1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 125 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 126 "struct pf_kstate *"); 127 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 128 "struct pfi_kkif *"); 129 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 130 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 131 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 132 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 133 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 134 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 135 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 136 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 137 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 138 "struct pf_krule *", "struct mbuf *", "int"); 139 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 140 "struct pf_sctp_source *"); 141 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 142 "struct pf_kstate *", "struct pf_sctp_source *"); 143 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 144 "int", "struct pf_pdesc *", "int"); 145 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 146 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 147 "int"); 148 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 149 "int"); 150 151 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 152 "struct mbuf *"); 153 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 154 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 155 "int", "struct pf_keth_rule *", "char *"); 156 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 157 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 158 "int", "struct pf_keth_rule *"); 159 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 160 161 /* 162 * Global variables 163 */ 164 165 /* state tables */ 166 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 167 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 168 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 169 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 170 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 172 VNET_DEFINE(struct pf_kstatus, pf_status); 173 174 VNET_DEFINE(u_int32_t, ticket_altqs_active); 175 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 176 VNET_DEFINE(int, altqs_inactive_open); 177 VNET_DEFINE(u_int32_t, ticket_pabuf); 178 179 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 180 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 181 VNET_DEFINE(u_char, pf_tcp_secret[16]); 182 #define V_pf_tcp_secret VNET(pf_tcp_secret) 183 VNET_DEFINE(int, pf_tcp_secret_init); 184 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 185 VNET_DEFINE(int, pf_tcp_iss_off); 186 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 187 VNET_DECLARE(int, pf_vnet_active); 188 #define V_pf_vnet_active VNET(pf_vnet_active) 189 190 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 191 #define V_pf_purge_idx VNET(pf_purge_idx) 192 193 #ifdef PF_WANT_32_TO_64_COUNTER 194 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 195 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 196 197 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 198 VNET_DEFINE(size_t, pf_allrulecount); 199 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 200 #endif 201 202 struct pf_sctp_endpoint; 203 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 204 struct pf_sctp_source { 205 sa_family_t af; 206 struct pf_addr addr; 207 TAILQ_ENTRY(pf_sctp_source) entry; 208 }; 209 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 210 struct pf_sctp_endpoint 211 { 212 uint32_t v_tag; 213 struct pf_sctp_sources sources; 214 RB_ENTRY(pf_sctp_endpoint) entry; 215 }; 216 static int 217 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 218 { 219 return (a->v_tag - b->v_tag); 220 } 221 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 222 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 223 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 224 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 225 static struct mtx_padalign pf_sctp_endpoints_mtx; 226 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 227 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 228 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 229 230 /* 231 * Queue for pf_intr() sends. 232 */ 233 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 234 struct pf_send_entry { 235 STAILQ_ENTRY(pf_send_entry) pfse_next; 236 struct mbuf *pfse_m; 237 enum { 238 PFSE_IP, 239 PFSE_IP6, 240 PFSE_ICMP, 241 PFSE_ICMP6, 242 } pfse_type; 243 struct { 244 int type; 245 int code; 246 int mtu; 247 } icmpopts; 248 }; 249 250 STAILQ_HEAD(pf_send_head, pf_send_entry); 251 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 252 #define V_pf_sendqueue VNET(pf_sendqueue) 253 254 static struct mtx_padalign pf_sendqueue_mtx; 255 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 256 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 257 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 258 259 /* 260 * Queue for pf_overload_task() tasks. 261 */ 262 struct pf_overload_entry { 263 SLIST_ENTRY(pf_overload_entry) next; 264 struct pf_addr addr; 265 sa_family_t af; 266 uint8_t dir; 267 struct pf_krule *rule; 268 }; 269 270 SLIST_HEAD(pf_overload_head, pf_overload_entry); 271 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 272 #define V_pf_overloadqueue VNET(pf_overloadqueue) 273 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 274 #define V_pf_overloadtask VNET(pf_overloadtask) 275 276 static struct mtx_padalign pf_overloadqueue_mtx; 277 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 278 "pf overload/flush queue", MTX_DEF); 279 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 280 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 281 282 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 283 struct mtx_padalign pf_unlnkdrules_mtx; 284 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 285 MTX_DEF); 286 287 struct sx pf_config_lock; 288 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 289 290 struct mtx_padalign pf_table_stats_lock; 291 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 292 MTX_DEF); 293 294 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 295 #define V_pf_sources_z VNET(pf_sources_z) 296 uma_zone_t pf_mtag_z; 297 VNET_DEFINE(uma_zone_t, pf_state_z); 298 VNET_DEFINE(uma_zone_t, pf_state_key_z); 299 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 300 301 VNET_DEFINE(struct unrhdr64, pf_stateid); 302 303 static void pf_src_tree_remove_state(struct pf_kstate *); 304 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 305 u_int32_t); 306 static void pf_add_threshold(struct pf_threshold *); 307 static int pf_check_threshold(struct pf_threshold *); 308 309 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 310 u_int16_t *, u_int16_t *, struct pf_addr *, 311 u_int16_t, u_int8_t, sa_family_t, sa_family_t); 312 static int pf_modulate_sack(struct pf_pdesc *, 313 struct tcphdr *, struct pf_state_peer *); 314 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 315 int *, u_int16_t *, u_int16_t *); 316 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 317 struct pf_addr *, struct pf_addr *, u_int16_t, 318 u_int16_t *, u_int16_t *, u_int16_t *, 319 u_int16_t *, u_int8_t, sa_family_t); 320 int pf_change_icmp_af(struct mbuf *, int, 321 struct pf_pdesc *, struct pf_pdesc *, 322 struct pf_addr *, struct pf_addr *, sa_family_t, 323 sa_family_t); 324 int pf_translate_icmp_af(int, void *); 325 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 326 sa_family_t, struct pf_krule *, int); 327 static void pf_detach_state(struct pf_kstate *); 328 static int pf_state_key_attach(struct pf_state_key *, 329 struct pf_state_key *, struct pf_kstate *); 330 static void pf_state_key_detach(struct pf_kstate *, int); 331 static int pf_state_key_ctor(void *, int, void *, int); 332 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 333 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 334 struct pf_mtag *pf_mtag); 335 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 336 struct pf_krule *, struct mbuf **); 337 static int pf_dummynet_route(struct pf_pdesc *, 338 struct pf_kstate *, struct pf_krule *, 339 struct ifnet *, struct sockaddr *, struct mbuf **); 340 static int pf_test_eth_rule(int, struct pfi_kkif *, 341 struct mbuf **); 342 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 343 struct pf_pdesc *, struct pf_krule **, 344 struct pf_kruleset **, struct inpcb *); 345 static int pf_create_state(struct pf_krule *, struct pf_krule *, 346 struct pf_krule *, struct pf_pdesc *, 347 struct pf_state_key *, struct pf_state_key *, int *, 348 struct pf_kstate **, int, u_int16_t, u_int16_t, 349 struct pf_krule_slist *, struct pf_udp_mapping *); 350 static int pf_state_key_addr_setup(struct pf_pdesc *, 351 struct pf_state_key_cmp *, int); 352 static int pf_tcp_track_full(struct pf_kstate **, 353 struct pf_pdesc *, u_short *, int *); 354 static int pf_tcp_track_sloppy(struct pf_kstate **, 355 struct pf_pdesc *, u_short *); 356 static int pf_test_state_tcp(struct pf_kstate **, 357 struct pf_pdesc *, u_short *); 358 static int pf_test_state_udp(struct pf_kstate **, 359 struct pf_pdesc *); 360 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 361 struct pf_pdesc *, struct pf_kstate **, 362 int, u_int16_t, u_int16_t, 363 int, int *, int, int); 364 static int pf_test_state_icmp(struct pf_kstate **, 365 struct pf_pdesc *, u_short *); 366 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 367 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 368 struct pfi_kkif *, struct pf_kstate *, int); 369 static int pf_test_state_sctp(struct pf_kstate **, 370 struct pf_pdesc *, u_short *); 371 static int pf_test_state_other(struct pf_kstate **, 372 struct pf_pdesc *); 373 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 374 int, u_int16_t); 375 static int pf_check_proto_cksum(struct mbuf *, int, int, 376 u_int8_t, sa_family_t); 377 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 378 int, int, u_short *); 379 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 380 u_short *); 381 static void pf_print_state_parts(struct pf_kstate *, 382 struct pf_state_key *, struct pf_state_key *); 383 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 384 bool, u_int8_t); 385 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 386 const struct pf_state_key_cmp *, u_int); 387 static bool pf_src_connlimit(struct pf_kstate *); 388 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 389 static void pf_counters_inc(int, struct pf_pdesc *, 390 struct pf_kstate *, struct pf_krule *, 391 struct pf_krule *); 392 static void pf_overload_task(void *v, int pending); 393 static u_short pf_insert_src_node(struct pf_ksrc_node **, 394 struct pf_srchash **, struct pf_krule *, 395 struct pf_addr *, sa_family_t, struct pf_addr *, 396 struct pfi_kkif *); 397 static u_int pf_purge_expired_states(u_int, int); 398 static void pf_purge_unlinked_rules(void); 399 static int pf_mtag_uminit(void *, int, int); 400 static void pf_mtag_free(struct m_tag *); 401 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 402 int, struct pf_state_key *); 403 #ifdef INET 404 static void pf_route(struct mbuf **, struct pf_krule *, 405 struct ifnet *, struct pf_kstate *, 406 struct pf_pdesc *, struct inpcb *); 407 #endif /* INET */ 408 #ifdef INET6 409 static void pf_change_a6(struct pf_addr *, u_int16_t *, 410 struct pf_addr *, u_int8_t); 411 static void pf_route6(struct mbuf **, struct pf_krule *, 412 struct ifnet *, struct pf_kstate *, 413 struct pf_pdesc *, struct inpcb *); 414 #endif /* INET6 */ 415 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 416 417 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 418 419 extern int pf_end_threads; 420 extern struct proc *pf_purge_proc; 421 422 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 423 424 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 425 426 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 427 do { \ 428 struct pf_state_key *nk; \ 429 if ((pd->dir) == PF_OUT) \ 430 nk = (_s)->key[PF_SK_STACK]; \ 431 else \ 432 nk = (_s)->key[PF_SK_WIRE]; \ 433 pf_packet_rework_nat(_m, _pd, _off, nk); \ 434 } while (0) 435 436 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 437 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 438 439 #define STATE_LOOKUP(k, s, pd) \ 440 do { \ 441 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 442 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 443 if ((s) == NULL) \ 444 return (PF_DROP); \ 445 if (PACKET_LOOPED(pd)) \ 446 return (PF_PASS); \ 447 } while (0) 448 449 static struct pfi_kkif * 450 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 451 { 452 struct pfi_kkif *k = pd->kif; 453 454 SDT_PROBE2(pf, ip, , bound_iface, st, k); 455 456 /* Floating unless otherwise specified. */ 457 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 458 return (V_pfi_all); 459 460 /* 461 * Initially set to all, because we don't know what interface we'll be 462 * sending this out when we create the state. 463 */ 464 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf)) 465 return (V_pfi_all); 466 467 /* 468 * If this state is created based on another state (e.g. SCTP 469 * multihome) always set it floating initially. We can't know for sure 470 * what interface the actual traffic for this state will come in on. 471 */ 472 if (pd->related_rule) 473 return (V_pfi_all); 474 475 /* Don't overrule the interface for states created on incoming packets. */ 476 if (st->direction == PF_IN) 477 return (k); 478 479 /* No route-to, so don't overrule. */ 480 if (st->act.rt != PF_ROUTETO) 481 return (k); 482 483 /* Bind to the route-to interface. */ 484 return (st->act.rt_kif); 485 } 486 487 #define STATE_INC_COUNTERS(s) \ 488 do { \ 489 struct pf_krule_item *mrm; \ 490 counter_u64_add(s->rule->states_cur, 1); \ 491 counter_u64_add(s->rule->states_tot, 1); \ 492 if (s->anchor != NULL) { \ 493 counter_u64_add(s->anchor->states_cur, 1); \ 494 counter_u64_add(s->anchor->states_tot, 1); \ 495 } \ 496 if (s->nat_rule != NULL) { \ 497 counter_u64_add(s->nat_rule->states_cur, 1);\ 498 counter_u64_add(s->nat_rule->states_tot, 1);\ 499 } \ 500 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 501 counter_u64_add(mrm->r->states_cur, 1); \ 502 counter_u64_add(mrm->r->states_tot, 1); \ 503 } \ 504 } while (0) 505 506 #define STATE_DEC_COUNTERS(s) \ 507 do { \ 508 struct pf_krule_item *mrm; \ 509 if (s->nat_rule != NULL) \ 510 counter_u64_add(s->nat_rule->states_cur, -1);\ 511 if (s->anchor != NULL) \ 512 counter_u64_add(s->anchor->states_cur, -1); \ 513 counter_u64_add(s->rule->states_cur, -1); \ 514 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 515 counter_u64_add(mrm->r->states_cur, -1); \ 516 } while (0) 517 518 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 519 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 520 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 521 VNET_DEFINE(struct pf_idhash *, pf_idhash); 522 VNET_DEFINE(struct pf_srchash *, pf_srchash); 523 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 524 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 525 526 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 527 "pf(4)"); 528 529 VNET_DEFINE(u_long, pf_hashmask); 530 VNET_DEFINE(u_long, pf_srchashmask); 531 VNET_DEFINE(u_long, pf_udpendpointhashmask); 532 VNET_DEFINE_STATIC(u_long, pf_hashsize); 533 #define V_pf_hashsize VNET(pf_hashsize) 534 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 535 #define V_pf_srchashsize VNET(pf_srchashsize) 536 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 537 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 538 u_long pf_ioctl_maxcount = 65535; 539 540 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 541 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 542 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 543 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 544 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 545 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 546 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 547 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 548 549 VNET_DEFINE(void *, pf_swi_cookie); 550 VNET_DEFINE(struct intr_event *, pf_swi_ie); 551 552 VNET_DEFINE(uint32_t, pf_hashseed); 553 #define V_pf_hashseed VNET(pf_hashseed) 554 555 static void 556 pf_sctp_checksum(struct mbuf *m, int off) 557 { 558 uint32_t sum = 0; 559 560 /* Zero out the checksum, to enable recalculation. */ 561 m_copyback(m, off + offsetof(struct sctphdr, checksum), 562 sizeof(sum), (caddr_t)&sum); 563 564 sum = sctp_calculate_cksum(m, off); 565 566 m_copyback(m, off + offsetof(struct sctphdr, checksum), 567 sizeof(sum), (caddr_t)&sum); 568 } 569 570 int 571 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 572 { 573 574 switch (af) { 575 #ifdef INET 576 case AF_INET: 577 if (a->addr32[0] > b->addr32[0]) 578 return (1); 579 if (a->addr32[0] < b->addr32[0]) 580 return (-1); 581 break; 582 #endif /* INET */ 583 #ifdef INET6 584 case AF_INET6: 585 if (a->addr32[3] > b->addr32[3]) 586 return (1); 587 if (a->addr32[3] < b->addr32[3]) 588 return (-1); 589 if (a->addr32[2] > b->addr32[2]) 590 return (1); 591 if (a->addr32[2] < b->addr32[2]) 592 return (-1); 593 if (a->addr32[1] > b->addr32[1]) 594 return (1); 595 if (a->addr32[1] < b->addr32[1]) 596 return (-1); 597 if (a->addr32[0] > b->addr32[0]) 598 return (1); 599 if (a->addr32[0] < b->addr32[0]) 600 return (-1); 601 break; 602 #endif /* INET6 */ 603 } 604 return (0); 605 } 606 607 static bool 608 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 609 { 610 switch (af) { 611 #ifdef INET 612 case AF_INET: 613 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 614 #endif 615 case AF_INET6: 616 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 617 default: 618 panic("Unknown af %d", af); 619 } 620 } 621 622 static void 623 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 624 struct pf_state_key *nk) 625 { 626 627 switch (pd->proto) { 628 case IPPROTO_TCP: { 629 struct tcphdr *th = &pd->hdr.tcp; 630 631 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 632 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 633 &th->th_sum, &nk->addr[pd->sidx], 634 nk->port[pd->sidx], 0, pd->af, pd->naf); 635 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 636 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 637 &th->th_sum, &nk->addr[pd->didx], 638 nk->port[pd->didx], 0, pd->af, pd->naf); 639 m_copyback(m, off, sizeof(*th), (caddr_t)th); 640 break; 641 } 642 case IPPROTO_UDP: { 643 struct udphdr *uh = &pd->hdr.udp; 644 645 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 646 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 647 &uh->uh_sum, &nk->addr[pd->sidx], 648 nk->port[pd->sidx], 1, pd->af, pd->naf); 649 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 650 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 651 &uh->uh_sum, &nk->addr[pd->didx], 652 nk->port[pd->didx], 1, pd->af, pd->naf); 653 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 654 break; 655 } 656 case IPPROTO_SCTP: { 657 struct sctphdr *sh = &pd->hdr.sctp; 658 uint16_t checksum = 0; 659 660 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 661 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 662 &checksum, &nk->addr[pd->sidx], 663 nk->port[pd->sidx], 1, pd->af, pd->naf); 664 } 665 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 666 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 667 &checksum, &nk->addr[pd->didx], 668 nk->port[pd->didx], 1, pd->af, pd->naf); 669 } 670 671 break; 672 } 673 case IPPROTO_ICMP: { 674 struct icmp *ih = &pd->hdr.icmp; 675 676 if (nk->port[pd->sidx] != ih->icmp_id) { 677 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 678 ih->icmp_cksum, ih->icmp_id, 679 nk->port[pd->sidx], 0); 680 ih->icmp_id = nk->port[pd->sidx]; 681 pd->sport = &ih->icmp_id; 682 683 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 684 } 685 /* FALLTHROUGH */ 686 } 687 default: 688 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 689 switch (pd->af) { 690 case AF_INET: 691 pf_change_a(&pd->src->v4.s_addr, 692 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 693 0); 694 break; 695 case AF_INET6: 696 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 697 break; 698 } 699 } 700 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 701 switch (pd->af) { 702 case AF_INET: 703 pf_change_a(&pd->dst->v4.s_addr, 704 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 705 0); 706 break; 707 case AF_INET6: 708 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 709 break; 710 } 711 } 712 break; 713 } 714 } 715 716 static __inline uint32_t 717 pf_hashkey(const struct pf_state_key *sk) 718 { 719 uint32_t h; 720 721 h = murmur3_32_hash32((const uint32_t *)sk, 722 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 723 V_pf_hashseed); 724 725 return (h & V_pf_hashmask); 726 } 727 728 __inline uint32_t 729 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 730 { 731 uint32_t h; 732 733 switch (af) { 734 case AF_INET: 735 h = murmur3_32_hash32((uint32_t *)&addr->v4, 736 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 737 break; 738 case AF_INET6: 739 h = murmur3_32_hash32((uint32_t *)&addr->v6, 740 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 741 break; 742 } 743 744 return (h & V_pf_srchashmask); 745 } 746 747 static inline uint32_t 748 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 749 { 750 uint32_t h; 751 752 h = murmur3_32_hash32((uint32_t *)endpoint, 753 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 754 V_pf_hashseed); 755 return (h & V_pf_udpendpointhashmask); 756 } 757 758 #ifdef ALTQ 759 static int 760 pf_state_hash(struct pf_kstate *s) 761 { 762 u_int32_t hv = (intptr_t)s / sizeof(*s); 763 764 hv ^= crc32(&s->src, sizeof(s->src)); 765 hv ^= crc32(&s->dst, sizeof(s->dst)); 766 if (hv == 0) 767 hv = 1; 768 return (hv); 769 } 770 #endif 771 772 static __inline void 773 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 774 { 775 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 776 s->dst.state = newstate; 777 if (which == PF_PEER_DST) 778 return; 779 if (s->src.state == newstate) 780 return; 781 if (s->creatorid == V_pf_status.hostid && 782 s->key[PF_SK_STACK] != NULL && 783 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 784 !(TCPS_HAVEESTABLISHED(s->src.state) || 785 s->src.state == TCPS_CLOSED) && 786 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 787 atomic_add_32(&V_pf_status.states_halfopen, -1); 788 789 s->src.state = newstate; 790 } 791 792 #ifdef INET6 793 void 794 pf_addrcpy(struct pf_addr *dst, const struct pf_addr *src, sa_family_t af) 795 { 796 switch (af) { 797 #ifdef INET 798 case AF_INET: 799 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 800 break; 801 #endif /* INET */ 802 case AF_INET6: 803 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 804 break; 805 } 806 } 807 #endif /* INET6 */ 808 809 static void 810 pf_init_threshold(struct pf_threshold *threshold, 811 u_int32_t limit, u_int32_t seconds) 812 { 813 threshold->limit = limit * PF_THRESHOLD_MULT; 814 threshold->seconds = seconds; 815 threshold->count = 0; 816 threshold->last = time_uptime; 817 } 818 819 static void 820 pf_add_threshold(struct pf_threshold *threshold) 821 { 822 u_int32_t t = time_uptime, diff = t - threshold->last; 823 824 if (diff >= threshold->seconds) 825 threshold->count = 0; 826 else 827 threshold->count -= threshold->count * diff / 828 threshold->seconds; 829 threshold->count += PF_THRESHOLD_MULT; 830 threshold->last = t; 831 } 832 833 static int 834 pf_check_threshold(struct pf_threshold *threshold) 835 { 836 return (threshold->count > threshold->limit); 837 } 838 839 static bool 840 pf_src_connlimit(struct pf_kstate *state) 841 { 842 struct pf_overload_entry *pfoe; 843 bool limited = false; 844 845 PF_STATE_LOCK_ASSERT(state); 846 PF_SRC_NODE_LOCK(state->src_node); 847 848 state->src_node->conn++; 849 state->src.tcp_est = 1; 850 pf_add_threshold(&state->src_node->conn_rate); 851 852 if (state->rule->max_src_conn && 853 state->rule->max_src_conn < 854 state->src_node->conn) { 855 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 856 limited = true; 857 } 858 859 if (state->rule->max_src_conn_rate.limit && 860 pf_check_threshold(&state->src_node->conn_rate)) { 861 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 862 limited = true; 863 } 864 865 if (!limited) 866 goto done; 867 868 /* Kill this state. */ 869 state->timeout = PFTM_PURGE; 870 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 871 872 if (state->rule->overload_tbl == NULL) 873 goto done; 874 875 /* Schedule overloading and flushing task. */ 876 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 877 if (pfoe == NULL) 878 goto done; /* too bad :( */ 879 880 bcopy(&state->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 881 pfoe->af = state->key[PF_SK_WIRE]->af; 882 pfoe->rule = state->rule; 883 pfoe->dir = state->direction; 884 PF_OVERLOADQ_LOCK(); 885 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 886 PF_OVERLOADQ_UNLOCK(); 887 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 888 889 done: 890 PF_SRC_NODE_UNLOCK(state->src_node); 891 return (limited); 892 } 893 894 static void 895 pf_overload_task(void *v, int pending) 896 { 897 struct pf_overload_head queue; 898 struct pfr_addr p; 899 struct pf_overload_entry *pfoe, *pfoe1; 900 uint32_t killed = 0; 901 902 CURVNET_SET((struct vnet *)v); 903 904 PF_OVERLOADQ_LOCK(); 905 queue = V_pf_overloadqueue; 906 SLIST_INIT(&V_pf_overloadqueue); 907 PF_OVERLOADQ_UNLOCK(); 908 909 bzero(&p, sizeof(p)); 910 SLIST_FOREACH(pfoe, &queue, next) { 911 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 912 if (V_pf_status.debug >= PF_DEBUG_MISC) { 913 printf("%s: blocking address ", __func__); 914 pf_print_host(&pfoe->addr, 0, pfoe->af); 915 printf("\n"); 916 } 917 918 p.pfra_af = pfoe->af; 919 switch (pfoe->af) { 920 #ifdef INET 921 case AF_INET: 922 p.pfra_net = 32; 923 p.pfra_ip4addr = pfoe->addr.v4; 924 break; 925 #endif 926 #ifdef INET6 927 case AF_INET6: 928 p.pfra_net = 128; 929 p.pfra_ip6addr = pfoe->addr.v6; 930 break; 931 #endif 932 } 933 934 PF_RULES_WLOCK(); 935 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 936 PF_RULES_WUNLOCK(); 937 } 938 939 /* 940 * Remove those entries, that don't need flushing. 941 */ 942 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 943 if (pfoe->rule->flush == 0) { 944 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 945 free(pfoe, M_PFTEMP); 946 } else 947 counter_u64_add( 948 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 949 950 /* If nothing to flush, return. */ 951 if (SLIST_EMPTY(&queue)) { 952 CURVNET_RESTORE(); 953 return; 954 } 955 956 for (int i = 0; i <= V_pf_hashmask; i++) { 957 struct pf_idhash *ih = &V_pf_idhash[i]; 958 struct pf_state_key *sk; 959 struct pf_kstate *s; 960 961 PF_HASHROW_LOCK(ih); 962 LIST_FOREACH(s, &ih->states, entry) { 963 sk = s->key[PF_SK_WIRE]; 964 SLIST_FOREACH(pfoe, &queue, next) 965 if (sk->af == pfoe->af && 966 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 967 pfoe->rule == s->rule) && 968 ((pfoe->dir == PF_OUT && 969 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 970 (pfoe->dir == PF_IN && 971 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 972 s->timeout = PFTM_PURGE; 973 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 974 killed++; 975 } 976 } 977 PF_HASHROW_UNLOCK(ih); 978 } 979 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 980 free(pfoe, M_PFTEMP); 981 if (V_pf_status.debug >= PF_DEBUG_MISC) 982 printf("%s: %u states killed", __func__, killed); 983 984 CURVNET_RESTORE(); 985 } 986 987 /* 988 * On node found always returns locked. On not found its configurable. 989 */ 990 struct pf_ksrc_node * 991 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 992 struct pf_srchash **sh, bool returnlocked) 993 { 994 struct pf_ksrc_node *n; 995 996 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 997 998 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 999 PF_HASHROW_LOCK(*sh); 1000 LIST_FOREACH(n, &(*sh)->nodes, entry) 1001 if (n->rule == rule && n->af == af && 1002 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 1003 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 1004 break; 1005 1006 if (n == NULL && !returnlocked) 1007 PF_HASHROW_UNLOCK(*sh); 1008 1009 return (n); 1010 } 1011 1012 bool 1013 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 1014 { 1015 struct pf_ksrc_node *cur; 1016 1017 if ((*sn) == NULL) 1018 return (false); 1019 1020 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 1021 1022 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1023 PF_HASHROW_LOCK(sh); 1024 LIST_FOREACH(cur, &(sh->nodes), entry) { 1025 if (cur == (*sn) && 1026 cur->expire != 1) /* Ignore nodes being killed */ 1027 return (true); 1028 } 1029 PF_HASHROW_UNLOCK(sh); 1030 (*sn) = NULL; 1031 return (false); 1032 } 1033 1034 static void 1035 pf_free_src_node(struct pf_ksrc_node *sn) 1036 { 1037 1038 for (int i = 0; i < 2; i++) { 1039 counter_u64_free(sn->bytes[i]); 1040 counter_u64_free(sn->packets[i]); 1041 } 1042 uma_zfree(V_pf_sources_z, sn); 1043 } 1044 1045 static u_short 1046 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_srchash **sh, 1047 struct pf_krule *rule, struct pf_addr *src, sa_family_t af, 1048 struct pf_addr *raddr, struct pfi_kkif *rkif) 1049 { 1050 u_short reason = 0; 1051 1052 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1053 rule->rdr.opts & PF_POOL_STICKYADDR), 1054 ("%s for non-tracking rule %p", __func__, rule)); 1055 1056 /* 1057 * Request the sh to always be locked, as we might insert a new sn. 1058 */ 1059 if (*sn == NULL) 1060 *sn = pf_find_src_node(src, rule, af, sh, true); 1061 1062 if (*sn == NULL) { 1063 PF_HASHROW_ASSERT(*sh); 1064 1065 if (rule->max_src_nodes && 1066 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1067 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1068 reason = PFRES_SRCLIMIT; 1069 goto done; 1070 } 1071 1072 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1073 if ((*sn) == NULL) { 1074 reason = PFRES_MEMORY; 1075 goto done; 1076 } 1077 1078 for (int i = 0; i < 2; i++) { 1079 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1080 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1081 1082 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1083 pf_free_src_node(*sn); 1084 reason = PFRES_MEMORY; 1085 goto done; 1086 } 1087 } 1088 1089 pf_init_threshold(&(*sn)->conn_rate, 1090 rule->max_src_conn_rate.limit, 1091 rule->max_src_conn_rate.seconds); 1092 1093 MPASS((*sn)->lock == NULL); 1094 (*sn)->lock = &(*sh)->lock; 1095 1096 (*sn)->af = af; 1097 (*sn)->rule = rule; 1098 PF_ACPY(&(*sn)->addr, src, af); 1099 PF_ACPY(&(*sn)->raddr, raddr, af); 1100 (*sn)->rkif = rkif; 1101 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1102 (*sn)->creation = time_uptime; 1103 (*sn)->ruletype = rule->action; 1104 if ((*sn)->rule != NULL) 1105 counter_u64_add((*sn)->rule->src_nodes, 1); 1106 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1107 } else { 1108 if (rule->max_src_states && 1109 (*sn)->states >= rule->max_src_states) { 1110 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1111 1); 1112 reason = PFRES_SRCLIMIT; 1113 goto done; 1114 } 1115 } 1116 done: 1117 if (reason == 0) 1118 (*sn)->states++; 1119 else 1120 (*sn) = NULL; 1121 1122 PF_HASHROW_UNLOCK(*sh); 1123 return (reason); 1124 } 1125 1126 void 1127 pf_unlink_src_node(struct pf_ksrc_node *src) 1128 { 1129 PF_SRC_NODE_LOCK_ASSERT(src); 1130 1131 LIST_REMOVE(src, entry); 1132 if (src->rule) 1133 counter_u64_add(src->rule->src_nodes, -1); 1134 } 1135 1136 u_int 1137 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1138 { 1139 struct pf_ksrc_node *sn, *tmp; 1140 u_int count = 0; 1141 1142 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1143 pf_free_src_node(sn); 1144 count++; 1145 } 1146 1147 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1148 1149 return (count); 1150 } 1151 1152 void 1153 pf_mtag_initialize(void) 1154 { 1155 1156 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1157 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1158 UMA_ALIGN_PTR, 0); 1159 } 1160 1161 /* Per-vnet data storage structures initialization. */ 1162 void 1163 pf_initialize(void) 1164 { 1165 struct pf_keyhash *kh; 1166 struct pf_idhash *ih; 1167 struct pf_srchash *sh; 1168 struct pf_udpendpointhash *uh; 1169 u_int i; 1170 1171 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1172 V_pf_hashsize = PF_HASHSIZ; 1173 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1174 V_pf_srchashsize = PF_SRCHASHSIZ; 1175 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1176 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1177 1178 V_pf_hashseed = arc4random(); 1179 1180 /* States and state keys storage. */ 1181 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1182 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1183 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1184 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1185 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1186 1187 V_pf_state_key_z = uma_zcreate("pf state keys", 1188 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1189 UMA_ALIGN_PTR, 0); 1190 1191 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1192 M_PFHASH, M_NOWAIT | M_ZERO); 1193 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1194 M_PFHASH, M_NOWAIT | M_ZERO); 1195 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1196 printf("pf: Unable to allocate memory for " 1197 "state_hashsize %lu.\n", V_pf_hashsize); 1198 1199 free(V_pf_keyhash, M_PFHASH); 1200 free(V_pf_idhash, M_PFHASH); 1201 1202 V_pf_hashsize = PF_HASHSIZ; 1203 V_pf_keyhash = mallocarray(V_pf_hashsize, 1204 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1205 V_pf_idhash = mallocarray(V_pf_hashsize, 1206 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1207 } 1208 1209 V_pf_hashmask = V_pf_hashsize - 1; 1210 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1211 i++, kh++, ih++) { 1212 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1213 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1214 } 1215 1216 /* Source nodes. */ 1217 V_pf_sources_z = uma_zcreate("pf source nodes", 1218 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1219 0); 1220 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1221 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1222 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1223 1224 V_pf_srchash = mallocarray(V_pf_srchashsize, 1225 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1226 if (V_pf_srchash == NULL) { 1227 printf("pf: Unable to allocate memory for " 1228 "source_hashsize %lu.\n", V_pf_srchashsize); 1229 1230 V_pf_srchashsize = PF_SRCHASHSIZ; 1231 V_pf_srchash = mallocarray(V_pf_srchashsize, 1232 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1233 } 1234 1235 V_pf_srchashmask = V_pf_srchashsize - 1; 1236 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1237 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1238 1239 1240 /* UDP endpoint mappings. */ 1241 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1242 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1243 UMA_ALIGN_PTR, 0); 1244 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1245 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1246 if (V_pf_udpendpointhash == NULL) { 1247 printf("pf: Unable to allocate memory for " 1248 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1249 1250 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1251 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1252 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1253 } 1254 1255 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1256 for (i = 0, uh = V_pf_udpendpointhash; 1257 i <= V_pf_udpendpointhashmask; 1258 i++, uh++) { 1259 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1260 MTX_DEF | MTX_DUPOK); 1261 } 1262 1263 /* ALTQ */ 1264 TAILQ_INIT(&V_pf_altqs[0]); 1265 TAILQ_INIT(&V_pf_altqs[1]); 1266 TAILQ_INIT(&V_pf_altqs[2]); 1267 TAILQ_INIT(&V_pf_altqs[3]); 1268 TAILQ_INIT(&V_pf_pabuf[0]); 1269 TAILQ_INIT(&V_pf_pabuf[1]); 1270 TAILQ_INIT(&V_pf_pabuf[2]); 1271 V_pf_altqs_active = &V_pf_altqs[0]; 1272 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1273 V_pf_altqs_inactive = &V_pf_altqs[2]; 1274 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1275 1276 /* Send & overload+flush queues. */ 1277 STAILQ_INIT(&V_pf_sendqueue); 1278 SLIST_INIT(&V_pf_overloadqueue); 1279 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1280 1281 /* Unlinked, but may be referenced rules. */ 1282 TAILQ_INIT(&V_pf_unlinked_rules); 1283 } 1284 1285 void 1286 pf_mtag_cleanup(void) 1287 { 1288 1289 uma_zdestroy(pf_mtag_z); 1290 } 1291 1292 void 1293 pf_cleanup(void) 1294 { 1295 struct pf_keyhash *kh; 1296 struct pf_idhash *ih; 1297 struct pf_srchash *sh; 1298 struct pf_udpendpointhash *uh; 1299 struct pf_send_entry *pfse, *next; 1300 u_int i; 1301 1302 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1303 i <= V_pf_hashmask; 1304 i++, kh++, ih++) { 1305 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1306 __func__)); 1307 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1308 __func__)); 1309 mtx_destroy(&kh->lock); 1310 mtx_destroy(&ih->lock); 1311 } 1312 free(V_pf_keyhash, M_PFHASH); 1313 free(V_pf_idhash, M_PFHASH); 1314 1315 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1316 KASSERT(LIST_EMPTY(&sh->nodes), 1317 ("%s: source node hash not empty", __func__)); 1318 mtx_destroy(&sh->lock); 1319 } 1320 free(V_pf_srchash, M_PFHASH); 1321 1322 for (i = 0, uh = V_pf_udpendpointhash; 1323 i <= V_pf_udpendpointhashmask; 1324 i++, uh++) { 1325 KASSERT(LIST_EMPTY(&uh->endpoints), 1326 ("%s: udp endpoint hash not empty", __func__)); 1327 mtx_destroy(&uh->lock); 1328 } 1329 free(V_pf_udpendpointhash, M_PFHASH); 1330 1331 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1332 m_freem(pfse->pfse_m); 1333 free(pfse, M_PFTEMP); 1334 } 1335 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1336 1337 uma_zdestroy(V_pf_sources_z); 1338 uma_zdestroy(V_pf_state_z); 1339 uma_zdestroy(V_pf_state_key_z); 1340 uma_zdestroy(V_pf_udp_mapping_z); 1341 } 1342 1343 static int 1344 pf_mtag_uminit(void *mem, int size, int how) 1345 { 1346 struct m_tag *t; 1347 1348 t = (struct m_tag *)mem; 1349 t->m_tag_cookie = MTAG_ABI_COMPAT; 1350 t->m_tag_id = PACKET_TAG_PF; 1351 t->m_tag_len = sizeof(struct pf_mtag); 1352 t->m_tag_free = pf_mtag_free; 1353 1354 return (0); 1355 } 1356 1357 static void 1358 pf_mtag_free(struct m_tag *t) 1359 { 1360 1361 uma_zfree(pf_mtag_z, t); 1362 } 1363 1364 struct pf_mtag * 1365 pf_get_mtag(struct mbuf *m) 1366 { 1367 struct m_tag *mtag; 1368 1369 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1370 return ((struct pf_mtag *)(mtag + 1)); 1371 1372 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1373 if (mtag == NULL) 1374 return (NULL); 1375 bzero(mtag + 1, sizeof(struct pf_mtag)); 1376 m_tag_prepend(m, mtag); 1377 1378 return ((struct pf_mtag *)(mtag + 1)); 1379 } 1380 1381 static int 1382 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1383 struct pf_kstate *s) 1384 { 1385 struct pf_keyhash *khs, *khw, *kh; 1386 struct pf_state_key *sk, *cur; 1387 struct pf_kstate *si, *olds = NULL; 1388 int idx; 1389 1390 NET_EPOCH_ASSERT(); 1391 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1392 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1393 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1394 1395 /* 1396 * We need to lock hash slots of both keys. To avoid deadlock 1397 * we always lock the slot with lower address first. Unlock order 1398 * isn't important. 