1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/random.h> 58 #include <sys/refcount.h> 59 #include <sys/sdt.h> 60 #include <sys/socket.h> 61 #include <sys/sysctl.h> 62 #include <sys/taskqueue.h> 63 #include <sys/ucred.h> 64 65 #include <crypto/sha2/sha512.h> 66 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/if_private.h> 70 #include <net/if_types.h> 71 #include <net/if_vlan_var.h> 72 #include <net/route.h> 73 #include <net/route/nhop.h> 74 #include <net/vnet.h> 75 76 #include <net/pfil.h> 77 #include <net/pfvar.h> 78 #include <net/if_pflog.h> 79 #include <net/if_pfsync.h> 80 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_var.h> 83 #include <netinet/in_fib.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_fw.h> 86 #include <netinet/ip_icmp.h> 87 #include <netinet/icmp_var.h> 88 #include <netinet/ip_var.h> 89 #include <netinet/tcp.h> 90 #include <netinet/tcp_fsm.h> 91 #include <netinet/tcp_seq.h> 92 #include <netinet/tcp_timer.h> 93 #include <netinet/tcp_var.h> 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 97 /* dummynet */ 98 #include <netinet/ip_dummynet.h> 99 #include <netinet/ip_fw.h> 100 #include <netpfil/ipfw/dn_heap.h> 101 #include <netpfil/ipfw/ip_fw_private.h> 102 #include <netpfil/ipfw/ip_dn_private.h> 103 104 #ifdef INET6 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet6/in6_fib.h> 111 #include <netinet6/scope6_var.h> 112 #endif /* INET6 */ 113 114 #include <netinet/sctp_header.h> 115 #include <netinet/sctp_crc32.h> 116 117 #include <machine/in_cksum.h> 118 #include <security/mac/mac_framework.h> 119 120 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 121 122 SDT_PROVIDER_DEFINE(pf); 123 SDT_PROBE_DEFINE2(pf, , test, reason_set, "int", "int"); 124 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 132 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 133 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 134 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 135 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 136 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 137 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 138 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 139 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 140 "struct pf_krule *", "struct mbuf *", "int"); 141 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 142 "struct pf_sctp_source *"); 143 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 144 "struct pf_kstate *", "struct pf_sctp_source *"); 145 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int", 146 "int", "struct pf_pdesc *", "int"); 147 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t"); 148 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *", 149 "int"); 150 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *", 151 "int"); 152 153 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 154 "struct mbuf *"); 155 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 156 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 157 "int", "struct pf_keth_rule *", "char *"); 158 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 159 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 160 "int", "struct pf_keth_rule *"); 161 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 162 163 /* 164 * Global variables 165 */ 166 167 /* state tables */ 168 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 169 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 170 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 173 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 174 VNET_DEFINE(struct pf_kstatus, pf_status); 175 176 VNET_DEFINE(u_int32_t, ticket_altqs_active); 177 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 178 VNET_DEFINE(int, altqs_inactive_open); 179 VNET_DEFINE(u_int32_t, ticket_pabuf); 180 181 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 182 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 183 VNET_DEFINE(u_char, pf_tcp_secret[16]); 184 #define V_pf_tcp_secret VNET(pf_tcp_secret) 185 VNET_DEFINE(int, pf_tcp_secret_init); 186 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 187 VNET_DEFINE(int, pf_tcp_iss_off); 188 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 189 VNET_DECLARE(int, pf_vnet_active); 190 #define V_pf_vnet_active VNET(pf_vnet_active) 191 192 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 193 #define V_pf_purge_idx VNET(pf_purge_idx) 194 195 #ifdef PF_WANT_32_TO_64_COUNTER 196 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 197 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 198 199 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 200 VNET_DEFINE(size_t, pf_allrulecount); 201 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 202 #endif 203 204 struct pf_sctp_endpoint; 205 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 206 struct pf_sctp_source { 207 sa_family_t af; 208 struct pf_addr addr; 209 TAILQ_ENTRY(pf_sctp_source) entry; 210 }; 211 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 212 struct pf_sctp_endpoint 213 { 214 uint32_t v_tag; 215 struct pf_sctp_sources sources; 216 RB_ENTRY(pf_sctp_endpoint) entry; 217 }; 218 static int 219 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 220 { 221 return (a->v_tag - b->v_tag); 222 } 223 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 224 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 225 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 226 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 227 static struct mtx_padalign pf_sctp_endpoints_mtx; 228 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 229 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 230 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 231 232 /* 233 * Queue for pf_intr() sends. 234 */ 235 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 236 struct pf_send_entry { 237 STAILQ_ENTRY(pf_send_entry) pfse_next; 238 struct mbuf *pfse_m; 239 enum { 240 PFSE_IP, 241 PFSE_IP6, 242 PFSE_ICMP, 243 PFSE_ICMP6, 244 } pfse_type; 245 struct { 246 int type; 247 int code; 248 int mtu; 249 } icmpopts; 250 }; 251 252 STAILQ_HEAD(pf_send_head, pf_send_entry); 253 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 254 #define V_pf_sendqueue VNET(pf_sendqueue) 255 256 static struct mtx_padalign pf_sendqueue_mtx; 257 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 258 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 259 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 260 261 /* 262 * Queue for pf_overload_task() tasks. 263 */ 264 struct pf_overload_entry { 265 SLIST_ENTRY(pf_overload_entry) next; 266 struct pf_addr addr; 267 sa_family_t af; 268 uint8_t dir; 269 struct pf_krule *rule; 270 }; 271 272 SLIST_HEAD(pf_overload_head, pf_overload_entry); 273 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 274 #define V_pf_overloadqueue VNET(pf_overloadqueue) 275 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 276 #define V_pf_overloadtask VNET(pf_overloadtask) 277 278 static struct mtx_padalign pf_overloadqueue_mtx; 279 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 280 "pf overload/flush queue", MTX_DEF); 281 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 282 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 283 284 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 285 struct mtx_padalign pf_unlnkdrules_mtx; 286 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 287 MTX_DEF); 288 289 struct sx pf_config_lock; 290 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 291 292 struct mtx_padalign pf_table_stats_lock; 293 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 294 MTX_DEF); 295 296 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 297 #define V_pf_sources_z VNET(pf_sources_z) 298 uma_zone_t pf_mtag_z; 299 VNET_DEFINE(uma_zone_t, pf_state_z); 300 VNET_DEFINE(uma_zone_t, pf_state_key_z); 301 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 302 303 VNET_DEFINE(struct unrhdr64, pf_stateid); 304 305 static void pf_src_tree_remove_state(struct pf_kstate *); 306 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 307 u_int32_t); 308 static void pf_add_threshold(struct pf_threshold *); 309 static int pf_check_threshold(struct pf_threshold *); 310 311 static void pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *, 312 u_int16_t *, u_int16_t *, struct pf_addr *, 313 u_int16_t, u_int8_t, sa_family_t, sa_family_t); 314 static int pf_modulate_sack(struct pf_pdesc *, 315 struct tcphdr *, struct pf_state_peer *); 316 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 317 u_int16_t *, u_int16_t *); 318 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 319 struct pf_addr *, struct pf_addr *, u_int16_t, 320 u_int16_t *, u_int16_t *, u_int16_t *, 321 u_int16_t *, u_int8_t, sa_family_t); 322 int pf_change_icmp_af(struct mbuf *, int, 323 struct pf_pdesc *, struct pf_pdesc *, 324 struct pf_addr *, struct pf_addr *, sa_family_t, 325 sa_family_t); 326 int pf_translate_icmp_af(int, void *); 327 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 328 sa_family_t, struct pf_krule *, int); 329 static void pf_detach_state(struct pf_kstate *); 330 static int pf_state_key_attach(struct pf_state_key *, 331 struct pf_state_key *, struct pf_kstate *); 332 static void pf_state_key_detach(struct pf_kstate *, int); 333 static int pf_state_key_ctor(void *, int, void *, int); 334 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 335 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 336 struct pf_mtag *pf_mtag); 337 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 338 struct pf_krule *, struct mbuf **); 339 static int pf_dummynet_route(struct pf_pdesc *, 340 struct pf_kstate *, struct pf_krule *, 341 struct ifnet *, const struct sockaddr *, struct mbuf **); 342 static int pf_test_eth_rule(int, struct pfi_kkif *, 343 struct mbuf **); 344 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 345 struct pf_pdesc *, struct pf_krule **, 346 struct pf_kruleset **, u_short *, struct inpcb *); 347 static int pf_create_state(struct pf_krule *, struct pf_krule *, 348 struct pf_krule *, struct pf_pdesc *, 349 struct pf_state_key *, struct pf_state_key *, int *, 350 struct pf_kstate **, int, u_int16_t, u_int16_t, 351 struct pf_krule_slist *, struct pf_udp_mapping *); 352 static int pf_state_key_addr_setup(struct pf_pdesc *, 353 struct pf_state_key_cmp *, int); 354 static int pf_tcp_track_full(struct pf_kstate *, 355 struct pf_pdesc *, u_short *, int *, 356 struct pf_state_peer *, struct pf_state_peer *, 357 u_int8_t, u_int8_t); 358 static int pf_tcp_track_sloppy(struct pf_kstate *, 359 struct pf_pdesc *, u_short *, 360 struct pf_state_peer *, struct pf_state_peer *, 361 u_int8_t, u_int8_t); 362 static int pf_test_state(struct pf_kstate **, struct pf_pdesc *, 363 u_short *); 364 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 365 struct pf_pdesc *, struct pf_kstate **, 366 u_int16_t, u_int16_t, int, int *, int, int); 367 static int pf_test_state_icmp(struct pf_kstate **, 368 struct pf_pdesc *, u_short *); 369 static int pf_sctp_track(struct pf_kstate *, struct pf_pdesc *, 370 u_short *); 371 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 372 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 373 struct pfi_kkif *, struct pf_kstate *, int); 374 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 375 int, u_int16_t); 376 static int pf_check_proto_cksum(struct mbuf *, int, int, 377 u_int8_t, sa_family_t); 378 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 379 int, int, u_short *); 380 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 381 u_short *); 382 static void pf_print_state_parts(struct pf_kstate *, 383 struct pf_state_key *, struct pf_state_key *); 384 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 385 bool, u_int8_t); 386 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 387 const struct pf_state_key_cmp *, u_int); 388 static bool pf_src_connlimit(struct pf_kstate *); 389 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 390 static void pf_counters_inc(int, struct pf_pdesc *, 391 struct pf_kstate *, struct pf_krule *, 392 struct pf_krule *); 393 static void pf_log_matches(struct pf_pdesc *, struct pf_krule *, 394 struct pf_krule *, struct pf_kruleset *, 395 struct pf_krule_slist *); 396 static void pf_overload_task(void *v, int pending); 397 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 398 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 399 struct pf_addr *, sa_family_t, struct pf_addr *, 400 struct pfi_kkif *, pf_sn_types_t); 401 static u_int pf_purge_expired_states(u_int, int); 402 static void pf_purge_unlinked_rules(void); 403 static int pf_mtag_uminit(void *, int, int); 404 static void pf_mtag_free(struct m_tag *); 405 static void pf_packet_rework_nat(struct pf_pdesc *, int, 406 struct pf_state_key *); 407 #ifdef INET 408 static void pf_route(struct mbuf **, struct pf_krule *, 409 struct ifnet *, struct pf_kstate *, 410 struct pf_pdesc *, struct inpcb *); 411 #endif /* INET */ 412 #ifdef INET6 413 static void pf_change_a6(struct pf_addr *, u_int16_t *, 414 struct pf_addr *, u_int8_t); 415 static void pf_route6(struct mbuf **, struct pf_krule *, 416 struct ifnet *, struct pf_kstate *, 417 struct pf_pdesc *, struct inpcb *); 418 #endif /* INET6 */ 419 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 420 421 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 422 423 extern int pf_end_threads; 424 extern struct proc *pf_purge_proc; 425 426 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 427 428 #define PACKET_UNDO_NAT(_pd, _off, _s) \ 429 do { \ 430 struct pf_state_key *nk; \ 431 if ((pd->dir) == PF_OUT) \ 432 nk = (_s)->key[PF_SK_STACK]; \ 433 else \ 434 nk = (_s)->key[PF_SK_WIRE]; \ 435 pf_packet_rework_nat(_pd, _off, nk); \ 436 } while (0) 437 438 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 439 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 440 441 #define STATE_LOOKUP(k, s, pd) \ 442 do { \ 443 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 444 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 445 if ((s) == NULL) \ 446 return (PF_DROP); \ 447 if (PACKET_LOOPED(pd)) \ 448 return (PF_PASS); \ 449 } while (0) 450 451 static struct pfi_kkif * 452 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 453 { 454 struct pfi_kkif *k = pd->kif; 455 456 SDT_PROBE2(pf, ip, , bound_iface, st, k); 457 458 /* Floating unless otherwise specified. */ 459 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 460 return (V_pfi_all); 461 462 /* 463 * Initially set to all, because we don't know what interface we'll be 464 * sending this out when we create the state. 