (V_pf_srchash == NULL) { 1243 printf("pf: Unable to allocate memory for " 1244 "source_hashsize %lu.\n", V_pf_srchashsize); 1245 1246 V_pf_srchashsize = PF_SRCHASHSIZ; 1247 V_pf_srchash = mallocarray(V_pf_srchashsize, 1248 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1249 } 1250 1251 V_pf_srchashmask = V_pf_srchashsize - 1; 1252 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1253 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1254 1255 1256 /* UDP endpoint mappings. */ 1257 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1258 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1259 UMA_ALIGN_PTR, 0); 1260 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1261 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1262 if (V_pf_udpendpointhash == NULL) { 1263 printf("pf: Unable to allocate memory for " 1264 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1265 1266 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1267 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1268 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1269 } 1270 1271 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1272 for (i = 0, uh = V_pf_udpendpointhash; 1273 i <= V_pf_udpendpointhashmask; 1274 i++, uh++) { 1275 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1276 MTX_DEF | MTX_DUPOK); 1277 } 1278 1279 /* ALTQ */ 1280 TAILQ_INIT(&V_pf_altqs[0]); 1281 TAILQ_INIT(&V_pf_altqs[1]); 1282 TAILQ_INIT(&V_pf_altqs[2]); 1283 TAILQ_INIT(&V_pf_altqs[3]); 1284 TAILQ_INIT(&V_pf_pabuf[0]); 1285 TAILQ_INIT(&V_pf_pabuf[1]); 1286 TAILQ_INIT(&V_pf_pabuf[2]); 1287 V_pf_altqs_active = &V_pf_altqs[0]; 1288 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1289 V_pf_altqs_inactive = &V_pf_altqs[2]; 1290 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1291 1292 /* Send & overload+flush queues. */ 1293 STAILQ_INIT(&V_pf_sendqueue); 1294 SLIST_INIT(&V_pf_overloadqueue); 1295 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1296 1297 /* Unlinked, but may be referenced rules. */ 1298 TAILQ_INIT(&V_pf_unlinked_rules); 1299 } 1300 1301 void 1302 pf_mtag_cleanup(void) 1303 { 1304 1305 uma_zdestroy(pf_mtag_z); 1306 } 1307 1308 void 1309 pf_cleanup(void) 1310 { 1311 struct pf_keyhash *kh; 1312 struct pf_idhash *ih; 1313 struct pf_srchash *sh; 1314 struct pf_udpendpointhash *uh; 1315 struct pf_send_entry *pfse, *next; 1316 u_int i; 1317 1318 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1319 i <= V_pf_hashmask; 1320 i++, kh++, ih++) { 1321 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1322 __func__)); 1323 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1324 __func__)); 1325 mtx_destroy(&kh->lock); 1326 mtx_destroy(&ih->lock); 1327 } 1328 free(V_pf_keyhash, M_PFHASH); 1329 free(V_pf_idhash, M_PFHASH); 1330 1331 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1332 KASSERT(LIST_EMPTY(&sh->nodes), 1333 ("%s: source node hash not empty", __func__)); 1334 mtx_destroy(&sh->lock); 1335 } 1336 free(V_pf_srchash, M_PFHASH); 1337 1338 for (i = 0, uh = V_pf_udpendpointhash; 1339 i <= V_pf_udpendpointhashmask; 1340 i++, uh++) { 1341 KASSERT(LIST_EMPTY(&uh->endpoints), 1342 ("%s: udp endpoint hash not empty", __func__)); 1343 mtx_destroy(&uh->lock); 1344 } 1345 free(V_pf_udpendpointhash, M_PFHASH); 1346 1347 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1348 m_freem(pfse->pfse_m); 1349 free(pfse, M_PFTEMP); 1350 } 1351 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1352 1353 uma_zdestroy(V_pf_sources_z); 1354 uma_zdestroy(V_pf_state_z); 1355 uma_zdestroy(V_pf_state_key_z); 1356 uma_zdestroy(V_pf_udp_mapping_z); 1357 } 1358 1359 static int 1360 pf_mtag_uminit(void *mem, int size, int how) 1361 { 1362 struct m_tag *t; 1363 1364 t = (struct m_tag *)mem; 1365 t->m_tag_cookie = MTAG_ABI_COMPAT; 1366 t->m_tag_id = PACKET_TAG_PF; 1367 t->m_tag_len = sizeof(struct pf_mtag); 1368 t->m_tag_free = pf_mtag_free; 1369 1370 return (0); 1371 } 1372 1373 static void 1374 pf_mtag_free(struct m_tag *t) 1375 { 1376 1377 uma_zfree(pf_mtag_z, t); 1378 } 1379 1380 struct pf_mtag * 1381 pf_get_mtag(struct mbuf *m) 1382 { 1383 struct m_tag *mtag; 1384 1385 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1386 return ((struct pf_mtag *)(mtag + 1)); 1387 1388 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1389 if (mtag == NULL) 1390 return (NULL); 1391 bzero(mtag + 1, sizeof(struct pf_mtag)); 1392 m_tag_prepend(m, mtag); 1393 1394 return ((struct pf_mtag *)(mtag + 1)); 1395 } 1396 1397 static int 1398 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1399 struct pf_kstate *s) 1400 { 1401 struct pf_keyhash *khs, *khw, *kh; 1402 struct pf_state_key *sk, *cur; 1403 struct pf_kstate *si, *olds = NULL; 1404 int idx; 1405 1406 NET_EPOCH_ASSERT(); 1407 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1408 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1409 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1410 1411 /* 1412 * We need to lock hash slots of both keys. To avoid deadlock 1413 * we always lock the slot with lower address first. Unlock order 1414 * isn't important. 1415 * 1416 * We also need to lock ID hash slot before dropping key 1417 * locks. On success we return with ID hash slot locked. 1418 */ 1419 1420 if (skw == sks) { 1421 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1422 PF_HASHROW_LOCK(khs); 1423 } else { 1424 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1425 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1426 if (khs == khw) { 1427 PF_HASHROW_LOCK(khs); 1428 } else if (khs < khw) { 1429 PF_HASHROW_LOCK(khs); 1430 PF_HASHROW_LOCK(khw); 1431 } else { 1432 PF_HASHROW_LOCK(khw); 1433 PF_HASHROW_LOCK(khs); 1434 } 1435 } 1436 1437 #define KEYS_UNLOCK() do { \ 1438 if (khs != khw) { \ 1439 PF_HASHROW_UNLOCK(khs); \ 1440 PF_HASHROW_UNLOCK(khw); \ 1441 } else \ 1442 PF_HASHROW_UNLOCK(khs); \ 1443 } while (0) 1444 1445 /* 1446 * First run: start with wire key. 1447 */ 1448 sk = skw; 1449 kh = khw; 1450 idx = PF_SK_WIRE; 1451 1452 MPASS(s->lock == NULL); 1453 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1454 1455 keyattach: 1456 LIST_FOREACH(cur, &kh->keys, entry) 1457 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1458 break; 1459 1460 if (cur != NULL) { 1461 /* Key exists. Check for same kif, if none, add to key. */ 1462 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1463 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1464 1465 PF_HASHROW_LOCK(ih); 1466 if (si->kif == s->kif && 1467 ((si->key[PF_SK_WIRE]->af == sk->af && 1468 si->direction == s->direction) || 1469 (si->key[PF_SK_WIRE]->af != 1470 si->key[PF_SK_STACK]->af && 1471 sk->af == si->key[PF_SK_STACK]->af && 1472 si->direction != s->direction))) { 1473 if (sk->proto == IPPROTO_TCP && 1474 si->src.state >= TCPS_FIN_WAIT_2 && 1475 si->dst.state >= TCPS_FIN_WAIT_2) { 1476 /* 1477 * New state matches an old >FIN_WAIT_2 1478 * state. We can't drop key hash locks, 1479 * thus we can't unlink it properly. 1480 * 1481 * As a workaround we drop it into 1482 * TCPS_CLOSED state, schedule purge 1483 * ASAP and push it into the very end 1484 * of the slot TAILQ, so that it won't 1485 * conflict with our new state. 1486 */ 1487 pf_set_protostate(si, PF_PEER_BOTH, 1488 TCPS_CLOSED); 1489 si->timeout = PFTM_PURGE; 1490 olds = si; 1491 } else { 1492 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1493 printf("pf: %s key attach " 1494 "failed on %s: ", 1495 (idx == PF_SK_WIRE) ? 1496 "wire" : "stack", 1497 s->kif->pfik_name); 1498 pf_print_state_parts(s, 1499 (idx == PF_SK_WIRE) ? 1500 sk : NULL, 1501 (idx == PF_SK_STACK) ? 1502 sk : NULL); 1503 printf(", existing: "); 1504 pf_print_state_parts(si, 1505 (idx == PF_SK_WIRE) ? 1506 sk : NULL, 1507 (idx == PF_SK_STACK) ? 1508 sk : NULL); 1509 printf("\n"); 1510 } 1511 s->timeout = PFTM_UNLINKED; 1512 PF_HASHROW_UNLOCK(ih); 1513 KEYS_UNLOCK(); 1514 if (idx == PF_SK_WIRE) { 1515 uma_zfree(V_pf_state_key_z, skw); 1516 if (skw != sks) 1517 uma_zfree(V_pf_state_key_z, sks); 1518 } else { 1519 pf_detach_state(s); 1520 } 1521 return (EEXIST); /* collision! */ 1522 } 1523 } 1524 PF_HASHROW_UNLOCK(ih); 1525 } 1526 uma_zfree(V_pf_state_key_z, sk); 1527 s->key[idx] = cur; 1528 } else { 1529 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1530 s->key[idx] = sk; 1531 } 1532 1533 stateattach: 1534 /* List is sorted, if-bound states before floating. */ 1535 if (s->kif == V_pfi_all) 1536 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1537 else 1538 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1539 1540 if (olds) { 1541 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1542 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1543 key_list[idx]); 1544 olds = NULL; 1545 } 1546 1547 /* 1548 * Attach done. See how should we (or should not?) 1549 * attach a second key. 1550 */ 1551 if (sks == skw) { 1552 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1553 idx = PF_SK_STACK; 1554 sks = NULL; 1555 goto stateattach; 1556 } else if (sks != NULL) { 1557 /* 1558 * Continue attaching with stack key. 1559 */ 1560 sk = sks; 1561 kh = khs; 1562 idx = PF_SK_STACK; 1563 sks = NULL; 1564 goto keyattach; 1565 } 1566 1567 PF_STATE_LOCK(s); 1568 KEYS_UNLOCK(); 1569 1570 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1571 ("%s failure", __func__)); 1572 1573 return (0); 1574 #undef KEYS_UNLOCK 1575 } 1576 1577 static void 1578 pf_detach_state(struct pf_kstate *s) 1579 { 1580 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1581 struct pf_keyhash *kh; 1582 1583 NET_EPOCH_ASSERT(); 1584 MPASS(s->timeout >= PFTM_MAX); 1585 1586 pf_sctp_multihome_detach_addr(s); 1587 1588 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1589 V_pflow_export_state_ptr(s); 1590 1591 if (sks != NULL) { 1592 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1593 PF_HASHROW_LOCK(kh); 1594 if (s->key[PF_SK_STACK] != NULL) 1595 pf_state_key_detach(s, PF_SK_STACK); 1596 /* 1597 * If both point to same key, then we are done. 1598 */ 1599 if (sks == s->key[PF_SK_WIRE]) { 1600 pf_state_key_detach(s, PF_SK_WIRE); 1601 PF_HASHROW_UNLOCK(kh); 1602 return; 1603 } 1604 PF_HASHROW_UNLOCK(kh); 1605 } 1606 1607 if (s->key[PF_SK_WIRE] != NULL) { 1608 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1609 PF_HASHROW_LOCK(kh); 1610 if (s->key[PF_SK_WIRE] != NULL) 1611 pf_state_key_detach(s, PF_SK_WIRE); 1612 PF_HASHROW_UNLOCK(kh); 1613 } 1614 } 1615 1616 static void 1617 pf_state_key_detach(struct pf_kstate *s, int idx) 1618 { 1619 struct pf_state_key *sk = s->key[idx]; 1620 #ifdef INVARIANTS 1621 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1622 1623 PF_HASHROW_ASSERT(kh); 1624 #endif 1625 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1626 s->key[idx] = NULL; 1627 1628 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1629 LIST_REMOVE(sk, entry); 1630 uma_zfree(V_pf_state_key_z, sk); 1631 } 1632 } 1633 1634 static int 1635 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1636 { 1637 struct pf_state_key *sk = mem; 1638 1639 bzero(sk, sizeof(struct pf_state_key_cmp)); 1640 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1641 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1642 1643 return (0); 1644 } 1645 1646 static int 1647 pf_state_key_addr_setup(struct pf_pdesc *pd, 1648 struct pf_state_key_cmp *key, int multi) 1649 { 1650 struct pf_addr *saddr = pd->src; 1651 struct pf_addr *daddr = pd->dst; 1652 #ifdef INET6 1653 struct nd_neighbor_solicit nd; 1654 struct pf_addr *target; 1655 u_short action, reason; 1656 1657 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1658 goto copy; 1659 1660 switch (pd->hdr.icmp6.icmp6_type) { 1661 case ND_NEIGHBOR_SOLICIT: 1662 if (multi) 1663 return (-1); 1664 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1665 return (-1); 1666 target = (struct pf_addr *)&nd.nd_ns_target; 1667 daddr = target; 1668 break; 1669 case ND_NEIGHBOR_ADVERT: 1670 if (multi) 1671 return (-1); 1672 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1673 return (-1); 1674 target = (struct pf_addr *)&nd.nd_ns_target; 1675 saddr = target; 1676 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1677 key->addr[pd->didx].addr32[0] = 0; 1678 key->addr[pd->didx].addr32[1] = 0; 1679 key->addr[pd->didx].addr32[2] = 0; 1680 key->addr[pd->didx].addr32[3] = 0; 1681 daddr = NULL; /* overwritten */ 1682 } 1683 break; 1684 default: 1685 if (multi == PF_ICMP_MULTI_LINK) { 1686 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1687 key->addr[pd->sidx].addr32[1] = 0; 1688 key->addr[pd->sidx].addr32[2] = 0; 1689 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1690 saddr = NULL; /* overwritten */ 1691 } 1692 } 1693 copy: 1694 #endif 1695 if (saddr) 1696 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1697 if (daddr) 1698 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1699 1700 return (0); 1701 } 1702 1703 int 1704 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1705 struct pf_state_key **sk, struct pf_state_key **nk) 1706 { 1707 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1708 if (*sk == NULL) 1709 return (ENOMEM); 1710 1711 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1712 0)) { 1713 uma_zfree(V_pf_state_key_z, *sk); 1714 *sk = NULL; 1715 return (ENOMEM); 1716 } 1717 1718 (*sk)->port[pd->sidx] = sport; 1719 (*sk)->port[pd->didx] = dport; 1720 (*sk)->proto = pd->proto; 1721 (*sk)->af = pd->af; 1722 1723 *nk = pf_state_key_clone(*sk); 1724 if (*nk == NULL) { 1725 uma_zfree(V_pf_state_key_z, *sk); 1726 *sk = NULL; 1727 return (ENOMEM); 1728 } 1729 1730 if (pd->af != pd->naf) { 1731 (*sk)->port[pd->sidx] = pd->osport; 1732 (*sk)->port[pd->didx] = pd->odport; 1733 1734 (*nk)->af = pd->naf; 1735 1736 /* 1737 * We're overwriting an address here, so potentially there's bits of an IPv6 1738 * address left in here. Clear that out first. 1739 */ 1740 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1741 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1742 1743 PF_ACPY(&(*nk)->addr[pd->didx], &pd->nsaddr, pd->naf); 1744 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->ndaddr, pd->naf); 1745 (*nk)->port[pd->didx] = pd->nsport; 1746 (*nk)->port[pd->sidx] = pd->ndport; 1747 switch (pd->proto) { 1748 case IPPROTO_ICMP: 1749 (*nk)->proto = IPPROTO_ICMPV6; 1750 break; 1751 case IPPROTO_ICMPV6: 1752 (*nk)->proto = IPPROTO_ICMP; 1753 break; 1754 default: 1755 (*nk)->proto = pd->proto; 1756 } 1757 } 1758 1759 return (0); 1760 } 1761 1762 struct pf_state_key * 1763 pf_state_key_clone(const struct pf_state_key *orig) 1764 { 1765 struct pf_state_key *sk; 1766 1767 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1768 if (sk == NULL) 1769 return (NULL); 1770 1771 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1772 1773 return (sk); 1774 } 1775 1776 int 1777 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1778 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1779 { 1780 struct pf_idhash *ih; 1781 struct pf_kstate *cur; 1782 int error; 1783 1784 NET_EPOCH_ASSERT(); 1785 1786 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1787 ("%s: sks not pristine", __func__)); 1788 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1789 ("%s: skw not pristine", __func__)); 1790 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1791 1792 s->kif = kif; 1793 s->orig_kif = orig_kif; 1794 1795 if (s->id == 0 && s->creatorid == 0) { 1796 s->id = alloc_unr64(&V_pf_stateid); 1797 s->id = htobe64(s->id); 1798 s->creatorid = V_pf_status.hostid; 1799 } 1800 1801 /* Returns with ID locked on success. */ 1802 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1803 return (error); 1804 1805 ih = &V_pf_idhash[PF_IDHASH(s)]; 1806 PF_HASHROW_ASSERT(ih); 1807 LIST_FOREACH(cur, &ih->states, entry) 1808 if (cur->id == s->id && cur->creatorid == s->creatorid) 1809 break; 1810 1811 if (cur != NULL) { 1812 s->timeout = PFTM_UNLINKED; 1813 PF_HASHROW_UNLOCK(ih); 1814 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1815 printf("pf: state ID collision: " 1816 "id: %016llx creatorid: %08x\n", 1817 (unsigned long long)be64toh(s->id), 1818 ntohl(s->creatorid)); 1819 } 1820 pf_detach_state(s); 1821 return (EEXIST); 1822 } 1823 LIST_INSERT_HEAD(&ih->states, s, entry); 1824 /* One for keys, one for ID hash. */ 1825 refcount_init(&s->refs, 2); 1826 1827 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1828 if (V_pfsync_insert_state_ptr != NULL) 1829 V_pfsync_insert_state_ptr(s); 1830 1831 /* Returns locked. */ 1832 return (0); 1833 } 1834 1835 /* 1836 * Find state by ID: returns with locked row on success. 1837 */ 1838 struct pf_kstate * 1839 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1840 { 1841 struct pf_idhash *ih; 1842 struct pf_kstate *s; 1843 1844 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1845 1846 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1847 1848 PF_HASHROW_LOCK(ih); 1849 LIST_FOREACH(s, &ih->states, entry) 1850 if (s->id == id && s->creatorid == creatorid) 1851 break; 1852 1853 if (s == NULL) 1854 PF_HASHROW_UNLOCK(ih); 1855 1856 return (s); 1857 } 1858 1859 /* 1860 * Find state by key. 1861 * Returns with ID hash slot locked on success. 1862 */ 1863 static struct pf_kstate * 1864 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1865 u_int dir) 1866 { 1867 struct pf_keyhash *kh; 1868 struct pf_state_key *sk; 1869 struct pf_kstate *s; 1870 int idx; 1871 1872 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1873 1874 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1875 1876 PF_HASHROW_LOCK(kh); 1877 LIST_FOREACH(sk, &kh->keys, entry) 1878 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1879 break; 1880 if (sk == NULL) { 1881 PF_HASHROW_UNLOCK(kh); 1882 return (NULL); 1883 } 1884 1885 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1886 1887 /* List is sorted, if-bound states before floating ones. */ 1888 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1889 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1890 PF_STATE_LOCK(s); 1891 PF_HASHROW_UNLOCK(kh); 1892 if (__predict_false(s->timeout >= PFTM_MAX)) { 1893 /* 1894 * State is either being processed by 1895 * pf_unlink_state() in an other thread, or 1896 * is scheduled for immediate expiry. 1897 */ 1898 PF_STATE_UNLOCK(s); 1899 return (NULL); 1900 } 1901 return (s); 1902 } 1903 1904 /* Look through the other list, in case of AF-TO */ 1905 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1906 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1907 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1908 continue; 1909 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1910 PF_STATE_LOCK(s); 1911 PF_HASHROW_UNLOCK(kh); 1912 if (__predict_false(s->timeout >= PFTM_MAX)) { 1913 /* 1914 * State is either being processed by 1915 * pf_unlink_state() in an other thread, or 1916 * is scheduled for immediate expiry. 1917 */ 1918 PF_STATE_UNLOCK(s); 1919 return (NULL); 1920 } 1921 return (s); 1922 } 1923 } 1924 1925 PF_HASHROW_UNLOCK(kh); 1926 1927 return (NULL); 1928 } 1929 1930 /* 1931 * Returns with ID hash slot locked on success. 1932 */ 1933 struct pf_kstate * 1934 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1935 { 1936 struct pf_keyhash *kh; 1937 struct pf_state_key *sk; 1938 struct pf_kstate *s, *ret = NULL; 1939 int idx, inout = 0; 1940 1941 if (more != NULL) 1942 *more = 0; 1943 1944 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1945 1946 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1947 1948 PF_HASHROW_LOCK(kh); 1949 LIST_FOREACH(sk, &kh->keys, entry) 1950 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1951 break; 1952 if (sk == NULL) { 1953 PF_HASHROW_UNLOCK(kh); 1954 return (NULL); 1955 } 1956 switch (dir) { 1957 case PF_IN: 1958 idx = PF_SK_WIRE; 1959 break; 1960 case PF_OUT: 1961 idx = PF_SK_STACK; 1962 break; 1963 case PF_INOUT: 1964 idx = PF_SK_WIRE; 1965 inout = 1; 1966 break; 1967 default: 1968 panic("%s: dir %u", __func__, dir); 1969 } 1970 second_run: 1971 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1972 if (more == NULL) { 1973 PF_STATE_LOCK(s); 1974 PF_HASHROW_UNLOCK(kh); 1975 return (s); 1976 } 1977 1978 if (ret) 1979 (*more)++; 1980 else { 1981 ret = s; 1982 PF_STATE_LOCK(s); 1983 } 1984 } 1985 if (inout == 1) { 1986 inout = 0; 1987 idx = PF_SK_STACK; 1988 goto second_run; 1989 } 1990 PF_HASHROW_UNLOCK(kh); 1991 1992 return (ret); 1993 } 1994 1995 /* 1996 * FIXME 1997 * This routine is inefficient -- locks the state only to unlock immediately on 1998 * return. 1999 * It is racy -- after the state is unlocked nothing stops other threads from 2000 * removing it. 2001 */ 2002 bool 2003 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2004 { 2005 struct pf_kstate *s; 2006 2007 s = pf_find_state_all(key, dir, NULL); 2008 if (s != NULL) { 2009 PF_STATE_UNLOCK(s); 2010 return (true); 2011 } 2012 return (false); 2013 } 2014 2015 struct pf_udp_mapping * 2016 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2017 struct pf_addr *nat_addr, uint16_t nat_port) 2018 { 2019 struct pf_udp_mapping *mapping; 2020 2021 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2022 if (mapping == NULL) 2023 return (NULL); 2024 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 2025 mapping->endpoints[0].port = src_port; 2026 mapping->endpoints[0].af = af; 2027 mapping->endpoints[0].mapping = mapping; 2028 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 2029 mapping->endpoints[1].port = nat_port; 2030 mapping->endpoints[1].af = af; 2031 mapping->endpoints[1].mapping = mapping; 2032 refcount_init(&mapping->refs, 1); 2033 return (mapping); 2034 } 2035 2036 int 2037 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2038 { 2039 struct pf_udpendpointhash *h0, *h1; 2040 struct pf_udp_endpoint *endpoint; 2041 int ret = EEXIST; 2042 2043 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2044 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2045 if (h0 == h1) { 2046 PF_HASHROW_LOCK(h0); 2047 } else if (h0 < h1) { 2048 PF_HASHROW_LOCK(h0); 2049 PF_HASHROW_LOCK(h1); 2050 } else { 2051 PF_HASHROW_LOCK(h1); 2052 PF_HASHROW_LOCK(h0); 2053 } 2054 2055 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2056 if (bcmp(endpoint, &mapping->endpoints[0], 2057 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2058 break; 2059 } 2060 if (endpoint != NULL) 2061 goto cleanup; 2062 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2063 if (bcmp(endpoint, &mapping->endpoints[1], 2064 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2065 break; 2066 } 2067 if (endpoint != NULL) 2068 goto cleanup; 2069 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2070 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2071 ret = 0; 2072 2073 cleanup: 2074 if (h0 != h1) { 2075 PF_HASHROW_UNLOCK(h0); 2076 PF_HASHROW_UNLOCK(h1); 2077 } else { 2078 PF_HASHROW_UNLOCK(h0); 2079 } 2080 return (ret); 2081 } 2082 2083 void 2084 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2085 { 2086 /* refcount is synchronized on the source endpoint's row lock */ 2087 struct pf_udpendpointhash *h0, *h1; 2088 2089 if (mapping == NULL) 2090 return; 2091 2092 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2093 PF_HASHROW_LOCK(h0); 2094 if (refcount_release(&mapping->refs)) { 2095 LIST_REMOVE(&mapping->endpoints[0], entry); 2096 PF_HASHROW_UNLOCK(h0); 2097 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2098 PF_HASHROW_LOCK(h1); 2099 LIST_REMOVE(&mapping->endpoints[1], entry); 2100 PF_HASHROW_UNLOCK(h1); 2101 2102 uma_zfree(V_pf_udp_mapping_z, mapping); 2103 } else { 2104 PF_HASHROW_UNLOCK(h0); 2105 } 2106 } 2107 2108 2109 struct pf_udp_mapping * 2110 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2111 { 2112 struct pf_udpendpointhash *uh; 2113 struct pf_udp_endpoint *endpoint; 2114 2115 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2116 2117 PF_HASHROW_LOCK(uh); 2118 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2119 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2120 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2121 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2122 break; 2123 } 2124 if (endpoint == NULL) { 2125 PF_HASHROW_UNLOCK(uh); 2126 return (NULL); 2127 } 2128 refcount_acquire(&endpoint->mapping->refs); 2129 PF_HASHROW_UNLOCK(uh); 2130 return (endpoint->mapping); 2131 } 2132 /* END state table stuff */ 2133 2134 static void 2135 pf_send(struct pf_send_entry *pfse) 2136 { 2137 2138 PF_SENDQ_LOCK(); 2139 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2140 PF_SENDQ_UNLOCK(); 2141 swi_sched(V_pf_swi_cookie, 0); 2142 } 2143 2144 static bool 2145 pf_isforlocal(struct mbuf *m, int af) 2146 { 2147 switch (af) { 2148 #ifdef INET 2149 case AF_INET: { 2150 struct ip *ip = mtod(m, struct ip *); 2151 2152 return (in_localip(ip->ip_dst)); 2153 } 2154 #endif 2155 #ifdef INET6 2156 case AF_INET6: { 2157 struct ip6_hdr *ip6; 2158 struct in6_ifaddr *ia; 2159 ip6 = mtod(m, struct ip6_hdr *); 2160 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2161 if (ia == NULL) 2162 return (false); 2163 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2164 } 2165 #endif 2166 } 2167 2168 return (false); 2169 } 2170 2171 int 2172 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2173 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2174 { 2175 /* 2176 * ICMP types marked with PF_OUT are typically responses to 2177 * PF_IN, and will match states in the opposite direction. 2178 * PF_IN ICMP types need to match a state with that type. 2179 */ 2180 *icmp_dir = PF_OUT; 2181 *multi = PF_ICMP_MULTI_LINK; 2182 /* Queries (and responses) */ 2183 switch (pd->af) { 2184 #ifdef INET 2185 case AF_INET: 2186 switch (type) { 2187 case ICMP_ECHO: 2188 *icmp_dir = PF_IN; 2189 case ICMP_ECHOREPLY: 2190 *virtual_type = ICMP_ECHO; 2191 *virtual_id = pd->hdr.icmp.icmp_id; 2192 break; 2193 2194 case ICMP_TSTAMP: 2195 *icmp_dir = PF_IN; 2196 case ICMP_TSTAMPREPLY: 2197 *virtual_type = ICMP_TSTAMP; 2198 *virtual_id = pd->hdr.icmp.icmp_id; 2199 break; 2200 2201 case ICMP_IREQ: 2202 *icmp_dir = PF_IN; 2203 case ICMP_IREQREPLY: 2204 *virtual_type = ICMP_IREQ; 2205 *virtual_id = pd->hdr.icmp.icmp_id; 2206 break; 2207 2208 case ICMP_MASKREQ: 2209 *icmp_dir = PF_IN; 2210 case ICMP_MASKREPLY: 2211 *virtual_type = ICMP_MASKREQ; 2212 *virtual_id = pd->hdr.icmp.icmp_id; 2213 break; 2214 2215 case ICMP_IPV6_WHEREAREYOU: 2216 *icmp_dir = PF_IN; 2217 case ICMP_IPV6_IAMHERE: 2218 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2219 *virtual_id = 0; /* Nothing sane to match on! */ 2220 break; 2221 2222 case ICMP_MOBILE_REGREQUEST: 2223 *icmp_dir = PF_IN; 2224 case ICMP_MOBILE_REGREPLY: 2225 *virtual_type = ICMP_MOBILE_REGREQUEST; 2226 *virtual_id = 0; /* Nothing sane to match on! */ 2227 break; 2228 2229 case ICMP_ROUTERSOLICIT: 2230 *icmp_dir = PF_IN; 2231 case ICMP_ROUTERADVERT: 2232 *virtual_type = ICMP_ROUTERSOLICIT; 2233 *virtual_id = 0; /* Nothing sane to match on! */ 2234 break; 2235 2236 /* These ICMP types map to other connections */ 2237 case ICMP_UNREACH: 2238 case ICMP_SOURCEQUENCH: 2239 case ICMP_REDIRECT: 2240 case ICMP_TIMXCEED: 2241 case ICMP_PARAMPROB: 2242 /* These will not be used, but set them anyway */ 2243 *icmp_dir = PF_IN; 2244 *virtual_type = type; 2245 *virtual_id = 0; 2246 HTONS(*virtual_type); 2247 return (1); /* These types match to another state */ 2248 2249 /* 2250 * All remaining ICMP types get their own states, 2251 * and will only match in one direction. 2252 */ 2253 default: 2254 *icmp_dir = PF_IN; 2255 *virtual_type = type; 2256 *virtual_id = 0; 2257 break; 2258 } 2259 break; 2260 #endif /* INET */ 2261 #ifdef INET6 2262 case AF_INET6: 2263 switch (type) { 2264 case ICMP6_ECHO_REQUEST: 2265 *icmp_dir = PF_IN; 2266 case ICMP6_ECHO_REPLY: 2267 *virtual_type = ICMP6_ECHO_REQUEST; 2268 *virtual_id = pd->hdr.icmp6.icmp6_id; 2269 break; 2270 2271 case MLD_LISTENER_QUERY: 2272 case MLD_LISTENER_REPORT: { 2273 /* 2274 * Listener Report can be sent by clients 2275 * without an associated Listener Query. 2276 * In addition to that, when Report is sent as a 2277 * reply to a Query its source and destination 2278 * address are different. 2279 */ 2280 *icmp_dir = PF_IN; 2281 *virtual_type = MLD_LISTENER_QUERY; 2282 *virtual_id = 0; 2283 break; 2284 } 2285 case MLD_MTRACE: 2286 *icmp_dir = PF_IN; 2287 case MLD_MTRACE_RESP: 2288 *virtual_type = MLD_MTRACE; 2289 *virtual_id = 0; /* Nothing sane to match on! */ 2290 break; 2291 2292 case ND_NEIGHBOR_SOLICIT: 2293 *icmp_dir = PF_IN; 2294 case ND_NEIGHBOR_ADVERT: { 2295 *virtual_type = ND_NEIGHBOR_SOLICIT; 2296 *virtual_id = 0; 2297 break; 2298 } 2299 2300 /* 2301 * These ICMP types map to other connections. 2302 * ND_REDIRECT can't be in this list because the triggering 2303 * packet header is optional. 2304 */ 2305 case ICMP6_DST_UNREACH: 2306 case ICMP6_PACKET_TOO_BIG: 2307 case ICMP6_TIME_EXCEEDED: 2308 case ICMP6_PARAM_PROB: 2309 /* These will not be used, but set them anyway */ 2310 *icmp_dir = PF_IN; 2311 *virtual_type = type; 2312 *virtual_id = 0; 2313 HTONS(*virtual_type); 2314 return (1); /* These types match to another state */ 2315 /* 2316 * All remaining ICMP6 types get their own states, 2317 * and will only match in one direction. 2318 */ 2319 default: 2320 *icmp_dir = PF_IN; 2321 *virtual_type = type; 2322 *virtual_id = 0; 2323 break; 2324 } 2325 break; 2326 #endif /* INET6 */ 2327 } 2328 HTONS(*virtual_type); 2329 return (0); /* These types match to their own state */ 2330 } 2331 2332 void 2333 pf_intr(void *v) 2334 { 2335 struct epoch_tracker et; 2336 struct pf_send_head queue; 2337 struct pf_send_entry *pfse, *next; 2338 2339 CURVNET_SET((struct vnet *)v); 2340 2341 PF_SENDQ_LOCK(); 2342 queue = V_pf_sendqueue; 2343 STAILQ_INIT(&V_pf_sendqueue); 2344 PF_SENDQ_UNLOCK(); 2345 2346 NET_EPOCH_ENTER(et); 2347 2348 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2349 switch (pfse->pfse_type) { 2350 #ifdef INET 2351 case PFSE_IP: { 2352 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2353 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2354 ("%s: rcvif != loif", __func__)); 2355 2356 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2357 pfse->pfse_m->m_pkthdr.csum_flags |= 2358 CSUM_IP_VALID | CSUM_IP_CHECKED | 2359 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2360 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2361 ip_input(pfse->pfse_m); 2362 } else { 2363 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2364 NULL); 2365 } 2366 break; 2367 } 2368 case PFSE_ICMP: 2369 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2370 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2371 break; 2372 #endif /* INET */ 2373 #ifdef INET6 2374 case PFSE_IP6: 2375 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2376 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2377 ("%s: rcvif != loif", __func__)); 2378 2379 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2380 M_LOOP; 2381 pfse->pfse_m->m_pkthdr.csum_flags |= 2382 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2383 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2384 ip6_input(pfse->pfse_m); 2385 } else { 2386 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2387 NULL, NULL); 2388 } 2389 break; 2390 case PFSE_ICMP6: 2391 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2392 pfse->icmpopts.code, pfse->icmpopts.mtu); 2393 break; 2394 #endif /* INET6 */ 2395 default: 2396 panic("%s: unknown type", __func__); 2397 } 2398 free(pfse, M_PFTEMP); 2399 } 2400 NET_EPOCH_EXIT(et); 2401 CURVNET_RESTORE(); 2402 } 2403 2404 #define pf_purge_thread_period (hz / 10) 2405 2406 #ifdef PF_WANT_32_TO_64_COUNTER 2407 static void 2408 pf_status_counter_u64_periodic(void) 2409 { 2410 2411 PF_RULES_RASSERT(); 2412 2413 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2414 return; 2415 } 2416 2417 for (int i = 0; i < FCNT_MAX; i++) { 2418 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2419 } 2420 } 2421 2422 static void 2423 pf_kif_counter_u64_periodic(void) 2424 { 2425 struct pfi_kkif *kif; 2426 size_t r, run; 2427 2428 PF_RULES_RASSERT(); 2429 2430 if (__predict_false(V_pf_allkifcount == 0)) { 2431 return; 2432 } 2433 2434 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2435 return; 2436 } 2437 2438 run = V_pf_allkifcount / 10; 2439 if (run < 5) 2440 run = 5; 2441 2442 for (r = 0; r < run; r++) { 2443 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2444 if (kif == NULL) { 2445 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2446 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2447 break; 2448 } 2449 2450 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2451 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2452 2453 for (int i = 0; i < 2; i++) { 2454 for (int j = 0; j < 2; j++) { 2455 for (int k = 0; k < 2; k++) { 2456 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2457 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2458 } 2459 } 2460 } 2461 } 2462 } 2463 2464 static void 2465 pf_rule_counter_u64_periodic(void) 2466 { 2467 struct pf_krule *rule; 2468 size_t r, run; 2469 2470 PF_RULES_RASSERT(); 2471 2472 if (__predict_false(V_pf_allrulecount == 0)) { 2473 return; 2474 } 2475 2476 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2477 return; 2478 } 2479 2480 run = V_pf_allrulecount / 10; 2481 if (run < 5) 2482 run = 5; 2483 2484 for (r = 0; r < run; r++) { 2485 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2486 if (rule == NULL) { 2487 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2488 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2489 break; 2490 } 2491 2492 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2493 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2494 2495 pf_counter_u64_periodic(&rule->evaluations); 2496 for (int i = 0; i < 2; i++) { 2497 pf_counter_u64_periodic(&rule->packets[i]); 2498 pf_counter_u64_periodic(&rule->bytes[i]); 2499 } 2500 } 2501 } 2502 2503 static void 2504 pf_counter_u64_periodic_main(void) 2505 { 2506 PF_RULES_RLOCK_TRACKER; 2507 2508 V_pf_counter_periodic_iter++; 2509 2510 PF_RULES_RLOCK(); 2511 pf_counter_u64_critical_enter(); 2512 pf_status_counter_u64_periodic(); 2513 pf_kif_counter_u64_periodic(); 2514 pf_rule_counter_u64_periodic(); 2515 pf_counter_u64_critical_exit(); 2516 PF_RULES_RUNLOCK(); 2517 } 2518 #else 2519 #define pf_counter_u64_periodic_main() do { } while (0) 2520 #endif 2521 2522 void 2523 pf_purge_thread(void *unused __unused) 2524 { 2525 struct epoch_tracker et; 2526 2527 VNET_ITERATOR_DECL(vnet_iter); 2528 2529 sx_xlock(&pf_end_lock); 2530 while (pf_end_threads == 0) { 2531 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2532 2533 VNET_LIST_RLOCK(); 2534 NET_EPOCH_ENTER(et); 2535 VNET_FOREACH(vnet_iter) { 2536 CURVNET_SET(vnet_iter); 2537 2538 /* Wait until V_pf_default_rule is initialized. */ 2539 if (V_pf_vnet_active == 0) { 2540 CURVNET_RESTORE(); 2541 continue; 2542 } 2543 2544 pf_counter_u64_periodic_main(); 2545 2546 /* 2547 * Process 1/interval fraction of the state 2548 * table every run. 2549 */ 2550 V_pf_purge_idx = 2551 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2552 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2553 2554 /* 2555 * Purge other expired types every 2556 * PFTM_INTERVAL seconds. 