1399 * 1400 * We also need to lock ID hash slot before dropping key 1401 * locks. On success we return with ID hash slot locked. 1402 */ 1403 1404 if (skw == sks) { 1405 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1406 PF_HASHROW_LOCK(khs); 1407 } else { 1408 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1409 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1410 if (khs == khw) { 1411 PF_HASHROW_LOCK(khs); 1412 } else if (khs < khw) { 1413 PF_HASHROW_LOCK(khs); 1414 PF_HASHROW_LOCK(khw); 1415 } else { 1416 PF_HASHROW_LOCK(khw); 1417 PF_HASHROW_LOCK(khs); 1418 } 1419 } 1420 1421 #define KEYS_UNLOCK() do { \ 1422 if (khs != khw) { \ 1423 PF_HASHROW_UNLOCK(khs); \ 1424 PF_HASHROW_UNLOCK(khw); \ 1425 } else \ 1426 PF_HASHROW_UNLOCK(khs); \ 1427 } while (0) 1428 1429 /* 1430 * First run: start with wire key. 1431 */ 1432 sk = skw; 1433 kh = khw; 1434 idx = PF_SK_WIRE; 1435 1436 MPASS(s->lock == NULL); 1437 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1438 1439 keyattach: 1440 LIST_FOREACH(cur, &kh->keys, entry) 1441 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1442 break; 1443 1444 if (cur != NULL) { 1445 /* Key exists. Check for same kif, if none, add to key. */ 1446 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1447 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1448 1449 PF_HASHROW_LOCK(ih); 1450 if (si->kif == s->kif && 1451 ((si->key[PF_SK_WIRE]->af == sk->af && 1452 si->direction == s->direction) || 1453 (si->key[PF_SK_WIRE]->af != 1454 si->key[PF_SK_STACK]->af && 1455 sk->af == si->key[PF_SK_STACK]->af && 1456 si->direction != s->direction))) { 1457 if (sk->proto == IPPROTO_TCP && 1458 si->src.state >= TCPS_FIN_WAIT_2 && 1459 si->dst.state >= TCPS_FIN_WAIT_2) { 1460 /* 1461 * New state matches an old >FIN_WAIT_2 1462 * state. We can't drop key hash locks, 1463 * thus we can't unlink it properly. 1464 * 1465 * As a workaround we drop it into 1466 * TCPS_CLOSED state, schedule purge 1467 * ASAP and push it into the very end 1468 * of the slot TAILQ, so that it won't 1469 * conflict with our new state. 1470 */ 1471 pf_set_protostate(si, PF_PEER_BOTH, 1472 TCPS_CLOSED); 1473 si->timeout = PFTM_PURGE; 1474 olds = si; 1475 } else { 1476 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1477 printf("pf: %s key attach " 1478 "failed on %s: ", 1479 (idx == PF_SK_WIRE) ? 1480 "wire" : "stack", 1481 s->kif->pfik_name); 1482 pf_print_state_parts(s, 1483 (idx == PF_SK_WIRE) ? 1484 sk : NULL, 1485 (idx == PF_SK_STACK) ? 1486 sk : NULL); 1487 printf(", existing: "); 1488 pf_print_state_parts(si, 1489 (idx == PF_SK_WIRE) ? 1490 sk : NULL, 1491 (idx == PF_SK_STACK) ? 1492 sk : NULL); 1493 printf("\n"); 1494 } 1495 s->timeout = PFTM_UNLINKED; 1496 PF_HASHROW_UNLOCK(ih); 1497 KEYS_UNLOCK(); 1498 if (idx == PF_SK_WIRE) { 1499 uma_zfree(V_pf_state_key_z, skw); 1500 if (skw != sks) 1501 uma_zfree(V_pf_state_key_z, sks); 1502 } else { 1503 pf_detach_state(s); 1504 } 1505 return (EEXIST); /* collision! */ 1506 } 1507 } 1508 PF_HASHROW_UNLOCK(ih); 1509 } 1510 uma_zfree(V_pf_state_key_z, sk); 1511 s->key[idx] = cur; 1512 } else { 1513 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1514 s->key[idx] = sk; 1515 } 1516 1517 stateattach: 1518 /* List is sorted, if-bound states before floating. */ 1519 if (s->kif == V_pfi_all) 1520 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1521 else 1522 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1523 1524 if (olds) { 1525 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1526 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1527 key_list[idx]); 1528 olds = NULL; 1529 } 1530 1531 /* 1532 * Attach done. See how should we (or should not?) 1533 * attach a second key. 1534 */ 1535 if (sks == skw) { 1536 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1537 idx = PF_SK_STACK; 1538 sks = NULL; 1539 goto stateattach; 1540 } else if (sks != NULL) { 1541 /* 1542 * Continue attaching with stack key. 1543 */ 1544 sk = sks; 1545 kh = khs; 1546 idx = PF_SK_STACK; 1547 sks = NULL; 1548 goto keyattach; 1549 } 1550 1551 PF_STATE_LOCK(s); 1552 KEYS_UNLOCK(); 1553 1554 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1555 ("%s failure", __func__)); 1556 1557 return (0); 1558 #undef KEYS_UNLOCK 1559 } 1560 1561 static void 1562 pf_detach_state(struct pf_kstate *s) 1563 { 1564 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1565 struct pf_keyhash *kh; 1566 1567 NET_EPOCH_ASSERT(); 1568 MPASS(s->timeout >= PFTM_MAX); 1569 1570 pf_sctp_multihome_detach_addr(s); 1571 1572 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1573 V_pflow_export_state_ptr(s); 1574 1575 if (sks != NULL) { 1576 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1577 PF_HASHROW_LOCK(kh); 1578 if (s->key[PF_SK_STACK] != NULL) 1579 pf_state_key_detach(s, PF_SK_STACK); 1580 /* 1581 * If both point to same key, then we are done. 1582 */ 1583 if (sks == s->key[PF_SK_WIRE]) { 1584 pf_state_key_detach(s, PF_SK_WIRE); 1585 PF_HASHROW_UNLOCK(kh); 1586 return; 1587 } 1588 PF_HASHROW_UNLOCK(kh); 1589 } 1590 1591 if (s->key[PF_SK_WIRE] != NULL) { 1592 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1593 PF_HASHROW_LOCK(kh); 1594 if (s->key[PF_SK_WIRE] != NULL) 1595 pf_state_key_detach(s, PF_SK_WIRE); 1596 PF_HASHROW_UNLOCK(kh); 1597 } 1598 } 1599 1600 static void 1601 pf_state_key_detach(struct pf_kstate *s, int idx) 1602 { 1603 struct pf_state_key *sk = s->key[idx]; 1604 #ifdef INVARIANTS 1605 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1606 1607 PF_HASHROW_ASSERT(kh); 1608 #endif 1609 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1610 s->key[idx] = NULL; 1611 1612 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1613 LIST_REMOVE(sk, entry); 1614 uma_zfree(V_pf_state_key_z, sk); 1615 } 1616 } 1617 1618 static int 1619 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1620 { 1621 struct pf_state_key *sk = mem; 1622 1623 bzero(sk, sizeof(struct pf_state_key_cmp)); 1624 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1625 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1626 1627 return (0); 1628 } 1629 1630 static int 1631 pf_state_key_addr_setup(struct pf_pdesc *pd, 1632 struct pf_state_key_cmp *key, int multi) 1633 { 1634 struct pf_addr *saddr = pd->src; 1635 struct pf_addr *daddr = pd->dst; 1636 #ifdef INET6 1637 struct nd_neighbor_solicit nd; 1638 struct pf_addr *target; 1639 u_short action, reason; 1640 1641 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1642 goto copy; 1643 1644 switch (pd->hdr.icmp6.icmp6_type) { 1645 case ND_NEIGHBOR_SOLICIT: 1646 if (multi) 1647 return (-1); 1648 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1649 return (-1); 1650 target = (struct pf_addr *)&nd.nd_ns_target; 1651 daddr = target; 1652 break; 1653 case ND_NEIGHBOR_ADVERT: 1654 if (multi) 1655 return (-1); 1656 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1657 return (-1); 1658 target = (struct pf_addr *)&nd.nd_ns_target; 1659 saddr = target; 1660 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1661 key->addr[pd->didx].addr32[0] = 0; 1662 key->addr[pd->didx].addr32[1] = 0; 1663 key->addr[pd->didx].addr32[2] = 0; 1664 key->addr[pd->didx].addr32[3] = 0; 1665 daddr = NULL; /* overwritten */ 1666 } 1667 break; 1668 default: 1669 if (multi == PF_ICMP_MULTI_LINK) { 1670 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1671 key->addr[pd->sidx].addr32[1] = 0; 1672 key->addr[pd->sidx].addr32[2] = 0; 1673 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1674 saddr = NULL; /* overwritten */ 1675 } 1676 } 1677 copy: 1678 #endif 1679 if (saddr) 1680 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1681 if (daddr) 1682 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1683 1684 return (0); 1685 } 1686 1687 int 1688 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1689 struct pf_state_key **sk, struct pf_state_key **nk) 1690 { 1691 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1692 if (*sk == NULL) 1693 return (ENOMEM); 1694 1695 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1696 0)) { 1697 uma_zfree(V_pf_state_key_z, *sk); 1698 *sk = NULL; 1699 return (ENOMEM); 1700 } 1701 1702 (*sk)->port[pd->sidx] = sport; 1703 (*sk)->port[pd->didx] = dport; 1704 (*sk)->proto = pd->proto; 1705 (*sk)->af = pd->af; 1706 1707 *nk = pf_state_key_clone(*sk); 1708 if (*nk == NULL) { 1709 uma_zfree(V_pf_state_key_z, *sk); 1710 *sk = NULL; 1711 return (ENOMEM); 1712 } 1713 1714 if (pd->af != pd->naf) { 1715 (*sk)->port[pd->sidx] = pd->osport; 1716 (*sk)->port[pd->didx] = pd->odport; 1717 1718 (*nk)->af = pd->naf; 1719 1720 /* 1721 * We're overwriting an address here, so potentially there's bits of an IPv6 1722 * address left in here. Clear that out first. 1723 */ 1724 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1725 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1726 1727 PF_ACPY(&(*nk)->addr[pd->didx], &pd->nsaddr, pd->naf); 1728 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->ndaddr, pd->naf); 1729 (*nk)->port[pd->didx] = pd->nsport; 1730 (*nk)->port[pd->sidx] = pd->ndport; 1731 switch (pd->proto) { 1732 case IPPROTO_ICMP: 1733 (*nk)->proto = IPPROTO_ICMPV6; 1734 break; 1735 case IPPROTO_ICMPV6: 1736 (*nk)->proto = IPPROTO_ICMP; 1737 break; 1738 default: 1739 (*nk)->proto = pd->proto; 1740 } 1741 } 1742 1743 return (0); 1744 } 1745 1746 struct pf_state_key * 1747 pf_state_key_clone(const struct pf_state_key *orig) 1748 { 1749 struct pf_state_key *sk; 1750 1751 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1752 if (sk == NULL) 1753 return (NULL); 1754 1755 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1756 1757 return (sk); 1758 } 1759 1760 int 1761 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1762 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1763 { 1764 struct pf_idhash *ih; 1765 struct pf_kstate *cur; 1766 int error; 1767 1768 NET_EPOCH_ASSERT(); 1769 1770 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1771 ("%s: sks not pristine", __func__)); 1772 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1773 ("%s: skw not pristine", __func__)); 1774 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1775 1776 s->kif = kif; 1777 s->orig_kif = orig_kif; 1778 1779 if (s->id == 0 && s->creatorid == 0) { 1780 s->id = alloc_unr64(&V_pf_stateid); 1781 s->id = htobe64(s->id); 1782 s->creatorid = V_pf_status.hostid; 1783 } 1784 1785 /* Returns with ID locked on success. */ 1786 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1787 return (error); 1788 1789 ih = &V_pf_idhash[PF_IDHASH(s)]; 1790 PF_HASHROW_ASSERT(ih); 1791 LIST_FOREACH(cur, &ih->states, entry) 1792 if (cur->id == s->id && cur->creatorid == s->creatorid) 1793 break; 1794 1795 if (cur != NULL) { 1796 s->timeout = PFTM_UNLINKED; 1797 PF_HASHROW_UNLOCK(ih); 1798 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1799 printf("pf: state ID collision: " 1800 "id: %016llx creatorid: %08x\n", 1801 (unsigned long long)be64toh(s->id), 1802 ntohl(s->creatorid)); 1803 } 1804 pf_detach_state(s); 1805 return (EEXIST); 1806 } 1807 LIST_INSERT_HEAD(&ih->states, s, entry); 1808 /* One for keys, one for ID hash. */ 1809 refcount_init(&s->refs, 2); 1810 1811 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1812 if (V_pfsync_insert_state_ptr != NULL) 1813 V_pfsync_insert_state_ptr(s); 1814 1815 /* Returns locked. */ 1816 return (0); 1817 } 1818 1819 /* 1820 * Find state by ID: returns with locked row on success. 1821 */ 1822 struct pf_kstate * 1823 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1824 { 1825 struct pf_idhash *ih; 1826 struct pf_kstate *s; 1827 1828 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1829 1830 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1831 1832 PF_HASHROW_LOCK(ih); 1833 LIST_FOREACH(s, &ih->states, entry) 1834 if (s->id == id && s->creatorid == creatorid) 1835 break; 1836 1837 if (s == NULL) 1838 PF_HASHROW_UNLOCK(ih); 1839 1840 return (s); 1841 } 1842 1843 /* 1844 * Find state by key. 1845 * Returns with ID hash slot locked on success. 1846 */ 1847 static struct pf_kstate * 1848 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1849 u_int dir) 1850 { 1851 struct pf_keyhash *kh; 1852 struct pf_state_key *sk; 1853 struct pf_kstate *s; 1854 int idx; 1855 1856 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1857 1858 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1859 1860 PF_HASHROW_LOCK(kh); 1861 LIST_FOREACH(sk, &kh->keys, entry) 1862 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1863 break; 1864 if (sk == NULL) { 1865 PF_HASHROW_UNLOCK(kh); 1866 return (NULL); 1867 } 1868 1869 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1870 1871 /* List is sorted, if-bound states before floating ones. */ 1872 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1873 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1874 PF_STATE_LOCK(s); 1875 PF_HASHROW_UNLOCK(kh); 1876 if (__predict_false(s->timeout >= PFTM_MAX)) { 1877 /* 1878 * State is either being processed by 1879 * pf_unlink_state() in an other thread, or 1880 * is scheduled for immediate expiry. 1881 */ 1882 PF_STATE_UNLOCK(s); 1883 return (NULL); 1884 } 1885 return (s); 1886 } 1887 1888 /* Look through the other list, in case of AF-TO */ 1889 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1890 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1891 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1892 continue; 1893 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1894 PF_STATE_LOCK(s); 1895 PF_HASHROW_UNLOCK(kh); 1896 if (__predict_false(s->timeout >= PFTM_MAX)) { 1897 /* 1898 * State is either being processed by 1899 * pf_unlink_state() in an other thread, or 1900 * is scheduled for immediate expiry. 1901 */ 1902 PF_STATE_UNLOCK(s); 1903 return (NULL); 1904 } 1905 return (s); 1906 } 1907 } 1908 1909 PF_HASHROW_UNLOCK(kh); 1910 1911 return (NULL); 1912 } 1913 1914 /* 1915 * Returns with ID hash slot locked on success. 1916 */ 1917 struct pf_kstate * 1918 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1919 { 1920 struct pf_keyhash *kh; 1921 struct pf_state_key *sk; 1922 struct pf_kstate *s, *ret = NULL; 1923 int idx, inout = 0; 1924 1925 if (more != NULL) 1926 *more = 0; 1927 1928 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1929 1930 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1931 1932 PF_HASHROW_LOCK(kh); 1933 LIST_FOREACH(sk, &kh->keys, entry) 1934 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1935 break; 1936 if (sk == NULL) { 1937 PF_HASHROW_UNLOCK(kh); 1938 return (NULL); 1939 } 1940 switch (dir) { 1941 case PF_IN: 1942 idx = PF_SK_WIRE; 1943 break; 1944 case PF_OUT: 1945 idx = PF_SK_STACK; 1946 break; 1947 case PF_INOUT: 1948 idx = PF_SK_WIRE; 1949 inout = 1; 1950 break; 1951 default: 1952 panic("%s: dir %u", __func__, dir); 1953 } 1954 second_run: 1955 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1956 if (more == NULL) { 1957 PF_STATE_LOCK(s); 1958 PF_HASHROW_UNLOCK(kh); 1959 return (s); 1960 } 1961 1962 if (ret) 1963 (*more)++; 1964 else { 1965 ret = s; 1966 PF_STATE_LOCK(s); 1967 } 1968 } 1969 if (inout == 1) { 1970 inout = 0; 1971 idx = PF_SK_STACK; 1972 goto second_run; 1973 } 1974 PF_HASHROW_UNLOCK(kh); 1975 1976 return (ret); 1977 } 1978 1979 /* 1980 * FIXME 1981 * This routine is inefficient -- locks the state only to unlock immediately on 1982 * return. 1983 * It is racy -- after the state is unlocked nothing stops other threads from 1984 * removing it. 1985 */ 1986 bool 1987 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1988 { 1989 struct pf_kstate *s; 1990 1991 s = pf_find_state_all(key, dir, NULL); 1992 if (s != NULL) { 1993 PF_STATE_UNLOCK(s); 1994 return (true); 1995 } 1996 return (false); 1997 } 1998 1999 struct pf_udp_mapping * 2000 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2001 struct pf_addr *nat_addr, uint16_t nat_port) 2002 { 2003 struct pf_udp_mapping *mapping; 2004 2005 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2006 if (mapping == NULL) 2007 return (NULL); 2008 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 2009 mapping->endpoints[0].port = src_port; 2010 mapping->endpoints[0].af = af; 2011 mapping->endpoints[0].mapping = mapping; 2012 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 2013 mapping->endpoints[1].port = nat_port; 2014 mapping->endpoints[1].af = af; 2015 mapping->endpoints[1].mapping = mapping; 2016 refcount_init(&mapping->refs, 1); 2017 return (mapping); 2018 } 2019 2020 int 2021 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2022 { 2023 struct pf_udpendpointhash *h0, *h1; 2024 struct pf_udp_endpoint *endpoint; 2025 int ret = EEXIST; 2026 2027 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2028 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2029 if (h0 == h1) { 2030 PF_HASHROW_LOCK(h0); 2031 } else if (h0 < h1) { 2032 PF_HASHROW_LOCK(h0); 2033 PF_HASHROW_LOCK(h1); 2034 } else { 2035 PF_HASHROW_LOCK(h1); 2036 PF_HASHROW_LOCK(h0); 2037 } 2038 2039 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2040 if (bcmp(endpoint, &mapping->endpoints[0], 2041 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2042 break; 2043 } 2044 if (endpoint != NULL) 2045 goto cleanup; 2046 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2047 if (bcmp(endpoint, &mapping->endpoints[1], 2048 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2049 break; 2050 } 2051 if (endpoint != NULL) 2052 goto cleanup; 2053 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2054 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2055 ret = 0; 2056 2057 cleanup: 2058 if (h0 != h1) { 2059 PF_HASHROW_UNLOCK(h0); 2060 PF_HASHROW_UNLOCK(h1); 2061 } else { 2062 PF_HASHROW_UNLOCK(h0); 2063 } 2064 return (ret); 2065 } 2066 2067 void 2068 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2069 { 2070 /* refcount is synchronized on the source endpoint's row lock */ 2071 struct pf_udpendpointhash *h0, *h1; 2072 2073 if (mapping == NULL) 2074 return; 2075 2076 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2077 PF_HASHROW_LOCK(h0); 2078 if (refcount_release(&mapping->refs)) { 2079 LIST_REMOVE(&mapping->endpoints[0], entry); 2080 PF_HASHROW_UNLOCK(h0); 2081 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2082 PF_HASHROW_LOCK(h1); 2083 LIST_REMOVE(&mapping->endpoints[1], entry); 2084 PF_HASHROW_UNLOCK(h1); 2085 2086 uma_zfree(V_pf_udp_mapping_z, mapping); 2087 } else { 2088 PF_HASHROW_UNLOCK(h0); 2089 } 2090 } 2091 2092 2093 struct pf_udp_mapping * 2094 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2095 { 2096 struct pf_udpendpointhash *uh; 2097 struct pf_udp_endpoint *endpoint; 2098 2099 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2100 2101 PF_HASHROW_LOCK(uh); 2102 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2103 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2104 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2105 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2106 break; 2107 } 2108 if (endpoint == NULL) { 2109 PF_HASHROW_UNLOCK(uh); 2110 return (NULL); 2111 } 2112 refcount_acquire(&endpoint->mapping->refs); 2113 PF_HASHROW_UNLOCK(uh); 2114 return (endpoint->mapping); 2115 } 2116 /* END state table stuff */ 2117 2118 static void 2119 pf_send(struct pf_send_entry *pfse) 2120 { 2121 2122 PF_SENDQ_LOCK(); 2123 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2124 PF_SENDQ_UNLOCK(); 2125 swi_sched(V_pf_swi_cookie, 0); 2126 } 2127 2128 static bool 2129 pf_isforlocal(struct mbuf *m, int af) 2130 { 2131 switch (af) { 2132 #ifdef INET 2133 case AF_INET: { 2134 struct ip *ip = mtod(m, struct ip *); 2135 2136 return (in_localip(ip->ip_dst)); 2137 } 2138 #endif 2139 #ifdef INET6 2140 case AF_INET6: { 2141 struct ip6_hdr *ip6; 2142 struct in6_ifaddr *ia; 2143 ip6 = mtod(m, struct ip6_hdr *); 2144 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2145 if (ia == NULL) 2146 return (false); 2147 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2148 } 2149 #endif 2150 } 2151 2152 return (false); 2153 } 2154 2155 int 2156 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2157 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2158 { 2159 /* 2160 * ICMP types marked with PF_OUT are typically responses to 2161 * PF_IN, and will match states in the opposite direction. 2162 * PF_IN ICMP types need to match a state with that type. 2163 */ 2164 *icmp_dir = PF_OUT; 2165 *multi = PF_ICMP_MULTI_LINK; 2166 /* Queries (and responses) */ 2167 switch (pd->af) { 2168 #ifdef INET 2169 case AF_INET: 2170 switch (type) { 2171 case ICMP_ECHO: 2172 *icmp_dir = PF_IN; 2173 case ICMP_ECHOREPLY: 2174 *virtual_type = ICMP_ECHO; 2175 *virtual_id = pd->hdr.icmp.icmp_id; 2176 break; 2177 2178 case ICMP_TSTAMP: 2179 *icmp_dir = PF_IN; 2180 case ICMP_TSTAMPREPLY: 2181 *virtual_type = ICMP_TSTAMP; 2182 *virtual_id = pd->hdr.icmp.icmp_id; 2183 break; 2184 2185 case ICMP_IREQ: 2186 *icmp_dir = PF_IN; 2187 case ICMP_IREQREPLY: 2188 *virtual_type = ICMP_IREQ; 2189 *virtual_id = pd->hdr.icmp.icmp_id; 2190 break; 2191 2192 case ICMP_MASKREQ: 2193 *icmp_dir = PF_IN; 2194 case ICMP_MASKREPLY: 2195 *virtual_type = ICMP_MASKREQ; 2196 *virtual_id = pd->hdr.icmp.icmp_id; 2197 break; 2198 2199 case ICMP_IPV6_WHEREAREYOU: 2200 *icmp_dir = PF_IN; 2201 case ICMP_IPV6_IAMHERE: 2202 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2203 *virtual_id = 0; /* Nothing sane to match on! */ 2204 break; 2205 2206 case ICMP_MOBILE_REGREQUEST: 2207 *icmp_dir = PF_IN; 2208 case ICMP_MOBILE_REGREPLY: 2209 *virtual_type = ICMP_MOBILE_REGREQUEST; 2210 *virtual_id = 0; /* Nothing sane to match on! */ 2211 break; 2212 2213 case ICMP_ROUTERSOLICIT: 2214 *icmp_dir = PF_IN; 2215 case ICMP_ROUTERADVERT: 2216 *virtual_type = ICMP_ROUTERSOLICIT; 2217 *virtual_id = 0; /* Nothing sane to match on! */ 2218 break; 2219 2220 /* These ICMP types map to other connections */ 2221 case ICMP_UNREACH: 2222 case ICMP_SOURCEQUENCH: 2223 case ICMP_REDIRECT: 2224 case ICMP_TIMXCEED: 2225 case ICMP_PARAMPROB: 2226 /* These will not be used, but set them anyway */ 2227 *icmp_dir = PF_IN; 2228 *virtual_type = type; 2229 *virtual_id = 0; 2230 HTONS(*virtual_type); 2231 return (1); /* These types match to another state */ 2232 2233 /* 2234 * All remaining ICMP types get their own states, 2235 * and will only match in one direction. 2236 */ 2237 default: 2238 *icmp_dir = PF_IN; 2239 *virtual_type = type; 2240 *virtual_id = 0; 2241 break; 2242 } 2243 break; 2244 #endif /* INET */ 2245 #ifdef INET6 2246 case AF_INET6: 2247 switch (type) { 2248 case ICMP6_ECHO_REQUEST: 2249 *icmp_dir = PF_IN; 2250 case ICMP6_ECHO_REPLY: 2251 *virtual_type = ICMP6_ECHO_REQUEST; 2252 *virtual_id = pd->hdr.icmp6.icmp6_id; 2253 break; 2254 2255 case MLD_LISTENER_QUERY: 2256 case MLD_LISTENER_REPORT: { 2257 /* 2258 * Listener Report can be sent by clients 2259 * without an associated Listener Query. 2260 * In addition to that, when Report is sent as a 2261 * reply to a Query its source and destination 2262 * address are different. 2263 */ 2264 *icmp_dir = PF_IN; 2265 *virtual_type = MLD_LISTENER_QUERY; 2266 *virtual_id = 0; 2267 break; 2268 } 2269 case MLD_MTRACE: 2270 *icmp_dir = PF_IN; 2271 case MLD_MTRACE_RESP: 2272 *virtual_type = MLD_MTRACE; 2273 *virtual_id = 0; /* Nothing sane to match on! */ 2274 break; 2275 2276 case ND_NEIGHBOR_SOLICIT: 2277 *icmp_dir = PF_IN; 2278 case ND_NEIGHBOR_ADVERT: { 2279 *virtual_type = ND_NEIGHBOR_SOLICIT; 2280 *virtual_id = 0; 2281 break; 2282 } 2283 2284 /* 2285 * These ICMP types map to other connections. 2286 * ND_REDIRECT can't be in this list because the triggering 2287 * packet header is optional. 2288 */ 2289 case ICMP6_DST_UNREACH: 2290 case ICMP6_PACKET_TOO_BIG: 2291 case ICMP6_TIME_EXCEEDED: 2292 case ICMP6_PARAM_PROB: 2293 /* These will not be used, but set them anyway */ 2294 *icmp_dir = PF_IN; 2295 *virtual_type = type; 2296 *virtual_id = 0; 2297 HTONS(*virtual_type); 2298 return (1); /* These types match to another state */ 2299 /* 2300 * All remaining ICMP6 types get their own states, 2301 * and will only match in one direction. 2302 */ 2303 default: 2304 *icmp_dir = PF_IN; 2305 *virtual_type = type; 2306 *virtual_id = 0; 2307 break; 2308 } 2309 break; 2310 #endif /* INET6 */ 2311 } 2312 HTONS(*virtual_type); 2313 return (0); /* These types match to their own state */ 2314 } 2315 2316 void 2317 pf_intr(void *v) 2318 { 2319 struct epoch_tracker et; 2320 struct pf_send_head queue; 2321 struct pf_send_entry *pfse, *next; 2322 2323 CURVNET_SET((struct vnet *)v); 2324 2325 PF_SENDQ_LOCK(); 2326 queue = V_pf_sendqueue; 2327 STAILQ_INIT(&V_pf_sendqueue); 2328 PF_SENDQ_UNLOCK(); 2329 2330 NET_EPOCH_ENTER(et); 2331 2332 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2333 switch (pfse->pfse_type) { 2334 #ifdef INET 2335 case PFSE_IP: { 2336 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2337 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2338 ("%s: rcvif != loif", __func__)); 2339 2340 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2341 pfse->pfse_m->m_pkthdr.csum_flags |= 2342 CSUM_IP_VALID | CSUM_IP_CHECKED; 2343 ip_input(pfse->pfse_m); 2344 } else { 2345 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2346 NULL); 2347 } 2348 break; 2349 } 2350 case PFSE_ICMP: 2351 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2352 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2353 break; 2354 #endif /* INET */ 2355 #ifdef INET6 2356 case PFSE_IP6: 2357 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2358 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2359 ("%s: rcvif != loif", __func__)); 2360 2361 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2362 M_LOOP; 2363 ip6_input(pfse->pfse_m); 2364 } else { 2365 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2366 NULL, NULL); 2367 } 2368 break; 2369 case PFSE_ICMP6: 2370 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2371 pfse->icmpopts.code, pfse->icmpopts.mtu); 2372 break; 2373 #endif /* INET6 */ 2374 default: 2375 panic("%s: unknown type", __func__); 2376 } 2377 free(pfse, M_PFTEMP); 2378 } 2379 NET_EPOCH_EXIT(et); 2380 CURVNET_RESTORE(); 2381 } 2382 2383 #define pf_purge_thread_period (hz / 10) 2384 2385 #ifdef PF_WANT_32_TO_64_COUNTER 2386 static void 2387 pf_status_counter_u64_periodic(void) 2388 { 2389 2390 PF_RULES_RASSERT(); 2391 2392 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2393 return; 2394 } 2395 2396 for (int i = 0; i < FCNT_MAX; i++) { 2397 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2398 } 2399 } 2400 2401 static void 2402 pf_kif_counter_u64_periodic(void) 2403 { 2404 struct pfi_kkif *kif; 2405 size_t r, run; 2406 2407 PF_RULES_RASSERT(); 2408 2409 if (__predict_false(V_pf_allkifcount == 0)) { 2410 return; 2411 } 2412 2413 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2414 return; 2415 } 2416 2417 run = V_pf_allkifcount / 10; 2418 if (run < 5) 2419 run = 5; 2420 2421 for (r = 0; r < run; r++) { 2422 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2423 if (kif == NULL) { 2424 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2425 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2426 break; 2427 } 2428 2429 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2430 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2431 2432 for (int i = 0; i < 2; i++) { 2433 for (int j = 0; j < 2; j++) { 2434 for (int k = 0; k < 2; k++) { 2435 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2436 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2437 } 2438 } 2439 } 2440 } 2441 } 2442 2443 static void 2444 pf_rule_counter_u64_periodic(void) 2445 { 2446 struct pf_krule *rule; 2447 size_t r, run; 2448 2449 PF_RULES_RASSERT(); 2450 2451 if (__predict_false(V_pf_allrulecount == 0)) { 2452 return; 2453 } 2454 2455 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2456 return; 2457 } 2458 2459 run = V_pf_allrulecount / 10; 2460 if (run < 5) 2461 run = 5; 2462 2463 for (r = 0; r < run; r++) { 2464 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2465 if (rule == NULL) { 2466 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2467 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2468 break; 2469 } 2470 2471 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2472 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2473 2474 pf_counter_u64_periodic(&rule->evaluations); 2475 for (int i = 0; i < 2; i++) { 2476 pf_counter_u64_periodic(&rule->packets[i]); 2477 pf_counter_u64_periodic(&rule->bytes[i]); 2478 } 2479 } 2480 } 2481 2482 static void 2483 pf_counter_u64_periodic_main(void) 2484 { 2485 PF_RULES_RLOCK_TRACKER; 2486 2487 V_pf_counter_periodic_iter++; 2488 2489 PF_RULES_RLOCK(); 2490 pf_counter_u64_critical_enter(); 2491 pf_status_counter_u64_periodic(); 2492 pf_kif_counter_u64_periodic(); 2493 pf_rule_counter_u64_periodic(); 2494 pf_counter_u64_critical_exit(); 2495 PF_RULES_RUNLOCK(); 2496 } 2497 #else 2498 #define pf_counter_u64_periodic_main() do { } while (0) 2499 #endif 2500 2501 void 2502 pf_purge_thread(void *unused __unused) 2503 { 2504 struct epoch_tracker et; 2505 2506 VNET_ITERATOR_DECL(vnet_iter); 2507 2508 sx_xlock(&pf_end_lock); 2509 while (pf_end_threads == 0) { 2510 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2511 2512 VNET_LIST_RLOCK(); 2513 NET_EPOCH_ENTER(et); 2514 VNET_FOREACH(vnet_iter) { 2515 CURVNET_SET(vnet_iter); 2516 2517 /* Wait until V_pf_default_rule is initialized. */ 2518 if (V_pf_vnet_active == 0) { 2519 CURVNET_RESTORE(); 2520 continue; 2521 } 2522 2523 pf_counter_u64_periodic_main(); 2524 2525 /* 2526 * Process 1/interval fraction of the state 2527 * table every run. 2528 */ 2529 V_pf_purge_idx = 2530 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2531 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2532 2533 /* 2534 * Purge other expired types every 2535 * PFTM_INTERVAL seconds. 2536 */ 2537 if (V_pf_purge_idx == 0) { 2538 /* 2539 * Order is important: 2540 * - states and src nodes reference rules 2541 * - states and rules reference kifs 2542 */ 2543 pf_purge_expired_fragments(); 2544 pf_purge_expired_src_nodes(); 2545 pf_purge_unlinked_rules(); 2546 pfi_kkif_purge(); 2547 } 2548 CURVNET_RESTORE(); 2549 } 2550 NET_EPOCH_EXIT(et); 2551 VNET_LIST_RUNLOCK(); 2552 } 2553 2554 pf_end_threads++; 2555 sx_xunlock(&pf_end_lock); 2556 kproc_exit(0); 2557 } 2558 2559 void 2560 pf_unload_vnet_purge(void) 2561 { 2562 2563 /* 2564 * To cleanse up all kifs and rules we need 2565 * two runs: first one clears reference flags, 2566 * then pf_purge_expired_states() doesn't 2567 * raise them, and then second run frees. 2568 */ 2569 pf_purge_unlinked_rules(); 2570 pfi_kkif_purge(); 2571 2572 /* 2573 * Now purge everything. 2574 */ 2575 pf_purge_expired_states(0, V_pf_hashmask); 2576 pf_purge_fragments(UINT_MAX); 2577 pf_purge_expired_src_nodes(); 2578 2579 /* 2580 * Now all kifs & rules should be unreferenced, 2581 * thus should be successfully freed. 2582 */ 2583 pf_purge_unlinked_rules(); 2584 pfi_kkif_purge(); 2585 } 2586 2587 u_int32_t 2588 pf_state_expires(const struct pf_kstate *state) 2589 { 2590 u_int32_t timeout; 2591 u_int32_t start; 2592 u_int32_t end; 2593 u_int32_t states; 2594 2595 /* handle all PFTM_* > PFTM_MAX here */ 2596 if (state->timeout == PFTM_PURGE) 2597 return (time_uptime); 2598 KASSERT(state->timeout != PFTM_UNLINKED, 2599 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2600 KASSERT((state->timeout < PFTM_MAX), 2601 ("pf_state_expires: timeout > PFTM_MAX")); 2602 timeout = state->rule->timeout[state->timeout]; 2603 if (!timeout) 2604 timeout = V_pf_default_rule.timeout[state->timeout]; 2605 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2606 if (start && state->rule != &V_pf_default_rule) { 2607 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2608 states = counter_u64_fetch(state->rule->states_cur); 2609 } else { 2610 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2611 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2612 states = V_pf_status.states; 2613 } 2614 if (end && states > start && start < end) { 2615 if (states < end) { 2616 timeout = (u_int64_t)timeout * (end - states) / 2617 (end - start); 2618 return ((state->expire / 1000) + timeout); 2619 } 2620 else 2621 return (time_uptime); 2622 } 2623 return ((state->expire / 1000) + timeout); 2624 } 2625 2626 void 2627 pf_purge_expired_src_nodes(void) 2628 { 2629 struct pf_ksrc_node_list freelist; 2630 struct pf_srchash *sh; 2631 struct pf_ksrc_node *cur, *next; 2632 int i; 2633 2634 LIST_INIT(&freelist); 2635 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2636 PF_HASHROW_LOCK(sh); 2637 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2638 if (cur->states == 0 && cur->expire <= time_uptime) { 2639 pf_unlink_src_node(cur); 2640 LIST_INSERT_HEAD(&freelist, cur, entry); 2641 } else if (cur->rule != NULL) 2642 cur->rule->rule_ref |= PFRULE_REFS; 2643 PF_HASHROW_UNLOCK(sh); 2644 } 2645 2646 pf_free_src_nodes(&freelist); 2647 2648 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2649 } 2650 2651 static void 2652 pf_src_tree_remove_state(struct pf_kstate *s) 2653 { 2654 struct pf_ksrc_node *sn; 2655 uint32_t timeout; 2656 2657 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2658 s->rule->timeout[PFTM_SRC_NODE] : 2659 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2660 2661 if (s->src_node != NULL) { 2662 sn = s->src_node; 2663 PF_SRC_NODE_LOCK(sn); 2664 if (s->src.tcp_est) 2665 --sn->conn; 2666 if (--sn->states == 0) 2667 sn->expire = time_uptime + timeout; 2668 PF_SRC_NODE_UNLOCK(sn); 2669 } 2670 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2671 sn = s->nat_src_node; 2672 PF_SRC_NODE_LOCK(sn); 2673 if (--sn->states == 0) 2674 sn->expire = time_uptime + timeout; 2675 PF_SRC_NODE_UNLOCK(sn); 2676 } 2677 s->src_node = s->nat_src_node = NULL; 2678 } 2679 2680 /* 2681 * Unlink and potentilly free a state. Function may be 2682 * called with ID hash row locked, but always returns 2683 * unlocked, since it needs to go through key hash locking. 2684 */ 2685 int 2686 pf_unlink_state(struct pf_kstate *s) 2687 { 2688 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2689 2690 NET_EPOCH_ASSERT(); 2691 PF_HASHROW_ASSERT(ih); 2692 2693 if (s->timeout == PFTM_UNLINKED) { 2694 /* 2695 * State is being processed 2696 * by pf_unlink_state() in 2697 * an other thread. 2698 */ 2699 PF_HASHROW_UNLOCK(ih); 2700 return (0); /* XXXGL: undefined actually */ 2701 } 2702 2703 if (s->src.state == PF_TCPS_PROXY_DST) { 2704 /* XXX wire key the right one? */ 2705 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2706 &s->key[PF_SK_WIRE]->addr[1], 2707 &s->key[PF_SK_WIRE]->addr[0], 2708 s->key[PF_SK_WIRE]->port[1], 2709 s->key[PF_SK_WIRE]->port[0], 2710 s->src.seqhi, s->src.seqlo + 1, 2711 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2712 s->act.rtableid); 2713 } 2714 2715 LIST_REMOVE(s, entry); 2716 pf_src_tree_remove_state(s); 2717 2718 if (V_pfsync_delete_state_ptr != NULL) 2719 V_pfsync_delete_state_ptr(s); 2720 2721 STATE_DEC_COUNTERS(s); 2722 2723 s->timeout = PFTM_UNLINKED; 2724 2725 /* Ensure we remove it from the list of halfopen states, if needed. */ 2726 if (s->key[PF_SK_STACK] != NULL && 2727 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2728 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2729 2730 PF_HASHROW_UNLOCK(ih); 2731 2732 pf_detach_state(s); 2733 2734 pf_udp_mapping_release(s->udp_mapping); 2735 2736 /* pf_state_insert() initialises refs to 2 */ 2737 return (pf_release_staten(s, 2)); 2738 } 2739 2740 struct pf_kstate * 2741 pf_alloc_state(int flags) 2742 { 2743 2744 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2745 } 2746 2747 void 2748 pf_free_state(struct pf_kstate *cur) 2749 { 2750 struct pf_krule_item *ri; 2751 2752 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2753 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2754 cur->timeout)); 2755 2756 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2757 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2758 free(ri, M_PF_RULE_ITEM); 2759 } 2760 2761 pf_normalize_tcp_cleanup(cur); 2762 uma_zfree(V_pf_state_z, cur); 2763 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2764 } 2765 2766 /* 2767 * Called only from pf_purge_thread(), thus serialized. 