465 */ 466 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN)) 467 return (V_pfi_all); 468 469 /* 470 * If this state is created based on another state (e.g. SCTP 471 * multihome) always set it floating initially. We can't know for sure 472 * what interface the actual traffic for this state will come in on. 473 */ 474 if (pd->related_rule) 475 return (V_pfi_all); 476 477 /* Don't overrule the interface for states created on incoming packets. */ 478 if (st->direction == PF_IN) 479 return (k); 480 481 /* No route-to, so don't overrule. */ 482 if (st->act.rt != PF_ROUTETO) 483 return (k); 484 485 /* Bind to the route-to interface. */ 486 return (st->act.rt_kif); 487 } 488 489 #define STATE_INC_COUNTERS(s) \ 490 do { \ 491 struct pf_krule_item *mrm; \ 492 counter_u64_add(s->rule->states_cur, 1); \ 493 counter_u64_add(s->rule->states_tot, 1); \ 494 if (s->anchor != NULL) { \ 495 counter_u64_add(s->anchor->states_cur, 1); \ 496 counter_u64_add(s->anchor->states_tot, 1); \ 497 } \ 498 if (s->nat_rule != NULL) { \ 499 counter_u64_add(s->nat_rule->states_cur, 1);\ 500 counter_u64_add(s->nat_rule->states_tot, 1);\ 501 } \ 502 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 503 counter_u64_add(mrm->r->states_cur, 1); \ 504 counter_u64_add(mrm->r->states_tot, 1); \ 505 } \ 506 } while (0) 507 508 #define STATE_DEC_COUNTERS(s) \ 509 do { \ 510 struct pf_krule_item *mrm; \ 511 if (s->nat_rule != NULL) \ 512 counter_u64_add(s->nat_rule->states_cur, -1);\ 513 if (s->anchor != NULL) \ 514 counter_u64_add(s->anchor->states_cur, -1); \ 515 counter_u64_add(s->rule->states_cur, -1); \ 516 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 517 counter_u64_add(mrm->r->states_cur, -1); \ 518 } while (0) 519 520 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 521 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 522 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 523 VNET_DEFINE(struct pf_idhash *, pf_idhash); 524 VNET_DEFINE(struct pf_srchash *, pf_srchash); 525 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 526 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 527 528 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 529 "pf(4)"); 530 531 VNET_DEFINE(u_long, pf_hashmask); 532 VNET_DEFINE(u_long, pf_srchashmask); 533 VNET_DEFINE(u_long, pf_udpendpointhashmask); 534 VNET_DEFINE_STATIC(u_long, pf_hashsize); 535 #define V_pf_hashsize VNET(pf_hashsize) 536 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 537 #define V_pf_srchashsize VNET(pf_srchashsize) 538 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 539 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 540 u_long pf_ioctl_maxcount = 65535; 541 542 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 543 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 544 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 545 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 546 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 547 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 548 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 549 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 550 551 VNET_DEFINE(void *, pf_swi_cookie); 552 VNET_DEFINE(struct intr_event *, pf_swi_ie); 553 554 VNET_DEFINE(uint32_t, pf_hashseed); 555 #define V_pf_hashseed VNET(pf_hashseed) 556 557 static void 558 pf_sctp_checksum(struct mbuf *m, int off) 559 { 560 uint32_t sum = 0; 561 562 /* Zero out the checksum, to enable recalculation. */ 563 m_copyback(m, off + offsetof(struct sctphdr, checksum), 564 sizeof(sum), (caddr_t)&sum); 565 566 sum = sctp_calculate_cksum(m, off); 567 568 m_copyback(m, off + offsetof(struct sctphdr, checksum), 569 sizeof(sum), (caddr_t)&sum); 570 } 571 572 int 573 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 574 { 575 576 switch (af) { 577 #ifdef INET 578 case AF_INET: 579 if (a->addr32[0] > b->addr32[0]) 580 return (1); 581 if (a->addr32[0] < b->addr32[0]) 582 return (-1); 583 break; 584 #endif /* INET */ 585 #ifdef INET6 586 case AF_INET6: 587 if (a->addr32[3] > b->addr32[3]) 588 return (1); 589 if (a->addr32[3] < b->addr32[3]) 590 return (-1); 591 if (a->addr32[2] > b->addr32[2]) 592 return (1); 593 if (a->addr32[2] < b->addr32[2]) 594 return (-1); 595 if (a->addr32[1] > b->addr32[1]) 596 return (1); 597 if (a->addr32[1] < b->addr32[1]) 598 return (-1); 599 if (a->addr32[0] > b->addr32[0]) 600 return (1); 601 if (a->addr32[0] < b->addr32[0]) 602 return (-1); 603 break; 604 #endif /* INET6 */ 605 default: 606 unhandled_af(af); 607 } 608 return (0); 609 } 610 611 static bool 612 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 613 { 614 switch (af) { 615 #ifdef INET 616 case AF_INET: 617 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 618 #endif /* INET */ 619 case AF_INET6: 620 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 621 default: 622 unhandled_af(af); 623 } 624 } 625 626 static void 627 pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk) 628 { 629 630 switch (pd->proto) { 631 case IPPROTO_TCP: { 632 struct tcphdr *th = &pd->hdr.tcp; 633 634 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 635 pf_change_ap(pd, pd->src, &th->th_sport, pd->ip_sum, 636 &th->th_sum, &nk->addr[pd->sidx], 637 nk->port[pd->sidx], 0, pd->af, pd->naf); 638 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 639 pf_change_ap(pd, pd->dst, &th->th_dport, pd->ip_sum, 640 &th->th_sum, &nk->addr[pd->didx], 641 nk->port[pd->didx], 0, pd->af, pd->naf); 642 m_copyback(pd->m, off, sizeof(*th), (caddr_t)th); 643 break; 644 } 645 case IPPROTO_UDP: { 646 struct udphdr *uh = &pd->hdr.udp; 647 648 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 649 pf_change_ap(pd, pd->src, &uh->uh_sport, pd->ip_sum, 650 &uh->uh_sum, &nk->addr[pd->sidx], 651 nk->port[pd->sidx], 1, pd->af, pd->naf); 652 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 653 pf_change_ap(pd, pd->dst, &uh->uh_dport, pd->ip_sum, 654 &uh->uh_sum, &nk->addr[pd->didx], 655 nk->port[pd->didx], 1, pd->af, pd->naf); 656 m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh); 657 break; 658 } 659 case IPPROTO_SCTP: { 660 struct sctphdr *sh = &pd->hdr.sctp; 661 uint16_t checksum = 0; 662 663 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 664 pf_change_ap(pd, pd->src, &sh->src_port, pd->ip_sum, 665 &checksum, &nk->addr[pd->sidx], 666 nk->port[pd->sidx], 1, pd->af, pd->naf); 667 } 668 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 669 pf_change_ap(pd, pd->dst, &sh->dest_port, pd->ip_sum, 670 &checksum, &nk->addr[pd->didx], 671 nk->port[pd->didx], 1, pd->af, pd->naf); 672 } 673 674 break; 675 } 676 case IPPROTO_ICMP: { 677 struct icmp *ih = &pd->hdr.icmp; 678 679 if (nk->port[pd->sidx] != ih->icmp_id) { 680 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 681 ih->icmp_cksum, ih->icmp_id, 682 nk->port[pd->sidx], 0); 683 ih->icmp_id = nk->port[pd->sidx]; 684 pd->sport = &ih->icmp_id; 685 686 m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih); 687 } 688 /* FALLTHROUGH */ 689 } 690 default: 691 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 692 switch (pd->af) { 693 case AF_INET: 694 pf_change_a(&pd->src->v4.s_addr, 695 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 696 0); 697 break; 698 case AF_INET6: 699 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 700 break; 701 default: 702 unhandled_af(pd->af); 703 } 704 } 705 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 706 switch (pd->af) { 707 case AF_INET: 708 pf_change_a(&pd->dst->v4.s_addr, 709 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 710 0); 711 break; 712 case AF_INET6: 713 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 714 break; 715 default: 716 unhandled_af(pd->af); 717 } 718 } 719 break; 720 } 721 } 722 723 static __inline uint32_t 724 pf_hashkey(const struct pf_state_key *sk) 725 { 726 uint32_t h; 727 728 h = murmur3_32_hash32((const uint32_t *)sk, 729 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 730 V_pf_hashseed); 731 732 return (h & V_pf_hashmask); 733 } 734 735 __inline uint32_t 736 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 737 { 738 uint32_t h; 739 740 switch (af) { 741 case AF_INET: 742 h = murmur3_32_hash32((uint32_t *)&addr->v4, 743 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 744 break; 745 case AF_INET6: 746 h = murmur3_32_hash32((uint32_t *)&addr->v6, 747 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 748 break; 749 default: 750 unhandled_af(af); 751 } 752 753 return (h & V_pf_srchashmask); 754 } 755 756 static inline uint32_t 757 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 758 { 759 uint32_t h; 760 761 h = murmur3_32_hash32((uint32_t *)endpoint, 762 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 763 V_pf_hashseed); 764 return (h & V_pf_udpendpointhashmask); 765 } 766 767 #ifdef ALTQ 768 static int 769 pf_state_hash(struct pf_kstate *s) 770 { 771 u_int32_t hv = (intptr_t)s / sizeof(*s); 772 773 hv ^= crc32(&s->src, sizeof(s->src)); 774 hv ^= crc32(&s->dst, sizeof(s->dst)); 775 if (hv == 0) 776 hv = 1; 777 return (hv); 778 } 779 #endif /* ALTQ */ 780 781 static __inline void 782 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 783 { 784 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 785 s->dst.state = newstate; 786 if (which == PF_PEER_DST) 787 return; 788 if (s->src.state == newstate) 789 return; 790 if (s->creatorid == V_pf_status.hostid && 791 s->key[PF_SK_STACK] != NULL && 792 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 793 !(TCPS_HAVEESTABLISHED(s->src.state) || 794 s->src.state == TCPS_CLOSED) && 795 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 796 atomic_add_32(&V_pf_status.states_halfopen, -1); 797 798 s->src.state = newstate; 799 } 800 801 static void 802 pf_init_threshold(struct pf_threshold *threshold, 803 u_int32_t limit, u_int32_t seconds) 804 { 805 threshold->limit = limit * PF_THRESHOLD_MULT; 806 threshold->seconds = seconds; 807 threshold->count = 0; 808 threshold->last = time_uptime; 809 } 810 811 static void 812 pf_add_threshold(struct pf_threshold *threshold) 813 { 814 u_int32_t t = time_uptime, diff = t - threshold->last; 815 816 if (diff >= threshold->seconds) 817 threshold->count = 0; 818 else 819 threshold->count -= threshold->count * diff / 820 threshold->seconds; 821 threshold->count += PF_THRESHOLD_MULT; 822 threshold->last = t; 823 } 824 825 static int 826 pf_check_threshold(struct pf_threshold *threshold) 827 { 828 return (threshold->count > threshold->limit); 829 } 830 831 static bool 832 pf_src_connlimit(struct pf_kstate *state) 833 { 834 struct pf_overload_entry *pfoe; 835 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 836 bool limited = false; 837 838 PF_STATE_LOCK_ASSERT(state); 839 PF_SRC_NODE_LOCK(src_node); 840 841 src_node->conn++; 842 state->src.tcp_est = 1; 843 pf_add_threshold(&src_node->conn_rate); 844 845 if (state->rule->max_src_conn && 846 state->rule->max_src_conn < 847 src_node->conn) { 848 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 849 limited = true; 850 } 851 852 if (state->rule->max_src_conn_rate.limit && 853 pf_check_threshold(&src_node->conn_rate)) { 854 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 855 limited = true; 856 } 857 858 if (!limited) 859 goto done; 860 861 /* Kill this state. */ 862 state->timeout = PFTM_PURGE; 863 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 864 865 if (state->rule->overload_tbl == NULL) 866 goto done; 867 868 /* Schedule overloading and flushing task. */ 869 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 870 if (pfoe == NULL) 871 goto done; /* too bad :( */ 872 873 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 874 pfoe->af = state->key[PF_SK_WIRE]->af; 875 pfoe->rule = state->rule; 876 pfoe->dir = state->direction; 877 PF_OVERLOADQ_LOCK(); 878 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 879 PF_OVERLOADQ_UNLOCK(); 880 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 881 882 done: 883 PF_SRC_NODE_UNLOCK(src_node); 884 return (limited); 885 } 886 887 static void 888 pf_overload_task(void *v, int pending) 889 { 890 struct pf_overload_head queue; 891 struct pfr_addr p; 892 struct pf_overload_entry *pfoe, *pfoe1; 893 uint32_t killed = 0; 894 895 CURVNET_SET((struct vnet *)v); 896 897 PF_OVERLOADQ_LOCK(); 898 queue = V_pf_overloadqueue; 899 SLIST_INIT(&V_pf_overloadqueue); 900 PF_OVERLOADQ_UNLOCK(); 901 902 bzero(&p, sizeof(p)); 903 SLIST_FOREACH(pfoe, &queue, next) { 904 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 905 if (V_pf_status.debug >= PF_DEBUG_MISC) { 906 printf("%s: blocking address ", __func__); 907 pf_print_host(&pfoe->addr, 0, pfoe->af); 908 printf("\n"); 909 } 910 911 p.pfra_af = pfoe->af; 912 switch (pfoe->af) { 913 #ifdef INET 914 case AF_INET: 915 p.pfra_net = 32; 916 p.pfra_ip4addr = pfoe->addr.v4; 917 break; 918 #endif /* INET */ 919 #ifdef INET6 920 case AF_INET6: 921 p.pfra_net = 128; 922 p.pfra_ip6addr = pfoe->addr.v6; 923 break; 924 #endif /* INET6 */ 925 default: 926 unhandled_af(pfoe->af); 927 } 928 929 PF_RULES_WLOCK(); 930 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 931 PF_RULES_WUNLOCK(); 932 } 933 934 /* 935 * Remove those entries, that don't need flushing. 936 */ 937 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 938 if (pfoe->rule->flush == 0) { 939 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 940 free(pfoe, M_PFTEMP); 941 } else 942 counter_u64_add( 943 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 944 945 /* If nothing to flush, return. */ 946 if (SLIST_EMPTY(&queue)) { 947 CURVNET_RESTORE(); 948 return; 949 } 950 951 for (int i = 0; i <= V_pf_hashmask; i++) { 952 struct pf_idhash *ih = &V_pf_idhash[i]; 953 struct pf_state_key *sk; 954 struct pf_kstate *s; 955 956 PF_HASHROW_LOCK(ih); 957 LIST_FOREACH(s, &ih->states, entry) { 958 sk = s->key[PF_SK_WIRE]; 959 SLIST_FOREACH(pfoe, &queue, next) 960 if (sk->af == pfoe->af && 961 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 962 pfoe->rule == s->rule) && 963 ((pfoe->dir == PF_OUT && 964 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 965 (pfoe->dir == PF_IN && 966 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 967 s->timeout = PFTM_PURGE; 968 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 969 killed++; 970 } 971 } 972 PF_HASHROW_UNLOCK(ih); 973 } 974 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 975 free(pfoe, M_PFTEMP); 976 if (V_pf_status.debug >= PF_DEBUG_MISC) 977 printf("%s: %u states killed", __func__, killed); 978 979 CURVNET_RESTORE(); 980 } 981 982 /* 983 * On node found always returns locked. On not found its configurable. 984 */ 985 struct pf_ksrc_node * 986 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 987 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 988 { 989 struct pf_ksrc_node *n; 990 991 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 992 993 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 994 PF_HASHROW_LOCK(*sh); 995 LIST_FOREACH(n, &(*sh)->nodes, entry) 996 if (n->rule == rule && n->af == af && n->type == sn_type && 997 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 998 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 999 break; 1000 1001 if (n == NULL && !returnlocked) 1002 PF_HASHROW_UNLOCK(*sh); 1003 1004 return (n); 1005 } 1006 1007 bool 1008 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 1009 { 1010 struct pf_ksrc_node *cur; 1011 1012 if ((*sn) == NULL) 1013 return (false); 1014 1015 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 1016 1017 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1018 PF_HASHROW_LOCK(sh); 1019 LIST_FOREACH(cur, &(sh->nodes), entry) { 1020 if (cur == (*sn) && 1021 cur->expire != 1) /* Ignore nodes being killed */ 1022 return (true); 1023 } 1024 PF_HASHROW_UNLOCK(sh); 1025 (*sn) = NULL; 1026 return (false); 1027 } 1028 1029 static void 1030 pf_free_src_node(struct pf_ksrc_node *sn) 1031 { 1032 1033 for (int i = 0; i < 2; i++) { 1034 counter_u64_free(sn->bytes[i]); 1035 counter_u64_free(sn->packets[i]); 1036 } 1037 uma_zfree(V_pf_sources_z, sn); 1038 } 1039 1040 static u_short 1041 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1042 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1043 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1044 struct pfi_kkif *rkif, pf_sn_types_t sn_type) 1045 { 1046 u_short reason = 0; 1047 struct pf_krule *r_track = rule; 1048 struct pf_ksrc_node **sn = &(sns[sn_type]); 1049 struct pf_srchash **sh = &(snhs[sn_type]); 1050 1051 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1052 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1053 1054 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1055 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1056 1057 /* 1058 * XXX: There could be a KASSERT for 1059 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1060 * but we'd need to pass pool *only* for this KASSERT. 