2557 */ 2558 if (V_pf_purge_idx == 0) { 2559 /* 2560 * Order is important: 2561 * - states and src nodes reference rules 2562 * - states and rules reference kifs 2563 */ 2564 pf_purge_expired_fragments(); 2565 pf_purge_expired_src_nodes(); 2566 pf_purge_unlinked_rules(); 2567 pfi_kkif_purge(); 2568 } 2569 CURVNET_RESTORE(); 2570 } 2571 NET_EPOCH_EXIT(et); 2572 VNET_LIST_RUNLOCK(); 2573 } 2574 2575 pf_end_threads++; 2576 sx_xunlock(&pf_end_lock); 2577 kproc_exit(0); 2578 } 2579 2580 void 2581 pf_unload_vnet_purge(void) 2582 { 2583 2584 /* 2585 * To cleanse up all kifs and rules we need 2586 * two runs: first one clears reference flags, 2587 * then pf_purge_expired_states() doesn't 2588 * raise them, and then second run frees. 2589 */ 2590 pf_purge_unlinked_rules(); 2591 pfi_kkif_purge(); 2592 2593 /* 2594 * Now purge everything. 2595 */ 2596 pf_purge_expired_states(0, V_pf_hashmask); 2597 pf_purge_fragments(UINT_MAX); 2598 pf_purge_expired_src_nodes(); 2599 2600 /* 2601 * Now all kifs & rules should be unreferenced, 2602 * thus should be successfully freed. 2603 */ 2604 pf_purge_unlinked_rules(); 2605 pfi_kkif_purge(); 2606 } 2607 2608 u_int32_t 2609 pf_state_expires(const struct pf_kstate *state) 2610 { 2611 u_int32_t timeout; 2612 u_int32_t start; 2613 u_int32_t end; 2614 u_int32_t states; 2615 2616 /* handle all PFTM_* > PFTM_MAX here */ 2617 if (state->timeout == PFTM_PURGE) 2618 return (time_uptime); 2619 KASSERT(state->timeout != PFTM_UNLINKED, 2620 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2621 KASSERT((state->timeout < PFTM_MAX), 2622 ("pf_state_expires: timeout > PFTM_MAX")); 2623 timeout = state->rule->timeout[state->timeout]; 2624 if (!timeout) 2625 timeout = V_pf_default_rule.timeout[state->timeout]; 2626 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2627 if (start && state->rule != &V_pf_default_rule) { 2628 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2629 states = counter_u64_fetch(state->rule->states_cur); 2630 } else { 2631 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2632 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2633 states = V_pf_status.states; 2634 } 2635 if (end && states > start && start < end) { 2636 if (states < end) { 2637 timeout = (u_int64_t)timeout * (end - states) / 2638 (end - start); 2639 return ((state->expire / 1000) + timeout); 2640 } 2641 else 2642 return (time_uptime); 2643 } 2644 return ((state->expire / 1000) + timeout); 2645 } 2646 2647 void 2648 pf_purge_expired_src_nodes(void) 2649 { 2650 struct pf_ksrc_node_list freelist; 2651 struct pf_srchash *sh; 2652 struct pf_ksrc_node *cur, *next; 2653 int i; 2654 2655 LIST_INIT(&freelist); 2656 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2657 PF_HASHROW_LOCK(sh); 2658 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2659 if (cur->states == 0 && cur->expire <= time_uptime) { 2660 pf_unlink_src_node(cur); 2661 LIST_INSERT_HEAD(&freelist, cur, entry); 2662 } else if (cur->rule != NULL) 2663 cur->rule->rule_ref |= PFRULE_REFS; 2664 PF_HASHROW_UNLOCK(sh); 2665 } 2666 2667 pf_free_src_nodes(&freelist); 2668 2669 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2670 } 2671 2672 static void 2673 pf_src_tree_remove_state(struct pf_kstate *s) 2674 { 2675 uint32_t timeout; 2676 2677 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2678 s->rule->timeout[PFTM_SRC_NODE] : 2679 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2680 2681 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 2682 if (s->sns[sn_type] == NULL) 2683 continue; 2684 PF_SRC_NODE_LOCK(s->sns[sn_type]); 2685 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 2686 --(s->sns[sn_type]->conn); 2687 if (--(s->sns[sn_type]->states) == 0) 2688 s->sns[sn_type]->expire = time_uptime + timeout; 2689 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 2690 s->sns[sn_type] = NULL; 2691 } 2692 2693 } 2694 2695 /* 2696 * Unlink and potentilly free a state. Function may be 2697 * called with ID hash row locked, but always returns 2698 * unlocked, since it needs to go through key hash locking. 2699 */ 2700 int 2701 pf_unlink_state(struct pf_kstate *s) 2702 { 2703 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2704 2705 NET_EPOCH_ASSERT(); 2706 PF_HASHROW_ASSERT(ih); 2707 2708 if (s->timeout == PFTM_UNLINKED) { 2709 /* 2710 * State is being processed 2711 * by pf_unlink_state() in 2712 * an other thread. 2713 */ 2714 PF_HASHROW_UNLOCK(ih); 2715 return (0); /* XXXGL: undefined actually */ 2716 } 2717 2718 if (s->src.state == PF_TCPS_PROXY_DST) { 2719 /* XXX wire key the right one? */ 2720 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2721 &s->key[PF_SK_WIRE]->addr[1], 2722 &s->key[PF_SK_WIRE]->addr[0], 2723 s->key[PF_SK_WIRE]->port[1], 2724 s->key[PF_SK_WIRE]->port[0], 2725 s->src.seqhi, s->src.seqlo + 1, 2726 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2727 s->act.rtableid); 2728 } 2729 2730 LIST_REMOVE(s, entry); 2731 pf_src_tree_remove_state(s); 2732 2733 if (V_pfsync_delete_state_ptr != NULL) 2734 V_pfsync_delete_state_ptr(s); 2735 2736 STATE_DEC_COUNTERS(s); 2737 2738 s->timeout = PFTM_UNLINKED; 2739 2740 /* Ensure we remove it from the list of halfopen states, if needed. */ 2741 if (s->key[PF_SK_STACK] != NULL && 2742 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2743 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2744 2745 PF_HASHROW_UNLOCK(ih); 2746 2747 pf_detach_state(s); 2748 2749 pf_udp_mapping_release(s->udp_mapping); 2750 2751 /* pf_state_insert() initialises refs to 2 */ 2752 return (pf_release_staten(s, 2)); 2753 } 2754 2755 struct pf_kstate * 2756 pf_alloc_state(int flags) 2757 { 2758 2759 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2760 } 2761 2762 void 2763 pf_free_state(struct pf_kstate *cur) 2764 { 2765 struct pf_krule_item *ri; 2766 2767 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2768 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2769 cur->timeout)); 2770 2771 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2772 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2773 free(ri, M_PF_RULE_ITEM); 2774 } 2775 2776 pf_normalize_tcp_cleanup(cur); 2777 uma_zfree(V_pf_state_z, cur); 2778 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2779 } 2780 2781 /* 2782 * Called only from pf_purge_thread(), thus serialized. 2783 */ 2784 static u_int 2785 pf_purge_expired_states(u_int i, int maxcheck) 2786 { 2787 struct pf_idhash *ih; 2788 struct pf_kstate *s; 2789 struct pf_krule_item *mrm; 2790 size_t count __unused; 2791 2792 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2793 2794 /* 2795 * Go through hash and unlink states that expire now. 2796 */ 2797 while (maxcheck > 0) { 2798 count = 0; 2799 ih = &V_pf_idhash[i]; 2800 2801 /* only take the lock if we expect to do work */ 2802 if (!LIST_EMPTY(&ih->states)) { 2803 relock: 2804 PF_HASHROW_LOCK(ih); 2805 LIST_FOREACH(s, &ih->states, entry) { 2806 if (pf_state_expires(s) <= time_uptime) { 2807 V_pf_status.states -= 2808 pf_unlink_state(s); 2809 goto relock; 2810 } 2811 s->rule->rule_ref |= PFRULE_REFS; 2812 if (s->nat_rule != NULL) 2813 s->nat_rule->rule_ref |= PFRULE_REFS; 2814 if (s->anchor != NULL) 2815 s->anchor->rule_ref |= PFRULE_REFS; 2816 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2817 SLIST_FOREACH(mrm, &s->match_rules, entry) 2818 mrm->r->rule_ref |= PFRULE_REFS; 2819 if (s->act.rt_kif) 2820 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2821 count++; 2822 } 2823 PF_HASHROW_UNLOCK(ih); 2824 } 2825 2826 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2827 2828 /* Return when we hit end of hash. */ 2829 if (++i > V_pf_hashmask) { 2830 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2831 return (0); 2832 } 2833 2834 maxcheck--; 2835 } 2836 2837 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2838 2839 return (i); 2840 } 2841 2842 static void 2843 pf_purge_unlinked_rules(void) 2844 { 2845 struct pf_krulequeue tmpq; 2846 struct pf_krule *r, *r1; 2847 2848 /* 2849 * If we have overloading task pending, then we'd 2850 * better skip purging this time. There is a tiny 2851 * probability that overloading task references 2852 * an already unlinked rule. 2853 */ 2854 PF_OVERLOADQ_LOCK(); 2855 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2856 PF_OVERLOADQ_UNLOCK(); 2857 return; 2858 } 2859 PF_OVERLOADQ_UNLOCK(); 2860 2861 /* 2862 * Do naive mark-and-sweep garbage collecting of old rules. 2863 * Reference flag is raised by pf_purge_expired_states() 2864 * and pf_purge_expired_src_nodes(). 2865 * 2866 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2867 * use a temporary queue. 2868 */ 2869 TAILQ_INIT(&tmpq); 2870 PF_UNLNKDRULES_LOCK(); 2871 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2872 if (!(r->rule_ref & PFRULE_REFS)) { 2873 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2874 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2875 } else 2876 r->rule_ref &= ~PFRULE_REFS; 2877 } 2878 PF_UNLNKDRULES_UNLOCK(); 2879 2880 if (!TAILQ_EMPTY(&tmpq)) { 2881 PF_CONFIG_LOCK(); 2882 PF_RULES_WLOCK(); 2883 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2884 TAILQ_REMOVE(&tmpq, r, entries); 2885 pf_free_rule(r); 2886 } 2887 PF_RULES_WUNLOCK(); 2888 PF_CONFIG_UNLOCK(); 2889 } 2890 } 2891 2892 void 2893 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2894 { 2895 switch (af) { 2896 #ifdef INET 2897 case AF_INET: { 2898 u_int32_t a = ntohl(addr->addr32[0]); 2899 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2900 (a>>8)&255, a&255); 2901 if (p) { 2902 p = ntohs(p); 2903 printf(":%u", p); 2904 } 2905 break; 2906 } 2907 #endif /* INET */ 2908 #ifdef INET6 2909 case AF_INET6: { 2910 u_int16_t b; 2911 u_int8_t i, curstart, curend, maxstart, maxend; 2912 curstart = curend = maxstart = maxend = 255; 2913 for (i = 0; i < 8; i++) { 2914 if (!addr->addr16[i]) { 2915 if (curstart == 255) 2916 curstart = i; 2917 curend = i; 2918 } else { 2919 if ((curend - curstart) > 2920 (maxend - maxstart)) { 2921 maxstart = curstart; 2922 maxend = curend; 2923 } 2924 curstart = curend = 255; 2925 } 2926 } 2927 if ((curend - curstart) > 2928 (maxend - maxstart)) { 2929 maxstart = curstart; 2930 maxend = curend; 2931 } 2932 for (i = 0; i < 8; i++) { 2933 if (i >= maxstart && i <= maxend) { 2934 if (i == 0) 2935 printf(":"); 2936 if (i == maxend) 2937 printf(":"); 2938 } else { 2939 b = ntohs(addr->addr16[i]); 2940 printf("%x", b); 2941 if (i < 7) 2942 printf(":"); 2943 } 2944 } 2945 if (p) { 2946 p = ntohs(p); 2947 printf("[%u]", p); 2948 } 2949 break; 2950 } 2951 #endif /* INET6 */ 2952 } 2953 } 2954 2955 void 2956 pf_print_state(struct pf_kstate *s) 2957 { 2958 pf_print_state_parts(s, NULL, NULL); 2959 } 2960 2961 static void 2962 pf_print_state_parts(struct pf_kstate *s, 2963 struct pf_state_key *skwp, struct pf_state_key *sksp) 2964 { 2965 struct pf_state_key *skw, *sks; 2966 u_int8_t proto, dir; 2967 2968 /* Do our best to fill these, but they're skipped if NULL */ 2969 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2970 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2971 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2972 dir = s ? s->direction : 0; 2973 2974 switch (proto) { 2975 case IPPROTO_IPV4: 2976 printf("IPv4"); 2977 break; 2978 case IPPROTO_IPV6: 2979 printf("IPv6"); 2980 break; 2981 case IPPROTO_TCP: 2982 printf("TCP"); 2983 break; 2984 case IPPROTO_UDP: 2985 printf("UDP"); 2986 break; 2987 case IPPROTO_ICMP: 2988 printf("ICMP"); 2989 break; 2990 case IPPROTO_ICMPV6: 2991 printf("ICMPv6"); 2992 break; 2993 default: 2994 printf("%u", proto); 2995 break; 2996 } 2997 switch (dir) { 2998 case PF_IN: 2999 printf(" in"); 3000 break; 3001 case PF_OUT: 3002 printf(" out"); 3003 break; 3004 } 3005 if (skw) { 3006 printf(" wire: "); 3007 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3008 printf(" "); 3009 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3010 } 3011 if (sks) { 3012 printf(" stack: "); 3013 if (sks != skw) { 3014 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3015 printf(" "); 3016 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3017 } else 3018 printf("-"); 3019 } 3020 if (s) { 3021 if (proto == IPPROTO_TCP) { 3022 printf(" [lo=%u high=%u win=%u modulator=%u", 3023 s->src.seqlo, s->src.seqhi, 3024 s->src.max_win, s->src.seqdiff); 3025 if (s->src.wscale && s->dst.wscale) 3026 printf(" wscale=%u", 3027 s->src.wscale & PF_WSCALE_MASK); 3028 printf("]"); 3029 printf(" [lo=%u high=%u win=%u modulator=%u", 3030 s->dst.seqlo, s->dst.seqhi, 3031 s->dst.max_win, s->dst.seqdiff); 3032 if (s->src.wscale && s->dst.wscale) 3033 printf(" wscale=%u", 3034 s->dst.wscale & PF_WSCALE_MASK); 3035 printf("]"); 3036 } 3037 printf(" %u:%u", s->src.state, s->dst.state); 3038 if (s->rule) 3039 printf(" @%d", s->rule->nr); 3040 } 3041 } 3042 3043 void 3044 pf_print_flags(uint16_t f) 3045 { 3046 if (f) 3047 printf(" "); 3048 if (f & TH_FIN) 3049 printf("F"); 3050 if (f & TH_SYN) 3051 printf("S"); 3052 if (f & TH_RST) 3053 printf("R"); 3054 if (f & TH_PUSH) 3055 printf("P"); 3056 if (f & TH_ACK) 3057 printf("A"); 3058 if (f & TH_URG) 3059 printf("U"); 3060 if (f & TH_ECE) 3061 printf("E"); 3062 if (f & TH_CWR) 3063 printf("W"); 3064 if (f & TH_AE) 3065 printf("e"); 3066 } 3067 3068 #define PF_SET_SKIP_STEPS(i) \ 3069 do { \ 3070 while (head[i] != cur) { \ 3071 head[i]->skip[i] = cur; \ 3072 head[i] = TAILQ_NEXT(head[i], entries); \ 3073 } \ 3074 } while (0) 3075 3076 void 3077 pf_calc_skip_steps(struct pf_krulequeue *rules) 3078 { 3079 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3080 int i; 3081 3082 cur = TAILQ_FIRST(rules); 3083 prev = cur; 3084 for (i = 0; i < PF_SKIP_COUNT; ++i) 3085 head[i] = cur; 3086 while (cur != NULL) { 3087 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3088 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3089 if (cur->direction != prev->direction) 3090 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3091 if (cur->af != prev->af) 3092 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3093 if (cur->proto != prev->proto) 3094 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3095 if (cur->src.neg != prev->src.neg || 3096 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3097 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3098 if (cur->dst.neg != prev->dst.neg || 3099 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3100 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3101 if (cur->src.port[0] != prev->src.port[0] || 3102 cur->src.port[1] != prev->src.port[1] || 3103 cur->src.port_op != prev->src.port_op) 3104 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3105 if (cur->dst.port[0] != prev->dst.port[0] || 3106 cur->dst.port[1] != prev->dst.port[1] || 3107 cur->dst.port_op != prev->dst.port_op) 3108 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3109 3110 prev = cur; 3111 cur = TAILQ_NEXT(cur, entries); 3112 } 3113 for (i = 0; i < PF_SKIP_COUNT; ++i) 3114 PF_SET_SKIP_STEPS(i); 3115 } 3116 3117 int 3118 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3119 { 3120 if (aw1->type != aw2->type) 3121 return (1); 3122 switch (aw1->type) { 3123 case PF_ADDR_ADDRMASK: 3124 case PF_ADDR_RANGE: 3125 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3126 return (1); 3127 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3128 return (1); 3129 return (0); 3130 case PF_ADDR_DYNIFTL: 3131 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3132 case PF_ADDR_NONE: 3133 case PF_ADDR_NOROUTE: 3134 case PF_ADDR_URPFFAILED: 3135 return (0); 3136 case PF_ADDR_TABLE: 3137 return (aw1->p.tbl != aw2->p.tbl); 3138 default: 3139 printf("invalid address type: %d\n", aw1->type); 3140 return (1); 3141 } 3142 } 3143 3144 /** 3145 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3146 * header isn't always a full checksum. In some cases (i.e. output) it's a 3147 * pseudo-header checksum, which is a partial checksum over src/dst IP 3148 * addresses, protocol number and length. 3149 * 3150 * That means we have the following cases: 3151 * * Input or forwarding: we don't have TSO, the checksum fields are full 3152 * checksums, we need to update the checksum whenever we change anything. 3153 * * Output (i.e. the checksum is a pseudo-header checksum): 3154 * x The field being updated is src/dst address or affects the length of 3155 * the packet. We need to update the pseudo-header checksum (note that this 3156 * checksum is not ones' complement). 3157 * x Some other field is being modified (e.g. src/dst port numbers): We 3158 * don't have to update anything. 3159 **/ 3160 u_int16_t 3161 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3162 { 3163 u_int32_t x; 3164 3165 x = cksum + old - new; 3166 x = (x + (x >> 16)) & 0xffff; 3167 3168 /* optimise: eliminate a branch when not udp */ 3169 if (udp && cksum == 0x0000) 3170 return cksum; 3171 if (udp && x == 0x0000) 3172 x = 0xffff; 3173 3174 return (u_int16_t)(x); 3175 } 3176 3177 static void 3178 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3179 u_int8_t udp) 3180 { 3181 u_int16_t old = htons(hi ? (*f << 8) : *f); 3182 u_int16_t new = htons(hi ? ( v << 8) : v); 3183 3184 if (*f == v) 3185 return; 3186 3187 *f = v; 3188 3189 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3190 return; 3191 3192 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3193 } 3194 3195 void 3196 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3197 bool hi, u_int8_t udp) 3198 { 3199 u_int8_t *fb = (u_int8_t *)f; 3200 u_int8_t *vb = (u_int8_t *)&v; 3201 3202 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3203 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3204 } 3205 3206 void 3207 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3208 bool hi, u_int8_t udp) 3209 { 3210 u_int8_t *fb = (u_int8_t *)f; 3211 u_int8_t *vb = (u_int8_t *)&v; 3212 3213 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3214 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3215 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3216 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3217 } 3218 3219 u_int16_t 3220 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3221 u_int16_t new, u_int8_t udp) 3222 { 3223 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3224 return (cksum); 3225 3226 return (pf_cksum_fixup(cksum, old, new, udp)); 3227 } 3228 3229 static void 3230 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3231 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3232 sa_family_t af, sa_family_t naf) 3233 { 3234 struct pf_addr ao; 3235 u_int16_t po; 3236 3237 PF_ACPY(&ao, a, af); 3238 if (af == naf) 3239 PF_ACPY(a, an, af); 3240 3241 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3242 *pc = ~*pc; 3243 3244 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3245 return; 3246 po = *p; 3247 *p = pn; 3248 3249 switch (af) { 3250 #ifdef INET 3251 case AF_INET: 3252 switch (naf) { 3253 case AF_INET: 3254 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3255 ao.addr16[0], an->addr16[0], 0), 3256 ao.addr16[1], an->addr16[1], 0); 3257 *p = pn; 3258 3259 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3260 ao.addr16[0], an->addr16[0], u), 3261 ao.addr16[1], an->addr16[1], u); 3262 3263 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3264 break; 3265 #ifdef INET6 3266 case AF_INET6: 3267 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3268 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3269 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3270 ao.addr16[0], an->addr16[0], u), 3271 ao.addr16[1], an->addr16[1], u), 3272 0, an->addr16[2], u), 3273 0, an->addr16[3], u), 3274 0, an->addr16[4], u), 3275 0, an->addr16[5], u), 3276 0, an->addr16[6], u), 3277 0, an->addr16[7], u), 3278 po, pn, u); 3279 3280 /* XXXKP TODO *ic checksum? */ 3281 break; 3282 #endif /* INET6 */ 3283 } 3284 break; 3285 #endif /* INET */ 3286 #ifdef INET6 3287 case AF_INET6: 3288 switch (naf) { 3289 #ifdef INET 3290 case AF_INET: 3291 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3292 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3293 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3294 ao.addr16[0], an->addr16[0], u), 3295 ao.addr16[1], an->addr16[1], u), 3296 ao.addr16[2], 0, u), 3297 ao.addr16[3], 0, u), 3298 ao.addr16[4], 0, u), 3299 ao.addr16[5], 0, u), 3300 ao.addr16[6], 0, u), 3301 ao.addr16[7], 0, u), 3302 po, pn, u); 3303 3304 /* XXXKP TODO *ic checksum? */ 3305 break; 3306 #endif /* INET */ 3307 case AF_INET6: 3308 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3309 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3310 pf_cksum_fixup(pf_cksum_fixup(*pc, 3311 ao.addr16[0], an->addr16[0], u), 3312 ao.addr16[1], an->addr16[1], u), 3313 ao.addr16[2], an->addr16[2], u), 3314 ao.addr16[3], an->addr16[3], u), 3315 ao.addr16[4], an->addr16[4], u), 3316 ao.addr16[5], an->addr16[5], u), 3317 ao.addr16[6], an->addr16[6], u), 3318 ao.addr16[7], an->addr16[7], u); 3319 3320 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3321 break; 3322 } 3323 break; 3324 #endif /* INET6 */ 3325 } 3326 3327 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3328 CSUM_DELAY_DATA_IPV6)) { 3329 *pc = ~*pc; 3330 if (! *pc) 3331 *pc = 0xffff; 3332 } 3333 } 3334 3335 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3336 void 3337 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3338 { 3339 u_int32_t ao; 3340 3341 memcpy(&ao, a, sizeof(ao)); 3342 memcpy(a, &an, sizeof(u_int32_t)); 3343 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3344 ao % 65536, an % 65536, u); 3345 } 3346 3347 void 3348 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3349 { 3350 u_int32_t ao; 3351 3352 memcpy(&ao, a, sizeof(ao)); 3353 memcpy(a, &an, sizeof(u_int32_t)); 3354 3355 *c = pf_proto_cksum_fixup(m, 3356 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3357 ao % 65536, an % 65536, udp); 3358 } 3359 3360 #ifdef INET6 3361 static void 3362 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3363 { 3364 struct pf_addr ao; 3365 3366 PF_ACPY(&ao, a, AF_INET6); 3367 PF_ACPY(a, an, AF_INET6); 3368 3369 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3370 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3371 pf_cksum_fixup(pf_cksum_fixup(*c, 3372 ao.addr16[0], an->addr16[0], u), 3373 ao.addr16[1], an->addr16[1], u), 3374 ao.addr16[2], an->addr16[2], u), 3375 ao.addr16[3], an->addr16[3], u), 3376 ao.addr16[4], an->addr16[4], u), 3377 ao.addr16[5], an->addr16[5], u), 3378 ao.addr16[6], an->addr16[6], u), 3379 ao.addr16[7], an->addr16[7], u); 3380 } 3381 #endif /* INET6 */ 3382 3383 static void 3384 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3385 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3386 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3387 { 3388 struct pf_addr oia, ooa; 3389 3390 PF_ACPY(&oia, ia, af); 3391 if (oa) 3392 PF_ACPY(&ooa, oa, af); 3393 3394 /* Change inner protocol port, fix inner protocol checksum. */ 3395 if (ip != NULL) { 3396 u_int16_t oip = *ip; 3397 u_int32_t opc; 3398 3399 if (pc != NULL) 3400 opc = *pc; 3401 *ip = np; 3402 if (pc != NULL) 3403 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3404 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3405 if (pc != NULL) 3406 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3407 } 3408 /* Change inner ip address, fix inner ip and icmp checksums. */ 3409 PF_ACPY(ia, na, af); 3410 switch (af) { 3411 #ifdef INET 3412 case AF_INET: { 3413 u_int32_t oh2c = *h2c; 3414 3415 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3416 oia.addr16[0], ia->addr16[0], 0), 3417 oia.addr16[1], ia->addr16[1], 0); 3418 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3419 oia.addr16[0], ia->addr16[0], 0), 3420 oia.addr16[1], ia->addr16[1], 0); 3421 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3422 break; 3423 } 3424 #endif /* INET */ 3425 #ifdef INET6 3426 case AF_INET6: 3427 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3428 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3429 pf_cksum_fixup(pf_cksum_fixup(*ic, 3430 oia.addr16[0], ia->addr16[0], u), 3431 oia.addr16[1], ia->addr16[1], u), 3432 oia.addr16[2], ia->addr16[2], u), 3433 oia.addr16[3], ia->addr16[3], u), 3434 oia.addr16[4], ia->addr16[4], u), 3435 oia.addr16[5], ia->addr16[5], u), 3436 oia.addr16[6], ia->addr16[6], u), 3437 oia.addr16[7], ia->addr16[7], u); 3438 break; 3439 #endif /* INET6 */ 3440 } 3441 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3442 if (oa) { 3443 PF_ACPY(oa, na, af); 3444 switch (af) { 3445 #ifdef INET 3446 case AF_INET: 3447 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3448 ooa.addr16[0], oa->addr16[0], 0), 3449 ooa.addr16[1], oa->addr16[1], 0); 3450 break; 3451 #endif /* INET */ 3452 #ifdef INET6 3453 case AF_INET6: 3454 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3455 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3456 pf_cksum_fixup(pf_cksum_fixup(*ic, 3457 ooa.addr16[0], oa->addr16[0], u), 3458 ooa.addr16[1], oa->addr16[1], u), 3459 ooa.addr16[2], oa->addr16[2], u), 3460 ooa.addr16[3], oa->addr16[3], u), 3461 ooa.addr16[4], oa->addr16[4], u), 3462 ooa.addr16[5], oa->addr16[5], u), 3463 ooa.addr16[6], oa->addr16[6], u), 3464 ooa.addr16[7], oa->addr16[7], u); 3465 break; 3466 #endif /* INET6 */ 3467 } 3468 } 3469 } 3470 3471 int 3472 pf_translate_af(struct pf_pdesc *pd) 3473 { 3474 #if defined(INET) && defined(INET6) 3475 struct mbuf *mp; 3476 struct ip *ip4; 3477 struct ip6_hdr *ip6; 3478 struct icmp6_hdr *icmp; 3479 struct m_tag *mtag; 3480 struct pf_fragment_tag *ftag; 3481 int hlen; 3482 3483 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3484 3485 /* trim the old header */ 3486 m_adj(pd->m, pd->off); 3487 3488 /* prepend a new one */ 3489 M_PREPEND(pd->m, hlen, M_NOWAIT); 3490 if (pd->m == NULL) 3491 return (-1); 3492 3493 switch (pd->naf) { 3494 case AF_INET: 3495 ip4 = mtod(pd->m, struct ip *); 3496 bzero(ip4, hlen); 3497 ip4->ip_v = IPVERSION; 3498 ip4->ip_hl = hlen >> 2; 3499 ip4->ip_tos = pd->tos; 3500 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3501 ip_fillid(ip4); 3502 ip4->ip_ttl = pd->ttl; 3503 ip4->ip_p = pd->proto; 3504 ip4->ip_src = pd->nsaddr.v4; 3505 ip4->ip_dst = pd->ndaddr.v4; 3506 pd->src = (struct pf_addr *)&ip4->ip_src; 3507 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3508 pd->off = sizeof(struct ip); 3509 break; 3510 case AF_INET6: 3511 ip6 = mtod(pd->m, struct ip6_hdr *); 3512 bzero(ip6, hlen); 3513 ip6->ip6_vfc = IPV6_VERSION; 3514 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3515 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3516 ip6->ip6_nxt = pd->proto; 3517 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3518 ip6->ip6_hlim = IPV6_DEFHLIM; 3519 else 3520 ip6->ip6_hlim = pd->ttl; 3521 ip6->ip6_src = pd->nsaddr.v6; 3522 ip6->ip6_dst = pd->ndaddr.v6; 3523 pd->src = (struct pf_addr *)&ip6->ip6_src; 3524 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3525 pd->off = sizeof(struct ip6_hdr); 3526 3527 /* 3528 * If we're dealing with a reassembled packet we need to adjust 3529 * the header length from the IPv4 header size to IPv6 header 3530 * size. 3531 */ 3532 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3533 if (mtag) { 3534 ftag = (struct pf_fragment_tag *)(mtag + 1); 3535 ftag->ft_hdrlen = sizeof(*ip6); 3536 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3537 sizeof(struct ip) + sizeof(struct ip6_frag); 3538 } 3539 break; 3540 default: 3541 return (-1); 3542 } 3543 3544 /* recalculate icmp/icmp6 checksums */ 3545 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3546 int off; 3547 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3548 NULL) { 3549 pd->m = NULL; 3550 return (-1); 3551 } 3552 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3553 icmp->icmp6_cksum = 0; 3554 icmp->icmp6_cksum = pd->naf == AF_INET ? 