2768 */ 2769 static u_int 2770 pf_purge_expired_states(u_int i, int maxcheck) 2771 { 2772 struct pf_idhash *ih; 2773 struct pf_kstate *s; 2774 struct pf_krule_item *mrm; 2775 size_t count __unused; 2776 2777 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2778 2779 /* 2780 * Go through hash and unlink states that expire now. 2781 */ 2782 while (maxcheck > 0) { 2783 count = 0; 2784 ih = &V_pf_idhash[i]; 2785 2786 /* only take the lock if we expect to do work */ 2787 if (!LIST_EMPTY(&ih->states)) { 2788 relock: 2789 PF_HASHROW_LOCK(ih); 2790 LIST_FOREACH(s, &ih->states, entry) { 2791 if (pf_state_expires(s) <= time_uptime) { 2792 V_pf_status.states -= 2793 pf_unlink_state(s); 2794 goto relock; 2795 } 2796 s->rule->rule_ref |= PFRULE_REFS; 2797 if (s->nat_rule != NULL) 2798 s->nat_rule->rule_ref |= PFRULE_REFS; 2799 if (s->anchor != NULL) 2800 s->anchor->rule_ref |= PFRULE_REFS; 2801 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2802 SLIST_FOREACH(mrm, &s->match_rules, entry) 2803 mrm->r->rule_ref |= PFRULE_REFS; 2804 if (s->act.rt_kif) 2805 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2806 count++; 2807 } 2808 PF_HASHROW_UNLOCK(ih); 2809 } 2810 2811 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2812 2813 /* Return when we hit end of hash. */ 2814 if (++i > V_pf_hashmask) { 2815 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2816 return (0); 2817 } 2818 2819 maxcheck--; 2820 } 2821 2822 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2823 2824 return (i); 2825 } 2826 2827 static void 2828 pf_purge_unlinked_rules(void) 2829 { 2830 struct pf_krulequeue tmpq; 2831 struct pf_krule *r, *r1; 2832 2833 /* 2834 * If we have overloading task pending, then we'd 2835 * better skip purging this time. There is a tiny 2836 * probability that overloading task references 2837 * an already unlinked rule. 2838 */ 2839 PF_OVERLOADQ_LOCK(); 2840 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2841 PF_OVERLOADQ_UNLOCK(); 2842 return; 2843 } 2844 PF_OVERLOADQ_UNLOCK(); 2845 2846 /* 2847 * Do naive mark-and-sweep garbage collecting of old rules. 2848 * Reference flag is raised by pf_purge_expired_states() 2849 * and pf_purge_expired_src_nodes(). 2850 * 2851 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2852 * use a temporary queue. 2853 */ 2854 TAILQ_INIT(&tmpq); 2855 PF_UNLNKDRULES_LOCK(); 2856 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2857 if (!(r->rule_ref & PFRULE_REFS)) { 2858 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2859 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2860 } else 2861 r->rule_ref &= ~PFRULE_REFS; 2862 } 2863 PF_UNLNKDRULES_UNLOCK(); 2864 2865 if (!TAILQ_EMPTY(&tmpq)) { 2866 PF_CONFIG_LOCK(); 2867 PF_RULES_WLOCK(); 2868 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2869 TAILQ_REMOVE(&tmpq, r, entries); 2870 pf_free_rule(r); 2871 } 2872 PF_RULES_WUNLOCK(); 2873 PF_CONFIG_UNLOCK(); 2874 } 2875 } 2876 2877 void 2878 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2879 { 2880 switch (af) { 2881 #ifdef INET 2882 case AF_INET: { 2883 u_int32_t a = ntohl(addr->addr32[0]); 2884 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2885 (a>>8)&255, a&255); 2886 if (p) { 2887 p = ntohs(p); 2888 printf(":%u", p); 2889 } 2890 break; 2891 } 2892 #endif /* INET */ 2893 #ifdef INET6 2894 case AF_INET6: { 2895 u_int16_t b; 2896 u_int8_t i, curstart, curend, maxstart, maxend; 2897 curstart = curend = maxstart = maxend = 255; 2898 for (i = 0; i < 8; i++) { 2899 if (!addr->addr16[i]) { 2900 if (curstart == 255) 2901 curstart = i; 2902 curend = i; 2903 } else { 2904 if ((curend - curstart) > 2905 (maxend - maxstart)) { 2906 maxstart = curstart; 2907 maxend = curend; 2908 } 2909 curstart = curend = 255; 2910 } 2911 } 2912 if ((curend - curstart) > 2913 (maxend - maxstart)) { 2914 maxstart = curstart; 2915 maxend = curend; 2916 } 2917 for (i = 0; i < 8; i++) { 2918 if (i >= maxstart && i <= maxend) { 2919 if (i == 0) 2920 printf(":"); 2921 if (i == maxend) 2922 printf(":"); 2923 } else { 2924 b = ntohs(addr->addr16[i]); 2925 printf("%x", b); 2926 if (i < 7) 2927 printf(":"); 2928 } 2929 } 2930 if (p) { 2931 p = ntohs(p); 2932 printf("[%u]", p); 2933 } 2934 break; 2935 } 2936 #endif /* INET6 */ 2937 } 2938 } 2939 2940 void 2941 pf_print_state(struct pf_kstate *s) 2942 { 2943 pf_print_state_parts(s, NULL, NULL); 2944 } 2945 2946 static void 2947 pf_print_state_parts(struct pf_kstate *s, 2948 struct pf_state_key *skwp, struct pf_state_key *sksp) 2949 { 2950 struct pf_state_key *skw, *sks; 2951 u_int8_t proto, dir; 2952 2953 /* Do our best to fill these, but they're skipped if NULL */ 2954 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2955 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2956 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2957 dir = s ? s->direction : 0; 2958 2959 switch (proto) { 2960 case IPPROTO_IPV4: 2961 printf("IPv4"); 2962 break; 2963 case IPPROTO_IPV6: 2964 printf("IPv6"); 2965 break; 2966 case IPPROTO_TCP: 2967 printf("TCP"); 2968 break; 2969 case IPPROTO_UDP: 2970 printf("UDP"); 2971 break; 2972 case IPPROTO_ICMP: 2973 printf("ICMP"); 2974 break; 2975 case IPPROTO_ICMPV6: 2976 printf("ICMPv6"); 2977 break; 2978 default: 2979 printf("%u", proto); 2980 break; 2981 } 2982 switch (dir) { 2983 case PF_IN: 2984 printf(" in"); 2985 break; 2986 case PF_OUT: 2987 printf(" out"); 2988 break; 2989 } 2990 if (skw) { 2991 printf(" wire: "); 2992 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2993 printf(" "); 2994 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2995 } 2996 if (sks) { 2997 printf(" stack: "); 2998 if (sks != skw) { 2999 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3000 printf(" "); 3001 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3002 } else 3003 printf("-"); 3004 } 3005 if (s) { 3006 if (proto == IPPROTO_TCP) { 3007 printf(" [lo=%u high=%u win=%u modulator=%u", 3008 s->src.seqlo, s->src.seqhi, 3009 s->src.max_win, s->src.seqdiff); 3010 if (s->src.wscale && s->dst.wscale) 3011 printf(" wscale=%u", 3012 s->src.wscale & PF_WSCALE_MASK); 3013 printf("]"); 3014 printf(" [lo=%u high=%u win=%u modulator=%u", 3015 s->dst.seqlo, s->dst.seqhi, 3016 s->dst.max_win, s->dst.seqdiff); 3017 if (s->src.wscale && s->dst.wscale) 3018 printf(" wscale=%u", 3019 s->dst.wscale & PF_WSCALE_MASK); 3020 printf("]"); 3021 } 3022 printf(" %u:%u", s->src.state, s->dst.state); 3023 if (s->rule) 3024 printf(" @%d", s->rule->nr); 3025 } 3026 } 3027 3028 void 3029 pf_print_flags(uint16_t f) 3030 { 3031 if (f) 3032 printf(" "); 3033 if (f & TH_FIN) 3034 printf("F"); 3035 if (f & TH_SYN) 3036 printf("S"); 3037 if (f & TH_RST) 3038 printf("R"); 3039 if (f & TH_PUSH) 3040 printf("P"); 3041 if (f & TH_ACK) 3042 printf("A"); 3043 if (f & TH_URG) 3044 printf("U"); 3045 if (f & TH_ECE) 3046 printf("E"); 3047 if (f & TH_CWR) 3048 printf("W"); 3049 if (f & TH_AE) 3050 printf("e"); 3051 } 3052 3053 #define PF_SET_SKIP_STEPS(i) \ 3054 do { \ 3055 while (head[i] != cur) { \ 3056 head[i]->skip[i] = cur; \ 3057 head[i] = TAILQ_NEXT(head[i], entries); \ 3058 } \ 3059 } while (0) 3060 3061 void 3062 pf_calc_skip_steps(struct pf_krulequeue *rules) 3063 { 3064 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3065 int i; 3066 3067 cur = TAILQ_FIRST(rules); 3068 prev = cur; 3069 for (i = 0; i < PF_SKIP_COUNT; ++i) 3070 head[i] = cur; 3071 while (cur != NULL) { 3072 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3073 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3074 if (cur->direction != prev->direction) 3075 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3076 if (cur->af != prev->af) 3077 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3078 if (cur->proto != prev->proto) 3079 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3080 if (cur->src.neg != prev->src.neg || 3081 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3082 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3083 if (cur->dst.neg != prev->dst.neg || 3084 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3085 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3086 if (cur->src.port[0] != prev->src.port[0] || 3087 cur->src.port[1] != prev->src.port[1] || 3088 cur->src.port_op != prev->src.port_op) 3089 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3090 if (cur->dst.port[0] != prev->dst.port[0] || 3091 cur->dst.port[1] != prev->dst.port[1] || 3092 cur->dst.port_op != prev->dst.port_op) 3093 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3094 3095 prev = cur; 3096 cur = TAILQ_NEXT(cur, entries); 3097 } 3098 for (i = 0; i < PF_SKIP_COUNT; ++i) 3099 PF_SET_SKIP_STEPS(i); 3100 } 3101 3102 int 3103 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3104 { 3105 if (aw1->type != aw2->type) 3106 return (1); 3107 switch (aw1->type) { 3108 case PF_ADDR_ADDRMASK: 3109 case PF_ADDR_RANGE: 3110 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3111 return (1); 3112 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3113 return (1); 3114 return (0); 3115 case PF_ADDR_DYNIFTL: 3116 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3117 case PF_ADDR_NONE: 3118 case PF_ADDR_NOROUTE: 3119 case PF_ADDR_URPFFAILED: 3120 return (0); 3121 case PF_ADDR_TABLE: 3122 return (aw1->p.tbl != aw2->p.tbl); 3123 default: 3124 printf("invalid address type: %d\n", aw1->type); 3125 return (1); 3126 } 3127 } 3128 3129 /** 3130 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3131 * header isn't always a full checksum. In some cases (i.e. output) it's a 3132 * pseudo-header checksum, which is a partial checksum over src/dst IP 3133 * addresses, protocol number and length. 3134 * 3135 * That means we have the following cases: 3136 * * Input or forwarding: we don't have TSO, the checksum fields are full 3137 * checksums, we need to update the checksum whenever we change anything. 3138 * * Output (i.e. the checksum is a pseudo-header checksum): 3139 * x The field being updated is src/dst address or affects the length of 3140 * the packet. We need to update the pseudo-header checksum (note that this 3141 * checksum is not ones' complement). 3142 * x Some other field is being modified (e.g. src/dst port numbers): We 3143 * don't have to update anything. 3144 **/ 3145 u_int16_t 3146 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3147 { 3148 u_int32_t x; 3149 3150 x = cksum + old - new; 3151 x = (x + (x >> 16)) & 0xffff; 3152 3153 /* optimise: eliminate a branch when not udp */ 3154 if (udp && cksum == 0x0000) 3155 return cksum; 3156 if (udp && x == 0x0000) 3157 x = 0xffff; 3158 3159 return (u_int16_t)(x); 3160 } 3161 3162 static void 3163 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3164 u_int8_t udp) 3165 { 3166 u_int16_t old = htons(hi ? (*f << 8) : *f); 3167 u_int16_t new = htons(hi ? ( v << 8) : v); 3168 3169 if (*f == v) 3170 return; 3171 3172 *f = v; 3173 3174 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3175 return; 3176 3177 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3178 } 3179 3180 void 3181 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3182 bool hi, u_int8_t udp) 3183 { 3184 u_int8_t *fb = (u_int8_t *)f; 3185 u_int8_t *vb = (u_int8_t *)&v; 3186 3187 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3188 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3189 } 3190 3191 void 3192 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3193 bool hi, u_int8_t udp) 3194 { 3195 u_int8_t *fb = (u_int8_t *)f; 3196 u_int8_t *vb = (u_int8_t *)&v; 3197 3198 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3199 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3200 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3201 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3202 } 3203 3204 u_int16_t 3205 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3206 u_int16_t new, u_int8_t udp) 3207 { 3208 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3209 return (cksum); 3210 3211 return (pf_cksum_fixup(cksum, old, new, udp)); 3212 } 3213 3214 static void 3215 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3216 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3217 sa_family_t af, sa_family_t naf) 3218 { 3219 struct pf_addr ao; 3220 u_int16_t po = *p; 3221 3222 PF_ACPY(&ao, a, af); 3223 if (af == naf) 3224 PF_ACPY(a, an, af); 3225 3226 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3227 *pc = ~*pc; 3228 3229 *p = pn; 3230 3231 switch (af) { 3232 #ifdef INET 3233 case AF_INET: 3234 switch (naf) { 3235 case AF_INET: 3236 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3237 ao.addr16[0], an->addr16[0], 0), 3238 ao.addr16[1], an->addr16[1], 0); 3239 *p = pn; 3240 3241 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3242 ao.addr16[0], an->addr16[0], u), 3243 ao.addr16[1], an->addr16[1], u); 3244 3245 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3246 break; 3247 #ifdef INET6 3248 case AF_INET6: 3249 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3250 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3251 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3252 ao.addr16[0], an->addr16[0], u), 3253 ao.addr16[1], an->addr16[1], u), 3254 0, an->addr16[2], u), 3255 0, an->addr16[3], u), 3256 0, an->addr16[4], u), 3257 0, an->addr16[5], u), 3258 0, an->addr16[6], u), 3259 0, an->addr16[7], u), 3260 po, pn, u); 3261 3262 /* XXXKP TODO *ic checksum? */ 3263 break; 3264 #endif /* INET6 */ 3265 } 3266 break; 3267 #endif /* INET */ 3268 #ifdef INET6 3269 case AF_INET6: 3270 switch (naf) { 3271 #ifdef INET 3272 case AF_INET: 3273 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3274 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3275 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3276 ao.addr16[0], an->addr16[0], u), 3277 ao.addr16[1], an->addr16[1], u), 3278 ao.addr16[2], 0, u), 3279 ao.addr16[3], 0, u), 3280 ao.addr16[4], 0, u), 3281 ao.addr16[5], 0, u), 3282 ao.addr16[6], 0, u), 3283 ao.addr16[7], 0, u), 3284 po, pn, u); 3285 3286 /* XXXKP TODO *ic checksum? */ 3287 break; 3288 #endif /* INET */ 3289 case AF_INET6: 3290 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3291 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3292 pf_cksum_fixup(pf_cksum_fixup(*pc, 3293 ao.addr16[0], an->addr16[0], u), 3294 ao.addr16[1], an->addr16[1], u), 3295 ao.addr16[2], an->addr16[2], u), 3296 ao.addr16[3], an->addr16[3], u), 3297 ao.addr16[4], an->addr16[4], u), 3298 ao.addr16[5], an->addr16[5], u), 3299 ao.addr16[6], an->addr16[6], u), 3300 ao.addr16[7], an->addr16[7], u); 3301 3302 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3303 break; 3304 } 3305 break; 3306 #endif /* INET6 */ 3307 } 3308 3309 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3310 CSUM_DELAY_DATA_IPV6)) { 3311 *pc = ~*pc; 3312 if (! *pc) 3313 *pc = 0xffff; 3314 } 3315 } 3316 3317 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3318 void 3319 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3320 { 3321 u_int32_t ao; 3322 3323 memcpy(&ao, a, sizeof(ao)); 3324 memcpy(a, &an, sizeof(u_int32_t)); 3325 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3326 ao % 65536, an % 65536, u); 3327 } 3328 3329 void 3330 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3331 { 3332 u_int32_t ao; 3333 3334 memcpy(&ao, a, sizeof(ao)); 3335 memcpy(a, &an, sizeof(u_int32_t)); 3336 3337 *c = pf_proto_cksum_fixup(m, 3338 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3339 ao % 65536, an % 65536, udp); 3340 } 3341 3342 #ifdef INET6 3343 static void 3344 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3345 { 3346 struct pf_addr ao; 3347 3348 PF_ACPY(&ao, a, AF_INET6); 3349 PF_ACPY(a, an, AF_INET6); 3350 3351 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3352 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3353 pf_cksum_fixup(pf_cksum_fixup(*c, 3354 ao.addr16[0], an->addr16[0], u), 3355 ao.addr16[1], an->addr16[1], u), 3356 ao.addr16[2], an->addr16[2], u), 3357 ao.addr16[3], an->addr16[3], u), 3358 ao.addr16[4], an->addr16[4], u), 3359 ao.addr16[5], an->addr16[5], u), 3360 ao.addr16[6], an->addr16[6], u), 3361 ao.addr16[7], an->addr16[7], u); 3362 } 3363 #endif /* INET6 */ 3364 3365 static void 3366 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3367 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3368 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3369 { 3370 struct pf_addr oia, ooa; 3371 3372 PF_ACPY(&oia, ia, af); 3373 if (oa) 3374 PF_ACPY(&ooa, oa, af); 3375 3376 /* Change inner protocol port, fix inner protocol checksum. */ 3377 if (ip != NULL) { 3378 u_int16_t oip = *ip; 3379 u_int32_t opc; 3380 3381 if (pc != NULL) 3382 opc = *pc; 3383 *ip = np; 3384 if (pc != NULL) 3385 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3386 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3387 if (pc != NULL) 3388 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3389 } 3390 /* Change inner ip address, fix inner ip and icmp checksums. */ 3391 PF_ACPY(ia, na, af); 3392 switch (af) { 3393 #ifdef INET 3394 case AF_INET: { 3395 u_int32_t oh2c = *h2c; 3396 3397 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3398 oia.addr16[0], ia->addr16[0], 0), 3399 oia.addr16[1], ia->addr16[1], 0); 3400 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3401 oia.addr16[0], ia->addr16[0], 0), 3402 oia.addr16[1], ia->addr16[1], 0); 3403 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3404 break; 3405 } 3406 #endif /* INET */ 3407 #ifdef INET6 3408 case AF_INET6: 3409 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3410 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3411 pf_cksum_fixup(pf_cksum_fixup(*ic, 3412 oia.addr16[0], ia->addr16[0], u), 3413 oia.addr16[1], ia->addr16[1], u), 3414 oia.addr16[2], ia->addr16[2], u), 3415 oia.addr16[3], ia->addr16[3], u), 3416 oia.addr16[4], ia->addr16[4], u), 3417 oia.addr16[5], ia->addr16[5], u), 3418 oia.addr16[6], ia->addr16[6], u), 3419 oia.addr16[7], ia->addr16[7], u); 3420 break; 3421 #endif /* INET6 */ 3422 } 3423 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3424 if (oa) { 3425 PF_ACPY(oa, na, af); 3426 switch (af) { 3427 #ifdef INET 3428 case AF_INET: 3429 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3430 ooa.addr16[0], oa->addr16[0], 0), 3431 ooa.addr16[1], oa->addr16[1], 0); 3432 break; 3433 #endif /* INET */ 3434 #ifdef INET6 3435 case AF_INET6: 3436 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3437 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3438 pf_cksum_fixup(pf_cksum_fixup(*ic, 3439 ooa.addr16[0], oa->addr16[0], u), 3440 ooa.addr16[1], oa->addr16[1], u), 3441 ooa.addr16[2], oa->addr16[2], u), 3442 ooa.addr16[3], oa->addr16[3], u), 3443 ooa.addr16[4], oa->addr16[4], u), 3444 ooa.addr16[5], oa->addr16[5], u), 3445 ooa.addr16[6], oa->addr16[6], u), 3446 ooa.addr16[7], oa->addr16[7], u); 3447 break; 3448 #endif /* INET6 */ 3449 } 3450 } 3451 } 3452 3453 int 3454 pf_translate_af(struct pf_pdesc *pd) 3455 { 3456 #if defined(INET) && defined(INET6) 3457 struct mbuf *mp; 3458 struct ip *ip4; 3459 struct ip6_hdr *ip6; 3460 struct icmp6_hdr *icmp; 3461 struct m_tag *mtag; 3462 struct pf_fragment_tag *ftag; 3463 int hlen; 3464 3465 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3466 3467 /* trim the old header */ 3468 m_adj(pd->m, pd->off); 3469 3470 /* prepend a new one */ 3471 M_PREPEND(pd->m, hlen, M_NOWAIT); 3472 if (pd->m == NULL) 3473 return (-1); 3474 3475 switch (pd->naf) { 3476 case AF_INET: 3477 ip4 = mtod(pd->m, struct ip *); 3478 bzero(ip4, hlen); 3479 ip4->ip_v = IPVERSION; 3480 ip4->ip_hl = hlen >> 2; 3481 ip4->ip_tos = pd->tos; 3482 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3483 ip_fillid(ip4); 3484 ip4->ip_ttl = pd->ttl; 3485 ip4->ip_p = pd->proto; 3486 ip4->ip_src = pd->nsaddr.v4; 3487 ip4->ip_dst = pd->ndaddr.v4; 3488 pd->src = (struct pf_addr *)&ip4->ip_src; 3489 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3490 pd->off = sizeof(struct ip); 3491 break; 3492 case AF_INET6: 3493 ip6 = mtod(pd->m, struct ip6_hdr *); 3494 bzero(ip6, hlen); 3495 ip6->ip6_vfc = IPV6_VERSION; 3496 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3497 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3498 ip6->ip6_nxt = pd->proto; 3499 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3500 ip6->ip6_hlim = IPV6_DEFHLIM; 3501 else 3502 ip6->ip6_hlim = pd->ttl; 3503 ip6->ip6_src = pd->nsaddr.v6; 3504 ip6->ip6_dst = pd->ndaddr.v6; 3505 pd->src = (struct pf_addr *)&ip6->ip6_src; 3506 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3507 pd->off = sizeof(struct ip6_hdr); 3508 3509 /* 3510 * If we're dealing with a reassembled packet we need to adjust 3511 * the header length from the IPv4 header size to IPv6 header 3512 * size. 3513 */ 3514 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3515 if (mtag) { 3516 ftag = (struct pf_fragment_tag *)(mtag + 1); 3517 ftag->ft_hdrlen = sizeof(*ip6); 3518 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3519 sizeof(struct ip) + sizeof(struct ip6_frag); 3520 } 3521 break; 3522 default: 3523 return (-1); 3524 } 3525 3526 /* recalculate icmp/icmp6 checksums */ 3527 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3528 int off; 3529 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3530 NULL) { 3531 pd->m = NULL; 3532 return (-1); 3533 } 3534 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3535 icmp->icmp6_cksum = 0; 3536 icmp->icmp6_cksum = pd->naf == AF_INET ? 3537 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3538 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3539 ntohs(ip6->ip6_plen)); 3540 } 3541 #endif /* INET && INET6 */ 3542 3543 return (0); 3544 } 3545 3546 int 3547 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3548 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3549 sa_family_t af, sa_family_t naf) 3550 { 3551 #if defined(INET) && defined(INET6) 3552 struct mbuf *n = NULL; 3553 struct ip *ip4; 3554 struct ip6_hdr *ip6; 3555 int hlen, olen, mlen; 3556 3557 if (af == naf || (af != AF_INET && af != AF_INET6) || 3558 (naf != AF_INET && naf != AF_INET6)) 3559 return (-1); 3560 3561 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3562 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3563 return (-1); 3564 3565 /* old header */ 3566 olen = pd2->off - off; 3567 /* new header */ 3568 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3569 /* data lenght */ 3570 mlen = m->m_pkthdr.len - pd2->off; 3571 3572 /* trim old header */ 3573 m_adj(n, olen); 3574 3575 /* prepend a new one */ 3576 M_PREPEND(n, hlen, M_NOWAIT); 3577 if (n == NULL) 3578 return (-1); 3579 3580 /* translate inner ip/ip6 header */ 3581 switch (naf) { 3582 case AF_INET: 3583 ip4 = mtod(n, struct ip *); 3584 bzero(ip4, sizeof(*ip4)); 3585 ip4->ip_v = IPVERSION; 3586 ip4->ip_hl = sizeof(*ip4) >> 2; 3587 ip4->ip_len = htons(sizeof(*ip4) + mlen); 3588 ip_fillid(ip4); 3589 ip4->ip_off = htons(IP_DF); 3590 ip4->ip_ttl = pd2->ttl; 3591 if (pd2->proto == IPPROTO_ICMPV6) 3592 ip4->ip_p = IPPROTO_ICMP; 3593 else 3594 ip4->ip_p = pd2->proto; 3595 ip4->ip_src = src->v4; 3596 ip4->ip_dst = dst->v4; 3597 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3598 break; 3599 case AF_INET6: 3600 ip6 = mtod(n, struct ip6_hdr *); 3601 bzero(ip6, sizeof(*ip6)); 3602 ip6->ip6_vfc = IPV6_VERSION; 3603 ip6->ip6_plen = htons(mlen); 3604 if (pd2->proto == IPPROTO_ICMP) 3605 ip6->ip6_nxt = IPPROTO_ICMPV6; 3606 else 3607 ip6->ip6_nxt = pd2->proto; 3608 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3609 ip6->ip6_hlim = IPV6_DEFHLIM; 3610 else 3611 ip6->ip6_hlim = pd2->ttl; 3612 ip6->ip6_src = src->v6; 3613 ip6->ip6_dst = dst->v6; 3614 break; 3615 } 3616 3617 /* adjust payload offset and total packet length */ 3618 pd2->off += hlen - olen; 3619 pd->tot_len += hlen - olen; 3620 3621 /* merge modified inner packet with the original header */ 3622 mlen = n->m_pkthdr.len; 3623 m_cat(m, n); 3624 m->m_pkthdr.len += mlen; 3625 #endif /* INET && INET6 */ 3626 3627 return (0); 3628 } 3629 3630 #define PTR_IP(field) (offsetof(struct ip, field)) 3631 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3632 3633 int 3634 pf_translate_icmp_af(int af, void *arg) 3635 { 3636 #if defined(INET) && defined(INET6) 3637 struct icmp *icmp4; 3638 struct icmp6_hdr *icmp6; 3639 u_int32_t mtu; 3640 int32_t ptr = -1; 3641 u_int8_t type; 3642 u_int8_t code; 3643 3644 switch (af) { 3645 case AF_INET: 3646 icmp6 = arg; 3647 type = icmp6->icmp6_type; 3648 code = icmp6->icmp6_code; 3649 mtu = ntohl(icmp6->icmp6_mtu); 3650 3651 switch (type) { 3652 case ICMP6_ECHO_REQUEST: 3653 type = ICMP_ECHO; 3654 break; 3655 case ICMP6_ECHO_REPLY: 3656 type = ICMP_ECHOREPLY; 3657 break; 3658 case ICMP6_DST_UNREACH: 3659 type = ICMP_UNREACH; 3660 switch (code) { 3661 case ICMP6_DST_UNREACH_NOROUTE: 3662 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3663 case ICMP6_DST_UNREACH_ADDR: 3664 code = ICMP_UNREACH_HOST; 3665 break; 3666 case ICMP6_DST_UNREACH_ADMIN: 3667 code = ICMP_UNREACH_HOST_PROHIB; 3668 break; 3669 case ICMP6_DST_UNREACH_NOPORT: 3670 code = ICMP_UNREACH_PORT; 3671 break; 3672 default: 3673 return (-1); 3674 } 3675 break; 3676 case ICMP6_PACKET_TOO_BIG: 3677 type = ICMP_UNREACH; 3678 code = ICMP_UNREACH_NEEDFRAG; 3679 mtu -= 20; 3680 break; 3681 case ICMP6_TIME_EXCEEDED: 3682 type = ICMP_TIMXCEED; 3683 break; 3684 case ICMP6_PARAM_PROB: 3685 switch (code) { 3686 case ICMP6_PARAMPROB_HEADER: 3687 type = ICMP_PARAMPROB; 3688 code = ICMP_PARAMPROB_ERRATPTR; 3689 ptr = ntohl(icmp6->icmp6_pptr); 3690 3691 if (ptr == PTR_IP6(ip6_vfc)) 3692 ; /* preserve */ 3693 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3694 ptr = PTR_IP(ip_tos); 3695 else if (ptr == PTR_IP6(ip6_plen) || 3696 ptr == PTR_IP6(ip6_plen) + 1) 3697 ptr = PTR_IP(ip_len); 3698 else if (ptr == PTR_IP6(ip6_nxt)) 3699 ptr = PTR_IP(ip_p); 3700 else if (ptr == PTR_IP6(ip6_hlim)) 3701 ptr = PTR_IP(ip_ttl); 3702 else if (ptr >= PTR_IP6(ip6_src) && 3703 ptr < PTR_IP6(ip6_dst)) 3704 ptr = PTR_IP(ip_src); 3705 else if (ptr >= PTR_IP6(ip6_dst) && 3706 ptr < sizeof(struct ip6_hdr)) 3707 ptr = PTR_IP(ip_dst); 3708 else { 3709 return (-1); 3710 } 3711 break; 3712 case ICMP6_PARAMPROB_NEXTHEADER: 3713 type = ICMP_UNREACH; 3714 code = ICMP_UNREACH_PROTOCOL; 3715 break; 3716 default: 3717 return (-1); 3718 } 3719 break; 3720 default: 3721 return (-1); 3722 } 3723 if (icmp6->icmp6_type != type) { 3724 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3725 icmp6->icmp6_type, type, 0); 3726 icmp6->icmp6_type = type; 3727 } 3728 if (icmp6->icmp6_code != code) { 3729 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3730 icmp6->icmp6_code, code, 0); 3731 icmp6->icmp6_code = code; 3732 } 3733 if (icmp6->icmp6_mtu != htonl(mtu)) { 3734 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3735 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3736 /* aligns well with a icmpv4 nextmtu */ 3737 icmp6->icmp6_mtu = htonl(mtu); 3738 } 3739 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3740 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3741 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3742 /* icmpv4 pptr is a one most significant byte */ 3743 icmp6->icmp6_pptr = htonl(ptr << 24); 3744 } 3745 break; 3746 case AF_INET6: 3747 icmp4 = arg; 3748 type = icmp4->icmp_type; 3749 code = icmp4->icmp_code; 3750 mtu = ntohs(icmp4->icmp_nextmtu); 3751 3752 switch (type) { 3753 case ICMP_ECHO: 3754 type = ICMP6_ECHO_REQUEST; 3755 break; 3756 case ICMP_ECHOREPLY: 3757 type = ICMP6_ECHO_REPLY; 3758 break; 3759 case ICMP_UNREACH: 3760 type = ICMP6_DST_UNREACH; 3761 switch (code) { 3762 case ICMP_UNREACH_NET: 3763 case ICMP_UNREACH_HOST: 3764 case ICMP_UNREACH_NET_UNKNOWN: 3765 case ICMP_UNREACH_HOST_UNKNOWN: 3766 case ICMP_UNREACH_ISOLATED: 3767 case ICMP_UNREACH_TOSNET: 3768 case ICMP_UNREACH_TOSHOST: 3769 code = ICMP6_DST_UNREACH_NOROUTE; 3770 break; 3771 case ICMP_UNREACH_PORT: 3772 code = ICMP6_DST_UNREACH_NOPORT; 3773 break; 3774 case ICMP_UNREACH_NET_PROHIB: 3775 case ICMP_UNREACH_HOST_PROHIB: 3776 case ICMP_UNREACH_FILTER_PROHIB: 3777 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3778 code = ICMP6_DST_UNREACH_ADMIN; 3779 break; 3780 case ICMP_UNREACH_PROTOCOL: 3781 type = ICMP6_PARAM_PROB; 3782 code = ICMP6_PARAMPROB_NEXTHEADER; 3783 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3784 break; 3785 case ICMP_UNREACH_NEEDFRAG: 3786 type = ICMP6_PACKET_TOO_BIG; 3787 code = 0; 3788 mtu += 20; 3789 break; 3790 default: 3791 return (-1); 3792 } 3793 break; 3794 case ICMP_TIMXCEED: 3795 type = ICMP6_TIME_EXCEEDED; 3796 break; 3797 case ICMP_PARAMPROB: 3798 type = ICMP6_PARAM_PROB; 3799 switch (code) { 3800 case ICMP_PARAMPROB_ERRATPTR: 3801 code = ICMP6_PARAMPROB_HEADER; 3802 break; 3803 case ICMP_PARAMPROB_LENGTH: 3804 code = ICMP6_PARAMPROB_HEADER; 3805 break; 3806 default: 3807 return (-1); 3808 } 3809 3810 ptr = icmp4->icmp_pptr; 3811 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3812 ; /* preserve */ 3813 else if (ptr == PTR_IP(ip_len) || 3814 ptr == PTR_IP(ip_len) + 1) 3815 ptr = PTR_IP6(ip6_plen); 3816 else if (ptr == PTR_IP(ip_ttl)) 3817 ptr = PTR_IP6(ip6_hlim); 3818 else if (ptr == PTR_IP(ip_p)) 3819 ptr = PTR_IP6(ip6_nxt); 3820 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 3821 ptr = PTR_IP6(ip6_src); 3822 else if (ptr >= PTR_IP(ip_dst) && 3823 ptr < sizeof(struct ip)) 3824 ptr = PTR_IP6(ip6_dst); 3825 else { 3826 return (-1); 3827 } 3828 break; 3829 default: 3830 return (-1); 3831 } 3832 if (icmp4->icmp_type != type) { 3833 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3834 icmp4->icmp_type, type, 0); 3835 icmp4->icmp_type = type; 3836 } 3837 if (icmp4->icmp_code != code) { 3838 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3839 icmp4->icmp_code, code, 0); 3840 icmp4->icmp_code = code; 3841 } 3842 if (icmp4->icmp_nextmtu != htons(mtu)) { 3843 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3844 icmp4->icmp_nextmtu, htons(mtu), 0); 3845 icmp4->icmp_nextmtu = htons(mtu); 3846 } 3847 if (ptr >= 0 && icmp4->icmp_void != ptr) { 3848 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3849 htons(icmp4->icmp_pptr), htons(ptr), 0); 3850 icmp4->icmp_void = htonl(ptr); 3851 } 3852 break; 3853 } 3854 #endif /* INET && INET6 */ 3855 3856 return (0); 3857 } 3858 3859 /* 3860 * Need to modulate the sequence numbers in the TCP SACK option 3861 * (credits to Krzysztof Pfaff for report and patch) 3862 */ 3863 static int 3864 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3865 struct pf_state_peer *dst) 3866 { 3867 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3868 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3869 int copyback = 0, i, olen; 3870 struct sackblk sack; 3871 3872 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3873 if (hlen < TCPOLEN_SACKLEN || 3874 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3875 return 0; 3876 3877 while (hlen >= TCPOLEN_SACKLEN) { 3878 size_t startoff = opt - opts; 3879 olen = opt[1]; 3880 switch (*opt) { 3881 case TCPOPT_EOL: /* FALLTHROUGH */ 3882 case TCPOPT_NOP: 3883 opt++; 3884 hlen--; 3885 break; 3886 case TCPOPT_SACK: 3887 if (olen > hlen) 3888 olen = hlen; 3889 if (olen >= TCPOLEN_SACKLEN) { 3890 for (i = 2; i + TCPOLEN_SACK <= olen; 3891 i += TCPOLEN_SACK) { 3892 memcpy(&sack, &opt[i], sizeof(sack)); 3893 pf_patch_32_unaligned(pd->m, 3894 &th->th_sum, &sack.start, 3895 htonl(ntohl(sack.start) - dst->seqdiff), 3896 PF_ALGNMNT(startoff), 3897 0); 3898 pf_patch_32_unaligned(pd->m, &th->th_sum, 3899 &sack.end, 3900 htonl(ntohl(sack.end) - dst->seqdiff), 3901 PF_ALGNMNT(startoff), 3902 0); 3903 memcpy(&opt[i], &sack, sizeof(sack)); 3904 } 3905 copyback = 1; 3906 } 3907 /* FALLTHROUGH */ 3908 default: 3909 if (olen < 2) 3910 olen = 2; 3911 hlen -= olen; 3912 opt += olen; 3913 } 3914 } 3915 3916 if (copyback) 3917 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3918 return (copyback); 3919 } 3920 3921 struct mbuf * 3922 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3923 const struct pf_addr *saddr, const struct pf_addr *daddr, 3924 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3925 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3926 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3927 { 3928 struct mbuf *m; 3929 int len, tlen; 3930 #ifdef INET 3931 struct ip *h = NULL; 3932 #endif /* INET */ 3933 #ifdef INET6 3934 struct ip6_hdr *h6 = NULL; 3935 #endif /* INET6 */ 3936 struct tcphdr *th; 3937 char *opt; 3938 struct pf_mtag *pf_mtag; 3939 3940 len = 0; 3941 th = NULL; 3942 3943 /* maximum segment size tcp option */ 3944 tlen = sizeof(struct tcphdr); 3945 if (mss) 3946 tlen += 4; 3947 3948 switch (af) { 3949 #ifdef INET 3950 case AF_INET: 3951 len = sizeof(struct ip) + tlen; 3952 break; 3953 #endif /* INET */ 3954 #ifdef INET6 3955 case AF_INET6: 3956 len = sizeof(struct ip6_hdr) + tlen; 3957 break; 3958 #endif /* INET6 */ 3959 } 3960 3961 m = m_gethdr(M_NOWAIT, MT_DATA); 3962 if (m == NULL) 3963 return (NULL); 3964 3965 #ifdef MAC 3966 mac_netinet_firewall_send(m); 3967 #endif 3968 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3969 m_freem(m); 3970 return (NULL); 3971 } 3972 m->m_flags |= mbuf_flags; 3973 pf_mtag->tag = mtag_tag; 3974 pf_mtag->flags = mtag_flags; 3975 3976 if (rtableid >= 0) 3977 M_SETFIB(m, rtableid); 3978 3979 #ifdef ALTQ 3980 if (r != NULL && r->qid) { 3981 pf_mtag->qid = r->qid; 3982 3983 /* add hints for ecn */ 3984 pf_mtag->hdr = mtod(m, struct ip *); 3985 } 3986 #endif /* ALTQ */ 3987 m->m_data += max_linkhdr; 3988 m->m_pkthdr.len = m->m_len = len; 3989 /* The rest of the stack assumes a rcvif, so provide one. 3990 * This is a locally generated packet, so .. close enough. */ 3991 m->m_pkthdr.rcvif = V_loif; 3992 bzero(m->m_data, len); 3993 switch (af) { 3994 #ifdef INET 3995 case AF_INET: 3996 h = mtod(m, struct ip *); 3997 3998 /* IP header fields included in the TCP checksum */ 3999 h->ip_p = IPPROTO_TCP; 4000 h->ip_len = htons(tlen); 4001 h->ip_src.s_addr = saddr->v4.s_addr; 4002 h->ip_dst.s_addr = daddr->v4.s_addr; 4003 4004 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4005 break; 4006 #endif /* INET */ 4007 #ifdef INET6 4008 case AF_INET6: 4009 h6 = mtod(m, struct ip6_hdr *); 4010 4011 /* IP header fields included in the TCP checksum */ 4012 h6->ip6_nxt = IPPROTO_TCP; 4013 h6->ip6_plen = htons(tlen); 4014 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4015 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4016 4017 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4018 break; 4019 #endif /* INET6 */ 4020 } 4021 4022 /* TCP header */ 4023 th->th_sport = sport; 4024 th->th_dport = dport; 4025 th->th_seq = htonl(seq); 4026 th->th_ack = htonl(ack); 4027 th->th_off = tlen >> 2; 4028 tcp_set_flags(th, tcp_flags); 4029 th->th_win = htons(win); 4030 4031 if (mss) { 4032 opt = (char *)(th + 1); 4033 opt[0] = TCPOPT_MAXSEG; 4034 opt[1] = 4; 4035 HTONS(mss); 4036 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 4037 } 4038 4039 switch (af) { 4040 #ifdef INET 4041 case AF_INET: 4042 /* TCP checksum */ 4043 th->th_sum = in_cksum(m, len); 4044 4045 /* Finish the IP header */ 4046 h->ip_v = 4; 4047 h->ip_hl = sizeof(*h) >> 2; 4048 h->ip_tos = IPTOS_LOWDELAY; 4049 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4050 h->ip_len = htons(len); 4051 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4052 h->ip_sum = 0; 4053 break; 4054 #endif /* INET */ 4055 #ifdef INET6 4056 case AF_INET6: 4057 /* TCP checksum */ 4058 th->th_sum = in6_cksum(m, IPPROTO_TCP, 4059 sizeof(struct ip6_hdr), tlen); 4060 4061 h6->ip6_vfc |= IPV6_VERSION; 4062 h6->ip6_hlim = IPV6_DEFHLIM; 4063 break; 4064 #endif /* INET6 */ 4065 } 4066 4067 return (m); 4068 } 4069 4070 static void 4071 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4072 uint8_t ttl, int rtableid) 4073 { 4074 struct mbuf *m; 4075 #ifdef INET 4076 struct ip *h = NULL; 4077 #endif /* INET */ 4078 #ifdef INET6 4079 struct ip6_hdr *h6 = NULL; 4080 #endif /* INET6 */ 4081 struct sctphdr *hdr; 4082 struct sctp_chunkhdr *chunk; 4083 struct pf_send_entry *pfse; 4084 int off = 0; 4085 4086 MPASS(af == pd->af); 4087 4088 m = m_gethdr(M_NOWAIT, MT_DATA); 4089 if (m == NULL) 4090 return; 4091 4092 m->m_data += max_linkhdr; 4093 m->m_flags |= M_SKIP_FIREWALL; 4094 /* The rest of the stack assumes a rcvif, so provide one. 4095 * This is a locally generated packet, so .. close enough. */ 4096 m->m_pkthdr.rcvif = V_loif; 4097 4098 /* IPv4|6 header */ 4099 switch (af) { 4100 #ifdef INET 4101 case AF_INET: 4102 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4103 4104 h = mtod(m, struct ip *); 4105 4106 /* IP header fields included in the TCP checksum */ 4107 4108 h->ip_p = IPPROTO_SCTP; 4109 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4110 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4111 h->ip_src = pd->dst->v4; 4112 h->ip_dst = pd->src->v4; 4113 4114 off += sizeof(struct ip); 4115 break; 4116 #endif /* INET */ 4117 #ifdef INET6 4118 case AF_INET6: 4119 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4120 4121 h6 = mtod(m, struct ip6_hdr *); 4122 4123 /* IP header fields included in the TCP checksum */ 4124 h6->ip6_vfc |= IPV6_VERSION; 4125 h6->ip6_nxt = IPPROTO_SCTP; 4126 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4127 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4128 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4129 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4130 4131 off += sizeof(struct ip6_hdr); 4132 break; 4133 #endif /* INET6 */ 4134 } 4135 4136 /* SCTP header */ 4137 hdr = mtodo(m, off); 4138 4139 hdr->src_port = pd->hdr.sctp.dest_port; 4140 hdr->dest_port = pd->hdr.sctp.src_port; 4141 hdr->v_tag = pd->sctp_initiate_tag; 4142 hdr->checksum = 0; 4143 4144 /* Abort chunk. */ 4145 off += sizeof(struct sctphdr); 4146 chunk = mtodo(m, off); 4147 4148 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4149 chunk->chunk_length = htons(sizeof(*chunk)); 4150 4151 /* SCTP checksum */ 4152 off += sizeof(*chunk); 4153 m->m_pkthdr.len = m->m_len = off; 4154 4155 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4156 4157 if (rtableid >= 0) 4158 M_SETFIB(m, rtableid); 4159 4160 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4161 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4162 if (pfse == NULL) { 4163 m_freem(m); 4164 return; 4165 } 4166 4167 switch (af) { 4168 #ifdef INET 4169 case AF_INET: 4170 pfse->pfse_type = PFSE_IP; 4171 break; 4172 #endif /* INET */ 4173 #ifdef INET6 4174 case AF_INET6: 4175 pfse->pfse_type = PFSE_IP6; 4176 break; 4177 #endif /* INET6 */ 4178 } 4179 4180 pfse->pfse_m = m; 4181 pf_send(pfse); 4182 } 4183 4184 void 4185 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4186 const struct pf_addr *saddr, const struct pf_addr *daddr, 4187 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4188 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4189 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 4190 { 4191 struct pf_send_entry *pfse; 4192 struct mbuf *m; 4193 4194 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4195 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 4196 if (m == NULL) 4197 return; 4198 4199 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4200 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4201 if (pfse == NULL) { 4202 m_freem(m); 4203 return; 4204 } 4205 4206 switch (af) { 4207 #ifdef INET 4208 case AF_INET: 4209 pfse->pfse_type = PFSE_IP; 4210 break; 4211 #endif /* INET */ 4212 #ifdef INET6 4213 case AF_INET6: 4214 pfse->pfse_type = PFSE_IP6; 4215 break; 4216 #endif /* INET6 */ 4217 } 4218 4219 pfse->pfse_m = m; 4220 pf_send(pfse); 4221 } 4222 4223 static void 4224 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4225 struct pf_state_key *sk, struct tcphdr *th, 4226 u_int16_t bproto_sum, u_int16_t bip_sum, 4227 u_short *reason, int rtableid) 4228 { 4229 struct pf_addr * const saddr = pd->src; 4230 struct pf_addr * const daddr = pd->dst; 4231 4232 /* undo NAT changes, if they have taken place */ 4233 if (nr != NULL) { 4234 PF_ACPY(saddr, &sk->addr[pd->sidx], pd->af); 4235 PF_ACPY(daddr, &sk->addr[pd->didx], pd->af); 4236 if (pd->sport) 4237 *pd->sport = sk->port[pd->sidx]; 4238 if (pd->dport) 4239 *pd->dport = sk->port[pd->didx]; 4240 if (pd->ip_sum) 4241 *pd->ip_sum = bip_sum; 4242 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4243 } 4244 if (pd->proto == IPPROTO_TCP && 4245 ((r->rule_flag & PFRULE_RETURNRST) || 4246 (r->rule_flag & PFRULE_RETURN)) && 4247 !