1061 */ 1062 1063 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1064 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1065 r_track = &V_pf_default_rule; 1066 1067 /* 1068 * Request the sh to always be locked, as we might insert a new sn. 1069 */ 1070 if (*sn == NULL) 1071 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1072 1073 if (*sn == NULL) { 1074 PF_HASHROW_ASSERT(*sh); 1075 1076 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1077 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1078 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1079 reason = PFRES_SRCLIMIT; 1080 goto done; 1081 } 1082 1083 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1084 if ((*sn) == NULL) { 1085 reason = PFRES_MEMORY; 1086 goto done; 1087 } 1088 1089 for (int i = 0; i < 2; i++) { 1090 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1091 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1092 1093 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1094 pf_free_src_node(*sn); 1095 reason = PFRES_MEMORY; 1096 goto done; 1097 } 1098 } 1099 1100 if (sn_type == PF_SN_LIMIT) 1101 pf_init_threshold(&(*sn)->conn_rate, 1102 rule->max_src_conn_rate.limit, 1103 rule->max_src_conn_rate.seconds); 1104 1105 MPASS((*sn)->lock == NULL); 1106 (*sn)->lock = &(*sh)->lock; 1107 1108 (*sn)->af = af; 1109 (*sn)->rule = r_track; 1110 PF_ACPY(&(*sn)->addr, src, af); 1111 if (raddr != NULL) 1112 PF_ACPY(&(*sn)->raddr, raddr, af); 1113 (*sn)->rkif = rkif; 1114 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1115 (*sn)->creation = time_uptime; 1116 (*sn)->ruletype = rule->action; 1117 (*sn)->type = sn_type; 1118 counter_u64_add(r_track->src_nodes[sn_type], 1); 1119 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1120 } else { 1121 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1122 (*sn)->states >= rule->max_src_states) { 1123 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1124 1); 1125 reason = PFRES_SRCLIMIT; 1126 goto done; 1127 } 1128 } 1129 done: 1130 if (reason == 0) 1131 (*sn)->states++; 1132 else 1133 (*sn) = NULL; 1134 1135 PF_HASHROW_UNLOCK(*sh); 1136 return (reason); 1137 } 1138 1139 void 1140 pf_unlink_src_node(struct pf_ksrc_node *src) 1141 { 1142 PF_SRC_NODE_LOCK_ASSERT(src); 1143 1144 LIST_REMOVE(src, entry); 1145 if (src->rule) 1146 counter_u64_add(src->rule->src_nodes[src->type], -1); 1147 } 1148 1149 u_int 1150 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1151 { 1152 struct pf_ksrc_node *sn, *tmp; 1153 u_int count = 0; 1154 1155 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1156 pf_free_src_node(sn); 1157 count++; 1158 } 1159 1160 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1161 1162 return (count); 1163 } 1164 1165 void 1166 pf_mtag_initialize(void) 1167 { 1168 1169 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1170 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1171 UMA_ALIGN_PTR, 0); 1172 } 1173 1174 /* Per-vnet data storage structures initialization. */ 1175 void 1176 pf_initialize(void) 1177 { 1178 struct pf_keyhash *kh; 1179 struct pf_idhash *ih; 1180 struct pf_srchash *sh; 1181 struct pf_udpendpointhash *uh; 1182 u_int i; 1183 1184 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1185 V_pf_hashsize = PF_HASHSIZ; 1186 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1187 V_pf_srchashsize = PF_SRCHASHSIZ; 1188 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1189 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1190 1191 V_pf_hashseed = arc4random(); 1192 1193 /* States and state keys storage. */ 1194 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1195 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1196 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1197 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1198 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1199 1200 V_pf_state_key_z = uma_zcreate("pf state keys", 1201 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1202 UMA_ALIGN_PTR, 0); 1203 1204 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1205 M_PFHASH, M_NOWAIT | M_ZERO); 1206 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1207 M_PFHASH, M_NOWAIT | M_ZERO); 1208 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1209 printf("pf: Unable to allocate memory for " 1210 "state_hashsize %lu.\n", V_pf_hashsize); 1211 1212 free(V_pf_keyhash, M_PFHASH); 1213 free(V_pf_idhash, M_PFHASH); 1214 1215 V_pf_hashsize = PF_HASHSIZ; 1216 V_pf_keyhash = mallocarray(V_pf_hashsize, 1217 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1218 V_pf_idhash = mallocarray(V_pf_hashsize, 1219 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1220 } 1221 1222 V_pf_hashmask = V_pf_hashsize - 1; 1223 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1224 i++, kh++, ih++) { 1225 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1226 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1227 } 1228 1229 /* Source nodes. */ 1230 V_pf_sources_z = uma_zcreate("pf source nodes", 1231 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1232 0); 1233 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1234 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1235 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1236 1237 V_pf_srchash = mallocarray(V_pf_srchashsize, 1238 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1239 if (V_pf_srchash == NULL) { 1240 printf("pf: Unable to allocate memory for " 1241 "source_hashsize %lu.\n", V_pf_srchashsize); 1242 1243 V_pf_srchashsize = PF_SRCHASHSIZ; 1244 V_pf_srchash = mallocarray(V_pf_srchashsize, 1245 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1246 } 1247 1248 V_pf_srchashmask = V_pf_srchashsize - 1; 1249 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1250 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1251 1252 1253 /* UDP endpoint mappings. */ 1254 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1255 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1256 UMA_ALIGN_PTR, 0); 1257 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1258 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1259 if (V_pf_udpendpointhash == NULL) { 1260 printf("pf: Unable to allocate memory for " 1261 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1262 1263 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1264 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1265 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1266 } 1267 1268 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1269 for (i = 0, uh = V_pf_udpendpointhash; 1270 i <= V_pf_udpendpointhashmask; 1271 i++, uh++) { 1272 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1273 MTX_DEF | MTX_DUPOK); 1274 } 1275 1276 /* ALTQ */ 1277 TAILQ_INIT(&V_pf_altqs[0]); 1278 TAILQ_INIT(&V_pf_altqs[1]); 1279 TAILQ_INIT(&V_pf_altqs[2]); 1280 TAILQ_INIT(&V_pf_altqs[3]); 1281 TAILQ_INIT(&V_pf_pabuf[0]); 1282 TAILQ_INIT(&V_pf_pabuf[1]); 1283 TAILQ_INIT(&V_pf_pabuf[2]); 1284 V_pf_altqs_active = &V_pf_altqs[0]; 1285 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1286 V_pf_altqs_inactive = &V_pf_altqs[2]; 1287 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1288 1289 /* Send & overload+flush queues. */ 1290 STAILQ_INIT(&V_pf_sendqueue); 1291 SLIST_INIT(&V_pf_overloadqueue); 1292 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1293 1294 /* Unlinked, but may be referenced rules. */ 1295 TAILQ_INIT(&V_pf_unlinked_rules); 1296 } 1297 1298 void 1299 pf_mtag_cleanup(void) 1300 { 1301 1302 uma_zdestroy(pf_mtag_z); 1303 } 1304 1305 void 1306 pf_cleanup(void) 1307 { 1308 struct pf_keyhash *kh; 1309 struct pf_idhash *ih; 1310 struct pf_srchash *sh; 1311 struct pf_udpendpointhash *uh; 1312 struct pf_send_entry *pfse, *next; 1313 u_int i; 1314 1315 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1316 i <= V_pf_hashmask; 1317 i++, kh++, ih++) { 1318 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1319 __func__)); 1320 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1321 __func__)); 1322 mtx_destroy(&kh->lock); 1323 mtx_destroy(&ih->lock); 1324 } 1325 free(V_pf_keyhash, M_PFHASH); 1326 free(V_pf_idhash, M_PFHASH); 1327 1328 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1329 KASSERT(LIST_EMPTY(&sh->nodes), 1330 ("%s: source node hash not empty", __func__)); 1331 mtx_destroy(&sh->lock); 1332 } 1333 free(V_pf_srchash, M_PFHASH); 1334 1335 for (i = 0, uh = V_pf_udpendpointhash; 1336 i <= V_pf_udpendpointhashmask; 1337 i++, uh++) { 1338 KASSERT(LIST_EMPTY(&uh->endpoints), 1339 ("%s: udp endpoint hash not empty", __func__)); 1340 mtx_destroy(&uh->lock); 1341 } 1342 free(V_pf_udpendpointhash, M_PFHASH); 1343 1344 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1345 m_freem(pfse->pfse_m); 1346 free(pfse, M_PFTEMP); 1347 } 1348 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1349 1350 uma_zdestroy(V_pf_sources_z); 1351 uma_zdestroy(V_pf_state_z); 1352 uma_zdestroy(V_pf_state_key_z); 1353 uma_zdestroy(V_pf_udp_mapping_z); 1354 } 1355 1356 static int 1357 pf_mtag_uminit(void *mem, int size, int how) 1358 { 1359 struct m_tag *t; 1360 1361 t = (struct m_tag *)mem; 1362 t->m_tag_cookie = MTAG_ABI_COMPAT; 1363 t->m_tag_id = PACKET_TAG_PF; 1364 t->m_tag_len = sizeof(struct pf_mtag); 1365 t->m_tag_free = pf_mtag_free; 1366 1367 return (0); 1368 } 1369 1370 static void 1371 pf_mtag_free(struct m_tag *t) 1372 { 1373 1374 uma_zfree(pf_mtag_z, t); 1375 } 1376 1377 struct pf_mtag * 1378 pf_get_mtag(struct mbuf *m) 1379 { 1380 struct m_tag *mtag; 1381 1382 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1383 return ((struct pf_mtag *)(mtag + 1)); 1384 1385 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1386 if (mtag == NULL) 1387 return (NULL); 1388 bzero(mtag + 1, sizeof(struct pf_mtag)); 1389 m_tag_prepend(m, mtag); 1390 1391 return ((struct pf_mtag *)(mtag + 1)); 1392 } 1393 1394 static int 1395 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1396 struct pf_kstate *s) 1397 { 1398 struct pf_keyhash *khs, *khw, *kh; 1399 struct pf_state_key *sk, *cur; 1400 struct pf_kstate *si, *olds = NULL; 1401 int idx; 1402 1403 NET_EPOCH_ASSERT(); 1404 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1405 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1406 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1407 1408 /* 1409 * We need to lock hash slots of both keys. To avoid deadlock 1410 * we always lock the slot with lower address first. Unlock order 1411 * isn't important. 1412 * 1413 * We also need to lock ID hash slot before dropping key 1414 * locks. On success we return with ID hash slot locked. 1415 */ 1416 1417 if (skw == sks) { 1418 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1419 PF_HASHROW_LOCK(khs); 1420 } else { 1421 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1422 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1423 if (khs == khw) { 1424 PF_HASHROW_LOCK(khs); 1425 } else if (khs < khw) { 1426 PF_HASHROW_LOCK(khs); 1427 PF_HASHROW_LOCK(khw); 1428 } else { 1429 PF_HASHROW_LOCK(khw); 1430 PF_HASHROW_LOCK(khs); 1431 } 1432 } 1433 1434 #define KEYS_UNLOCK() do { \ 1435 if (khs != khw) { \ 1436 PF_HASHROW_UNLOCK(khs); \ 1437 PF_HASHROW_UNLOCK(khw); \ 1438 } else \ 1439 PF_HASHROW_UNLOCK(khs); \ 1440 } while (0) 1441 1442 /* 1443 * First run: start with wire key. 1444 */ 1445 sk = skw; 1446 kh = khw; 1447 idx = PF_SK_WIRE; 1448 1449 MPASS(s->lock == NULL); 1450 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1451 1452 keyattach: 1453 LIST_FOREACH(cur, &kh->keys, entry) 1454 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1455 break; 1456 1457 if (cur != NULL) { 1458 /* Key exists. Check for same kif, if none, add to key. */ 1459 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1460 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1461 1462 PF_HASHROW_LOCK(ih); 1463 if (si->kif == s->kif && 1464 ((si->key[PF_SK_WIRE]->af == sk->af && 1465 si->direction == s->direction) || 1466 (si->key[PF_SK_WIRE]->af != 1467 si->key[PF_SK_STACK]->af && 1468 sk->af == si->key[PF_SK_STACK]->af && 1469 si->direction != s->direction))) { 1470 if (sk->proto == IPPROTO_TCP && 1471 si->src.state >= TCPS_FIN_WAIT_2 && 1472 si->dst.state >= TCPS_FIN_WAIT_2) { 1473 /* 1474 * New state matches an old >FIN_WAIT_2 1475 * state. We can't drop key hash locks, 1476 * thus we can't unlink it properly. 1477 * 1478 * As a workaround we drop it into 1479 * TCPS_CLOSED state, schedule purge 1480 * ASAP and push it into the very end 1481 * of the slot TAILQ, so that it won't 1482 * conflict with our new state. 1483 */ 1484 pf_set_protostate(si, PF_PEER_BOTH, 1485 TCPS_CLOSED); 1486 si->timeout = PFTM_PURGE; 1487 olds = si; 1488 } else { 1489 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1490 printf("pf: %s key attach " 1491 "failed on %s: ", 1492 (idx == PF_SK_WIRE) ? 1493 "wire" : "stack", 1494 s->kif->pfik_name); 1495 pf_print_state_parts(s, 1496 (idx == PF_SK_WIRE) ? 1497 sk : NULL, 1498 (idx == PF_SK_STACK) ? 1499 sk : NULL); 1500 printf(", existing: "); 1501 pf_print_state_parts(si, 1502 (idx == PF_SK_WIRE) ? 1503 sk : NULL, 1504 (idx == PF_SK_STACK) ? 