3555 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3556 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3557 ntohs(ip6->ip6_plen)); 3558 } 3559 #endif /* INET && INET6 */ 3560 3561 return (0); 3562 } 3563 3564 int 3565 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3566 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3567 sa_family_t af, sa_family_t naf) 3568 { 3569 #if defined(INET) && defined(INET6) 3570 struct mbuf *n = NULL; 3571 struct ip *ip4; 3572 struct ip6_hdr *ip6; 3573 int hlen, olen, mlen; 3574 3575 if (af == naf || (af != AF_INET && af != AF_INET6) || 3576 (naf != AF_INET && naf != AF_INET6)) 3577 return (-1); 3578 3579 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3580 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3581 return (-1); 3582 3583 /* old header */ 3584 olen = pd2->off - off; 3585 /* new header */ 3586 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3587 3588 /* trim old header */ 3589 m_adj(n, olen); 3590 3591 /* prepend a new one */ 3592 M_PREPEND(n, hlen, M_NOWAIT); 3593 if (n == NULL) 3594 return (-1); 3595 3596 /* translate inner ip/ip6 header */ 3597 switch (naf) { 3598 case AF_INET: 3599 ip4 = mtod(n, struct ip *); 3600 bzero(ip4, sizeof(*ip4)); 3601 ip4->ip_v = IPVERSION; 3602 ip4->ip_hl = sizeof(*ip4) >> 2; 3603 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 3604 ip_fillid(ip4); 3605 ip4->ip_off = htons(IP_DF); 3606 ip4->ip_ttl = pd2->ttl; 3607 if (pd2->proto == IPPROTO_ICMPV6) 3608 ip4->ip_p = IPPROTO_ICMP; 3609 else 3610 ip4->ip_p = pd2->proto; 3611 ip4->ip_src = src->v4; 3612 ip4->ip_dst = dst->v4; 3613 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3614 break; 3615 case AF_INET6: 3616 ip6 = mtod(n, struct ip6_hdr *); 3617 bzero(ip6, sizeof(*ip6)); 3618 ip6->ip6_vfc = IPV6_VERSION; 3619 ip6->ip6_plen = htons(pd2->tot_len - olen); 3620 if (pd2->proto == IPPROTO_ICMP) 3621 ip6->ip6_nxt = IPPROTO_ICMPV6; 3622 else 3623 ip6->ip6_nxt = pd2->proto; 3624 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3625 ip6->ip6_hlim = IPV6_DEFHLIM; 3626 else 3627 ip6->ip6_hlim = pd2->ttl; 3628 ip6->ip6_src = src->v6; 3629 ip6->ip6_dst = dst->v6; 3630 break; 3631 } 3632 3633 /* adjust payload offset and total packet length */ 3634 pd2->off += hlen - olen; 3635 pd->tot_len += hlen - olen; 3636 3637 /* merge modified inner packet with the original header */ 3638 mlen = n->m_pkthdr.len; 3639 m_cat(m, n); 3640 m->m_pkthdr.len += mlen; 3641 #endif /* INET && INET6 */ 3642 3643 return (0); 3644 } 3645 3646 #define PTR_IP(field) (offsetof(struct ip, field)) 3647 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3648 3649 int 3650 pf_translate_icmp_af(int af, void *arg) 3651 { 3652 #if defined(INET) && defined(INET6) 3653 struct icmp *icmp4; 3654 struct icmp6_hdr *icmp6; 3655 u_int32_t mtu; 3656 int32_t ptr = -1; 3657 u_int8_t type; 3658 u_int8_t code; 3659 3660 switch (af) { 3661 case AF_INET: 3662 icmp6 = arg; 3663 type = icmp6->icmp6_type; 3664 code = icmp6->icmp6_code; 3665 mtu = ntohl(icmp6->icmp6_mtu); 3666 3667 switch (type) { 3668 case ICMP6_ECHO_REQUEST: 3669 type = ICMP_ECHO; 3670 break; 3671 case ICMP6_ECHO_REPLY: 3672 type = ICMP_ECHOREPLY; 3673 break; 3674 case ICMP6_DST_UNREACH: 3675 type = ICMP_UNREACH; 3676 switch (code) { 3677 case ICMP6_DST_UNREACH_NOROUTE: 3678 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3679 case ICMP6_DST_UNREACH_ADDR: 3680 code = ICMP_UNREACH_HOST; 3681 break; 3682 case ICMP6_DST_UNREACH_ADMIN: 3683 code = ICMP_UNREACH_HOST_PROHIB; 3684 break; 3685 case ICMP6_DST_UNREACH_NOPORT: 3686 code = ICMP_UNREACH_PORT; 3687 break; 3688 default: 3689 return (-1); 3690 } 3691 break; 3692 case ICMP6_PACKET_TOO_BIG: 3693 type = ICMP_UNREACH; 3694 code = ICMP_UNREACH_NEEDFRAG; 3695 mtu -= 20; 3696 break; 3697 case ICMP6_TIME_EXCEEDED: 3698 type = ICMP_TIMXCEED; 3699 break; 3700 case ICMP6_PARAM_PROB: 3701 switch (code) { 3702 case ICMP6_PARAMPROB_HEADER: 3703 type = ICMP_PARAMPROB; 3704 code = ICMP_PARAMPROB_ERRATPTR; 3705 ptr = ntohl(icmp6->icmp6_pptr); 3706 3707 if (ptr == PTR_IP6(ip6_vfc)) 3708 ; /* preserve */ 3709 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3710 ptr = PTR_IP(ip_tos); 3711 else if (ptr == PTR_IP6(ip6_plen) || 3712 ptr == PTR_IP6(ip6_plen) + 1) 3713 ptr = PTR_IP(ip_len); 3714 else if (ptr == PTR_IP6(ip6_nxt)) 3715 ptr = PTR_IP(ip_p); 3716 else if (ptr == PTR_IP6(ip6_hlim)) 3717 ptr = PTR_IP(ip_ttl); 3718 else if (ptr >= PTR_IP6(ip6_src) && 3719 ptr < PTR_IP6(ip6_dst)) 3720 ptr = PTR_IP(ip_src); 3721 else if (ptr >= PTR_IP6(ip6_dst) && 3722 ptr < sizeof(struct ip6_hdr)) 3723 ptr = PTR_IP(ip_dst); 3724 else { 3725 return (-1); 3726 } 3727 break; 3728 case ICMP6_PARAMPROB_NEXTHEADER: 3729 type = ICMP_UNREACH; 3730 code = ICMP_UNREACH_PROTOCOL; 3731 break; 3732 default: 3733 return (-1); 3734 } 3735 break; 3736 default: 3737 return (-1); 3738 } 3739 if (icmp6->icmp6_type != type) { 3740 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3741 icmp6->icmp6_type, type, 0); 3742 icmp6->icmp6_type = type; 3743 } 3744 if (icmp6->icmp6_code != code) { 3745 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3746 icmp6->icmp6_code, code, 0); 3747 icmp6->icmp6_code = code; 3748 } 3749 if (icmp6->icmp6_mtu != htonl(mtu)) { 3750 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3751 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3752 /* aligns well with a icmpv4 nextmtu */ 3753 icmp6->icmp6_mtu = htonl(mtu); 3754 } 3755 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3756 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3757 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3758 /* icmpv4 pptr is a one most significant byte */ 3759 icmp6->icmp6_pptr = htonl(ptr << 24); 3760 } 3761 break; 3762 case AF_INET6: 3763 icmp4 = arg; 3764 type = icmp4->icmp_type; 3765 code = icmp4->icmp_code; 3766 mtu = ntohs(icmp4->icmp_nextmtu); 3767 3768 switch (type) { 3769 case ICMP_ECHO: 3770 type = ICMP6_ECHO_REQUEST; 3771 break; 3772 case ICMP_ECHOREPLY: 3773 type = ICMP6_ECHO_REPLY; 3774 break; 3775 case ICMP_UNREACH: 3776 type = ICMP6_DST_UNREACH; 3777 switch (code) { 3778 case ICMP_UNREACH_NET: 3779 case ICMP_UNREACH_HOST: 3780 case ICMP_UNREACH_NET_UNKNOWN: 3781 case ICMP_UNREACH_HOST_UNKNOWN: 3782 case ICMP_UNREACH_ISOLATED: 3783 case ICMP_UNREACH_TOSNET: 3784 case ICMP_UNREACH_TOSHOST: 3785 code = ICMP6_DST_UNREACH_NOROUTE; 3786 break; 3787 case ICMP_UNREACH_PORT: 3788 code = ICMP6_DST_UNREACH_NOPORT; 3789 break; 3790 case ICMP_UNREACH_NET_PROHIB: 3791 case ICMP_UNREACH_HOST_PROHIB: 3792 case ICMP_UNREACH_FILTER_PROHIB: 3793 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3794 code = ICMP6_DST_UNREACH_ADMIN; 3795 break; 3796 case ICMP_UNREACH_PROTOCOL: 3797 type = ICMP6_PARAM_PROB; 3798 code = ICMP6_PARAMPROB_NEXTHEADER; 3799 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3800 break; 3801 case ICMP_UNREACH_NEEDFRAG: 3802 type = ICMP6_PACKET_TOO_BIG; 3803 code = 0; 3804 mtu += 20; 3805 break; 3806 default: 3807 return (-1); 3808 } 3809 break; 3810 case ICMP_TIMXCEED: 3811 type = ICMP6_TIME_EXCEEDED; 3812 break; 3813 case ICMP_PARAMPROB: 3814 type = ICMP6_PARAM_PROB; 3815 switch (code) { 3816 case ICMP_PARAMPROB_ERRATPTR: 3817 code = ICMP6_PARAMPROB_HEADER; 3818 break; 3819 case ICMP_PARAMPROB_LENGTH: 3820 code = ICMP6_PARAMPROB_HEADER; 3821 break; 3822 default: 3823 return (-1); 3824 } 3825 3826 ptr = icmp4->icmp_pptr; 3827 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3828 ; /* preserve */ 3829 else if (ptr == PTR_IP(ip_len) || 3830 ptr == PTR_IP(ip_len) + 1) 3831 ptr = PTR_IP6(ip6_plen); 3832 else if (ptr == PTR_IP(ip_ttl)) 3833 ptr = PTR_IP6(ip6_hlim); 3834 else if (ptr == PTR_IP(ip_p)) 3835 ptr = PTR_IP6(ip6_nxt); 3836 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 3837 ptr = PTR_IP6(ip6_src); 3838 else if (ptr >= PTR_IP(ip_dst) && 3839 ptr < sizeof(struct ip)) 3840 ptr = PTR_IP6(ip6_dst); 3841 else { 3842 return (-1); 3843 } 3844 break; 3845 default: 3846 return (-1); 3847 } 3848 if (icmp4->icmp_type != type) { 3849 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3850 icmp4->icmp_type, type, 0); 3851 icmp4->icmp_type = type; 3852 } 3853 if (icmp4->icmp_code != code) { 3854 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3855 icmp4->icmp_code, code, 0); 3856 icmp4->icmp_code = code; 3857 } 3858 if (icmp4->icmp_nextmtu != htons(mtu)) { 3859 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3860 icmp4->icmp_nextmtu, htons(mtu), 0); 3861 icmp4->icmp_nextmtu = htons(mtu); 3862 } 3863 if (ptr >= 0 && icmp4->icmp_void != ptr) { 3864 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3865 htons(icmp4->icmp_pptr), htons(ptr), 0); 3866 icmp4->icmp_void = htonl(ptr); 3867 } 3868 break; 3869 } 3870 #endif /* INET && INET6 */ 3871 3872 return (0); 3873 } 3874 3875 /* 3876 * Need to modulate the sequence numbers in the TCP SACK option 3877 * (credits to Krzysztof Pfaff for report and patch) 3878 */ 3879 static int 3880 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3881 struct pf_state_peer *dst) 3882 { 3883 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3884 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3885 int copyback = 0, i, olen; 3886 struct sackblk sack; 3887 3888 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3889 if (hlen < TCPOLEN_SACKLEN || hlen > MAX_TCPOPTLEN || 3890 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3891 return 0; 3892 3893 while (hlen >= TCPOLEN_SACKLEN) { 3894 size_t startoff = opt - opts; 3895 olen = opt[1]; 3896 switch (*opt) { 3897 case TCPOPT_EOL: /* FALLTHROUGH */ 3898 case TCPOPT_NOP: 3899 opt++; 3900 hlen--; 3901 break; 3902 case TCPOPT_SACK: 3903 if (olen > hlen) 3904 olen = hlen; 3905 if (olen >= TCPOLEN_SACKLEN) { 3906 for (i = 2; i + TCPOLEN_SACK <= olen; 3907 i += TCPOLEN_SACK) { 3908 memcpy(&sack, &opt[i], sizeof(sack)); 3909 pf_patch_32_unaligned(pd->m, 3910 &th->th_sum, &sack.start, 3911 htonl(ntohl(sack.start) - dst->seqdiff), 3912 PF_ALGNMNT(startoff), 3913 0); 3914 pf_patch_32_unaligned(pd->m, &th->th_sum, 3915 &sack.end, 3916 htonl(ntohl(sack.end) - dst->seqdiff), 3917 PF_ALGNMNT(startoff), 3918 0); 3919 memcpy(&opt[i], &sack, sizeof(sack)); 3920 } 3921 copyback = 1; 3922 } 3923 /* FALLTHROUGH */ 3924 default: 3925 if (olen < 2) 3926 olen = 2; 3927 hlen -= olen; 3928 opt += olen; 3929 } 3930 } 3931 3932 if (copyback) 3933 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3934 return (copyback); 3935 } 3936 3937 struct mbuf * 3938 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3939 const struct pf_addr *saddr, const struct pf_addr *daddr, 3940 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3941 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3942 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3943 { 3944 struct mbuf *m; 3945 int len, tlen; 3946 #ifdef INET 3947 struct ip *h = NULL; 3948 #endif /* INET */ 3949 #ifdef INET6 3950 struct ip6_hdr *h6 = NULL; 3951 #endif /* INET6 */ 3952 struct tcphdr *th; 3953 char *opt; 3954 struct pf_mtag *pf_mtag; 3955 3956 len = 0; 3957 th = NULL; 3958 3959 /* maximum segment size tcp option */ 3960 tlen = sizeof(struct tcphdr); 3961 if (mss) 3962 tlen += 4; 3963 3964 switch (af) { 3965 #ifdef INET 3966 case AF_INET: 3967 len = sizeof(struct ip) + tlen; 3968 break; 3969 #endif /* INET */ 3970 #ifdef INET6 3971 case AF_INET6: 3972 len = sizeof(struct ip6_hdr) + tlen; 3973 break; 3974 #endif /* INET6 */ 3975 } 3976 3977 m = m_gethdr(M_NOWAIT, MT_DATA); 3978 if (m == NULL) 3979 return (NULL); 3980 3981 #ifdef MAC 3982 mac_netinet_firewall_send(m); 3983 #endif 3984 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3985 m_freem(m); 3986 return (NULL); 3987 } 3988 m->m_flags |= mbuf_flags; 3989 pf_mtag->tag = mtag_tag; 3990 pf_mtag->flags = mtag_flags; 3991 3992 if (rtableid >= 0) 3993 M_SETFIB(m, rtableid); 3994 3995 #ifdef ALTQ 3996 if (r != NULL && r->qid) { 3997 pf_mtag->qid = r->qid; 3998 3999 /* add hints for ecn */ 4000 pf_mtag->hdr = mtod(m, struct ip *); 4001 } 4002 #endif /* ALTQ */ 4003 m->m_data += max_linkhdr; 4004 m->m_pkthdr.len = m->m_len = len; 4005 /* The rest of the stack assumes a rcvif, so provide one. 4006 * This is a locally generated packet, so .. close enough. */ 4007 m->m_pkthdr.rcvif = V_loif; 4008 bzero(m->m_data, len); 4009 switch (af) { 4010 #ifdef INET 4011 case AF_INET: 4012 m->m_pkthdr.csum_flags |= CSUM_TCP; 4013 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4014 4015 h = mtod(m, struct ip *); 4016 4017 h->ip_p = IPPROTO_TCP; 4018 h->ip_len = htons(tlen); 4019 h->ip_v = 4; 4020 h->ip_hl = sizeof(*h) >> 2; 4021 h->ip_tos = IPTOS_LOWDELAY; 4022 h->ip_len = htons(len); 4023 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4024 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4025 h->ip_sum = 0; 4026 h->ip_src.s_addr = saddr->v4.s_addr; 4027 h->ip_dst.s_addr = daddr->v4.s_addr; 4028 4029 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4030 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4031 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4032 break; 4033 #endif /* INET */ 4034 #ifdef INET6 4035 case AF_INET6: 4036 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4037 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4038 4039 h6 = mtod(m, struct ip6_hdr *); 4040 4041 /* IP header fields included in the TCP checksum */ 4042 h6->ip6_nxt = IPPROTO_TCP; 4043 h6->ip6_plen = htons(tlen); 4044 h6->ip6_vfc |= IPV6_VERSION; 4045 h6->ip6_hlim = V_ip6_defhlim; 4046 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4047 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4048 4049 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4050 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4051 IPPROTO_TCP, 0); 4052 break; 4053 #endif /* INET6 */ 4054 } 4055 4056 /* TCP header */ 4057 th->th_sport = sport; 4058 th->th_dport = dport; 4059 th->th_seq = htonl(seq); 4060 th->th_ack = htonl(ack); 4061 th->th_off = tlen >> 2; 4062 tcp_set_flags(th, tcp_flags); 4063 th->th_win = htons(win); 4064 4065 if (mss) { 4066 opt = (char *)(th + 1); 4067 opt[0] = TCPOPT_MAXSEG; 4068 opt[1] = 4; 4069 HTONS(mss); 4070 memcpy((opt + 2), &mss, 2); 4071 } 4072 4073 return (m); 4074 } 4075 4076 static void 4077 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4078 uint8_t ttl, int rtableid) 4079 { 4080 struct mbuf *m; 4081 #ifdef INET 4082 struct ip *h = NULL; 4083 #endif /* INET */ 4084 #ifdef INET6 4085 struct ip6_hdr *h6 = NULL; 4086 #endif /* INET6 */ 4087 struct sctphdr *hdr; 4088 struct sctp_chunkhdr *chunk; 4089 struct pf_send_entry *pfse; 4090 int off = 0; 4091 4092 MPASS(af == pd->af); 4093 4094 m = m_gethdr(M_NOWAIT, MT_DATA); 4095 if (m == NULL) 4096 return; 4097 4098 m->m_data += max_linkhdr; 4099 m->m_flags |= M_SKIP_FIREWALL; 4100 /* The rest of the stack assumes a rcvif, so provide one. 4101 * This is a locally generated packet, so .. close enough. */ 4102 m->m_pkthdr.rcvif = V_loif; 4103 4104 /* IPv4|6 header */ 4105 switch (af) { 4106 #ifdef INET 4107 case AF_INET: 4108 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4109 4110 h = mtod(m, struct ip *); 4111 4112 /* IP header fields included in the TCP checksum */ 4113 4114 h->ip_p = IPPROTO_SCTP; 4115 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4116 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4117 h->ip_src = pd->dst->v4; 4118 h->ip_dst = pd->src->v4; 4119 4120 off += sizeof(struct ip); 4121 break; 4122 #endif /* INET */ 4123 #ifdef INET6 4124 case AF_INET6: 4125 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4126 4127 h6 = mtod(m, struct ip6_hdr *); 4128 4129 /* IP header fields included in the TCP checksum */ 4130 h6->ip6_vfc |= IPV6_VERSION; 4131 h6->ip6_nxt = IPPROTO_SCTP; 4132 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4133 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4134 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4135 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4136 4137 off += sizeof(struct ip6_hdr); 4138 break; 4139 #endif /* INET6 */ 4140 } 4141 4142 /* SCTP header */ 4143 hdr = mtodo(m, off); 4144 4145 hdr->src_port = pd->hdr.sctp.dest_port; 4146 hdr->dest_port = pd->hdr.sctp.src_port; 4147 hdr->v_tag = pd->sctp_initiate_tag; 4148 hdr->checksum = 0; 4149 4150 /* Abort chunk. */ 4151 off += sizeof(struct sctphdr); 4152 chunk = mtodo(m, off); 4153 4154 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4155 chunk->chunk_length = htons(sizeof(*chunk)); 4156 4157 /* SCTP checksum */ 4158 off += sizeof(*chunk); 4159 m->m_pkthdr.len = m->m_len = off; 4160 4161 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4162 4163 if (rtableid >= 0) 4164 M_SETFIB(m, rtableid); 4165 4166 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4167 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4168 if (pfse == NULL) { 4169 m_freem(m); 4170 return; 4171 } 4172 4173 switch (af) { 4174 #ifdef INET 4175 case AF_INET: 4176 pfse->pfse_type = PFSE_IP; 4177 break; 4178 #endif /* INET */ 4179 #ifdef INET6 4180 case AF_INET6: 4181 pfse->pfse_type = PFSE_IP6; 4182 break; 4183 #endif /* INET6 */ 4184 } 4185 4186 pfse->pfse_m = m; 4187 pf_send(pfse); 4188 } 4189 4190 void 4191 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4192 const struct pf_addr *saddr, const struct pf_addr *daddr, 4193 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4194 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4195 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 4196 { 4197 struct pf_send_entry *pfse; 4198 struct mbuf *m; 4199 4200 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4201 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 4202 if (m == NULL) 4203 return; 4204 4205 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4206 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4207 if (pfse == NULL) { 4208 m_freem(m); 4209 return; 4210 } 4211 4212 switch (af) { 4213 #ifdef INET 4214 case AF_INET: 4215 pfse->pfse_type = PFSE_IP; 4216 break; 4217 #endif /* INET */ 4218 #ifdef INET6 4219 case AF_INET6: 4220 pfse->pfse_type = PFSE_IP6; 4221 break; 4222 #endif /* INET6 */ 4223 } 4224 4225 pfse->pfse_m = m; 4226 pf_send(pfse); 4227 } 4228 4229 static void 4230 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4231 struct pf_state_key *sk, struct tcphdr *th, 4232 u_int16_t bproto_sum, u_int16_t bip_sum, 4233 u_short *reason, int rtableid) 4234 { 4235 struct pf_addr * const saddr = pd->src; 4236 struct pf_addr * const daddr = pd->dst; 4237 4238 /* undo NAT changes, if they have taken place */ 4239 if (nr != NULL) { 4240 PF_ACPY(saddr, &sk->addr[pd->sidx], pd->af); 4241 PF_ACPY(daddr, &sk->addr[pd->didx], pd->af); 4242 if (pd->sport) 4243 *pd->sport = sk->port[pd->sidx]; 4244 if (pd->dport) 4245 *pd->dport = sk->port[pd->didx]; 4246 if (pd->ip_sum) 4247 *pd->ip_sum = bip_sum; 4248 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4249 } 4250 if (pd->proto == IPPROTO_TCP && 4251 ((r->rule_flag & PFRULE_RETURNRST) || 4252 (r->rule_flag & PFRULE_RETURN)) && 4253 !(tcp_get_flags(th) & TH_RST)) { 4254 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4255 4256 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4257 IPPROTO_TCP, pd->af)) 4258 REASON_SET(reason, PFRES_PROTCKSUM); 4259 else { 4260 if (tcp_get_flags(th) & TH_SYN) 4261 ack++; 4262 if (tcp_get_flags(th) & TH_FIN) 4263 ack++; 4264 pf_send_tcp(r, pd->af, pd->dst, 4265 pd->src, th->th_dport, th->th_sport, 4266 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4267 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 4268 } 4269 } else if (pd->proto == IPPROTO_SCTP && 4270 (r->rule_flag & PFRULE_RETURN)) { 4271 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4272 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4273 r->return_icmp) 4274 pf_send_icmp(pd->m, r->return_icmp >> 8, 4275 r->return_icmp & 255, pd->af, r, rtableid); 4276 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4277 r->return_icmp6) 4278 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4279 r->return_icmp6 & 255, pd->af, r, rtableid); 4280 } 4281 4282 static int 4283 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4284 { 4285 struct m_tag *mtag; 4286 u_int8_t mpcp; 4287 4288 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4289 if (mtag == NULL) 4290 return (0); 4291 4292 if (prio == PF_PRIO_ZERO) 4293 prio = 0; 4294 4295 mpcp = *(uint8_t *)(mtag + 1); 4296 4297 return (mpcp == prio); 4298 } 4299 4300 static int 4301 pf_icmp_to_bandlim(uint8_t type) 4302 { 4303 switch (type) { 4304 case ICMP_ECHO: 4305 case ICMP_ECHOREPLY: 4306 return (BANDLIM_ICMP_ECHO); 4307 case ICMP_TSTAMP: 4308 case ICMP_TSTAMPREPLY: 4309 return (BANDLIM_ICMP_TSTAMP); 4310 case ICMP_UNREACH: 4311 default: 4312 return (BANDLIM_ICMP_UNREACH); 4313 } 4314 } 4315 4316 static void 4317 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 4318 struct pf_krule *r, int rtableid) 4319 { 4320 struct pf_send_entry *pfse; 4321 struct mbuf *m0; 4322 struct pf_mtag *pf_mtag; 4323 4324 /* ICMP packet rate limitation. */ 4325 switch (af) { 4326 #ifdef INET6 4327 case AF_INET6: 4328 if (icmp6_ratelimit(NULL, type, code)) 4329 return; 4330 break; 4331 #endif 4332 #ifdef INET 4333 case AF_INET: 4334 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4335 return; 4336 break; 4337 #endif 4338 } 4339 4340 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4341 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4342 if (pfse == NULL) 4343 return; 4344 4345 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4346 free(pfse, M_PFTEMP); 4347 return; 4348 } 4349 4350 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4351 free(pfse, M_PFTEMP); 4352 return; 4353 } 4354 /* XXX: revisit */ 4355 m0->m_flags |= M_SKIP_FIREWALL; 4356 4357 if (rtableid >= 0) 4358 M_SETFIB(m0, rtableid); 4359 4360 #ifdef ALTQ 4361 if (r->qid) { 4362 pf_mtag->qid = r->qid; 4363 /* add hints for ecn */ 4364 pf_mtag->hdr = mtod(m0, struct ip *); 4365 } 4366 #endif /* ALTQ */ 4367 4368 switch (af) { 4369 #ifdef INET 4370 case AF_INET: 4371 pfse->pfse_type = PFSE_ICMP; 4372 break; 4373 #endif /* INET */ 4374 #ifdef INET6 4375 case AF_INET6: 4376 pfse->pfse_type = PFSE_ICMP6; 4377 break; 4378 #endif /* INET6 */ 4379 } 4380 pfse->pfse_m = m0; 4381 pfse->icmpopts.type = type; 4382 pfse->icmpopts.code = code; 4383 pf_send(pfse); 4384 } 4385 4386 /* 4387 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 4388 * If n is 0, they match if they are equal. If n is != 0, they match if they 4389 * are different. 4390 */ 4391 int 4392 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4393 const struct pf_addr *b, sa_family_t af) 4394 { 4395 int match = 0; 4396 4397 switch (af) { 4398 #ifdef INET 4399 case AF_INET: 4400 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4401 match++; 4402 break; 4403 #endif /* INET */ 4404 #ifdef INET6 4405 case AF_INET6: 4406 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4407 match++; 4408 break; 4409 #endif /* INET6 */ 4410 } 4411 if (match) { 4412 if (n) 4413 return (0); 4414 else 4415 return (1); 4416 } else { 4417 if (n) 4418 return (1); 4419 else 4420 return (0); 4421 } 4422 } 4423 4424 /* 4425 * Return 1 if b <= a <= e, otherwise return 0. 4426 */ 4427 int 4428 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4429 const struct pf_addr *a, sa_family_t af) 4430 { 4431 switch (af) { 4432 #ifdef INET 4433 case AF_INET: 4434 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4435 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4436 return (0); 4437 break; 4438 #endif /* INET */ 4439 #ifdef INET6 4440 case AF_INET6: { 4441 int i; 4442 4443 /* check a >= b */ 4444 for (i = 0; i < 4; ++i) 4445 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4446 break; 4447 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4448 return (0); 4449 /* check a <= e */ 4450 for (i = 0; i < 4; ++i) 4451 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4452 break; 4453 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4454 return (0); 4455 break; 4456 } 4457 #endif /* INET6 */ 4458 } 4459 return (1); 4460 } 4461 4462 static int 4463 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4464 { 4465 switch (op) { 4466 case PF_OP_IRG: 4467 return ((p > a1) && (p < a2)); 4468 case PF_OP_XRG: 4469 return ((p < a1) || (p > a2)); 4470 case PF_OP_RRG: 4471 return ((p >= a1) && (p <= a2)); 4472 case PF_OP_EQ: 4473 return (p == a1); 4474 case PF_OP_NE: 4475 return (p != a1); 4476 case PF_OP_LT: 4477 return (p < a1); 4478 case PF_OP_LE: 4479 return (p <= a1); 4480 case PF_OP_GT: 4481 return (p > a1); 4482 case PF_OP_GE: 4483 return (p >= a1); 4484 } 4485 return (0); /* never reached */ 4486 } 4487 4488 int 4489 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4490 { 4491 NTOHS(a1); 4492 NTOHS(a2); 4493 NTOHS(p); 4494 return (pf_match(op, a1, a2, p)); 4495 } 4496 4497 static int 4498 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4499 { 4500 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4501 return (0); 4502 return (pf_match(op, a1, a2, u)); 4503 } 4504 4505 static int 4506 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4507 { 4508 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4509 return (0); 4510 return (pf_match(op, a1, a2, g)); 4511 } 4512 4513 int 4514 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4515 { 4516 if (*tag == -1) 4517 *tag = mtag; 4518 4519 return ((!r->match_tag_not && r->match_tag == *tag) || 4520 (r->match_tag_not && r->match_tag != *tag)); 4521 } 4522 4523 static int 4524 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4525 { 4526 struct ifnet *ifp = m->m_pkthdr.rcvif; 4527 struct pfi_kkif *kif; 4528 4529 if (ifp == NULL) 4530 return (0); 4531 4532 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4533 4534 if (kif == NULL) { 4535 DPFPRINTF(PF_DEBUG_URGENT, 4536 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 4537 r->rcv_ifname)); 4538 return (0); 4539 } 4540 4541 return (pfi_kkif_match(r->rcv_kif, kif)); 4542 } 4543 4544 int 4545 pf_tag_packet(struct pf_pdesc *pd, int tag) 4546 { 4547 4548 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4549 4550 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4551 return (ENOMEM); 4552 4553 pd->pf_mtag->tag = tag; 4554 4555 return (0); 4556 } 4557 4558 #define PF_ANCHOR_STACKSIZE 32 4559 struct pf_kanchor_stackframe { 4560 struct pf_kruleset *rs; 4561 struct pf_krule *r; /* XXX: + match bit */ 4562 struct pf_kanchor *child; 4563 }; 4564 4565 /* 4566 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4567 */ 4568 #define PF_ANCHORSTACK_MATCH 0x00000001 4569 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4570 4571 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4572 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4573 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4574 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4575 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4576 } while (0) 4577 4578 void 4579 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4580 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a) 4581 { 4582 struct pf_kanchor_stackframe *f; 4583 4584 PF_RULES_RASSERT(); 4585 4586 if (*depth >= PF_ANCHOR_STACKSIZE) { 4587 printf("%s: anchor stack overflow on %s\n", 4588 __func__, (*r)->anchor->name); 4589 *r = TAILQ_NEXT(*r, entries); 4590 return; 4591 } else if (*depth == 0 && a != NULL) 4592 *a = *r; 4593 f = stack + (*depth)++; 4594 f->rs = *rs; 4595 f->r = *r; 4596 if ((*r)->anchor_wildcard) { 4597 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4598 4599 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4600 *r = NULL; 4601 return; 4602 } 4603 *rs = &f->child->ruleset; 4604 } else { 4605 f->child = NULL; 4606 *rs = &(*r)->anchor->ruleset; 4607 } 4608 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4609 } 4610 4611 int 4612 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4613 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4614 int *match) 4615 { 4616 struct pf_kanchor_stackframe *f; 4617 struct pf_krule *fr; 4618 int quick = 0; 4619 4620 PF_RULES_RASSERT(); 4621 4622 do { 4623 if (*depth <= 0) 4624 break; 4625 f = stack + *depth - 1; 4626 fr = PF_ANCHOR_RULE(f); 4627 if (f->child != NULL) { 4628 f->child = RB_NEXT(pf_kanchor_node, 4629 &fr->anchor->children, f->child); 4630 if (f->child != NULL) { 4631 *rs = &f->child->ruleset; 4632 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4633 if (*r == NULL) 4634 continue; 4635 else 4636 break; 4637 } 4638 } 4639 (*depth)--; 4640 if (*depth == 0 && a != NULL) 4641 *a = NULL; 4642 *rs = f->rs; 4643 if (match != NULL && *match > *depth) { 4644 *match = *depth; 4645 if (f->r->quick) 4646 quick = 1; 4647 } 4648 *r = TAILQ_NEXT(fr, entries); 4649 } while (*r == NULL); 4650 4651 return (quick); 4652 } 4653 4654 struct pf_keth_anchor_stackframe { 4655 struct pf_keth_ruleset *rs; 4656 struct pf_keth_rule *r; /* XXX: + match bit */ 4657 struct pf_keth_anchor *child; 4658 }; 4659 4660 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4661 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4662 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4663 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4664 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4665 } while (0) 4666 4667 void 4668 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4669 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4670 struct pf_keth_rule **a, int *match) 4671 { 4672 struct pf_keth_anchor_stackframe *f; 4673 4674 NET_EPOCH_ASSERT(); 4675 4676 if (match) 4677 *match = 0; 4678 if (*depth >= PF_ANCHOR_STACKSIZE) { 4679 printf("%s: anchor stack overflow on %s\n", 4680 __func__, (*r)->anchor->name); 4681 *r = TAILQ_NEXT(*r, entries); 4682 return; 4683 } else if (*depth == 0 && a != NULL) 4684 *a = *r; 4685 f = stack + (*depth)++; 4686 f->rs = *rs; 4687 f->r = *r; 4688 if ((*r)->anchor_wildcard) { 4689 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4690 4691 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4692 *r = NULL; 4693 return; 4694 } 4695 *rs = &f->child->ruleset; 4696 } else { 4697 f->child = NULL; 4698 *rs = &(*r)->anchor->ruleset; 4699 } 4700 *r = TAILQ_FIRST((*rs)->active.rules); 4701 } 4702 4703 int 4704 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4705 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4706 struct pf_keth_rule **a, int *match) 4707 { 4708 struct pf_keth_anchor_stackframe *f; 4709 struct pf_keth_rule *fr; 4710 int quick = 0; 4711 4712 NET_EPOCH_ASSERT(); 4713 4714 do { 4715 if (*depth <= 0) 4716 break; 4717 f = stack + *depth - 1; 4718 fr = PF_ETH_ANCHOR_RULE(f); 4719 if (f->child != NULL) { 4720 /* 4721 * This block traverses through 4722 * a wildcard anchor. 4723 */ 4724 if (match != NULL && *match) { 4725 /* 4726 * If any of "*" matched, then 4727 * "foo/ *" matched, mark frame 4728 * appropriately. 4729 */ 4730 PF_ETH_ANCHOR_SET_MATCH(f); 4731 *match = 0; 4732 } 4733 f->child = RB_NEXT(pf_keth_anchor_node, 4734 &fr->anchor->children, f->child); 4735 if (f->child != NULL) { 4736 *rs = &f->child->ruleset; 4737 *r = TAILQ_FIRST((*rs)->active.rules); 4738 if (*r == NULL) 4739 continue; 4740 else 4741 break; 4742 } 4743 } 4744 (*depth)--; 4745 if (*depth == 0 && a != NULL) 4746 *a = NULL; 4747 *rs = f->rs; 4748 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4749 quick = fr->quick; 4750 *r = TAILQ_NEXT(fr, entries); 4751 } while (*r == NULL); 4752 4753 return (quick); 4754 } 4755 4756 #ifdef INET6 4757 void 4758 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4759 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4760 { 4761 switch (af) { 4762 #ifdef INET 4763 case AF_INET: 4764 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4765 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4766 break; 4767 #endif /* INET */ 4768 case AF_INET6: 4769 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4770 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4771 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4772 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4773 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4774 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4775 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4776 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4777 break; 4778 } 4779 } 4780 4781 void 4782 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4783 { 4784 switch (af) { 4785 #ifdef INET 4786 case AF_INET: 4787 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4788 break; 4789 #endif /* INET */ 4790 case AF_INET6: 4791 if (addr->addr32[3] == 0xffffffff) { 4792 addr->addr32[3] = 0; 4793 if (addr->addr32[2] == 0xffffffff) { 4794 addr->addr32[2] = 0; 4795 if (addr->addr32[1] == 0xffffffff) { 4796 addr->addr32[1] = 0; 4797 addr->addr32[0] = 4798 htonl(ntohl(addr->addr32[0]) + 1); 4799 } else 4800 addr->addr32[1] = 4801 htonl(ntohl(addr->addr32[1]) + 1); 4802 } else 4803 addr->addr32[2] = 4804 htonl(ntohl(addr->addr32[2]) + 1); 4805 } else 4806 addr->addr32[3] = 4807 htonl(ntohl(addr->addr32[3]) + 1); 4808 break; 4809 } 4810 } 4811 #endif /* INET6 */ 4812 4813 void 4814 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4815 { 4816 /* 4817 * Modern rules use the same flags in rules as they do in states. 4818 */ 4819 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4820 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4821 4822 /* 4823 * Old-style scrub rules have different flags which need to be translated. 