(tcp_get_flags(th) & TH_RST)) { 4248 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4249 4250 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4251 IPPROTO_TCP, pd->af)) 4252 REASON_SET(reason, PFRES_PROTCKSUM); 4253 else { 4254 if (tcp_get_flags(th) & TH_SYN) 4255 ack++; 4256 if (tcp_get_flags(th) & TH_FIN) 4257 ack++; 4258 pf_send_tcp(r, pd->af, pd->dst, 4259 pd->src, th->th_dport, th->th_sport, 4260 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4261 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 4262 } 4263 } else if (pd->proto == IPPROTO_SCTP && 4264 (r->rule_flag & PFRULE_RETURN)) { 4265 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4266 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4267 r->return_icmp) 4268 pf_send_icmp(pd->m, r->return_icmp >> 8, 4269 r->return_icmp & 255, pd->af, r, rtableid); 4270 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4271 r->return_icmp6) 4272 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4273 r->return_icmp6 & 255, pd->af, r, rtableid); 4274 } 4275 4276 static int 4277 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4278 { 4279 struct m_tag *mtag; 4280 u_int8_t mpcp; 4281 4282 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4283 if (mtag == NULL) 4284 return (0); 4285 4286 if (prio == PF_PRIO_ZERO) 4287 prio = 0; 4288 4289 mpcp = *(uint8_t *)(mtag + 1); 4290 4291 return (mpcp == prio); 4292 } 4293 4294 static int 4295 pf_icmp_to_bandlim(uint8_t type) 4296 { 4297 switch (type) { 4298 case ICMP_ECHO: 4299 case ICMP_ECHOREPLY: 4300 return (BANDLIM_ICMP_ECHO); 4301 case ICMP_TSTAMP: 4302 case ICMP_TSTAMPREPLY: 4303 return (BANDLIM_ICMP_TSTAMP); 4304 case ICMP_UNREACH: 4305 default: 4306 return (BANDLIM_ICMP_UNREACH); 4307 } 4308 } 4309 4310 static void 4311 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 4312 struct pf_krule *r, int rtableid) 4313 { 4314 struct pf_send_entry *pfse; 4315 struct mbuf *m0; 4316 struct pf_mtag *pf_mtag; 4317 4318 /* ICMP packet rate limitation. */ 4319 switch (af) { 4320 #ifdef INET6 4321 case AF_INET6: 4322 if (icmp6_ratelimit(NULL, type, code)) 4323 return; 4324 break; 4325 #endif 4326 #ifdef INET 4327 case AF_INET: 4328 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4329 return; 4330 break; 4331 #endif 4332 } 4333 4334 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4335 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4336 if (pfse == NULL) 4337 return; 4338 4339 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4340 free(pfse, M_PFTEMP); 4341 return; 4342 } 4343 4344 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4345 free(pfse, M_PFTEMP); 4346 return; 4347 } 4348 /* XXX: revisit */ 4349 m0->m_flags |= M_SKIP_FIREWALL; 4350 4351 if (rtableid >= 0) 4352 M_SETFIB(m0, rtableid); 4353 4354 #ifdef ALTQ 4355 if (r->qid) { 4356 pf_mtag->qid = r->qid; 4357 /* add hints for ecn */ 4358 pf_mtag->hdr = mtod(m0, struct ip *); 4359 } 4360 #endif /* ALTQ */ 4361 4362 switch (af) { 4363 #ifdef INET 4364 case AF_INET: 4365 pfse->pfse_type = PFSE_ICMP; 4366 break; 4367 #endif /* INET */ 4368 #ifdef INET6 4369 case AF_INET6: 4370 pfse->pfse_type = PFSE_ICMP6; 4371 break; 4372 #endif /* INET6 */ 4373 } 4374 pfse->pfse_m = m0; 4375 pfse->icmpopts.type = type; 4376 pfse->icmpopts.code = code; 4377 pf_send(pfse); 4378 } 4379 4380 /* 4381 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 4382 * If n is 0, they match if they are equal. If n is != 0, they match if they 4383 * are different. 4384 */ 4385 int 4386 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4387 const struct pf_addr *b, sa_family_t af) 4388 { 4389 int match = 0; 4390 4391 switch (af) { 4392 #ifdef INET 4393 case AF_INET: 4394 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4395 match++; 4396 break; 4397 #endif /* INET */ 4398 #ifdef INET6 4399 case AF_INET6: 4400 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4401 match++; 4402 break; 4403 #endif /* INET6 */ 4404 } 4405 if (match) { 4406 if (n) 4407 return (0); 4408 else 4409 return (1); 4410 } else { 4411 if (n) 4412 return (1); 4413 else 4414 return (0); 4415 } 4416 } 4417 4418 /* 4419 * Return 1 if b <= a <= e, otherwise return 0. 4420 */ 4421 int 4422 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4423 const struct pf_addr *a, sa_family_t af) 4424 { 4425 switch (af) { 4426 #ifdef INET 4427 case AF_INET: 4428 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4429 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4430 return (0); 4431 break; 4432 #endif /* INET */ 4433 #ifdef INET6 4434 case AF_INET6: { 4435 int i; 4436 4437 /* check a >= b */ 4438 for (i = 0; i < 4; ++i) 4439 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4440 break; 4441 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4442 return (0); 4443 /* check a <= e */ 4444 for (i = 0; i < 4; ++i) 4445 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4446 break; 4447 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4448 return (0); 4449 break; 4450 } 4451 #endif /* INET6 */ 4452 } 4453 return (1); 4454 } 4455 4456 static int 4457 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4458 { 4459 switch (op) { 4460 case PF_OP_IRG: 4461 return ((p > a1) && (p < a2)); 4462 case PF_OP_XRG: 4463 return ((p < a1) || (p > a2)); 4464 case PF_OP_RRG: 4465 return ((p >= a1) && (p <= a2)); 4466 case PF_OP_EQ: 4467 return (p == a1); 4468 case PF_OP_NE: 4469 return (p != a1); 4470 case PF_OP_LT: 4471 return (p < a1); 4472 case PF_OP_LE: 4473 return (p <= a1); 4474 case PF_OP_GT: 4475 return (p > a1); 4476 case PF_OP_GE: 4477 return (p >= a1); 4478 } 4479 return (0); /* never reached */ 4480 } 4481 4482 int 4483 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4484 { 4485 NTOHS(a1); 4486 NTOHS(a2); 4487 NTOHS(p); 4488 return (pf_match(op, a1, a2, p)); 4489 } 4490 4491 static int 4492 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4493 { 4494 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4495 return (0); 4496 return (pf_match(op, a1, a2, u)); 4497 } 4498 4499 static int 4500 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4501 { 4502 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4503 return (0); 4504 return (pf_match(op, a1, a2, g)); 4505 } 4506 4507 int 4508 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4509 { 4510 if (*tag == -1) 4511 *tag = mtag; 4512 4513 return ((!r->match_tag_not && r->match_tag == *tag) || 4514 (r->match_tag_not && r->match_tag != *tag)); 4515 } 4516 4517 static int 4518 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4519 { 4520 struct ifnet *ifp = m->m_pkthdr.rcvif; 4521 struct pfi_kkif *kif; 4522 4523 if (ifp == NULL) 4524 return (0); 4525 4526 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4527 4528 if (kif == NULL) { 4529 DPFPRINTF(PF_DEBUG_URGENT, 4530 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 4531 r->rcv_ifname)); 4532 return (0); 4533 } 4534 4535 return (pfi_kkif_match(r->rcv_kif, kif)); 4536 } 4537 4538 int 4539 pf_tag_packet(struct pf_pdesc *pd, int tag) 4540 { 4541 4542 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4543 4544 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4545 return (ENOMEM); 4546 4547 pd->pf_mtag->tag = tag; 4548 4549 return (0); 4550 } 4551 4552 #define PF_ANCHOR_STACKSIZE 32 4553 struct pf_kanchor_stackframe { 4554 struct pf_kruleset *rs; 4555 struct pf_krule *r; /* XXX: + match bit */ 4556 struct pf_kanchor *child; 4557 }; 4558 4559 /* 4560 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4561 */ 4562 #define PF_ANCHORSTACK_MATCH 0x00000001 4563 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4564 4565 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4566 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4567 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4568 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4569 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4570 } while (0) 4571 4572 void 4573 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4574 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4575 int *match) 4576 { 4577 struct pf_kanchor_stackframe *f; 4578 4579 PF_RULES_RASSERT(); 4580 4581 if (match) 4582 *match = 0; 4583 if (*depth >= PF_ANCHOR_STACKSIZE) { 4584 printf("%s: anchor stack overflow on %s\n", 4585 __func__, (*r)->anchor->name); 4586 *r = TAILQ_NEXT(*r, entries); 4587 return; 4588 } else if (*depth == 0 && a != NULL) 4589 *a = *r; 4590 f = stack + (*depth)++; 4591 f->rs = *rs; 4592 f->r = *r; 4593 if ((*r)->anchor_wildcard) { 4594 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4595 4596 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4597 *r = NULL; 4598 return; 4599 } 4600 *rs = &f->child->ruleset; 4601 } else { 4602 f->child = NULL; 4603 *rs = &(*r)->anchor->ruleset; 4604 } 4605 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4606 } 4607 4608 int 4609 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4610 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4611 int *match) 4612 { 4613 struct pf_kanchor_stackframe *f; 4614 struct pf_krule *fr; 4615 int quick = 0; 4616 4617 PF_RULES_RASSERT(); 4618 4619 do { 4620 if (*depth <= 0) 4621 break; 4622 f = stack + *depth - 1; 4623 fr = PF_ANCHOR_RULE(f); 4624 if (f->child != NULL) { 4625 /* 4626 * This block traverses through 4627 * a wildcard anchor. 4628 */ 4629 if (match != NULL && *match) { 4630 /* 4631 * If any of "*" matched, then 4632 * "foo/ *" matched, mark frame 4633 * appropriately. 4634 */ 4635 PF_ANCHOR_SET_MATCH(f); 4636 *match = 0; 4637 } 4638 f->child = RB_NEXT(pf_kanchor_node, 4639 &fr->anchor->children, f->child); 4640 if (f->child != NULL) { 4641 *rs = &f->child->ruleset; 4642 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4643 if (*r == NULL) 4644 continue; 4645 else 4646 break; 4647 } 4648 } 4649 (*depth)--; 4650 if (*depth == 0 && a != NULL) 4651 *a = NULL; 4652 *rs = f->rs; 4653 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4654 quick = fr->quick; 4655 *r = TAILQ_NEXT(fr, entries); 4656 } while (*r == NULL); 4657 4658 return (quick); 4659 } 4660 4661 struct pf_keth_anchor_stackframe { 4662 struct pf_keth_ruleset *rs; 4663 struct pf_keth_rule *r; /* XXX: + match bit */ 4664 struct pf_keth_anchor *child; 4665 }; 4666 4667 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4668 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4669 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4670 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4671 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4672 } while (0) 4673 4674 void 4675 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4676 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4677 struct pf_keth_rule **a, int *match) 4678 { 4679 struct pf_keth_anchor_stackframe *f; 4680 4681 NET_EPOCH_ASSERT(); 4682 4683 if (match) 4684 *match = 0; 4685 if (*depth >= PF_ANCHOR_STACKSIZE) { 4686 printf("%s: anchor stack overflow on %s\n", 4687 __func__, (*r)->anchor->name); 4688 *r = TAILQ_NEXT(*r, entries); 4689 return; 4690 } else if (*depth == 0 && a != NULL) 4691 *a = *r; 4692 f = stack + (*depth)++; 4693 f->rs = *rs; 4694 f->r = *r; 4695 if ((*r)->anchor_wildcard) { 4696 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4697 4698 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4699 *r = NULL; 4700 return; 4701 } 4702 *rs = &f->child->ruleset; 4703 } else { 4704 f->child = NULL; 4705 *rs = &(*r)->anchor->ruleset; 4706 } 4707 *r = TAILQ_FIRST((*rs)->active.rules); 4708 } 4709 4710 int 4711 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4712 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4713 struct pf_keth_rule **a, int *match) 4714 { 4715 struct pf_keth_anchor_stackframe *f; 4716 struct pf_keth_rule *fr; 4717 int quick = 0; 4718 4719 NET_EPOCH_ASSERT(); 4720 4721 do { 4722 if (*depth <= 0) 4723 break; 4724 f = stack + *depth - 1; 4725 fr = PF_ETH_ANCHOR_RULE(f); 4726 if (f->child != NULL) { 4727 /* 4728 * This block traverses through 4729 * a wildcard anchor. 4730 */ 4731 if (match != NULL && *match) { 4732 /* 4733 * If any of "*" matched, then 4734 * "foo/ *" matched, mark frame 4735 * appropriately. 4736 */ 4737 PF_ETH_ANCHOR_SET_MATCH(f); 4738 *match = 0; 4739 } 4740 f->child = RB_NEXT(pf_keth_anchor_node, 4741 &fr->anchor->children, f->child); 4742 if (f->child != NULL) { 4743 *rs = &f->child->ruleset; 4744 *r = TAILQ_FIRST((*rs)->active.rules); 4745 if (*r == NULL) 4746 continue; 4747 else 4748 break; 4749 } 4750 } 4751 (*depth)--; 4752 if (*depth == 0 && a != NULL) 4753 *a = NULL; 4754 *rs = f->rs; 4755 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4756 quick = fr->quick; 4757 *r = TAILQ_NEXT(fr, entries); 4758 } while (*r == NULL); 4759 4760 return (quick); 4761 } 4762 4763 #ifdef INET6 4764 void 4765 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4766 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4767 { 4768 switch (af) { 4769 #ifdef INET 4770 case AF_INET: 4771 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4772 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4773 break; 4774 #endif /* INET */ 4775 case AF_INET6: 4776 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4777 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4778 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4779 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4780 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4781 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4782 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4783 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4784 break; 4785 } 4786 } 4787 4788 void 4789 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4790 { 4791 switch (af) { 4792 #ifdef INET 4793 case AF_INET: 4794 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4795 break; 4796 #endif /* INET */ 4797 case AF_INET6: 4798 if (addr->addr32[3] == 0xffffffff) { 4799 addr->addr32[3] = 0; 4800 if (addr->addr32[2] == 0xffffffff) { 4801 addr->addr32[2] = 0; 4802 if (addr->addr32[1] == 0xffffffff) { 4803 addr->addr32[1] = 0; 4804 addr->addr32[0] = 4805 htonl(ntohl(addr->addr32[0]) + 1); 4806 } else 4807 addr->addr32[1] = 4808 htonl(ntohl(addr->addr32[1]) + 1); 4809 } else 4810 addr->addr32[2] = 4811 htonl(ntohl(addr->addr32[2]) + 1); 4812 } else 4813 addr->addr32[3] = 4814 htonl(ntohl(addr->addr32[3]) + 1); 4815 break; 4816 } 4817 } 4818 #endif /* INET6 */ 4819 4820 void 4821 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4822 { 4823 /* 4824 * Modern rules use the same flags in rules as they do in states. 4825 */ 4826 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4827 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4828 4829 /* 4830 * Old-style scrub rules have different flags which need to be translated. 4831 */ 4832 if (r->rule_flag & PFRULE_RANDOMID) 4833 a->flags |= PFSTATE_RANDOMID; 4834 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4835 a->flags |= PFSTATE_SETTOS; 4836 a->set_tos = r->set_tos; 4837 } 4838 4839 if (r->qid) 4840 a->qid = r->qid; 4841 if (r->pqid) 4842 a->pqid = r->pqid; 4843 if (r->rtableid >= 0) 4844 a->rtableid = r->rtableid; 4845 a->log |= r->log; 4846 if (r->min_ttl) 4847 a->min_ttl = r->min_ttl; 4848 if (r->max_mss) 4849 a->max_mss = r->max_mss; 4850 if (r->dnpipe) 4851 a->dnpipe = r->dnpipe; 4852 if (r->dnrpipe) 4853 a->dnrpipe = r->dnrpipe; 4854 if (r->dnpipe || r->dnrpipe) { 4855 if (r->free_flags & PFRULE_DN_IS_PIPE) 4856 a->flags |= PFSTATE_DN_IS_PIPE; 4857 else 4858 a->flags &= ~PFSTATE_DN_IS_PIPE; 4859 } 4860 if (r->scrub_flags & PFSTATE_SETPRIO) { 4861 a->set_prio[0] = r->set_prio[0]; 4862 a->set_prio[1] = r->set_prio[1]; 4863 } 4864 } 4865 4866 int 4867 pf_socket_lookup(struct pf_pdesc *pd) 4868 { 4869 struct pf_addr *saddr, *daddr; 4870 u_int16_t sport, dport; 4871 struct inpcbinfo *pi; 4872 struct inpcb *inp; 4873 4874 pd->lookup.uid = UID_MAX; 4875 pd->lookup.gid = GID_MAX; 4876 4877 switch (pd->proto) { 4878 case IPPROTO_TCP: 4879 sport = pd->hdr.tcp.th_sport; 4880 dport = pd->hdr.tcp.th_dport; 4881 pi = &V_tcbinfo; 4882 break; 4883 case IPPROTO_UDP: 4884 sport = pd->hdr.udp.uh_sport; 4885 dport = pd->hdr.udp.uh_dport; 4886 pi = &V_udbinfo; 4887 break; 4888 default: 4889 return (-1); 4890 } 4891 if (pd->dir == PF_IN) { 4892 saddr = pd->src; 4893 daddr = pd->dst; 4894 } else { 4895 u_int16_t p; 4896 4897 p = sport; 4898 sport = dport; 4899 dport = p; 4900 saddr = pd->dst; 4901 daddr = pd->src; 4902 } 4903 switch (pd->af) { 4904 #ifdef INET 4905 case AF_INET: 4906 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4907 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4908 if (inp == NULL) { 4909 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4910 daddr->v4, dport, INPLOOKUP_WILDCARD | 4911 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4912 if (inp == NULL) 4913 return (-1); 4914 } 4915 break; 4916 #endif /* INET */ 4917 #ifdef INET6 4918 case AF_INET6: 4919 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4920 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4921 if (inp == NULL) { 4922 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4923 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4924 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4925 if (inp == NULL) 4926 return (-1); 4927 } 4928 break; 4929 #endif /* INET6 */ 4930 } 4931 INP_RLOCK_ASSERT(inp); 4932 pd->lookup.uid = inp->inp_cred->cr_uid; 4933 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4934 INP_RUNLOCK(inp); 4935 4936 return (1); 4937 } 4938 4939 u_int8_t 4940 pf_get_wscale(struct pf_pdesc *pd) 4941 { 4942 struct tcphdr *th = &pd->hdr.tcp; 4943 int hlen; 4944 u_int8_t hdr[60]; 4945 u_int8_t *opt, optlen; 4946 u_int8_t wscale = 0; 4947 4948 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4949 if (hlen <= sizeof(struct tcphdr)) 4950 return (0); 4951 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4952 return (0); 4953 opt = hdr + sizeof(struct tcphdr); 4954 hlen -= sizeof(struct tcphdr); 4955 while (hlen >= 3) { 4956 switch (*opt) { 4957 case TCPOPT_EOL: 4958 case TCPOPT_NOP: 4959 ++opt; 4960 --hlen; 4961 break; 4962 case TCPOPT_WINDOW: 4963 wscale = opt[2]; 4964 if (wscale > TCP_MAX_WINSHIFT) 4965 wscale = TCP_MAX_WINSHIFT; 4966 wscale |= PF_WSCALE_FLAG; 4967 /* FALLTHROUGH */ 4968 default: 4969 optlen = opt[1]; 4970 if (optlen < 2) 4971 optlen = 2; 4972 hlen -= optlen; 4973 opt += optlen; 4974 break; 4975 } 4976 } 4977 return (wscale); 4978 } 4979 4980 u_int16_t 4981 pf_get_mss(struct pf_pdesc *pd) 4982 { 4983 struct tcphdr *th = &pd->hdr.tcp; 4984 int hlen; 4985 u_int8_t hdr[60]; 4986 u_int8_t *opt, optlen; 4987 u_int16_t mss = V_tcp_mssdflt; 4988 4989 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4990 if (hlen <= sizeof(struct tcphdr)) 4991 return (0); 4992 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4993 return (0); 4994 opt = hdr + sizeof(struct tcphdr); 4995 hlen -= sizeof(struct tcphdr); 4996 while (hlen >= TCPOLEN_MAXSEG) { 4997 switch (*opt) { 4998 case TCPOPT_EOL: 4999 case TCPOPT_NOP: 5000 ++opt; 5001 --hlen; 5002 break; 5003 case TCPOPT_MAXSEG: 5004 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 5005 NTOHS(mss); 5006 /* FALLTHROUGH */ 5007 default: 5008 optlen = opt[1]; 5009 if (optlen < 2) 5010 optlen = 2; 5011 hlen -= optlen; 5012 opt += optlen; 5013 break; 5014 } 5015 } 5016 return (mss); 5017 } 5018 5019 static u_int16_t 5020 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5021 { 5022 struct nhop_object *nh; 5023 #ifdef INET6 5024 struct in6_addr dst6; 5025 uint32_t scopeid; 5026 #endif /* INET6 */ 5027 int hlen = 0; 5028 uint16_t mss = 0; 5029 5030 NET_EPOCH_ASSERT(); 5031 5032 switch (af) { 5033 #ifdef INET 5034 case AF_INET: 5035 hlen = sizeof(struct ip); 5036 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5037 if (nh != NULL) 5038 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5039 break; 5040 #endif /* INET */ 5041 #ifdef INET6 5042 case AF_INET6: 5043 hlen = sizeof(struct ip6_hdr); 5044 in6_splitscope(&addr->v6, &dst6, &scopeid); 5045 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5046 if (nh != NULL) 5047 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5048 break; 5049 #endif /* INET6 */ 5050 } 5051 5052 mss = max(V_tcp_mssdflt, mss); 5053 mss = min(mss, offer); 5054 mss = max(mss, 64); /* sanity - at least max opt space */ 5055 return (mss); 5056 } 5057 5058 static u_int32_t 5059 pf_tcp_iss(struct pf_pdesc *pd) 5060 { 5061 MD5_CTX ctx; 5062 u_int32_t digest[4]; 5063 5064 if (V_pf_tcp_secret_init == 0) { 5065 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5066 MD5Init(&V_pf_tcp_secret_ctx); 5067 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5068 sizeof(V_pf_tcp_secret)); 5069 V_pf_tcp_secret_init = 1; 5070 } 5071 5072 ctx = V_pf_tcp_secret_ctx; 5073 5074 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 5075 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 5076 switch (pd->af) { 5077 case AF_INET6: 5078 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 5079 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 5080 break; 5081 case AF_INET: 5082 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 5083 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 5084 break; 5085 } 5086 MD5Final((u_char *)digest, &ctx); 5087 V_pf_tcp_iss_off += 4096; 5088 #define ISN_RANDOM_INCREMENT (4096 - 1) 5089 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5090 V_pf_tcp_iss_off); 5091 #undef ISN_RANDOM_INCREMENT 5092 } 5093 5094 static bool 5095 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5096 { 5097 bool match = true; 5098 5099 /* Always matches if not set */ 5100 if (! r->isset) 5101 return (!r->neg); 5102 5103 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5104 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5105 match = false; 5106 break; 5107 } 5108 } 5109 5110 return (match ^ r->neg); 5111 } 5112 5113 static int 5114 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5115 { 5116 if (*tag == -1) 5117 *tag = mtag; 5118 5119 return ((!r->match_tag_not && r->match_tag == *tag) || 5120 (r->match_tag_not && r->match_tag != *tag)); 5121 } 5122 5123 static void 5124 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5125 { 5126 /* If we don't have the interface drop the packet. */ 5127 if (ifp == NULL) { 5128 m_freem(m); 5129 return; 5130 } 5131 5132 switch (ifp->if_type) { 5133 case IFT_ETHER: 5134 case IFT_XETHER: 5135 case IFT_L2VLAN: 5136 case IFT_BRIDGE: 5137 case IFT_IEEE8023ADLAG: 5138 break; 5139 default: 5140 m_freem(m); 5141 return; 5142 } 5143 5144 ifp->if_transmit(ifp, m); 5145 } 5146 5147 static int 5148 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5149 { 5150 #ifdef INET 5151 struct ip ip; 5152 #endif 5153 #ifdef INET6 5154 struct ip6_hdr ip6; 5155 #endif 5156 struct mbuf *m = *m0; 5157 struct ether_header *e; 5158 struct pf_keth_rule *r, *rm, *a = NULL; 5159 struct pf_keth_ruleset *ruleset = NULL; 5160 struct pf_mtag *mtag; 5161 struct pf_keth_ruleq *rules; 5162 struct pf_addr *src = NULL, *dst = NULL; 5163 struct pfi_kkif *bridge_to; 5164 sa_family_t af = 0; 5165 uint16_t proto; 5166 int asd = 0, match = 0; 5167 int tag = -1; 5168 uint8_t action; 5169 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5170 5171 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5172 NET_EPOCH_ASSERT(); 5173 5174 PF_RULES_RLOCK_TRACKER; 5175 5176 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5177 5178 mtag = pf_find_mtag(m); 5179 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5180 /* Dummynet re-injects packets after they've 5181 * completed their delay. We've already 5182 * processed them, so pass unconditionally. */ 5183 5184 /* But only once. We may see the packet multiple times (e.g. 5185 * PFIL_IN/PFIL_OUT). */ 5186 pf_dummynet_flag_remove(m, mtag); 5187 5188 return (PF_PASS); 5189 } 5190 5191 ruleset = V_pf_keth; 5192 rules = ck_pr_load_ptr(&ruleset->active.rules); 5193 r = TAILQ_FIRST(rules); 5194 rm = NULL; 5195 5196 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5197 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5198 DPFPRINTF(PF_DEBUG_URGENT, 5199 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 5200 ", pullup failed\n")); 5201 return (PF_DROP); 5202 } 5203 e = mtod(m, struct ether_header *); 5204 proto = ntohs(e->ether_type); 5205 5206 switch (proto) { 5207 #ifdef INET 5208 case ETHERTYPE_IP: { 5209 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5210 sizeof(ip))) 5211 return (PF_DROP); 5212 5213 af = AF_INET; 5214 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5215 (caddr_t)&ip); 5216 src = (struct pf_addr *)&ip.ip_src; 5217 dst = (struct pf_addr *)&ip.ip_dst; 5218 break; 5219 } 5220 #endif /* INET */ 5221 #ifdef INET6 5222 case ETHERTYPE_IPV6: { 5223 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5224 sizeof(ip6))) 5225 return (PF_DROP); 5226 5227 af = AF_INET6; 5228 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5229 (caddr_t)&ip6); 5230 src = (struct pf_addr *)&ip6.ip6_src; 5231 dst = (struct pf_addr *)&ip6.ip6_dst; 5232 break; 5233 } 5234 #endif /* INET6 */ 5235 } 5236 5237 PF_RULES_RLOCK(); 5238 5239 while (r != NULL) { 5240 counter_u64_add(r->evaluations, 1); 5241 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5242 5243 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5244 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5245 "kif"); 5246 r = r->skip[PFE_SKIP_IFP].ptr; 5247 } 5248 else if (r->direction && r->direction != dir) { 5249 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5250 "dir"); 5251 r = r->skip[PFE_SKIP_DIR].ptr; 5252 } 5253 else if (r->proto && r->proto != proto) { 5254 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5255 "proto"); 5256 r = r->skip[PFE_SKIP_PROTO].ptr; 5257 } 5258 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5259 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5260 "src"); 5261 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5262 } 5263 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5264 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5265 "dst"); 5266 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5267 } 5268 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5269 r->ipsrc.neg, kif, M_GETFIB(m))) { 5270 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5271 "ip_src"); 5272 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5273 } 5274 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5275 r->ipdst.neg, kif, M_GETFIB(m))) { 5276 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5277 "ip_dst"); 5278 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5279 } 5280 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5281 mtag ? mtag->tag : 0)) { 5282 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5283 "match_tag"); 5284 r = TAILQ_NEXT(r, entries); 5285 } 5286 else { 5287 if (r->tag) 5288 tag = r->tag; 5289 if (r->anchor == NULL) { 5290 /* Rule matches */ 5291 rm = r; 5292 5293 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5294 5295 if (r->quick) 5296 break; 5297 5298 r = TAILQ_NEXT(r, entries); 5299 } else { 5300 pf_step_into_keth_anchor(anchor_stack, &asd, 5301 &ruleset, &r, &a, &match); 5302 } 5303 } 5304 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5305 &ruleset, &r, &a, &match)) 5306 break; 5307 } 5308 5309 r = rm; 5310 5311 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5312 5313 /* Default to pass. */ 5314 if (r == NULL) { 5315 PF_RULES_RUNLOCK(); 5316 return (PF_PASS); 5317 } 5318 5319 /* Execute action. */ 5320 counter_u64_add(r->packets[dir == PF_OUT], 1); 5321 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5322 pf_update_timestamp(r); 5323 5324 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5325 if (r->action == PF_DROP) { 5326 PF_RULES_RUNLOCK(); 5327 return (PF_DROP); 5328 } 5329 5330 if (tag > 0) { 5331 if (mtag == NULL) 5332 mtag = pf_get_mtag(m); 5333 if (mtag == NULL) { 5334 PF_RULES_RUNLOCK(); 5335 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5336 return (PF_DROP); 5337 } 5338 mtag->tag = tag; 5339 } 5340 5341 if (r->qid != 0) { 5342 if (mtag == NULL) 5343 mtag = pf_get_mtag(m); 5344 if (mtag == NULL) { 5345 PF_RULES_RUNLOCK(); 5346 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5347 return (PF_DROP); 5348 } 5349 mtag->qid = r->qid; 5350 } 5351 5352 action = r->action; 5353 bridge_to = r->bridge_to; 5354 5355 /* Dummynet */ 5356 if (r->dnpipe) { 5357 struct ip_fw_args dnflow; 5358 5359 /* Drop packet if dummynet is not loaded. */ 5360 if (ip_dn_io_ptr == NULL) { 5361 PF_RULES_RUNLOCK(); 5362 m_freem(m); 5363 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5364 return (PF_DROP); 5365 } 5366 if (mtag == NULL) 5367 mtag = pf_get_mtag(m); 5368 if (mtag == NULL) { 5369 PF_RULES_RUNLOCK(); 5370 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5371 return (PF_DROP); 5372 } 5373 5374 bzero(&dnflow, sizeof(dnflow)); 5375 5376 /* We don't have port numbers here, so we set 0. That means 5377 * that we'll be somewhat limited in distinguishing flows (i.e. 5378 * only based on IP addresses, not based on port numbers), but 5379 * it's better than nothing. */ 5380 dnflow.f_id.dst_port = 0; 5381 dnflow.f_id.src_port = 0; 5382 dnflow.f_id.proto = 0; 5383 5384 dnflow.rule.info = r->dnpipe; 5385 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5386 if (r->dnflags & PFRULE_DN_IS_PIPE) 5387 dnflow.rule.info |= IPFW_IS_PIPE; 5388 5389 dnflow.f_id.extra = dnflow.rule.info; 5390 5391 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5392 dnflow.flags |= IPFW_ARGS_ETHER; 5393 dnflow.ifp = kif->pfik_ifp; 5394 5395 switch (af) { 5396 case AF_INET: 5397 dnflow.f_id.addr_type = 4; 5398 dnflow.f_id.src_ip = src->v4.s_addr; 5399 dnflow.f_id.dst_ip = dst->v4.s_addr; 5400 break; 5401 case AF_INET6: 5402 dnflow.flags |= IPFW_ARGS_IP6; 5403 dnflow.f_id.addr_type = 6; 5404 dnflow.f_id.src_ip6 = src->v6; 5405 dnflow.f_id.dst_ip6 = dst->v6; 5406 break; 5407 } 5408 5409 PF_RULES_RUNLOCK(); 5410 5411 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5412 ip_dn_io_ptr(m0, &dnflow); 5413 if (*m0 != NULL) 5414 pf_dummynet_flag_remove(m, mtag); 5415 } else { 5416 PF_RULES_RUNLOCK(); 5417 } 5418 5419 if (action == PF_PASS && bridge_to) { 5420 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5421 *m0 = NULL; /* We've eaten the packet. */ 5422 } 5423 5424 return (action); 5425 } 5426 5427 #define PF_TEST_ATTRIB(t, a)\ 5428 do { \ 5429 if (t) { \ 5430 r = a; \ 5431 goto nextrule; \ 5432 } \ 5433 } while (0) 5434 5435 static int 5436 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5437 struct pf_pdesc *pd, struct pf_krule **am, 5438 struct pf_kruleset **rsm, struct inpcb *inp) 5439 { 5440 struct pf_krule *nr = NULL; 5441 struct pf_krule *r, *a = NULL; 5442 struct pf_kruleset *ruleset = NULL; 5443 struct pf_krule_slist match_rules; 5444 struct pf_krule_item *ri; 5445 struct tcphdr *th = &pd->hdr.tcp; 5446 struct pf_state_key *sk = NULL, *nk = NULL; 5447 u_short reason, transerror; 5448 int rewrite = 0; 5449 int tag = -1; 5450 int asd = 0; 5451 int match = 0; 5452 int state_icmp = 0, icmp_dir, multi; 5453 u_int16_t virtual_type, virtual_id; 5454 u_int16_t bproto_sum = 0, bip_sum = 0; 5455 u_int8_t icmptype = 0, icmpcode = 0; 5456 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5457 struct pf_udp_mapping *udp_mapping = NULL; 5458 5459 PF_RULES_RASSERT(); 5460 5461 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5462 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5463 5464 SLIST_INIT(&match_rules); 5465 5466 if (inp != NULL) { 5467 INP_LOCK_ASSERT(inp); 5468 pd->lookup.uid = inp->inp_cred->cr_uid; 5469 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 5470 pd->lookup.done = 1; 5471 } 5472 5473 switch (pd->virtual_proto) { 5474 case IPPROTO_TCP: 5475 pd->nsport = th->th_sport; 5476 pd->ndport = th->th_dport; 5477 break; 5478 case IPPROTO_UDP: 5479 pd->nsport = pd->hdr.udp.uh_sport; 5480 pd->ndport = pd->hdr.udp.uh_dport; 5481 break; 5482 case IPPROTO_SCTP: 5483 pd->nsport = pd->hdr.sctp.src_port; 5484 pd->ndport = pd->hdr.sctp.dest_port; 5485 break; 5486 #ifdef INET 5487 case IPPROTO_ICMP: 5488 MPASS(pd->af == AF_INET); 5489 icmptype = pd->hdr.icmp.icmp_type; 5490 icmpcode = pd->hdr.icmp.icmp_code; 5491 state_icmp = pf_icmp_mapping(pd, icmptype, 5492 &icmp_dir, &multi, &virtual_id, &virtual_type); 5493 if (icmp_dir == PF_IN) { 5494 pd->nsport = virtual_id; 5495 pd->ndport = virtual_type; 5496 } else { 5497 pd->nsport = virtual_type; 5498 pd->ndport = virtual_id; 5499 } 5500 break; 5501 #endif /* INET */ 5502 #ifdef INET6 5503 case IPPROTO_ICMPV6: 5504 MPASS(pd->af == AF_INET6); 5505 icmptype = pd->hdr.icmp6.icmp6_type; 5506 icmpcode = pd->hdr.icmp6.icmp6_code; 5507 state_icmp = pf_icmp_mapping(pd, icmptype, 5508 &icmp_dir, &multi, &virtual_id, &virtual_type); 5509 if (icmp_dir == PF_IN) { 5510 pd->nsport = virtual_id; 5511 pd->ndport = virtual_type; 5512 } else { 5513 pd->nsport = virtual_type; 5514 pd->ndport = virtual_id; 5515 } 5516 5517 break; 5518 #endif /* INET6 */ 5519 default: 5520 pd->nsport = pd->ndport = 0; 5521 break; 5522 } 5523 pd->osport = pd->nsport; 5524 pd->odport = pd->ndport; 5525 5526 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5527 5528 /* check packet for BINAT/NAT/RDR */ 5529 transerror = pf_get_translation(pd, pd->off, &sk, &nk, anchor_stack, 5530 &nr, &udp_mapping); 5531 switch (transerror) { 5532 default: 5533 /* A translation error occurred. */ 5534 REASON_SET(&reason, transerror); 5535 goto cleanup; 5536 case PFRES_MAX: 5537 /* No match. */ 5538 break; 5539 case PFRES_MATCH: 5540 KASSERT(sk != NULL, ("%s: null sk", __func__)); 5541 KASSERT(nk != NULL, ("%s: null nk", __func__)); 5542 5543 if (nr->log) { 5544 PFLOG_PACKET(PF_PASS, PFRES_MATCH, nr, a, 5545 ruleset, pd, 1); 5546 } 5547 5548 if (pd->ip_sum) 5549 bip_sum = *pd->ip_sum; 5550 5551 switch (pd->proto) { 5552 case IPPROTO_TCP: 5553 bproto_sum = th->th_sum; 5554 5555 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5556 nk->port[pd->sidx] != pd->nsport) { 5557 pf_change_ap(pd->m, pd->src, &th->th_sport, 5558 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5559 nk->port[pd->sidx], 0, pd->af, pd->naf); 5560 pd->sport = &th->th_sport; 5561 pd->nsport = th->th_sport; 5562 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5563 } 5564 5565 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5566 nk->port[pd->didx] != pd->ndport) { 5567 pf_change_ap(pd->m, pd->dst, &th->th_dport, 5568 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5569 nk->port[pd->didx], 0, pd->af, pd->naf); 5570 pd->dport = &th->th_dport; 5571 pd->ndport = th->th_dport; 5572 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5573 } 5574 rewrite++; 5575 break; 5576 case IPPROTO_UDP: 5577 bproto_sum = pd->hdr.udp.uh_sum; 5578 5579 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5580 nk->port[pd->sidx] != pd->nsport) { 5581 pf_change_ap(pd->m, pd->src, 5582 &pd->hdr.udp.uh_sport, 5583 pd->ip_sum, &pd->hdr.udp.uh_sum, 5584 &nk->addr[pd->sidx], 5585 nk->port[pd->sidx], 1, pd->af, pd->naf); 5586 pd->sport = &pd->hdr.udp.uh_sport; 5587 pd->nsport = pd->hdr.udp.uh_sport; 5588 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5589 } 5590 5591 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5592 nk->port[pd->didx] != pd->ndport) { 5593 pf_change_ap(pd->m, pd->dst, 5594 &pd->hdr.udp.uh_dport, 5595 pd->ip_sum, &pd->hdr.udp.uh_sum, 5596 &nk->addr[pd->didx], 5597 nk->port[pd->didx], 1, pd->af, pd->naf); 5598 pd->dport = &pd->hdr.udp.uh_dport; 5599 pd->ndport = pd->hdr.udp.uh_dport; 5600 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5601 } 5602 rewrite++; 5603 break; 5604 case IPPROTO_SCTP: { 5605 uint16_t checksum = 0; 5606 5607 