1505 sk : NULL); 1506 printf("\n"); 1507 } 1508 s->timeout = PFTM_UNLINKED; 1509 if (idx == PF_SK_STACK) 1510 /* 1511 * Remove the wire key from 1512 * the hash. Other threads 1513 * can't be referencing it 1514 * because we still hold the 1515 * hash lock. 1516 */ 1517 pf_state_key_detach(s, 1518 PF_SK_WIRE); 1519 PF_HASHROW_UNLOCK(ih); 1520 KEYS_UNLOCK(); 1521 if (idx == PF_SK_WIRE) 1522 /* 1523 * We've not inserted either key. 1524 * Free both. 1525 */ 1526 uma_zfree(V_pf_state_key_z, skw); 1527 if (skw != sks) 1528 uma_zfree( 1529 V_pf_state_key_z, 1530 sks); 1531 return (EEXIST); /* collision! */ 1532 } 1533 } 1534 PF_HASHROW_UNLOCK(ih); 1535 } 1536 uma_zfree(V_pf_state_key_z, sk); 1537 s->key[idx] = cur; 1538 } else { 1539 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1540 s->key[idx] = sk; 1541 } 1542 1543 stateattach: 1544 /* List is sorted, if-bound states before floating. */ 1545 if (s->kif == V_pfi_all) 1546 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1547 else 1548 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1549 1550 if (olds) { 1551 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1552 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1553 key_list[idx]); 1554 olds = NULL; 1555 } 1556 1557 /* 1558 * Attach done. See how should we (or should not?) 1559 * attach a second key. 1560 */ 1561 if (sks == skw) { 1562 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1563 idx = PF_SK_STACK; 1564 sks = NULL; 1565 goto stateattach; 1566 } else if (sks != NULL) { 1567 /* 1568 * Continue attaching with stack key. 1569 */ 1570 sk = sks; 1571 kh = khs; 1572 idx = PF_SK_STACK; 1573 sks = NULL; 1574 goto keyattach; 1575 } 1576 1577 PF_STATE_LOCK(s); 1578 KEYS_UNLOCK(); 1579 1580 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1581 ("%s failure", __func__)); 1582 1583 return (0); 1584 #undef KEYS_UNLOCK 1585 } 1586 1587 static void 1588 pf_detach_state(struct pf_kstate *s) 1589 { 1590 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1591 struct pf_keyhash *kh; 1592 1593 NET_EPOCH_ASSERT(); 1594 MPASS(s->timeout >= PFTM_MAX); 1595 1596 pf_sctp_multihome_detach_addr(s); 1597 1598 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1599 V_pflow_export_state_ptr(s); 1600 1601 if (sks != NULL) { 1602 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1603 PF_HASHROW_LOCK(kh); 1604 if (s->key[PF_SK_STACK] != NULL) 1605 pf_state_key_detach(s, PF_SK_STACK); 1606 /* 1607 * If both point to same key, then we are done. 1608 */ 1609 if (sks == s->key[PF_SK_WIRE]) { 1610 pf_state_key_detach(s, PF_SK_WIRE); 1611 PF_HASHROW_UNLOCK(kh); 1612 return; 1613 } 1614 PF_HASHROW_UNLOCK(kh); 1615 } 1616 1617 if (s->key[PF_SK_WIRE] != NULL) { 1618 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1619 PF_HASHROW_LOCK(kh); 1620 if (s->key[PF_SK_WIRE] != NULL) 1621 pf_state_key_detach(s, PF_SK_WIRE); 1622 PF_HASHROW_UNLOCK(kh); 1623 } 1624 } 1625 1626 static void 1627 pf_state_key_detach(struct pf_kstate *s, int idx) 1628 { 1629 struct pf_state_key *sk = s->key[idx]; 1630 #ifdef INVARIANTS 1631 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1632 1633 PF_HASHROW_ASSERT(kh); 1634 #endif /* INVARIANTS */ 1635 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1636 s->key[idx] = NULL; 1637 1638 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1639 LIST_REMOVE(sk, entry); 1640 uma_zfree(V_pf_state_key_z, sk); 1641 } 1642 } 1643 1644 static int 1645 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1646 { 1647 struct pf_state_key *sk = mem; 1648 1649 bzero(sk, sizeof(struct pf_state_key_cmp)); 1650 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1651 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1652 1653 return (0); 1654 } 1655 1656 static int 1657 pf_state_key_addr_setup(struct pf_pdesc *pd, 1658 struct pf_state_key_cmp *key, int multi) 1659 { 1660 struct pf_addr *saddr = pd->src; 1661 struct pf_addr *daddr = pd->dst; 1662 #ifdef INET6 1663 struct nd_neighbor_solicit nd; 1664 struct pf_addr *target; 1665 u_short action, reason; 1666 1667 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1668 goto copy; 1669 1670 switch (pd->hdr.icmp6.icmp6_type) { 1671 case ND_NEIGHBOR_SOLICIT: 1672 if (multi) 1673 return (-1); 1674 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1675 return (-1); 1676 target = (struct pf_addr *)&nd.nd_ns_target; 1677 daddr = target; 1678 break; 1679 case ND_NEIGHBOR_ADVERT: 1680 if (multi) 1681 return (-1); 1682 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1683 return (-1); 1684 target = (struct pf_addr *)&nd.nd_ns_target; 1685 saddr = target; 1686 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1687 key->addr[pd->didx].addr32[0] = 0; 1688 key->addr[pd->didx].addr32[1] = 0; 1689 key->addr[pd->didx].addr32[2] = 0; 1690 key->addr[pd->didx].addr32[3] = 0; 1691 daddr = NULL; /* overwritten */ 1692 } 1693 break; 1694 default: 1695 if (multi) { 1696 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1697 key->addr[pd->sidx].addr32[1] = 0; 1698 key->addr[pd->sidx].addr32[2] = 0; 1699 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1700 saddr = NULL; /* overwritten */ 1701 } 1702 } 1703 copy: 1704 #endif /* INET6 */ 1705 if (saddr) 1706 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1707 if (daddr) 1708 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1709 1710 return (0); 1711 } 1712 1713 int 1714 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1715 struct pf_state_key **sk, struct pf_state_key **nk) 1716 { 1717 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1718 if (*sk == NULL) 1719 return (ENOMEM); 1720 1721 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1722 0)) { 1723 uma_zfree(V_pf_state_key_z, *sk); 1724 *sk = NULL; 1725 return (ENOMEM); 1726 } 1727 1728 (*sk)->port[pd->sidx] = sport; 1729 (*sk)->port[pd->didx] = dport; 1730 (*sk)->proto = pd->proto; 1731 (*sk)->af = pd->af; 1732 1733 *nk = pf_state_key_clone(*sk); 1734 if (*nk == NULL) { 1735 uma_zfree(V_pf_state_key_z, *sk); 1736 *sk = NULL; 1737 return (ENOMEM); 1738 } 1739 1740 if (pd->af != pd->naf) { 1741 (*sk)->port[pd->sidx] = pd->osport; 1742 (*sk)->port[pd->didx] = pd->odport; 1743 1744 (*nk)->af = pd->naf; 1745 1746 /* 1747 * We're overwriting an address here, so potentially there's bits of an IPv6 1748 * address left in here. Clear that out first. 1749 */ 1750 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 1751 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 1752 if (pd->dir == PF_IN) { 1753 PF_ACPY(&(*nk)->addr[pd->didx], &pd->nsaddr, pd->naf); 1754 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->ndaddr, pd->naf); 1755 (*nk)->port[pd->didx] = pd->nsport; 1756 (*nk)->port[pd->sidx] = pd->ndport; 1757 } else { 1758 PF_ACPY(&(*nk)->addr[pd->sidx], &pd->nsaddr, pd->naf); 1759 PF_ACPY(&(*nk)->addr[pd->didx], &pd->ndaddr, pd->naf); 1760 (*nk)->port[pd->sidx] = pd->nsport; 1761 (*nk)->port[pd->didx] = pd->ndport; 1762 } 1763 1764 switch (pd->proto) { 1765 case IPPROTO_ICMP: 1766 (*nk)->proto = IPPROTO_ICMPV6; 1767 break; 1768 case IPPROTO_ICMPV6: 1769 (*nk)->proto = IPPROTO_ICMP; 1770 break; 1771 default: 1772 (*nk)->proto = pd->proto; 1773 } 1774 } 1775 1776 return (0); 1777 } 1778 1779 struct pf_state_key * 1780 pf_state_key_clone(const struct pf_state_key *orig) 1781 { 1782 struct pf_state_key *sk; 1783 1784 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1785 if (sk == NULL) 1786 return (NULL); 1787 1788 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1789 1790 return (sk); 1791 } 1792 1793 int 1794 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1795 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1796 { 1797 struct pf_idhash *ih; 1798 struct pf_kstate *cur; 1799 int error; 1800 1801 NET_EPOCH_ASSERT(); 1802 1803 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1804 ("%s: sks not pristine", __func__)); 1805 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1806 ("%s: skw not pristine", __func__)); 1807 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1808 1809 s->kif = kif; 1810 s->orig_kif = orig_kif; 1811 1812 if (s->id == 0 && s->creatorid == 0) { 1813 s->id = alloc_unr64(&V_pf_stateid); 1814 s->id = htobe64(s->id); 1815 s->creatorid = V_pf_status.hostid; 1816 } 1817 1818 /* Returns with ID locked on success. */ 1819 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1820 return (error); 1821 skw = sks = NULL; 1822 1823 ih = &V_pf_idhash[PF_IDHASH(s)]; 1824 PF_HASHROW_ASSERT(ih); 1825 LIST_FOREACH(cur, &ih->states, entry) 1826 if (cur->id == s->id && cur->creatorid == s->creatorid) 1827 break; 1828 1829 if (cur != NULL) { 1830 s->timeout = PFTM_UNLINKED; 1831 PF_HASHROW_UNLOCK(ih); 1832 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1833 printf("pf: state ID collision: " 1834 "id: %016llx creatorid: %08x\n", 1835 (unsigned long long)be64toh(s->id), 1836 ntohl(s->creatorid)); 1837 } 1838 pf_detach_state(s); 1839 return (EEXIST); 1840 } 1841 LIST_INSERT_HEAD(&ih->states, s, entry); 1842 /* One for keys, one for ID hash. */ 1843 refcount_init(&s->refs, 2); 1844 1845 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1846 if (V_pfsync_insert_state_ptr != NULL) 1847 V_pfsync_insert_state_ptr(s); 1848 1849 /* Returns locked. */ 1850 return (0); 1851 } 1852 1853 /* 1854 * Find state by ID: returns with locked row on success. 1855 */ 1856 struct pf_kstate * 1857 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1858 { 1859 struct pf_idhash *ih; 1860 struct pf_kstate *s; 1861 1862 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1863 1864 ih = &V_pf_idhash[PF_IDHASHID(id)]; 1865 1866 PF_HASHROW_LOCK(ih); 1867 LIST_FOREACH(s, &ih->states, entry) 1868 if (s->id == id && s->creatorid == creatorid) 1869 break; 1870 1871 if (s == NULL) 1872 PF_HASHROW_UNLOCK(ih); 1873 1874 return (s); 1875 } 1876 1877 /* 1878 * Find state by key. 1879 * Returns with ID hash slot locked on success. 1880 */ 1881 static struct pf_kstate * 1882 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1883 u_int dir) 1884 { 1885 struct pf_keyhash *kh; 1886 struct pf_state_key *sk; 1887 struct pf_kstate *s; 1888 int idx; 1889 1890 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1891 1892 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1893 1894 PF_HASHROW_LOCK(kh); 1895 LIST_FOREACH(sk, &kh->keys, entry) 1896 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1897 break; 1898 if (sk == NULL) { 1899 PF_HASHROW_UNLOCK(kh); 1900 return (NULL); 1901 } 1902 1903 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1904 1905 /* List is sorted, if-bound states before floating ones. */ 1906 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1907 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1908 PF_STATE_LOCK(s); 1909 PF_HASHROW_UNLOCK(kh); 1910 if (__predict_false(s->timeout >= PFTM_MAX)) { 1911 /* 1912 * State is either being processed by 1913 * pf_remove_state() in an other thread, or 1914 * is scheduled for immediate expiry. 1915 */ 1916 PF_STATE_UNLOCK(s); 1917 return (NULL); 1918 } 1919 return (s); 1920 } 1921 1922 /* Look through the other list, in case of AF-TO */ 1923 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 1924 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1925 if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 1926 continue; 1927 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1928 PF_STATE_LOCK(s); 1929 PF_HASHROW_UNLOCK(kh); 1930 if (__predict_false(s->timeout >= PFTM_MAX)) { 1931 /* 1932 * State is either being processed by 1933 * pf_remove_state() in an other thread, or 1934 * is scheduled for immediate expiry. 1935 */ 1936 PF_STATE_UNLOCK(s); 1937 return (NULL); 1938 } 1939 return (s); 1940 } 1941 } 1942 1943 PF_HASHROW_UNLOCK(kh); 1944 1945 return (NULL); 1946 } 1947 1948 /* 1949 * Returns with ID hash slot locked on success. 1950 */ 1951 struct pf_kstate * 1952 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1953 { 1954 struct pf_keyhash *kh; 1955 struct pf_state_key *sk; 1956 struct pf_kstate *s, *ret = NULL; 1957 int idx, inout = 0; 1958 1959 if (more != NULL) 1960 *more = 0; 1961 1962 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1963 1964 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1965 1966 PF_HASHROW_LOCK(kh); 1967 LIST_FOREACH(sk, &kh->keys, entry) 1968 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1969 break; 1970 if (sk == NULL) { 1971 PF_HASHROW_UNLOCK(kh); 1972 return (NULL); 1973 } 1974 switch (dir) { 1975 case PF_IN: 1976 idx = PF_SK_WIRE; 1977 break; 1978 case PF_OUT: 1979 idx = PF_SK_STACK; 1980 break; 1981 case PF_INOUT: 1982 idx = PF_SK_WIRE; 1983 inout = 1; 1984 break; 1985 default: 1986 panic("%s: dir %u", __func__, dir); 1987 } 1988 second_run: 1989 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1990 if (more == NULL) { 1991 PF_STATE_LOCK(s); 1992 PF_HASHROW_UNLOCK(kh); 1993 return (s); 1994 } 1995 1996 if (ret) 1997 (*more)++; 1998 else { 1999 ret = s; 2000 PF_STATE_LOCK(s); 2001 } 2002 } 2003 if (inout == 1) { 2004 inout = 0; 2005 idx = PF_SK_STACK; 2006 goto second_run; 2007 } 2008 PF_HASHROW_UNLOCK(kh); 2009 2010 return (ret); 2011 } 2012 2013 /* 2014 * FIXME 2015 * This routine is inefficient -- locks the state only to unlock immediately on 2016 * return. 2017 * It is racy -- after the state is unlocked nothing stops other threads from 2018 * removing it. 2019 */ 2020 bool 2021 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2022 { 2023 struct pf_kstate *s; 2024 2025 s = pf_find_state_all(key, dir, NULL); 2026 if (s != NULL) { 2027 PF_STATE_UNLOCK(s); 2028 return (true); 2029 } 2030 return (false); 2031 } 2032 2033 struct pf_udp_mapping * 2034 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2035 struct pf_addr *nat_addr, uint16_t nat_port) 2036 { 2037 struct pf_udp_mapping *mapping; 2038 2039 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2040 if (mapping == NULL) 2041 return (NULL); 2042 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 2043 mapping->endpoints[0].port = src_port; 2044 mapping->endpoints[0].af = af; 2045 mapping->endpoints[0].mapping = mapping; 2046 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 2047 mapping->endpoints[1].port = nat_port; 2048 mapping->endpoints[1].af = af; 2049 mapping->endpoints[1].mapping = mapping; 2050 refcount_init(&mapping->refs, 1); 2051 return (mapping); 2052 } 2053 2054 int 2055 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2056 { 2057 struct pf_udpendpointhash *h0, *h1; 2058 struct pf_udp_endpoint *endpoint; 2059 int ret = EEXIST; 2060 2061 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2062 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2063 if (h0 == h1) { 2064 PF_HASHROW_LOCK(h0); 2065 } else if (h0 < h1) { 2066 PF_HASHROW_LOCK(h0); 2067 PF_HASHROW_LOCK(h1); 2068 } else { 2069 PF_HASHROW_LOCK(h1); 2070 PF_HASHROW_LOCK(h0); 2071 } 2072 2073 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2074 if (bcmp(endpoint, &mapping->endpoints[0], 2075 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2076 break; 2077 } 2078 if (endpoint != NULL) 2079 goto cleanup; 2080 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2081 if (bcmp(endpoint, &mapping->endpoints[1], 2082 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2083 break; 2084 } 2085 if (endpoint != NULL) 2086 goto cleanup; 2087 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2088 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2089 ret = 0; 2090 2091 cleanup: 2092 if (h0 != h1) { 2093 PF_HASHROW_UNLOCK(h0); 2094 PF_HASHROW_UNLOCK(h1); 2095 } else { 2096 PF_HASHROW_UNLOCK(h0); 2097 } 2098 return (ret); 2099 } 2100 2101 void 2102 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2103 { 2104 /* refcount is synchronized on the source endpoint's row lock */ 2105 struct pf_udpendpointhash *h0, *h1; 2106 2107 if (mapping == NULL) 2108 return; 2109 2110 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2111 PF_HASHROW_LOCK(h0); 2112 if (refcount_release(&mapping->refs)) { 2113 LIST_REMOVE(&mapping->endpoints[0], entry); 2114 PF_HASHROW_UNLOCK(h0); 2115 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2116 PF_HASHROW_LOCK(h1); 2117 LIST_REMOVE(&mapping->endpoints[1], entry); 2118 PF_HASHROW_UNLOCK(h1); 2119 2120 uma_zfree(V_pf_udp_mapping_z, mapping); 2121 } else { 2122 PF_HASHROW_UNLOCK(h0); 2123 } 2124 } 2125 2126 2127 struct pf_udp_mapping * 2128 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2129 { 2130 struct pf_udpendpointhash *uh; 2131 struct pf_udp_endpoint *endpoint; 2132 2133 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2134 2135 PF_HASHROW_LOCK(uh); 2136 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2137 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2138 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2139 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2140 break; 2141 } 2142 if (endpoint == NULL) { 2143 PF_HASHROW_UNLOCK(uh); 2144 return (NULL); 2145 } 2146 refcount_acquire(&endpoint->mapping->refs); 2147 PF_HASHROW_UNLOCK(uh); 2148 return (endpoint->mapping); 2149 } 2150 /* END state table stuff */ 2151 2152 static void 2153 pf_send(struct pf_send_entry *pfse) 2154 { 2155 2156 PF_SENDQ_LOCK(); 2157 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2158 PF_SENDQ_UNLOCK(); 2159 swi_sched(V_pf_swi_cookie, 0); 2160 } 2161 2162 static bool 2163 pf_isforlocal(struct mbuf *m, int af) 2164 { 2165 switch (af) { 2166 #ifdef INET 2167 case AF_INET: { 2168 struct ip *ip = mtod(m, struct ip *); 2169 2170 return (in_localip(ip->ip_dst)); 2171 } 2172 #endif /* INET */ 2173 #ifdef INET6 2174 case AF_INET6: { 2175 struct ip6_hdr *ip6; 2176 struct in6_ifaddr *ia; 2177 ip6 = mtod(m, struct ip6_hdr *); 2178 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2179 if (ia == NULL) 2180 return (false); 2181 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2182 } 2183 #endif /* INET6 */ 2184 default: 2185 unhandled_af(af); 2186 } 2187 2188 return (false); 2189 } 2190 2191 int 2192 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2193 int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type) 2194 { 2195 /* 2196 * ICMP types marked with PF_OUT are typically responses to 2197 * PF_IN, and will match states in the opposite direction. 