4824 */ 4825 if (r->rule_flag & PFRULE_RANDOMID) 4826 a->flags |= PFSTATE_RANDOMID; 4827 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4828 a->flags |= PFSTATE_SETTOS; 4829 a->set_tos = r->set_tos; 4830 } 4831 4832 if (r->qid) 4833 a->qid = r->qid; 4834 if (r->pqid) 4835 a->pqid = r->pqid; 4836 if (r->rtableid >= 0) 4837 a->rtableid = r->rtableid; 4838 a->log |= r->log; 4839 if (r->min_ttl) 4840 a->min_ttl = r->min_ttl; 4841 if (r->max_mss) 4842 a->max_mss = r->max_mss; 4843 if (r->dnpipe) 4844 a->dnpipe = r->dnpipe; 4845 if (r->dnrpipe) 4846 a->dnrpipe = r->dnrpipe; 4847 if (r->dnpipe || r->dnrpipe) { 4848 if (r->free_flags & PFRULE_DN_IS_PIPE) 4849 a->flags |= PFSTATE_DN_IS_PIPE; 4850 else 4851 a->flags &= ~PFSTATE_DN_IS_PIPE; 4852 } 4853 if (r->scrub_flags & PFSTATE_SETPRIO) { 4854 a->set_prio[0] = r->set_prio[0]; 4855 a->set_prio[1] = r->set_prio[1]; 4856 } 4857 } 4858 4859 int 4860 pf_socket_lookup(struct pf_pdesc *pd) 4861 { 4862 struct pf_addr *saddr, *daddr; 4863 u_int16_t sport, dport; 4864 struct inpcbinfo *pi; 4865 struct inpcb *inp; 4866 4867 pd->lookup.uid = UID_MAX; 4868 pd->lookup.gid = GID_MAX; 4869 4870 switch (pd->proto) { 4871 case IPPROTO_TCP: 4872 sport = pd->hdr.tcp.th_sport; 4873 dport = pd->hdr.tcp.th_dport; 4874 pi = &V_tcbinfo; 4875 break; 4876 case IPPROTO_UDP: 4877 sport = pd->hdr.udp.uh_sport; 4878 dport = pd->hdr.udp.uh_dport; 4879 pi = &V_udbinfo; 4880 break; 4881 default: 4882 return (-1); 4883 } 4884 if (pd->dir == PF_IN) { 4885 saddr = pd->src; 4886 daddr = pd->dst; 4887 } else { 4888 u_int16_t p; 4889 4890 p = sport; 4891 sport = dport; 4892 dport = p; 4893 saddr = pd->dst; 4894 daddr = pd->src; 4895 } 4896 switch (pd->af) { 4897 #ifdef INET 4898 case AF_INET: 4899 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4900 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4901 if (inp == NULL) { 4902 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4903 daddr->v4, dport, INPLOOKUP_WILDCARD | 4904 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4905 if (inp == NULL) 4906 return (-1); 4907 } 4908 break; 4909 #endif /* INET */ 4910 #ifdef INET6 4911 case AF_INET6: 4912 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4913 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4914 if (inp == NULL) { 4915 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4916 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4917 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4918 if (inp == NULL) 4919 return (-1); 4920 } 4921 break; 4922 #endif /* INET6 */ 4923 } 4924 INP_RLOCK_ASSERT(inp); 4925 pd->lookup.uid = inp->inp_cred->cr_uid; 4926 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4927 INP_RUNLOCK(inp); 4928 4929 return (1); 4930 } 4931 4932 u_int8_t 4933 pf_get_wscale(struct pf_pdesc *pd) 4934 { 4935 struct tcphdr *th = &pd->hdr.tcp; 4936 int hlen; 4937 u_int8_t hdr[60]; 4938 u_int8_t *opt, optlen; 4939 u_int8_t wscale = 0; 4940 4941 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4942 if (hlen <= sizeof(struct tcphdr)) 4943 return (0); 4944 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4945 return (0); 4946 opt = hdr + sizeof(struct tcphdr); 4947 hlen -= sizeof(struct tcphdr); 4948 while (hlen >= 3) { 4949 switch (*opt) { 4950 case TCPOPT_EOL: 4951 case TCPOPT_NOP: 4952 ++opt; 4953 --hlen; 4954 break; 4955 case TCPOPT_WINDOW: 4956 wscale = opt[2]; 4957 if (wscale > TCP_MAX_WINSHIFT) 4958 wscale = TCP_MAX_WINSHIFT; 4959 wscale |= PF_WSCALE_FLAG; 4960 /* FALLTHROUGH */ 4961 default: 4962 optlen = opt[1]; 4963 if (optlen < 2) 4964 optlen = 2; 4965 hlen -= optlen; 4966 opt += optlen; 4967 break; 4968 } 4969 } 4970 return (wscale); 4971 } 4972 4973 u_int16_t 4974 pf_get_mss(struct pf_pdesc *pd) 4975 { 4976 struct tcphdr *th = &pd->hdr.tcp; 4977 int hlen; 4978 u_int8_t hdr[60]; 4979 u_int8_t *opt, optlen; 4980 u_int16_t mss = V_tcp_mssdflt; 4981 4982 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4983 if (hlen <= sizeof(struct tcphdr)) 4984 return (0); 4985 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4986 return (0); 4987 opt = hdr + sizeof(struct tcphdr); 4988 hlen -= sizeof(struct tcphdr); 4989 while (hlen >= TCPOLEN_MAXSEG) { 4990 switch (*opt) { 4991 case TCPOPT_EOL: 4992 case TCPOPT_NOP: 4993 ++opt; 4994 --hlen; 4995 break; 4996 case TCPOPT_MAXSEG: 4997 memcpy(&mss, (opt + 2), 2); 4998 NTOHS(mss); 4999 /* FALLTHROUGH */ 5000 default: 5001 optlen = opt[1]; 5002 if (optlen < 2) 5003 optlen = 2; 5004 hlen -= optlen; 5005 opt += optlen; 5006 break; 5007 } 5008 } 5009 return (mss); 5010 } 5011 5012 static u_int16_t 5013 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5014 { 5015 struct nhop_object *nh; 5016 #ifdef INET6 5017 struct in6_addr dst6; 5018 uint32_t scopeid; 5019 #endif /* INET6 */ 5020 int hlen = 0; 5021 uint16_t mss = 0; 5022 5023 NET_EPOCH_ASSERT(); 5024 5025 switch (af) { 5026 #ifdef INET 5027 case AF_INET: 5028 hlen = sizeof(struct ip); 5029 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5030 if (nh != NULL) 5031 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5032 break; 5033 #endif /* INET */ 5034 #ifdef INET6 5035 case AF_INET6: 5036 hlen = sizeof(struct ip6_hdr); 5037 in6_splitscope(&addr->v6, &dst6, &scopeid); 5038 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5039 if (nh != NULL) 5040 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5041 break; 5042 #endif /* INET6 */ 5043 } 5044 5045 mss = max(V_tcp_mssdflt, mss); 5046 mss = min(mss, offer); 5047 mss = max(mss, 64); /* sanity - at least max opt space */ 5048 return (mss); 5049 } 5050 5051 static u_int32_t 5052 pf_tcp_iss(struct pf_pdesc *pd) 5053 { 5054 SHA512_CTX ctx; 5055 union { 5056 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5057 uint32_t words[1]; 5058 } digest; 5059 5060 if (V_pf_tcp_secret_init == 0) { 5061 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5062 SHA512_Init(&V_pf_tcp_secret_ctx); 5063 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5064 sizeof(V_pf_tcp_secret)); 5065 V_pf_tcp_secret_init = 1; 5066 } 5067 5068 ctx = V_pf_tcp_secret_ctx; 5069 5070 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5071 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5072 switch (pd->af) { 5073 case AF_INET6: 5074 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5075 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5076 break; 5077 case AF_INET: 5078 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5079 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5080 break; 5081 } 5082 SHA512_Final(digest.bytes, &ctx); 5083 V_pf_tcp_iss_off += 4096; 5084 #define ISN_RANDOM_INCREMENT (4096 - 1) 5085 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5086 V_pf_tcp_iss_off); 5087 #undef ISN_RANDOM_INCREMENT 5088 } 5089 5090 static bool 5091 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5092 { 5093 bool match = true; 5094 5095 /* Always matches if not set */ 5096 if (! r->isset) 5097 return (!r->neg); 5098 5099 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5100 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5101 match = false; 5102 break; 5103 } 5104 } 5105 5106 return (match ^ r->neg); 5107 } 5108 5109 static int 5110 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5111 { 5112 if (*tag == -1) 5113 *tag = mtag; 5114 5115 return ((!r->match_tag_not && r->match_tag == *tag) || 5116 (r->match_tag_not && r->match_tag != *tag)); 5117 } 5118 5119 static void 5120 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5121 { 5122 /* If we don't have the interface drop the packet. */ 5123 if (ifp == NULL) { 5124 m_freem(m); 5125 return; 5126 } 5127 5128 switch (ifp->if_type) { 5129 case IFT_ETHER: 5130 case IFT_XETHER: 5131 case IFT_L2VLAN: 5132 case IFT_BRIDGE: 5133 case IFT_IEEE8023ADLAG: 5134 break; 5135 default: 5136 m_freem(m); 5137 return; 5138 } 5139 5140 ifp->if_transmit(ifp, m); 5141 } 5142 5143 static int 5144 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5145 { 5146 #ifdef INET 5147 struct ip ip; 5148 #endif 5149 #ifdef INET6 5150 struct ip6_hdr ip6; 5151 #endif 5152 struct mbuf *m = *m0; 5153 struct ether_header *e; 5154 struct pf_keth_rule *r, *rm, *a = NULL; 5155 struct pf_keth_ruleset *ruleset = NULL; 5156 struct pf_mtag *mtag; 5157 struct pf_keth_ruleq *rules; 5158 struct pf_addr *src = NULL, *dst = NULL; 5159 struct pfi_kkif *bridge_to; 5160 sa_family_t af = 0; 5161 uint16_t proto; 5162 int asd = 0, match = 0; 5163 int tag = -1; 5164 uint8_t action; 5165 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5166 5167 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5168 NET_EPOCH_ASSERT(); 5169 5170 PF_RULES_RLOCK_TRACKER; 5171 5172 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5173 5174 mtag = pf_find_mtag(m); 5175 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5176 /* Dummynet re-injects packets after they've 5177 * completed their delay. We've already 5178 * processed them, so pass unconditionally. */ 5179 5180 /* But only once. We may see the packet multiple times (e.g. 5181 * PFIL_IN/PFIL_OUT). */ 5182 pf_dummynet_flag_remove(m, mtag); 5183 5184 return (PF_PASS); 5185 } 5186 5187 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5188 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5189 DPFPRINTF(PF_DEBUG_URGENT, 5190 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 5191 ", pullup failed\n")); 5192 return (PF_DROP); 5193 } 5194 e = mtod(m, struct ether_header *); 5195 proto = ntohs(e->ether_type); 5196 5197 switch (proto) { 5198 #ifdef INET 5199 case ETHERTYPE_IP: { 5200 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5201 sizeof(ip))) 5202 return (PF_DROP); 5203 5204 af = AF_INET; 5205 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5206 (caddr_t)&ip); 5207 src = (struct pf_addr *)&ip.ip_src; 5208 dst = (struct pf_addr *)&ip.ip_dst; 5209 break; 5210 } 5211 #endif /* INET */ 5212 #ifdef INET6 5213 case ETHERTYPE_IPV6: { 5214 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5215 sizeof(ip6))) 5216 return (PF_DROP); 5217 5218 af = AF_INET6; 5219 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5220 (caddr_t)&ip6); 5221 src = (struct pf_addr *)&ip6.ip6_src; 5222 dst = (struct pf_addr *)&ip6.ip6_dst; 5223 break; 5224 } 5225 #endif /* INET6 */ 5226 } 5227 5228 PF_RULES_RLOCK(); 5229 5230 ruleset = V_pf_keth; 5231 rules = atomic_load_ptr(&ruleset->active.rules); 5232 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5233 counter_u64_add(r->evaluations, 1); 5234 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5235 5236 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5237 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5238 "kif"); 5239 r = r->skip[PFE_SKIP_IFP].ptr; 5240 } 5241 else if (r->direction && r->direction != dir) { 5242 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5243 "dir"); 5244 r = r->skip[PFE_SKIP_DIR].ptr; 5245 } 5246 else if (r->proto && r->proto != proto) { 5247 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5248 "proto"); 5249 r = r->skip[PFE_SKIP_PROTO].ptr; 5250 } 5251 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5252 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5253 "src"); 5254 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5255 } 5256 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5257 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5258 "dst"); 5259 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5260 } 5261 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5262 r->ipsrc.neg, kif, M_GETFIB(m))) { 5263 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5264 "ip_src"); 5265 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5266 } 5267 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5268 r->ipdst.neg, kif, M_GETFIB(m))) { 5269 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5270 "ip_dst"); 5271 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5272 } 5273 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5274 mtag ? mtag->tag : 0)) { 5275 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5276 "match_tag"); 5277 r = TAILQ_NEXT(r, entries); 5278 } 5279 else { 5280 if (r->tag) 5281 tag = r->tag; 5282 if (r->anchor == NULL) { 5283 /* Rule matches */ 5284 rm = r; 5285 5286 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5287 5288 if (r->quick) 5289 break; 5290 5291 r = TAILQ_NEXT(r, entries); 5292 } else { 5293 pf_step_into_keth_anchor(anchor_stack, &asd, 5294 &ruleset, &r, &a, &match); 5295 } 5296 } 5297 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5298 &ruleset, &r, &a, &match)) 5299 break; 5300 } 5301 5302 r = rm; 5303 5304 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5305 5306 /* Default to pass. */ 5307 if (r == NULL) { 5308 PF_RULES_RUNLOCK(); 5309 return (PF_PASS); 5310 } 5311 5312 /* Execute action. */ 5313 counter_u64_add(r->packets[dir == PF_OUT], 1); 5314 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5315 pf_update_timestamp(r); 5316 5317 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5318 if (r->action == PF_DROP) { 5319 PF_RULES_RUNLOCK(); 5320 return (PF_DROP); 5321 } 5322 5323 if (tag > 0) { 5324 if (mtag == NULL) 5325 mtag = pf_get_mtag(m); 5326 if (mtag == NULL) { 5327 PF_RULES_RUNLOCK(); 5328 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5329 return (PF_DROP); 5330 } 5331 mtag->tag = tag; 5332 } 5333 5334 if (r->qid != 0) { 5335 if (mtag == NULL) 5336 mtag = pf_get_mtag(m); 5337 if (mtag == NULL) { 5338 PF_RULES_RUNLOCK(); 5339 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5340 return (PF_DROP); 5341 } 5342 mtag->qid = r->qid; 5343 } 5344 5345 action = r->action; 5346 bridge_to = r->bridge_to; 5347 5348 /* Dummynet */ 5349 if (r->dnpipe) { 5350 struct ip_fw_args dnflow; 5351 5352 /* Drop packet if dummynet is not loaded. */ 5353 if (ip_dn_io_ptr == NULL) { 5354 PF_RULES_RUNLOCK(); 5355 m_freem(m); 5356 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5357 return (PF_DROP); 5358 } 5359 if (mtag == NULL) 5360 mtag = pf_get_mtag(m); 5361 if (mtag == NULL) { 5362 PF_RULES_RUNLOCK(); 5363 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5364 return (PF_DROP); 5365 } 5366 5367 bzero(&dnflow, sizeof(dnflow)); 5368 5369 /* We don't have port numbers here, so we set 0. That means 5370 * that we'll be somewhat limited in distinguishing flows (i.e. 5371 * only based on IP addresses, not based on port numbers), but 5372 * it's better than nothing. */ 5373 dnflow.f_id.dst_port = 0; 5374 dnflow.f_id.src_port = 0; 5375 dnflow.f_id.proto = 0; 5376 5377 dnflow.rule.info = r->dnpipe; 5378 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5379 if (r->dnflags & PFRULE_DN_IS_PIPE) 5380 dnflow.rule.info |= IPFW_IS_PIPE; 5381 5382 dnflow.f_id.extra = dnflow.rule.info; 5383 5384 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5385 dnflow.flags |= IPFW_ARGS_ETHER; 5386 dnflow.ifp = kif->pfik_ifp; 5387 5388 switch (af) { 5389 case AF_INET: 5390 dnflow.f_id.addr_type = 4; 5391 dnflow.f_id.src_ip = src->v4.s_addr; 5392 dnflow.f_id.dst_ip = dst->v4.s_addr; 5393 break; 5394 case AF_INET6: 5395 dnflow.flags |= IPFW_ARGS_IP6; 5396 dnflow.f_id.addr_type = 6; 5397 dnflow.f_id.src_ip6 = src->v6; 5398 dnflow.f_id.dst_ip6 = dst->v6; 5399 break; 5400 } 5401 5402 PF_RULES_RUNLOCK(); 5403 5404 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5405 ip_dn_io_ptr(m0, &dnflow); 5406 if (*m0 != NULL) 5407 pf_dummynet_flag_remove(m, mtag); 5408 } else { 5409 PF_RULES_RUNLOCK(); 5410 } 5411 5412 if (action == PF_PASS && bridge_to) { 5413 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5414 *m0 = NULL; /* We've eaten the packet. */ 5415 } 5416 5417 return (action); 5418 } 5419 5420 #define PF_TEST_ATTRIB(t, a)\ 5421 do { \ 5422 if (t) { \ 5423 r = a; \ 5424 goto nextrule; \ 5425 } \ 5426 } while (0) 5427 5428 static int 5429 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5430 struct pf_pdesc *pd, struct pf_krule **am, 5431 struct pf_kruleset **rsm, struct inpcb *inp) 5432 { 5433 struct pf_krule *nr = NULL; 5434 struct pf_krule *r, *a = NULL; 5435 struct pf_kruleset *ruleset = NULL; 5436 struct pf_krule_slist match_rules; 5437 struct pf_krule_item *ri; 5438 struct tcphdr *th = &pd->hdr.tcp; 5439 struct pf_state_key *sk = NULL, *nk = NULL; 5440 u_short reason, transerror; 5441 int rewrite = 0; 5442 int tag = -1; 5443 int asd = 0; 5444 int match = 0; 5445 int state_icmp = 0, icmp_dir, multi; 5446 u_int16_t virtual_type, virtual_id; 5447 u_int16_t bproto_sum = 0, bip_sum = 0; 5448 u_int8_t icmptype = 0, icmpcode = 0; 5449 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5450 struct pf_udp_mapping *udp_mapping = NULL; 5451 5452 PF_RULES_RASSERT(); 5453 5454 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5455 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5456 5457 SLIST_INIT(&match_rules); 5458 5459 if (inp != NULL) { 5460 INP_LOCK_ASSERT(inp); 5461 pd->lookup.uid = inp->inp_cred->cr_uid; 5462 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 5463 pd->lookup.done = 1; 5464 } 5465 5466 switch (pd->virtual_proto) { 5467 case IPPROTO_TCP: 5468 pd->nsport = th->th_sport; 5469 pd->ndport = th->th_dport; 5470 break; 5471 case IPPROTO_UDP: 5472 pd->nsport = pd->hdr.udp.uh_sport; 5473 pd->ndport = pd->hdr.udp.uh_dport; 5474 break; 5475 case IPPROTO_SCTP: 5476 pd->nsport = pd->hdr.sctp.src_port; 5477 pd->ndport = pd->hdr.sctp.dest_port; 5478 break; 5479 #ifdef INET 5480 case IPPROTO_ICMP: 5481 MPASS(pd->af == AF_INET); 5482 icmptype = pd->hdr.icmp.icmp_type; 5483 icmpcode = pd->hdr.icmp.icmp_code; 5484 state_icmp = pf_icmp_mapping(pd, icmptype, 5485 &icmp_dir, &multi, &virtual_id, &virtual_type); 5486 if (icmp_dir == PF_IN) { 5487 pd->nsport = virtual_id; 5488 pd->ndport = virtual_type; 5489 } else { 5490 pd->nsport = virtual_type; 5491 pd->ndport = virtual_id; 5492 } 5493 break; 5494 #endif /* INET */ 5495 #ifdef INET6 5496 case IPPROTO_ICMPV6: 5497 MPASS(pd->af == AF_INET6); 5498 icmptype = pd->hdr.icmp6.icmp6_type; 5499 icmpcode = pd->hdr.icmp6.icmp6_code; 5500 state_icmp = pf_icmp_mapping(pd, icmptype, 5501 &icmp_dir, &multi, &virtual_id, &virtual_type); 5502 if (icmp_dir == PF_IN) { 5503 pd->nsport = virtual_id; 5504 pd->ndport = virtual_type; 5505 } else { 5506 pd->nsport = virtual_type; 5507 pd->ndport = virtual_id; 5508 } 5509 5510 break; 5511 #endif /* INET6 */ 5512 default: 5513 pd->nsport = pd->ndport = 0; 5514 break; 5515 } 5516 pd->osport = pd->nsport; 5517 pd->odport = pd->ndport; 5518 5519 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5520 5521 /* check packet for BINAT/NAT/RDR */ 5522 transerror = pf_get_translation(pd, pd->off, &sk, &nk, anchor_stack, 5523 &nr, &udp_mapping); 5524 switch (transerror) { 5525 default: 5526 /* A translation error occurred. */ 5527 REASON_SET(&reason, transerror); 5528 goto cleanup; 5529 case PFRES_MAX: 5530 /* No match. */ 5531 break; 5532 case PFRES_MATCH: 5533 KASSERT(sk != NULL, ("%s: null sk", __func__)); 5534 KASSERT(nk != NULL, ("%s: null nk", __func__)); 5535 5536 if (nr->log) { 5537 PFLOG_PACKET(nr->action, PFRES_MATCH, nr, a, 5538 ruleset, pd, 1); 5539 } 5540 5541 if (pd->ip_sum) 5542 bip_sum = *pd->ip_sum; 5543 5544 switch (pd->proto) { 5545 case IPPROTO_TCP: 5546 bproto_sum = th->th_sum; 5547 5548 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5549 nk->port[pd->sidx] != pd->nsport) { 5550 pf_change_ap(pd->m, pd->src, &th->th_sport, 5551 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5552 nk->port[pd->sidx], 0, pd->af, pd->naf); 5553 pd->sport = &th->th_sport; 5554 pd->nsport = th->th_sport; 5555 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5556 } 5557 5558 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5559 nk->port[pd->didx] != pd->ndport) { 5560 pf_change_ap(pd->m, pd->dst, &th->th_dport, 5561 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5562 nk->port[pd->didx], 0, pd->af, pd->naf); 5563 pd->dport = &th->th_dport; 5564 pd->ndport = th->th_dport; 5565 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5566 } 5567 rewrite++; 5568 break; 5569 case IPPROTO_UDP: 5570 bproto_sum = pd->hdr.udp.uh_sum; 5571 5572 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5573 nk->port[pd->sidx] != pd->nsport) { 5574 pf_change_ap(pd->m, pd->src, 5575 &pd->hdr.udp.uh_sport, 5576 pd->ip_sum, &pd->hdr.udp.uh_sum, 5577 &nk->addr[pd->sidx], 5578 nk->port[pd->sidx], 1, pd->af, pd->naf); 5579 pd->sport = &pd->hdr.udp.uh_sport; 5580 pd->nsport = pd->hdr.udp.uh_sport; 5581 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5582 } 5583 5584 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5585 nk->port[pd->didx] != pd->ndport) { 5586 pf_change_ap(pd->m, pd->dst, 5587 &pd->hdr.udp.uh_dport, 5588 pd->ip_sum, &pd->hdr.udp.uh_sum, 5589 &nk->addr[pd->didx], 5590 nk->port[pd->didx], 1, pd->af, pd->naf); 5591 pd->dport = &pd->hdr.udp.uh_dport; 5592 pd->ndport = pd->hdr.udp.uh_dport; 5593 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5594 } 5595 rewrite++; 5596 break; 5597 case IPPROTO_SCTP: { 5598 uint16_t checksum = 0; 5599 5600 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5601 nk->port[pd->sidx] != pd->nsport) { 5602 pf_change_ap(pd->m, pd->src, 5603 &pd->hdr.sctp.src_port, pd->ip_sum, &checksum, 5604 &nk->addr[pd->sidx], 5605 nk->port[pd->sidx], 1, pd->af, pd->naf); 5606 pd->sport = &pd->hdr.sctp.src_port; 5607 pd->nsport = pd->hdr.sctp.src_port; 5608 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5609 } 5610 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5611 nk->port[pd->didx] != pd->ndport) { 5612 pf_change_ap(pd->m, pd->dst, 5613 &pd->hdr.sctp.dest_port, pd->ip_sum, &checksum, 5614 &nk->addr[pd->didx], 5615 nk->port[pd->didx], 1, pd->af, pd->naf); 5616 pd->dport = &pd->hdr.sctp.dest_port; 5617 pd->ndport = pd->hdr.sctp.dest_port; 5618 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5619 } 5620 break; 5621 } 5622 #ifdef INET 5623 case IPPROTO_ICMP: 5624 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 5625 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 5626 nk->addr[pd->sidx].v4.s_addr, 0); 5627 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5628 } 5629 5630 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 5631 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 5632 nk->addr[pd->didx].v4.s_addr, 0); 5633 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5634 } 5635 5636 if (virtual_type == htons(ICMP_ECHO) && 5637 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5638 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5639 pd->hdr.icmp.icmp_cksum, pd->nsport, 5640 nk->port[pd->sidx], 0); 5641 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5642 pd->sport = &pd->hdr.icmp.icmp_id; 5643 } 5644 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5645 break; 5646 #endif /* INET */ 5647 #ifdef INET6 5648 case IPPROTO_ICMPV6: 5649 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 5650 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 5651 &nk->addr[pd->sidx], 0); 5652 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5653 } 5654 5655 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 5656 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 5657 &nk->addr[pd->didx], 0); 5658 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5659 } 5660 rewrite++; 5661 break; 5662 #endif /* INET */ 5663 default: 5664 switch (pd->af) { 5665 #ifdef INET 5666 case AF_INET: 5667 if (PF_ANEQ(&pd->nsaddr, 5668 &nk->addr[pd->sidx], AF_INET)) { 5669 pf_change_a(&pd->src->v4.s_addr, 5670 pd->ip_sum, 5671 nk->addr[pd->sidx].v4.s_addr, 0); 5672 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5673 } 5674 5675 if (PF_ANEQ(&pd->ndaddr, 5676 &nk->addr[pd->didx], AF_INET)) { 5677 pf_change_a(&pd->dst->v4.s_addr, 5678 pd->ip_sum, 5679 nk->addr[pd->didx].v4.s_addr, 0); 5680 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5681 } 5682 break; 5683 #endif /* INET */ 5684 #ifdef INET6 5685 case AF_INET6: 5686 if (PF_ANEQ(&pd->nsaddr, 5687 &nk->addr[pd->sidx], AF_INET6)) { 5688 PF_ACPY(&pd->nsaddr, &nk->addr[pd->sidx], pd->af); 5689 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5690 } 5691 5692 if (PF_ANEQ(&pd->ndaddr, 5693 &nk->addr[pd->didx], AF_INET6)) { 5694 PF_ACPY(&pd->ndaddr, &nk->addr[pd->didx], pd->af); 5695 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5696 } 5697 break; 5698 #endif /* INET */ 5699 } 5700 break; 5701 } 5702 if (nr->natpass) 5703 r = NULL; 5704 } 5705 5706 while (r != NULL) { 5707 if (pd->related_rule) { 5708 *rm = pd->related_rule; 5709 break; 5710 } 5711 pf_counter_u64_add(&r->evaluations, 1); 5712 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5713 r->skip[PF_SKIP_IFP]); 5714 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5715 r->skip[PF_SKIP_DIR]); 5716 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5717 r->skip[PF_SKIP_AF]); 5718 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5719 r->skip[PF_SKIP_PROTO]); 5720 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5721 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5722 r->skip[PF_SKIP_SRC_ADDR]); 5723 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5724 r->dst.neg, NULL, M_GETFIB(pd->m)), 5725 r->skip[PF_SKIP_DST_ADDR]); 5726 switch (pd->virtual_proto) { 5727 case PF_VPROTO_FRAGMENT: 5728 /* tcp/udp only. port_op always 0 in other cases */ 5729 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5730 TAILQ_NEXT(r, entries)); 5731 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5732 TAILQ_NEXT(r, entries)); 5733 /* icmp only. type/code always 0 in other cases */ 5734 PF_TEST_ATTRIB((r->type || r->code), 5735 TAILQ_NEXT(r, entries)); 5736 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5737 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5738 TAILQ_NEXT(r, entries)); 5739 break; 5740 5741 case IPPROTO_TCP: 5742 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(th)) != r->flags, 5743 TAILQ_NEXT(r, entries)); 5744 /* FALLTHROUGH */ 5745 case IPPROTO_SCTP: 5746 case IPPROTO_UDP: 5747 /* tcp/udp only. port_op always 0 in other cases */ 5748 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5749 r->src.port[0], r->src.port[1], pd->nsport), 5750 r->skip[PF_SKIP_SRC_PORT]); 5751 /* tcp/udp only. port_op always 0 in other cases */ 5752 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5753 r->dst.port[0], r->dst.port[1], pd->ndport), 5754 r->skip[PF_SKIP_DST_PORT]); 5755 /* tcp/udp only. uid.op always 0 in other cases */ 5756 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5757 pf_socket_lookup(pd), 1)) && 5758 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5759 pd->lookup.uid), 5760 TAILQ_NEXT(r, entries)); 5761 /* tcp/udp only. gid.op always 0 in other cases */ 5762 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5763 pf_socket_lookup(pd), 1)) && 5764 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5765 pd->lookup.gid), 5766 TAILQ_NEXT(r, entries)); 5767 break; 5768 5769 case IPPROTO_ICMP: 5770 case IPPROTO_ICMPV6: 5771 /* icmp only. type always 0 in other cases */ 5772 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5773 TAILQ_NEXT(r, entries)); 5774 /* icmp only. type always 0 in other cases */ 5775 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5776 TAILQ_NEXT(r, entries)); 5777 break; 5778 5779 default: 5780 break; 5781 } 5782 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5783 TAILQ_NEXT(r, entries)); 5784 PF_TEST_ATTRIB(r->prio && 5785 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5786 TAILQ_NEXT(r, entries)); 5787 PF_TEST_ATTRIB(r->prob && 5788 r->prob <= arc4random(), 5789 TAILQ_NEXT(r, entries)); 5790 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5791 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5792 TAILQ_NEXT(r, entries)); 5793 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 5794 r->rcvifnot), 5795 TAILQ_NEXT(r, entries)); 5796 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5797 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5798 TAILQ_NEXT(r, entries)); 5799 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5800 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5801 pf_osfp_fingerprint(pd, th), 5802 r->os_fingerprint)), 5803 TAILQ_NEXT(r, entries)); 5804 /* FALLTHROUGH */ 5805 if (r->tag) 5806 tag = r->tag; 5807 if (r->anchor == NULL) { 5808 if (r->action == PF_MATCH) { 5809 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5810 if (ri == NULL) { 5811 REASON_SET(&reason, PFRES_MEMORY); 5812 goto cleanup; 5813 } 5814 ri->r = r; 5815 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5816 pf_counter_u64_critical_enter(); 5817 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5818 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5819 pf_counter_u64_critical_exit(); 5820 pf_rule_to_actions(r, &pd->act); 5821 if (r->rule_flag & PFRULE_AFTO) 5822 pd->naf = r->naf; 5823 if (pd->af != pd->naf) { 5824 if (pf_get_transaddr_af(r, pd) == -1) { 5825 REASON_SET(&reason, PFRES_TRANSLATE); 5826 goto cleanup; 5827 } 5828 } 5829 if (r->log || pd->act.log & PF_LOG_MATCHES) 5830 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5831 a, ruleset, pd, 1); 5832 } else { 5833 match = asd; 5834 *rm = r; 5835 *am = a; 5836 *rsm = ruleset; 5837 if (pd->act.log & PF_LOG_MATCHES) 5838 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5839 a, ruleset, pd, 1); 5840 } 5841 if (r->quick) 5842 break; 5843 r = TAILQ_NEXT(r, entries); 5844 } else 5845 pf_step_into_anchor(anchor_stack, &asd, 5846 &ruleset, PF_RULESET_FILTER, &r, &a); 5847 nextrule: 5848 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5849 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5850 break; 5851 } 5852 r = *rm; 5853 a = *am; 5854 ruleset = *rsm; 5855 5856 REASON_SET(&reason, PFRES_MATCH); 5857 5858 /* apply actions for last matching pass/block rule */ 5859 pf_rule_to_actions(r, &pd->act); 5860 if (r->rule_flag & PFRULE_AFTO) 5861 pd->naf = r->naf; 5862 if (pd->af != pd->naf) { 5863 if (pf_get_transaddr_af(r, pd) == -1) { 5864 REASON_SET(&reason, PFRES_TRANSLATE); 5865 goto cleanup; 5866 } 5867 } 5868 5869 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5870 if (rewrite) 5871 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5872 PFLOG_PACKET(r->action, reason, r, a, ruleset, pd, 1); 5873 } 5874 5875 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5876 (r->action == PF_DROP) && 5877 ((r->rule_flag & PFRULE_RETURNRST) || 5878 (r->rule_flag & PFRULE_RETURNICMP) || 5879 (r->rule_flag & PFRULE_RETURN))) { 5880 pf_return(r, nr, pd, sk, th, bproto_sum, 5881 bip_sum, &reason, r->rtableid); 5882 } 5883 5884 if (r->action == PF_DROP) 5885 goto cleanup; 5886 5887 if (tag > 0 && pf_tag_packet(pd, tag)) { 5888 REASON_SET(&reason, PFRES_MEMORY); 5889 goto cleanup; 5890 } 5891 if (pd->act.rtableid >= 0) 5892 M_SETFIB(pd->m, pd->act.rtableid); 5893 5894 if (r->rt) { 5895 struct pf_ksrc_node *sn = NULL; 5896 struct pf_srchash *snh = NULL; 5897 struct pf_kpool *pool = &r->route; 5898 5899 /* Backwards compatibility. */ 5900 if (TAILQ_EMPTY(&pool->list)) 5901 pool = &r->rdr; 5902 5903 /* 5904 * Set act.rt here instead of in pf_rule_to_actions() because 5905 * it is applied only from the last pass rule. 5906 */ 5907 pd->act.rt = r->rt; 5908 /* Don't use REASON_SET, pf_map_addr increases the reason counters */ 5909 reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, 5910 &pd->act.rt_kif, NULL, &sn, &snh, pool, PF_SN_ROUTE); 5911 if (reason != 0) 5912 goto cleanup; 5913 } 5914 5915 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5916 (!state_icmp && (r->keep_state || nr != NULL || 5917 (pd->flags & PFDESC_TCP_NORM)))) { 5918 int action; 5919 bool nat64; 5920 5921 action = pf_create_state(r, nr, a, pd, nk, sk, 5922 &rewrite, sm, tag, bproto_sum, bip_sum, 5923 &match_rules, udp_mapping); 5924 if (action != PF_PASS) { 5925 pf_udp_mapping_release(udp_mapping); 5926 pd->act.log |= PF_LOG_FORCE; 5927 if (action == PF_DROP && 5928 (r->rule_flag & PFRULE_RETURN)) 5929 pf_return(r, nr, pd, sk, th, 5930 bproto_sum, bip_sum, &reason, 5931 pd->act.rtableid); 5932 return (action); 5933 } 5934 5935 nat64 = pd->af != pd->naf; 5936 if (nat64) { 5937 struct pf_state_key *_sk; 5938 int ret; 5939 5940 if (sk == NULL) 5941 sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 5942 if (nk == NULL) 5943 nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 5944 if (pd->dir == PF_IN) 5945 _sk = sk; 5946 else 5947 _sk = nk; 5948 5949 ret = pf_translate(pd, 5950 &_sk->addr[pd->didx], 5951 _sk->port[pd->didx], 5952 &_sk->addr[pd->sidx], 5953 _sk->port[pd->sidx], 5954 virtual_type, icmp_dir); 5955 if (ret < 0) 5956 goto cleanup; 5957 5958 rewrite += ret; 5959 } 5960 } else { 5961 while ((ri = SLIST_FIRST(&match_rules))) { 5962 SLIST_REMOVE_HEAD(&match_rules, entry); 5963 free(ri, M_PF_RULE_ITEM); 5964 } 5965 5966 uma_zfree(V_pf_state_key_z, sk); 5967 uma_zfree(V_pf_state_key_z, nk); 5968 pf_udp_mapping_release(udp_mapping); 5969 } 5970 5971 /* copy back packet headers if we performed NAT operations */ 5972 if (rewrite) 5973 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5974 5975 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5976 pd->dir == PF_OUT && 5977 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 5978 /* 5979 * We want the state created, but we dont 5980 * want to send this in case a partner 5981 * firewall has to know about it to allow 5982 * replies through it. 5983 */ 5984 return (PF_DEFER); 5985 5986 if (rewrite && sk != NULL && nk != NULL && sk->af != nk->af) { 5987 return (PF_AFRT); 5988 } else 5989 return (PF_PASS); 5990 5991 cleanup: 5992 while ((ri = SLIST_FIRST(&match_rules))) { 5993 SLIST_REMOVE_HEAD(&match_rules, entry); 5994 free(ri, M_PF_RULE_ITEM); 5995 } 5996 5997 uma_zfree(V_pf_state_key_z, sk); 5998 uma_zfree(V_pf_state_key_z, nk); 5999 pf_udp_mapping_release(udp_mapping); 6000 6001 return (PF_DROP); 6002 } 6003 6004 static int 6005 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 6006 struct pf_pdesc *pd, struct pf_state_key *nk, struct pf_state_key *sk, 6007 int *rewrite, struct pf_kstate **sm, int tag, u_int16_t bproto_sum, 6008 u_int16_t bip_sum, struct pf_krule_slist *match_rules, 6009 struct pf_udp_mapping *udp_mapping) 6010 { 6011 struct pf_kstate *s = NULL; 6012 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6013 /* 6014 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6015 * but for PF_SN_NAT it is different. Don't try optimizing it, 6016 * just store all 3 hashes. 