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5608 nk->port[pd->sidx] != pd->nsport) { 5609 pf_change_ap(pd->m, pd->src, 5610 &pd->hdr.sctp.src_port, pd->ip_sum, &checksum, 5611 &nk->addr[pd->sidx], 5612 nk->port[pd->sidx], 1, pd->af, pd->naf); 5613 pd->sport = &pd->hdr.sctp.src_port; 5614 pd->nsport = pd->hdr.sctp.src_port; 5615 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5616 } 5617 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5618 nk->port[pd->didx] != pd->ndport) { 5619 pf_change_ap(pd->m, pd->dst, 5620 &pd->hdr.sctp.dest_port, pd->ip_sum, &checksum, 5621 &nk->addr[pd->didx], 5622 nk->port[pd->didx], 1, pd->af, pd->naf); 5623 pd->dport = &pd->hdr.sctp.dest_port; 5624 pd->ndport = pd->hdr.sctp.dest_port; 5625 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5626 } 5627 break; 5628 } 5629 #ifdef INET 5630 case IPPROTO_ICMP: 5631 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 5632 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 5633 nk->addr[pd->sidx].v4.s_addr, 0); 5634 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5635 } 5636 5637 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 5638 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 5639 nk->addr[pd->didx].v4.s_addr, 0); 5640 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5641 } 5642 5643 if (virtual_type == htons(ICMP_ECHO) && 5644 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5645 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5646 pd->hdr.icmp.icmp_cksum, pd->nsport, 5647 nk->port[pd->sidx], 0); 5648 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5649 pd->sport = &pd->hdr.icmp.icmp_id; 5650 } 5651 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5652 break; 5653 #endif /* INET */ 5654 #ifdef INET6 5655 case IPPROTO_ICMPV6: 5656 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 5657 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 5658 &nk->addr[pd->sidx], 0); 5659 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5660 } 5661 5662 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 5663 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 5664 &nk->addr[pd->didx], 0); 5665 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5666 } 5667 rewrite++; 5668 break; 5669 #endif /* INET */ 5670 default: 5671 switch (pd->af) { 5672 #ifdef INET 5673 case AF_INET: 5674 if (PF_ANEQ(&pd->nsaddr, 5675 &nk->addr[pd->sidx], AF_INET)) { 5676 pf_change_a(&pd->src->v4.s_addr, 5677 pd->ip_sum, 5678 nk->addr[pd->sidx].v4.s_addr, 0); 5679 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5680 } 5681 5682 if (PF_ANEQ(&pd->ndaddr, 5683 &nk->addr[pd->didx], AF_INET)) { 5684 pf_change_a(&pd->dst->v4.s_addr, 5685 pd->ip_sum, 5686 nk->addr[pd->didx].v4.s_addr, 0); 5687 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5688 } 5689 break; 5690 #endif /* INET */ 5691 #ifdef INET6 5692 case AF_INET6: 5693 if (PF_ANEQ(&pd->nsaddr, 5694 &nk->addr[pd->sidx], AF_INET6)) { 5695 PF_ACPY(&pd->nsaddr, &nk->addr[pd->sidx], pd->af); 5696 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5697 } 5698 5699 if (PF_ANEQ(&pd->ndaddr, 5700 &nk->addr[pd->didx], AF_INET6)) { 5701 PF_ACPY(&pd->ndaddr, &nk->addr[pd->didx], pd->af); 5702 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5703 } 5704 break; 5705 #endif /* INET */ 5706 } 5707 break; 5708 } 5709 if (nr->natpass) 5710 r = NULL; 5711 } 5712 5713 while (r != NULL) { 5714 if (pd->related_rule) { 5715 *rm = pd->related_rule; 5716 break; 5717 } 5718 pf_counter_u64_add(&r->evaluations, 1); 5719 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5720 r->skip[PF_SKIP_IFP]); 5721 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5722 r->skip[PF_SKIP_DIR]); 5723 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5724 r->skip[PF_SKIP_AF]); 5725 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5726 r->skip[PF_SKIP_PROTO]); 5727 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5728 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5729 r->skip[PF_SKIP_SRC_ADDR]); 5730 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5731 r->dst.neg, NULL, M_GETFIB(pd->m)), 5732 r->skip[PF_SKIP_DST_ADDR]); 5733 switch (pd->virtual_proto) { 5734 case PF_VPROTO_FRAGMENT: 5735 /* tcp/udp only. port_op always 0 in other cases */ 5736 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5737 TAILQ_NEXT(r, entries)); 5738 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5739 TAILQ_NEXT(r, entries)); 5740 /* icmp only. type/code always 0 in other cases */ 5741 PF_TEST_ATTRIB((r->type || r->code), 5742 TAILQ_NEXT(r, entries)); 5743 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5744 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5745 TAILQ_NEXT(r, entries)); 5746 break; 5747 5748 case IPPROTO_TCP: 5749 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(th)) != r->flags, 5750 TAILQ_NEXT(r, entries)); 5751 /* FALLTHROUGH */ 5752 case IPPROTO_SCTP: 5753 case IPPROTO_UDP: 5754 /* tcp/udp only. port_op always 0 in other cases */ 5755 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5756 r->src.port[0], r->src.port[1], pd->nsport), 5757 r->skip[PF_SKIP_SRC_PORT]); 5758 /* tcp/udp only. port_op always 0 in other cases */ 5759 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5760 r->dst.port[0], r->dst.port[1], pd->ndport), 5761 r->skip[PF_SKIP_DST_PORT]); 5762 /* tcp/udp only. uid.op always 0 in other cases */ 5763 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5764 pf_socket_lookup(pd), 1)) && 5765 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5766 pd->lookup.uid), 5767 TAILQ_NEXT(r, entries)); 5768 /* tcp/udp only. gid.op always 0 in other cases */ 5769 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5770 pf_socket_lookup(pd), 1)) && 5771 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5772 pd->lookup.gid), 5773 TAILQ_NEXT(r, entries)); 5774 break; 5775 5776 case IPPROTO_ICMP: 5777 case IPPROTO_ICMPV6: 5778 /* icmp only. type always 0 in other cases */ 5779 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5780 TAILQ_NEXT(r, entries)); 5781 /* icmp only. type always 0 in other cases */ 5782 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5783 TAILQ_NEXT(r, entries)); 5784 break; 5785 5786 default: 5787 break; 5788 } 5789 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5790 TAILQ_NEXT(r, entries)); 5791 PF_TEST_ATTRIB(r->prio && 5792 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5793 TAILQ_NEXT(r, entries)); 5794 PF_TEST_ATTRIB(r->prob && 5795 r->prob <= arc4random(), 5796 TAILQ_NEXT(r, entries)); 5797 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5798 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5799 TAILQ_NEXT(r, entries)); 5800 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(pd->m, r), 5801 TAILQ_NEXT(r, entries)); 5802 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5803 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5804 TAILQ_NEXT(r, entries)); 5805 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5806 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5807 pf_osfp_fingerprint(pd, th), 5808 r->os_fingerprint)), 5809 TAILQ_NEXT(r, entries)); 5810 /* FALLTHROUGH */ 5811 if (r->tag) 5812 tag = r->tag; 5813 if (r->anchor == NULL) { 5814 if (r->action == PF_MATCH) { 5815 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5816 if (ri == NULL) { 5817 REASON_SET(&reason, PFRES_MEMORY); 5818 goto cleanup; 5819 } 5820 ri->r = r; 5821 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5822 pf_counter_u64_critical_enter(); 5823 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5824 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5825 pf_counter_u64_critical_exit(); 5826 pf_rule_to_actions(r, &pd->act); 5827 if (r->rule_flag & PFRULE_AFTO) 5828 pd->naf = r->naf; 5829 if (pd->af != pd->naf) { 5830 if (pf_get_transaddr_af(r, pd) == -1) { 5831 REASON_SET(&reason, PFRES_MEMORY); 5832 goto cleanup; 5833 } 5834 } 5835 if (r->log || pd->act.log & PF_LOG_MATCHES) 5836 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5837 a, ruleset, pd, 1); 5838 } else { 5839 match = 1; 5840 *rm = r; 5841 *am = a; 5842 *rsm = ruleset; 5843 if (pd->act.log & PF_LOG_MATCHES) 5844 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5845 a, ruleset, pd, 1); 5846 } 5847 if ((*rm)->quick) 5848 break; 5849 r = TAILQ_NEXT(r, entries); 5850 } else 5851 pf_step_into_anchor(anchor_stack, &asd, 5852 &ruleset, PF_RULESET_FILTER, &r, &a, 5853 &match); 5854 nextrule: 5855 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5856 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5857 break; 5858 } 5859 r = *rm; 5860 a = *am; 5861 ruleset = *rsm; 5862 5863 REASON_SET(&reason, PFRES_MATCH); 5864 5865 /* apply actions for last matching pass/block rule */ 5866 pf_rule_to_actions(r, &pd->act); 5867 if (r->rule_flag & PFRULE_AFTO) 5868 pd->naf = r->naf; 5869 if (pd->af != pd->naf) { 5870 if (pf_get_transaddr_af(r, pd) == -1) { 5871 REASON_SET(&reason, PFRES_MEMORY); 5872 goto cleanup; 5873 } 5874 } 5875 5876 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5877 if (rewrite) 5878 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5879 PFLOG_PACKET(r->action, reason, r, a, ruleset, pd, 1); 5880 } 5881 5882 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5883 (r->action == PF_DROP) && 5884 ((r->rule_flag & PFRULE_RETURNRST) || 5885 (r->rule_flag & PFRULE_RETURNICMP) || 5886 (r->rule_flag & PFRULE_RETURN))) { 5887 pf_return(r, nr, pd, sk, th, bproto_sum, 5888 bip_sum, &reason, r->rtableid); 5889 } 5890 5891 if (r->action == PF_DROP) 5892 goto cleanup; 5893 5894 if (tag > 0 && pf_tag_packet(pd, tag)) { 5895 REASON_SET(&reason, PFRES_MEMORY); 5896 goto cleanup; 5897 } 5898 if (pd->act.rtableid >= 0) 5899 M_SETFIB(pd->m, pd->act.rtableid); 5900 5901 if (r->rt) { 5902 struct pf_ksrc_node *sn = NULL; 5903 struct pf_srchash *snh = NULL; 5904 struct pf_kpool *pool = &r->route; 5905 5906 /* Backwards compatibility. */ 5907 if (TAILQ_EMPTY(&pool->list)) 5908 pool = &r->rdr; 5909 5910 /* 5911 * Set act.rt here instead of in pf_rule_to_actions() because 5912 * it is applied only from the last pass rule. 5913 */ 5914 pd->act.rt = r->rt; 5915 /* Don't use REASON_SET, pf_map_addr increases the reason counters */ 5916 reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, 5917 &pd->act.rt_kif, NULL, &sn, &snh, pool); 5918 if (reason != 0) 5919 goto cleanup; 5920 } 5921 5922 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5923 (!state_icmp && (r->keep_state || nr != NULL || 5924 (pd->flags & PFDESC_TCP_NORM)))) { 5925 int action; 5926 bool nat64; 5927 5928 action = pf_create_state(r, nr, a, pd, nk, sk, 5929 &rewrite, sm, tag, bproto_sum, bip_sum, 5930 &match_rules, udp_mapping); 5931 if (action != PF_PASS) { 5932 pf_udp_mapping_release(udp_mapping); 5933 pd->act.log |= PF_LOG_FORCE; 5934 if (action == PF_DROP && 5935 (r->rule_flag & PFRULE_RETURN)) 5936 pf_return(r, nr, pd, sk, th, 5937 bproto_sum, bip_sum, &reason, 5938 pd->act.rtableid); 5939 return (action); 5940 } 5941 5942 nat64 = pd->af != pd->naf; 5943 if (nat64) { 5944 struct pf_state_key *_sk; 5945 int ret; 5946 5947 if (sk == NULL) 5948 sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 5949 if (nk == NULL) 5950 nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 5951 if (pd->dir == PF_IN) 5952 _sk = sk; 5953 else 5954 _sk = nk; 5955 5956 ret = pf_translate(pd, 5957 &_sk->addr[pd->didx], 5958 _sk->port[pd->didx], 5959 &_sk->addr[pd->sidx], 5960 _sk->port[pd->sidx], 5961 virtual_type, icmp_dir); 5962 if (ret < 0) 5963 goto cleanup; 5964 5965 rewrite += ret; 5966 } 5967 } else { 5968 while ((ri = SLIST_FIRST(&match_rules))) { 5969 SLIST_REMOVE_HEAD(&match_rules, entry); 5970 free(ri, M_PF_RULE_ITEM); 5971 } 5972 5973 uma_zfree(V_pf_state_key_z, sk); 5974 uma_zfree(V_pf_state_key_z, nk); 5975 pf_udp_mapping_release(udp_mapping); 5976 } 5977 5978 /* copy back packet headers if we performed NAT operations */ 5979 if (rewrite) 5980 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5981 5982 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5983 pd->dir == PF_OUT && 5984 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 5985 /* 5986 * We want the state created, but we dont 5987 * want to send this in case a partner 5988 * firewall has to know about it to allow 5989 * replies through it. 5990 */ 5991 return (PF_DEFER); 5992 5993 if (rewrite && sk != NULL && nk != NULL && sk->af != nk->af) { 5994 return (PF_AFRT); 5995 } else 5996 return (PF_PASS); 5997 5998 cleanup: 5999 while ((ri = SLIST_FIRST(&match_rules))) { 6000 SLIST_REMOVE_HEAD(&match_rules, entry); 6001 free(ri, M_PF_RULE_ITEM); 6002 } 6003 6004 uma_zfree(V_pf_state_key_z, sk); 6005 uma_zfree(V_pf_state_key_z, nk); 6006 pf_udp_mapping_release(udp_mapping); 6007 6008 return (PF_DROP); 6009 } 6010 6011 static int 6012 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 6013 struct pf_pdesc *pd, struct pf_state_key *nk, struct pf_state_key *sk, 6014 int *rewrite, struct pf_kstate **sm, int tag, u_int16_t bproto_sum, 6015 u_int16_t bip_sum, struct pf_krule_slist *match_rules, 6016 struct pf_udp_mapping *udp_mapping) 6017 { 6018 struct pf_kstate *s = NULL; 6019 struct pf_ksrc_node *sn = NULL; 6020 struct pf_srchash *snh = NULL; 6021 struct pf_ksrc_node *nsn = NULL; 6022 struct pf_srchash *nsnh = NULL; 6023 struct tcphdr *th = &pd->hdr.tcp; 6024 u_int16_t mss = V_tcp_mssdflt; 6025 u_short reason, sn_reason; 6026 struct pf_krule_item *ri; 6027 6028 /* check maximums */ 6029 if (r->max_states && 6030 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6031 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6032 REASON_SET(&reason, PFRES_MAXSTATES); 6033 goto csfailed; 6034 } 6035 /* src node for filter rule */ 6036 if ((r->rule_flag & PFRULE_SRCTRACK || 6037 r->rdr.opts & PF_POOL_STICKYADDR) && 6038 (sn_reason = pf_insert_src_node(&sn, &snh, r, pd->src, pd->af, 6039 &pd->act.rt_addr, pd->act.rt_kif)) != 0) { 6040 REASON_SET(&reason, sn_reason); 6041 goto csfailed; 6042 } 6043 /* src node for translation rule */ 6044 if (nr != NULL && (nr->rdr.opts & PF_POOL_STICKYADDR) && 6045 (sn_reason = pf_insert_src_node(&nsn, &nsnh, nr, &sk->addr[pd->sidx], 6046 pd->af, &nk->addr[1], NULL)) != 0 ) { 6047 REASON_SET(&reason, sn_reason); 6048 goto csfailed; 6049 } 6050 s = pf_alloc_state(M_NOWAIT); 6051 if (s == NULL) { 6052 REASON_SET(&reason, PFRES_MEMORY); 6053 goto csfailed; 6054 } 6055 s->rule = r; 6056 s->nat_rule = nr; 6057 s->anchor = a; 6058 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 6059 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6060 6061 STATE_INC_COUNTERS(s); 6062 if (r->allow_opts) 6063 s->state_flags |= PFSTATE_ALLOWOPTS; 6064 if (r->rule_flag & PFRULE_STATESLOPPY) 6065 s->state_flags |= PFSTATE_SLOPPY; 6066 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6067 s->state_flags |= PFSTATE_SCRUB_TCP; 6068 if ((r->rule_flag & PFRULE_PFLOW) || 6069 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 6070 s->state_flags |= PFSTATE_PFLOW; 6071 6072 s->act.log = pd->act.log & PF_LOG_ALL; 6073 s->sync_state = PFSYNC_S_NONE; 6074 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6075 6076 if (nr != NULL) 6077 s->act.log |= nr->log & PF_LOG_ALL; 6078 switch (pd->proto) { 6079 case IPPROTO_TCP: 6080 s->src.seqlo = ntohl(th->th_seq); 6081 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6082 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6083 r->keep_state == PF_STATE_MODULATE) { 6084 /* Generate sequence number modulator */ 6085 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6086 0) 6087 s->src.seqdiff = 1; 6088 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6089 htonl(s->src.seqlo + s->src.seqdiff), 0); 6090 *rewrite = 1; 6091 } else 6092 s->src.seqdiff = 0; 6093 if (tcp_get_flags(th) & TH_SYN) { 6094 s->src.seqhi++; 6095 s->src.wscale = pf_get_wscale(pd); 6096 } 6097 s->src.max_win = MAX(ntohs(th->th_win), 1); 6098 if (s->src.wscale & PF_WSCALE_MASK) { 6099 /* Remove scale factor from initial window */ 6100 int win = s->src.max_win; 6101 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6102 s->src.max_win = (win - 1) >> 6103 (s->src.wscale & PF_WSCALE_MASK); 6104 } 6105 if (tcp_get_flags(th) & TH_FIN) 6106 s->src.seqhi++; 6107 s->dst.seqhi = 1; 6108 s->dst.max_win = 1; 6109 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6110 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6111 s->timeout = PFTM_TCP_FIRST_PACKET; 6112 atomic_add_32(&V_pf_status.states_halfopen, 1); 6113 break; 6114 case IPPROTO_UDP: 6115 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6116 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6117 s->timeout = PFTM_UDP_FIRST_PACKET; 6118 break; 6119 case IPPROTO_SCTP: 6120 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6121 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6122 s->timeout = PFTM_SCTP_FIRST_PACKET; 6123 break; 6124 case IPPROTO_ICMP: 6125 #ifdef INET6 6126 case IPPROTO_ICMPV6: 6127 #endif 6128 s->timeout = PFTM_ICMP_FIRST_PACKET; 6129 break; 6130 default: 6131 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6132 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6133 s->timeout = PFTM_OTHER_FIRST_PACKET; 6134 } 6135 6136 s->creation = s->expire = pf_get_uptime(); 6137 6138 if (pd->proto == IPPROTO_TCP) { 6139 if (s->state_flags & PFSTATE_SCRUB_TCP && 6140 pf_normalize_tcp_init(pd, th, &s->src, &s->dst)) { 6141 REASON_SET(&reason, PFRES_MEMORY); 6142 goto csfailed; 6143 } 6144 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6145 pf_normalize_tcp_stateful(pd, &reason, th, s, 6146 &s->src, &s->dst, rewrite)) { 6147 /* This really shouldn't happen!!! */ 6148 DPFPRINTF(PF_DEBUG_URGENT, 6149 ("pf_normalize_tcp_stateful failed on first " 6150 "pkt\n")); 6151 goto csfailed; 6152 } 6153 } else if (pd->proto == IPPROTO_SCTP) { 6154 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6155 goto csfailed; 6156 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6157 goto csfailed; 6158 } 6159 s->direction = pd->dir; 6160 6161 /* 6162 * sk/nk could already been setup by pf_get_translation(). 6163 */ 6164 if (nr == NULL) { 6165 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 6166 __func__, nr, sk, nk)); 6167 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6168 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6169 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, &sk, &nk)) { 6170 goto csfailed; 6171 } 6172 } else 6173 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 6174 __func__, nr, sk, nk)); 6175 6176 /* Swap sk/nk for PF_OUT. */ 6177 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6178 (pd->dir == PF_IN) ? sk : nk, 6179 (pd->dir == PF_IN) ? nk : sk, s)) { 6180 REASON_SET(&reason, PFRES_STATEINS); 6181 goto drop; 6182 } else 6183 *sm = s; 6184 6185 /* 6186 * Lock order is important: first state, then source node. 6187 */ 6188 if (pf_src_node_exists(&sn, snh)) { 6189 s->src_node = sn; 6190 PF_HASHROW_UNLOCK(snh); 6191 } 6192 if (pf_src_node_exists(&nsn, nsnh)) { 6193 s->nat_src_node = nsn; 6194 PF_HASHROW_UNLOCK(nsnh); 6195 } 6196 6197 if (tag > 0) 6198 s->tag = tag; 6199 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6200 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 6201 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6202 /* undo NAT changes, if they have taken place */ 6203 if (nr != NULL) { 6204 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 6205 if (pd->dir == PF_OUT) 6206 skt = s->key[PF_SK_STACK]; 6207 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 6208 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 6209 if (pd->sport) 6210 *pd->sport = skt->port[pd->sidx]; 6211 if (pd->dport) 6212 *pd->dport = skt->port[pd->didx]; 6213 if (pd->ip_sum) 6214 *pd->ip_sum = bip_sum; 6215 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6216 } 6217 s->src.seqhi = htonl(arc4random()); 6218 /* Find mss option */ 6219 int rtid = M_GETFIB(pd->m); 6220 mss = pf_get_mss(pd); 6221 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6222 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6223 s->src.mss = mss; 6224 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6225 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6226 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6227 pd->act.rtableid); 6228 REASON_SET(&reason, PFRES_SYNPROXY); 6229 return (PF_SYNPROXY_DROP); 6230 } 6231 6232 s->udp_mapping = udp_mapping; 6233 6234 return (PF_PASS); 6235 6236 csfailed: 6237 while ((ri = SLIST_FIRST(match_rules))) { 6238 SLIST_REMOVE_HEAD(match_rules, entry); 6239 free(ri, M_PF_RULE_ITEM); 6240 } 6241 6242 uma_zfree(V_pf_state_key_z, sk); 6243 uma_zfree(V_pf_state_key_z, nk); 6244 6245 if (pf_src_node_exists(&sn, snh)) { 6246 if (--sn->states == 0 && sn->expire == 0) { 6247 pf_unlink_src_node(sn); 6248 pf_free_src_node(sn); 6249 counter_u64_add( 6250 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6251 } 6252 PF_HASHROW_UNLOCK(snh); 6253 } 6254 6255 if (sn != nsn && pf_src_node_exists(&nsn, nsnh)) { 6256 if (--nsn->states == 0 && nsn->expire == 0) { 6257 pf_unlink_src_node(nsn); 6258 pf_free_src_node(nsn); 6259 counter_u64_add( 6260 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6261 } 6262 PF_HASHROW_UNLOCK(nsnh); 6263 } 6264 6265 drop: 6266 if (s != NULL) { 6267 pf_src_tree_remove_state(s); 6268 s->timeout = PFTM_UNLINKED; 6269 STATE_DEC_COUNTERS(s); 6270 pf_free_state(s); 6271 } 6272 6273 return (PF_DROP); 6274 } 6275 6276 int 6277 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6278 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6279 int icmp_dir) 6280 { 6281 /* 6282 * pf_translate() implements OpenBSD's "new" NAT approach. 6283 * We don't follow it, because it involves a breaking syntax change 6284 * (removing nat/rdr rules, moving it into regular pf rules.) 6285 * It also moves NAT processing to be done after normal rules evaluation 6286 * whereas in FreeBSD that's done before rules processing. 6287 * 6288 * We adopt the function only for nat64, and keep other NAT processing 6289 * before rules processing. 6290 */ 6291 int rewrite = 0; 6292 int afto = pd->af != pd->naf; 6293 6294 MPASS(afto); 6295 6296 switch (pd->proto) { 6297 case IPPROTO_TCP: 6298 if (afto || *pd->sport != sport) { 6299 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6300 saddr, sport, 0, pd->af, pd->naf); 6301 rewrite = 1; 6302 } 6303 if (afto || *pd->dport != dport) { 6304 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6305 daddr, dport, 0, pd->af, pd->naf); 6306 rewrite = 1; 6307 } 6308 break; 6309 6310 case IPPROTO_UDP: 6311 if (afto || *pd->sport != sport) { 6312 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6313 saddr, sport, 1, pd->af, pd->naf); 6314 rewrite = 1; 6315 } 6316 if (afto || *pd->dport != dport) { 6317 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6318 daddr, dport, 1, pd->af, pd->naf); 6319 rewrite = 1; 6320 } 6321 break; 6322 6323 case IPPROTO_SCTP: { 6324 uint16_t checksum = 0; 6325 if (afto || *pd->sport != sport) { 6326 pf_change_ap(pd->m, pd->src, pd->sport, pd->ip_sum, &checksum, 6327 saddr, sport, 1, pd->af, pd->naf); 6328 rewrite = 1; 6329 } 6330 if (afto || *pd->dport != dport) { 6331 pf_change_ap(pd->m, pd->dst, pd->dport, pd->ip_sum, &checksum, 6332 daddr, dport, 1, pd->af, pd->naf); 6333 rewrite = 1; 6334 } 6335 break; 6336 } 6337 6338 #ifdef INET 6339 case IPPROTO_ICMP: 6340 /* pf_translate() is also used when logging invalid packets */ 6341 if (pd->af != AF_INET) 6342 return (0); 6343 6344 if (afto) { 6345 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6346 return (-1); 6347 pd->proto = IPPROTO_ICMPV6; 6348 rewrite = 1; 6349 } 6350 if (virtual_type == htons(ICMP_ECHO)) { 6351 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6352 6353 if (icmpid != pd->hdr.icmp.icmp_id) { 6354 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6355 pd->hdr.icmp.icmp_cksum, 6356 pd->hdr.icmp.icmp_id, icmpid, 0); 6357 pd->hdr.icmp.icmp_id = icmpid; 6358 /* XXX TODO copyback. */ 6359 rewrite = 1; 6360 } 6361 } 6362 break; 6363 #endif /* INET */ 6364 6365 #ifdef INET6 6366 case IPPROTO_ICMPV6: 6367 /* pf_translate() is also used when logging invalid packets */ 6368 if (pd->af != AF_INET6) 6369 return (0); 6370 6371 if (afto) { 6372 /* ip_sum will be recalculated in pf_translate_af */ 6373 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6374 return (0); 6375 pd->proto = IPPROTO_ICMP; 6376 rewrite = 1; 6377 } 6378 break; 6379 #endif /* INET6 */ 6380 6381 default: 6382 break; 6383 } 6384 6385 return (rewrite); 6386 } 6387 6388 static int 6389 pf_tcp_track_full(struct pf_kstate **state, struct pf_pdesc *pd, 6390 u_short *reason, int *copyback) 6391 { 6392 struct tcphdr *th = &pd->hdr.tcp; 6393 struct pf_state_peer *src, *dst; 6394 u_int16_t win = ntohs(th->th_win); 6395 u_int32_t ack, end, data_end, seq, orig_seq; 6396 u_int8_t sws, dws, psrc, pdst; 6397 int ackskew; 6398 6399 if (pd->dir == (*state)->direction) { 6400 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 6401 src = &(*state)->dst; 6402 dst = &(*state)->src; 6403 } else { 6404 src = &(*state)->src; 6405 dst = &(*state)->dst; 6406 } 6407 psrc = PF_PEER_SRC; 6408 pdst = PF_PEER_DST; 6409 } else { 6410 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 6411 src = &(*state)->src; 6412 dst = &(*state)->dst; 6413 } else { 6414 src = &(*state)->dst; 6415 dst = &(*state)->src; 6416 } 6417 psrc = PF_PEER_DST; 6418 pdst = PF_PEER_SRC; 6419 } 6420 6421 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6422 sws = src->wscale & PF_WSCALE_MASK; 6423 dws = dst->wscale & PF_WSCALE_MASK; 6424 } else 6425 sws = dws = 0; 6426 6427 /* 6428 * Sequence tracking algorithm from Guido van Rooij's paper: 6429 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6430 * tcp_filtering.ps 6431 */ 6432 6433 orig_seq = seq = ntohl(th->th_seq); 6434 if (src->seqlo == 0) { 6435 /* First packet from this end. Set its state */ 6436 6437 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6438 src->scrub == NULL) { 6439 if (pf_normalize_tcp_init(pd, th, src, dst)) { 6440 REASON_SET(reason, PFRES_MEMORY); 6441 return (PF_DROP); 6442 } 6443 } 6444 6445 /* Deferred generation of sequence number modulator */ 6446 if (dst->seqdiff && !src->seqdiff) { 6447 /* use random iss for the TCP server */ 6448 while ((src->seqdiff = arc4random() - seq) == 0) 6449 ; 6450 ack = ntohl(th->th_ack) - dst->seqdiff; 6451 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6452 src->seqdiff), 0); 6453 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6454 *copyback = 1; 6455 } else { 6456 ack = ntohl(th->th_ack); 6457 } 6458 6459 end = seq + pd->p_len; 6460 if (tcp_get_flags(th) & TH_SYN) { 6461 end++; 6462 if (dst->wscale & PF_WSCALE_FLAG) { 6463 src->wscale = pf_get_wscale(pd); 6464 if (src->wscale & PF_WSCALE_FLAG) { 6465 /* Remove scale factor from initial 6466 * window */ 6467 sws = src->wscale & PF_WSCALE_MASK; 6468 win = ((u_int32_t)win + (1 << sws) - 1) 6469 >> sws; 6470 dws = dst->wscale & PF_WSCALE_MASK; 6471 } else { 6472 /* fixup other window */ 6473 dst->max_win = MIN(TCP_MAXWIN, 6474 (u_int32_t)dst->max_win << 6475 (dst->wscale & PF_WSCALE_MASK)); 6476 /* in case of a retrans SYN|ACK */ 6477 dst->wscale = 0; 6478 } 6479 } 6480 } 6481 data_end = end; 6482 if (tcp_get_flags(th) & TH_FIN) 6483 end++; 6484 6485 src->seqlo = seq; 6486 if (src->state < TCPS_SYN_SENT) 6487 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6488 6489 /* 6490 * May need to slide the window (seqhi may have been set by 6491 * the crappy stack check or if we picked up the connection 6492 * after establishment) 6493 */ 6494 if (src->seqhi == 1 || 6495 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6496 src->seqhi = end + MAX(1, dst->max_win << dws); 6497 if (win > src->max_win) 6498 src->max_win = win; 6499 6500 } else { 6501 ack = ntohl(th->th_ack) - dst->seqdiff; 6502 if (src->seqdiff) { 6503 /* Modulate sequence numbers */ 6504 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6505 src->seqdiff), 0); 6506 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6507 *copyback = 1; 6508 } 6509 end = seq + pd->p_len; 6510 if (tcp_get_flags(th) & TH_SYN) 6511 end++; 6512 data_end = end; 6513 if (tcp_get_flags(th) & TH_FIN) 6514 end++; 6515 } 6516 6517 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6518 /* Let it pass through the ack skew check */ 6519 ack = dst->seqlo; 6520 } else if ((ack == 0 && 6521 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6522 /* broken tcp stacks do not set ack */ 6523 (dst->state < TCPS_SYN_SENT)) { 6524 /* 6525 * Many stacks (ours included) will set the ACK number in an 6526 * FIN|ACK if the SYN times out -- no sequence to ACK. 6527 */ 6528 ack = dst->seqlo; 6529 } 6530 6531 if (seq == end) { 6532 /* Ease sequencing restrictions on no data packets */ 6533 seq = src->seqlo; 6534 data_end = end = seq; 6535 } 6536 6537 ackskew = dst->seqlo - ack; 6538 6539 /* 6540 * Need to demodulate the sequence numbers in any TCP SACK options 6541 * (Selective ACK). We could optionally validate the SACK values 6542 * against the current ACK window, either forwards or backwards, but 6543 * I'm not confident that SACK has been implemented properly 6544 * everywhere. It wouldn't surprise me if several stacks accidentally 6545 * SACK too far backwards of previously ACKed data. There really aren't 6546 * any security implications of bad SACKing unless the target stack 6547 * doesn't validate the option length correctly. Someone trying to 6548 * spoof into a TCP connection won't bother blindly sending SACK 6549 * options anyway. 6550 */ 6551 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6552 if (pf_modulate_sack(pd, th, dst)) 6553 *copyback = 1; 6554 } 6555 6556 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6557 if (SEQ_GEQ(src->seqhi, data_end) && 6558 /* Last octet inside other's window space */ 6559 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6560 /* Retrans: not more than one window back */ 6561 (ackskew >= -MAXACKWINDOW) && 6562 /* Acking not more than one reassembled fragment backwards */ 6563 (ackskew <= (MAXACKWINDOW << sws)) && 6564 /* Acking not more than one window forward */ 6565 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6566 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 6567 /* Require an exact/+1 sequence match on resets when possible */ 6568 6569 if (dst->scrub || src->scrub) { 6570 if (pf_normalize_tcp_stateful(pd, reason, th, 6571 *state, src, dst, copyback)) 6572 return (PF_DROP); 6573 } 6574 6575 /* update max window */ 6576 if (src->max_win < win) 6577 src->max_win = win; 6578 /* synchronize sequencing */ 6579 if (SEQ_GT(end, src->seqlo)) 6580 src->seqlo = end; 6581 /* slide the window of what the other end can send */ 6582 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6583 dst->seqhi = ack + MAX((win << sws), 1); 6584 6585 /* update states */ 6586 if (tcp_get_flags(th) & TH_SYN) 6587 if (src->state < TCPS_SYN_SENT) 6588 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6589 if (tcp_get_flags(th) & TH_FIN) 6590 if (src->state < TCPS_CLOSING) 6591 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6592 if (tcp_get_flags(th) & TH_ACK) { 6593 if (dst->state == TCPS_SYN_SENT) { 6594 pf_set_protostate(*state, pdst, 6595 TCPS_ESTABLISHED); 6596 if (src->state == TCPS_ESTABLISHED && 6597 (*state)->src_node != NULL && 6598 pf_src_connlimit(*state)) { 6599 REASON_SET(reason, PFRES_SRCLIMIT); 6600 return (PF_DROP); 6601 } 6602 } else if (dst->state == TCPS_CLOSING) 6603 pf_set_protostate(*state, pdst, 6604 TCPS_FIN_WAIT_2); 6605 } 6606 if (tcp_get_flags(th) & TH_RST) 6607 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6608 6609 /* update expire time */ 6610 (*state)->expire = pf_get_uptime(); 6611 if (src->state >= TCPS_FIN_WAIT_2 && 6612 dst->state >= TCPS_FIN_WAIT_2) 6613 (*state)->timeout = PFTM_TCP_CLOSED; 6614 else if (src->state >= TCPS_CLOSING && 6615 dst->state >= TCPS_CLOSING) 6616 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6617 else if (src->state < TCPS_ESTABLISHED || 6618 dst->state < TCPS_ESTABLISHED) 6619 (*state)->timeout = PFTM_TCP_OPENING; 6620 else if (src->state >= TCPS_CLOSING || 6621 dst->state >= TCPS_CLOSING) 6622 (*state)->timeout = PFTM_TCP_CLOSING; 6623 else 6624 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6625 6626 /* Fall through to PASS packet */ 6627 6628 } else if ((dst->state < TCPS_SYN_SENT || 6629 dst->state >= TCPS_FIN_WAIT_2 || 6630 src->state >= TCPS_FIN_WAIT_2) && 6631 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6632 /* Within a window forward of the originating packet */ 6633 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6634 /* Within a window backward of the originating packet */ 6635 6636 /* 6637 * This currently handles three situations: 6638 * 1) Stupid stacks will shotgun SYNs before their peer 6639 * replies. 6640 * 2) When PF catches an already established stream (the 6641 * firewall rebooted, the state table was flushed, routes 6642 * changed...) 6643 * 3) Packets get funky immediately after the connection 6644 * closes (this should catch Solaris spurious ACK|FINs 6645 * that web servers like to spew after a close) 6646 * 6647 * This must be a little more careful than the above code 6648 * since packet floods will also be caught here. We don't 6649 * update the TTL here to mitigate the damage of a packet 6650 * flood and so the same code can handle awkward establishment 6651 * and a loosened connection close. 6652 * In the establishment case, a correct peer response will 6653 * validate the connection, go through the normal state code 6654 * and keep updating the state TTL. 6655 */ 6656 6657 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6658 printf("pf: loose state match: "); 6659 pf_print_state(*state); 6660 pf_print_flags(tcp_get_flags(th)); 6661 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6662 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6663 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 6664 (unsigned long long)(*state)->packets[1], 6665 pd->dir == PF_IN ? "in" : "out", 6666 pd->dir == (*state)->direction ? "fwd" : "rev"); 6667 } 6668 6669 if (dst->scrub || src->scrub) { 6670 if (pf_normalize_tcp_stateful(pd, reason, th, 6671 *state, src, dst, copyback)) 6672 return (PF_DROP); 6673 } 6674 6675 /* update max window */ 6676 if (src->max_win < win) 6677 src->max_win = win; 6678 /* synchronize sequencing */ 6679 if (SEQ_GT(end, src->seqlo)) 6680 src->seqlo = end; 6681 /* slide the window of what the other end can send */ 6682 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6683 dst->seqhi = ack + MAX((win << sws), 1); 6684 6685 /* 6686 * Cannot set dst->seqhi here since this could be a shotgunned 6687 * SYN and not an already established connection. 6688 */ 6689 6690 if (tcp_get_flags(th) & TH_FIN) 6691 if (src->state < TCPS_CLOSING) 6692 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6693 if (tcp_get_flags(th) & TH_RST) 6694 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6695 6696 /* Fall through to PASS packet */ 6697 6698 } else { 6699 if ((*state)->dst.state == TCPS_SYN_SENT && 6700 (*state)->src.state == TCPS_SYN_SENT) { 6701 /* Send RST for state mismatches during handshake */ 6702 if (!(tcp_get_flags(th) & TH_RST)) 6703 pf_send_tcp((*state)->rule, pd->af, 6704 pd->dst, pd->src, th->th_dport, 6705 th->th_sport, ntohl(th->th_ack), 0, 6706 TH_RST, 0, 0, 6707 (*state)->rule->return_ttl, M_SKIP_FIREWALL, 6708 0, 0, (*state)->act.rtableid); 6709 src->seqlo = 0; 6710 src->seqhi = 1; 6711 src->max_win = 1; 6712 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6713 printf("pf: BAD state: "); 6714 pf_print_state(*state); 6715 pf_print_flags(tcp_get_flags(th)); 6716 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6717 "pkts=%llu:%llu dir=%s,%s\n", 6718 seq, orig_seq, ack, pd->p_len, ackskew, 6719 (unsigned long long)(*state)->packets[0], 6720 (unsigned long long)(*state)->packets[1], 6721 pd->dir == PF_IN ? "in" : "out", 6722 pd->dir == (*state)->direction ? "fwd" : "rev"); 6723 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6724 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6725 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6726 ' ': '2', 6727 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6728 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6729 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6730 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6731 } 6732 REASON_SET(reason, PFRES_BADSTATE); 6733 return (PF_DROP); 6734 } 6735 6736 return (PF_PASS); 6737 } 6738 6739 static int 6740 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6741 { 6742 struct tcphdr *th = &pd->hdr.tcp; 6743 struct pf_state_peer *src, *dst; 6744 u_int8_t psrc, pdst; 6745 6746 if (pd->dir == (*state)->direction) { 6747 src = &(*state)->src; 6748 dst = &(*state)->dst; 6749 psrc = PF_PEER_SRC; 6750 pdst = PF_PEER_DST; 6751 } else { 6752 src = &(*state)->dst; 6753 dst = &(*state)->src; 6754 psrc = PF_PEER_DST; 6755 pdst = PF_PEER_SRC; 6756 } 6757 6758 if (tcp_get_flags(th) & TH_SYN) 6759 if (src->state < TCPS_SYN_SENT) 6760 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6761 if (tcp_get_flags(th) & TH_FIN) 6762 if (src->state < TCPS_CLOSING) 6763 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6764 if (tcp_get_flags(th) & TH_ACK) { 6765 if (dst->state == TCPS_SYN_SENT) { 6766 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6767 if (src->state == TCPS_ESTABLISHED && 6768 (*state)->src_node != NULL && 6769 pf_src_connlimit(*state)) { 6770 REASON_SET(reason, PFRES_SRCLIMIT); 6771 return (PF_DROP); 6772 } 6773 } else if (dst->state == TCPS_CLOSING) { 6774 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6775 } else if (src->state == TCPS_SYN_SENT && 6776 dst->state < TCPS_SYN_SENT) { 6777 /* 6778 * Handle a special sloppy case where we only see one 6779 * half of the connection. If there is a ACK after 6780 * the initial SYN without ever seeing a packet from 6781 * the destination, set the connection to established. 