2198 * PF_IN ICMP types need to match a state with that type. 2199 */ 2200 *icmp_dir = PF_OUT; 2201 2202 /* Queries (and responses) */ 2203 switch (pd->af) { 2204 #ifdef INET 2205 case AF_INET: 2206 switch (type) { 2207 case ICMP_ECHO: 2208 *icmp_dir = PF_IN; 2209 /* FALLTHROUGH */ 2210 case ICMP_ECHOREPLY: 2211 *virtual_type = ICMP_ECHO; 2212 *virtual_id = pd->hdr.icmp.icmp_id; 2213 break; 2214 2215 case ICMP_TSTAMP: 2216 *icmp_dir = PF_IN; 2217 /* FALLTHROUGH */ 2218 case ICMP_TSTAMPREPLY: 2219 *virtual_type = ICMP_TSTAMP; 2220 *virtual_id = pd->hdr.icmp.icmp_id; 2221 break; 2222 2223 case ICMP_IREQ: 2224 *icmp_dir = PF_IN; 2225 /* FALLTHROUGH */ 2226 case ICMP_IREQREPLY: 2227 *virtual_type = ICMP_IREQ; 2228 *virtual_id = pd->hdr.icmp.icmp_id; 2229 break; 2230 2231 case ICMP_MASKREQ: 2232 *icmp_dir = PF_IN; 2233 /* FALLTHROUGH */ 2234 case ICMP_MASKREPLY: 2235 *virtual_type = ICMP_MASKREQ; 2236 *virtual_id = pd->hdr.icmp.icmp_id; 2237 break; 2238 2239 case ICMP_IPV6_WHEREAREYOU: 2240 *icmp_dir = PF_IN; 2241 /* FALLTHROUGH */ 2242 case ICMP_IPV6_IAMHERE: 2243 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2244 *virtual_id = 0; /* Nothing sane to match on! */ 2245 break; 2246 2247 case ICMP_MOBILE_REGREQUEST: 2248 *icmp_dir = PF_IN; 2249 /* FALLTHROUGH */ 2250 case ICMP_MOBILE_REGREPLY: 2251 *virtual_type = ICMP_MOBILE_REGREQUEST; 2252 *virtual_id = 0; /* Nothing sane to match on! */ 2253 break; 2254 2255 case ICMP_ROUTERSOLICIT: 2256 *icmp_dir = PF_IN; 2257 /* FALLTHROUGH */ 2258 case ICMP_ROUTERADVERT: 2259 *virtual_type = ICMP_ROUTERSOLICIT; 2260 *virtual_id = 0; /* Nothing sane to match on! */ 2261 break; 2262 2263 /* These ICMP types map to other connections */ 2264 case ICMP_UNREACH: 2265 case ICMP_SOURCEQUENCH: 2266 case ICMP_REDIRECT: 2267 case ICMP_TIMXCEED: 2268 case ICMP_PARAMPROB: 2269 /* These will not be used, but set them anyway */ 2270 *icmp_dir = PF_IN; 2271 *virtual_type = type; 2272 *virtual_id = 0; 2273 *virtual_type = htons(*virtual_type); 2274 return (1); /* These types match to another state */ 2275 2276 /* 2277 * All remaining ICMP types get their own states, 2278 * and will only match in one direction. 2279 */ 2280 default: 2281 *icmp_dir = PF_IN; 2282 *virtual_type = type; 2283 *virtual_id = 0; 2284 break; 2285 } 2286 break; 2287 #endif /* INET */ 2288 #ifdef INET6 2289 case AF_INET6: 2290 switch (type) { 2291 case ICMP6_ECHO_REQUEST: 2292 *icmp_dir = PF_IN; 2293 /* FALLTHROUGH */ 2294 case ICMP6_ECHO_REPLY: 2295 *virtual_type = ICMP6_ECHO_REQUEST; 2296 *virtual_id = pd->hdr.icmp6.icmp6_id; 2297 break; 2298 2299 case MLD_LISTENER_QUERY: 2300 case MLD_LISTENER_REPORT: { 2301 /* 2302 * Listener Report can be sent by clients 2303 * without an associated Listener Query. 2304 * In addition to that, when Report is sent as a 2305 * reply to a Query its source and destination 2306 * address are different. 2307 */ 2308 *icmp_dir = PF_IN; 2309 *virtual_type = MLD_LISTENER_QUERY; 2310 *virtual_id = 0; 2311 break; 2312 } 2313 case MLD_MTRACE: 2314 *icmp_dir = PF_IN; 2315 /* FALLTHROUGH */ 2316 case MLD_MTRACE_RESP: 2317 *virtual_type = MLD_MTRACE; 2318 *virtual_id = 0; /* Nothing sane to match on! */ 2319 break; 2320 2321 case ND_NEIGHBOR_SOLICIT: 2322 *icmp_dir = PF_IN; 2323 /* FALLTHROUGH */ 2324 case ND_NEIGHBOR_ADVERT: { 2325 *virtual_type = ND_NEIGHBOR_SOLICIT; 2326 *virtual_id = 0; 2327 break; 2328 } 2329 2330 /* 2331 * These ICMP types map to other connections. 2332 * ND_REDIRECT can't be in this list because the triggering 2333 * packet header is optional. 2334 */ 2335 case ICMP6_DST_UNREACH: 2336 case ICMP6_PACKET_TOO_BIG: 2337 case ICMP6_TIME_EXCEEDED: 2338 case ICMP6_PARAM_PROB: 2339 /* These will not be used, but set them anyway */ 2340 *icmp_dir = PF_IN; 2341 *virtual_type = type; 2342 *virtual_id = 0; 2343 *virtual_type = htons(*virtual_type); 2344 return (1); /* These types match to another state */ 2345 /* 2346 * All remaining ICMP6 types get their own states, 2347 * and will only match in one direction. 2348 */ 2349 default: 2350 *icmp_dir = PF_IN; 2351 *virtual_type = type; 2352 *virtual_id = 0; 2353 break; 2354 } 2355 break; 2356 #endif /* INET6 */ 2357 default: 2358 unhandled_af(pd->af); 2359 } 2360 *virtual_type = htons(*virtual_type); 2361 return (0); /* These types match to their own state */ 2362 } 2363 2364 void 2365 pf_intr(void *v) 2366 { 2367 struct epoch_tracker et; 2368 struct pf_send_head queue; 2369 struct pf_send_entry *pfse, *next; 2370 2371 CURVNET_SET((struct vnet *)v); 2372 2373 PF_SENDQ_LOCK(); 2374 queue = V_pf_sendqueue; 2375 STAILQ_INIT(&V_pf_sendqueue); 2376 PF_SENDQ_UNLOCK(); 2377 2378 NET_EPOCH_ENTER(et); 2379 2380 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2381 switch (pfse->pfse_type) { 2382 #ifdef INET 2383 case PFSE_IP: { 2384 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2385 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2386 ("%s: rcvif != loif", __func__)); 2387 2388 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2389 pfse->pfse_m->m_pkthdr.csum_flags |= 2390 CSUM_IP_VALID | CSUM_IP_CHECKED | 2391 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2392 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2393 ip_input(pfse->pfse_m); 2394 } else { 2395 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2396 NULL); 2397 } 2398 break; 2399 } 2400 case PFSE_ICMP: 2401 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2402 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2403 break; 2404 #endif /* INET */ 2405 #ifdef INET6 2406 case PFSE_IP6: 2407 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2408 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2409 ("%s: rcvif != loif", __func__)); 2410 2411 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2412 M_LOOP; 2413 pfse->pfse_m->m_pkthdr.csum_flags |= 2414 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2415 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2416 ip6_input(pfse->pfse_m); 2417 } else { 2418 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2419 NULL, NULL); 2420 } 2421 break; 2422 case PFSE_ICMP6: 2423 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2424 pfse->icmpopts.code, pfse->icmpopts.mtu); 2425 break; 2426 #endif /* INET6 */ 2427 default: 2428 panic("%s: unknown type", __func__); 2429 } 2430 free(pfse, M_PFTEMP); 2431 } 2432 NET_EPOCH_EXIT(et); 2433 CURVNET_RESTORE(); 2434 } 2435 2436 #define pf_purge_thread_period (hz / 10) 2437 2438 #ifdef PF_WANT_32_TO_64_COUNTER 2439 static void 2440 pf_status_counter_u64_periodic(void) 2441 { 2442 2443 PF_RULES_RASSERT(); 2444 2445 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2446 return; 2447 } 2448 2449 for (int i = 0; i < FCNT_MAX; i++) { 2450 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2451 } 2452 } 2453 2454 static void 2455 pf_kif_counter_u64_periodic(void) 2456 { 2457 struct pfi_kkif *kif; 2458 size_t r, run; 2459 2460 PF_RULES_RASSERT(); 2461 2462 if (__predict_false(V_pf_allkifcount == 0)) { 2463 return; 2464 } 2465 2466 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2467 return; 2468 } 2469 2470 run = V_pf_allkifcount / 10; 2471 if (run < 5) 2472 run = 5; 2473 2474 for (r = 0; r < run; r++) { 2475 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2476 if (kif == NULL) { 2477 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2478 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2479 break; 2480 } 2481 2482 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2483 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2484 2485 for (int i = 0; i < 2; i++) { 2486 for (int j = 0; j < 2; j++) { 2487 for (int k = 0; k < 2; k++) { 2488 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2489 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2490 } 2491 } 2492 } 2493 } 2494 } 2495 2496 static void 2497 pf_rule_counter_u64_periodic(void) 2498 { 2499 struct pf_krule *rule; 2500 size_t r, run; 2501 2502 PF_RULES_RASSERT(); 2503 2504 if (__predict_false(V_pf_allrulecount == 0)) { 2505 return; 2506 } 2507 2508 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2509 return; 2510 } 2511 2512 run = V_pf_allrulecount / 10; 2513 if (run < 5) 2514 run = 5; 2515 2516 for (r = 0; r < run; r++) { 2517 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2518 if (rule == NULL) { 2519 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2520 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2521 break; 2522 } 2523 2524 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2525 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2526 2527 pf_counter_u64_periodic(&rule->evaluations); 2528 for (int i = 0; i < 2; i++) { 2529 pf_counter_u64_periodic(&rule->packets[i]); 2530 pf_counter_u64_periodic(&rule->bytes[i]); 2531 } 2532 } 2533 } 2534 2535 static void 2536 pf_counter_u64_periodic_main(void) 2537 { 2538 PF_RULES_RLOCK_TRACKER; 2539 2540 V_pf_counter_periodic_iter++; 2541 2542 PF_RULES_RLOCK(); 2543 pf_counter_u64_critical_enter(); 2544 pf_status_counter_u64_periodic(); 2545 pf_kif_counter_u64_periodic(); 2546 pf_rule_counter_u64_periodic(); 2547 pf_counter_u64_critical_exit(); 2548 PF_RULES_RUNLOCK(); 2549 } 2550 #else 2551 #define pf_counter_u64_periodic_main() do { } while (0) 2552 #endif 2553 2554 void 2555 pf_purge_thread(void *unused __unused) 2556 { 2557 struct epoch_tracker et; 2558 2559 VNET_ITERATOR_DECL(vnet_iter); 2560 2561 sx_xlock(&pf_end_lock); 2562 while (pf_end_threads == 0) { 2563 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2564 2565 VNET_LIST_RLOCK(); 2566 NET_EPOCH_ENTER(et); 2567 VNET_FOREACH(vnet_iter) { 2568 CURVNET_SET(vnet_iter); 2569 2570 /* Wait until V_pf_default_rule is initialized. */ 2571 if (V_pf_vnet_active == 0) { 2572 CURVNET_RESTORE(); 2573 continue; 2574 } 2575 2576 pf_counter_u64_periodic_main(); 2577 2578 /* 2579 * Process 1/interval fraction of the state 2580 * table every run. 2581 */ 2582 V_pf_purge_idx = 2583 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2584 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2585 2586 /* 2587 * Purge other expired types every 2588 * PFTM_INTERVAL seconds. 2589 */ 2590 if (V_pf_purge_idx == 0) { 2591 /* 2592 * Order is important: 2593 * - states and src nodes reference rules 2594 * - states and rules reference kifs 2595 */ 2596 pf_purge_expired_fragments(); 2597 pf_purge_expired_src_nodes(); 2598 pf_purge_unlinked_rules(); 2599 pfi_kkif_purge(); 2600 } 2601 CURVNET_RESTORE(); 2602 } 2603 NET_EPOCH_EXIT(et); 2604 VNET_LIST_RUNLOCK(); 2605 } 2606 2607 pf_end_threads++; 2608 sx_xunlock(&pf_end_lock); 2609 kproc_exit(0); 2610 } 2611 2612 void 2613 pf_unload_vnet_purge(void) 2614 { 2615 2616 /* 2617 * To cleanse up all kifs and rules we need 2618 * two runs: first one clears reference flags, 2619 * then pf_purge_expired_states() doesn't 2620 * raise them, and then second run frees. 2621 */ 2622 pf_purge_unlinked_rules(); 2623 pfi_kkif_purge(); 2624 2625 /* 2626 * Now purge everything. 2627 */ 2628 pf_purge_expired_states(0, V_pf_hashmask); 2629 pf_purge_fragments(UINT_MAX); 2630 pf_purge_expired_src_nodes(); 2631 2632 /* 2633 * Now all kifs & rules should be unreferenced, 2634 * thus should be successfully freed. 2635 */ 2636 pf_purge_unlinked_rules(); 2637 pfi_kkif_purge(); 2638 } 2639 2640 u_int32_t 2641 pf_state_expires(const struct pf_kstate *state) 2642 { 2643 u_int32_t timeout; 2644 u_int32_t start; 2645 u_int32_t end; 2646 u_int32_t states; 2647 2648 /* handle all PFTM_* > PFTM_MAX here */ 2649 if (state->timeout == PFTM_PURGE) 2650 return (time_uptime); 2651 KASSERT(state->timeout != PFTM_UNLINKED, 2652 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2653 KASSERT((state->timeout < PFTM_MAX), 2654 ("pf_state_expires: timeout > PFTM_MAX")); 2655 timeout = state->rule->timeout[state->timeout]; 2656 if (!timeout) 2657 timeout = V_pf_default_rule.timeout[state->timeout]; 2658 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2659 if (start && state->rule != &V_pf_default_rule) { 2660 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2661 states = counter_u64_fetch(state->rule->states_cur); 2662 } else { 2663 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2664 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2665 states = V_pf_status.states; 2666 } 2667 if (end && states > start && start < end) { 2668 if (states < end) { 2669 timeout = (u_int64_t)timeout * (end - states) / 2670 (end - start); 2671 return ((state->expire / 1000) + timeout); 2672 } 2673 else 2674 return (time_uptime); 2675 } 2676 return ((state->expire / 1000) + timeout); 2677 } 2678 2679 void 2680 pf_purge_expired_src_nodes(void) 2681 { 2682 struct pf_ksrc_node_list freelist; 2683 struct pf_srchash *sh; 2684 struct pf_ksrc_node *cur, *next; 2685 int i; 2686 2687 LIST_INIT(&freelist); 2688 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2689 PF_HASHROW_LOCK(sh); 2690 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2691 if (cur->states == 0 && cur->expire <= time_uptime) { 2692 pf_unlink_src_node(cur); 2693 LIST_INSERT_HEAD(&freelist, cur, entry); 2694 } else if (cur->rule != NULL) 2695 cur->rule->rule_ref |= PFRULE_REFS; 2696 PF_HASHROW_UNLOCK(sh); 2697 } 2698 2699 pf_free_src_nodes(&freelist); 2700 2701 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2702 } 2703 2704 static void 2705 pf_src_tree_remove_state(struct pf_kstate *s) 2706 { 2707 uint32_t timeout; 2708 2709 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2710 s->rule->timeout[PFTM_SRC_NODE] : 2711 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2712 2713 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 2714 if (s->sns[sn_type] == NULL) 2715 continue; 2716 PF_SRC_NODE_LOCK(s->sns[sn_type]); 2717 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 2718 --(s->sns[sn_type]->conn); 2719 if (--(s->sns[sn_type]->states) == 0) 2720 s->sns[sn_type]->expire = time_uptime + timeout; 2721 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 2722 s->sns[sn_type] = NULL; 2723 } 2724 2725 } 2726 2727 /* 2728 * Unlink and potentilly free a state. Function may be 2729 * called with ID hash row locked, but always returns 2730 * unlocked, since it needs to go through key hash locking. 