6017 */ 6018 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6019 struct tcphdr *th = &pd->hdr.tcp; 6020 u_int16_t mss = V_tcp_mssdflt; 6021 u_short reason, sn_reason; 6022 struct pf_krule_item *ri; 6023 struct pf_kpool *pool_route = &r->route; 6024 6025 /* check maximums */ 6026 if (r->max_states && 6027 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6028 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6029 REASON_SET(&reason, PFRES_MAXSTATES); 6030 goto csfailed; 6031 } 6032 /* src node for limits */ 6033 if ((r->rule_flag & PFRULE_SRCTRACK) && 6034 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6035 NULL, NULL, PF_SN_LIMIT)) != 0) { 6036 REASON_SET(&reason, sn_reason); 6037 goto csfailed; 6038 } 6039 /* src node for route-to rule */ 6040 if (TAILQ_EMPTY(&pool_route->list)) /* Backwards compatibility. */ 6041 pool_route = &r->rdr; 6042 if ((pool_route->opts & PF_POOL_STICKYADDR) && 6043 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6044 &pd->act.rt_addr, pd->act.rt_kif, PF_SN_ROUTE)) != 0) { 6045 REASON_SET(&reason, sn_reason); 6046 goto csfailed; 6047 } 6048 /* src node for translation rule */ 6049 if (nr != NULL && (nr->rdr.opts & PF_POOL_STICKYADDR) && 6050 (sn_reason = pf_insert_src_node(sns, snhs, nr, &sk->addr[pd->sidx], 6051 pd->af, &nk->addr[1], NULL, PF_SN_NAT)) != 0 ) { 6052 REASON_SET(&reason, sn_reason); 6053 goto csfailed; 6054 } 6055 s = pf_alloc_state(M_NOWAIT); 6056 if (s == NULL) { 6057 REASON_SET(&reason, PFRES_MEMORY); 6058 goto csfailed; 6059 } 6060 s->rule = r; 6061 s->nat_rule = nr; 6062 s->anchor = a; 6063 memcpy(&s->match_rules, match_rules, sizeof(s->match_rules)); 6064 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6065 6066 STATE_INC_COUNTERS(s); 6067 if (r->allow_opts) 6068 s->state_flags |= PFSTATE_ALLOWOPTS; 6069 if (r->rule_flag & PFRULE_STATESLOPPY) 6070 s->state_flags |= PFSTATE_SLOPPY; 6071 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6072 s->state_flags |= PFSTATE_SCRUB_TCP; 6073 if ((r->rule_flag & PFRULE_PFLOW) || 6074 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 6075 s->state_flags |= PFSTATE_PFLOW; 6076 6077 s->act.log = pd->act.log & PF_LOG_ALL; 6078 s->sync_state = PFSYNC_S_NONE; 6079 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6080 6081 if (nr != NULL) 6082 s->act.log |= nr->log & PF_LOG_ALL; 6083 switch (pd->proto) { 6084 case IPPROTO_TCP: 6085 s->src.seqlo = ntohl(th->th_seq); 6086 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6087 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6088 r->keep_state == PF_STATE_MODULATE) { 6089 /* Generate sequence number modulator */ 6090 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6091 0) 6092 s->src.seqdiff = 1; 6093 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6094 htonl(s->src.seqlo + s->src.seqdiff), 0); 6095 *rewrite = 1; 6096 } else 6097 s->src.seqdiff = 0; 6098 if (tcp_get_flags(th) & TH_SYN) { 6099 s->src.seqhi++; 6100 s->src.wscale = pf_get_wscale(pd); 6101 } 6102 s->src.max_win = MAX(ntohs(th->th_win), 1); 6103 if (s->src.wscale & PF_WSCALE_MASK) { 6104 /* Remove scale factor from initial window */ 6105 int win = s->src.max_win; 6106 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6107 s->src.max_win = (win - 1) >> 6108 (s->src.wscale & PF_WSCALE_MASK); 6109 } 6110 if (tcp_get_flags(th) & TH_FIN) 6111 s->src.seqhi++; 6112 s->dst.seqhi = 1; 6113 s->dst.max_win = 1; 6114 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6115 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6116 s->timeout = PFTM_TCP_FIRST_PACKET; 6117 atomic_add_32(&V_pf_status.states_halfopen, 1); 6118 break; 6119 case IPPROTO_UDP: 6120 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6121 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6122 s->timeout = PFTM_UDP_FIRST_PACKET; 6123 break; 6124 case IPPROTO_SCTP: 6125 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6126 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6127 s->timeout = PFTM_SCTP_FIRST_PACKET; 6128 break; 6129 case IPPROTO_ICMP: 6130 #ifdef INET6 6131 case IPPROTO_ICMPV6: 6132 #endif 6133 s->timeout = PFTM_ICMP_FIRST_PACKET; 6134 break; 6135 default: 6136 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6137 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6138 s->timeout = PFTM_OTHER_FIRST_PACKET; 6139 } 6140 6141 s->creation = s->expire = pf_get_uptime(); 6142 6143 if (pd->proto == IPPROTO_TCP) { 6144 if (s->state_flags & PFSTATE_SCRUB_TCP && 6145 pf_normalize_tcp_init(pd, th, &s->src, &s->dst)) { 6146 REASON_SET(&reason, PFRES_MEMORY); 6147 goto csfailed; 6148 } 6149 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6150 pf_normalize_tcp_stateful(pd, &reason, th, s, 6151 &s->src, &s->dst, rewrite)) { 6152 /* This really shouldn't happen!!! */ 6153 DPFPRINTF(PF_DEBUG_URGENT, 6154 ("pf_normalize_tcp_stateful failed on first " 6155 "pkt\n")); 6156 goto csfailed; 6157 } 6158 } else if (pd->proto == IPPROTO_SCTP) { 6159 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6160 goto csfailed; 6161 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6162 goto csfailed; 6163 } 6164 s->direction = pd->dir; 6165 6166 /* 6167 * sk/nk could already been setup by pf_get_translation(). 6168 */ 6169 if (nr == NULL) { 6170 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 6171 __func__, nr, sk, nk)); 6172 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6173 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6174 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, &sk, &nk)) { 6175 goto csfailed; 6176 } 6177 } else 6178 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 6179 __func__, nr, sk, nk)); 6180 6181 /* Swap sk/nk for PF_OUT. */ 6182 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6183 (pd->dir == PF_IN) ? sk : nk, 6184 (pd->dir == PF_IN) ? nk : sk, s)) { 6185 REASON_SET(&reason, PFRES_STATEINS); 6186 goto drop; 6187 } else 6188 *sm = s; 6189 6190 /* 6191 * Lock order is important: first state, then source node. 6192 */ 6193 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6194 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6195 s->sns[sn_type] = sns[sn_type]; 6196 PF_HASHROW_UNLOCK(snhs[sn_type]); 6197 } 6198 } 6199 6200 if (tag > 0) 6201 s->tag = tag; 6202 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6203 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 6204 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6205 /* undo NAT changes, if they have taken place */ 6206 if (nr != NULL) { 6207 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 6208 if (pd->dir == PF_OUT) 6209 skt = s->key[PF_SK_STACK]; 6210 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 6211 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 6212 if (pd->sport) 6213 *pd->sport = skt->port[pd->sidx]; 6214 if (pd->dport) 6215 *pd->dport = skt->port[pd->didx]; 6216 if (pd->ip_sum) 6217 *pd->ip_sum = bip_sum; 6218 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6219 } 6220 s->src.seqhi = htonl(arc4random()); 6221 /* Find mss option */ 6222 int rtid = M_GETFIB(pd->m); 6223 mss = pf_get_mss(pd); 6224 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6225 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6226 s->src.mss = mss; 6227 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6228 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6229 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6230 pd->act.rtableid); 6231 REASON_SET(&reason, PFRES_SYNPROXY); 6232 return (PF_SYNPROXY_DROP); 6233 } 6234 6235 s->udp_mapping = udp_mapping; 6236 6237 return (PF_PASS); 6238 6239 csfailed: 6240 while ((ri = SLIST_FIRST(match_rules))) { 6241 SLIST_REMOVE_HEAD(match_rules, entry); 6242 free(ri, M_PF_RULE_ITEM); 6243 } 6244 6245 uma_zfree(V_pf_state_key_z, sk); 6246 uma_zfree(V_pf_state_key_z, nk); 6247 6248 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6249 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6250 if (--sns[sn_type]->states == 0 && 6251 sns[sn_type]->expire == 0) { 6252 pf_unlink_src_node(sns[sn_type]); 6253 pf_free_src_node(sns[sn_type]); 6254 counter_u64_add( 6255 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6256 } 6257 PF_HASHROW_UNLOCK(snhs[sn_type]); 6258 } 6259 } 6260 6261 drop: 6262 if (s != NULL) { 6263 pf_src_tree_remove_state(s); 6264 s->timeout = PFTM_UNLINKED; 6265 STATE_DEC_COUNTERS(s); 6266 pf_free_state(s); 6267 } 6268 6269 return (PF_DROP); 6270 } 6271 6272 int 6273 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6274 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6275 int icmp_dir) 6276 { 6277 /* 6278 * pf_translate() implements OpenBSD's "new" NAT approach. 6279 * We don't follow it, because it involves a breaking syntax change 6280 * (removing nat/rdr rules, moving it into regular pf rules.) 6281 * It also moves NAT processing to be done after normal rules evaluation 6282 * whereas in FreeBSD that's done before rules processing. 6283 * 6284 * We adopt the function only for nat64, and keep other NAT processing 6285 * before rules processing. 6286 */ 6287 int rewrite = 0; 6288 int afto = pd->af != pd->naf; 6289 6290 MPASS(afto); 6291 6292 switch (pd->proto) { 6293 case IPPROTO_TCP: 6294 if (afto || *pd->sport != sport) { 6295 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6296 saddr, sport, 0, pd->af, pd->naf); 6297 rewrite = 1; 6298 } 6299 if (afto || *pd->dport != dport) { 6300 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6301 daddr, dport, 0, pd->af, pd->naf); 6302 rewrite = 1; 6303 } 6304 break; 6305 6306 case IPPROTO_UDP: 6307 if (afto || *pd->sport != sport) { 6308 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6309 saddr, sport, 1, pd->af, pd->naf); 6310 rewrite = 1; 6311 } 6312 if (afto || *pd->dport != dport) { 6313 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6314 daddr, dport, 1, pd->af, pd->naf); 6315 rewrite = 1; 6316 } 6317 break; 6318 6319 case IPPROTO_SCTP: { 6320 uint16_t checksum = 0; 6321 if (afto || *pd->sport != sport) { 6322 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &checksum, 6323 saddr, sport, 1, pd->af, pd->naf); 6324 rewrite = 1; 6325 } 6326 if (afto || *pd->dport != dport) { 6327 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &checksum, 6328 daddr, dport, 1, pd->af, pd->naf); 6329 rewrite = 1; 6330 } 6331 break; 6332 } 6333 6334 #ifdef INET 6335 case IPPROTO_ICMP: 6336 /* pf_translate() is also used when logging invalid packets */ 6337 if (pd->af != AF_INET) 6338 return (0); 6339 6340 if (afto) { 6341 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6342 return (-1); 6343 pd->proto = IPPROTO_ICMPV6; 6344 rewrite = 1; 6345 } 6346 if (virtual_type == htons(ICMP_ECHO)) { 6347 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6348 6349 if (icmpid != pd->hdr.icmp.icmp_id) { 6350 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6351 pd->hdr.icmp.icmp_cksum, 6352 pd->hdr.icmp.icmp_id, icmpid, 0); 6353 pd->hdr.icmp.icmp_id = icmpid; 6354 /* XXX TODO copyback. */ 6355 rewrite = 1; 6356 } 6357 } 6358 break; 6359 #endif /* INET */ 6360 6361 #ifdef INET6 6362 case IPPROTO_ICMPV6: 6363 /* pf_translate() is also used when logging invalid packets */ 6364 if (pd->af != AF_INET6) 6365 return (0); 6366 6367 if (afto) { 6368 /* ip_sum will be recalculated in pf_translate_af */ 6369 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6370 return (0); 6371 pd->proto = IPPROTO_ICMP; 6372 rewrite = 1; 6373 } 6374 break; 6375 #endif /* INET6 */ 6376 6377 default: 6378 break; 6379 } 6380 6381 return (rewrite); 6382 } 6383 6384 static int 6385 pf_tcp_track_full(struct pf_kstate **state, struct pf_pdesc *pd, 6386 u_short *reason, int *copyback) 6387 { 6388 struct tcphdr *th = &pd->hdr.tcp; 6389 struct pf_state_peer *src, *dst; 6390 u_int16_t win = ntohs(th->th_win); 6391 u_int32_t ack, end, data_end, seq, orig_seq; 6392 u_int8_t sws, dws, psrc, pdst; 6393 int ackskew; 6394 6395 if (pd->dir == (*state)->direction) { 6396 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 6397 src = &(*state)->dst; 6398 dst = &(*state)->src; 6399 } else { 6400 src = &(*state)->src; 6401 dst = &(*state)->dst; 6402 } 6403 psrc = PF_PEER_SRC; 6404 pdst = PF_PEER_DST; 6405 } else { 6406 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 6407 src = &(*state)->src; 6408 dst = &(*state)->dst; 6409 } else { 6410 src = &(*state)->dst; 6411 dst = &(*state)->src; 6412 } 6413 psrc = PF_PEER_DST; 6414 pdst = PF_PEER_SRC; 6415 } 6416 6417 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6418 sws = src->wscale & PF_WSCALE_MASK; 6419 dws = dst->wscale & PF_WSCALE_MASK; 6420 } else 6421 sws = dws = 0; 6422 6423 /* 6424 * Sequence tracking algorithm from Guido van Rooij's paper: 6425 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6426 * tcp_filtering.ps 6427 */ 6428 6429 orig_seq = seq = ntohl(th->th_seq); 6430 if (src->seqlo == 0) { 6431 /* First packet from this end. Set its state */ 6432 6433 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6434 src->scrub == NULL) { 6435 if (pf_normalize_tcp_init(pd, th, src, dst)) { 6436 REASON_SET(reason, PFRES_MEMORY); 6437 return (PF_DROP); 6438 } 6439 } 6440 6441 /* Deferred generation of sequence number modulator */ 6442 if (dst->seqdiff && !src->seqdiff) { 6443 /* use random iss for the TCP server */ 6444 while ((src->seqdiff = arc4random() - seq) == 0) 6445 ; 6446 ack = ntohl(th->th_ack) - dst->seqdiff; 6447 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6448 src->seqdiff), 0); 6449 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6450 *copyback = 1; 6451 } else { 6452 ack = ntohl(th->th_ack); 6453 } 6454 6455 end = seq + pd->p_len; 6456 if (tcp_get_flags(th) & TH_SYN) { 6457 end++; 6458 if (dst->wscale & PF_WSCALE_FLAG) { 6459 src->wscale = pf_get_wscale(pd); 6460 if (src->wscale & PF_WSCALE_FLAG) { 6461 /* Remove scale factor from initial 6462 * window */ 6463 sws = src->wscale & PF_WSCALE_MASK; 6464 win = ((u_int32_t)win + (1 << sws) - 1) 6465 >> sws; 6466 dws = dst->wscale & PF_WSCALE_MASK; 6467 } else { 6468 /* fixup other window */ 6469 dst->max_win = MIN(TCP_MAXWIN, 6470 (u_int32_t)dst->max_win << 6471 (dst->wscale & PF_WSCALE_MASK)); 6472 /* in case of a retrans SYN|ACK */ 6473 dst->wscale = 0; 6474 } 6475 } 6476 } 6477 data_end = end; 6478 if (tcp_get_flags(th) & TH_FIN) 6479 end++; 6480 6481 src->seqlo = seq; 6482 if (src->state < TCPS_SYN_SENT) 6483 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6484 6485 /* 6486 * May need to slide the window (seqhi may have been set by 6487 * the crappy stack check or if we picked up the connection 6488 * after establishment) 6489 */ 6490 if (src->seqhi == 1 || 6491 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6492 src->seqhi = end + MAX(1, dst->max_win << dws); 6493 if (win > src->max_win) 6494 src->max_win = win; 6495 6496 } else { 6497 ack = ntohl(th->th_ack) - dst->seqdiff; 6498 if (src->seqdiff) { 6499 /* Modulate sequence numbers */ 6500 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6501 src->seqdiff), 0); 6502 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6503 *copyback = 1; 6504 } 6505 end = seq + pd->p_len; 6506 if (tcp_get_flags(th) & TH_SYN) 6507 end++; 6508 data_end = end; 6509 if (tcp_get_flags(th) & TH_FIN) 6510 end++; 6511 } 6512 6513 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6514 /* Let it pass through the ack skew check */ 6515 ack = dst->seqlo; 6516 } else if ((ack == 0 && 6517 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6518 /* broken tcp stacks do not set ack */ 6519 (dst->state < TCPS_SYN_SENT)) { 6520 /* 6521 * Many stacks (ours included) will set the ACK number in an 6522 * FIN|ACK if the SYN times out -- no sequence to ACK. 6523 */ 6524 ack = dst->seqlo; 6525 } 6526 6527 if (seq == end) { 6528 /* Ease sequencing restrictions on no data packets */ 6529 seq = src->seqlo; 6530 data_end = end = seq; 6531 } 6532 6533 ackskew = dst->seqlo - ack; 6534 6535 /* 6536 * Need to demodulate the sequence numbers in any TCP SACK options 6537 * (Selective ACK). We could optionally validate the SACK values 6538 * against the current ACK window, either forwards or backwards, but 6539 * I'm not confident that SACK has been implemented properly 6540 * everywhere. It wouldn't surprise me if several stacks accidentally 6541 * SACK too far backwards of previously ACKed data. There really aren't 6542 * any security implications of bad SACKing unless the target stack 6543 * doesn't validate the option length correctly. Someone trying to 6544 * spoof into a TCP connection won't bother blindly sending SACK 6545 * options anyway. 6546 */ 6547 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6548 if (pf_modulate_sack(pd, th, dst)) 6549 *copyback = 1; 6550 } 6551 6552 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6553 if (SEQ_GEQ(src->seqhi, data_end) && 6554 /* Last octet inside other's window space */ 6555 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6556 /* Retrans: not more than one window back */ 6557 (ackskew >= -MAXACKWINDOW) && 6558 /* Acking not more than one reassembled fragment backwards */ 6559 (ackskew <= (MAXACKWINDOW << sws)) && 6560 /* Acking not more than one window forward */ 6561 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6562 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 6563 /* Require an exact/+1 sequence match on resets when possible */ 6564 6565 if (dst->scrub || src->scrub) { 6566 if (pf_normalize_tcp_stateful(pd, reason, th, 6567 *state, src, dst, copyback)) 6568 return (PF_DROP); 6569 } 6570 6571 /* update max window */ 6572 if (src->max_win < win) 6573 src->max_win = win; 6574 /* synchronize sequencing */ 6575 if (SEQ_GT(end, src->seqlo)) 6576 src->seqlo = end; 6577 /* slide the window of what the other end can send */ 6578 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6579 dst->seqhi = ack + MAX((win << sws), 1); 6580 6581 /* update states */ 6582 if (tcp_get_flags(th) & TH_SYN) 6583 if (src->state < TCPS_SYN_SENT) 6584 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6585 if (tcp_get_flags(th) & TH_FIN) 6586 if (src->state < TCPS_CLOSING) 6587 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6588 if (tcp_get_flags(th) & TH_ACK) { 6589 if (dst->state == TCPS_SYN_SENT) { 6590 pf_set_protostate(*state, pdst, 6591 TCPS_ESTABLISHED); 6592 if (src->state == TCPS_ESTABLISHED && 6593 (*state)->sns[PF_SN_LIMIT] != NULL && 6594 pf_src_connlimit(*state)) { 6595 REASON_SET(reason, PFRES_SRCLIMIT); 6596 return (PF_DROP); 6597 } 6598 } else if (dst->state == TCPS_CLOSING) 6599 pf_set_protostate(*state, pdst, 6600 TCPS_FIN_WAIT_2); 6601 } 6602 if (tcp_get_flags(th) & TH_RST) 6603 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6604 6605 /* update expire time */ 6606 (*state)->expire = pf_get_uptime(); 6607 if (src->state >= TCPS_FIN_WAIT_2 && 6608 dst->state >= TCPS_FIN_WAIT_2) 6609 (*state)->timeout = PFTM_TCP_CLOSED; 6610 else if (src->state >= TCPS_CLOSING && 6611 dst->state >= TCPS_CLOSING) 6612 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6613 else if (src->state < TCPS_ESTABLISHED || 6614 dst->state < TCPS_ESTABLISHED) 6615 (*state)->timeout = PFTM_TCP_OPENING; 6616 else if (src->state >= TCPS_CLOSING || 6617 dst->state >= TCPS_CLOSING) 6618 (*state)->timeout = PFTM_TCP_CLOSING; 6619 else 6620 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6621 6622 /* Fall through to PASS packet */ 6623 6624 } else if ((dst->state < TCPS_SYN_SENT || 6625 dst->state >= TCPS_FIN_WAIT_2 || 6626 src->state >= TCPS_FIN_WAIT_2) && 6627 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6628 /* Within a window forward of the originating packet */ 6629 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6630 /* Within a window backward of the originating packet */ 6631 6632 /* 6633 * This currently handles three situations: 6634 * 1) Stupid stacks will shotgun SYNs before their peer 6635 * replies. 6636 * 2) When PF catches an already established stream (the 6637 * firewall rebooted, the state table was flushed, routes 6638 * changed...) 6639 * 3) Packets get funky immediately after the connection 6640 * closes (this should catch Solaris spurious ACK|FINs 6641 * that web servers like to spew after a close) 6642 * 6643 * This must be a little more careful than the above code 6644 * since packet floods will also be caught here. We don't 6645 * update the TTL here to mitigate the damage of a packet 6646 * flood and so the same code can handle awkward establishment 6647 * and a loosened connection close. 6648 * In the establishment case, a correct peer response will 6649 * validate the connection, go through the normal state code 6650 * and keep updating the state TTL. 6651 */ 6652 6653 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6654 printf("pf: loose state match: "); 6655 pf_print_state(*state); 6656 pf_print_flags(tcp_get_flags(th)); 6657 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6658 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6659 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 6660 (unsigned long long)(*state)->packets[1], 6661 pd->dir == PF_IN ? "in" : "out", 6662 pd->dir == (*state)->direction ? "fwd" : "rev"); 6663 } 6664 6665 if (dst->scrub || src->scrub) { 6666 if (pf_normalize_tcp_stateful(pd, reason, th, 6667 *state, src, dst, copyback)) 6668 return (PF_DROP); 6669 } 6670 6671 /* update max window */ 6672 if (src->max_win < win) 6673 src->max_win = win; 6674 /* synchronize sequencing */ 6675 if (SEQ_GT(end, src->seqlo)) 6676 src->seqlo = end; 6677 /* slide the window of what the other end can send */ 6678 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6679 dst->seqhi = ack + MAX((win << sws), 1); 6680 6681 /* 6682 * Cannot set dst->seqhi here since this could be a shotgunned 6683 * SYN and not an already established connection. 6684 */ 6685 6686 if (tcp_get_flags(th) & TH_FIN) 6687 if (src->state < TCPS_CLOSING) 6688 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6689 if (tcp_get_flags(th) & TH_RST) 6690 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6691 6692 /* Fall through to PASS packet */ 6693 6694 } else { 6695 if ((*state)->dst.state == TCPS_SYN_SENT && 6696 (*state)->src.state == TCPS_SYN_SENT) { 6697 /* Send RST for state mismatches during handshake */ 6698 if (!(tcp_get_flags(th) & TH_RST)) 6699 pf_send_tcp((*state)->rule, pd->af, 6700 pd->dst, pd->src, th->th_dport, 6701 th->th_sport, ntohl(th->th_ack), 0, 6702 TH_RST, 0, 0, 6703 (*state)->rule->return_ttl, M_SKIP_FIREWALL, 6704 0, 0, (*state)->act.rtableid); 6705 src->seqlo = 0; 6706 src->seqhi = 1; 6707 src->max_win = 1; 6708 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6709 printf("pf: BAD state: "); 6710 pf_print_state(*state); 6711 pf_print_flags(tcp_get_flags(th)); 6712 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6713 "pkts=%llu:%llu dir=%s,%s\n", 6714 seq, orig_seq, ack, pd->p_len, ackskew, 6715 (unsigned long long)(*state)->packets[0], 6716 (unsigned long long)(*state)->packets[1], 6717 pd->dir == PF_IN ? "in" : "out", 6718 pd->dir == (*state)->direction ? "fwd" : "rev"); 6719 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6720 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6721 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6722 ' ': '2', 6723 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6724 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6725 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6726 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6727 } 6728 REASON_SET(reason, PFRES_BADSTATE); 6729 return (PF_DROP); 6730 } 6731 6732 return (PF_PASS); 6733 } 6734 6735 static int 6736 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6737 { 6738 struct tcphdr *th = &pd->hdr.tcp; 6739 struct pf_state_peer *src, *dst; 6740 u_int8_t psrc, pdst; 6741 6742 if (pd->dir == (*state)->direction) { 6743 src = &(*state)->src; 6744 dst = &(*state)->dst; 6745 psrc = PF_PEER_SRC; 6746 pdst = PF_PEER_DST; 6747 } else { 6748 src = &(*state)->dst; 6749 dst = &(*state)->src; 6750 psrc = PF_PEER_DST; 6751 pdst = PF_PEER_SRC; 6752 } 6753 6754 if (tcp_get_flags(th) & TH_SYN) 6755 if (src->state < TCPS_SYN_SENT) 6756 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6757 if (tcp_get_flags(th) & TH_FIN) 6758 if (src->state < TCPS_CLOSING) 6759 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6760 if (tcp_get_flags(th) & TH_ACK) { 6761 if (dst->state == TCPS_SYN_SENT) { 6762 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6763 if (src->state == TCPS_ESTABLISHED && 6764 (*state)->sns[PF_SN_LIMIT] != NULL && 6765 pf_src_connlimit(*state)) { 6766 REASON_SET(reason, PFRES_SRCLIMIT); 6767 return (PF_DROP); 6768 } 6769 } else if (dst->state == TCPS_CLOSING) { 6770 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6771 } else if (src->state == TCPS_SYN_SENT && 6772 dst->state < TCPS_SYN_SENT) { 6773 /* 6774 * Handle a special sloppy case where we only see one 6775 * half of the connection. If there is a ACK after 6776 * the initial SYN without ever seeing a packet from 6777 * the destination, set the connection to established. 6778 */ 6779 pf_set_protostate(*state, PF_PEER_BOTH, 6780 TCPS_ESTABLISHED); 6781 dst->state = src->state = TCPS_ESTABLISHED; 6782 if ((*state)->sns[PF_SN_LIMIT] != NULL && 6783 pf_src_connlimit(*state)) { 6784 REASON_SET(reason, PFRES_SRCLIMIT); 6785 return (PF_DROP); 6786 } 6787 } else if (src->state == TCPS_CLOSING && 6788 dst->state == TCPS_ESTABLISHED && 6789 dst->seqlo == 0) { 6790 /* 6791 * Handle the closing of half connections where we 6792 * don't see the full bidirectional FIN/ACK+ACK 6793 * handshake. 6794 */ 6795 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6796 } 6797 } 6798 if (tcp_get_flags(th) & TH_RST) 6799 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6800 6801 /* update expire time */ 6802 (*state)->expire = pf_get_uptime(); 6803 if (src->state >= TCPS_FIN_WAIT_2 && 6804 dst->state >= TCPS_FIN_WAIT_2) 6805 (*state)->timeout = PFTM_TCP_CLOSED; 6806 else if (src->state >= TCPS_CLOSING && 6807 dst->state >= TCPS_CLOSING) 6808 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6809 else if (src->state < TCPS_ESTABLISHED || 6810 dst->state < TCPS_ESTABLISHED) 6811 (*state)->timeout = PFTM_TCP_OPENING; 6812 else if (src->state >= TCPS_CLOSING || 6813 dst->state >= TCPS_CLOSING) 6814 (*state)->timeout = PFTM_TCP_CLOSING; 6815 else 6816 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6817 6818 return (PF_PASS); 6819 } 6820 6821 static int 6822 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6823 { 6824 struct pf_state_key *sk = (*state)->key[pd->didx]; 6825 struct tcphdr *th = &pd->hdr.tcp; 6826 6827 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6828 if (pd->dir != (*state)->direction) { 6829 REASON_SET(reason, PFRES_SYNPROXY); 6830 return (PF_SYNPROXY_DROP); 6831 } 6832 if (tcp_get_flags(th) & TH_SYN) { 6833 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6834 REASON_SET(reason, PFRES_SYNPROXY); 6835 return (PF_DROP); 6836 } 6837 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6838 pd->src, th->th_dport, th->th_sport, 6839 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6840 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 6841 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6842 REASON_SET(reason, PFRES_SYNPROXY); 6843 return (PF_SYNPROXY_DROP); 6844 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6845 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6846 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6847 REASON_SET(reason, PFRES_SYNPROXY); 6848 return (PF_DROP); 6849 } else if ((*state)->sns[PF_SN_LIMIT] != NULL && 6850 pf_src_connlimit(*state)) { 6851 REASON_SET(reason, PFRES_SRCLIMIT); 6852 return (PF_DROP); 6853 } else 6854 pf_set_protostate(*state, PF_PEER_SRC, 6855 PF_TCPS_PROXY_DST); 6856 } 6857 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6858 if (pd->dir == (*state)->direction) { 6859 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 6860 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6861 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6862 REASON_SET(reason, PFRES_SYNPROXY); 6863 return (PF_DROP); 6864 } 6865 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6866 if ((*state)->dst.seqhi == 1) 6867 (*state)->dst.seqhi = htonl(arc4random()); 6868 pf_send_tcp((*state)->rule, pd->af, 6869 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6870 sk->port[pd->sidx], sk->port[pd->didx], 6871 (*state)->dst.seqhi, 0, TH_SYN, 0, 6872 (*state)->src.mss, 0, 6873 (*state)->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6874 (*state)->tag, 0, (*state)->act.rtableid); 6875 REASON_SET(reason, PFRES_SYNPROXY); 6876 return (PF_SYNPROXY_DROP); 6877 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 6878 (TH_SYN|TH_ACK)) || 6879 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6880 REASON_SET(reason, PFRES_SYNPROXY); 6881 return (PF_DROP); 6882 } else { 6883 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6884 (*state)->dst.seqlo = ntohl(th->th_seq); 6885 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6886 pd->src, th->th_dport, th->th_sport, 6887 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6888 TH_ACK, (*state)->src.max_win, 0, 0, 0, 6889 (*state)->tag, 0, (*state)->act.rtableid); 6890 pf_send_tcp((*state)->rule, pd->af, 6891 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6892 sk->port[pd->sidx], sk->port[pd->didx], 6893 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6894 TH_ACK, (*state)->dst.max_win, 0, 0, 6895 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6896 (*state)->src.seqdiff = (*state)->dst.seqhi - 6897 (*state)->src.seqlo; 6898 (*state)->dst.seqdiff = (*state)->src.seqhi - 6899 (*state)->dst.seqlo; 6900 (*state)->src.seqhi = (*state)->src.seqlo + 6901 (*state)->dst.max_win; 6902 (*state)->dst.seqhi = (*state)->dst.seqlo + 6903 (*state)->src.max_win; 6904 (*state)->src.wscale = (*state)->dst.wscale = 0; 6905 pf_set_protostate(*state, PF_PEER_BOTH, 6906 TCPS_ESTABLISHED); 6907 REASON_SET(reason, PFRES_SYNPROXY); 6908 return (PF_SYNPROXY_DROP); 6909 } 6910 } 6911 6912 return (PF_PASS); 6913 } 6914 6915 static int 6916 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6917 { 6918 struct pf_state_key_cmp key; 6919 int copyback = 0; 6920 struct pf_state_peer *src, *dst; 6921 uint8_t psrc, pdst; 6922 int action = PF_PASS; 6923 6924 bzero(&key, sizeof(key)); 6925 key.af = pd->af; 6926 key.proto = pd->virtual_proto; 6927 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 6928 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 6929 key.port[pd->sidx] = pd->osport; 6930 key.port[pd->didx] = pd->odport; 6931 6932 STATE_LOOKUP(&key, *state, pd); 6933 6934 if (pd->dir == (*state)->direction) { 6935 src = &(*state)->src; 6936 dst = &(*state)->dst; 6937 psrc = PF_PEER_SRC; 6938 pdst = PF_PEER_DST; 6939 } else { 6940 src = &(*state)->dst; 6941 dst = &(*state)->src; 6942 psrc = PF_PEER_DST; 6943 pdst = PF_PEER_SRC; 6944 } 6945 6946 switch (pd->virtual_proto) { 6947 case IPPROTO_TCP: { 6948 struct tcphdr *th = &pd->hdr.tcp; 6949 6950 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6951 return (action); 6952 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 6953 (*state)->dst.state >= TCPS_FIN_WAIT_2 && 6954 (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN) || 6955 ((tcp_get_flags(th) & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6956 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6957 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6958 printf("pf: state reuse "); 6959 pf_print_state(*state); 6960 pf_print_flags(tcp_get_flags(th)); 6961 printf("\n"); 6962 } 6963 /* XXX make sure it's the same direction ?? */ 6964 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6965 pf_unlink_state(*state); 6966 *state = NULL; 6967 return (PF_DROP); 6968 } 6969 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6970 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6971 return (PF_DROP); 6972 } else { 6973 int ret; 6974 6975 ret = pf_tcp_track_full(state, pd, reason, 6976 ©back); 6977 if (ret == PF_DROP) 6978 return (PF_DROP); 6979 } 6980 break; 6981 } 6982 case IPPROTO_UDP: 6983 /* update states */ 6984 if (src->state < PFUDPS_SINGLE) 6985 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6986 if (dst->state == PFUDPS_SINGLE) 6987 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6988 6989 /* update expire time */ 6990 (*state)->expire = pf_get_uptime(); 6991 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6992 (*state)->timeout = PFTM_UDP_MULTIPLE; 6993 else 6994 (*state)->timeout = PFTM_UDP_SINGLE; 6995 break; 6996 case IPPROTO_SCTP: 6997 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6998 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6999 pd->sctp_flags & PFDESC_SCTP_INIT) { 7000 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7001 pf_unlink_state(*state); 7002 *state = NULL; 7003 return (PF_DROP); 7004 } 7005 7006 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7007 return (PF_DROP); 7008 7009 /* Track state. */ 7010 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7011 if (src->state < SCTP_COOKIE_WAIT) { 7012 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7013 (*state)->timeout = PFTM_SCTP_OPENING; 7014 } 7015 } 7016 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7017 MPASS(dst->scrub != NULL); 7018 if (dst->scrub->pfss_v_tag == 0) 7019 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7020 } 7021 7022 /* 7023 * Bind to the correct interface if we're if-bound. For multihomed 7024 * extra associations we don't know which interface that will be until 7025 * here, so we've inserted the state on V_pf_all. Fix that now. 7026 */ 7027 if ((*state)->kif == V_pfi_all && 7028 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7029 (*state)->kif = pd->kif; 7030 7031 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7032 if (src->state < SCTP_ESTABLISHED) { 7033 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7034 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7035 } 7036 } 7037 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7038 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7039 if (src->state < SCTP_SHUTDOWN_PENDING) { 7040 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7041 (*state)->timeout = PFTM_SCTP_CLOSING; 7042 } 7043 } 7044 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7045 