6782 */ 6783 pf_set_protostate(*state, PF_PEER_BOTH, 6784 TCPS_ESTABLISHED); 6785 dst->state = src->state = TCPS_ESTABLISHED; 6786 if ((*state)->src_node != NULL && 6787 pf_src_connlimit(*state)) { 6788 REASON_SET(reason, PFRES_SRCLIMIT); 6789 return (PF_DROP); 6790 } 6791 } else if (src->state == TCPS_CLOSING && 6792 dst->state == TCPS_ESTABLISHED && 6793 dst->seqlo == 0) { 6794 /* 6795 * Handle the closing of half connections where we 6796 * don't see the full bidirectional FIN/ACK+ACK 6797 * handshake. 6798 */ 6799 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6800 } 6801 } 6802 if (tcp_get_flags(th) & TH_RST) 6803 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6804 6805 /* update expire time */ 6806 (*state)->expire = pf_get_uptime(); 6807 if (src->state >= TCPS_FIN_WAIT_2 && 6808 dst->state >= TCPS_FIN_WAIT_2) 6809 (*state)->timeout = PFTM_TCP_CLOSED; 6810 else if (src->state >= TCPS_CLOSING && 6811 dst->state >= TCPS_CLOSING) 6812 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6813 else if (src->state < TCPS_ESTABLISHED || 6814 dst->state < TCPS_ESTABLISHED) 6815 (*state)->timeout = PFTM_TCP_OPENING; 6816 else if (src->state >= TCPS_CLOSING || 6817 dst->state >= TCPS_CLOSING) 6818 (*state)->timeout = PFTM_TCP_CLOSING; 6819 else 6820 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6821 6822 return (PF_PASS); 6823 } 6824 6825 static int 6826 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6827 { 6828 struct pf_state_key *sk = (*state)->key[pd->didx]; 6829 struct tcphdr *th = &pd->hdr.tcp; 6830 6831 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6832 if (pd->dir != (*state)->direction) { 6833 REASON_SET(reason, PFRES_SYNPROXY); 6834 return (PF_SYNPROXY_DROP); 6835 } 6836 if (tcp_get_flags(th) & TH_SYN) { 6837 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6838 REASON_SET(reason, PFRES_SYNPROXY); 6839 return (PF_DROP); 6840 } 6841 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6842 pd->src, th->th_dport, th->th_sport, 6843 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6844 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 6845 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6846 REASON_SET(reason, PFRES_SYNPROXY); 6847 return (PF_SYNPROXY_DROP); 6848 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6849 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6850 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6851 REASON_SET(reason, PFRES_SYNPROXY); 6852 return (PF_DROP); 6853 } else if ((*state)->src_node != NULL && 6854 pf_src_connlimit(*state)) { 6855 REASON_SET(reason, PFRES_SRCLIMIT); 6856 return (PF_DROP); 6857 } else 6858 pf_set_protostate(*state, PF_PEER_SRC, 6859 PF_TCPS_PROXY_DST); 6860 } 6861 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6862 if (pd->dir == (*state)->direction) { 6863 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 6864 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6865 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6866 REASON_SET(reason, PFRES_SYNPROXY); 6867 return (PF_DROP); 6868 } 6869 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6870 if ((*state)->dst.seqhi == 1) 6871 (*state)->dst.seqhi = htonl(arc4random()); 6872 pf_send_tcp((*state)->rule, pd->af, 6873 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6874 sk->port[pd->sidx], sk->port[pd->didx], 6875 (*state)->dst.seqhi, 0, TH_SYN, 0, 6876 (*state)->src.mss, 0, 6877 (*state)->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6878 (*state)->tag, 0, (*state)->act.rtableid); 6879 REASON_SET(reason, PFRES_SYNPROXY); 6880 return (PF_SYNPROXY_DROP); 6881 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 6882 (TH_SYN|TH_ACK)) || 6883 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6884 REASON_SET(reason, PFRES_SYNPROXY); 6885 return (PF_DROP); 6886 } else { 6887 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6888 (*state)->dst.seqlo = ntohl(th->th_seq); 6889 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6890 pd->src, th->th_dport, th->th_sport, 6891 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6892 TH_ACK, (*state)->src.max_win, 0, 0, 0, 6893 (*state)->tag, 0, (*state)->act.rtableid); 6894 pf_send_tcp((*state)->rule, pd->af, 6895 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6896 sk->port[pd->sidx], sk->port[pd->didx], 6897 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6898 TH_ACK, (*state)->dst.max_win, 0, 0, 6899 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6900 (*state)->src.seqdiff = (*state)->dst.seqhi - 6901 (*state)->src.seqlo; 6902 (*state)->dst.seqdiff = (*state)->src.seqhi - 6903 (*state)->dst.seqlo; 6904 (*state)->src.seqhi = (*state)->src.seqlo + 6905 (*state)->dst.max_win; 6906 (*state)->dst.seqhi = (*state)->dst.seqlo + 6907 (*state)->src.max_win; 6908 (*state)->src.wscale = (*state)->dst.wscale = 0; 6909 pf_set_protostate(*state, PF_PEER_BOTH, 6910 TCPS_ESTABLISHED); 6911 REASON_SET(reason, PFRES_SYNPROXY); 6912 return (PF_SYNPROXY_DROP); 6913 } 6914 } 6915 6916 return (PF_PASS); 6917 } 6918 6919 static int 6920 pf_test_state_tcp(struct pf_kstate **state, struct pf_pdesc *pd, 6921 u_short *reason) 6922 { 6923 struct pf_state_key_cmp key; 6924 struct tcphdr *th = &pd->hdr.tcp; 6925 int copyback = 0; 6926 int action = PF_PASS; 6927 struct pf_state_peer *src, *dst; 6928 6929 bzero(&key, sizeof(key)); 6930 key.af = pd->af; 6931 key.proto = IPPROTO_TCP; 6932 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 6933 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 6934 key.port[pd->sidx] = th->th_sport; 6935 key.port[pd->didx] = th->th_dport; 6936 6937 STATE_LOOKUP(&key, *state, pd); 6938 6939 if (pd->dir == (*state)->direction) { 6940 src = &(*state)->src; 6941 dst = &(*state)->dst; 6942 } else { 6943 src = &(*state)->dst; 6944 dst = &(*state)->src; 6945 } 6946 6947 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6948 return (action); 6949 6950 if (dst->state >= TCPS_FIN_WAIT_2 && 6951 src->state >= TCPS_FIN_WAIT_2 && 6952 (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN) || 6953 ((tcp_get_flags(th) & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6954 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6955 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6956 printf("pf: state reuse "); 6957 pf_print_state(*state); 6958 pf_print_flags(tcp_get_flags(th)); 6959 printf("\n"); 6960 } 6961 /* XXX make sure it's the same direction ?? */ 6962 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6963 pf_unlink_state(*state); 6964 *state = NULL; 6965 return (PF_DROP); 6966 } 6967 6968 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6969 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6970 return (PF_DROP); 6971 } else { 6972 int ret; 6973 6974 ret = pf_tcp_track_full(state, pd, reason, 6975 ©back); 6976 if (ret == PF_DROP) 6977 return (PF_DROP); 6978 } 6979 6980 /* translate source/destination address, if necessary */ 6981 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6982 struct pf_state_key *nk; 6983 int afto, sidx, didx; 6984 6985 if (PF_REVERSED_KEY((*state)->key, pd->af)) 6986 nk = (*state)->key[pd->sidx]; 6987 else 6988 nk = (*state)->key[pd->didx]; 6989 6990 afto = pd->af != nk->af; 6991 sidx = afto ? pd->didx : pd->sidx; 6992 didx = afto ? pd->sidx : pd->didx; 6993 6994 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 6995 nk->port[sidx] != th->th_sport) 6996 pf_change_ap(pd->m, pd->src, &th->th_sport, 6997 pd->ip_sum, &th->th_sum, &nk->addr[sidx], 6998 nk->port[sidx], 0, pd->af, nk->af); 6999 7000 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7001 nk->port[didx] != th->th_dport) 7002 pf_change_ap(pd->m, pd->dst, &th->th_dport, 7003 pd->ip_sum, &th->th_sum, &nk->addr[didx], 7004 nk->port[didx], 0, pd->af, nk->af); 7005 7006 if (afto) { 7007 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7008 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7009 pd->naf = nk->af; 7010 action = PF_AFRT; 7011 } 7012 7013 copyback = 1; 7014 } 7015 7016 /* Copyback sequence modulation or stateful scrub changes if needed */ 7017 if (copyback) 7018 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 7019 7020 return (action); 7021 } 7022 7023 static int 7024 pf_test_state_udp(struct pf_kstate **state, struct pf_pdesc *pd) 7025 { 7026 struct pf_state_peer *src, *dst; 7027 struct pf_state_key_cmp key; 7028 struct udphdr *uh = &pd->hdr.udp; 7029 uint8_t psrc, pdst; 7030 int action = PF_PASS; 7031 7032 bzero(&key, sizeof(key)); 7033 key.af = pd->af; 7034 key.proto = IPPROTO_UDP; 7035 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 7036 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 7037 key.port[pd->sidx] = uh->uh_sport; 7038 key.port[pd->didx] = uh->uh_dport; 7039 7040 STATE_LOOKUP(&key, *state, pd); 7041 7042 if (pd->dir == (*state)->direction) { 7043 src = &(*state)->src; 7044 dst = &(*state)->dst; 7045 psrc = PF_PEER_SRC; 7046 pdst = PF_PEER_DST; 7047 } else { 7048 src = &(*state)->dst; 7049 dst = &(*state)->src; 7050 psrc = PF_PEER_DST; 7051 pdst = PF_PEER_SRC; 7052 } 7053 7054 /* update states */ 7055 if (src->state < PFUDPS_SINGLE) 7056 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 7057 if (dst->state == PFUDPS_SINGLE) 7058 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 7059 7060 /* update expire time */ 7061 (*state)->expire = pf_get_uptime(); 7062 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 7063 (*state)->timeout = PFTM_UDP_MULTIPLE; 7064 else 7065 (*state)->timeout = PFTM_UDP_SINGLE; 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] != uh->uh_sport) 7083 pf_change_ap(pd->m, pd->src, &uh->uh_sport, pd->ip_sum, 7084 &uh->uh_sum, &nk->addr[pd->sidx], 7085 nk->port[sidx], 1, pd->af, nk->af); 7086 7087 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7088 nk->port[didx] != uh->uh_dport) 7089 pf_change_ap(pd->m, pd->dst, &uh->uh_dport, pd->ip_sum, 7090 &uh->uh_sum, &nk->addr[pd->didx], 7091 nk->port[didx], 1, pd->af, nk->af); 7092 7093 if (afto) { 7094 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7095 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7096 pd->naf = nk->af; 7097 action = PF_AFRT; 7098 } 7099 7100 m_copyback(pd->m, pd->off, sizeof(*uh), (caddr_t)uh); 7101 } 7102 7103 return (action); 7104 } 7105 7106 static int 7107 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7108 u_short *reason) 7109 { 7110 struct pf_state_peer *src; 7111 if (pd->dir == state->direction) { 7112 if (PF_REVERSED_KEY(state->key, pd->af)) 7113 src = &state->dst; 7114 else 7115 src = &state->src; 7116 } else { 7117 if (PF_REVERSED_KEY(state->key, pd->af)) 7118 src = &state->src; 7119 else 7120 src = &state->dst; 7121 } 7122 7123 if (src->scrub != NULL) { 7124 if (src->scrub->pfss_v_tag == 0) 7125 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7126 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7127 return (PF_DROP); 7128 } 7129 7130 return (PF_PASS); 7131 } 7132 7133 static int 7134 pf_test_state_sctp(struct pf_kstate **state, struct pf_pdesc *pd, 7135 u_short *reason) 7136 { 7137 struct pf_state_key_cmp key; 7138 struct pf_state_peer *src, *dst; 7139 struct sctphdr *sh = &pd->hdr.sctp; 7140 u_int8_t psrc; //, pdst; 7141 7142 bzero(&key, sizeof(key)); 7143 key.af = pd->af; 7144 key.proto = IPPROTO_SCTP; 7145 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 7146 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 7147 key.port[pd->sidx] = sh->src_port; 7148 key.port[pd->didx] = sh->dest_port; 7149 7150 STATE_LOOKUP(&key, *state, pd); 7151 7152 if (pd->dir == (*state)->direction) { 7153 src = &(*state)->src; 7154 dst = &(*state)->dst; 7155 psrc = PF_PEER_SRC; 7156 } else { 7157 src = &(*state)->dst; 7158 dst = &(*state)->src; 7159 psrc = PF_PEER_DST; 7160 } 7161 7162 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 7163 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 7164 pd->sctp_flags & PFDESC_SCTP_INIT) { 7165 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7166 pf_unlink_state(*state); 7167 *state = NULL; 7168 return (PF_DROP); 7169 } 7170 7171 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7172 return (PF_DROP); 7173 7174 /* Track state. */ 7175 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7176 if (src->state < SCTP_COOKIE_WAIT) { 7177 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7178 (*state)->timeout = PFTM_SCTP_OPENING; 7179 } 7180 } 7181 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7182 MPASS(dst->scrub != NULL); 7183 if (dst->scrub->pfss_v_tag == 0) 7184 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7185 } 7186 7187 /* 7188 * Bind to the correct interface if we're if-bound. For multihomed 7189 * extra associations we don't know which interface that will be until 7190 * here, so we've inserted the state on V_pf_all. Fix that now. 7191 */ 7192 if ((*state)->kif == V_pfi_all && 7193 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7194 (*state)->kif = pd->kif; 7195 7196 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7197 if (src->state < SCTP_ESTABLISHED) { 7198 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7199 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7200 } 7201 } 7202 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7203 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7204 if (src->state < SCTP_SHUTDOWN_PENDING) { 7205 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7206 (*state)->timeout = PFTM_SCTP_CLOSING; 7207 } 7208 } 7209 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7210 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7211 (*state)->timeout = PFTM_SCTP_CLOSED; 7212 } 7213 7214 (*state)->expire = pf_get_uptime(); 7215 7216 /* translate source/destination address, if necessary */ 7217 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7218 uint16_t checksum = 0; 7219 struct pf_state_key *nk; 7220 int afto, sidx, didx; 7221 7222 if (PF_REVERSED_KEY((*state)->key, pd->af)) 7223 nk = (*state)->key[pd->sidx]; 7224 else 7225 nk = (*state)->key[pd->didx]; 7226 7227 afto = pd->af != nk->af; 7228 sidx = afto ? pd->didx : pd->sidx; 7229 didx = afto ? pd->sidx : pd->didx; 7230 7231 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7232 nk->port[sidx] != pd->hdr.sctp.src_port) { 7233 pf_change_ap(pd->m, pd->src, &pd->hdr.sctp.src_port, 7234 pd->ip_sum, &checksum, &nk->addr[sidx], 7235 nk->port[sidx], 1, pd->af, pd->naf); 7236 } 7237 7238 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7239 nk->port[didx] != pd->hdr.sctp.dest_port) { 7240 pf_change_ap(pd->m, pd->dst, &pd->hdr.sctp.dest_port, 7241 pd->ip_sum, &checksum, &nk->addr[didx], 7242 nk->port[didx], 1, pd->af, pd->naf); 7243 } 7244 7245 if (afto) { 7246 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7247 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7248 pd->naf = nk->af; 7249 return (PF_AFRT); 7250 } 7251 } 7252 7253 return (PF_PASS); 7254 } 7255 7256 static void 7257 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7258 { 7259 struct pf_sctp_endpoint key; 7260 struct pf_sctp_endpoint *ep; 7261 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7262 struct pf_sctp_source *i, *tmp; 7263 7264 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7265 return; 7266 7267 PF_SCTP_ENDPOINTS_LOCK(); 7268 7269 key.v_tag = s->dst.scrub->pfss_v_tag; 7270 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7271 if (ep != NULL) { 7272 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7273 if (pf_addr_cmp(&i->addr, 7274 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7275 s->key[PF_SK_WIRE]->af) == 0) { 7276 SDT_PROBE3(pf, sctp, multihome, remove, 7277 key.v_tag, s, i); 7278 TAILQ_REMOVE(&ep->sources, i, entry); 7279 free(i, M_PFTEMP); 7280 break; 7281 } 7282 } 7283 7284 if (TAILQ_EMPTY(&ep->sources)) { 7285 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7286 free(ep, M_PFTEMP); 7287 } 7288 } 7289 7290 /* Other direction. */ 7291 key.v_tag = s->src.scrub->pfss_v_tag; 7292 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7293 if (ep != NULL) { 7294 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7295 if (pf_addr_cmp(&i->addr, 7296 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7297 s->key[PF_SK_WIRE]->af) == 0) { 7298 SDT_PROBE3(pf, sctp, multihome, remove, 7299 key.v_tag, s, i); 7300 TAILQ_REMOVE(&ep->sources, i, entry); 7301 free(i, M_PFTEMP); 7302 break; 7303 } 7304 } 7305 7306 if (TAILQ_EMPTY(&ep->sources)) { 7307 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7308 free(ep, M_PFTEMP); 7309 } 7310 } 7311 7312 PF_SCTP_ENDPOINTS_UNLOCK(); 7313 } 7314 7315 static void 7316 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7317 { 7318 struct pf_sctp_endpoint key = { 7319 .v_tag = v_tag, 7320 }; 7321 struct pf_sctp_source *i; 7322 struct pf_sctp_endpoint *ep; 7323 7324 PF_SCTP_ENDPOINTS_LOCK(); 7325 7326 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7327 if (ep == NULL) { 7328 ep = malloc(sizeof(struct pf_sctp_endpoint), 7329 M_PFTEMP, M_NOWAIT); 7330 if (ep == NULL) { 7331 PF_SCTP_ENDPOINTS_UNLOCK(); 7332 return; 7333 } 7334 7335 ep->v_tag = v_tag; 7336 TAILQ_INIT(&ep->sources); 7337 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7338 } 7339 7340 /* Avoid inserting duplicates. */ 7341 TAILQ_FOREACH(i, &ep->sources, entry) { 7342 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7343 PF_SCTP_ENDPOINTS_UNLOCK(); 7344 return; 7345 } 7346 } 7347 7348 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7349 if (i == NULL) { 7350 PF_SCTP_ENDPOINTS_UNLOCK(); 7351 return; 7352 } 7353 7354 i->af = pd->af; 7355 memcpy(&i->addr, a, sizeof(*a)); 7356 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7357 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7358 7359 PF_SCTP_ENDPOINTS_UNLOCK(); 7360 } 7361 7362 static void 7363 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7364 struct pf_kstate *s, int action) 7365 { 7366 struct pf_sctp_multihome_job *j, *tmp; 7367 struct pf_sctp_source *i; 7368 int ret __unused; 7369 struct pf_kstate *sm = NULL; 7370 struct pf_krule *ra = NULL; 7371 struct pf_krule *r = &V_pf_default_rule; 7372 struct pf_kruleset *rs = NULL; 7373 bool do_extra = true; 7374 7375 PF_RULES_RLOCK_TRACKER; 7376 7377 again: 7378 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7379 if (s == NULL || action != PF_PASS) 7380 goto free; 7381 7382 /* Confirm we don't recurse here. */ 7383 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7384 7385 switch (j->op) { 7386 case SCTP_ADD_IP_ADDRESS: { 7387 uint32_t v_tag = pd->sctp_initiate_tag; 7388 7389 if (v_tag == 0) { 7390 if (s->direction == pd->dir) 7391 v_tag = s->src.scrub->pfss_v_tag; 7392 else 7393 v_tag = s->dst.scrub->pfss_v_tag; 7394 } 7395 7396 /* 7397 * Avoid duplicating states. We'll already have 7398 * created a state based on the source address of 7399 * the packet, but SCTP endpoints may also list this 7400 * address again in the INIT(_ACK) parameters. 7401 */ 7402 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7403 break; 7404 } 7405 7406 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7407 PF_RULES_RLOCK(); 7408 sm = NULL; 7409 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7410 j->pd.related_rule = s->rule; 7411 } 7412 ret = pf_test_rule(&r, &sm, 7413 &j->pd, &ra, &rs, NULL); 7414 PF_RULES_RUNLOCK(); 7415 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7416 if (ret != PF_DROP && sm != NULL) { 7417 /* Inherit v_tag values. */ 7418 if (sm->direction == s->direction) { 7419 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7420 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7421 } else { 7422 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7423 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7424 } 7425 PF_STATE_UNLOCK(sm); 7426 } else { 7427 /* If we try duplicate inserts? */ 7428 break; 7429 } 7430 7431 /* Only add the address if we've actually allowed the state. */ 7432 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7433 7434 if (! do_extra) { 7435 break; 7436 } 7437 /* 7438 * We need to do this for each of our source addresses. 7439 * Find those based on the verification tag. 7440 */ 7441 struct pf_sctp_endpoint key = { 7442 .v_tag = pd->hdr.sctp.v_tag, 7443 }; 7444 struct pf_sctp_endpoint *ep; 7445 7446 PF_SCTP_ENDPOINTS_LOCK(); 7447 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7448 if (ep == NULL) { 7449 PF_SCTP_ENDPOINTS_UNLOCK(); 7450 break; 7451 } 7452 MPASS(ep != NULL); 7453 7454 TAILQ_FOREACH(i, &ep->sources, entry) { 7455 struct pf_sctp_multihome_job *nj; 7456 7457 /* SCTP can intermingle IPv4 and IPv6. */ 7458 if (i->af != pd->af) 7459 continue; 7460 7461 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7462 if (! nj) { 7463 continue; 7464 } 7465 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7466 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7467 nj->pd.src = &nj->src; 7468 // New destination address! 7469 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7470 nj->pd.dst = &nj->dst; 7471 nj->pd.m = j->pd.m; 7472 nj->op = j->op; 7473 7474 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7475 } 7476 PF_SCTP_ENDPOINTS_UNLOCK(); 7477 7478 break; 7479 } 7480 case SCTP_DEL_IP_ADDRESS: { 7481 struct pf_state_key_cmp key; 7482 uint8_t psrc; 7483 7484 bzero(&key, sizeof(key)); 7485 key.af = j->pd.af; 7486 key.proto = IPPROTO_SCTP; 7487 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7488 PF_ACPY(&key.addr[0], j->pd.src, key.af); 7489 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 7490 key.port[0] = j->pd.hdr.sctp.src_port; 7491 key.port[1] = j->pd.hdr.sctp.dest_port; 7492 } else { /* stack side, reverse */ 7493 PF_ACPY(&key.addr[1], j->pd.src, key.af); 7494 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 7495 key.port[1] = j->pd.hdr.sctp.src_port; 7496 key.port[0] = j->pd.hdr.sctp.dest_port; 7497 } 7498 7499 sm = pf_find_state(kif, &key, j->pd.dir); 7500 if (sm != NULL) { 7501 PF_STATE_LOCK_ASSERT(sm); 7502 if (j->pd.dir == sm->direction) { 7503 psrc = PF_PEER_SRC; 7504 } else { 7505 psrc = PF_PEER_DST; 7506 } 7507 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7508 sm->timeout = PFTM_SCTP_CLOSING; 7509 PF_STATE_UNLOCK(sm); 7510 } 7511 break; 7512 default: 7513 panic("Unknown op %#x", j->op); 7514 } 7515 } 7516 7517 free: 7518 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7519 free(j, M_PFTEMP); 7520 } 7521 7522 /* We may have inserted extra work while processing the list. */ 7523 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7524 do_extra = false; 7525 goto again; 7526 } 7527 } 7528 7529 static int 7530 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7531 { 7532 int off = 0; 7533 struct pf_sctp_multihome_job *job; 7534 7535 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7536 7537 while (off < len) { 7538 struct sctp_paramhdr h; 7539 7540 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 7541 pd->af)) 7542 return (PF_DROP); 7543 7544 /* Parameters are at least 4 bytes. */ 7545 if (ntohs(h.param_length) < 4) 7546 return (PF_DROP); 7547 7548 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7549 ntohs(h.param_length)); 7550 7551 switch (ntohs(h.param_type)) { 7552 case SCTP_IPV4_ADDRESS: { 7553 struct in_addr t; 7554 7555 if (ntohs(h.param_length) != 7556 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7557 return (PF_DROP); 7558 7559 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7560 NULL, NULL, pd->af)) 7561 return (PF_DROP); 7562 7563 if (in_nullhost(t)) 7564 t.s_addr = pd->src->v4.s_addr; 7565 7566 /* 7567 * We hold the state lock (idhash) here, which means 7568 * that we can't acquire the keyhash, or we'll get a 7569 * LOR (and potentially double-lock things too). We also 7570 * can't release the state lock here, so instead we'll 7571 * enqueue this for async handling. 7572 * There's a relatively small race here, in that a 7573 * packet using the new addresses could arrive already, 7574 * but that's just though luck for it. 7575 */ 7576 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7577 if (! job) 7578 return (PF_DROP); 7579 7580 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7581 7582 memcpy(&job->pd, pd, sizeof(*pd)); 7583 7584 // New source address! 7585 memcpy(&job->src, &t, sizeof(t)); 7586 job->pd.src = &job->src; 7587 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7588 job->pd.dst = &job->dst; 7589 job->pd.m = pd->m; 7590 job->op = op; 7591 7592 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7593 break; 7594 } 7595 #ifdef INET6 7596 case SCTP_IPV6_ADDRESS: { 7597 struct in6_addr t; 7598 7599 if (ntohs(h.param_length) != 7600 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7601 return (PF_DROP); 7602 7603 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7604 NULL, NULL, pd->af)) 7605 return (PF_DROP); 7606 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7607 break; 7608 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7609 memcpy(&t, &pd->src->v6, sizeof(t)); 7610 7611 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7612 if (! job) 7613 return (PF_DROP); 7614 7615 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7616 7617 memcpy(&job->pd, pd, sizeof(*pd)); 7618 memcpy(&job->src, &t, sizeof(t)); 7619 job->pd.src = &job->src; 7620 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7621 job->pd.dst = &job->dst; 7622 job->pd.m = pd->m; 7623 job->op = op; 7624 7625 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7626 break; 7627 } 7628 #endif 7629 case SCTP_ADD_IP_ADDRESS: { 7630 int ret; 7631 struct sctp_asconf_paramhdr ah; 7632 7633 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7634 NULL, NULL, pd->af)) 7635 return (PF_DROP); 7636 7637 ret = pf_multihome_scan(start + off + sizeof(ah), 7638 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7639 SCTP_ADD_IP_ADDRESS); 7640 if (ret != PF_PASS) 7641 return (ret); 7642 break; 7643 } 7644 case SCTP_DEL_IP_ADDRESS: { 7645 int ret; 7646 struct sctp_asconf_paramhdr ah; 7647 7648 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7649 NULL, NULL, pd->af)) 7650 return (PF_DROP); 7651 ret = pf_multihome_scan(start + off + sizeof(ah), 7652 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7653 SCTP_DEL_IP_ADDRESS); 7654 if (ret != PF_PASS) 7655 return (ret); 7656 break; 7657 } 7658 default: 7659 break; 7660 } 7661 7662 off += roundup(ntohs(h.param_length), 4); 7663 } 7664 7665 return (PF_PASS); 7666 } 7667 7668 int 7669 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7670 { 7671 start += sizeof(struct sctp_init_chunk); 7672 len -= sizeof(struct sctp_init_chunk); 7673 7674 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7675 } 7676 7677 int 7678 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7679 { 7680 start += sizeof(struct sctp_asconf_chunk); 7681 len -= sizeof(struct sctp_asconf_chunk); 7682 7683 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7684 } 7685 7686 int 7687 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7688 struct pf_kstate **state, int direction, 7689 u_int16_t icmpid, u_int16_t type, int icmp_dir, 7690 int *iidx, int multi, int inner) 7691 { 7692 key->af = pd->af; 7693 key->proto = pd->proto; 7694 if (icmp_dir == PF_IN) { 7695 *iidx = pd->sidx; 7696 key->port[pd->sidx] = icmpid; 7697 key->port[pd->didx] = type; 7698 } else { 7699 *iidx = pd->didx; 7700 key->port[pd->sidx] = type; 7701 key->port[pd->didx] = icmpid; 7702 } 7703 if (pf_state_key_addr_setup(pd, key, multi)) 7704 return (PF_DROP); 7705 7706 STATE_LOOKUP(key, *state, pd); 7707 7708 if ((*state)->state_flags & PFSTATE_SLOPPY) 7709 return (-1); 7710 7711 /* Is this ICMP message flowing in right direction? */ 7712 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7713 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7714 PF_IN : PF_OUT; 7715 else 7716 direction = (*state)->direction; 7717 if ((*state)->rule->type && 7718 (((!inner && direction == pd->dir) || 7719 (inner && direction != pd->dir)) ? 7720 PF_IN : PF_OUT) != icmp_dir) { 7721 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7722 printf("pf: icmp type %d in wrong direction (%d): ", 7723 ntohs(type), icmp_dir); 7724 pf_print_state(*state); 7725 printf("\n"); 7726 } 7727 PF_STATE_UNLOCK(*state); 7728 *state = NULL; 7729 return (PF_DROP); 7730 } 7731 return (-1); 7732 } 7733 7734 static int 7735 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7736 u_short *reason) 7737 { 7738 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7739 u_int16_t *icmpsum, virtual_id, virtual_type; 7740 u_int8_t icmptype, icmpcode; 7741 int icmp_dir, iidx, ret, multi; 7742 struct pf_state_key_cmp key; 7743 #ifdef INET 7744 u_int16_t icmpid; 7745 #endif 7746 7747 MPASS(*state == NULL); 7748 7749 bzero(&key, sizeof(key)); 7750 switch (pd->proto) { 7751 #ifdef INET 7752 case IPPROTO_ICMP: 7753 icmptype = pd->hdr.icmp.icmp_type; 7754 icmpcode = pd->hdr.icmp.icmp_code; 7755 icmpid = pd->hdr.icmp.icmp_id; 7756 icmpsum = &pd->hdr.icmp.icmp_cksum; 7757 break; 7758 #endif /* INET */ 7759 #ifdef INET6 7760 case IPPROTO_ICMPV6: 7761 icmptype = pd->hdr.icmp6.icmp6_type; 7762 icmpcode = pd->hdr.icmp6.icmp6_code; 7763 #ifdef INET 7764 icmpid = pd->hdr.icmp6.icmp6_id; 7765 #endif 7766 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7767 break; 7768 #endif /* INET6 */ 7769 } 7770 7771 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 7772 &virtual_id, &virtual_type) == 0) { 7773 /* 7774 * ICMP query/reply message not related to a TCP/UDP/SCTP 7775 * packet. Search for an ICMP state. 7776 */ 7777 ret = pf_icmp_state_lookup(&key, pd, state, pd->dir, 7778 virtual_id, virtual_type, icmp_dir, &iidx, 7779 PF_ICMP_MULTI_NONE, 0); 7780 if (ret >= 0) { 7781 MPASS(*state == NULL); 7782 if (ret == PF_DROP && pd->af == AF_INET6 && 7783 icmp_dir == PF_OUT) { 7784 ret = pf_icmp_state_lookup(&key, pd, state, 7785 pd->dir, virtual_id, virtual_type, 7786 icmp_dir, &iidx, multi, 0); 7787 if (ret >= 0) { 7788 MPASS(*state == NULL); 7789 return (ret); 7790 } 7791 } else 7792 return (ret); 7793 } 7794 7795 (*state)->expire = pf_get_uptime(); 7796 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7797 7798 /* translate source/destination address, if necessary */ 7799 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7800 struct pf_state_key *nk; 7801 int afto, sidx, didx; 7802 7803 if (PF_REVERSED_KEY((*state)->key, pd->af)) 7804 nk = (*state)->key[pd->sidx]; 7805 else 7806 nk = (*state)->key[pd->didx]; 7807 7808 afto = pd->af != nk->af; 7809 sidx = afto ? pd->didx : pd->sidx; 7810 didx = afto ? pd->sidx : pd->didx; 7811 iidx = afto ? !iidx : iidx; 7812 7813 switch (pd->af) { 7814 #ifdef INET 7815 case AF_INET: 7816 #ifdef INET6 7817 if (afto) { 7818 if (pf_translate_icmp_af(AF_INET6, 7819 &pd->hdr.icmp)) 7820 return (PF_DROP); 7821 pd->proto = IPPROTO_ICMPV6; 7822 } 7823 #endif 7824 if (!afto && 7825 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 7826 pf_change_a(&saddr->v4.s_addr, 7827 pd->ip_sum, 7828 nk->addr[sidx].v4.s_addr, 7829 0); 7830 7831 if (!afto && PF_ANEQ(pd->dst, 7832 &nk->addr[didx], AF_INET)) 7833 pf_change_a(&daddr->v4.s_addr, 7834 pd->ip_sum, 7835 nk->addr[didx].v4.s_addr, 0); 7836 7837 if (nk->port[iidx] != 7838 pd->hdr.icmp.icmp_id) { 7839 pd->hdr.icmp.icmp_cksum = 7840 pf_cksum_fixup( 7841 pd->hdr.icmp.icmp_cksum, icmpid, 7842 nk->port[iidx], 0); 7843 pd->hdr.icmp.icmp_id = 7844 nk->port[iidx]; 7845 } 7846 7847 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7848 (caddr_t )&pd->hdr.icmp); 7849 break; 7850 #endif /* INET */ 7851 #ifdef INET6 7852 case AF_INET6: 7853 #ifdef INET 7854 if (afto) { 7855 if (pf_translate_icmp_af(AF_INET, 7856 &pd->hdr.icmp6)) 7857 return (PF_DROP); 7858 pd->proto = IPPROTO_ICMP; 7859 } 7860 #endif 7861 if (!afto && 7862 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 7863 pf_change_a6(saddr, 7864 &pd->hdr.icmp6.icmp6_cksum, 7865 &nk->addr[sidx], 0); 7866 7867 if (!afto && PF_ANEQ(pd->dst, 7868 &nk->addr[didx], AF_INET6)) 7869 pf_change_a6(daddr, 7870 &pd->hdr.icmp6.icmp6_cksum, 7871 &nk->addr[didx], 0); 7872 7873 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 7874 pd->hdr.icmp6.icmp6_id = 7875 nk->port[iidx]; 7876 7877 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7878 (caddr_t )&pd->hdr.icmp6); 7879 break; 7880 #endif /* INET6 */ 7881 } 7882 if (afto) { 7883 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7884 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7885 pd->naf = nk->af; 7886 return (PF_AFRT); 7887 } 7888 } 7889 return (PF_PASS); 7890 7891 } else { 7892 /* 7893 * ICMP error message in response to a TCP/UDP packet. 7894 * Extract the inner TCP/UDP header and search for that state. 7895 */ 7896 7897 struct pf_pdesc pd2; 7898 bzero(&pd2, sizeof pd2); 7899 #ifdef INET 7900 struct ip h2; 7901 #endif /* INET */ 7902 #ifdef INET6 7903 struct ip6_hdr h2_6; 7904 #endif /* INET6 */ 7905 int ipoff2 = 0; 7906 7907 pd2.af = pd->af; 7908 pd2.dir = pd->dir; 7909 /* Payload packet is from the opposite direction. */ 7910 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7911 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7912 pd2.m = pd->m; 7913 switch (pd->af) { 7914 #ifdef INET 7915 case AF_INET: 7916 /* offset of h2 in mbuf chain */ 7917 ipoff2 = pd->off + ICMP_MINLEN; 7918 7919 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7920 NULL, reason, pd2.af)) { 7921 DPFPRINTF(PF_DEBUG_MISC, 7922 ("pf: ICMP error message too short " 7923 "(ip)\n")); 7924 return (PF_DROP); 7925 } 7926 /* 7927 * ICMP error messages don't refer to non-first 7928 * fragments 7929 */ 7930 if (h2.ip_off & htons(IP_OFFMASK)) { 7931 REASON_SET(reason, PFRES_FRAG); 7932 return (PF_DROP); 7933 } 7934 7935 /* offset of protocol header that follows h2 */ 7936 pd2.off = ipoff2 + (h2.ip_hl << 2); 7937 7938 pd2.proto = h2.ip_p; 7939 pd2.src = (struct pf_addr *)&h2.ip_src; 7940 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7941 pd2.ip_sum = &h2.ip_sum; 7942 break; 7943 #endif /* INET */ 7944 #ifdef INET6 7945 case AF_INET6: 7946 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7947 7948 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7949 NULL, reason, pd2.af)) { 7950 DPFPRINTF(PF_DEBUG_MISC, 7951 ("pf: ICMP error message too short " 7952 "(ip6)\n")); 7953 return (PF_DROP); 7954 } 7955 pd2.off = ipoff2; 7956 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 7957 return (PF_DROP); 7958 7959 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7960 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7961 pd2.ip_sum = NULL; 7962 break; 7963 #endif /* INET6 */ 7964 } 7965 7966 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7967 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7968 printf("pf: BAD ICMP %d:%d outer dst: ", 7969 icmptype, icmpcode); 7970 pf_print_host(pd->src, 0, pd->af); 7971 printf(" -> "); 7972 pf_print_host(pd->dst, 0, pd->af); 7973 printf(" inner src: "); 7974 pf_print_host(pd2.src, 0, pd2.af); 7975 printf(" -> "); 7976 pf_print_host(pd2.dst, 0, pd2.af); 7977 printf("\n"); 7978 } 7979 REASON_SET(reason, PFRES_BADSTATE); 7980 return (PF_DROP); 7981 } 7982 7983 switch (pd2.proto) { 7984 case IPPROTO_TCP: { 7985 struct tcphdr th; 7986 u_int32_t seq; 7987 struct pf_state_peer *src, *dst; 7988 u_int8_t dws; 7989 int copyback = 0; 7990 7991 /* 7992 * Only the first 8 bytes of the TCP header can be 7993 * expected. Don't access any TCP header fields after 7994 * th_seq, an ackskew test is not possible. 