2731 */ 2732 int 2733 pf_remove_state(struct pf_kstate *s) 2734 { 2735 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2736 2737 NET_EPOCH_ASSERT(); 2738 PF_HASHROW_ASSERT(ih); 2739 2740 if (s->timeout == PFTM_UNLINKED) { 2741 /* 2742 * State is being processed 2743 * by pf_remove_state() in 2744 * an other thread. 2745 */ 2746 PF_HASHROW_UNLOCK(ih); 2747 return (0); /* XXXGL: undefined actually */ 2748 } 2749 2750 if (s->src.state == PF_TCPS_PROXY_DST) { 2751 /* XXX wire key the right one? */ 2752 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2753 &s->key[PF_SK_WIRE]->addr[1], 2754 &s->key[PF_SK_WIRE]->addr[0], 2755 s->key[PF_SK_WIRE]->port[1], 2756 s->key[PF_SK_WIRE]->port[0], 2757 s->src.seqhi, s->src.seqlo + 1, 2758 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2759 s->act.rtableid); 2760 } 2761 2762 LIST_REMOVE(s, entry); 2763 pf_src_tree_remove_state(s); 2764 2765 if (V_pfsync_delete_state_ptr != NULL) 2766 V_pfsync_delete_state_ptr(s); 2767 2768 STATE_DEC_COUNTERS(s); 2769 2770 s->timeout = PFTM_UNLINKED; 2771 2772 /* Ensure we remove it from the list of halfopen states, if needed. */ 2773 if (s->key[PF_SK_STACK] != NULL && 2774 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2775 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2776 2777 PF_HASHROW_UNLOCK(ih); 2778 2779 pf_detach_state(s); 2780 2781 pf_udp_mapping_release(s->udp_mapping); 2782 2783 /* pf_state_insert() initialises refs to 2 */ 2784 return (pf_release_staten(s, 2)); 2785 } 2786 2787 struct pf_kstate * 2788 pf_alloc_state(int flags) 2789 { 2790 2791 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2792 } 2793 2794 void 2795 pf_free_state(struct pf_kstate *cur) 2796 { 2797 struct pf_krule_item *ri; 2798 2799 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2800 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2801 cur->timeout)); 2802 2803 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2804 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2805 free(ri, M_PF_RULE_ITEM); 2806 } 2807 2808 pf_normalize_tcp_cleanup(cur); 2809 uma_zfree(V_pf_state_z, cur); 2810 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2811 } 2812 2813 /* 2814 * Called only from pf_purge_thread(), thus serialized. 2815 */ 2816 static u_int 2817 pf_purge_expired_states(u_int i, int maxcheck) 2818 { 2819 struct pf_idhash *ih; 2820 struct pf_kstate *s; 2821 struct pf_krule_item *mrm; 2822 size_t count __unused; 2823 2824 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2825 2826 /* 2827 * Go through hash and unlink states that expire now. 2828 */ 2829 while (maxcheck > 0) { 2830 count = 0; 2831 ih = &V_pf_idhash[i]; 2832 2833 /* only take the lock if we expect to do work */ 2834 if (!LIST_EMPTY(&ih->states)) { 2835 relock: 2836 PF_HASHROW_LOCK(ih); 2837 LIST_FOREACH(s, &ih->states, entry) { 2838 if (pf_state_expires(s) <= time_uptime) { 2839 V_pf_status.states -= 2840 pf_remove_state(s); 2841 goto relock; 2842 } 2843 s->rule->rule_ref |= PFRULE_REFS; 2844 if (s->nat_rule != NULL) 2845 s->nat_rule->rule_ref |= PFRULE_REFS; 2846 if (s->anchor != NULL) 2847 s->anchor->rule_ref |= PFRULE_REFS; 2848 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2849 SLIST_FOREACH(mrm, &s->match_rules, entry) 2850 mrm->r->rule_ref |= PFRULE_REFS; 2851 if (s->act.rt_kif) 2852 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2853 count++; 2854 } 2855 PF_HASHROW_UNLOCK(ih); 2856 } 2857 2858 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2859 2860 /* Return when we hit end of hash. */ 2861 if (++i > V_pf_hashmask) { 2862 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2863 return (0); 2864 } 2865 2866 maxcheck--; 2867 } 2868 2869 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2870 2871 return (i); 2872 } 2873 2874 static void 2875 pf_purge_unlinked_rules(void) 2876 { 2877 struct pf_krulequeue tmpq; 2878 struct pf_krule *r, *r1; 2879 2880 /* 2881 * If we have overloading task pending, then we'd 2882 * better skip purging this time. There is a tiny 2883 * probability that overloading task references 2884 * an already unlinked rule. 2885 */ 2886 PF_OVERLOADQ_LOCK(); 2887 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2888 PF_OVERLOADQ_UNLOCK(); 2889 return; 2890 } 2891 PF_OVERLOADQ_UNLOCK(); 2892 2893 /* 2894 * Do naive mark-and-sweep garbage collecting of old rules. 2895 * Reference flag is raised by pf_purge_expired_states() 2896 * and pf_purge_expired_src_nodes(). 2897 * 2898 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2899 * use a temporary queue. 2900 */ 2901 TAILQ_INIT(&tmpq); 2902 PF_UNLNKDRULES_LOCK(); 2903 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2904 if (!(r->rule_ref & PFRULE_REFS)) { 2905 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2906 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2907 } else 2908 r->rule_ref &= ~PFRULE_REFS; 2909 } 2910 PF_UNLNKDRULES_UNLOCK(); 2911 2912 if (!TAILQ_EMPTY(&tmpq)) { 2913 PF_CONFIG_LOCK(); 2914 PF_RULES_WLOCK(); 2915 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2916 TAILQ_REMOVE(&tmpq, r, entries); 2917 pf_free_rule(r); 2918 } 2919 PF_RULES_WUNLOCK(); 2920 PF_CONFIG_UNLOCK(); 2921 } 2922 } 2923 2924 void 2925 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2926 { 2927 switch (af) { 2928 #ifdef INET 2929 case AF_INET: { 2930 u_int32_t a = ntohl(addr->addr32[0]); 2931 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2932 (a>>8)&255, a&255); 2933 if (p) { 2934 p = ntohs(p); 2935 printf(":%u", p); 2936 } 2937 break; 2938 } 2939 #endif /* INET */ 2940 #ifdef INET6 2941 case AF_INET6: { 2942 u_int16_t b; 2943 u_int8_t i, curstart, curend, maxstart, maxend; 2944 curstart = curend = maxstart = maxend = 255; 2945 for (i = 0; i < 8; i++) { 2946 if (!addr->addr16[i]) { 2947 if (curstart == 255) 2948 curstart = i; 2949 curend = i; 2950 } else { 2951 if ((curend - curstart) > 2952 (maxend - maxstart)) { 2953 maxstart = curstart; 2954 maxend = curend; 2955 } 2956 curstart = curend = 255; 2957 } 2958 } 2959 if ((curend - curstart) > 2960 (maxend - maxstart)) { 2961 maxstart = curstart; 2962 maxend = curend; 2963 } 2964 for (i = 0; i < 8; i++) { 2965 if (i >= maxstart && i <= maxend) { 2966 if (i == 0) 2967 printf(":"); 2968 if (i == maxend) 2969 printf(":"); 2970 } else { 2971 b = ntohs(addr->addr16[i]); 2972 printf("%x", b); 2973 if (i < 7) 2974 printf(":"); 2975 } 2976 } 2977 if (p) { 2978 p = ntohs(p); 2979 printf("[%u]", p); 2980 } 2981 break; 2982 } 2983 #endif /* INET6 */ 2984 default: 2985 unhandled_af(af); 2986 } 2987 } 2988 2989 void 2990 pf_print_state(struct pf_kstate *s) 2991 { 2992 pf_print_state_parts(s, NULL, NULL); 2993 } 2994 2995 static void 2996 pf_print_state_parts(struct pf_kstate *s, 2997 struct pf_state_key *skwp, struct pf_state_key *sksp) 2998 { 2999 struct pf_state_key *skw, *sks; 3000 u_int8_t proto, dir; 3001 3002 /* Do our best to fill these, but they're skipped if NULL */ 3003 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 3004 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 3005 proto = skw ? skw->proto : (sks ? sks->proto : 0); 3006 dir = s ? s->direction : 0; 3007 3008 switch (proto) { 3009 case IPPROTO_IPV4: 3010 printf("IPv4"); 3011 break; 3012 case IPPROTO_IPV6: 3013 printf("IPv6"); 3014 break; 3015 case IPPROTO_TCP: 3016 printf("TCP"); 3017 break; 3018 case IPPROTO_UDP: 3019 printf("UDP"); 3020 break; 3021 case IPPROTO_ICMP: 3022 printf("ICMP"); 3023 break; 3024 case IPPROTO_ICMPV6: 3025 printf("ICMPv6"); 3026 break; 3027 default: 3028 printf("%u", proto); 3029 break; 3030 } 3031 switch (dir) { 3032 case PF_IN: 3033 printf(" in"); 3034 break; 3035 case PF_OUT: 3036 printf(" out"); 3037 break; 3038 } 3039 if (skw) { 3040 printf(" wire: "); 3041 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3042 printf(" "); 3043 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3044 } 3045 if (sks) { 3046 printf(" stack: "); 3047 if (sks != skw) { 3048 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3049 printf(" "); 3050 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3051 } else 3052 printf("-"); 3053 } 3054 if (s) { 3055 if (proto == IPPROTO_TCP) { 3056 printf(" [lo=%u high=%u win=%u modulator=%u", 3057 s->src.seqlo, s->src.seqhi, 3058 s->src.max_win, s->src.seqdiff); 3059 if (s->src.wscale && s->dst.wscale) 3060 printf(" wscale=%u", 3061 s->src.wscale & PF_WSCALE_MASK); 3062 printf("]"); 3063 printf(" [lo=%u high=%u win=%u modulator=%u", 3064 s->dst.seqlo, s->dst.seqhi, 3065 s->dst.max_win, s->dst.seqdiff); 3066 if (s->src.wscale && s->dst.wscale) 3067 printf(" wscale=%u", 3068 s->dst.wscale & PF_WSCALE_MASK); 3069 printf("]"); 3070 } 3071 printf(" %u:%u", s->src.state, s->dst.state); 3072 if (s->rule) 3073 printf(" @%d", s->rule->nr); 3074 } 3075 } 3076 3077 void 3078 pf_print_flags(uint16_t f) 3079 { 3080 if (f) 3081 printf(" "); 3082 if (f & TH_FIN) 3083 printf("F"); 3084 if (f & TH_SYN) 3085 printf("S"); 3086 if (f & TH_RST) 3087 printf("R"); 3088 if (f & TH_PUSH) 3089 printf("P"); 3090 if (f & TH_ACK) 3091 printf("A"); 3092 if (f & TH_URG) 3093 printf("U"); 3094 if (f & TH_ECE) 3095 printf("E"); 3096 if (f & TH_CWR) 3097 printf("W"); 3098 if (f & TH_AE) 3099 printf("e"); 3100 } 3101 3102 #define PF_SET_SKIP_STEPS(i) \ 3103 do { \ 3104 while (head[i] != cur) { \ 3105 head[i]->skip[i] = cur; \ 3106 head[i] = TAILQ_NEXT(head[i], entries); \ 3107 } \ 3108 } while (0) 3109 3110 void 3111 pf_calc_skip_steps(struct pf_krulequeue *rules) 3112 { 3113 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3114 int i; 3115 3116 cur = TAILQ_FIRST(rules); 3117 prev = cur; 3118 for (i = 0; i < PF_SKIP_COUNT; ++i) 3119 head[i] = cur; 3120 while (cur != NULL) { 3121 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3122 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3123 if (cur->direction != prev->direction) 3124 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3125 if (cur->af != prev->af) 3126 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3127 if (cur->proto != prev->proto) 3128 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3129 if (cur->src.neg != prev->src.neg || 3130 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3131 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3132 if (cur->dst.neg != prev->dst.neg || 3133 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3134 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3135 if (cur->src.port[0] != prev->src.port[0] || 3136 cur->src.port[1] != prev->src.port[1] || 3137 cur->src.port_op != prev->src.port_op) 3138 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3139 if (cur->dst.port[0] != prev->dst.port[0] || 3140 cur->dst.port[1] != prev->dst.port[1] || 3141 cur->dst.port_op != prev->dst.port_op) 3142 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3143 3144 prev = cur; 3145 cur = TAILQ_NEXT(cur, entries); 3146 } 3147 for (i = 0; i < PF_SKIP_COUNT; ++i) 3148 PF_SET_SKIP_STEPS(i); 3149 } 3150 3151 int 3152 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3153 { 3154 if (aw1->type != aw2->type) 3155 return (1); 3156 switch (aw1->type) { 3157 case PF_ADDR_ADDRMASK: 3158 case PF_ADDR_RANGE: 3159 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3160 return (1); 3161 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3162 return (1); 3163 return (0); 3164 case PF_ADDR_DYNIFTL: 3165 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3166 case PF_ADDR_NONE: 3167 case PF_ADDR_NOROUTE: 3168 case PF_ADDR_URPFFAILED: 3169 return (0); 3170 case PF_ADDR_TABLE: 3171 return (aw1->p.tbl != aw2->p.tbl); 3172 default: 3173 printf("invalid address type: %d\n", aw1->type); 3174 return (1); 3175 } 3176 } 3177 3178 /** 3179 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3180 * header isn't always a full checksum. In some cases (i.e. output) it's a 3181 * pseudo-header checksum, which is a partial checksum over src/dst IP 3182 * addresses, protocol number and length. 3183 * 3184 * That means we have the following cases: 3185 * * Input or forwarding: we don't have TSO, the checksum fields are full 3186 * checksums, we need to update the checksum whenever we change anything. 3187 * * Output (i.e. the checksum is a pseudo-header checksum): 3188 * x The field being updated is src/dst address or affects the length of 3189 * the packet. We need to update the pseudo-header checksum (note that this 3190 * checksum is not ones' complement). 3191 * x Some other field is being modified (e.g. src/dst port numbers): We 3192 * don't have to update anything. 3193 **/ 3194 u_int16_t 3195 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3196 { 3197 u_int32_t x; 3198 3199 x = cksum + old - new; 3200 x = (x + (x >> 16)) & 0xffff; 3201 3202 /* optimise: eliminate a branch when not udp */ 3203 if (udp && cksum == 0x0000) 3204 return cksum; 3205 if (udp && x == 0x0000) 3206 x = 0xffff; 3207 3208 return (u_int16_t)(x); 3209 } 3210 3211 static void 3212 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3213 u_int8_t udp) 3214 { 3215 u_int16_t old = htons(hi ? (*f << 8) : *f); 3216 u_int16_t new = htons(hi ? ( v << 8) : v); 3217 3218 if (*f == v) 3219 return; 3220 3221 *f = v; 3222 3223 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3224 return; 3225 3226 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3227 } 3228 3229 void 3230 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3231 bool hi, u_int8_t udp) 3232 { 3233 u_int8_t *fb = (u_int8_t *)f; 3234 u_int8_t *vb = (u_int8_t *)&v; 3235 3236 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3237 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3238 } 3239 3240 void 3241 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3242 bool hi, u_int8_t udp) 3243 { 3244 u_int8_t *fb = (u_int8_t *)f; 3245 u_int8_t *vb = (u_int8_t *)&v; 3246 3247 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3248 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3249 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3250 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3251 } 3252 3253 u_int16_t 3254 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3255 u_int16_t new, u_int8_t udp) 3256 { 3257 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3258 return (cksum); 3259 3260 return (pf_cksum_fixup(cksum, old, new, udp)); 3261 } 3262 3263 static void 3264 pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3265 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3266 sa_family_t af, sa_family_t naf) 3267 { 3268 struct pf_addr ao; 3269 u_int16_t po; 3270 3271 PF_ACPY(&ao, a, af); 3272 if (af == naf) 3273 PF_ACPY(a, an, af); 3274 3275 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3276 *pc = ~*pc; 3277 3278 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3279 return; 3280 po = *p; 3281 *p = pn; 3282 3283 switch (af) { 3284 #ifdef INET 3285 case AF_INET: 3286 switch (naf) { 3287 case AF_INET: 3288 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3289 ao.addr16[0], an->addr16[0], 0), 3290 ao.addr16[1], an->addr16[1], 0); 3291 *p = pn; 3292 3293 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3294 ao.addr16[0], an->addr16[0], u), 3295 ao.addr16[1], an->addr16[1], u); 3296 3297 *pc = pf_proto_cksum_fixup(pd->m, *pc, po, pn, u); 3298 break; 3299 #ifdef INET6 3300 case AF_INET6: 3301 