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7046 (*state)->timeout = PFTM_SCTP_CLOSED; 7047 } 7048 7049 (*state)->expire = pf_get_uptime(); 7050 break; 7051 default: 7052 /* update states */ 7053 if (src->state < PFOTHERS_SINGLE) 7054 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7055 if (dst->state == PFOTHERS_SINGLE) 7056 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7057 7058 /* update expire time */ 7059 (*state)->expire = pf_get_uptime(); 7060 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7061 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7062 else 7063 (*state)->timeout = PFTM_OTHER_SINGLE; 7064 break; 7065 } 7066 7067 /* translate source/destination address, if necessary */ 7068 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7069 struct pf_state_key *nk; 7070 int afto, sidx, didx; 7071 7072 if (PF_REVERSED_KEY((*state)->key, pd->af)) 7073 nk = (*state)->key[pd->sidx]; 7074 else 7075 nk = (*state)->key[pd->didx]; 7076 7077 afto = pd->af != nk->af; 7078 sidx = afto ? pd->didx : pd->sidx; 7079 didx = afto ? pd->sidx : pd->didx; 7080 7081 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7082 nk->port[sidx] != pd->osport) 7083 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, 7084 pd->pcksum, &nk->addr[pd->sidx], 7085 nk->port[sidx], pd->virtual_proto == IPPROTO_UDP, 7086 pd->af, nk->af); 7087 7088 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7089 nk->port[didx] != pd->odport) 7090 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, 7091 pd->pcksum, &nk->addr[pd->didx], 7092 nk->port[didx], pd->virtual_proto == IPPROTO_UDP, 7093 pd->af, nk->af); 7094 7095 if (afto) { 7096 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7097 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7098 pd->naf = nk->af; 7099 action = PF_AFRT; 7100 } 7101 7102 copyback = 1; 7103 } 7104 7105 if (copyback && pd->hdrlen > 0) 7106 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 7107 7108 return (action); 7109 } 7110 7111 static int 7112 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7113 u_short *reason) 7114 { 7115 struct pf_state_peer *src; 7116 if (pd->dir == state->direction) { 7117 if (PF_REVERSED_KEY(state->key, pd->af)) 7118 src = &state->dst; 7119 else 7120 src = &state->src; 7121 } else { 7122 if (PF_REVERSED_KEY(state->key, pd->af)) 7123 src = &state->src; 7124 else 7125 src = &state->dst; 7126 } 7127 7128 if (src->scrub != NULL) { 7129 if (src->scrub->pfss_v_tag == 0) 7130 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7131 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7132 return (PF_DROP); 7133 } 7134 7135 return (PF_PASS); 7136 } 7137 7138 static void 7139 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7140 { 7141 struct pf_sctp_endpoint key; 7142 struct pf_sctp_endpoint *ep; 7143 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7144 struct pf_sctp_source *i, *tmp; 7145 7146 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7147 return; 7148 7149 PF_SCTP_ENDPOINTS_LOCK(); 7150 7151 key.v_tag = s->dst.scrub->pfss_v_tag; 7152 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7153 if (ep != NULL) { 7154 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7155 if (pf_addr_cmp(&i->addr, 7156 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7157 s->key[PF_SK_WIRE]->af) == 0) { 7158 SDT_PROBE3(pf, sctp, multihome, remove, 7159 key.v_tag, s, i); 7160 TAILQ_REMOVE(&ep->sources, i, entry); 7161 free(i, M_PFTEMP); 7162 break; 7163 } 7164 } 7165 7166 if (TAILQ_EMPTY(&ep->sources)) { 7167 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7168 free(ep, M_PFTEMP); 7169 } 7170 } 7171 7172 /* Other direction. */ 7173 key.v_tag = s->src.scrub->pfss_v_tag; 7174 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7175 if (ep != NULL) { 7176 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7177 if (pf_addr_cmp(&i->addr, 7178 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7179 s->key[PF_SK_WIRE]->af) == 0) { 7180 SDT_PROBE3(pf, sctp, multihome, remove, 7181 key.v_tag, s, i); 7182 TAILQ_REMOVE(&ep->sources, i, entry); 7183 free(i, M_PFTEMP); 7184 break; 7185 } 7186 } 7187 7188 if (TAILQ_EMPTY(&ep->sources)) { 7189 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7190 free(ep, M_PFTEMP); 7191 } 7192 } 7193 7194 PF_SCTP_ENDPOINTS_UNLOCK(); 7195 } 7196 7197 static void 7198 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7199 { 7200 struct pf_sctp_endpoint key = { 7201 .v_tag = v_tag, 7202 }; 7203 struct pf_sctp_source *i; 7204 struct pf_sctp_endpoint *ep; 7205 7206 PF_SCTP_ENDPOINTS_LOCK(); 7207 7208 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7209 if (ep == NULL) { 7210 ep = malloc(sizeof(struct pf_sctp_endpoint), 7211 M_PFTEMP, M_NOWAIT); 7212 if (ep == NULL) { 7213 PF_SCTP_ENDPOINTS_UNLOCK(); 7214 return; 7215 } 7216 7217 ep->v_tag = v_tag; 7218 TAILQ_INIT(&ep->sources); 7219 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7220 } 7221 7222 /* Avoid inserting duplicates. */ 7223 TAILQ_FOREACH(i, &ep->sources, entry) { 7224 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7225 PF_SCTP_ENDPOINTS_UNLOCK(); 7226 return; 7227 } 7228 } 7229 7230 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7231 if (i == NULL) { 7232 PF_SCTP_ENDPOINTS_UNLOCK(); 7233 return; 7234 } 7235 7236 i->af = pd->af; 7237 memcpy(&i->addr, a, sizeof(*a)); 7238 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7239 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7240 7241 PF_SCTP_ENDPOINTS_UNLOCK(); 7242 } 7243 7244 static void 7245 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7246 struct pf_kstate *s, int action) 7247 { 7248 struct pf_sctp_multihome_job *j, *tmp; 7249 struct pf_sctp_source *i; 7250 int ret __unused; 7251 struct pf_kstate *sm = NULL; 7252 struct pf_krule *ra = NULL; 7253 struct pf_krule *r = &V_pf_default_rule; 7254 struct pf_kruleset *rs = NULL; 7255 bool do_extra = true; 7256 7257 PF_RULES_RLOCK_TRACKER; 7258 7259 again: 7260 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7261 if (s == NULL || action != PF_PASS) 7262 goto free; 7263 7264 /* Confirm we don't recurse here. */ 7265 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7266 7267 switch (j->op) { 7268 case SCTP_ADD_IP_ADDRESS: { 7269 uint32_t v_tag = pd->sctp_initiate_tag; 7270 7271 if (v_tag == 0) { 7272 if (s->direction == pd->dir) 7273 v_tag = s->src.scrub->pfss_v_tag; 7274 else 7275 v_tag = s->dst.scrub->pfss_v_tag; 7276 } 7277 7278 /* 7279 * Avoid duplicating states. We'll already have 7280 * created a state based on the source address of 7281 * the packet, but SCTP endpoints may also list this 7282 * address again in the INIT(_ACK) parameters. 7283 */ 7284 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7285 break; 7286 } 7287 7288 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7289 PF_RULES_RLOCK(); 7290 sm = NULL; 7291 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7292 j->pd.related_rule = s->rule; 7293 } 7294 ret = pf_test_rule(&r, &sm, 7295 &j->pd, &ra, &rs, NULL); 7296 PF_RULES_RUNLOCK(); 7297 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7298 if (ret != PF_DROP && sm != NULL) { 7299 /* Inherit v_tag values. */ 7300 if (sm->direction == s->direction) { 7301 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7302 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7303 } else { 7304 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7305 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7306 } 7307 PF_STATE_UNLOCK(sm); 7308 } else { 7309 /* If we try duplicate inserts? */ 7310 break; 7311 } 7312 7313 /* Only add the address if we've actually allowed the state. */ 7314 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7315 7316 if (! do_extra) { 7317 break; 7318 } 7319 /* 7320 * We need to do this for each of our source addresses. 7321 * Find those based on the verification tag. 7322 */ 7323 struct pf_sctp_endpoint key = { 7324 .v_tag = pd->hdr.sctp.v_tag, 7325 }; 7326 struct pf_sctp_endpoint *ep; 7327 7328 PF_SCTP_ENDPOINTS_LOCK(); 7329 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7330 if (ep == NULL) { 7331 PF_SCTP_ENDPOINTS_UNLOCK(); 7332 break; 7333 } 7334 MPASS(ep != NULL); 7335 7336 TAILQ_FOREACH(i, &ep->sources, entry) { 7337 struct pf_sctp_multihome_job *nj; 7338 7339 /* SCTP can intermingle IPv4 and IPv6. */ 7340 if (i->af != pd->af) 7341 continue; 7342 7343 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7344 if (! nj) { 7345 continue; 7346 } 7347 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7348 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7349 nj->pd.src = &nj->src; 7350 // New destination address! 7351 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7352 nj->pd.dst = &nj->dst; 7353 nj->pd.m = j->pd.m; 7354 nj->op = j->op; 7355 7356 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7357 } 7358 PF_SCTP_ENDPOINTS_UNLOCK(); 7359 7360 break; 7361 } 7362 case SCTP_DEL_IP_ADDRESS: { 7363 struct pf_state_key_cmp key; 7364 uint8_t psrc; 7365 7366 bzero(&key, sizeof(key)); 7367 key.af = j->pd.af; 7368 key.proto = IPPROTO_SCTP; 7369 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7370 PF_ACPY(&key.addr[0], j->pd.src, key.af); 7371 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 7372 key.port[0] = j->pd.hdr.sctp.src_port; 7373 key.port[1] = j->pd.hdr.sctp.dest_port; 7374 } else { /* stack side, reverse */ 7375 PF_ACPY(&key.addr[1], j->pd.src, key.af); 7376 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 7377 key.port[1] = j->pd.hdr.sctp.src_port; 7378 key.port[0] = j->pd.hdr.sctp.dest_port; 7379 } 7380 7381 sm = pf_find_state(kif, &key, j->pd.dir); 7382 if (sm != NULL) { 7383 PF_STATE_LOCK_ASSERT(sm); 7384 if (j->pd.dir == sm->direction) { 7385 psrc = PF_PEER_SRC; 7386 } else { 7387 psrc = PF_PEER_DST; 7388 } 7389 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7390 sm->timeout = PFTM_SCTP_CLOSING; 7391 PF_STATE_UNLOCK(sm); 7392 } 7393 break; 7394 default: 7395 panic("Unknown op %#x", j->op); 7396 } 7397 } 7398 7399 free: 7400 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7401 free(j, M_PFTEMP); 7402 } 7403 7404 /* We may have inserted extra work while processing the list. */ 7405 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7406 do_extra = false; 7407 goto again; 7408 } 7409 } 7410 7411 static int 7412 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7413 { 7414 int off = 0; 7415 struct pf_sctp_multihome_job *job; 7416 7417 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7418 7419 while (off < len) { 7420 struct sctp_paramhdr h; 7421 7422 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 7423 pd->af)) 7424 return (PF_DROP); 7425 7426 /* Parameters are at least 4 bytes. */ 7427 if (ntohs(h.param_length) < 4) 7428 return (PF_DROP); 7429 7430 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7431 ntohs(h.param_length)); 7432 7433 switch (ntohs(h.param_type)) { 7434 case SCTP_IPV4_ADDRESS: { 7435 struct in_addr t; 7436 7437 if (ntohs(h.param_length) != 7438 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7439 return (PF_DROP); 7440 7441 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7442 NULL, NULL, pd->af)) 7443 return (PF_DROP); 7444 7445 if (in_nullhost(t)) 7446 t.s_addr = pd->src->v4.s_addr; 7447 7448 /* 7449 * We hold the state lock (idhash) here, which means 7450 * that we can't acquire the keyhash, or we'll get a 7451 * LOR (and potentially double-lock things too). We also 7452 * can't release the state lock here, so instead we'll 7453 * enqueue this for async handling. 7454 * There's a relatively small race here, in that a 7455 * packet using the new addresses could arrive already, 7456 * but that's just though luck for it. 7457 */ 7458 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7459 if (! job) 7460 return (PF_DROP); 7461 7462 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7463 7464 memcpy(&job->pd, pd, sizeof(*pd)); 7465 7466 // New source address! 7467 memcpy(&job->src, &t, sizeof(t)); 7468 job->pd.src = &job->src; 7469 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7470 job->pd.dst = &job->dst; 7471 job->pd.m = pd->m; 7472 job->op = op; 7473 7474 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7475 break; 7476 } 7477 #ifdef INET6 7478 case SCTP_IPV6_ADDRESS: { 7479 struct in6_addr t; 7480 7481 if (ntohs(h.param_length) != 7482 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7483 return (PF_DROP); 7484 7485 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7486 NULL, NULL, pd->af)) 7487 return (PF_DROP); 7488 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7489 break; 7490 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7491 memcpy(&t, &pd->src->v6, sizeof(t)); 7492 7493 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7494 if (! job) 7495 return (PF_DROP); 7496 7497 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7498 7499 memcpy(&job->pd, pd, sizeof(*pd)); 7500 memcpy(&job->src, &t, sizeof(t)); 7501 job->pd.src = &job->src; 7502 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7503 job->pd.dst = &job->dst; 7504 job->pd.m = pd->m; 7505 job->op = op; 7506 7507 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7508 break; 7509 } 7510 #endif 7511 case SCTP_ADD_IP_ADDRESS: { 7512 int ret; 7513 struct sctp_asconf_paramhdr ah; 7514 7515 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7516 NULL, NULL, pd->af)) 7517 return (PF_DROP); 7518 7519 ret = pf_multihome_scan(start + off + sizeof(ah), 7520 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7521 SCTP_ADD_IP_ADDRESS); 7522 if (ret != PF_PASS) 7523 return (ret); 7524 break; 7525 } 7526 case SCTP_DEL_IP_ADDRESS: { 7527 int ret; 7528 struct sctp_asconf_paramhdr ah; 7529 7530 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7531 NULL, NULL, pd->af)) 7532 return (PF_DROP); 7533 ret = pf_multihome_scan(start + off + sizeof(ah), 7534 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7535 SCTP_DEL_IP_ADDRESS); 7536 if (ret != PF_PASS) 7537 return (ret); 7538 break; 7539 } 7540 default: 7541 break; 7542 } 7543 7544 off += roundup(ntohs(h.param_length), 4); 7545 } 7546 7547 return (PF_PASS); 7548 } 7549 7550 int 7551 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7552 { 7553 start += sizeof(struct sctp_init_chunk); 7554 len -= sizeof(struct sctp_init_chunk); 7555 7556 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7557 } 7558 7559 int 7560 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7561 { 7562 start += sizeof(struct sctp_asconf_chunk); 7563 len -= sizeof(struct sctp_asconf_chunk); 7564 7565 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7566 } 7567 7568 int 7569 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7570 struct pf_kstate **state, int direction, 7571 u_int16_t icmpid, u_int16_t type, int icmp_dir, 7572 int *iidx, int multi, int inner) 7573 { 7574 key->af = pd->af; 7575 key->proto = pd->proto; 7576 if (icmp_dir == PF_IN) { 7577 *iidx = pd->sidx; 7578 key->port[pd->sidx] = icmpid; 7579 key->port[pd->didx] = type; 7580 } else { 7581 *iidx = pd->didx; 7582 key->port[pd->sidx] = type; 7583 key->port[pd->didx] = icmpid; 7584 } 7585 if (pf_state_key_addr_setup(pd, key, multi)) 7586 return (PF_DROP); 7587 7588 STATE_LOOKUP(key, *state, pd); 7589 7590 if ((*state)->state_flags & PFSTATE_SLOPPY) 7591 return (-1); 7592 7593 /* Is this ICMP message flowing in right direction? */ 7594 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7595 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7596 PF_IN : PF_OUT; 7597 else 7598 direction = (*state)->direction; 7599 if ((*state)->rule->type && 7600 (((!inner && direction == pd->dir) || 7601 (inner && direction != pd->dir)) ? 7602 PF_IN : PF_OUT) != icmp_dir) { 7603 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7604 printf("pf: icmp type %d in wrong direction (%d): ", 7605 ntohs(type), icmp_dir); 7606 pf_print_state(*state); 7607 printf("\n"); 7608 } 7609 PF_STATE_UNLOCK(*state); 7610 *state = NULL; 7611 return (PF_DROP); 7612 } 7613 return (-1); 7614 } 7615 7616 static int 7617 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7618 u_short *reason) 7619 { 7620 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7621 u_int16_t *icmpsum, virtual_id, virtual_type; 7622 u_int8_t icmptype, icmpcode; 7623 int icmp_dir, iidx, ret, multi; 7624 struct pf_state_key_cmp key; 7625 #ifdef INET 7626 u_int16_t icmpid; 7627 #endif 7628 7629 MPASS(*state == NULL); 7630 7631 bzero(&key, sizeof(key)); 7632 switch (pd->proto) { 7633 #ifdef INET 7634 case IPPROTO_ICMP: 7635 icmptype = pd->hdr.icmp.icmp_type; 7636 icmpcode = pd->hdr.icmp.icmp_code; 7637 icmpid = pd->hdr.icmp.icmp_id; 7638 icmpsum = &pd->hdr.icmp.icmp_cksum; 7639 break; 7640 #endif /* INET */ 7641 #ifdef INET6 7642 case IPPROTO_ICMPV6: 7643 icmptype = pd->hdr.icmp6.icmp6_type; 7644 icmpcode = pd->hdr.icmp6.icmp6_code; 7645 #ifdef INET 7646 icmpid = pd->hdr.icmp6.icmp6_id; 7647 #endif 7648 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7649 break; 7650 #endif /* INET6 */ 7651 } 7652 7653 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 7654 &virtual_id, &virtual_type) == 0) { 7655 /* 7656 * ICMP query/reply message not related to a TCP/UDP/SCTP 7657 * packet. Search for an ICMP state. 7658 */ 7659 ret = pf_icmp_state_lookup(&key, pd, state, pd->dir, 7660 virtual_id, virtual_type, icmp_dir, &iidx, 7661 PF_ICMP_MULTI_NONE, 0); 7662 if (ret >= 0) { 7663 MPASS(*state == NULL); 7664 if (ret == PF_DROP && pd->af == AF_INET6 && 7665 icmp_dir == PF_OUT) { 7666 ret = pf_icmp_state_lookup(&key, pd, state, 7667 pd->dir, virtual_id, virtual_type, 7668 icmp_dir, &iidx, multi, 0); 7669 if (ret >= 0) { 7670 MPASS(*state == NULL); 7671 return (ret); 7672 } 7673 } else 7674 return (ret); 7675 } 7676 7677 (*state)->expire = pf_get_uptime(); 7678 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7679 7680 /* translate source/destination address, if necessary */ 7681 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7682 struct pf_state_key *nk; 7683 int afto, sidx, didx; 7684 7685 if (PF_REVERSED_KEY((*state)->key, pd->af)) 7686 nk = (*state)->key[pd->sidx]; 7687 else 7688 nk = (*state)->key[pd->didx]; 7689 7690 afto = pd->af != nk->af; 7691 sidx = afto ? pd->didx : pd->sidx; 7692 didx = afto ? pd->sidx : pd->didx; 7693 iidx = afto ? !iidx : iidx; 7694 7695 switch (pd->af) { 7696 #ifdef INET 7697 case AF_INET: 7698 #ifdef INET6 7699 if (afto) { 7700 if (pf_translate_icmp_af(AF_INET6, 7701 &pd->hdr.icmp)) 7702 return (PF_DROP); 7703 pd->proto = IPPROTO_ICMPV6; 7704 } 7705 #endif 7706 if (!afto && 7707 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 7708 pf_change_a(&saddr->v4.s_addr, 7709 pd->ip_sum, 7710 nk->addr[sidx].v4.s_addr, 7711 0); 7712 7713 if (!afto && PF_ANEQ(pd->dst, 7714 &nk->addr[didx], AF_INET)) 7715 pf_change_a(&daddr->v4.s_addr, 7716 pd->ip_sum, 7717 nk->addr[didx].v4.s_addr, 0); 7718 7719 if (nk->port[iidx] != 7720 pd->hdr.icmp.icmp_id) { 7721 pd->hdr.icmp.icmp_cksum = 7722 pf_cksum_fixup( 7723 pd->hdr.icmp.icmp_cksum, icmpid, 7724 nk->port[iidx], 0); 7725 pd->hdr.icmp.icmp_id = 7726 nk->port[iidx]; 7727 } 7728 7729 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7730 (caddr_t )&pd->hdr.icmp); 7731 break; 7732 #endif /* INET */ 7733 #ifdef INET6 7734 case AF_INET6: 7735 #ifdef INET 7736 if (afto) { 7737 if (pf_translate_icmp_af(AF_INET, 7738 &pd->hdr.icmp6)) 7739 return (PF_DROP); 7740 pd->proto = IPPROTO_ICMP; 7741 } 7742 #endif 7743 if (!afto && 7744 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 7745 pf_change_a6(saddr, 7746 &pd->hdr.icmp6.icmp6_cksum, 7747 &nk->addr[sidx], 0); 7748 7749 if (!afto && PF_ANEQ(pd->dst, 7750 &nk->addr[didx], AF_INET6)) 7751 pf_change_a6(daddr, 7752 &pd->hdr.icmp6.icmp6_cksum, 7753 &nk->addr[didx], 0); 7754 7755 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 7756 pd->hdr.icmp6.icmp6_id = 7757 nk->port[iidx]; 7758 7759 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7760 (caddr_t )&pd->hdr.icmp6); 7761 break; 7762 #endif /* INET6 */ 7763 } 7764 if (afto) { 7765 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7766 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7767 pd->naf = nk->af; 7768 return (PF_AFRT); 7769 } 7770 } 7771 return (PF_PASS); 7772 7773 } else { 7774 /* 7775 * ICMP error message in response to a TCP/UDP packet. 7776 * Extract the inner TCP/UDP header and search for that state. 7777 */ 7778 7779 struct pf_pdesc pd2; 7780 bzero(&pd2, sizeof pd2); 7781 #ifdef INET 7782 struct ip h2; 7783 #endif /* INET */ 7784 #ifdef INET6 7785 struct ip6_hdr h2_6; 7786 #endif /* INET6 */ 7787 int ipoff2 = 0; 7788 7789 pd2.af = pd->af; 7790 pd2.dir = pd->dir; 7791 /* Payload packet is from the opposite direction. */ 7792 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7793 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7794 pd2.m = pd->m; 7795 pd2.kif = pd->kif; 7796 switch (pd->af) { 7797 #ifdef INET 7798 case AF_INET: 7799 /* offset of h2 in mbuf chain */ 7800 ipoff2 = pd->off + ICMP_MINLEN; 7801 7802 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7803 NULL, reason, pd2.af)) { 7804 DPFPRINTF(PF_DEBUG_MISC, 7805 ("pf: ICMP error message too short " 7806 "(ip)\n")); 7807 return (PF_DROP); 7808 } 7809 /* 7810 * ICMP error messages don't refer to non-first 7811 * fragments 7812 */ 7813 if (h2.ip_off & htons(IP_OFFMASK)) { 7814 REASON_SET(reason, PFRES_FRAG); 7815 return (PF_DROP); 7816 } 7817 7818 /* offset of protocol header that follows h2 */ 7819 pd2.off = ipoff2 + (h2.ip_hl << 2); 7820 7821 pd2.proto = h2.ip_p; 7822 pd2.tot_len = ntohs(h2.ip_len); 7823 pd2.src = (struct pf_addr *)&h2.ip_src; 7824 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7825 pd2.ip_sum = &h2.ip_sum; 7826 break; 7827 #endif /* INET */ 7828 #ifdef INET6 7829 case AF_INET6: 7830 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7831 7832 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7833 NULL, reason, pd2.af)) { 7834 DPFPRINTF(PF_DEBUG_MISC, 7835 ("pf: ICMP error message too short " 7836 "(ip6)\n")); 7837 return (PF_DROP); 7838 } 7839 pd2.off = ipoff2; 7840 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 7841 return (PF_DROP); 7842 7843 pd2.tot_len = ntohs(h2_6.ip6_plen) + 7844 sizeof(struct ip6_hdr); 7845 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7846 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7847 pd2.ip_sum = NULL; 7848 break; 7849 #endif /* INET6 */ 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; 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 7890 key.af = pd2.af; 7891 key.proto = IPPROTO_TCP; 7892 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7893 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7894 key.port[pd2.sidx] = th.th_sport; 7895 key.port[pd2.didx] = th.th_dport; 7896 7897 STATE_LOOKUP(&key, *state, pd); 7898 7899 if (pd->dir == (*state)->direction) { 7900 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 7901 src = &(*state)->src; 7902 dst = &(*state)->dst; 7903 } else { 7904 src = &(*state)->dst; 7905 dst = &(*state)->src; 7906 } 7907 } else { 7908 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 7909 src = &(*state)->dst; 7910 dst = &(*state)->src; 7911 } else { 7912 src = &(*state)->src; 7913 dst = &(*state)->dst; 7914 } 7915 } 7916 7917 if (src->wscale && dst->wscale) 7918 dws = dst->wscale & PF_WSCALE_MASK; 7919 else 7920 dws = 0; 7921 7922 /* Demodulate sequence number */ 7923 seq = ntohl(th.th_seq) - src->seqdiff; 7924 if (src->seqdiff) { 7925 pf_change_a(&th.th_seq, icmpsum, 7926 htonl(seq), 0); 7927 copyback = 1; 7928 } 7929 7930 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7931 (!SEQ_GEQ(src->seqhi, seq) || 7932 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7933 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7934 printf("pf: BAD ICMP %d:%d ", 7935 icmptype, icmpcode); 7936 pf_print_host(pd->src, 0, pd->af); 7937 printf(" -> "); 7938 pf_print_host(pd->dst, 0, pd->af); 7939 printf(" state: "); 7940 pf_print_state(*state); 7941 printf(" seq=%u\n", seq); 7942 } 7943 REASON_SET(reason, PFRES_BADSTATE); 7944 return (PF_DROP); 7945 } else { 7946 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7947 printf("pf: OK ICMP %d:%d ", 7948 icmptype, icmpcode); 7949 pf_print_host(pd->src, 0, pd->af); 7950 printf(" -> "); 7951 pf_print_host(pd->dst, 0, pd->af); 7952 printf(" state: "); 7953 pf_print_state(*state); 7954 printf(" seq=%u\n", seq); 7955 } 7956 } 7957 7958 /* translate source/destination address, if necessary */ 7959 if ((*state)->key[PF_SK_WIRE] != 7960 (*state)->key[PF_SK_STACK]) { 7961 7962 struct pf_state_key *nk; 7963 7964 if (PF_REVERSED_KEY((*state)->key, pd->af)) 7965 nk = (*state)->key[pd->sidx]; 7966 else 7967 nk = (*state)->key[pd->didx]; 7968 7969 #if defined(INET) && defined(INET6) 7970 int afto, sidx, didx; 7971 7972 afto = pd->af != nk->af; 7973 sidx = afto ? pd2.didx : pd2.sidx; 7974 didx = afto ? pd2.sidx : pd2.didx; 7975 7976 if (afto) { 7977 if (pf_translate_icmp_af(nk->af, 7978 &pd->hdr.icmp)) 7979 return (PF_DROP); 7980 m_copyback(pd->m, pd->off, 7981 sizeof(struct icmp6_hdr), 7982 (c_caddr_t)&pd->hdr.icmp6); 7983 if (pf_change_icmp_af(pd->m, ipoff2, pd, 7984 &pd2, &nk->addr[sidx], 7985 &nk->addr[didx], pd->af, 7986 nk->af)) 7987 return (PF_DROP); 7988 if (nk->af == AF_INET) 7989 pd->proto = IPPROTO_ICMP; 7990 else 7991 pd->proto = IPPROTO_ICMPV6; 7992 pf_change_ap(pd->m, pd2.src, &th.th_sport, 7993 pd->ip_sum, &th.th_sum, &nk->addr[pd2.sidx], 7994 nk->port[sidx], 1, pd->af, nk->af); 7995 pf_change_ap(pd->m, pd2.dst, &th.th_dport, 7996 pd->ip_sum, &th.th_sum, &nk->addr[pd2.didx], 7997 nk->port[didx], 1, pd->af, nk->af); 7998 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)&th); 7999 PF_ACPY(pd->src, 8000 &nk->addr[pd2.sidx], nk->af); 8001 PF_ACPY(pd->dst, 8002 &nk->addr[pd2.didx], nk->af); 8003 pd->naf = nk->af; 8004 return (PF_AFRT); 8005 } 8006 #endif 8007 8008 if (PF_ANEQ(pd2.src, 8009 &nk->addr[pd2.sidx], pd2.af) || 8010 nk->port[pd2.sidx] != th.th_sport) 8011 pf_change_icmp(pd2.src, &th.th_sport, 8012 daddr, &nk->addr[pd2.sidx], 8013 nk->port[pd2.sidx], NULL, 8014 pd2.ip_sum, icmpsum, 8015 pd->ip_sum, 0, pd2.af); 8016 8017 if (PF_ANEQ(pd2.dst, 8018 &nk->addr[pd2.didx], pd2.af) || 8019 nk->port[pd2.didx] != th.th_dport) 8020 pf_change_icmp(pd2.dst, &th.th_dport, 8021 saddr, &nk->addr[pd2.didx], 8022 nk->port[pd2.didx], NULL, 8023 pd2.ip_sum, icmpsum, 8024 pd->ip_sum, 0, pd2.af); 8025 copyback = 1; 8026 } 8027 8028 if (copyback) { 8029 switch (pd2.af) { 8030 #ifdef INET 8031 case AF_INET: 8032 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8033 (caddr_t )&pd->hdr.icmp); 8034 m_copyback(pd->m, ipoff2, sizeof(h2), 8035 (caddr_t )&h2); 8036 break; 8037 #endif /* INET */ 8038 #ifdef INET6 8039 case AF_INET6: 8040 m_copyback(pd->m, pd->off, 8041 sizeof(struct icmp6_hdr), 8042 (caddr_t )&pd->hdr.icmp6); 8043 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8044 (caddr_t )&h2_6); 8045 break; 8046 #endif /* INET6 */ 8047 } 8048 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 8049 } 8050 8051 return (PF_PASS); 8052 break; 8053 } 8054 case IPPROTO_UDP: { 8055 struct udphdr uh; 8056 8057 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 8058 NULL, reason, pd2.af)) { 8059 DPFPRINTF(PF_DEBUG_MISC, 8060 ("pf: ICMP error message too short " 8061 "(udp)\n")); 8062 return (PF_DROP); 8063 } 8064 8065 key.af = pd2.af; 8066 key.proto = IPPROTO_UDP; 8067 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8068 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8069 key.port[pd2.sidx] = uh.uh_sport; 8070 key.port[pd2.didx] = uh.uh_dport; 8071 8072 STATE_LOOKUP(&key, *state, pd); 8073 8074 /* translate source/destination address, if necessary */ 8075 if ((*state)->key[PF_SK_WIRE] != 8076 (*state)->key[PF_SK_STACK]) { 8077 struct pf_state_key *nk; 8078 8079 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8080 nk = (*state)->key[pd->sidx]; 8081 else 8082 nk = (*state)->key[pd->didx]; 8083 8084 #if defined(INET) && defined(INET6) 8085 int afto, sidx, didx; 8086 8087 afto = pd->af != nk->af; 8088 sidx = afto ? pd2.didx : pd2.sidx; 8089 didx = afto ? pd2.sidx : pd2.didx; 8090 8091 if (afto) { 8092 if (pf_translate_icmp_af(nk->af, 8093 &pd->hdr.icmp)) 8094 return (PF_DROP); 8095 m_copyback(pd->m, pd->off, 8096 sizeof(struct icmp6_hdr), 8097 (c_caddr_t)&pd->hdr.icmp6); 8098 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8099 &pd2, &nk->addr[sidx], 8100 &nk->addr[didx], pd->af, 8101 nk->af)) 8102 return (PF_DROP); 8103 if (nk->af == AF_INET) 8104 pd->proto = IPPROTO_ICMP; 8105 else 8106 pd->proto = IPPROTO_ICMPV6; 8107 pf_change_ap(pd->m, pd2.src, &uh.uh_sport, 8108 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.sidx], 8109 nk->port[sidx], 1, pd->af, nk->af); 8110 pf_change_ap(pd->m, pd2.dst, &uh.uh_dport, 8111 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.didx], 8112 nk->port[didx], 1, pd->af, nk->af); 8113 m_copyback(pd2.m, pd2.off, sizeof(uh), 8114 (c_caddr_t)&uh); 8115 PF_ACPY(&pd->nsaddr, 8116 &nk->addr[pd2.sidx], nk->af); 8117 PF_ACPY(&pd->ndaddr, 8118 &nk->addr[pd2.didx], nk->af); 8119 pd->naf = nk->af; 8120 return (PF_AFRT); 8121 } 8122 #endif 8123 8124 if (PF_ANEQ(pd2.src, 8125 &nk->addr[pd2.sidx], pd2.af) || 8126 nk->port[pd2.sidx] != uh.uh_sport) 8127 pf_change_icmp(pd2.src, &uh.uh_sport, 8128 daddr, &nk->addr[pd2.sidx], 8129 nk->port[pd2.sidx], &uh.uh_sum, 8130 pd2.ip_sum, icmpsum, 8131 pd->ip_sum, 1, pd2.af); 8132 8133 if (PF_ANEQ(pd2.dst, 8134 &nk->addr[pd2.didx], pd2.af) || 8135 nk->port[pd2.didx] != uh.uh_dport) 8136 pf_change_icmp(pd2.dst, &uh.uh_dport, 8137 saddr, &nk->addr[pd2.didx], 8138 nk->port[pd2.didx], &uh.uh_sum, 8139 pd2.ip_sum, icmpsum, 8140 pd->ip_sum, 1, pd2.af); 8141 8142 switch (pd2.af) { 8143 #ifdef INET 8144 case AF_INET: 8145 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8146 (caddr_t )&pd->hdr.icmp); 8147 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8148 break; 8149 #endif /* INET */ 8150 #ifdef INET6 8151 case AF_INET6: 8152 m_copyback(pd->m, pd->off, 8153 sizeof(struct icmp6_hdr), 8154 (caddr_t )&pd->hdr.icmp6); 8155 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8156 (caddr_t )&h2_6); 8157 break; 8158 #endif /* INET6 */ 8159 } 8160 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 8161 } 8162 return (PF_PASS); 8163 break; 8164 } 8165 #ifdef INET 8166 case IPPROTO_SCTP: { 8167 struct sctphdr sh; 8168 struct pf_state_peer *src; 8169 int copyback = 0; 8170 8171 if (! pf_pull_hdr(pd->m, pd2.off, &sh, sizeof(sh), NULL, reason, 8172 pd2.af)) { 8173 DPFPRINTF(PF_DEBUG_MISC, 8174 ("pf: ICMP error message too short " 8175 "(sctp)\n")); 8176 return (PF_DROP); 8177 } 8178 8179 key.af = pd2.af; 8180 key.proto = IPPROTO_SCTP; 8181 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8182 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8183 key.port[pd2.sidx] = sh.src_port; 8184 key.port[pd2.didx] = sh.dest_port; 8185 8186 STATE_LOOKUP(&key, *state, pd); 8187 8188 if (pd->dir == (*state)->direction) { 8189 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8190 src = &(*state)->src; 8191 else 8192 src = &(*state)->dst; 8193 } else { 8194 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8195 src = &(*state)->dst; 8196 else 8197 src = &(*state)->src; 8198 } 8199 8200 if (src->scrub->pfss_v_tag != sh.v_tag) { 8201 DPFPRINTF(PF_DEBUG_MISC, 8202 ("pf: ICMP error message has incorrect " 8203 "SCTP v_tag\n")); 8204 return (PF_DROP); 8205 } 8206 8207 /* translate source/destination address, if necessary */ 8208 if ((*state)->key[PF_SK_WIRE] != 8209 (*state)->key[PF_SK_STACK]) { 8210 8211 struct pf_state_key *nk; 8212 8213 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8214 nk = (*state)->key[pd->sidx]; 8215 else 8216 nk = (*state)->key[pd->didx]; 8217 8218 #if defined(INET) && defined(INET6) 8219 int afto, sidx, didx; 8220 8221 afto = pd->af != nk->af; 8222 sidx = afto ? pd2.didx : pd2.sidx; 8223 didx = afto ? pd2.sidx : pd2.didx; 8224 8225 if (afto) { 8226 if (pf_translate_icmp_af(nk->af, 8227 &pd->hdr.icmp)) 8228 return (PF_DROP); 8229 m_copyback(pd->m, pd->off, 8230 sizeof(struct icmp6_hdr), 8231 (c_caddr_t)&pd->hdr.icmp6); 8232 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8233 &pd2, &nk->addr[sidx], 8234 &nk->addr[didx], pd->af, 8235 nk->af)) 8236 return (PF_DROP); 8237 if (nk->af == AF_INET) 8238 pd->proto = IPPROTO_ICMP; 8239 else 8240 pd->proto = IPPROTO_ICMPV6; 8241 sh.src_port = nk->port[sidx]; 8242 sh.dest_port = nk->port[didx]; 8243 m_copyback(pd2.m, pd2.off, sizeof(sh), (c_caddr_t)&sh); 8244 PF_ACPY(pd->src, 8245 &nk->addr[pd2.sidx], nk->af); 8246 PF_ACPY(pd->dst, 8247 &nk->addr[pd2.didx], nk->af); 8248 pd->naf = nk->af; 8249 return (PF_AFRT); 8250 } 8251 #endif 8252 8253 if (PF_ANEQ(pd2.src, 8254 &nk->addr[pd2.sidx], pd2.af) || 8255 nk->port[pd2.sidx] != sh.src_port) 8256 pf_change_icmp(pd2.src, &sh.src_port, 8257 daddr, &nk->addr[pd2.sidx], 8258 nk->port[pd2.sidx], NULL, 8259 pd2.ip_sum, icmpsum, 8260 pd->ip_sum, 0, pd2.af); 8261 8262 if (PF_ANEQ(pd2.dst, 8263 &nk->addr[pd2.didx], pd2.af) || 8264 nk->port[pd2.didx] != sh.dest_port) 8265 pf_change_icmp(pd2.dst, &sh.dest_port, 8266 saddr, &nk->addr[pd2.didx], 8267 nk->port[pd2.didx], NULL, 8268 pd2.ip_sum, icmpsum, 8269 pd->ip_sum, 0, pd2.af); 8270 copyback = 1; 8271 } 8272 8273 if (copyback) { 8274 switch (pd2.af) { 8275 #ifdef INET 8276 case AF_INET: 8277 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8278 (caddr_t )&pd->hdr.icmp); 8279 m_copyback(pd->m, ipoff2, sizeof(h2), 8280 (caddr_t )&h2); 8281 break; 8282 #endif /* INET */ 8283 #ifdef INET6 8284 case AF_INET6: 8285 m_copyback(pd->m, pd->off, 8286 sizeof(struct icmp6_hdr), 8287 (caddr_t )&pd->hdr.icmp6); 8288 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8289 (caddr_t )&h2_6); 8290 break; 8291 #endif /* INET6 */ 8292 } 8293 m_copyback(pd->m, pd2.off, sizeof(sh), (caddr_t)&sh); 8294 } 8295 8296 return (PF_PASS); 8297 break; 8298 } 8299 case IPPROTO_ICMP: { 8300 struct icmp *iih = &pd2.hdr.icmp; 8301 8302 if (pd2.af != AF_INET) { 8303 REASON_SET(reason, PFRES_NORM); 8304 return (PF_DROP); 8305 } 8306 8307 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8308 NULL, reason, pd2.af)) { 8309 DPFPRINTF(PF_DEBUG_MISC, 8310 ("pf: ICMP error message too short i" 8311 "(icmp)\n")); 8312 return (PF_DROP); 8313 } 8314 8315 icmpid = iih->icmp_id; 8316 pf_icmp_mapping(&pd2, iih->icmp_type, 8317 &icmp_dir, &multi, &virtual_id, &virtual_type); 8318 8319 ret = pf_icmp_state_lookup(&key, &pd2, state, 8320 pd2.dir, virtual_id, virtual_type, 8321 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 8322 if (ret >= 0) { 8323 MPASS(*state == NULL); 8324 return (ret); 8325 } 8326 8327 /* translate source/destination address, if necessary */ 8328 if ((*state)->key[PF_SK_WIRE] != 8329 (*state)->key[PF_SK_STACK]) { 8330 struct pf_state_key *nk; 8331 8332 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8333 nk = (*state)->key[pd->sidx]; 8334 else 8335 nk = (*state)->key[pd->didx]; 8336 8337 #if defined(INET) && defined(INET6) 8338 int afto, sidx, didx; 8339 8340 afto = pd->af != nk->af; 8341 sidx = afto ? pd2.didx : pd2.sidx; 8342 didx = afto ? pd2.sidx : pd2.didx; 8343 iidx = afto ? !iidx : iidx; 8344 8345 if (afto) { 8346 if (nk->af != AF_INET6) 8347 return (PF_DROP); 8348 if (pf_translate_icmp_af(nk->af, 8349 &pd->hdr.icmp)) 8350 return (PF_DROP); 8351 m_copyback(pd->m, pd->off, 8352 sizeof(struct icmp6_hdr), 8353 (c_caddr_t)&pd->hdr.icmp6); 8354 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8355 &pd2, &nk->addr[sidx], 8356 &nk->addr[didx], pd->af, 8357 nk->af)) 8358 return (PF_DROP); 8359 pd->proto = IPPROTO_ICMPV6; 8360 if (pf_translate_icmp_af(nk->af, &iih)) 8361 return (PF_DROP); 8362 if (virtual_type == htons(ICMP_ECHO) && 8363 nk->port[iidx] != iih->icmp_id) 8364 iih->icmp_id = nk->port[iidx]; 8365 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8366 (c_caddr_t)&iih); 8367 PF_ACPY(&pd->nsaddr, 8368 &nk->addr[pd2.sidx], nk->af); 8369 PF_ACPY(&pd->ndaddr, 8370 &nk->addr[pd2.didx], nk->af); 8371 pd->naf = nk->af; 8372 return (PF_AFRT); 8373 } 8374 #endif 8375 8376 if (PF_ANEQ(pd2.src, 8377 &nk->addr[pd2.sidx], pd2.af) || 8378 (virtual_type == htons(ICMP_ECHO) && 8379 nk->port[iidx] != iih->icmp_id)) 8380 pf_change_icmp(pd2.src, 8381 (virtual_type == htons(ICMP_ECHO)) ? 