7995 */ 7996 if (!pf_pull_hdr(pd->m, pd2.off, &th, 8, NULL, reason, 7997 pd2.af)) { 7998 DPFPRINTF(PF_DEBUG_MISC, 7999 ("pf: ICMP error message too short " 8000 "(tcp)\n")); 8001 return (PF_DROP); 8002 } 8003 8004 key.af = pd2.af; 8005 key.proto = IPPROTO_TCP; 8006 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8007 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8008 key.port[pd2.sidx] = th.th_sport; 8009 key.port[pd2.didx] = th.th_dport; 8010 8011 STATE_LOOKUP(&key, *state, pd); 8012 8013 if (pd->dir == (*state)->direction) { 8014 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 8015 src = &(*state)->src; 8016 dst = &(*state)->dst; 8017 } else { 8018 src = &(*state)->dst; 8019 dst = &(*state)->src; 8020 } 8021 } else { 8022 if (PF_REVERSED_KEY((*state)->key, pd->af)) { 8023 src = &(*state)->dst; 8024 dst = &(*state)->src; 8025 } else { 8026 src = &(*state)->src; 8027 dst = &(*state)->dst; 8028 } 8029 } 8030 8031 if (src->wscale && dst->wscale) 8032 dws = dst->wscale & PF_WSCALE_MASK; 8033 else 8034 dws = 0; 8035 8036 /* Demodulate sequence number */ 8037 seq = ntohl(th.th_seq) - src->seqdiff; 8038 if (src->seqdiff) { 8039 pf_change_a(&th.th_seq, icmpsum, 8040 htonl(seq), 0); 8041 copyback = 1; 8042 } 8043 8044 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 8045 (!SEQ_GEQ(src->seqhi, seq) || 8046 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 8047 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8048 printf("pf: BAD ICMP %d:%d ", 8049 icmptype, icmpcode); 8050 pf_print_host(pd->src, 0, pd->af); 8051 printf(" -> "); 8052 pf_print_host(pd->dst, 0, pd->af); 8053 printf(" state: "); 8054 pf_print_state(*state); 8055 printf(" seq=%u\n", seq); 8056 } 8057 REASON_SET(reason, PFRES_BADSTATE); 8058 return (PF_DROP); 8059 } else { 8060 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8061 printf("pf: OK ICMP %d:%d ", 8062 icmptype, icmpcode); 8063 pf_print_host(pd->src, 0, pd->af); 8064 printf(" -> "); 8065 pf_print_host(pd->dst, 0, pd->af); 8066 printf(" state: "); 8067 pf_print_state(*state); 8068 printf(" seq=%u\n", seq); 8069 } 8070 } 8071 8072 /* translate source/destination address, if necessary */ 8073 if ((*state)->key[PF_SK_WIRE] != 8074 (*state)->key[PF_SK_STACK]) { 8075 8076 struct pf_state_key *nk; 8077 8078 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8079 nk = (*state)->key[pd->sidx]; 8080 else 8081 nk = (*state)->key[pd->didx]; 8082 8083 #if defined(INET) && defined(INET6) 8084 int afto, sidx, didx; 8085 8086 afto = pd->af != nk->af; 8087 sidx = afto ? pd2.didx : pd2.sidx; 8088 didx = afto ? pd2.sidx : pd2.didx; 8089 8090 if (afto) { 8091 if (pf_translate_icmp_af(nk->af, 8092 &pd->hdr.icmp)) 8093 return (PF_DROP); 8094 m_copyback(pd->m, pd->off, 8095 sizeof(struct icmp6_hdr), 8096 (c_caddr_t)&pd->hdr.icmp6); 8097 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8098 &pd2, &nk->addr[sidx], 8099 &nk->addr[didx], pd->af, 8100 nk->af)) 8101 return (PF_DROP); 8102 if (nk->af == AF_INET) 8103 pd->proto = IPPROTO_ICMP; 8104 else 8105 pd->proto = IPPROTO_ICMPV6; 8106 th.th_sport = nk->port[sidx]; 8107 th.th_dport = nk->port[didx]; 8108 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)&th); 8109 PF_ACPY(pd->src, 8110 &nk->addr[pd2.sidx], nk->af); 8111 PF_ACPY(pd->dst, 8112 &nk->addr[pd2.didx], nk->af); 8113 pd->naf = nk->af; 8114 return (PF_AFRT); 8115 } 8116 #endif 8117 8118 if (PF_ANEQ(pd2.src, 8119 &nk->addr[pd2.sidx], pd2.af) || 8120 nk->port[pd2.sidx] != th.th_sport) 8121 pf_change_icmp(pd2.src, &th.th_sport, 8122 daddr, &nk->addr[pd2.sidx], 8123 nk->port[pd2.sidx], NULL, 8124 pd2.ip_sum, icmpsum, 8125 pd->ip_sum, 0, pd2.af); 8126 8127 if (PF_ANEQ(pd2.dst, 8128 &nk->addr[pd2.didx], pd2.af) || 8129 nk->port[pd2.didx] != th.th_dport) 8130 pf_change_icmp(pd2.dst, &th.th_dport, 8131 saddr, &nk->addr[pd2.didx], 8132 nk->port[pd2.didx], NULL, 8133 pd2.ip_sum, icmpsum, 8134 pd->ip_sum, 0, pd2.af); 8135 copyback = 1; 8136 } 8137 8138 if (copyback) { 8139 switch (pd2.af) { 8140 #ifdef INET 8141 case AF_INET: 8142 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8143 (caddr_t )&pd->hdr.icmp); 8144 m_copyback(pd->m, ipoff2, sizeof(h2), 8145 (caddr_t )&h2); 8146 break; 8147 #endif /* INET */ 8148 #ifdef INET6 8149 case AF_INET6: 8150 m_copyback(pd->m, pd->off, 8151 sizeof(struct icmp6_hdr), 8152 (caddr_t )&pd->hdr.icmp6); 8153 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8154 (caddr_t )&h2_6); 8155 break; 8156 #endif /* INET6 */ 8157 } 8158 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 8159 } 8160 8161 return (PF_PASS); 8162 break; 8163 } 8164 case IPPROTO_UDP: { 8165 struct udphdr uh; 8166 8167 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 8168 NULL, reason, pd2.af)) { 8169 DPFPRINTF(PF_DEBUG_MISC, 8170 ("pf: ICMP error message too short " 8171 "(udp)\n")); 8172 return (PF_DROP); 8173 } 8174 8175 key.af = pd2.af; 8176 key.proto = IPPROTO_UDP; 8177 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8178 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8179 key.port[pd2.sidx] = uh.uh_sport; 8180 key.port[pd2.didx] = uh.uh_dport; 8181 8182 STATE_LOOKUP(&key, *state, pd); 8183 8184 /* translate source/destination address, if necessary */ 8185 if ((*state)->key[PF_SK_WIRE] != 8186 (*state)->key[PF_SK_STACK]) { 8187 struct pf_state_key *nk; 8188 8189 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8190 nk = (*state)->key[pd->sidx]; 8191 else 8192 nk = (*state)->key[pd->didx]; 8193 8194 #if defined(INET) && defined(INET6) 8195 int afto, sidx, didx; 8196 8197 afto = pd->af != nk->af; 8198 sidx = afto ? pd2.didx : pd2.sidx; 8199 didx = afto ? pd2.sidx : pd2.didx; 8200 8201 if (afto) { 8202 if (pf_translate_icmp_af(nk->af, 8203 &pd->hdr.icmp)) 8204 return (PF_DROP); 8205 m_copyback(pd->m, pd->off, 8206 sizeof(struct icmp6_hdr), 8207 (c_caddr_t)&pd->hdr.icmp6); 8208 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8209 &pd2, &nk->addr[sidx], 8210 &nk->addr[didx], pd->af, 8211 nk->af)) 8212 return (PF_DROP); 8213 if (nk->af == AF_INET) 8214 pd->proto = IPPROTO_ICMP; 8215 else 8216 pd->proto = IPPROTO_ICMPV6; 8217 pf_change_ap(pd->m, pd2.src, &uh.uh_sport, 8218 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.sidx], 8219 nk->port[sidx], 1, pd->af, nk->af); 8220 pf_change_ap(pd->m, pd2.dst, &uh.uh_dport, 8221 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.didx], 8222 nk->port[didx], 1, pd->af, nk->af); 8223 m_copyback(pd2.m, pd2.off, sizeof(uh), 8224 (c_caddr_t)&uh); 8225 PF_ACPY(&pd->nsaddr, 8226 &nk->addr[pd2.sidx], nk->af); 8227 PF_ACPY(&pd->ndaddr, 8228 &nk->addr[pd2.didx], nk->af); 8229 pd->naf = nk->af; 8230 return (PF_AFRT); 8231 } 8232 #endif 8233 8234 if (PF_ANEQ(pd2.src, 8235 &nk->addr[pd2.sidx], pd2.af) || 8236 nk->port[pd2.sidx] != uh.uh_sport) 8237 pf_change_icmp(pd2.src, &uh.uh_sport, 8238 daddr, &nk->addr[pd2.sidx], 8239 nk->port[pd2.sidx], &uh.uh_sum, 8240 pd2.ip_sum, icmpsum, 8241 pd->ip_sum, 1, pd2.af); 8242 8243 if (PF_ANEQ(pd2.dst, 8244 &nk->addr[pd2.didx], pd2.af) || 8245 nk->port[pd2.didx] != uh.uh_dport) 8246 pf_change_icmp(pd2.dst, &uh.uh_dport, 8247 saddr, &nk->addr[pd2.didx], 8248 nk->port[pd2.didx], &uh.uh_sum, 8249 pd2.ip_sum, icmpsum, 8250 pd->ip_sum, 1, pd2.af); 8251 8252 switch (pd2.af) { 8253 #ifdef INET 8254 case AF_INET: 8255 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8256 (caddr_t )&pd->hdr.icmp); 8257 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8258 break; 8259 #endif /* INET */ 8260 #ifdef INET6 8261 case AF_INET6: 8262 m_copyback(pd->m, pd->off, 8263 sizeof(struct icmp6_hdr), 8264 (caddr_t )&pd->hdr.icmp6); 8265 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8266 (caddr_t )&h2_6); 8267 break; 8268 #endif /* INET6 */ 8269 } 8270 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 8271 } 8272 return (PF_PASS); 8273 break; 8274 } 8275 #ifdef INET 8276 case IPPROTO_SCTP: { 8277 struct sctphdr sh; 8278 struct pf_state_peer *src; 8279 int copyback = 0; 8280 8281 if (! pf_pull_hdr(pd->m, pd2.off, &sh, sizeof(sh), NULL, reason, 8282 pd2.af)) { 8283 DPFPRINTF(PF_DEBUG_MISC, 8284 ("pf: ICMP error message too short " 8285 "(sctp)\n")); 8286 return (PF_DROP); 8287 } 8288 8289 key.af = pd2.af; 8290 key.proto = IPPROTO_SCTP; 8291 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8292 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8293 key.port[pd2.sidx] = sh.src_port; 8294 key.port[pd2.didx] = sh.dest_port; 8295 8296 STATE_LOOKUP(&key, *state, pd); 8297 8298 if (pd->dir == (*state)->direction) { 8299 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8300 src = &(*state)->src; 8301 else 8302 src = &(*state)->dst; 8303 } else { 8304 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8305 src = &(*state)->dst; 8306 else 8307 src = &(*state)->src; 8308 } 8309 8310 if (src->scrub->pfss_v_tag != sh.v_tag) { 8311 DPFPRINTF(PF_DEBUG_MISC, 8312 ("pf: ICMP error message has incorrect " 8313 "SCTP v_tag\n")); 8314 return (PF_DROP); 8315 } 8316 8317 /* translate source/destination address, if necessary */ 8318 if ((*state)->key[PF_SK_WIRE] != 8319 (*state)->key[PF_SK_STACK]) { 8320 8321 struct pf_state_key *nk; 8322 8323 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8324 nk = (*state)->key[pd->sidx]; 8325 else 8326 nk = (*state)->key[pd->didx]; 8327 8328 #if defined(INET) && defined(INET6) 8329 int afto, sidx, didx; 8330 8331 afto = pd->af != nk->af; 8332 sidx = afto ? pd2.didx : pd2.sidx; 8333 didx = afto ? pd2.sidx : pd2.didx; 8334 8335 if (afto) { 8336 if (pf_translate_icmp_af(nk->af, 8337 &pd->hdr.icmp)) 8338 return (PF_DROP); 8339 m_copyback(pd->m, pd->off, 8340 sizeof(struct icmp6_hdr), 8341 (c_caddr_t)&pd->hdr.icmp6); 8342 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8343 &pd2, &nk->addr[sidx], 8344 &nk->addr[didx], pd->af, 8345 nk->af)) 8346 return (PF_DROP); 8347 if (nk->af == AF_INET) 8348 pd->proto = IPPROTO_ICMP; 8349 else 8350 pd->proto = IPPROTO_ICMPV6; 8351 sh.src_port = nk->port[sidx]; 8352 sh.dest_port = nk->port[didx]; 8353 m_copyback(pd2.m, pd2.off, sizeof(sh), (c_caddr_t)&sh); 8354 PF_ACPY(pd->src, 8355 &nk->addr[pd2.sidx], nk->af); 8356 PF_ACPY(pd->dst, 8357 &nk->addr[pd2.didx], nk->af); 8358 pd->naf = nk->af; 8359 return (PF_AFRT); 8360 } 8361 #endif 8362 8363 if (PF_ANEQ(pd2.src, 8364 &nk->addr[pd2.sidx], pd2.af) || 8365 nk->port[pd2.sidx] != sh.src_port) 8366 pf_change_icmp(pd2.src, &sh.src_port, 8367 daddr, &nk->addr[pd2.sidx], 8368 nk->port[pd2.sidx], NULL, 8369 pd2.ip_sum, icmpsum, 8370 pd->ip_sum, 0, pd2.af); 8371 8372 if (PF_ANEQ(pd2.dst, 8373 &nk->addr[pd2.didx], pd2.af) || 8374 nk->port[pd2.didx] != sh.dest_port) 8375 pf_change_icmp(pd2.dst, &sh.dest_port, 8376 saddr, &nk->addr[pd2.didx], 8377 nk->port[pd2.didx], NULL, 8378 pd2.ip_sum, icmpsum, 8379 pd->ip_sum, 0, pd2.af); 8380 copyback = 1; 8381 } 8382 8383 if (copyback) { 8384 switch (pd2.af) { 8385 #ifdef INET 8386 case AF_INET: 8387 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8388 (caddr_t )&pd->hdr.icmp); 8389 m_copyback(pd->m, ipoff2, sizeof(h2), 8390 (caddr_t )&h2); 8391 break; 8392 #endif /* INET */ 8393 #ifdef INET6 8394 case AF_INET6: 8395 m_copyback(pd->m, pd->off, 8396 sizeof(struct icmp6_hdr), 8397 (caddr_t )&pd->hdr.icmp6); 8398 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8399 (caddr_t )&h2_6); 8400 break; 8401 #endif /* INET6 */ 8402 } 8403 m_copyback(pd->m, pd2.off, sizeof(sh), (caddr_t)&sh); 8404 } 8405 8406 return (PF_PASS); 8407 break; 8408 } 8409 case IPPROTO_ICMP: { 8410 struct icmp *iih = &pd2.hdr.icmp; 8411 8412 if (pd2.af != AF_INET) { 8413 REASON_SET(reason, PFRES_NORM); 8414 return (PF_DROP); 8415 } 8416 8417 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8418 NULL, reason, pd2.af)) { 8419 DPFPRINTF(PF_DEBUG_MISC, 8420 ("pf: ICMP error message too short i" 8421 "(icmp)\n")); 8422 return (PF_DROP); 8423 } 8424 8425 icmpid = iih->icmp_id; 8426 pf_icmp_mapping(&pd2, iih->icmp_type, 8427 &icmp_dir, &multi, &virtual_id, &virtual_type); 8428 8429 ret = pf_icmp_state_lookup(&key, &pd2, state, 8430 pd2.dir, virtual_id, virtual_type, 8431 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 8432 if (ret >= 0) { 8433 MPASS(*state == NULL); 8434 return (ret); 8435 } 8436 8437 /* translate source/destination address, if necessary */ 8438 if ((*state)->key[PF_SK_WIRE] != 8439 (*state)->key[PF_SK_STACK]) { 8440 struct pf_state_key *nk; 8441 8442 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8443 nk = (*state)->key[pd->sidx]; 8444 else 8445 nk = (*state)->key[pd->didx]; 8446 8447 #if defined(INET) && defined(INET6) 8448 int afto, sidx, didx; 8449 8450 afto = pd->af != nk->af; 8451 sidx = afto ? pd2.didx : pd2.sidx; 8452 didx = afto ? pd2.sidx : pd2.didx; 8453 iidx = afto ? !iidx : iidx; 8454 8455 if (afto) { 8456 if (nk->af != AF_INET6) 8457 return (PF_DROP); 8458 if (pf_translate_icmp_af(nk->af, 8459 &pd->hdr.icmp)) 8460 return (PF_DROP); 8461 m_copyback(pd->m, pd->off, 8462 sizeof(struct icmp6_hdr), 8463 (c_caddr_t)&pd->hdr.icmp6); 8464 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8465 &pd2, &nk->addr[sidx], 8466 &nk->addr[didx], pd->af, 8467 nk->af)) 8468 return (PF_DROP); 8469 pd->proto = IPPROTO_ICMPV6; 8470 if (pf_translate_icmp_af(nk->af, &iih)) 8471 return (PF_DROP); 8472 if (virtual_type == htons(ICMP_ECHO) && 8473 nk->port[iidx] != iih->icmp_id) 8474 iih->icmp_id = nk->port[iidx]; 8475 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8476 (c_caddr_t)&iih); 8477 PF_ACPY(&pd->nsaddr, 8478 &nk->addr[pd2.sidx], nk->af); 8479 PF_ACPY(&pd->ndaddr, 8480 &nk->addr[pd2.didx], nk->af); 8481 pd->naf = nk->af; 8482 return (PF_AFRT); 8483 } 8484 #endif 8485 8486 if (PF_ANEQ(pd2.src, 8487 &nk->addr[pd2.sidx], pd2.af) || 8488 (virtual_type == htons(ICMP_ECHO) && 8489 nk->port[iidx] != iih->icmp_id)) 8490 pf_change_icmp(pd2.src, 8491 (virtual_type == htons(ICMP_ECHO)) ? 8492 &iih->icmp_id : NULL, 8493 daddr, &nk->addr[pd2.sidx], 8494 (virtual_type == htons(ICMP_ECHO)) ? 8495 nk->port[iidx] : 0, NULL, 8496 pd2.ip_sum, icmpsum, 8497 pd->ip_sum, 0, AF_INET); 8498 8499 if (PF_ANEQ(pd2.dst, 8500 &nk->addr[pd2.didx], pd2.af)) 8501 pf_change_icmp(pd2.dst, NULL, NULL, 8502 &nk->addr[pd2.didx], 0, NULL, 8503 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8504 AF_INET); 8505 8506 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8507 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8508 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8509 } 8510 return (PF_PASS); 8511 break; 8512 } 8513 #endif /* INET */ 8514 #ifdef INET6 8515 case IPPROTO_ICMPV6: { 8516 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8517 8518 if (pd2.af != AF_INET6) { 8519 REASON_SET(reason, PFRES_NORM); 8520 return (PF_DROP); 8521 } 8522 8523 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8524 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 8525 DPFPRINTF(PF_DEBUG_MISC, 8526 ("pf: ICMP error message too short " 8527 "(icmp6)\n")); 8528 return (PF_DROP); 8529 } 8530 8531 pf_icmp_mapping(&pd2, iih->icmp6_type, 8532 &icmp_dir, &multi, &virtual_id, &virtual_type); 8533 8534 ret = pf_icmp_state_lookup(&key, &pd2, state, 8535 pd->dir, virtual_id, virtual_type, 8536 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 8537 if (ret >= 0) { 8538 MPASS(*state == NULL); 8539 if (ret == PF_DROP && pd2.af == AF_INET6 && 8540 icmp_dir == PF_OUT) { 8541 ret = pf_icmp_state_lookup(&key, &pd2, 8542 state, pd->dir, 8543 virtual_id, virtual_type, 8544 icmp_dir, &iidx, multi, 1); 8545 if (ret >= 0) { 8546 MPASS(*state == NULL); 8547 return (ret); 8548 } 8549 } else 8550 return (ret); 8551 } 8552 8553 /* translate source/destination address, if necessary */ 8554 if ((*state)->key[PF_SK_WIRE] != 8555 (*state)->key[PF_SK_STACK]) { 8556 struct pf_state_key *nk; 8557 8558 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8559 nk = (*state)->key[pd->sidx]; 8560 else 8561 nk = (*state)->key[pd->didx]; 8562 8563 #if defined(INET) && defined(INET6) 8564 int afto, sidx, didx; 8565 8566 afto = pd->af != nk->af; 8567 sidx = afto ? pd2.didx : pd2.sidx; 8568 didx = afto ? pd2.sidx : pd2.didx; 8569 iidx = afto ? !iidx : iidx; 8570 8571 if (afto) { 8572 if (nk->af != AF_INET) 8573 return (PF_DROP); 8574 if (pf_translate_icmp_af(nk->af, 8575 &pd->hdr.icmp)) 8576 return (PF_DROP); 8577 m_copyback(pd->m, pd->off, 8578 sizeof(struct icmp6_hdr), 8579 (c_caddr_t)&pd->hdr.icmp6); 8580 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8581 &pd2, &nk->addr[sidx], 8582 &nk->addr[didx], pd->af, 8583 nk->af)) 8584 return (PF_DROP); 8585 pd->proto = IPPROTO_ICMP; 8586 if (pf_translate_icmp_af(nk->af, &iih)) 8587 return (PF_DROP); 8588 if (virtual_type == 8589 htons(ICMP6_ECHO_REQUEST) && 8590 nk->port[iidx] != iih->icmp6_id) 8591 iih->icmp6_id = nk->port[iidx]; 8592 m_copyback(pd2.m, pd2.off, 8593 sizeof(struct icmp6_hdr), (c_caddr_t)&iih); 8594 PF_ACPY(&pd->nsaddr, 8595 &nk->addr[pd2.sidx], nk->af); 8596 PF_ACPY(&pd->ndaddr, 8597 &nk->addr[pd2.didx], nk->af); 8598 pd->naf = nk->af; 8599 return (PF_AFRT); 8600 } 8601 #endif 8602 8603 if (PF_ANEQ(pd2.src, 8604 &nk->addr[pd2.sidx], pd2.af) || 8605 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8606 nk->port[pd2.sidx] != iih->icmp6_id)) 8607 pf_change_icmp(pd2.src, 8608 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8609 ? &iih->icmp6_id : NULL, 8610 daddr, &nk->addr[pd2.sidx], 8611 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8612 ? nk->port[iidx] : 0, NULL, 8613 pd2.ip_sum, icmpsum, 8614 pd->ip_sum, 0, AF_INET6); 8615 8616 if (PF_ANEQ(pd2.dst, 8617 &nk->addr[pd2.didx], pd2.af)) 8618 pf_change_icmp(pd2.dst, NULL, NULL, 8619 &nk->addr[pd2.didx], 0, NULL, 8620 pd2.ip_sum, icmpsum, 8621 pd->ip_sum, 0, AF_INET6); 8622 8623 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8624 (caddr_t)&pd->hdr.icmp6); 8625 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8626 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8627 (caddr_t)iih); 8628 } 8629 return (PF_PASS); 8630 break; 8631 } 8632 #endif /* INET6 */ 8633 default: { 8634 key.af = pd2.af; 8635 key.proto = pd2.proto; 8636 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8637 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8638 key.port[0] = key.port[1] = 0; 8639 8640 STATE_LOOKUP(&key, *state, pd); 8641 8642 /* translate source/destination address, if necessary */ 8643 if ((*state)->key[PF_SK_WIRE] != 8644 (*state)->key[PF_SK_STACK]) { 8645 struct pf_state_key *nk = 8646 (*state)->key[pd->didx]; 8647 8648 if (PF_ANEQ(pd2.src, 8649 &nk->addr[pd2.sidx], pd2.af)) 8650 pf_change_icmp(pd2.src, NULL, daddr, 8651 &nk->addr[pd2.sidx], 0, NULL, 8652 pd2.ip_sum, icmpsum, 8653 pd->ip_sum, 0, pd2.af); 8654 8655 if (PF_ANEQ(pd2.dst, 8656 &nk->addr[pd2.didx], pd2.af)) 8657 pf_change_icmp(pd2.dst, NULL, saddr, 8658 &nk->addr[pd2.didx], 0, NULL, 8659 pd2.ip_sum, icmpsum, 8660 pd->ip_sum, 0, pd2.af); 8661 8662 switch (pd2.af) { 8663 #ifdef INET 8664 case AF_INET: 8665 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8666 (caddr_t)&pd->hdr.icmp); 8667 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8668 break; 8669 #endif /* INET */ 8670 #ifdef INET6 8671 case AF_INET6: 8672 m_copyback(pd->m, pd->off, 8673 sizeof(struct icmp6_hdr), 8674 (caddr_t )&pd->hdr.icmp6); 8675 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8676 (caddr_t )&h2_6); 8677 break; 8678 #endif /* INET6 */ 8679 } 8680 } 8681 return (PF_PASS); 8682 break; 8683 } 8684 } 8685 } 8686 } 8687 8688 static int 8689 pf_test_state_other(struct pf_kstate **state, struct pf_pdesc *pd) 8690 { 8691 struct pf_state_peer *src, *dst; 8692 struct pf_state_key_cmp key; 8693 uint8_t psrc, pdst; 8694 int action = PF_PASS; 8695 8696 bzero(&key, sizeof(key)); 8697 key.af = pd->af; 8698 key.proto = pd->proto; 8699 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 8700 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 8701 key.port[0] = key.port[1] = 0; 8702 8703 STATE_LOOKUP(&key, *state, pd); 8704 8705 if (pd->dir == (*state)->direction) { 8706 src = &(*state)->src; 8707 dst = &(*state)->dst; 8708 psrc = PF_PEER_SRC; 8709 pdst = PF_PEER_DST; 8710 } else { 8711 src = &(*state)->dst; 8712 dst = &(*state)->src; 8713 psrc = PF_PEER_DST; 8714 pdst = PF_PEER_SRC; 8715 } 8716 8717 /* update states */ 8718 if (src->state < PFOTHERS_SINGLE) 8719 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 8720 if (dst->state == PFOTHERS_SINGLE) 8721 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 8722 8723 /* update expire time */ 8724 (*state)->expire = pf_get_uptime(); 8725 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 8726 (*state)->timeout = PFTM_OTHER_MULTIPLE; 8727 else 8728 (*state)->timeout = PFTM_OTHER_SINGLE; 8729 8730 /* translate source/destination address, if necessary */ 8731 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 8732 struct pf_state_key *nk; 8733 int afto; 8734 8735 if (PF_REVERSED_KEY((*state)->key, pd->af)) 8736 nk = (*state)->key[pd->sidx]; 8737 else 8738 nk = (*state)->key[pd->didx]; 8739 8740 KASSERT(nk, ("%s: nk is null", __func__)); 8741 KASSERT(pd, ("%s: pd is null", __func__)); 8742 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 8743 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 8744 8745 afto = pd->af != nk->af; 8746 8747 switch (pd->af) { 8748 #ifdef INET 8749 case AF_INET: 8750 if (!afto && 8751 PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 8752 pf_change_a(&pd->src->v4.s_addr, 8753 pd->ip_sum, 8754 nk->addr[pd->sidx].v4.s_addr, 8755 0); 8756 8757 if (!afto && 8758 PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 8759 pf_change_a(&pd->dst->v4.s_addr, 8760 pd->ip_sum, 8761 nk->addr[pd->didx].v4.s_addr, 8762 0); 8763 8764 break; 8765 #endif /* INET */ 8766 #ifdef INET6 8767 case AF_INET6: 8768 if (!afto && 8769 PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 8770 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 8771 8772 if (!afto && 8773 PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 8774 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 8775 #endif /* INET6 */ 8776 } 8777 if (afto) { 8778 PF_ACPY(&pd->nsaddr, 8779 &nk->addr[afto ? pd->didx : pd->sidx], nk->af); 8780 PF_ACPY(&pd->ndaddr, 8781 &nk->addr[afto ? pd->sidx : pd->didx], nk->af); 8782 pd->naf = nk->af; 8783 action = PF_AFRT; 8784 } 8785 } 8786 return (action); 8787 } 8788 8789 /* 8790 * ipoff and off are measured from the start of the mbuf chain. 8791 * h must be at "ipoff" on the mbuf chain. 8792 */ 8793 void * 8794 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8795 u_short *actionp, u_short *reasonp, sa_family_t af) 8796 { 8797 switch (af) { 8798 #ifdef INET 8799 case AF_INET: { 8800 const struct ip *h = mtod(m, struct ip *); 8801 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 8802 8803 if (fragoff) { 8804 if (fragoff >= len) 8805 ACTION_SET(actionp, PF_PASS); 8806 else { 8807 ACTION_SET(actionp, PF_DROP); 8808 REASON_SET(reasonp, PFRES_FRAG); 8809 } 8810 return (NULL); 8811 } 8812 if (m->m_pkthdr.len < off + len || 8813 ntohs(h->ip_len) < off + len) { 8814 ACTION_SET(actionp, PF_DROP); 8815 REASON_SET(reasonp, PFRES_SHORT); 8816 return (NULL); 8817 } 8818 break; 8819 } 8820 #endif /* INET */ 8821 #ifdef INET6 8822 case AF_INET6: { 8823 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8824 8825 if (m->m_pkthdr.len < off + len || 8826 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 8827 (unsigned)(off + len)) { 8828 ACTION_SET(actionp, PF_DROP); 8829 REASON_SET(reasonp, PFRES_SHORT); 8830 return (NULL); 8831 } 8832 break; 8833 } 8834 #endif /* INET6 */ 8835 } 8836 m_copydata(m, off, len, p); 8837 return (p); 8838 } 8839 8840 int 8841 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 8842 int rtableid) 8843 { 8844 struct ifnet *ifp; 8845 8846 /* 8847 * Skip check for addresses with embedded interface scope, 8848 * as they would always match anyway. 8849 */ 8850 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 8851 return (1); 8852 8853 if (af != AF_INET && af != AF_INET6) 8854 return (0); 8855 8856 if (kif == V_pfi_all) 8857 return (1); 8858 8859 /* Skip checks for ipsec interfaces */ 8860 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 8861 return (1); 8862 8863 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 8864 8865 switch (af) { 8866 #ifdef INET6 8867 case AF_INET6: 8868 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 8869 ifp)); 8870 #endif 8871 #ifdef INET 8872 case AF_INET: 8873 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 8874 ifp)); 8875 #endif 8876 } 8877 8878 return (0); 8879 } 8880 8881 #ifdef INET 8882 static void 8883 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8884 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8885 { 8886 struct mbuf *m0, *m1, *md; 8887 struct sockaddr_in dst; 8888 struct ip *ip; 8889 struct ifnet *ifp = NULL; 8890 int error = 0; 8891 uint16_t ip_len, ip_off; 8892 uint16_t tmp; 8893 int r_dir; 8894 bool skip_test = false; 8895 8896 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8897 8898 SDT_PROBE4(pf, ip, route_to, entry, *m, pd, s, oifp); 8899 8900 if (s) { 8901 r_dir = s->direction; 8902 } else { 8903 r_dir = r->direction; 8904 } 8905 8906 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8907 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8908 __func__)); 8909 8910 if ((pd->pf_mtag == NULL && 8911 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8912 pd->pf_mtag->routed++ > 3) { 8913 m0 = *m; 8914 *m = NULL; 8915 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8916 goto bad_locked; 8917 } 8918 8919 if (pd->act.rt_kif != NULL) 8920 ifp = pd->act.rt_kif->pfik_ifp; 8921 8922 if (pd->act.rt == PF_DUPTO) { 8923 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8924 if (s != NULL) { 8925 PF_STATE_UNLOCK(s); 8926 } 8927 if (ifp == oifp) { 8928 /* When the 2nd interface is not skipped */ 8929 return; 8930 } else { 8931 m0 = *m; 8932 *m = NULL; 8933 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8934 goto bad; 8935 } 8936 } else { 8937 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8938 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8939 if (s) 8940 PF_STATE_UNLOCK(s); 8941 return; 8942 } 8943 } 8944 } else { 8945 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8946 if (pd->af == pd->naf) { 8947 pf_dummynet(pd, s, r, m); 8948 if (s) 8949 PF_STATE_UNLOCK(s); 8950 return; 8951 } else { 8952 skip_test = true; 8953 } 8954 } 8955 8956 /* 8957 * If we're actually doing route-to and af-to and are in the 8958 * reply direction. 8959 */ 8960 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 8961 pd->af != pd->naf) { 8962 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 8963 /* Un-set ifp so we do a plain route lookup. */ 8964 ifp = NULL; 8965 } 8966 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 8967 /* Un-set ifp so we do a plain route lookup. */ 8968 ifp = NULL; 8969 } 8970 } 8971 m0 = *m; 8972 } 8973 8974 ip = mtod(m0, struct ip *); 8975 8976 bzero(&dst, sizeof(dst)); 8977 dst.sin_family = AF_INET; 8978 dst.sin_len = sizeof(dst); 8979 dst.sin_addr = ip->ip_dst; 8980 dst.sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 8981 8982 if (s != NULL){ 8983 if (ifp == NULL && (pd->af != pd->naf)) { 8984 /* We're in the AFTO case. Do a route lookup. */ 8985 const struct nhop_object *nh; 8986 nh = fib4_lookup(M_GETFIB(*m), ip->ip_dst, 0, NHR_NONE, 0); 8987 if (nh) { 8988 ifp = nh->nh_ifp; 8989 8990 /* Use the gateway if needed. */ 8991 if (nh->nh_flags & NHF_GATEWAY) 8992 dst.sin_addr = nh->gw4_sa.sin_addr; 8993 else 8994 dst.sin_addr = ip->ip_dst; 8995 8996 /* 8997 * Bind to the correct interface if we're 8998 * if-bound. We don't know which interface 8999 * that will be until here, so we've inserted 9000 * the state on V_pf_all. Fix that now. 9001 */ 9002 if (s->kif == V_pfi_all && ifp != NULL && 9003 r->rule_flag & PFRULE_IFBOUND) 9004 s->kif = ifp->if_pf_kif; 9005 } 9006 } 9007 9008 if (r->rule_flag & PFRULE_IFBOUND && 9009 pd->act.rt == PF_REPLYTO && 9010 s->kif == V_pfi_all) { 9011 s->kif = pd->act.rt_kif; 9012 s->orig_kif = oifp->if_pf_kif; 9013 } 9014 9015 PF_STATE_UNLOCK(s); 9016 } 9017 9018 if (ifp == NULL) { 9019 m0 = *m; 9020 *m = NULL; 9021 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9022 goto bad; 9023 } 9024 9025 if (pd->dir == PF_IN && !skip_test) { 9026 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 9027 &pd->act) != PF_PASS) { 9028 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9029 goto bad; 9030 } else if (m0 == NULL) { 9031 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9032 goto done; 9033 } 9034 if (m0->m_len < sizeof(struct ip)) { 9035 DPFPRINTF(PF_DEBUG_URGENT, 9036 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 9037 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9038 goto bad; 9039 } 9040 ip = mtod(m0, struct ip *); 9041 } 9042 9043 if (ifp->if_flags & IFF_LOOPBACK) 9044 m0->m_flags |= M_SKIP_FIREWALL; 9045 9046 ip_len = ntohs(ip->ip_len); 9047 ip_off = ntohs(ip->ip_off); 9048 9049 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 9050 m0->m_pkthdr.csum_flags |= CSUM_IP; 9051 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 9052 in_delayed_cksum(m0); 9053 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 9054 } 9055 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 9056 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 9057 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 9058 } 9059 9060 if (pd->dir == PF_IN) { 9061 /* 9062 * Make sure dummynet gets the correct direction, in case it needs to 9063 * re-inject later. 9064 */ 9065 pd->dir = PF_OUT; 9066 9067 /* 9068 * The following processing is actually the rest of the inbound processing, even 9069 * though we've marked it as outbound (so we don't look through dummynet) and it 9070 * happens after the outbound processing (pf_test(PF_OUT) above). 9071 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9072 * conclusion about what direction it's processing, and we can't fix it or it 9073 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9074 * decision will pick the right pipe, and everything will mostly work as expected. 9075 */ 9076 tmp = pd->act.dnrpipe; 9077 pd->act.dnrpipe = pd->act.dnpipe; 9078 pd->act.dnpipe = tmp; 9079 } 9080 9081 /* 9082 * If small enough for interface, or the interface will take 9083 * care of the fragmentation for us, we can just send directly. 9084 */ 9085 if (ip_len <= ifp->if_mtu || 9086 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 9087 ip->ip_sum = 0; 9088 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 9089 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 9090 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 9091 } 9092 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 9093 9094 md = m0; 9095 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9096 if (md != NULL) { 9097 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 9098 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9099 } 9100 goto done; 9101 } 9102 9103 /* Balk when DF bit is set or the interface didn't support TSO. */ 9104 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 9105 error = EMSGSIZE; 9106 KMOD_IPSTAT_INC(ips_cantfrag); 9107 if (pd->act.rt != PF_DUPTO) { 9108 if (s && s->nat_rule != NULL) 9109 PACKET_UNDO_NAT(m0, pd, 9110 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 9111 s); 9112 9113 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 9114 ifp->if_mtu); 9115 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9116 goto done; 9117 } else { 9118 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9119 goto bad; 9120 } 9121 } 9122 9123 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 9124 if (error) { 9125 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9126 goto bad; 9127 } 9128 9129 for (; m0; m0 = m1) { 9130 m1 = m0->m_nextpkt; 9131 m0->m_nextpkt = NULL; 9132 if (error == 0) { 9133 m_clrprotoflags(m0); 9134 md = m0; 9135 pd->pf_mtag = pf_find_mtag(md); 9136 error = pf_dummynet_route(pd, s, r, ifp, 9137 sintosa(&dst), &md); 9138 if (md != NULL) { 9139 error = (*ifp->if_output)(ifp, md, 9140 sintosa(&dst), NULL); 9141 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9142 } 9143 } else 9144 m_freem(m0); 9145 } 9146 9147 if (error == 0) 9148 KMOD_IPSTAT_INC(ips_fragmented); 9149 9150 done: 9151 if (pd->act.rt != PF_DUPTO) 9152 *m = NULL; 9153 return; 9154 9155 bad_locked: 9156 if (s) 9157 PF_STATE_UNLOCK(s); 9158 bad: 9159 m_freem(m0); 9160 goto done; 9161 } 9162 #endif /* INET */ 9163 9164 #ifdef INET6 9165 static void 9166 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 9167 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9168 { 9169 struct mbuf *m0, *md; 9170 struct m_tag *mtag; 9171 struct sockaddr_in6 dst; 9172 struct ip6_hdr *ip6; 9173 struct ifnet *ifp = NULL; 9174 int r_dir; 9175 bool skip_test = false; 9176 9177 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 9178 9179 SDT_PROBE4(pf, ip6, route_to, entry, *m, pd, s, oifp); 9180 9181 if (s) { 9182 r_dir = s->direction; 9183 } else { 9184 r_dir = r->direction; 9185 } 9186 9187 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9188 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9189 __func__)); 9190 9191 if ((pd->pf_mtag == NULL && 9192 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 9193 pd->pf_mtag->routed++ > 3) { 9194 m0 = *m; 9195 *m = NULL; 9196 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9197 goto bad_locked; 9198 } 9199 9200 if (pd->act.rt_kif != NULL) 9201 ifp = pd->act.rt_kif->pfik_ifp; 9202 9203 if (pd->act.rt == PF_DUPTO) { 9204 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9205 if (s != NULL) { 9206 PF_STATE_UNLOCK(s); 9207 } 9208 if (ifp == oifp) { 9209 /* When the 2nd interface is not skipped */ 9210 return; 9211 } else { 9212 m0 = *m; 9213 *m = NULL; 9214 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9215 goto bad; 9216 } 9217 } else { 9218 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9219 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 9220 if (s) 9221 PF_STATE_UNLOCK(s); 9222 return; 9223 } 9224 } 9225 } else { 9226 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9227 if (pd->af == pd->naf) { 9228 pf_dummynet(pd, s, r, m); 9229 if (s) 9230 PF_STATE_UNLOCK(s); 9231 return; 9232 } else { 9233 skip_test = true; 9234 } 9235 } 9236 9237 /* 9238 * If we're actually doing route-to and af-to and are in the 9239 * reply direction. 9240 */ 9241 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9242 pd->af != pd->naf) { 9243 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9244 /* Un-set ifp so we do a plain route lookup. */ 9245 ifp = NULL; 9246 } 9247 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9248 /* Un-set ifp so we do a plain route lookup. */ 9249 ifp = NULL; 9250 } 9251 } 9252 m0 = *m; 9253 } 9254 9255 ip6 = mtod(m0, struct ip6_hdr *); 9256 9257 bzero(&dst, sizeof(dst)); 9258 dst.sin6_family = AF_INET6; 9259 dst.sin6_len = sizeof(dst); 9260 dst.sin6_addr = ip6->ip6_dst; 9261 PF_ACPY((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, AF_INET6); 9262 9263 if (s != NULL) { 9264 if (ifp == NULL && (pd->af != pd->naf)) { 9265 const struct nhop_object *nh; 9266 nh = fib6_lookup(M_GETFIB(*m), &ip6->ip6_dst, 0, NHR_NONE, 0); 9267 if (nh) { 9268 ifp = nh->nh_ifp; 9269 9270 /* Use the gateway if needed. */ 9271 if (nh->nh_flags & NHF_GATEWAY) 9272 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9273 sizeof(dst.sin6_addr)); 9274 else 9275 dst.sin6_addr = ip6->ip6_dst; 9276 9277 /* 9278 * Bind to the correct interface if we're 9279 * if-bound. We don't know which interface 9280 * that will be until here, so we've inserted 9281 * the state on V_pf_all. Fix that now. 9282 */ 9283 if (s->kif == V_pfi_all && ifp != NULL && 9284 r->rule_flag & PFRULE_IFBOUND) 9285 s->kif = ifp->if_pf_kif; 9286 } 9287 } 9288 9289 if (r->rule_flag & PFRULE_IFBOUND && 9290 pd->act.rt == PF_REPLYTO && 9291 s->kif == V_pfi_all) { 9292 s->kif = pd->act.rt_kif; 9293 s->orig_kif = oifp->if_pf_kif; 9294 } 9295 9296 PF_STATE_UNLOCK(s); 9297 } 9298 9299 if (pd->af != pd->naf) { 9300 struct udphdr *uh = &pd->hdr.udp; 9301 9302 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9303 uh->uh_sum = in6_cksum_pseudo(ip6, 9304 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9305 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9306 } 9307 } 9308 9309 if (ifp == NULL) { 9310 m0 = *m; 9311 *m = NULL; 9312 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9313 goto bad; 9314 } 9315 9316 if (pd->dir == PF_IN && !skip_test) { 9317 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9318 ifp, &m0, inp, &pd->act) != PF_PASS) { 9319 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9320 goto bad; 9321 } else if (m0 == NULL) { 9322 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9323 goto done; 9324 } 9325 if (m0->m_len < sizeof(struct ip6_hdr)) { 9326 DPFPRINTF(PF_DEBUG_URGENT, 9327 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 9328 __func__)); 9329 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9330 goto bad; 9331 } 9332 ip6 = mtod(m0, struct ip6_hdr *); 9333 } 9334 9335 if (ifp->if_flags & IFF_LOOPBACK) 9336 m0->m_flags |= M_SKIP_FIREWALL; 9337 9338 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9339 ~ifp->if_hwassist) { 9340 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9341 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9342 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9343 } 9344 9345 if (pd->dir == PF_IN) { 9346 uint16_t tmp; 9347 /* 9348 * Make sure dummynet gets the correct direction, in case it needs to 9349 * re-inject later. 9350 */ 9351 pd->dir = PF_OUT; 9352 9353 /* 9354 * The following processing is actually the rest of the inbound processing, even 9355 * though we've marked it as outbound (so we don't look through dummynet) and it 9356 * happens after the outbound processing (pf_test(PF_OUT) above). 9357 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9358 * conclusion about what direction it's processing, and we can't fix it or it 9359 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9360 * decision will pick the right pipe, and everything will mostly work as expected. 9361 */ 9362 tmp = pd->act.dnrpipe; 9363 pd->act.dnrpipe = pd->act.dnpipe; 9364 pd->act.dnpipe = tmp; 9365 } 9366 9367 /* 9368 * If the packet is too large for the outgoing interface, 9369 * send back an icmp6 error. 9370 */ 9371 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9372 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9373 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9374 if (mtag != NULL) { 9375 int ret __sdt_used; 9376 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9377 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9378 goto done; 9379 } 9380 9381 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9382 md = m0; 9383 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9384 if (md != NULL) { 9385 int ret __sdt_used; 9386 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9387 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9388 } 9389 } 9390 else { 9391 in6_ifstat_inc(ifp, ifs6_in_toobig); 9392 if (pd->act.rt != PF_DUPTO) { 9393 if (s && s->nat_rule != NULL) 9394 PACKET_UNDO_NAT(m0, pd, 9395 ((caddr_t)ip6 - m0->m_data) + 9396 sizeof(struct ip6_hdr), s); 9397 9398 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 9399 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9400 } else { 9401 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9402 goto bad; 9403 } 9404 } 9405 9406 done: 9407 if (pd->act.rt != PF_DUPTO) 9408 *m = NULL; 9409 return; 9410 9411 bad_locked: 9412 if (s) 9413 PF_STATE_UNLOCK(s); 9414 bad: 9415 m_freem(m0); 9416 goto done; 9417 } 9418 #endif /* INET6 */ 9419 9420 /* 9421 * FreeBSD supports cksum offloads for the following drivers. 9422 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9423 * 9424 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9425 * network driver performed cksum including pseudo header, need to verify 9426 * csum_data 9427 * CSUM_DATA_VALID : 9428 * network driver performed cksum, needs to additional pseudo header 9429 * cksum computation with partial csum_data(i.e. lack of H/W support for 9430 * pseudo header, for instance sk(4) and possibly gem(4)) 9431 * 9432 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9433 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9434 * TCP/UDP layer. 9435 * Also, set csum_data to 0xffff to force cksum validation. 