8382 &iih->icmp_id : NULL, 8383 daddr, &nk->addr[pd2.sidx], 8384 (virtual_type == htons(ICMP_ECHO)) ? 8385 nk->port[iidx] : 0, NULL, 8386 pd2.ip_sum, icmpsum, 8387 pd->ip_sum, 0, AF_INET); 8388 8389 if (PF_ANEQ(pd2.dst, 8390 &nk->addr[pd2.didx], pd2.af)) 8391 pf_change_icmp(pd2.dst, NULL, NULL, 8392 &nk->addr[pd2.didx], 0, NULL, 8393 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8394 AF_INET); 8395 8396 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8397 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8398 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8399 } 8400 return (PF_PASS); 8401 break; 8402 } 8403 #endif /* INET */ 8404 #ifdef INET6 8405 case IPPROTO_ICMPV6: { 8406 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8407 8408 if (pd2.af != AF_INET6) { 8409 REASON_SET(reason, PFRES_NORM); 8410 return (PF_DROP); 8411 } 8412 8413 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8414 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 8415 DPFPRINTF(PF_DEBUG_MISC, 8416 ("pf: ICMP error message too short " 8417 "(icmp6)\n")); 8418 return (PF_DROP); 8419 } 8420 8421 pf_icmp_mapping(&pd2, iih->icmp6_type, 8422 &icmp_dir, &multi, &virtual_id, &virtual_type); 8423 8424 ret = pf_icmp_state_lookup(&key, &pd2, state, 8425 pd->dir, virtual_id, virtual_type, 8426 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 8427 if (ret >= 0) { 8428 MPASS(*state == NULL); 8429 if (ret == PF_DROP && pd2.af == AF_INET6 && 8430 icmp_dir == PF_OUT) { 8431 ret = pf_icmp_state_lookup(&key, &pd2, 8432 state, pd->dir, 8433 virtual_id, virtual_type, 8434 icmp_dir, &iidx, multi, 1); 8435 if (ret >= 0) { 8436 MPASS(*state == NULL); 8437 return (ret); 8438 } 8439 } else 8440 return (ret); 8441 } 8442 8443 /* translate source/destination address, if necessary */ 8444 if ((*state)->key[PF_SK_WIRE] != 8445 (*state)->key[PF_SK_STACK]) { 8446 struct pf_state_key *nk; 8447 8448 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8449 nk = (*state)->key[pd->sidx]; 8450 else 8451 nk = (*state)->key[pd->didx]; 8452 8453 #if defined(INET) && defined(INET6) 8454 int afto, sidx, didx; 8455 8456 afto = pd->af != nk->af; 8457 sidx = afto ? pd2.didx : pd2.sidx; 8458 didx = afto ? pd2.sidx : pd2.didx; 8459 iidx = afto ? !iidx : iidx; 8460 8461 if (afto) { 8462 if (nk->af != AF_INET) 8463 return (PF_DROP); 8464 if (pf_translate_icmp_af(nk->af, 8465 &pd->hdr.icmp)) 8466 return (PF_DROP); 8467 m_copyback(pd->m, pd->off, 8468 sizeof(struct icmp6_hdr), 8469 (c_caddr_t)&pd->hdr.icmp6); 8470 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8471 &pd2, &nk->addr[sidx], 8472 &nk->addr[didx], pd->af, 8473 nk->af)) 8474 return (PF_DROP); 8475 pd->proto = IPPROTO_ICMP; 8476 if (pf_translate_icmp_af(nk->af, &iih)) 8477 return (PF_DROP); 8478 if (virtual_type == 8479 htons(ICMP6_ECHO_REQUEST) && 8480 nk->port[iidx] != iih->icmp6_id) 8481 iih->icmp6_id = nk->port[iidx]; 8482 m_copyback(pd2.m, pd2.off, 8483 sizeof(struct icmp6_hdr), (c_caddr_t)&iih); 8484 PF_ACPY(&pd->nsaddr, 8485 &nk->addr[pd2.sidx], nk->af); 8486 PF_ACPY(&pd->ndaddr, 8487 &nk->addr[pd2.didx], nk->af); 8488 pd->naf = nk->af; 8489 return (PF_AFRT); 8490 } 8491 #endif 8492 8493 if (PF_ANEQ(pd2.src, 8494 &nk->addr[pd2.sidx], pd2.af) || 8495 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8496 nk->port[pd2.sidx] != iih->icmp6_id)) 8497 pf_change_icmp(pd2.src, 8498 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8499 ? &iih->icmp6_id : NULL, 8500 daddr, &nk->addr[pd2.sidx], 8501 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8502 ? nk->port[iidx] : 0, NULL, 8503 pd2.ip_sum, icmpsum, 8504 pd->ip_sum, 0, AF_INET6); 8505 8506 if (PF_ANEQ(pd2.dst, 8507 &nk->addr[pd2.didx], pd2.af)) 8508 pf_change_icmp(pd2.dst, NULL, NULL, 8509 &nk->addr[pd2.didx], 0, NULL, 8510 pd2.ip_sum, icmpsum, 8511 pd->ip_sum, 0, AF_INET6); 8512 8513 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8514 (caddr_t)&pd->hdr.icmp6); 8515 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8516 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8517 (caddr_t)iih); 8518 } 8519 return (PF_PASS); 8520 break; 8521 } 8522 #endif /* INET6 */ 8523 default: { 8524 key.af = pd2.af; 8525 key.proto = pd2.proto; 8526 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8527 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8528 key.port[0] = key.port[1] = 0; 8529 8530 STATE_LOOKUP(&key, *state, pd); 8531 8532 /* translate source/destination address, if necessary */ 8533 if ((*state)->key[PF_SK_WIRE] != 8534 (*state)->key[PF_SK_STACK]) { 8535 struct pf_state_key *nk = 8536 (*state)->key[pd->didx]; 8537 8538 if (PF_ANEQ(pd2.src, 8539 &nk->addr[pd2.sidx], pd2.af)) 8540 pf_change_icmp(pd2.src, NULL, daddr, 8541 &nk->addr[pd2.sidx], 0, NULL, 8542 pd2.ip_sum, icmpsum, 8543 pd->ip_sum, 0, pd2.af); 8544 8545 if (PF_ANEQ(pd2.dst, 8546 &nk->addr[pd2.didx], pd2.af)) 8547 pf_change_icmp(pd2.dst, NULL, saddr, 8548 &nk->addr[pd2.didx], 0, NULL, 8549 pd2.ip_sum, icmpsum, 8550 pd->ip_sum, 0, pd2.af); 8551 8552 switch (pd2.af) { 8553 #ifdef INET 8554 case AF_INET: 8555 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8556 (caddr_t)&pd->hdr.icmp); 8557 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8558 break; 8559 #endif /* INET */ 8560 #ifdef INET6 8561 case AF_INET6: 8562 m_copyback(pd->m, pd->off, 8563 sizeof(struct icmp6_hdr), 8564 (caddr_t )&pd->hdr.icmp6); 8565 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8566 (caddr_t )&h2_6); 8567 break; 8568 #endif /* INET6 */ 8569 } 8570 } 8571 return (PF_PASS); 8572 break; 8573 } 8574 } 8575 } 8576 } 8577 8578 /* 8579 * ipoff and off are measured from the start of the mbuf chain. 8580 * h must be at "ipoff" on the mbuf chain. 8581 */ 8582 void * 8583 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8584 u_short *actionp, u_short *reasonp, sa_family_t af) 8585 { 8586 switch (af) { 8587 #ifdef INET 8588 case AF_INET: { 8589 const struct ip *h = mtod(m, struct ip *); 8590 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 8591 8592 if (fragoff) { 8593 if (fragoff >= len) 8594 ACTION_SET(actionp, PF_PASS); 8595 else { 8596 ACTION_SET(actionp, PF_DROP); 8597 REASON_SET(reasonp, PFRES_FRAG); 8598 } 8599 return (NULL); 8600 } 8601 if (m->m_pkthdr.len < off + len || 8602 ntohs(h->ip_len) < off + len) { 8603 ACTION_SET(actionp, PF_DROP); 8604 REASON_SET(reasonp, PFRES_SHORT); 8605 return (NULL); 8606 } 8607 break; 8608 } 8609 #endif /* INET */ 8610 #ifdef INET6 8611 case AF_INET6: { 8612 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8613 8614 if (m->m_pkthdr.len < off + len || 8615 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 8616 (unsigned)(off + len)) { 8617 ACTION_SET(actionp, PF_DROP); 8618 REASON_SET(reasonp, PFRES_SHORT); 8619 return (NULL); 8620 } 8621 break; 8622 } 8623 #endif /* INET6 */ 8624 } 8625 m_copydata(m, off, len, p); 8626 return (p); 8627 } 8628 8629 int 8630 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 8631 int rtableid) 8632 { 8633 struct ifnet *ifp; 8634 8635 /* 8636 * Skip check for addresses with embedded interface scope, 8637 * as they would always match anyway. 8638 */ 8639 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 8640 return (1); 8641 8642 if (af != AF_INET && af != AF_INET6) 8643 return (0); 8644 8645 if (kif == V_pfi_all) 8646 return (1); 8647 8648 /* Skip checks for ipsec interfaces */ 8649 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 8650 return (1); 8651 8652 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 8653 8654 switch (af) { 8655 #ifdef INET6 8656 case AF_INET6: 8657 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 8658 ifp)); 8659 #endif 8660 #ifdef INET 8661 case AF_INET: 8662 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 8663 ifp)); 8664 #endif 8665 } 8666 8667 return (0); 8668 } 8669 8670 #ifdef INET 8671 static void 8672 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8673 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8674 { 8675 struct mbuf *m0, *m1, *md; 8676 struct sockaddr_in dst; 8677 struct ip *ip; 8678 struct ifnet *ifp = NULL; 8679 int error = 0; 8680 uint16_t ip_len, ip_off; 8681 uint16_t tmp; 8682 int r_dir; 8683 bool skip_test = false; 8684 8685 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8686 8687 SDT_PROBE4(pf, ip, route_to, entry, *m, pd, s, oifp); 8688 8689 if (s) { 8690 r_dir = s->direction; 8691 } else { 8692 r_dir = r->direction; 8693 } 8694 8695 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8696 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8697 __func__)); 8698 8699 if ((pd->pf_mtag == NULL && 8700 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8701 pd->pf_mtag->routed++ > 3) { 8702 m0 = *m; 8703 *m = NULL; 8704 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8705 goto bad_locked; 8706 } 8707 8708 if (pd->act.rt_kif != NULL) 8709 ifp = pd->act.rt_kif->pfik_ifp; 8710 8711 if (pd->act.rt == PF_DUPTO) { 8712 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8713 if (s != NULL) { 8714 PF_STATE_UNLOCK(s); 8715 } 8716 if (ifp == oifp) { 8717 /* When the 2nd interface is not skipped */ 8718 return; 8719 } else { 8720 m0 = *m; 8721 *m = NULL; 8722 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8723 goto bad; 8724 } 8725 } else { 8726 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8727 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8728 if (s) 8729 PF_STATE_UNLOCK(s); 8730 return; 8731 } 8732 } 8733 } else { 8734 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8735 if (pd->af == pd->naf) { 8736 pf_dummynet(pd, s, r, m); 8737 if (s) 8738 PF_STATE_UNLOCK(s); 8739 return; 8740 } else { 8741 skip_test = true; 8742 } 8743 } 8744 8745 /* 8746 * If we're actually doing route-to and af-to and are in the 8747 * reply direction. 8748 */ 8749 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 8750 pd->af != pd->naf) { 8751 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 8752 /* Un-set ifp so we do a plain route lookup. */ 8753 ifp = NULL; 8754 } 8755 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 8756 /* Un-set ifp so we do a plain route lookup. */ 8757 ifp = NULL; 8758 } 8759 } 8760 m0 = *m; 8761 } 8762 8763 ip = mtod(m0, struct ip *); 8764 8765 bzero(&dst, sizeof(dst)); 8766 dst.sin_family = AF_INET; 8767 dst.sin_len = sizeof(dst); 8768 dst.sin_addr = ip->ip_dst; 8769 dst.sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 8770 8771 if (s != NULL){ 8772 if (ifp == NULL && (pd->af != pd->naf)) { 8773 /* We're in the AFTO case. Do a route lookup. */ 8774 const struct nhop_object *nh; 8775 nh = fib4_lookup(M_GETFIB(*m), ip->ip_dst, 0, NHR_NONE, 0); 8776 if (nh) { 8777 ifp = nh->nh_ifp; 8778 8779 /* Use the gateway if needed. */ 8780 if (nh->nh_flags & NHF_GATEWAY) 8781 dst.sin_addr = nh->gw4_sa.sin_addr; 8782 else 8783 dst.sin_addr = ip->ip_dst; 8784 8785 /* 8786 * Bind to the correct interface if we're 8787 * if-bound. We don't know which interface 8788 * that will be until here, so we've inserted 8789 * the state on V_pf_all. Fix that now. 8790 */ 8791 if (s->kif == V_pfi_all && ifp != NULL && 8792 r->rule_flag & PFRULE_IFBOUND) 8793 s->kif = ifp->if_pf_kif; 8794 } 8795 } 8796 8797 if (r->rule_flag & PFRULE_IFBOUND && 8798 pd->act.rt == PF_REPLYTO && 8799 s->kif == V_pfi_all) { 8800 s->kif = pd->act.rt_kif; 8801 s->orig_kif = oifp->if_pf_kif; 8802 } 8803 8804 PF_STATE_UNLOCK(s); 8805 } 8806 8807 if (ifp == NULL) { 8808 m0 = *m; 8809 *m = NULL; 8810 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8811 goto bad; 8812 } 8813 8814 if (pd->dir == PF_IN && !skip_test) { 8815 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8816 &pd->act) != PF_PASS) { 8817 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8818 goto bad; 8819 } else if (m0 == NULL) { 8820 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8821 goto done; 8822 } 8823 if (m0->m_len < sizeof(struct ip)) { 8824 DPFPRINTF(PF_DEBUG_URGENT, 8825 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 8826 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8827 goto bad; 8828 } 8829 ip = mtod(m0, struct ip *); 8830 } 8831 8832 if (ifp->if_flags & IFF_LOOPBACK) 8833 m0->m_flags |= M_SKIP_FIREWALL; 8834 8835 ip_len = ntohs(ip->ip_len); 8836 ip_off = ntohs(ip->ip_off); 8837 8838 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 8839 m0->m_pkthdr.csum_flags |= CSUM_IP; 8840 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 8841 in_delayed_cksum(m0); 8842 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 8843 } 8844 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 8845 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 8846 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 8847 } 8848 8849 if (pd->dir == PF_IN) { 8850 /* 8851 * Make sure dummynet gets the correct direction, in case it needs to 8852 * re-inject later. 8853 */ 8854 pd->dir = PF_OUT; 8855 8856 /* 8857 * The following processing is actually the rest of the inbound processing, even 8858 * though we've marked it as outbound (so we don't look through dummynet) and it 8859 * happens after the outbound processing (pf_test(PF_OUT) above). 8860 * Swap the dummynet pipe numbers, because it's going to come to the wrong 8861 * conclusion about what direction it's processing, and we can't fix it or it 8862 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 8863 * decision will pick the right pipe, and everything will mostly work as expected. 8864 */ 8865 tmp = pd->act.dnrpipe; 8866 pd->act.dnrpipe = pd->act.dnpipe; 8867 pd->act.dnpipe = tmp; 8868 } 8869 8870 /* 8871 * If small enough for interface, or the interface will take 8872 * care of the fragmentation for us, we can just send directly. 8873 */ 8874 if (ip_len <= ifp->if_mtu || 8875 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 8876 ip->ip_sum = 0; 8877 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 8878 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 8879 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 8880 } 8881 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 8882 8883 md = m0; 8884 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8885 if (md != NULL) { 8886 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 8887 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 8888 } 8889 goto done; 8890 } 8891 8892 /* Balk when DF bit is set or the interface didn't support TSO. */ 8893 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 8894 error = EMSGSIZE; 8895 KMOD_IPSTAT_INC(ips_cantfrag); 8896 if (pd->act.rt != PF_DUPTO) { 8897 if (s && s->nat_rule != NULL) 8898 PACKET_UNDO_NAT(m0, pd, 8899 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 8900 s); 8901 8902 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 8903 ifp->if_mtu); 8904 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8905 goto done; 8906 } else { 8907 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8908 goto bad; 8909 } 8910 } 8911 8912 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 8913 if (error) { 8914 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8915 goto bad; 8916 } 8917 8918 for (; m0; m0 = m1) { 8919 m1 = m0->m_nextpkt; 8920 m0->m_nextpkt = NULL; 8921 if (error == 0) { 8922 m_clrprotoflags(m0); 8923 md = m0; 8924 pd->pf_mtag = pf_find_mtag(md); 8925 error = pf_dummynet_route(pd, s, r, ifp, 8926 sintosa(&dst), &md); 8927 if (md != NULL) { 8928 error = (*ifp->if_output)(ifp, md, 8929 sintosa(&dst), NULL); 8930 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 8931 } 8932 } else 8933 m_freem(m0); 8934 } 8935 8936 if (error == 0) 8937 KMOD_IPSTAT_INC(ips_fragmented); 8938 8939 done: 8940 if (pd->act.rt != PF_DUPTO) 8941 *m = NULL; 8942 return; 8943 8944 bad_locked: 8945 if (s) 8946 PF_STATE_UNLOCK(s); 8947 bad: 8948 m_freem(m0); 8949 goto done; 8950 } 8951 #endif /* INET */ 8952 8953 #ifdef INET6 8954 static void 8955 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8956 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8957 { 8958 struct mbuf *m0, *md; 8959 struct m_tag *mtag; 8960 struct sockaddr_in6 dst; 8961 struct ip6_hdr *ip6; 8962 struct ifnet *ifp = NULL; 8963 int r_dir; 8964 bool skip_test = false; 8965 8966 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8967 8968 SDT_PROBE4(pf, ip6, route_to, entry, *m, pd, s, oifp); 8969 8970 if (s) { 8971 r_dir = s->direction; 8972 } else { 8973 r_dir = r->direction; 8974 } 8975 8976 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8977 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8978 __func__)); 8979 8980 if ((pd->pf_mtag == NULL && 8981 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8982 pd->pf_mtag->routed++ > 3) { 8983 m0 = *m; 8984 *m = NULL; 8985 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8986 goto bad_locked; 8987 } 8988 8989 if (pd->act.rt_kif != NULL) 8990 ifp = pd->act.rt_kif->pfik_ifp; 8991 8992 if (pd->act.rt == PF_DUPTO) { 8993 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8994 if (s != NULL) { 8995 PF_STATE_UNLOCK(s); 8996 } 8997 if (ifp == oifp) { 8998 /* When the 2nd interface is not skipped */ 8999 return; 9000 } else { 9001 m0 = *m; 9002 *m = NULL; 9003 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9004 goto bad; 9005 } 9006 } else { 9007 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9008 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 9009 if (s) 9010 PF_STATE_UNLOCK(s); 9011 return; 9012 } 9013 } 9014 } else { 9015 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9016 if (pd->af == pd->naf) { 9017 pf_dummynet(pd, s, r, m); 9018 if (s) 9019 PF_STATE_UNLOCK(s); 9020 return; 9021 } else { 9022 skip_test = true; 9023 } 9024 } 9025 9026 /* 9027 * If we're actually doing route-to and af-to and are in the 9028 * reply direction. 9029 */ 9030 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9031 pd->af != pd->naf) { 9032 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9033 /* Un-set ifp so we do a plain route lookup. */ 9034 ifp = NULL; 9035 } 9036 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9037 /* Un-set ifp so we do a plain route lookup. */ 9038 ifp = NULL; 9039 } 9040 } 9041 m0 = *m; 9042 } 9043 9044 ip6 = mtod(m0, struct ip6_hdr *); 9045 9046 bzero(&dst, sizeof(dst)); 9047 dst.sin6_family = AF_INET6; 9048 dst.sin6_len = sizeof(dst); 9049 dst.sin6_addr = ip6->ip6_dst; 9050 PF_ACPY((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, AF_INET6); 9051 9052 if (s != NULL) { 9053 if (ifp == NULL && (pd->af != pd->naf)) { 9054 const struct nhop_object *nh; 9055 nh = fib6_lookup(M_GETFIB(*m), &ip6->ip6_dst, 0, NHR_NONE, 0); 9056 if (nh) { 9057 ifp = nh->nh_ifp; 9058 9059 /* Use the gateway if needed. */ 9060 if (nh->nh_flags & NHF_GATEWAY) 9061 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9062 sizeof(dst.sin6_addr)); 9063 else 9064 dst.sin6_addr = ip6->ip6_dst; 9065 9066 /* 9067 * Bind to the correct interface if we're 9068 * if-bound. We don't know which interface 9069 * that will be until here, so we've inserted 9070 * the state on V_pf_all. Fix that now. 9071 */ 9072 if (s->kif == V_pfi_all && ifp != NULL && 9073 r->rule_flag & PFRULE_IFBOUND) 9074 s->kif = ifp->if_pf_kif; 9075 } 9076 } 9077 9078 if (r->rule_flag & PFRULE_IFBOUND && 9079 pd->act.rt == PF_REPLYTO && 9080 s->kif == V_pfi_all) { 9081 s->kif = pd->act.rt_kif; 9082 s->orig_kif = oifp->if_pf_kif; 9083 } 9084 9085 PF_STATE_UNLOCK(s); 9086 } 9087 9088 if (pd->af != pd->naf) { 9089 struct udphdr *uh = &pd->hdr.udp; 9090 9091 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9092 uh->uh_sum = in6_cksum_pseudo(ip6, 9093 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9094 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9095 } 9096 } 9097 9098 if (ifp == NULL) { 9099 m0 = *m; 9100 *m = NULL; 9101 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9102 goto bad; 9103 } 9104 9105 if (pd->dir == PF_IN && !skip_test) { 9106 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9107 ifp, &m0, inp, &pd->act) != PF_PASS) { 9108 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9109 goto bad; 9110 } else if (m0 == NULL) { 9111 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9112 goto done; 9113 } 9114 if (m0->m_len < sizeof(struct ip6_hdr)) { 9115 DPFPRINTF(PF_DEBUG_URGENT, 9116 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 9117 __func__)); 9118 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9119 goto bad; 9120 } 9121 ip6 = mtod(m0, struct ip6_hdr *); 9122 } 9123 9124 if (ifp->if_flags & IFF_LOOPBACK) 9125 m0->m_flags |= M_SKIP_FIREWALL; 9126 9127 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9128 ~ifp->if_hwassist) { 9129 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9130 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9131 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9132 } 9133 9134 if (pd->dir == PF_IN) { 9135 uint16_t tmp; 9136 /* 9137 * Make sure dummynet gets the correct direction, in case it needs to 9138 * re-inject later. 9139 */ 9140 pd->dir = PF_OUT; 9141 9142 /* 9143 * The following processing is actually the rest of the inbound processing, even 9144 * though we've marked it as outbound (so we don't look through dummynet) and it 9145 * happens after the outbound processing (pf_test(PF_OUT) above). 9146 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9147 * conclusion about what direction it's processing, and we can't fix it or it 9148 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9149 * decision will pick the right pipe, and everything will mostly work as expected. 9150 */ 9151 tmp = pd->act.dnrpipe; 9152 pd->act.dnrpipe = pd->act.dnpipe; 9153 pd->act.dnpipe = tmp; 9154 } 9155 9156 /* 9157 * If the packet is too large for the outgoing interface, 9158 * send back an icmp6 error. 9159 */ 9160 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9161 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9162 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9163 if (mtag != NULL) { 9164 int ret __sdt_used; 9165 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9166 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9167 goto done; 9168 } 9169 9170 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9171 md = m0; 9172 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9173 if (md != NULL) { 9174 int ret __sdt_used; 9175 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9176 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9177 } 9178 } 9179 else { 9180 in6_ifstat_inc(ifp, ifs6_in_toobig); 9181 if (pd->act.rt != PF_DUPTO) { 9182 if (s && s->nat_rule != NULL) 9183 PACKET_UNDO_NAT(m0, pd, 9184 ((caddr_t)ip6 - m0->m_data) + 9185 sizeof(struct ip6_hdr), s); 9186 9187 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 9188 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9189 } else { 9190 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9191 goto bad; 9192 } 9193 } 9194 9195 done: 9196 if (pd->act.rt != PF_DUPTO) 9197 *m = NULL; 9198 return; 9199 9200 bad_locked: 9201 if (s) 9202 PF_STATE_UNLOCK(s); 9203 bad: 9204 m_freem(m0); 9205 goto done; 9206 } 9207 #endif /* INET6 */ 9208 9209 /* 9210 * FreeBSD supports cksum offloads for the following drivers. 9211 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9212 * 9213 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9214 * network driver performed cksum including pseudo header, need to verify 9215 * csum_data 9216 * CSUM_DATA_VALID : 9217 * network driver performed cksum, needs to additional pseudo header 9218 * cksum computation with partial csum_data(i.e. lack of H/W support for 9219 * pseudo header, for instance sk(4) and possibly gem(4)) 9220 * 9221 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9222 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9223 * TCP/UDP layer. 9224 * Also, set csum_data to 0xffff to force cksum validation. 9225 */ 9226 static int 9227 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9228 { 9229 u_int16_t sum = 0; 9230 int hw_assist = 0; 9231 struct ip *ip; 9232 9233 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9234 return (1); 9235 if (m->m_pkthdr.len < off + len) 9236 return (1); 9237 9238 switch (p) { 9239 case IPPROTO_TCP: 9240 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9241 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9242 sum = m->m_pkthdr.csum_data; 9243 } else { 9244 ip = mtod(m, struct ip *); 9245 sum = in_pseudo(ip->ip_src.s_addr, 9246 ip->ip_dst.s_addr, htonl((u_short)len + 9247 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9248 } 9249 sum ^= 0xffff; 9250 ++hw_assist; 9251 } 9252 break; 9253 case IPPROTO_UDP: 9254 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9255 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9256 sum = m->m_pkthdr.csum_data; 9257 } else { 9258 ip = mtod(m, struct ip *); 9259 sum = in_pseudo(ip->ip_src.s_addr, 9260 ip->ip_dst.s_addr, htonl((u_short)len + 9261 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9262 } 9263 sum ^= 0xffff; 9264 ++hw_assist; 9265 } 9266 break; 9267 case IPPROTO_ICMP: 9268 #ifdef INET6 9269 case IPPROTO_ICMPV6: 9270 #endif /* INET6 */ 9271 break; 9272 default: 9273 return (1); 9274 } 9275 9276 if (!hw_assist) { 9277 switch (af) { 9278 case AF_INET: 9279 if (m->m_len < sizeof(struct ip)) 9280 return (1); 9281 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 9282 break; 9283 #ifdef INET6 9284 case AF_INET6: 9285 if (m->m_len < sizeof(struct ip6_hdr)) 9286 return (1); 9287 sum = in6_cksum(m, p, off, len); 9288 break; 9289 #endif /* INET6 */ 9290 } 9291 } 9292 if (sum) { 9293 switch (p) { 9294 case IPPROTO_TCP: 9295 { 9296 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9297 break; 9298 } 9299 case IPPROTO_UDP: 9300 { 9301 KMOD_UDPSTAT_INC(udps_badsum); 9302 break; 9303 } 9304 #ifdef INET 9305 case IPPROTO_ICMP: 9306 { 9307 KMOD_ICMPSTAT_INC(icps_checksum); 9308 break; 9309 } 9310 #endif 9311 #ifdef INET6 9312 case IPPROTO_ICMPV6: 9313 { 9314 KMOD_ICMP6STAT_INC(icp6s_checksum); 9315 break; 9316 } 9317 #endif /* INET6 */ 9318 } 9319 return (1); 9320 } else { 9321 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9322 m->m_pkthdr.csum_flags |= 9323 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9324 m->m_pkthdr.csum_data = 0xffff; 9325 } 9326 } 9327 return (0); 9328 } 9329 9330 static bool 9331 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9332 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9333 { 9334 int dndir = r->direction; 9335 9336 if (s && dndir == PF_INOUT) { 9337 dndir = s->direction; 9338 } else if (dndir == PF_INOUT) { 9339 /* Assume primary direction. Happens when we've set dnpipe in 9340 * the ethernet level code. */ 9341 dndir = pd->dir; 9342 } 9343 9344 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9345 return (false); 9346 9347 memset(dnflow, 0, sizeof(*dnflow)); 9348 9349 if (pd->dport != NULL) 9350 dnflow->f_id.dst_port = ntohs(*pd->dport); 9351 if (pd->sport != NULL) 9352 dnflow->f_id.src_port = ntohs(*pd->sport); 9353 9354 if (pd->dir == PF_IN) 9355 dnflow->flags |= IPFW_ARGS_IN; 9356 else 9357 dnflow->flags |= IPFW_ARGS_OUT; 9358 9359 if (pd->dir != dndir && pd->act.dnrpipe) { 9360 dnflow->rule.info = pd->act.dnrpipe; 9361 } 9362 else if (pd->dir == dndir && pd->act.dnpipe) { 9363 dnflow->rule.info = pd->act.dnpipe; 9364 } 9365 else { 9366 return (false); 9367 } 9368 9369 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9370 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9371 dnflow->rule.info |= IPFW_IS_PIPE; 9372 9373 dnflow->f_id.proto = pd->proto; 9374 dnflow->f_id.extra = dnflow->rule.info; 9375 switch (pd->naf) { 9376 case AF_INET: 9377 dnflow->f_id.addr_type = 4; 9378 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9379 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9380 break; 9381 case AF_INET6: 9382 dnflow->flags |= IPFW_ARGS_IP6; 9383 dnflow->f_id.addr_type = 6; 9384 dnflow->f_id.src_ip6 = pd->src->v6; 9385 dnflow->f_id.dst_ip6 = pd->dst->v6; 9386 break; 9387 } 9388 9389 return (true); 9390 } 9391 9392 int 9393 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9394 struct inpcb *inp) 9395 { 9396 struct pfi_kkif *kif; 9397 struct mbuf *m = *m0; 9398 9399 M_ASSERTPKTHDR(m); 9400 MPASS(ifp->if_vnet == curvnet); 9401 NET_EPOCH_ASSERT(); 9402 9403 if (!V_pf_status.running) 9404 return (PF_PASS); 9405 9406 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9407 9408 if (kif == NULL) { 9409 DPFPRINTF(PF_DEBUG_URGENT, 9410 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 9411 return (PF_DROP); 9412 } 9413 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9414 return (PF_PASS); 9415 9416 if (m->m_flags & M_SKIP_FIREWALL) 9417 return (PF_PASS); 9418 9419 if (__predict_false(! M_WRITABLE(*m0))) { 9420 m = *m0 = m_unshare(*m0, M_NOWAIT); 9421 if (*m0 == NULL) 9422 return (PF_DROP); 9423 } 9424 9425 /* Stateless! */ 9426 return (pf_test_eth_rule(dir, kif, m0)); 9427 } 9428 9429 static __inline void 9430 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9431 { 9432 struct m_tag *mtag; 9433 9434 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9435 9436 /* dummynet adds this tag, but pf does not need it, 9437 * and keeping it creates unexpected behavior, 9438 * e.g. in case of divert(4) usage right after dummynet. */ 9439 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9440 if (mtag != NULL) 9441 m_tag_delete(m, mtag); 9442 } 9443 9444 static int 9445 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9446 struct pf_krule *r, struct mbuf **m0) 9447 { 9448 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9449 } 9450 9451 static int 9452 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9453 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 9454 struct mbuf **m0) 9455 { 9456 struct ip_fw_args dnflow; 9457 9458 NET_EPOCH_ASSERT(); 9459 9460 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9461 return (0); 9462 9463 if (ip_dn_io_ptr == NULL) { 9464 m_freem(*m0); 9465 *m0 = NULL; 9466 return (ENOMEM); 9467 } 9468 9469 if (pd->pf_mtag == NULL && 9470 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 9471 m_freem(*m0); 9472 *m0 = NULL; 9473 return (ENOMEM); 9474 } 9475 9476 if (ifp != NULL) { 9477 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 9478 9479 pd->pf_mtag->if_index = ifp->if_index; 9480 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 9481 9482 MPASS(sa != NULL); 9483 9484 switch (pd->naf) { 9485 case AF_INET: 9486 memcpy(&pd->pf_mtag->dst, sa, 9487 sizeof(struct sockaddr_in)); 9488 break; 9489 case AF_INET6: 9490 memcpy(&pd->pf_mtag->dst, sa, 9491 sizeof(struct sockaddr_in6)); 9492 break; 9493 } 9494 } 9495 9496 if (s != NULL && s->nat_rule != NULL && 9497 s->nat_rule->action == PF_RDR && 9498 ( 9499 #ifdef INET 9500 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 9501 #endif 9502 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 9503 /* 9504 * If we're redirecting to loopback mark this packet 9505 * as being local. Otherwise it might get dropped 9506 * if dummynet re-injects. 9507 */ 9508 (*m0)->m_pkthdr.rcvif = V_loif; 9509 } 9510 9511 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 9512 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 9513 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 9514 ip_dn_io_ptr(m0, &dnflow); 9515 if (*m0 != NULL) { 9516 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9517 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 9518 } 9519 } 9520 9521 return (0); 9522 } 9523 9524 #ifdef INET6 9525 static int 9526 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 9527 u_short *reason) 9528 { 9529 struct ip6_opt opt; 9530 struct ip6_opt_jumbo jumbo; 9531 9532 while (off < end) { 9533 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 9534 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 9535 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 9536 return (PF_DROP); 9537 } 9538 if (opt.ip6o_type == IP6OPT_PAD1) { 9539 off++; 9540 continue; 9541 } 9542 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL, 9543 reason, AF_INET6)) { 9544 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 9545 return (PF_DROP); 9546 } 9547 if (off + sizeof(opt) + opt.ip6o_len > end) { 9548 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 9549 REASON_SET(reason, PFRES_IPOPTIONS); 9550 return (PF_DROP); 9551 } 9552 switch (opt.ip6o_type) { 9553 case IP6OPT_JUMBO: 9554 if (pd->jumbolen != 0) { 9555 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 9556 REASON_SET(reason, PFRES_IPOPTIONS); 9557 return (PF_DROP); 9558 } 9559 if (ntohs(h->ip6_plen) != 0) { 9560 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 9561 REASON_SET(reason, PFRES_IPOPTIONS); 9562 return (PF_DROP); 9563 } 9564 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL, 9565 reason, AF_INET6)) { 9566 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 9567 return (PF_DROP); 9568 } 9569 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 9570 sizeof(pd->jumbolen)); 9571 pd->jumbolen = ntohl(pd->jumbolen); 9572 if (pd->jumbolen < IPV6_MAXPACKET) { 9573 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 9574 REASON_SET(reason, PFRES_IPOPTIONS); 9575 return (PF_DROP); 9576 } 9577 break; 9578 default: 9579 break; 9580 } 9581 off += sizeof(opt) + opt.ip6o_len; 9582 } 9583 9584 return (PF_PASS); 9585 } 9586 9587 int 9588 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 9589 { 9590 struct ip6_frag frag; 9591 struct ip6_ext ext; 9592 struct ip6_rthdr rthdr; 9593 uint32_t end; 9594 int fraghdr_cnt = 0, rthdr_cnt = 0; 9595 9596 pd->off += sizeof(struct ip6_hdr); 9597 end = pd->off + ntohs(h->ip6_plen); 9598 pd->fragoff = pd->extoff = pd->jumbolen = 0; 9599 pd->proto = h->ip6_nxt; 9600 for (;;) { 9601 switch (pd->proto) { 9602 case IPPROTO_FRAGMENT: 9603 if (fraghdr_cnt++) { 9604 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 9605 REASON_SET(reason, PFRES_FRAG); 9606 return (PF_DROP); 9607 } 9608 /* jumbo payload packets cannot be fragmented */ 9609 if (pd->jumbolen != 0) { 9610 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 9611 REASON_SET(reason, PFRES_FRAG); 9612 return (PF_DROP); 9613 } 9614 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 9615 NULL, reason, AF_INET6)) { 9616 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 9617 return (PF_DROP); 9618 } 9619 /* stop walking over non initial fragments */ 9620 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 9621 pd->fragoff = pd->off; 9622 return (PF_PASS); 9623 } 9624 /* RFC6946: reassemble only non atomic fragments */ 9625 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 9626 pd->fragoff = pd->off; 9627 pd->off += sizeof(frag); 9628 pd->proto = frag.ip6f_nxt; 9629 break; 9630 case IPPROTO_ROUTING: 9631 if (rthdr_cnt++) { 9632 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 9633 REASON_SET(reason, PFRES_IPOPTIONS); 9634 return (PF_DROP); 9635 } 9636 /* fragments may be short */ 9637 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 9638 pd->off = pd->fragoff; 9639 pd->proto = IPPROTO_FRAGMENT; 9640 return (PF_PASS); 9641 } 9642 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 9643 NULL, reason, AF_INET6)) { 9644 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 9645 return (PF_DROP); 9646 } 9647 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9648 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 9649 REASON_SET(reason, PFRES_IPOPTIONS); 9650 return (PF_DROP); 9651 } 9652 /* FALLTHROUGH */ 9653 case IPPROTO_AH: 9654 case IPPROTO_HOPOPTS: 9655 case IPPROTO_DSTOPTS: 9656 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9657 NULL, reason, AF_INET6)) { 9658 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 9659 return (PF_DROP); 9660 } 9661 /* fragments may be short */ 9662 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 9663 pd->off = pd->fragoff; 9664 pd->proto = IPPROTO_FRAGMENT; 9665 return (PF_PASS); 9666 } 9667 /* reassembly needs the ext header before the frag */ 9668 if (pd->fragoff == 0) 9669 pd->extoff = pd->off; 9670 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0) { 9671 if (pf_walk_option6(pd, h, 9672 pd->off + sizeof(ext), 9673 pd->off + (ext.ip6e_len + 1) * 8, reason) 9674 != PF_PASS) 9675 return (PF_DROP); 9676 if (ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 9677 DPFPRINTF(PF_DEBUG_MISC, 9678 ("IPv6 missing jumbo")); 9679 REASON_SET(reason, PFRES_IPOPTIONS); 9680 return (PF_DROP); 9681 } 9682 } 9683 if (pd->proto == IPPROTO_AH) 9684 pd->off += (ext.ip6e_len + 2) * 4; 9685 else 9686 pd->off += (ext.ip6e_len + 1) * 8; 9687 pd->proto = ext.ip6e_nxt; 9688 break; 9689 case IPPROTO_TCP: 9690 case IPPROTO_UDP: 9691 case IPPROTO_SCTP: 9692 case IPPROTO_ICMPV6: 9693 /* fragments may be short, ignore inner header then */ 9694 if (pd->fragoff != 0 && end < pd->off + 9695 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 9696 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 9697 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 9698 sizeof(struct icmp6_hdr))) { 9699 pd->off = pd->fragoff; 9700 pd->proto = IPPROTO_FRAGMENT; 9701 } 9702 /* FALLTHROUGH */ 9703 default: 9704 return (PF_PASS); 9705 } 9706 } 9707 } 9708 #endif 9709 9710 static void 9711 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 9712 { 9713 memset(pd, 0, sizeof(*pd)); 9714 pd->pf_mtag = pf_find_mtag(m); 9715 pd->m = m; 9716 } 9717 9718 static int 9719 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 9720 u_short *action, u_short *reason, struct pfi_kkif *kif, 9721 struct pf_rule_actions *default_actions) 9722 { 9723 pd->dir = dir; 9724 pd->kif = kif; 9725 pd->m = *m0; 9726 pd->sidx = (dir == PF_IN) ? 0 : 1; 9727 pd->didx = (dir == PF_IN) ? 1 : 0; 9728 pd->af = pd->naf = af; 9729 9730 TAILQ_INIT(&pd->sctp_multihome_jobs); 9731 if (default_actions != NULL) 9732 memcpy(&pd->act, default_actions, sizeof(pd->act)); 9733 9734 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 9735 pd->act.dnpipe = pd->pf_mtag->dnpipe; 9736 pd->act.flags = pd->pf_mtag->dnflags; 9737 } 9738 9739 switch (af) { 9740 #ifdef INET 9741 case AF_INET: { 9742 struct ip *h; 9743 9744 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 9745 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 9746 DPFPRINTF(PF_DEBUG_URGENT, 9747 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 9748 *action = PF_DROP; 9749 REASON_SET(reason, PFRES_SHORT); 9750 return (-1); 9751 } 9752 9753 if (pf_normalize_ip(reason, pd) != PF_PASS) { 9754 /* We do IP header normalization and packet reassembly here */ 9755 *m0 = pd->m; 9756 *action = PF_DROP; 9757 return (-1); 9758 } 9759 *m0 = pd->m; 9760 9761 h = mtod(pd->m, struct ip *); 9762 pd->off = h->ip_hl << 2; 9763 if (pd->off < (int)sizeof(*h)) { 9764 *action = PF_DROP; 9765 REASON_SET(reason, PFRES_SHORT); 9766 return (-1); 9767 } 9768 pd->src = (struct pf_addr *)&h->ip_src; 9769 pd->dst = (struct pf_addr *)&h->ip_dst; 9770 pd->ip_sum = &h->ip_sum; 9771 pd->virtual_proto = pd->proto = h->ip_p; 9772 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 9773 pd->ttl = h->ip_ttl; 9774 pd->tot_len = ntohs(h->ip_len); 9775 pd->act.rtableid = -1; 9776 pd->df = h->ip_off & htons(IP_DF); 9777 9778 if (h->ip_hl > 5) /* has options */ 9779 pd->badopts++; 9780 9781 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 9782 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9783 9784 break; 9785 } 9786 #endif 9787 #ifdef INET6 9788 case AF_INET6: { 9789 struct ip6_hdr *h; 9790 9791 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 9792 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 9793 DPFPRINTF(PF_DEBUG_URGENT, 9794 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 9795 ", pullup failed\n")); 9796 *action = PF_DROP; 9797 REASON_SET(reason, PFRES_SHORT); 9798 return (-1); 9799 } 9800 9801 h = mtod(pd->m, struct ip6_hdr *); 9802 pd->off = 0; 9803 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9804 *action = PF_DROP; 9805 return (-1); 9806 } 9807 9808 h = mtod(pd->m, struct ip6_hdr *); 9809 pd->src = (struct pf_addr *)&h->ip6_src; 9810 pd->dst = (struct pf_addr *)&h->ip6_dst; 9811 pd->ip_sum = NULL; 9812 pd->tos = IPV6_DSCP(h); 9813 pd->ttl = h->ip6_hlim; 9814 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 9815 pd->virtual_proto = pd->proto = h->ip6_nxt; 9816 pd->act.rtableid = -1; 9817 9818 if (pd->fragoff != 0) 9819 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9820 9821 /* 9822 * we do not support jumbogram. if we keep going, zero ip6_plen 9823 * will do something bad, so drop the packet for now. 9824 */ 9825 if (htons(h->ip6_plen) == 0) { 9826 *action = PF_DROP; 9827 return (-1); 9828 } 9829 9830 /* We do IP header normalization and packet reassembly here */ 9831 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 9832 PF_PASS) { 9833 *m0 = pd->m; 9834 *action = PF_DROP; 9835 return (-1); 9836 } 9837 *m0 = pd->m; 9838 if (pd->m == NULL) { 9839 /* packet sits in reassembly queue, no error */ 9840 *action = PF_PASS; 9841 return (-1); 9842 } 9843 9844 /* Update pointers into the packet. */ 9845 h = mtod(pd->m, struct ip6_hdr *); 9846 pd->src = (struct pf_addr *)&h->ip6_src; 9847 pd->dst = (struct pf_addr *)&h->ip6_dst; 9848 9849 pd->off = 0; 9850 9851 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9852 *action = PF_DROP; 9853 return (-1); 9854 } 9855 9856 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 9857 /* 9858 * Reassembly may have changed the next protocol from 9859 * fragment to something else, so update. 9860 */ 9861 pd->virtual_proto = pd->proto; 9862 MPASS(pd->fragoff == 0); 9863 } 9864 9865 if (pd->fragoff != 0) 9866 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9867 9868 break; 9869 } 9870 #endif 9871 default: 9872 panic("pf_setup_pdesc called with illegal af %u", af); 9873 } 9874 9875 switch (pd->virtual_proto) { 9876 case IPPROTO_TCP: { 9877 struct tcphdr *th = &pd->hdr.tcp; 9878 9879 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 9880 reason, af)) { 9881 *action = PF_DROP; 9882 REASON_SET(reason, PFRES_SHORT); 9883 return (-1); 9884 } 9885 pd->hdrlen = sizeof(*th); 9886 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 9887 pd->sport = &th->th_sport; 9888 pd->dport = &th->th_dport; 9889 pd->pcksum = &th->th_sum; 9890 break; 9891 } 9892 case IPPROTO_UDP: { 9893 struct udphdr *uh = &pd->hdr.udp; 9894 9895 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 9896 reason, af)) { 9897 *action = PF_DROP; 9898 REASON_SET(reason, PFRES_SHORT); 9899 return (-1); 9900 } 9901 pd->hdrlen = sizeof(*uh); 9902 if (uh->uh_dport == 0 || 9903 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 9904 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 9905 *action = PF_DROP; 9906 REASON_SET(reason, PFRES_SHORT); 9907 return (-1); 9908 } 9909 pd->sport = &uh->uh_sport; 9910 pd->dport = &uh->uh_dport; 9911 pd->pcksum = &uh->uh_sum; 9912 break; 9913 } 9914 case IPPROTO_SCTP: { 9915 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 9916 action, reason, af)) { 9917 *action = PF_DROP; 9918 REASON_SET(reason, PFRES_SHORT); 9919 return (-1); 9920 } 9921 pd->hdrlen = sizeof(pd->hdr.sctp); 9922 pd->p_len = pd->tot_len - pd->off; 9923 9924 pd->sport = &pd->hdr.sctp.src_port; 9925 pd->dport = &pd->hdr.sctp.dest_port; 9926 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 9927 *action = PF_DROP; 9928 REASON_SET(reason, PFRES_SHORT); 9929 return (-1); 9930 } 9931 if (pf_scan_sctp(pd) != PF_PASS) { 9932 *action = PF_DROP; 9933 REASON_SET(reason, PFRES_SHORT); 9934 return (-1); 9935 } 9936 /* 9937 * Placeholder. The SCTP checksum is 32-bits, but 9938 * pf_test_state() expects to update a 16-bit checksum. 9939 * Provide a dummy value which we'll subsequently ignore. 9940 */ 9941 pd->pcksum = &pd->sctp_dummy_sum; 9942 break; 9943 } 9944 case IPPROTO_ICMP: { 9945 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 9946 action, reason, af)) { 9947 *action = PF_DROP; 9948 REASON_SET(reason, PFRES_SHORT); 9949 return (-1); 9950 } 9951 pd->hdrlen = ICMP_MINLEN; 9952 break; 9953 } 9954 #ifdef INET6 9955 case IPPROTO_ICMPV6: { 9956 size_t icmp_hlen = sizeof(struct icmp6_hdr); 9957 9958 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 9959 action, reason, af)) { 9960 *action = PF_DROP; 9961 REASON_SET(reason, PFRES_SHORT); 9962 return (-1); 9963 } 9964 /* ICMP headers we look further into to match state */ 9965 switch (pd->hdr.icmp6.icmp6_type) { 9966 case MLD_LISTENER_QUERY: 9967 case MLD_LISTENER_REPORT: 9968 icmp_hlen = sizeof(struct mld_hdr); 9969 break; 9970 case ND_NEIGHBOR_SOLICIT: 9971 case ND_NEIGHBOR_ADVERT: 9972 icmp_hlen = sizeof(struct nd_neighbor_solicit); 9973 break; 9974 } 9975 if (icmp_hlen > sizeof(struct icmp6_hdr) && 9976 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 9977 action, reason, af)) { 9978 *action = PF_DROP; 9979 REASON_SET(reason, PFRES_SHORT); 9980 return (-1); 9981 } 9982 pd->hdrlen = icmp_hlen; 9983 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 9984 break; 9985 } 9986 #endif 9987 } 9988 9989 if (pd->sport) 9990 pd->osport = pd->nsport = *pd->sport; 9991 if (pd->dport) 9992 pd->odport = pd->ndport = *pd->dport; 9993 9994 return (0); 9995 } 9996 9997 static void 9998 pf_counters_inc(int action, struct pf_pdesc *pd, 9999 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 10000 { 10001 struct pf_krule *tr; 10002 int dir = pd->dir; 10003 int dirndx; 10004 10005 pf_counter_u64_critical_enter(); 10006 pf_counter_u64_add_protected( 10007 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10008 pd->tot_len); 10009 pf_counter_u64_add_protected( 10010 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10011 1); 10012 10013 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 10014 dirndx = (dir == PF_OUT); 10015 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 10016 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 10017 pf_update_timestamp(r); 10018 10019 if (a != NULL) { 10020 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 10021 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 10022 } 10023 if (s != NULL) { 10024 struct pf_krule_item *ri; 10025 10026 if (s->nat_rule != NULL) { 10027 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 10028 1); 10029 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 10030 pd->tot_len); 10031 } 10032 /* 10033 * Source nodes are accessed unlocked here. 10034 * But since we are operating with stateful tracking 10035 * and the state is locked, those SNs could not have 10036 * been freed. 10037 */ 10038 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 10039 if (s->sns[sn_type] != NULL) { 10040 counter_u64_add( 10041 s->sns[sn_type]->packets[dirndx], 10042 1); 10043 counter_u64_add( 10044 s->sns[sn_type]->bytes[dirndx], 10045 pd->tot_len); 10046 } 10047 } 10048 dirndx = (dir == s->direction) ? 0 : 1; 10049 s->packets[dirndx]++; 10050 s->bytes[dirndx] += pd->tot_len; 10051 10052 SLIST_FOREACH(ri, &s->match_rules, entry) { 10053 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 10054 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 10055 10056 if (ri->r->src.addr.type == PF_ADDR_TABLE) 10057 pfr_update_stats(ri->r->src.addr.p.tbl, 10058 (s == NULL) ? pd->src : 10059 &s->key[(s->direction == PF_IN)]-> 10060 addr[(s->direction == PF_OUT)], 10061 pd->af, pd->tot_len, dir == PF_OUT, 10062 r->action == PF_PASS, ri->r->src.neg); 10063 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 10064 pfr_update_stats(ri->r->dst.addr.p.tbl, 10065 (s == NULL) ? pd->dst : 10066 &s->key[(s->direction == PF_IN)]-> 10067 addr[(s->direction == PF_IN)], 10068 pd->af, pd->tot_len, dir == PF_OUT, 10069 r->action == PF_PASS, ri->r->dst.neg); 10070 } 10071 } 10072 10073 tr = r; 10074 if (s != NULL && s->nat_rule != NULL && 10075 r == &V_pf_default_rule) 10076 tr = s->nat_rule; 10077 10078 if (tr->src.addr.type == PF_ADDR_TABLE) 10079 pfr_update_stats(tr->src.addr.p.tbl, 10080 (s == NULL) ? pd->src : 10081 &s->key[(s->direction == PF_IN)]-> 10082 addr[(s->direction == PF_OUT)], 10083 pd->af, pd->tot_len, dir == PF_OUT, 10084 r->action == PF_PASS, tr->src.neg); 10085 if (tr->dst.addr.type == PF_ADDR_TABLE) 10086 pfr_update_stats(tr->dst.addr.p.tbl, 10087 (s == NULL) ? pd->dst : 10088 &s->key[(s->direction == PF_IN)]-> 10089 addr[(s->direction == PF_IN)], 10090 pd->af, pd->tot_len, dir == PF_OUT, 10091 r->action == PF_PASS, tr->dst.neg); 10092 } 10093 pf_counter_u64_critical_exit(); 10094 } 10095 10096 #if defined(INET) || defined(INET6) 10097 int 10098 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10099 struct inpcb *inp, struct pf_rule_actions *default_actions) 10100 { 10101 struct pfi_kkif *kif; 10102 u_short action, reason = 0; 10103 struct m_tag *mtag; 10104 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10105 struct pf_kstate *s = NULL; 10106 struct pf_kruleset *ruleset = NULL; 10107 struct pf_pdesc pd; 10108 int use_2nd_queue = 0; 10109 uint16_t tag; 10110 10111 PF_RULES_RLOCK_TRACKER; 10112 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10113 M_ASSERTPKTHDR(*m0); 10114 10115 if (!V_pf_status.running) 10116 return (PF_PASS); 10117 10118 PF_RULES_RLOCK(); 10119 10120 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10121 10122 if (__predict_false(kif == NULL)) { 10123 DPFPRINTF(PF_DEBUG_URGENT, 10124 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 10125 PF_RULES_RUNLOCK(); 10126 return (PF_DROP); 10127 } 10128 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10129 PF_RULES_RUNLOCK(); 10130 return (PF_PASS); 10131 } 10132 10133 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10134 PF_RULES_RUNLOCK(); 10135 return (PF_PASS); 10136 } 10137 10138 if (__predict_false(! M_WRITABLE(*m0))) { 10139 *m0 = m_unshare(*m0, M_NOWAIT); 10140 if (*m0 == NULL) { 10141 PF_RULES_RUNLOCK(); 10142 return (PF_DROP); 10143 } 10144 } 10145 10146 pf_init_pdesc(&pd, *m0); 10147 10148 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10149 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10150 10151 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10152 pd.pf_mtag->if_idxgen); 10153 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10154 PF_RULES_RUNLOCK(); 10155 m_freem(*m0); 10156 *m0 = NULL; 10157 return (PF_PASS); 10158 } 10159 PF_RULES_RUNLOCK(); 10160 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10161 *m0 = NULL; 10162 return (PF_PASS); 10163 } 10164 10165 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10166 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10167 /* Dummynet re-injects packets after they've 10168 * completed their delay. We've already 10169 * processed them, so pass unconditionally. */ 10170 10171 /* But only once. We may see the packet multiple times (e.g. 10172 * PFIL_IN/PFIL_OUT). */ 10173 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 10174 PF_RULES_RUNLOCK(); 10175 10176 return (PF_PASS); 10177 } 10178 10179 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 10180 kif, default_actions) == -1) { 10181 if (action != PF_PASS) 10182 pd.act.log |= PF_LOG_FORCE; 10183 goto done; 10184 } 10185 10186 #ifdef INET 10187 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 10188 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 10189 PF_RULES_RUNLOCK(); 10190 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 10191 0, ifp->if_mtu); 10192 *m0 = NULL; 10193 return (PF_DROP); 10194 } 10195 #endif 10196 #ifdef INET6 10197 /* 10198 * If we end up changing IP addresses (e.g. binat) the stack may get 10199 * confused and fail to send the icmp6 packet too big error. Just send 10200 * it here, before we do any NAT. 10201 */ 10202 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 10203 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 10204 PF_RULES_RUNLOCK(); 10205 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 10206 *m0 = NULL; 10207 return (PF_DROP); 10208 } 10209 #endif 10210 10211 if (__predict_false(ip_divert_ptr != NULL) && 10212 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 10213 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 10214 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 10215 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 10216 if (pd.pf_mtag == NULL && 10217 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10218 action = PF_DROP; 10219 goto done; 10220 } 10221 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 10222 } 10223 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 10224 pd.m->m_flags |= M_FASTFWD_OURS; 10225 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10226 } 10227 m_tag_delete(pd.m, mtag); 10228 10229 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 10230 if (mtag != NULL) 10231 m_tag_delete(pd.m, mtag); 10232 } 10233 10234 switch (pd.virtual_proto) { 10235 case PF_VPROTO_FRAGMENT: 10236 /* 10237 * handle fragments that aren't reassembled by 10238 * normalization 10239 */ 10240 if (kif == NULL || r == NULL) /* pflog */ 10241 action = PF_DROP; 10242 else 10243 action = pf_test_rule(&r, &s, &pd, &a, 10244 &ruleset, inp); 10245 if (action != PF_PASS) 10246 REASON_SET(&reason, PFRES_FRAG); 10247 break; 10248 10249 case IPPROTO_TCP: { 10250 /* Respond to SYN with a syncookie. */ 10251 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 10252 pd.dir == PF_IN && pf_synflood_check(&pd)) { 10253 pf_syncookie_send(&pd); 10254 action = PF_DROP; 10255 break; 10256 } 10257 10258 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 10259 use_2nd_queue = 1; 10260 action = pf_normalize_tcp(&pd); 10261 if (action == PF_DROP) 10262 goto done; 10263 action = pf_test_state(&s, &pd, &reason); 10264 if (action == PF_PASS || action == PF_AFRT) { 10265 if (V_pfsync_update_state_ptr != NULL) 10266 V_pfsync_update_state_ptr(s); 10267 r = s->rule; 10268 a = s->anchor; 10269 } else if (s == NULL) { 10270 /* Validate remote SYN|ACK, re-create original SYN if 10271 * valid. */ 10272 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 10273 TH_ACK && pf_syncookie_validate(&pd) && 10274 pd.dir == PF_IN) { 10275 struct mbuf *msyn; 10276 10277 msyn = pf_syncookie_recreate_syn(&pd); 10278 if (msyn == NULL) { 10279 action = PF_DROP; 10280 break; 10281 } 10282 10283 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 10284 &pd.act); 10285 m_freem(msyn); 10286 if (action != PF_PASS) 10287 break; 10288 10289 action = pf_test_state(&s, &pd, &reason); 10290 if (action != PF_PASS || s == NULL) { 10291 action = PF_DROP; 10292 break; 10293 } 10294 10295 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 10296 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 10297 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 10298 action = pf_synproxy(&pd, &s, &reason); 10299 break; 10300 } else { 10301 action = pf_test_rule(&r, &s, &pd, 10302 &a, &ruleset, inp); 10303 } 10304 } 10305 break; 10306 } 10307 10308 case IPPROTO_SCTP: 10309 action = pf_normalize_sctp(&pd); 10310 if (action == PF_DROP) 10311 goto done; 10312 /* fallthrough */ 10313 case IPPROTO_UDP: 10314 default: 10315 action = pf_test_state(&s, &pd, &reason); 10316 if (action == PF_PASS || action == PF_AFRT) { 10317 if (V_pfsync_update_state_ptr != NULL) 10318 V_pfsync_update_state_ptr(s); 10319 r = s->rule; 10320 a = s->anchor; 10321 } else if (s == NULL) { 10322 action = pf_test_rule(&r, &s, 10323 &pd, &a, &ruleset, inp); 10324 } 10325 break; 10326 10327 case IPPROTO_ICMP: 10328 case IPPROTO_ICMPV6: { 10329 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 10330 action = PF_DROP; 10331 REASON_SET(&reason, PFRES_NORM); 10332 DPFPRINTF(PF_DEBUG_MISC, 10333 ("dropping IPv6 packet with ICMPv4 payload")); 10334 goto done; 10335 } 10336 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 10337 action = PF_DROP; 10338 REASON_SET(&reason, PFRES_NORM); 10339 DPFPRINTF(PF_DEBUG_MISC, 10340 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 10341 goto done; 10342 } 10343 action = pf_test_state_icmp(&s, &pd, &reason); 10344 if (action == PF_PASS || action == PF_AFRT) { 10345 if (V_pfsync_update_state_ptr != NULL) 10346 V_pfsync_update_state_ptr(s); 10347 r = s->rule; 10348 a = s->anchor; 10349 } else if (s == NULL) 10350 action = pf_test_rule(&r, &s, &pd, 10351 &a, &ruleset, inp); 10352 break; 10353 } 10354 10355 } 10356 10357 done: 10358 PF_RULES_RUNLOCK(); 10359 10360 if (pd.m == NULL) 10361 goto eat_pkt; 10362 10363 if (action == PF_PASS && pd.badopts && 10364 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 10365 action = PF_DROP; 10366 REASON_SET(&reason, PFRES_IPOPTIONS); 10367 pd.act.log = PF_LOG_FORCE; 10368 DPFPRINTF(PF_DEBUG_MISC, 10369 ("pf: dropping packet with dangerous headers\n")); 10370 } 10371 10372 if (s) { 10373 uint8_t log = pd.act.log; 10374 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 10375 pd.act.log |= log; 10376 tag = s->tag; 10377 } else { 10378 tag = r->tag; 10379 } 10380 10381 if (tag > 0 && pf_tag_packet(&pd, tag)) { 10382 action = PF_DROP; 10383 REASON_SET(&reason, PFRES_MEMORY); 10384 } 10385 10386 pf_scrub(&pd); 10387 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 10388 pf_normalize_mss(&pd); 10389 10390 if (pd.act.rtableid >= 0) 10391 M_SETFIB(pd.m, pd.act.rtableid); 10392 10393 if (pd.act.flags & PFSTATE_SETPRIO) { 10394 if (pd.tos & IPTOS_LOWDELAY) 10395 use_2nd_queue = 1; 10396 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 10397 action = PF_DROP; 10398 REASON_SET(&reason, PFRES_MEMORY); 10399 pd.act.log = PF_LOG_FORCE; 10400 DPFPRINTF(PF_DEBUG_MISC, 10401 ("pf: failed to allocate 802.1q mtag\n")); 10402 } 10403 } 10404 10405 #ifdef ALTQ 10406 if (action == PF_PASS && pd.act.qid) { 10407 if (pd.pf_mtag == NULL && 10408 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10409 action = PF_DROP; 10410 REASON_SET(&reason, PFRES_MEMORY); 10411 } else { 10412 if (s != NULL) 10413 pd.pf_mtag->qid_hash = pf_state_hash(s); 10414 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 10415 pd.pf_mtag->qid = pd.act.pqid; 10416 else 10417 pd.pf_mtag->qid = pd.act.qid; 10418 /* Add hints for ecn. */ 10419 pd.pf_mtag->hdr = mtod(pd.m, void *); 10420 } 10421 } 10422 #endif /* ALTQ */ 10423 10424 /* 10425 * connections redirected to loopback should not match sockets 10426 * bound specifically to loopback due to security implications, 10427 * see tcp_input() and in_pcblookup_listen(). 10428 */ 10429 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 10430 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 10431 (s->nat_rule->action == PF_RDR || 10432 s->nat_rule->action == PF_BINAT) && 10433 pf_is_loopback(af, pd.dst)) 10434 pd.m->m_flags |= M_SKIP_FIREWALL; 10435 10436 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 10437 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 10438 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 10439 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 10440 if (mtag != NULL) { 10441 ((struct pf_divert_mtag *)(mtag+1))->port = 10442 ntohs(r->divert.port); 10443 ((struct pf_divert_mtag *)(mtag+1))->idir = 10444 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 10445 PF_DIVERT_MTAG_DIR_OUT; 10446 10447 if (s) 10448 PF_STATE_UNLOCK(s); 10449 10450 m_tag_prepend(pd.m, mtag); 10451 if (pd.m->m_flags & M_FASTFWD_OURS) { 10452 if (pd.pf_mtag == NULL && 10453 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10454 action = PF_DROP; 10455 REASON_SET(&reason, PFRES_MEMORY); 10456 pd.act.log = PF_LOG_FORCE; 10457 DPFPRINTF(PF_DEBUG_MISC, 10458 ("pf: failed to allocate tag\n")); 10459 } else { 10460 pd.pf_mtag->flags |= 10461 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10462 pd.m->m_flags &= ~M_FASTFWD_OURS; 10463 } 10464 } 10465 ip_divert_ptr(*m0, dir == PF_IN); 10466 *m0 = NULL; 10467 10468 return (action); 10469 } else { 10470 /* XXX: ipfw has the same behaviour! */ 10471 action = PF_DROP; 10472 REASON_SET(&reason, PFRES_MEMORY); 10473 pd.act.log = PF_LOG_FORCE; 10474 DPFPRINTF(PF_DEBUG_MISC, 10475 ("pf: failed to allocate divert tag\n")); 10476 } 10477 } 10478 /* XXX: Anybody working on it?! */ 10479 if (af == AF_INET6 && r->divert.port) 10480 printf("pf: divert(9) is not supported for IPv6\n"); 10481 10482 /* this flag will need revising if the pkt is forwarded */ 10483 if (pd.pf_mtag) 10484 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 10485 10486 if (pd.act.log) { 10487 struct pf_krule *lr; 10488 struct pf_krule_item *ri; 10489 10490 if (s != NULL && s->nat_rule != NULL && 10491 s->nat_rule->log & PF_LOG_ALL) 10492 lr = s->nat_rule; 10493 else 10494 lr = r; 10495 10496 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 10497 PFLOG_PACKET(action, reason, lr, a, 10498 ruleset, &pd, (s == NULL)); 10499 if (s) { 10500 SLIST_FOREACH(ri, &s->match_rules, entry) 10501 if (ri->r->log & PF_LOG_ALL) 10502 PFLOG_PACKET(action, 10503 reason, ri->r, a, ruleset, &pd, 0); 10504 } 10505 } 10506 10507 pf_counters_inc(action, &pd, s, r, a); 10508 10509 switch (action) { 10510 case PF_SYNPROXY_DROP: 10511 m_freem(*m0); 10512 case PF_DEFER: 10513 *m0 = NULL; 10514 action = PF_PASS; 10515 break; 10516 case PF_DROP: 10517 m_freem(*m0); 10518 *m0 = NULL; 10519 break; 10520 case PF_AFRT: 10521 if (pf_translate_af(&pd)) { 10522 if (!pd.m) 10523 *m0 = NULL; 10524 action = PF_DROP; 10525 break; 10526 } 10527 *m0 = pd.m; /* pf_translate_af may change pd.m */ 10528 #ifdef INET 10529 if (pd.naf == AF_INET) 10530 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10531 #endif 10532 #ifdef INET6 10533 if (pd.naf == AF_INET6) 10534 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10535 #endif 10536 *m0 = NULL; 10537 action = PF_PASS; 10538 goto out; 10539 break; 10540 default: 10541 if (pd.act.rt) { 10542 switch (af) { 10543 #ifdef INET 10544 case AF_INET: 10545 /* pf_route() returns unlocked. */ 10546 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10547 break; 10548 #endif 10549 #ifdef INET6 10550 case AF_INET6: 10551 /* pf_route6() returns unlocked. */ 10552 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10553 break; 10554 #endif 10555 } 10556 goto out; 10557 } 10558 if (pf_dummynet(&pd, s, r, m0) != 0) { 10559 action = PF_DROP; 10560 REASON_SET(&reason, PFRES_MEMORY); 10561 } 10562 break; 10563 } 10564 10565 eat_pkt: 10566 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 10567 10568 if (s && action != PF_DROP) { 10569 if (!s->if_index_in && dir == PF_IN) 10570 s->if_index_in = ifp->if_index; 10571 else if (!s->if_index_out && dir == PF_OUT) 10572 s->if_index_out = ifp->if_index; 10573 } 10574 10575 if (s) 10576 PF_STATE_UNLOCK(s); 10577 10578 out: 10579 #ifdef INET6 10580 /* If reassembled packet passed, create new fragments. */ 10581 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 10582 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 10583 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 10584 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 10585 #endif 10586 10587 pf_sctp_multihome_delayed(&pd, kif, s, action); 10588 10589 return (action); 10590 } 10591 #endif /* INET || INET6 */ 10592