9436 */ 9437 static int 9438 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9439 { 9440 u_int16_t sum = 0; 9441 int hw_assist = 0; 9442 struct ip *ip; 9443 9444 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9445 return (1); 9446 if (m->m_pkthdr.len < off + len) 9447 return (1); 9448 9449 switch (p) { 9450 case IPPROTO_TCP: 9451 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9452 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9453 sum = m->m_pkthdr.csum_data; 9454 } else { 9455 ip = mtod(m, struct ip *); 9456 sum = in_pseudo(ip->ip_src.s_addr, 9457 ip->ip_dst.s_addr, htonl((u_short)len + 9458 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9459 } 9460 sum ^= 0xffff; 9461 ++hw_assist; 9462 } 9463 break; 9464 case IPPROTO_UDP: 9465 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9466 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9467 sum = m->m_pkthdr.csum_data; 9468 } else { 9469 ip = mtod(m, struct ip *); 9470 sum = in_pseudo(ip->ip_src.s_addr, 9471 ip->ip_dst.s_addr, htonl((u_short)len + 9472 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9473 } 9474 sum ^= 0xffff; 9475 ++hw_assist; 9476 } 9477 break; 9478 case IPPROTO_ICMP: 9479 #ifdef INET6 9480 case IPPROTO_ICMPV6: 9481 #endif /* INET6 */ 9482 break; 9483 default: 9484 return (1); 9485 } 9486 9487 if (!hw_assist) { 9488 switch (af) { 9489 case AF_INET: 9490 if (p == IPPROTO_ICMP) { 9491 if (m->m_len < off) 9492 return (1); 9493 m->m_data += off; 9494 m->m_len -= off; 9495 sum = in_cksum(m, len); 9496 m->m_data -= off; 9497 m->m_len += off; 9498 } else { 9499 if (m->m_len < sizeof(struct ip)) 9500 return (1); 9501 sum = in4_cksum(m, p, off, len); 9502 } 9503 break; 9504 #ifdef INET6 9505 case AF_INET6: 9506 if (m->m_len < sizeof(struct ip6_hdr)) 9507 return (1); 9508 sum = in6_cksum(m, p, off, len); 9509 break; 9510 #endif /* INET6 */ 9511 } 9512 } 9513 if (sum) { 9514 switch (p) { 9515 case IPPROTO_TCP: 9516 { 9517 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9518 break; 9519 } 9520 case IPPROTO_UDP: 9521 { 9522 KMOD_UDPSTAT_INC(udps_badsum); 9523 break; 9524 } 9525 #ifdef INET 9526 case IPPROTO_ICMP: 9527 { 9528 KMOD_ICMPSTAT_INC(icps_checksum); 9529 break; 9530 } 9531 #endif 9532 #ifdef INET6 9533 case IPPROTO_ICMPV6: 9534 { 9535 KMOD_ICMP6STAT_INC(icp6s_checksum); 9536 break; 9537 } 9538 #endif /* INET6 */ 9539 } 9540 return (1); 9541 } else { 9542 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9543 m->m_pkthdr.csum_flags |= 9544 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9545 m->m_pkthdr.csum_data = 0xffff; 9546 } 9547 } 9548 return (0); 9549 } 9550 9551 static bool 9552 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9553 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9554 { 9555 int dndir = r->direction; 9556 9557 if (s && dndir == PF_INOUT) { 9558 dndir = s->direction; 9559 } else if (dndir == PF_INOUT) { 9560 /* Assume primary direction. Happens when we've set dnpipe in 9561 * the ethernet level code. */ 9562 dndir = pd->dir; 9563 } 9564 9565 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9566 return (false); 9567 9568 memset(dnflow, 0, sizeof(*dnflow)); 9569 9570 if (pd->dport != NULL) 9571 dnflow->f_id.dst_port = ntohs(*pd->dport); 9572 if (pd->sport != NULL) 9573 dnflow->f_id.src_port = ntohs(*pd->sport); 9574 9575 if (pd->dir == PF_IN) 9576 dnflow->flags |= IPFW_ARGS_IN; 9577 else 9578 dnflow->flags |= IPFW_ARGS_OUT; 9579 9580 if (pd->dir != dndir && pd->act.dnrpipe) { 9581 dnflow->rule.info = pd->act.dnrpipe; 9582 } 9583 else if (pd->dir == dndir && pd->act.dnpipe) { 9584 dnflow->rule.info = pd->act.dnpipe; 9585 } 9586 else { 9587 return (false); 9588 } 9589 9590 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9591 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9592 dnflow->rule.info |= IPFW_IS_PIPE; 9593 9594 dnflow->f_id.proto = pd->proto; 9595 dnflow->f_id.extra = dnflow->rule.info; 9596 switch (pd->naf) { 9597 case AF_INET: 9598 dnflow->f_id.addr_type = 4; 9599 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9600 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9601 break; 9602 case AF_INET6: 9603 dnflow->flags |= IPFW_ARGS_IP6; 9604 dnflow->f_id.addr_type = 6; 9605 dnflow->f_id.src_ip6 = pd->src->v6; 9606 dnflow->f_id.dst_ip6 = pd->dst->v6; 9607 break; 9608 } 9609 9610 return (true); 9611 } 9612 9613 int 9614 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9615 struct inpcb *inp) 9616 { 9617 struct pfi_kkif *kif; 9618 struct mbuf *m = *m0; 9619 9620 M_ASSERTPKTHDR(m); 9621 MPASS(ifp->if_vnet == curvnet); 9622 NET_EPOCH_ASSERT(); 9623 9624 if (!V_pf_status.running) 9625 return (PF_PASS); 9626 9627 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9628 9629 if (kif == NULL) { 9630 DPFPRINTF(PF_DEBUG_URGENT, 9631 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 9632 return (PF_DROP); 9633 } 9634 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9635 return (PF_PASS); 9636 9637 if (m->m_flags & M_SKIP_FIREWALL) 9638 return (PF_PASS); 9639 9640 if (__predict_false(! M_WRITABLE(*m0))) { 9641 m = *m0 = m_unshare(*m0, M_NOWAIT); 9642 if (*m0 == NULL) 9643 return (PF_DROP); 9644 } 9645 9646 /* Stateless! */ 9647 return (pf_test_eth_rule(dir, kif, m0)); 9648 } 9649 9650 static __inline void 9651 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9652 { 9653 struct m_tag *mtag; 9654 9655 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9656 9657 /* dummynet adds this tag, but pf does not need it, 9658 * and keeping it creates unexpected behavior, 9659 * e.g. in case of divert(4) usage right after dummynet. */ 9660 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9661 if (mtag != NULL) 9662 m_tag_delete(m, mtag); 9663 } 9664 9665 static int 9666 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9667 struct pf_krule *r, struct mbuf **m0) 9668 { 9669 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9670 } 9671 9672 static int 9673 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9674 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 9675 struct mbuf **m0) 9676 { 9677 struct ip_fw_args dnflow; 9678 9679 NET_EPOCH_ASSERT(); 9680 9681 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9682 return (0); 9683 9684 if (ip_dn_io_ptr == NULL) { 9685 m_freem(*m0); 9686 *m0 = NULL; 9687 return (ENOMEM); 9688 } 9689 9690 if (pd->pf_mtag == NULL && 9691 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 9692 m_freem(*m0); 9693 *m0 = NULL; 9694 return (ENOMEM); 9695 } 9696 9697 if (ifp != NULL) { 9698 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 9699 9700 pd->pf_mtag->if_index = ifp->if_index; 9701 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 9702 9703 MPASS(sa != NULL); 9704 9705 switch (pd->naf) { 9706 case AF_INET: 9707 memcpy(&pd->pf_mtag->dst, sa, 9708 sizeof(struct sockaddr_in)); 9709 break; 9710 case AF_INET6: 9711 memcpy(&pd->pf_mtag->dst, sa, 9712 sizeof(struct sockaddr_in6)); 9713 break; 9714 } 9715 } 9716 9717 if (s != NULL && s->nat_rule != NULL && 9718 s->nat_rule->action == PF_RDR && 9719 ( 9720 #ifdef INET 9721 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 9722 #endif 9723 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 9724 /* 9725 * If we're redirecting to loopback mark this packet 9726 * as being local. Otherwise it might get dropped 9727 * if dummynet re-injects. 9728 */ 9729 (*m0)->m_pkthdr.rcvif = V_loif; 9730 } 9731 9732 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 9733 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 9734 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 9735 ip_dn_io_ptr(m0, &dnflow); 9736 if (*m0 != NULL) { 9737 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9738 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 9739 } 9740 } 9741 9742 return (0); 9743 } 9744 9745 #ifdef INET6 9746 static int 9747 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 9748 u_short *reason) 9749 { 9750 struct ip6_opt opt; 9751 struct ip6_opt_jumbo jumbo; 9752 9753 while (off < end) { 9754 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 9755 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 9756 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 9757 return (PF_DROP); 9758 } 9759 if (opt.ip6o_type == IP6OPT_PAD1) { 9760 off++; 9761 continue; 9762 } 9763 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL, 9764 reason, AF_INET6)) { 9765 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 9766 return (PF_DROP); 9767 } 9768 if (off + sizeof(opt) + opt.ip6o_len > end) { 9769 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 9770 REASON_SET(reason, PFRES_IPOPTIONS); 9771 return (PF_DROP); 9772 } 9773 switch (opt.ip6o_type) { 9774 case IP6OPT_JUMBO: 9775 if (pd->jumbolen != 0) { 9776 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 9777 REASON_SET(reason, PFRES_IPOPTIONS); 9778 return (PF_DROP); 9779 } 9780 if (ntohs(h->ip6_plen) != 0) { 9781 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 9782 REASON_SET(reason, PFRES_IPOPTIONS); 9783 return (PF_DROP); 9784 } 9785 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL, 9786 reason, AF_INET6)) { 9787 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 9788 return (PF_DROP); 9789 } 9790 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 9791 sizeof(pd->jumbolen)); 9792 pd->jumbolen = ntohl(pd->jumbolen); 9793 if (pd->jumbolen < IPV6_MAXPACKET) { 9794 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 9795 REASON_SET(reason, PFRES_IPOPTIONS); 9796 return (PF_DROP); 9797 } 9798 break; 9799 default: 9800 break; 9801 } 9802 off += sizeof(opt) + opt.ip6o_len; 9803 } 9804 9805 return (PF_PASS); 9806 } 9807 9808 int 9809 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 9810 { 9811 struct ip6_frag frag; 9812 struct ip6_ext ext; 9813 struct ip6_rthdr rthdr; 9814 uint32_t end; 9815 int rthdr_cnt = 0; 9816 9817 pd->off += sizeof(struct ip6_hdr); 9818 end = pd->off + ntohs(h->ip6_plen); 9819 pd->fragoff = pd->extoff = pd->jumbolen = 0; 9820 pd->proto = h->ip6_nxt; 9821 for (;;) { 9822 switch (pd->proto) { 9823 case IPPROTO_FRAGMENT: 9824 if (pd->fragoff != 0) { 9825 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 9826 REASON_SET(reason, PFRES_FRAG); 9827 return (PF_DROP); 9828 } 9829 /* jumbo payload packets cannot be fragmented */ 9830 if (pd->jumbolen != 0) { 9831 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 9832 REASON_SET(reason, PFRES_FRAG); 9833 return (PF_DROP); 9834 } 9835 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 9836 NULL, reason, AF_INET6)) { 9837 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 9838 return (PF_DROP); 9839 } 9840 pd->fragoff = pd->off; 9841 /* stop walking over non initial fragments */ 9842 if (htons((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) 9843 return (PF_PASS); 9844 pd->off += sizeof(frag); 9845 pd->proto = frag.ip6f_nxt; 9846 break; 9847 case IPPROTO_ROUTING: 9848 if (rthdr_cnt++) { 9849 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 9850 REASON_SET(reason, PFRES_IPOPTIONS); 9851 return (PF_DROP); 9852 } 9853 /* fragments may be short */ 9854 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 9855 pd->off = pd->fragoff; 9856 pd->proto = IPPROTO_FRAGMENT; 9857 return (PF_PASS); 9858 } 9859 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 9860 NULL, reason, AF_INET6)) { 9861 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 9862 return (PF_DROP); 9863 } 9864 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9865 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 9866 REASON_SET(reason, PFRES_IPOPTIONS); 9867 return (PF_DROP); 9868 } 9869 /* FALLTHROUGH */ 9870 case IPPROTO_AH: 9871 case IPPROTO_HOPOPTS: 9872 case IPPROTO_DSTOPTS: 9873 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9874 NULL, reason, AF_INET6)) { 9875 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 9876 return (PF_DROP); 9877 } 9878 /* fragments may be short */ 9879 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 9880 pd->off = pd->fragoff; 9881 pd->proto = IPPROTO_FRAGMENT; 9882 return (PF_PASS); 9883 } 9884 /* reassembly needs the ext header before the frag */ 9885 if (pd->fragoff == 0) 9886 pd->extoff = pd->off; 9887 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0) { 9888 if (pf_walk_option6(pd, h, 9889 pd->off + sizeof(ext), 9890 pd->off + (ext.ip6e_len + 1) * 8, reason) 9891 != PF_PASS) 9892 return (PF_DROP); 9893 if (ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 9894 DPFPRINTF(PF_DEBUG_MISC, 9895 ("IPv6 missing jumbo")); 9896 REASON_SET(reason, PFRES_IPOPTIONS); 9897 return (PF_DROP); 9898 } 9899 } 9900 if (pd->proto == IPPROTO_AH) 9901 pd->off += (ext.ip6e_len + 2) * 4; 9902 else 9903 pd->off += (ext.ip6e_len + 1) * 8; 9904 pd->proto = ext.ip6e_nxt; 9905 break; 9906 case IPPROTO_TCP: 9907 case IPPROTO_UDP: 9908 case IPPROTO_SCTP: 9909 case IPPROTO_ICMPV6: 9910 /* fragments may be short, ignore inner header then */ 9911 if (pd->fragoff != 0 && end < pd->off + 9912 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 9913 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 9914 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 9915 sizeof(struct icmp6_hdr))) { 9916 pd->off = pd->fragoff; 9917 pd->proto = IPPROTO_FRAGMENT; 9918 } 9919 /* FALLTHROUGH */ 9920 default: 9921 return (PF_PASS); 9922 } 9923 } 9924 } 9925 #endif 9926 9927 static void 9928 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 9929 { 9930 memset(pd, 0, sizeof(*pd)); 9931 pd->pf_mtag = pf_find_mtag(m); 9932 pd->m = m; 9933 } 9934 9935 static int 9936 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 9937 u_short *action, u_short *reason, struct pfi_kkif *kif, 9938 struct pf_rule_actions *default_actions) 9939 { 9940 pd->dir = dir; 9941 pd->kif = kif; 9942 pd->m = *m0; 9943 pd->sidx = (dir == PF_IN) ? 0 : 1; 9944 pd->didx = (dir == PF_IN) ? 1 : 0; 9945 pd->af = pd->naf = af; 9946 9947 TAILQ_INIT(&pd->sctp_multihome_jobs); 9948 if (default_actions != NULL) 9949 memcpy(&pd->act, default_actions, sizeof(pd->act)); 9950 9951 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 9952 pd->act.dnpipe = pd->pf_mtag->dnpipe; 9953 pd->act.flags = pd->pf_mtag->dnflags; 9954 } 9955 9956 switch (af) { 9957 #ifdef INET 9958 case AF_INET: { 9959 struct ip *h; 9960 9961 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 9962 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 9963 DPFPRINTF(PF_DEBUG_URGENT, 9964 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 9965 *action = PF_DROP; 9966 REASON_SET(reason, PFRES_SHORT); 9967 return (-1); 9968 } 9969 9970 if (pf_normalize_ip(reason, pd) != PF_PASS) { 9971 /* We do IP header normalization and packet reassembly here */ 9972 *m0 = pd->m; 9973 *action = PF_DROP; 9974 return (-1); 9975 } 9976 *m0 = pd->m; 9977 9978 h = mtod(pd->m, struct ip *); 9979 pd->off = h->ip_hl << 2; 9980 if (pd->off < (int)sizeof(*h)) { 9981 *action = PF_DROP; 9982 REASON_SET(reason, PFRES_SHORT); 9983 return (-1); 9984 } 9985 pd->src = (struct pf_addr *)&h->ip_src; 9986 pd->dst = (struct pf_addr *)&h->ip_dst; 9987 pd->ip_sum = &h->ip_sum; 9988 pd->virtual_proto = pd->proto = h->ip_p; 9989 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 9990 pd->ttl = h->ip_ttl; 9991 pd->tot_len = ntohs(h->ip_len); 9992 pd->act.rtableid = -1; 9993 9994 if (h->ip_hl > 5) /* has options */ 9995 pd->badopts++; 9996 9997 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 9998 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9999 10000 break; 10001 } 10002 #endif 10003 #ifdef INET6 10004 case AF_INET6: { 10005 struct ip6_hdr *h; 10006 10007 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 10008 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 10009 DPFPRINTF(PF_DEBUG_URGENT, 10010 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 10011 ", pullup failed\n")); 10012 *action = PF_DROP; 10013 REASON_SET(reason, PFRES_SHORT); 10014 return (-1); 10015 } 10016 10017 h = mtod(pd->m, struct ip6_hdr *); 10018 pd->off = 0; 10019 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10020 *action = PF_DROP; 10021 return (-1); 10022 } 10023 10024 h = mtod(pd->m, struct ip6_hdr *); 10025 pd->src = (struct pf_addr *)&h->ip6_src; 10026 pd->dst = (struct pf_addr *)&h->ip6_dst; 10027 pd->ip_sum = NULL; 10028 pd->tos = IPV6_DSCP(h); 10029 pd->ttl = h->ip6_hlim; 10030 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 10031 pd->virtual_proto = pd->proto = h->ip6_nxt; 10032 pd->act.rtableid = -1; 10033 10034 if (pd->fragoff != 0) 10035 pd->virtual_proto = PF_VPROTO_FRAGMENT; 10036 10037 /* 10038 * we do not support jumbogram. if we keep going, zero ip6_plen 10039 * will do something bad, so drop the packet for now. 10040 */ 10041 if (htons(h->ip6_plen) == 0) { 10042 *action = PF_DROP; 10043 return (-1); 10044 } 10045 10046 /* We do IP header normalization and packet reassembly here */ 10047 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 10048 PF_PASS) { 10049 *m0 = pd->m; 10050 *action = PF_DROP; 10051 return (-1); 10052 } 10053 *m0 = pd->m; 10054 if (pd->m == NULL) { 10055 /* packet sits in reassembly queue, no error */ 10056 *action = PF_PASS; 10057 return (-1); 10058 } 10059 10060 /* Update pointers into the packet. */ 10061 h = mtod(pd->m, struct ip6_hdr *); 10062 pd->src = (struct pf_addr *)&h->ip6_src; 10063 pd->dst = (struct pf_addr *)&h->ip6_dst; 10064 10065 pd->off = 0; 10066 10067 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 10068 *action = PF_DROP; 10069 return (-1); 10070 } 10071 10072 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 10073 /* 10074 * Reassembly may have changed the next protocol from 10075 * fragment to something else, so update. 10076 */ 10077 pd->virtual_proto = pd->proto; 10078 MPASS(pd->fragoff == 0); 10079 } 10080 10081 if (pd->fragoff != 0) 10082 pd->virtual_proto = PF_VPROTO_FRAGMENT; 10083 10084 break; 10085 } 10086 #endif 10087 default: 10088 panic("pf_setup_pdesc called with illegal af %u", af); 10089 } 10090 10091 switch (pd->virtual_proto) { 10092 case IPPROTO_TCP: { 10093 struct tcphdr *th = &pd->hdr.tcp; 10094 10095 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 10096 reason, af)) { 10097 *action = PF_DROP; 10098 REASON_SET(reason, PFRES_SHORT); 10099 return (-1); 10100 } 10101 pd->hdrlen = sizeof(*th); 10102 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 10103 pd->sport = &th->th_sport; 10104 pd->dport = &th->th_dport; 10105 break; 10106 } 10107 case IPPROTO_UDP: { 10108 struct udphdr *uh = &pd->hdr.udp; 10109 10110 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 10111 reason, af)) { 10112 *action = PF_DROP; 10113 REASON_SET(reason, PFRES_SHORT); 10114 return (-1); 10115 } 10116 pd->hdrlen = sizeof(*uh); 10117 if (uh->uh_dport == 0 || 10118 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 10119 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 10120 *action = PF_DROP; 10121 REASON_SET(reason, PFRES_SHORT); 10122 return (-1); 10123 } 10124 pd->sport = &uh->uh_sport; 10125 pd->dport = &uh->uh_dport; 10126 break; 10127 } 10128 case IPPROTO_SCTP: { 10129 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 10130 action, reason, af)) { 10131 *action = PF_DROP; 10132 REASON_SET(reason, PFRES_SHORT); 10133 return (-1); 10134 } 10135 pd->hdrlen = sizeof(pd->hdr.sctp); 10136 pd->p_len = pd->tot_len - pd->off; 10137 10138 pd->sport = &pd->hdr.sctp.src_port; 10139 pd->dport = &pd->hdr.sctp.dest_port; 10140 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 10141 *action = PF_DROP; 10142 REASON_SET(reason, PFRES_SHORT); 10143 return (-1); 10144 } 10145 if (pf_scan_sctp(pd) != PF_PASS) { 10146 *action = PF_DROP; 10147 REASON_SET(reason, PFRES_SHORT); 10148 return (-1); 10149 } 10150 break; 10151 } 10152 case IPPROTO_ICMP: { 10153 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 10154 action, reason, af)) { 10155 *action = PF_DROP; 10156 REASON_SET(reason, PFRES_SHORT); 10157 return (-1); 10158 } 10159 pd->hdrlen = ICMP_MINLEN; 10160 break; 10161 } 10162 #ifdef INET6 10163 case IPPROTO_ICMPV6: { 10164 size_t icmp_hlen = sizeof(struct icmp6_hdr); 10165 10166 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10167 action, reason, af)) { 10168 *action = PF_DROP; 10169 REASON_SET(reason, PFRES_SHORT); 10170 return (-1); 10171 } 10172 /* ICMP headers we look further into to match state */ 10173 switch (pd->hdr.icmp6.icmp6_type) { 10174 case MLD_LISTENER_QUERY: 10175 case MLD_LISTENER_REPORT: 10176 icmp_hlen = sizeof(struct mld_hdr); 10177 break; 10178 case ND_NEIGHBOR_SOLICIT: 10179 case ND_NEIGHBOR_ADVERT: 10180 icmp_hlen = sizeof(struct nd_neighbor_solicit); 10181 break; 10182 } 10183 if (icmp_hlen > sizeof(struct icmp6_hdr) && 10184 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10185 action, reason, af)) { 10186 *action = PF_DROP; 10187 REASON_SET(reason, PFRES_SHORT); 10188 return (-1); 10189 } 10190 pd->hdrlen = icmp_hlen; 10191 break; 10192 } 10193 #endif 10194 } 10195 return (0); 10196 } 10197 10198 static void 10199 pf_counters_inc(int action, struct pf_pdesc *pd, 10200 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 10201 { 10202 struct pf_krule *tr; 10203 int dir = pd->dir; 10204 int dirndx; 10205 10206 pf_counter_u64_critical_enter(); 10207 pf_counter_u64_add_protected( 10208 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10209 pd->tot_len); 10210 pf_counter_u64_add_protected( 10211 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10212 1); 10213 10214 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 10215 dirndx = (dir == PF_OUT); 10216 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 10217 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 10218 pf_update_timestamp(r); 10219 10220 if (a != NULL) { 10221 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 10222 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 10223 } 10224 if (s != NULL) { 10225 struct pf_krule_item *ri; 10226 10227 if (s->nat_rule != NULL) { 10228 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 10229 1); 10230 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 10231 pd->tot_len); 10232 } 10233 if (s->src_node != NULL) { 10234 counter_u64_add(s->src_node->packets[dirndx], 10235 1); 10236 counter_u64_add(s->src_node->bytes[dirndx], 10237 pd->tot_len); 10238 } 10239 if (s->nat_src_node != NULL) { 10240 counter_u64_add(s->nat_src_node->packets[dirndx], 10241 1); 10242 counter_u64_add(s->nat_src_node->bytes[dirndx], 10243 pd->tot_len); 10244 } 10245 dirndx = (dir == s->direction) ? 0 : 1; 10246 s->packets[dirndx]++; 10247 s->bytes[dirndx] += pd->tot_len; 10248 10249 SLIST_FOREACH(ri, &s->match_rules, entry) { 10250 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 10251 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 10252 } 10253 } 10254 10255 tr = r; 10256 if (s != NULL && s->nat_rule != NULL && 10257 r == &V_pf_default_rule) 10258 tr = s->nat_rule; 10259 10260 if (tr->src.addr.type == PF_ADDR_TABLE) 10261 pfr_update_stats(tr->src.addr.p.tbl, 10262 (s == NULL) ? pd->src : 10263 &s->key[(s->direction == PF_IN)]-> 10264 addr[(s->direction == PF_OUT)], 10265 pd->af, pd->tot_len, dir == PF_OUT, 10266 r->action == PF_PASS, tr->src.neg); 10267 if (tr->dst.addr.type == PF_ADDR_TABLE) 10268 pfr_update_stats(tr->dst.addr.p.tbl, 10269 (s == NULL) ? pd->dst : 10270 &s->key[(s->direction == PF_IN)]-> 10271 addr[(s->direction == PF_IN)], 10272 pd->af, pd->tot_len, dir == PF_OUT, 10273 r->action == PF_PASS, tr->dst.neg); 10274 } 10275 pf_counter_u64_critical_exit(); 10276 } 10277 10278 #if defined(INET) || defined(INET6) 10279 int 10280 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10281 struct inpcb *inp, struct pf_rule_actions *default_actions) 10282 { 10283 struct pfi_kkif *kif; 10284 u_short action, reason = 0; 10285 struct m_tag *mtag; 10286 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10287 struct pf_kstate *s = NULL; 10288 struct pf_kruleset *ruleset = NULL; 10289 struct pf_pdesc pd; 10290 int use_2nd_queue = 0; 10291 uint16_t tag; 10292 10293 PF_RULES_RLOCK_TRACKER; 10294 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10295 M_ASSERTPKTHDR(*m0); 10296 10297 if (!V_pf_status.running) 10298 return (PF_PASS); 10299 10300 PF_RULES_RLOCK(); 10301 10302 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10303 10304 if (__predict_false(kif == NULL)) { 10305 DPFPRINTF(PF_DEBUG_URGENT, 10306 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 10307 PF_RULES_RUNLOCK(); 10308 return (PF_DROP); 10309 } 10310 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10311 PF_RULES_RUNLOCK(); 10312 return (PF_PASS); 10313 } 10314 10315 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10316 PF_RULES_RUNLOCK(); 10317 return (PF_PASS); 10318 } 10319 10320 #ifdef INET6 10321 /* 10322 * If we end up changing IP addresses (e.g. binat) the stack may get 10323 * confused and fail to send the icmp6 packet too big error. Just send 10324 * it here, before we do any NAT. 10325 */ 10326 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 10327 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 10328 PF_RULES_RUNLOCK(); 10329 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 10330 *m0 = NULL; 10331 return (PF_DROP); 10332 } 10333 #endif 10334 10335 if (__predict_false(! M_WRITABLE(*m0))) { 10336 *m0 = m_unshare(*m0, M_NOWAIT); 10337 if (*m0 == NULL) { 10338 PF_RULES_RUNLOCK(); 10339 return (PF_DROP); 10340 } 10341 } 10342 10343 pf_init_pdesc(&pd, *m0); 10344 10345 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10346 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10347 10348 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10349 pd.pf_mtag->if_idxgen); 10350 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10351 PF_RULES_RUNLOCK(); 10352 m_freem(*m0); 10353 *m0 = NULL; 10354 return (PF_PASS); 10355 } 10356 PF_RULES_RUNLOCK(); 10357 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10358 *m0 = NULL; 10359 return (PF_PASS); 10360 } 10361 10362 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10363 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10364 /* Dummynet re-injects packets after they've 10365 * completed their delay. We've already 10366 * processed them, so pass unconditionally. */ 10367 10368 /* But only once. We may see the packet multiple times (e.g. 10369 * PFIL_IN/PFIL_OUT). */ 10370 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 10371 PF_RULES_RUNLOCK(); 10372 10373 return (PF_PASS); 10374 } 10375 10376 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 10377 kif, default_actions) == -1) { 10378 if (action != PF_PASS) 10379 pd.act.log |= PF_LOG_FORCE; 10380 goto done; 10381 } 10382 10383 if (__predict_false(ip_divert_ptr != NULL) && 10384 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 10385 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 10386 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 10387 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 10388 if (pd.pf_mtag == NULL && 10389 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10390 action = PF_DROP; 10391 goto done; 10392 } 10393 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 10394 } 10395 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 10396 pd.m->m_flags |= M_FASTFWD_OURS; 10397 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10398 } 10399 m_tag_delete(pd.m, mtag); 10400 10401 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 10402 if (mtag != NULL) 10403 m_tag_delete(pd.m, mtag); 10404 } 10405 10406 switch (pd.virtual_proto) { 10407 case PF_VPROTO_FRAGMENT: 10408 /* 10409 * handle fragments that aren't reassembled by 10410 * normalization 10411 */ 10412 if (kif == NULL || r == NULL) /* pflog */ 10413 action = PF_DROP; 10414 else 10415 action = pf_test_rule(&r, &s, &pd, &a, 10416 &ruleset, inp); 10417 if (action != PF_PASS) 10418 REASON_SET(&reason, PFRES_FRAG); 10419 break; 10420 10421 case IPPROTO_TCP: { 10422 /* Respond to SYN with a syncookie. */ 10423 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 10424 pd.dir == PF_IN && pf_synflood_check(&pd)) { 10425 pf_syncookie_send(&pd); 10426 action = PF_DROP; 10427 break; 10428 } 10429 10430 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 10431 use_2nd_queue = 1; 10432 action = pf_normalize_tcp(&pd); 10433 if (action == PF_DROP) 10434 goto done; 10435 action = pf_test_state_tcp(&s, &pd, &reason); 10436 if (action == PF_PASS || action == PF_AFRT) { 10437 if (V_pfsync_update_state_ptr != NULL) 10438 V_pfsync_update_state_ptr(s); 10439 r = s->rule; 10440 a = s->anchor; 10441 } else if (s == NULL) { 10442 /* Validate remote SYN|ACK, re-create original SYN if 10443 * valid. */ 10444 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 10445 TH_ACK && pf_syncookie_validate(&pd) && 10446 pd.dir == PF_IN) { 10447 struct mbuf *msyn; 10448 10449 msyn = pf_syncookie_recreate_syn(&pd); 10450 if (msyn == NULL) { 10451 action = PF_DROP; 10452 break; 10453 } 10454 10455 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 10456 &pd.act); 10457 m_freem(msyn); 10458 if (action != PF_PASS) 10459 break; 10460 10461 action = pf_test_state_tcp(&s, &pd, &reason); 10462 if (action != PF_PASS || s == NULL) { 10463 action = PF_DROP; 10464 break; 10465 } 10466 10467 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 10468 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 10469 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 10470 action = pf_synproxy(&pd, &s, &reason); 10471 break; 10472 } else { 10473 action = pf_test_rule(&r, &s, &pd, 10474 &a, &ruleset, inp); 10475 } 10476 } 10477 break; 10478 } 10479 10480 case IPPROTO_UDP: { 10481 action = pf_test_state_udp(&s, &pd); 10482 if (action == PF_PASS || action == PF_AFRT) { 10483 if (V_pfsync_update_state_ptr != NULL) 10484 V_pfsync_update_state_ptr(s); 10485 r = s->rule; 10486 a = s->anchor; 10487 } else if (s == NULL) 10488 action = pf_test_rule(&r, &s, &pd, 10489 &a, &ruleset, inp); 10490 break; 10491 } 10492 10493 case IPPROTO_SCTP: { 10494 action = pf_normalize_sctp(&pd); 10495 if (action == PF_DROP) 10496 goto done; 10497 action = pf_test_state_sctp(&s, &pd, &reason); 10498 if (action == PF_PASS || action == PF_AFRT) { 10499 if (V_pfsync_update_state_ptr != NULL) 10500 V_pfsync_update_state_ptr(s); 10501 r = s->rule; 10502 a = s->anchor; 10503 } else if (s == NULL) { 10504 action = pf_test_rule(&r, &s, 10505 &pd, &a, &ruleset, inp); 10506 } 10507 break; 10508 } 10509 10510 case IPPROTO_ICMP: 10511 case IPPROTO_ICMPV6: { 10512 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 10513 action = PF_DROP; 10514 REASON_SET(&reason, PFRES_NORM); 10515 DPFPRINTF(PF_DEBUG_MISC, 10516 ("dropping IPv6 packet with ICMPv4 payload")); 10517 goto done; 10518 } 10519 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 10520 action = PF_DROP; 10521 REASON_SET(&reason, PFRES_NORM); 10522 DPFPRINTF(PF_DEBUG_MISC, 10523 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 10524 goto done; 10525 } 10526 action = pf_test_state_icmp(&s, &pd, &reason); 10527 if (action == PF_PASS || action == PF_AFRT) { 10528 if (V_pfsync_update_state_ptr != NULL) 10529 V_pfsync_update_state_ptr(s); 10530 r = s->rule; 10531 a = s->anchor; 10532 } else if (s == NULL) 10533 action = pf_test_rule(&r, &s, &pd, 10534 &a, &ruleset, inp); 10535 break; 10536 } 10537 10538 default: 10539 action = pf_test_state_other(&s, &pd); 10540 if (action == PF_PASS || action == PF_AFRT) { 10541 if (V_pfsync_update_state_ptr != NULL) 10542 V_pfsync_update_state_ptr(s); 10543 r = s->rule; 10544 a = s->anchor; 10545 } else if (s == NULL) 10546 action = pf_test_rule(&r, &s, &pd, 10547 &a, &ruleset, inp); 10548 break; 10549 } 10550 10551 done: 10552 PF_RULES_RUNLOCK(); 10553 10554 if (pd.m == NULL) 10555 goto eat_pkt; 10556 10557 if (action == PF_PASS && pd.badopts && 10558 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 10559 action = PF_DROP; 10560 REASON_SET(&reason, PFRES_IPOPTIONS); 10561 pd.act.log = PF_LOG_FORCE; 10562 DPFPRINTF(PF_DEBUG_MISC, 10563 ("pf: dropping packet with dangerous headers\n")); 10564 } 10565 10566 if (s) { 10567 uint8_t log = pd.act.log; 10568 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 10569 pd.act.log |= log; 10570 tag = s->tag; 10571 } else { 10572 tag = r->tag; 10573 } 10574 10575 if (tag > 0 && pf_tag_packet(&pd, tag)) { 10576 action = PF_DROP; 10577 REASON_SET(&reason, PFRES_MEMORY); 10578 } 10579 10580 pf_scrub(&pd); 10581 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 10582 pf_normalize_mss(&pd); 10583 10584 if (pd.act.rtableid >= 0) 10585 M_SETFIB(pd.m, pd.act.rtableid); 10586 10587 if (pd.act.flags & PFSTATE_SETPRIO) { 10588 if (pd.tos & IPTOS_LOWDELAY) 10589 use_2nd_queue = 1; 10590 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 10591 action = PF_DROP; 10592 REASON_SET(&reason, PFRES_MEMORY); 10593 pd.act.log = PF_LOG_FORCE; 10594 DPFPRINTF(PF_DEBUG_MISC, 10595 ("pf: failed to allocate 802.1q mtag\n")); 10596 } 10597 } 10598 10599 #ifdef ALTQ 10600 if (action == PF_PASS && pd.act.qid) { 10601 if (pd.pf_mtag == NULL && 10602 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10603 action = PF_DROP; 10604 REASON_SET(&reason, PFRES_MEMORY); 10605 } else { 10606 if (s != NULL) 10607 pd.pf_mtag->qid_hash = pf_state_hash(s); 10608 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 10609 pd.pf_mtag->qid = pd.act.pqid; 10610 else 10611 pd.pf_mtag->qid = pd.act.qid; 10612 /* Add hints for ecn. */ 10613 pd.pf_mtag->hdr = mtod(pd.m, void *); 10614 } 10615 } 10616 #endif /* ALTQ */ 10617 10618 /* 10619 * connections redirected to loopback should not match sockets 10620 * bound specifically to loopback due to security implications, 10621 * see tcp_input() and in_pcblookup_listen(). 10622 */ 10623 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 10624 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 10625 (s->nat_rule->action == PF_RDR || 10626 s->nat_rule->action == PF_BINAT) && 10627 pf_is_loopback(af, pd.dst)) 10628 pd.m->m_flags |= M_SKIP_FIREWALL; 10629 10630 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 10631 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 10632 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 10633 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 10634 if (mtag != NULL) { 10635 ((struct pf_divert_mtag *)(mtag+1))->port = 10636 ntohs(r->divert.port); 10637 ((struct pf_divert_mtag *)(mtag+1))->idir = 10638 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 10639 PF_DIVERT_MTAG_DIR_OUT; 10640 10641 if (s) 10642 PF_STATE_UNLOCK(s); 10643 10644 m_tag_prepend(pd.m, mtag); 10645 if (pd.m->m_flags & M_FASTFWD_OURS) { 10646 if (pd.pf_mtag == NULL && 10647 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10648 action = PF_DROP; 10649 REASON_SET(&reason, PFRES_MEMORY); 10650 pd.act.log = PF_LOG_FORCE; 10651 DPFPRINTF(PF_DEBUG_MISC, 10652 ("pf: failed to allocate tag\n")); 10653 } else { 10654 pd.pf_mtag->flags |= 10655 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10656 pd.m->m_flags &= ~M_FASTFWD_OURS; 10657 } 10658 } 10659 ip_divert_ptr(*m0, dir == PF_IN); 10660 *m0 = NULL; 10661 10662 return (action); 10663 } else { 10664 /* XXX: ipfw has the same behaviour! */ 10665 action = PF_DROP; 10666 REASON_SET(&reason, PFRES_MEMORY); 10667 pd.act.log = PF_LOG_FORCE; 10668 DPFPRINTF(PF_DEBUG_MISC, 10669 ("pf: failed to allocate divert tag\n")); 10670 } 10671 } 10672 /* XXX: Anybody working on it?! */ 10673 if (af == AF_INET6 && r->divert.port) 10674 printf("pf: divert(9) is not supported for IPv6\n"); 10675 10676 /* this flag will need revising if the pkt is forwarded */ 10677 if (pd.pf_mtag) 10678 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 10679 10680 if (pd.act.log) { 10681 struct pf_krule *lr; 10682 struct pf_krule_item *ri; 10683 10684 if (s != NULL && s->nat_rule != NULL && 10685 s->nat_rule->log & PF_LOG_ALL) 10686 lr = s->nat_rule; 10687 else 10688 lr = r; 10689 10690 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 10691 PFLOG_PACKET(action, reason, lr, a, 10692 ruleset, &pd, (s == NULL)); 10693 if (s) { 10694 SLIST_FOREACH(ri, &s->match_rules, entry) 10695 if (ri->r->log & PF_LOG_ALL) 10696 PFLOG_PACKET(action, 10697 reason, ri->r, a, ruleset, &pd, 0); 10698 } 10699 } 10700 10701 pf_counters_inc(action, &pd, s, r, a); 10702 10703 switch (action) { 10704 case PF_SYNPROXY_DROP: 10705 m_freem(*m0); 10706 case PF_DEFER: 10707 *m0 = NULL; 10708 action = PF_PASS; 10709 break; 10710 case PF_DROP: 10711 m_freem(*m0); 10712 *m0 = NULL; 10713 break; 10714 case PF_AFRT: 10715 if (pf_translate_af(&pd)) { 10716 if (!pd.m) 10717 *m0 = NULL; 10718 action = PF_DROP; 10719 break; 10720 } 10721 *m0 = pd.m; /* pf_translate_af may change pd.m */ 10722 #ifdef INET 10723 if (pd.naf == AF_INET) 10724 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10725 #endif 10726 #ifdef INET6 10727 if (pd.naf == AF_INET6) 10728 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10729 #endif 10730 *m0 = NULL; 10731 action = PF_PASS; 10732 goto out; 10733 break; 10734 default: 10735 if (pd.act.rt) { 10736 switch (af) { 10737 #ifdef INET 10738 case AF_INET: 10739 /* pf_route() returns unlocked. */ 10740 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10741 break; 10742 #endif 10743 #ifdef INET6 10744 case AF_INET6: 10745 /* pf_route6() returns unlocked. */ 10746 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10747 break; 10748 #endif 10749 } 10750 goto out; 10751 } 10752 if (pf_dummynet(&pd, s, r, m0) != 0) { 10753 action = PF_DROP; 10754 REASON_SET(&reason, PFRES_MEMORY); 10755 } 10756 break; 10757 } 10758 10759 eat_pkt: 10760 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 10761 10762 if (s && action != PF_DROP) { 10763 if (!s->if_index_in && dir == PF_IN) 10764 s->if_index_in = ifp->if_index; 10765 else if (!s->if_index_out && dir == PF_OUT) 10766 s->if_index_out = ifp->if_index; 10767 } 10768 10769 if (s) 10770 PF_STATE_UNLOCK(s); 10771 10772 out: 10773 #ifdef INET6 10774 /* If reassembled packet passed, create new fragments. */ 10775 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 10776 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 10777 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 10778 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 10779 #endif 10780 10781 pf_sctp_multihome_delayed(&pd, kif, s, action); 10782 10783 return (action); 10784 } 10785 #endif /* INET || INET6 */ 10786