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3302 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3303 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3304 ao.addr16[0], an->addr16[0], u), 3305 ao.addr16[1], an->addr16[1], u), 3306 0, an->addr16[2], u), 3307 0, an->addr16[3], u), 3308 0, an->addr16[4], u), 3309 0, an->addr16[5], u), 3310 0, an->addr16[6], u), 3311 0, an->addr16[7], u), 3312 po, pn, u); 3313 3314 /* XXXKP TODO *ic checksum? */ 3315 break; 3316 #endif /* INET6 */ 3317 } 3318 break; 3319 #endif /* INET */ 3320 #ifdef INET6 3321 case AF_INET6: 3322 switch (naf) { 3323 #ifdef INET 3324 case AF_INET: 3325 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3326 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3327 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 3328 ao.addr16[0], an->addr16[0], u), 3329 ao.addr16[1], an->addr16[1], u), 3330 ao.addr16[2], 0, u), 3331 ao.addr16[3], 0, u), 3332 ao.addr16[4], 0, u), 3333 ao.addr16[5], 0, u), 3334 ao.addr16[6], 0, u), 3335 ao.addr16[7], 0, u), 3336 po, pn, u); 3337 3338 /* XXXKP TODO *ic checksum? */ 3339 break; 3340 #endif /* INET */ 3341 case AF_INET6: 3342 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3343 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3344 pf_cksum_fixup(pf_cksum_fixup(*pc, 3345 ao.addr16[0], an->addr16[0], u), 3346 ao.addr16[1], an->addr16[1], u), 3347 ao.addr16[2], an->addr16[2], u), 3348 ao.addr16[3], an->addr16[3], u), 3349 ao.addr16[4], an->addr16[4], u), 3350 ao.addr16[5], an->addr16[5], u), 3351 ao.addr16[6], an->addr16[6], u), 3352 ao.addr16[7], an->addr16[7], u); 3353 3354 *pc = pf_proto_cksum_fixup(pd->m, *pc, po, pn, u); 3355 break; 3356 } 3357 break; 3358 #endif /* INET6 */ 3359 default: 3360 unhandled_af(af); 3361 } 3362 3363 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3364 CSUM_DELAY_DATA_IPV6)) { 3365 *pc = ~*pc; 3366 if (! *pc) 3367 *pc = 0xffff; 3368 } 3369 } 3370 3371 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3372 void 3373 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3374 { 3375 u_int32_t ao; 3376 3377 memcpy(&ao, a, sizeof(ao)); 3378 memcpy(a, &an, sizeof(u_int32_t)); 3379 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3380 ao % 65536, an % 65536, u); 3381 } 3382 3383 void 3384 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3385 { 3386 u_int32_t ao; 3387 3388 memcpy(&ao, a, sizeof(ao)); 3389 memcpy(a, &an, sizeof(u_int32_t)); 3390 3391 *c = pf_proto_cksum_fixup(m, 3392 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3393 ao % 65536, an % 65536, udp); 3394 } 3395 3396 #ifdef INET6 3397 static void 3398 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3399 { 3400 struct pf_addr ao; 3401 3402 PF_ACPY(&ao, a, AF_INET6); 3403 PF_ACPY(a, an, AF_INET6); 3404 3405 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3406 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3407 pf_cksum_fixup(pf_cksum_fixup(*c, 3408 ao.addr16[0], an->addr16[0], u), 3409 ao.addr16[1], an->addr16[1], u), 3410 ao.addr16[2], an->addr16[2], u), 3411 ao.addr16[3], an->addr16[3], u), 3412 ao.addr16[4], an->addr16[4], u), 3413 ao.addr16[5], an->addr16[5], u), 3414 ao.addr16[6], an->addr16[6], u), 3415 ao.addr16[7], an->addr16[7], u); 3416 } 3417 #endif /* INET6 */ 3418 3419 static void 3420 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3421 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3422 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3423 { 3424 struct pf_addr oia, ooa; 3425 3426 PF_ACPY(&oia, ia, af); 3427 if (oa) 3428 PF_ACPY(&ooa, oa, af); 3429 3430 /* Change inner protocol port, fix inner protocol checksum. */ 3431 if (ip != NULL) { 3432 u_int16_t oip = *ip; 3433 u_int32_t opc; 3434 3435 if (pc != NULL) 3436 opc = *pc; 3437 *ip = np; 3438 if (pc != NULL) 3439 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3440 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3441 if (pc != NULL) 3442 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3443 } 3444 /* Change inner ip address, fix inner ip and icmp checksums. */ 3445 PF_ACPY(ia, na, af); 3446 switch (af) { 3447 #ifdef INET 3448 case AF_INET: { 3449 u_int32_t oh2c = *h2c; 3450 3451 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3452 oia.addr16[0], ia->addr16[0], 0), 3453 oia.addr16[1], ia->addr16[1], 0); 3454 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3455 oia.addr16[0], ia->addr16[0], 0), 3456 oia.addr16[1], ia->addr16[1], 0); 3457 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3458 break; 3459 } 3460 #endif /* INET */ 3461 #ifdef INET6 3462 case AF_INET6: 3463 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3464 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3465 pf_cksum_fixup(pf_cksum_fixup(*ic, 3466 oia.addr16[0], ia->addr16[0], u), 3467 oia.addr16[1], ia->addr16[1], u), 3468 oia.addr16[2], ia->addr16[2], u), 3469 oia.addr16[3], ia->addr16[3], u), 3470 oia.addr16[4], ia->addr16[4], u), 3471 oia.addr16[5], ia->addr16[5], u), 3472 oia.addr16[6], ia->addr16[6], u), 3473 oia.addr16[7], ia->addr16[7], u); 3474 break; 3475 #endif /* INET6 */ 3476 } 3477 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3478 if (oa) { 3479 PF_ACPY(oa, na, af); 3480 switch (af) { 3481 #ifdef INET 3482 case AF_INET: 3483 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3484 ooa.addr16[0], oa->addr16[0], 0), 3485 ooa.addr16[1], oa->addr16[1], 0); 3486 break; 3487 #endif /* INET */ 3488 #ifdef INET6 3489 case AF_INET6: 3490 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3491 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3492 pf_cksum_fixup(pf_cksum_fixup(*ic, 3493 ooa.addr16[0], oa->addr16[0], u), 3494 ooa.addr16[1], oa->addr16[1], u), 3495 ooa.addr16[2], oa->addr16[2], u), 3496 ooa.addr16[3], oa->addr16[3], u), 3497 ooa.addr16[4], oa->addr16[4], u), 3498 ooa.addr16[5], oa->addr16[5], u), 3499 ooa.addr16[6], oa->addr16[6], u), 3500 ooa.addr16[7], oa->addr16[7], u); 3501 break; 3502 #endif /* INET6 */ 3503 } 3504 } 3505 } 3506 3507 int 3508 pf_translate_af(struct pf_pdesc *pd) 3509 { 3510 #if defined(INET) && defined(INET6) 3511 struct mbuf *mp; 3512 struct ip *ip4; 3513 struct ip6_hdr *ip6; 3514 struct icmp6_hdr *icmp; 3515 struct m_tag *mtag; 3516 struct pf_fragment_tag *ftag; 3517 int hlen; 3518 3519 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3520 3521 /* trim the old header */ 3522 m_adj(pd->m, pd->off); 3523 3524 /* prepend a new one */ 3525 M_PREPEND(pd->m, hlen, M_NOWAIT); 3526 if (pd->m == NULL) 3527 return (-1); 3528 3529 switch (pd->naf) { 3530 case AF_INET: 3531 ip4 = mtod(pd->m, struct ip *); 3532 bzero(ip4, hlen); 3533 ip4->ip_v = IPVERSION; 3534 ip4->ip_hl = hlen >> 2; 3535 ip4->ip_tos = pd->tos; 3536 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3537 ip_fillid(ip4, V_ip_random_id); 3538 ip4->ip_ttl = pd->ttl; 3539 ip4->ip_p = pd->proto; 3540 ip4->ip_src = pd->nsaddr.v4; 3541 ip4->ip_dst = pd->ndaddr.v4; 3542 pd->src = (struct pf_addr *)&ip4->ip_src; 3543 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3544 pd->off = sizeof(struct ip); 3545 break; 3546 case AF_INET6: 3547 ip6 = mtod(pd->m, struct ip6_hdr *); 3548 bzero(ip6, hlen); 3549 ip6->ip6_vfc = IPV6_VERSION; 3550 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3551 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3552 ip6->ip6_nxt = pd->proto; 3553 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3554 ip6->ip6_hlim = IPV6_DEFHLIM; 3555 else 3556 ip6->ip6_hlim = pd->ttl; 3557 ip6->ip6_src = pd->nsaddr.v6; 3558 ip6->ip6_dst = pd->ndaddr.v6; 3559 pd->src = (struct pf_addr *)&ip6->ip6_src; 3560 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3561 pd->off = sizeof(struct ip6_hdr); 3562 3563 /* 3564 * If we're dealing with a reassembled packet we need to adjust 3565 * the header length from the IPv4 header size to IPv6 header 3566 * size. 3567 */ 3568 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 3569 if (mtag) { 3570 ftag = (struct pf_fragment_tag *)(mtag + 1); 3571 ftag->ft_hdrlen = sizeof(*ip6); 3572 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 3573 sizeof(struct ip) + sizeof(struct ip6_frag); 3574 } 3575 break; 3576 default: 3577 return (-1); 3578 } 3579 3580 /* recalculate icmp/icmp6 checksums */ 3581 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 3582 int off; 3583 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 3584 NULL) { 3585 pd->m = NULL; 3586 return (-1); 3587 } 3588 icmp = (struct icmp6_hdr *)(mp->m_data + off); 3589 icmp->icmp6_cksum = 0; 3590 icmp->icmp6_cksum = pd->naf == AF_INET ? 3591 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 3592 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 3593 ntohs(ip6->ip6_plen)); 3594 } 3595 #endif /* INET && INET6 */ 3596 3597 return (0); 3598 } 3599 3600 int 3601 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 3602 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 3603 sa_family_t af, sa_family_t naf) 3604 { 3605 #if defined(INET) && defined(INET6) 3606 struct mbuf *n = NULL; 3607 struct ip *ip4; 3608 struct ip6_hdr *ip6; 3609 int hlen, olen, mlen; 3610 3611 if (af == naf || (af != AF_INET && af != AF_INET6) || 3612 (naf != AF_INET && naf != AF_INET6)) 3613 return (-1); 3614 3615 /* split the mbuf chain on the inner ip/ip6 header boundary */ 3616 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 3617 return (-1); 3618 3619 /* old header */ 3620 olen = pd2->off - off; 3621 /* new header */ 3622 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3623 3624 /* trim old header */ 3625 m_adj(n, olen); 3626 3627 /* prepend a new one */ 3628 M_PREPEND(n, hlen, M_NOWAIT); 3629 if (n == NULL) 3630 return (-1); 3631 3632 /* translate inner ip/ip6 header */ 3633 switch (naf) { 3634 case AF_INET: 3635 ip4 = mtod(n, struct ip *); 3636 bzero(ip4, sizeof(*ip4)); 3637 ip4->ip_v = IPVERSION; 3638 ip4->ip_hl = sizeof(*ip4) >> 2; 3639 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 3640 ip_fillid(ip4, V_ip_random_id); 3641 ip4->ip_off = htons(IP_DF); 3642 ip4->ip_ttl = pd2->ttl; 3643 if (pd2->proto == IPPROTO_ICMPV6) 3644 ip4->ip_p = IPPROTO_ICMP; 3645 else 3646 ip4->ip_p = pd2->proto; 3647 ip4->ip_src = src->v4; 3648 ip4->ip_dst = dst->v4; 3649 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 3650 break; 3651 case AF_INET6: 3652 ip6 = mtod(n, struct ip6_hdr *); 3653 bzero(ip6, sizeof(*ip6)); 3654 ip6->ip6_vfc = IPV6_VERSION; 3655 ip6->ip6_plen = htons(pd2->tot_len - olen); 3656 if (pd2->proto == IPPROTO_ICMP) 3657 ip6->ip6_nxt = IPPROTO_ICMPV6; 3658 else 3659 ip6->ip6_nxt = pd2->proto; 3660 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 3661 ip6->ip6_hlim = IPV6_DEFHLIM; 3662 else 3663 ip6->ip6_hlim = pd2->ttl; 3664 ip6->ip6_src = src->v6; 3665 ip6->ip6_dst = dst->v6; 3666 break; 3667 default: 3668 unhandled_af(naf); 3669 } 3670 3671 /* adjust payload offset and total packet length */ 3672 pd2->off += hlen - olen; 3673 pd->tot_len += hlen - olen; 3674 3675 /* merge modified inner packet with the original header */ 3676 mlen = n->m_pkthdr.len; 3677 m_cat(m, n); 3678 m->m_pkthdr.len += mlen; 3679 #endif /* INET && INET6 */ 3680 3681 return (0); 3682 } 3683 3684 #define PTR_IP(field) (offsetof(struct ip, field)) 3685 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 3686 3687 int 3688 pf_translate_icmp_af(int af, void *arg) 3689 { 3690 #if defined(INET) && defined(INET6) 3691 struct icmp *icmp4; 3692 struct icmp6_hdr *icmp6; 3693 u_int32_t mtu; 3694 int32_t ptr = -1; 3695 u_int8_t type; 3696 u_int8_t code; 3697 3698 switch (af) { 3699 case AF_INET: 3700 icmp6 = arg; 3701 type = icmp6->icmp6_type; 3702 code = icmp6->icmp6_code; 3703 mtu = ntohl(icmp6->icmp6_mtu); 3704 3705 switch (type) { 3706 case ICMP6_ECHO_REQUEST: 3707 type = ICMP_ECHO; 3708 break; 3709 case ICMP6_ECHO_REPLY: 3710 type = ICMP_ECHOREPLY; 3711 break; 3712 case ICMP6_DST_UNREACH: 3713 type = ICMP_UNREACH; 3714 switch (code) { 3715 case ICMP6_DST_UNREACH_NOROUTE: 3716 case ICMP6_DST_UNREACH_BEYONDSCOPE: 3717 case ICMP6_DST_UNREACH_ADDR: 3718 code = ICMP_UNREACH_HOST; 3719 break; 3720 case ICMP6_DST_UNREACH_ADMIN: 3721 code = ICMP_UNREACH_HOST_PROHIB; 3722 break; 3723 case ICMP6_DST_UNREACH_NOPORT: 3724 code = ICMP_UNREACH_PORT; 3725 break; 3726 default: 3727 return (-1); 3728 } 3729 break; 3730 case ICMP6_PACKET_TOO_BIG: 3731 type = ICMP_UNREACH; 3732 code = ICMP_UNREACH_NEEDFRAG; 3733 mtu -= 20; 3734 break; 3735 case ICMP6_TIME_EXCEEDED: 3736 type = ICMP_TIMXCEED; 3737 break; 3738 case ICMP6_PARAM_PROB: 3739 switch (code) { 3740 case ICMP6_PARAMPROB_HEADER: 3741 type = ICMP_PARAMPROB; 3742 code = ICMP_PARAMPROB_ERRATPTR; 3743 ptr = ntohl(icmp6->icmp6_pptr); 3744 3745 if (ptr == PTR_IP6(ip6_vfc)) 3746 ; /* preserve */ 3747 else if (ptr == PTR_IP6(ip6_vfc) + 1) 3748 ptr = PTR_IP(ip_tos); 3749 else if (ptr == PTR_IP6(ip6_plen) || 3750 ptr == PTR_IP6(ip6_plen) + 1) 3751 ptr = PTR_IP(ip_len); 3752 else if (ptr == PTR_IP6(ip6_nxt)) 3753 ptr = PTR_IP(ip_p); 3754 else if (ptr == PTR_IP6(ip6_hlim)) 3755 ptr = PTR_IP(ip_ttl); 3756 else if (ptr >= PTR_IP6(ip6_src) && 3757 ptr < PTR_IP6(ip6_dst)) 3758 ptr = PTR_IP(ip_src); 3759 else if (ptr >= PTR_IP6(ip6_dst) && 3760 ptr < sizeof(struct ip6_hdr)) 3761 ptr = PTR_IP(ip_dst); 3762 else { 3763 return (-1); 3764 } 3765 break; 3766 case ICMP6_PARAMPROB_NEXTHEADER: 3767 type = ICMP_UNREACH; 3768 code = ICMP_UNREACH_PROTOCOL; 3769 break; 3770 default: 3771 return (-1); 3772 } 3773 break; 3774 default: 3775 return (-1); 3776 } 3777 if (icmp6->icmp6_type != type) { 3778 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3779 icmp6->icmp6_type, type, 0); 3780 icmp6->icmp6_type = type; 3781 } 3782 if (icmp6->icmp6_code != code) { 3783 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3784 icmp6->icmp6_code, code, 0); 3785 icmp6->icmp6_code = code; 3786 } 3787 if (icmp6->icmp6_mtu != htonl(mtu)) { 3788 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3789 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 3790 /* aligns well with a icmpv4 nextmtu */ 3791 icmp6->icmp6_mtu = htonl(mtu); 3792 } 3793 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 3794 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 3795 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 3796 /* icmpv4 pptr is a one most significant byte */ 3797 icmp6->icmp6_pptr = htonl(ptr << 24); 3798 } 3799 break; 3800 case AF_INET6: 3801 icmp4 = arg; 3802 type = icmp4->icmp_type; 3803 code = icmp4->icmp_code; 3804 mtu = ntohs(icmp4->icmp_nextmtu); 3805 3806 switch (type) { 3807 case ICMP_ECHO: 3808 type = ICMP6_ECHO_REQUEST; 3809 break; 3810 case ICMP_ECHOREPLY: 3811 type = ICMP6_ECHO_REPLY; 3812 break; 3813 case ICMP_UNREACH: 3814 type = ICMP6_DST_UNREACH; 3815 switch (code) { 3816 case ICMP_UNREACH_NET: 3817 case ICMP_UNREACH_HOST: 3818 case ICMP_UNREACH_NET_UNKNOWN: 3819 case ICMP_UNREACH_HOST_UNKNOWN: 3820 case ICMP_UNREACH_ISOLATED: 3821 case ICMP_UNREACH_TOSNET: 3822 case ICMP_UNREACH_TOSHOST: 3823 code = ICMP6_DST_UNREACH_NOROUTE; 3824 break; 3825 case ICMP_UNREACH_PORT: 3826 code = ICMP6_DST_UNREACH_NOPORT; 3827 break; 3828 case ICMP_UNREACH_NET_PROHIB: 3829 case ICMP_UNREACH_HOST_PROHIB: 3830 case ICMP_UNREACH_FILTER_PROHIB: 3831 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 3832 code = ICMP6_DST_UNREACH_ADMIN; 3833 break; 3834 case ICMP_UNREACH_PROTOCOL: 3835 type = ICMP6_PARAM_PROB; 3836 code = ICMP6_PARAMPROB_NEXTHEADER; 3837 ptr = offsetof(struct ip6_hdr, ip6_nxt); 3838 break; 3839 case ICMP_UNREACH_NEEDFRAG: 3840 type = ICMP6_PACKET_TOO_BIG; 3841 code = 0; 3842 mtu += 20; 3843 break; 3844 default: 3845 return (-1); 3846 } 3847 break; 3848 case ICMP_TIMXCEED: 3849 type = ICMP6_TIME_EXCEEDED; 3850 break; 3851 case ICMP_PARAMPROB: 3852 type = ICMP6_PARAM_PROB; 3853 switch (code) { 3854 case ICMP_PARAMPROB_ERRATPTR: 3855 code = ICMP6_PARAMPROB_HEADER; 3856 break; 3857 case ICMP_PARAMPROB_LENGTH: 3858 code = ICMP6_PARAMPROB_HEADER; 3859 break; 3860 default: 3861 return (-1); 3862 } 3863 3864 ptr = icmp4->icmp_pptr; 3865 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 3866 ; /* preserve */ 3867 else if (ptr == PTR_IP(ip_len) || 3868 ptr == PTR_IP(ip_len) + 1) 3869 ptr = PTR_IP6(ip6_plen); 3870 else if (ptr == PTR_IP(ip_ttl)) 3871 ptr = PTR_IP6(ip6_hlim); 3872 else if (ptr == PTR_IP(ip_p)) 3873 ptr = PTR_IP6(ip6_nxt); 3874 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 3875 ptr = PTR_IP6(ip6_src); 3876 else if (ptr >= PTR_IP(ip_dst) && 3877 ptr < sizeof(struct ip)) 3878 ptr = PTR_IP6(ip6_dst); 3879 else { 3880 return (-1); 3881 } 3882 break; 3883 default: 3884 return (-1); 3885 } 3886 if (icmp4->icmp_type != type) { 3887 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3888 icmp4->icmp_type, type, 0); 3889 icmp4->icmp_type = type; 3890 } 3891 if (icmp4->icmp_code != code) { 3892 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3893 icmp4->icmp_code, code, 0); 3894 icmp4->icmp_code = code; 3895 } 3896 if (icmp4->icmp_nextmtu != htons(mtu)) { 3897 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3898 icmp4->icmp_nextmtu, htons(mtu), 0); 3899 icmp4->icmp_nextmtu = htons(mtu); 3900 } 3901 if (ptr >= 0 && icmp4->icmp_void != ptr) { 3902 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 3903 htons(icmp4->icmp_pptr), htons(ptr), 0); 3904 icmp4->icmp_void = htonl(ptr); 3905 } 3906 break; 3907 default: 3908 unhandled_af(af); 3909 } 3910 #endif /* INET && INET6 */ 3911 3912 return (0); 3913 } 3914 3915 /* 3916 * Need to modulate the sequence numbers in the TCP SACK option 3917 * (credits to Krzysztof Pfaff for report and patch) 3918 */ 3919 static int 3920 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3921 struct pf_state_peer *dst) 3922 { 3923 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3924 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3925 int copyback = 0, i, olen; 3926 struct sackblk sack; 3927 3928 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3929 if (hlen < TCPOLEN_SACKLEN || hlen > MAX_TCPOPTLEN || 3930 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3931 return 0; 3932 3933 while (hlen >= TCPOLEN_SACKLEN) { 3934 size_t startoff = opt - opts; 3935 olen = opt[1]; 3936 switch (*opt) { 3937 case TCPOPT_EOL: /* FALLTHROUGH */ 3938 case TCPOPT_NOP: 3939 opt++; 3940 hlen--; 3941 break; 3942 case TCPOPT_SACK: 3943 if (olen > hlen) 3944 olen = hlen; 3945 if (olen >= TCPOLEN_SACKLEN) { 3946 for (i = 2; i + TCPOLEN_SACK <= olen; 3947 i += TCPOLEN_SACK) { 3948 memcpy(&sack, &opt[i], sizeof(sack)); 3949 pf_patch_32_unaligned(pd->m, 3950 &th->th_sum, &sack.start, 3951 htonl(ntohl(sack.start) - dst->seqdiff), 3952 PF_ALGNMNT(startoff), 3953 0); 3954 pf_patch_32_unaligned(pd->m, &th->th_sum, 3955 &sack.end, 3956 htonl(ntohl(sack.end) - dst->seqdiff), 3957 PF_ALGNMNT(startoff), 3958 0); 3959 memcpy(&opt[i], &sack, sizeof(sack)); 3960 } 3961 copyback = 1; 3962 } 3963 /* FALLTHROUGH */ 3964 default: 3965 if (olen < 2) 3966 olen = 2; 3967 hlen -= olen; 3968 opt += olen; 3969 } 3970 } 3971 3972 if (copyback) 3973 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3974 return (copyback); 3975 } 3976 3977 struct mbuf * 3978 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3979 const struct pf_addr *saddr, const struct pf_addr *daddr, 3980 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3981 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3982 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3983 { 3984 struct mbuf *m; 3985 int len, tlen; 3986 #ifdef INET 3987 struct ip *h = NULL; 3988 #endif /* INET */ 3989 #ifdef INET6 3990 struct ip6_hdr *h6 = NULL; 3991 #endif /* INET6 */ 3992 struct tcphdr *th; 3993 char *opt; 3994 struct pf_mtag *pf_mtag; 3995 3996 len = 0; 3997 th = NULL; 3998 3999 /* maximum segment size tcp option */ 4000 tlen = sizeof(struct tcphdr); 4001 if (mss) 4002 tlen += 4; 4003 4004 switch (af) { 4005 #ifdef INET 4006 case AF_INET: 4007 len = sizeof(struct ip) + tlen; 4008 break; 4009 #endif /* INET */ 4010 #ifdef INET6 4011 case AF_INET6: 4012 len = sizeof(struct ip6_hdr) + tlen; 4013 break; 4014 #endif /* INET6 */ 4015 default: 4016 unhandled_af(af); 4017 } 4018 4019 m = m_gethdr(M_NOWAIT, MT_DATA); 4020 if (m == NULL) 4021 return (NULL); 4022 4023 #ifdef MAC 4024 mac_netinet_firewall_send(m); 4025 #endif 4026 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 4027 m_freem(m); 4028 return (NULL); 4029 } 4030 m->m_flags |= mbuf_flags; 4031 pf_mtag->tag = mtag_tag; 4032 pf_mtag->flags = mtag_flags; 4033 4034 if (rtableid >= 0) 4035 M_SETFIB(m, rtableid); 4036 4037 #ifdef ALTQ 4038 if (r != NULL && r->qid) { 4039 pf_mtag->qid = r->qid; 4040 4041 /* add hints for ecn */ 4042 pf_mtag->hdr = mtod(m, struct ip *); 4043 } 4044 #endif /* ALTQ */ 4045 m->m_data += max_linkhdr; 4046 m->m_pkthdr.len = m->m_len = len; 4047 /* The rest of the stack assumes a rcvif, so provide one. 4048 * This is a locally generated packet, so .. close enough. */ 4049 m->m_pkthdr.rcvif = V_loif; 4050 bzero(m->m_data, len); 4051 switch (af) { 4052 #ifdef INET 4053 case AF_INET: 4054 m->m_pkthdr.csum_flags |= CSUM_TCP; 4055 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4056 4057 h = mtod(m, struct ip *); 4058 4059 h->ip_p = IPPROTO_TCP; 4060 h->ip_len = htons(tlen); 4061 h->ip_v = 4; 4062 h->ip_hl = sizeof(*h) >> 2; 4063 h->ip_tos = IPTOS_LOWDELAY; 4064 h->ip_len = htons(len); 4065 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4066 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4067 h->ip_sum = 0; 4068 h->ip_src.s_addr = saddr->v4.s_addr; 4069 h->ip_dst.s_addr = daddr->v4.s_addr; 4070 4071 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4072 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4073 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4074 break; 4075 #endif /* INET */ 4076 #ifdef INET6 4077 case AF_INET6: 4078 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4079 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4080 4081 h6 = mtod(m, struct ip6_hdr *); 4082 4083 /* IP header fields included in the TCP checksum */ 4084 h6->ip6_nxt = IPPROTO_TCP; 4085 h6->ip6_plen = htons(tlen); 4086 h6->ip6_vfc |= IPV6_VERSION; 4087 h6->ip6_hlim = V_ip6_defhlim; 4088 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4089 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4090 4091 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4092 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4093 IPPROTO_TCP, 0); 4094 break; 4095 #endif /* INET6 */ 4096 } 4097 4098 /* TCP header */ 4099 th->th_sport = sport; 4100 th->th_dport = dport; 4101 th->th_seq = htonl(seq); 4102 th->th_ack = htonl(ack); 4103 th->th_off = tlen >> 2; 4104 tcp_set_flags(th, tcp_flags); 4105 th->th_win = htons(win); 4106 4107 if (mss) { 4108 opt = (char *)(th + 1); 4109 opt[0] = TCPOPT_MAXSEG; 4110 opt[1] = 4; 4111 mss = htons(mss); 4112 memcpy((opt + 2), &mss, 2); 4113 } 4114 4115 return (m); 4116 } 4117 4118 static void 4119 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4120 uint8_t ttl, int rtableid) 4121 { 4122 struct mbuf *m; 4123 #ifdef INET 4124 struct ip *h = NULL; 4125 #endif /* INET */ 4126 #ifdef INET6 4127 struct ip6_hdr *h6 = NULL; 4128 #endif /* INET6 */ 4129 struct sctphdr *hdr; 4130 struct sctp_chunkhdr *chunk; 4131 struct pf_send_entry *pfse; 4132 int off = 0; 4133 4134 MPASS(af == pd->af); 4135 4136 m = m_gethdr(M_NOWAIT, MT_DATA); 4137 if (m == NULL) 4138 return; 4139 4140 m->m_data += max_linkhdr; 4141 m->m_flags |= M_SKIP_FIREWALL; 4142 /* The rest of the stack assumes a rcvif, so provide one. 4143 * This is a locally generated packet, so .. close enough. */ 4144 m->m_pkthdr.rcvif = V_loif; 4145 4146 /* IPv4|6 header */ 4147 switch (af) { 4148 #ifdef INET 4149 case AF_INET: 4150 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4151 4152 h = mtod(m, struct ip *); 4153 4154 /* IP header fields included in the TCP checksum */ 4155 4156 h->ip_p = IPPROTO_SCTP; 4157 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4158 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4159 h->ip_src = pd->dst->v4; 4160 h->ip_dst = pd->src->v4; 4161 4162 off += sizeof(struct ip); 4163 break; 4164 #endif /* INET */ 4165 #ifdef INET6 4166 case AF_INET6: 4167 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4168 4169 h6 = mtod(m, struct ip6_hdr *); 4170 4171 /* IP header fields included in the TCP checksum */ 4172 h6->ip6_vfc |= IPV6_VERSION; 4173 h6->ip6_nxt = IPPROTO_SCTP; 4174 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4175 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4176 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4177 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4178 4179 off += sizeof(struct ip6_hdr); 4180 break; 4181 #endif /* INET6 */ 4182 default: 4183 unhandled_af(af); 4184 } 4185 4186 /* SCTP header */ 4187 hdr = mtodo(m, off); 4188 4189 hdr->src_port = pd->hdr.sctp.dest_port; 4190 hdr->dest_port = pd->hdr.sctp.src_port; 4191 hdr->v_tag = pd->sctp_initiate_tag; 4192 hdr->checksum = 0; 4193 4194 /* Abort chunk. */ 4195 off += sizeof(struct sctphdr); 4196 chunk = mtodo(m, off); 4197 4198 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4199 chunk->chunk_length = htons(sizeof(*chunk)); 4200 4201 /* SCTP checksum */ 4202 off += sizeof(*chunk); 4203 m->m_pkthdr.len = m->m_len = off; 4204 4205 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4206 4207 if (rtableid >= 0) 4208 M_SETFIB(m, rtableid); 4209 4210 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4211 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4212 if (pfse == NULL) { 4213 m_freem(m); 4214 return; 4215 } 4216 4217 switch (af) { 4218 #ifdef INET 4219 case AF_INET: 4220 pfse->pfse_type = PFSE_IP; 4221 break; 4222 #endif /* INET */ 4223 #ifdef INET6 4224 case AF_INET6: 4225 pfse->pfse_type = PFSE_IP6; 4226 break; 4227 #endif /* INET6 */ 4228 } 4229 4230 pfse->pfse_m = m; 4231 pf_send(pfse); 4232 } 4233 4234 void 4235 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4236 const struct pf_addr *saddr, const struct pf_addr *daddr, 4237 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4238 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4239 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 4240 { 4241 struct pf_send_entry *pfse; 4242 struct mbuf *m; 4243 4244 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4245 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 4246 if (m == NULL) 4247 return; 4248 4249 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4250 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4251 if (pfse == NULL) { 4252 m_freem(m); 4253 return; 4254 } 4255 4256 switch (af) { 4257 #ifdef INET 4258 case AF_INET: 4259 pfse->pfse_type = PFSE_IP; 4260 break; 4261 #endif /* INET */ 4262 #ifdef INET6 4263 case AF_INET6: 4264 pfse->pfse_type = PFSE_IP6; 4265 break; 4266 #endif /* INET6 */ 4267 default: 4268 unhandled_af(af); 4269 } 4270 4271 pfse->pfse_m = m; 4272 pf_send(pfse); 4273 } 4274 4275 static void 4276 pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum) 4277 { 4278 /* undo NAT changes, if they have taken place */ 4279 if (nr != NULL) { 4280 PF_ACPY(pd->src, &pd->osrc, pd->af); 4281 PF_ACPY(pd->dst, &pd->odst, pd->af); 4282 if (pd->sport) 4283 *pd->sport = pd->osport; 4284 if (pd->dport) 4285 *pd->dport = pd->odport; 4286 if (pd->ip_sum) 4287 *pd->ip_sum = bip_sum; 4288 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4289 } 4290 } 4291 4292 static void 4293 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4294 struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum, 4295 u_short *reason, int rtableid) 4296 { 4297 pf_undo_nat(nr, pd, bip_sum); 4298 4299 if (pd->proto == IPPROTO_TCP && 4300 ((r->rule_flag & PFRULE_RETURNRST) || 4301 (r->rule_flag & PFRULE_RETURN)) && 4302 !(tcp_get_flags(th) & TH_RST)) { 4303 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4304 4305 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4306 IPPROTO_TCP, pd->af)) 4307 REASON_SET(reason, PFRES_PROTCKSUM); 4308 else { 4309 if (tcp_get_flags(th) & TH_SYN) 4310 ack++; 4311 if (tcp_get_flags(th) & TH_FIN) 4312 ack++; 4313 pf_send_tcp(r, pd->af, pd->dst, 4314 pd->src, th->th_dport, th->th_sport, 4315 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4316 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 4317 } 4318 } else if (pd->proto == IPPROTO_SCTP && 4319 (r->rule_flag & PFRULE_RETURN)) { 4320 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4321 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4322 r->return_icmp) 4323 pf_send_icmp(pd->m, r->return_icmp >> 8, 4324 r->return_icmp & 255, pd->af, r, rtableid); 4325 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4326 r->return_icmp6) 4327 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4328 r->return_icmp6 & 255, pd->af, r, rtableid); 4329 } 4330 4331 static int 4332 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4333 { 4334 struct m_tag *mtag; 4335 u_int8_t mpcp; 4336 4337 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4338 if (mtag == NULL) 4339 return (0); 4340 4341 if (prio == PF_PRIO_ZERO) 4342 prio = 0; 4343 4344 mpcp = *(uint8_t *)(mtag + 1); 4345 4346 return (mpcp == prio); 4347 } 4348 4349 static int 4350 pf_icmp_to_bandlim(uint8_t type) 4351 { 4352 switch (type) { 4353 case ICMP_ECHO: 4354 case ICMP_ECHOREPLY: 4355 return (BANDLIM_ICMP_ECHO); 4356 case ICMP_TSTAMP: 4357 case ICMP_TSTAMPREPLY: 4358 return (BANDLIM_ICMP_TSTAMP); 4359 case ICMP_UNREACH: 4360 default: 4361 return (BANDLIM_ICMP_UNREACH); 4362 } 4363 } 4364 4365 static void 4366 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 4367 struct pf_krule *r, int rtableid) 4368 { 4369 struct pf_send_entry *pfse; 4370 struct mbuf *m0; 4371 struct pf_mtag *pf_mtag; 4372 4373 /* ICMP packet rate limitation. */ 4374 switch (af) { 4375 #ifdef INET6 4376 case AF_INET6: 4377 if (icmp6_ratelimit(NULL, type, code)) 4378 return; 4379 break; 4380 #endif /* INET6 */ 4381 #ifdef INET 4382 case AF_INET: 4383 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4384 return; 4385 break; 4386 #endif /* INET */ 4387 } 4388 4389 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4390 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4391 if (pfse == NULL) 4392 return; 4393 4394 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4395 free(pfse, M_PFTEMP); 4396 return; 4397 } 4398 4399 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4400 free(pfse, M_PFTEMP); 4401 return; 4402 } 4403 /* XXX: revisit */ 4404 m0->m_flags |= M_SKIP_FIREWALL; 4405 4406 if (rtableid >= 0) 4407 M_SETFIB(m0, rtableid); 4408 4409 #ifdef ALTQ 4410 if (r->qid) { 4411 pf_mtag->qid = r->qid; 4412 /* add hints for ecn */ 4413 pf_mtag->hdr = mtod(m0, struct ip *); 4414 } 4415 #endif /* ALTQ */ 4416 4417 switch (af) { 4418 #ifdef INET 4419 case AF_INET: 4420 pfse->pfse_type = PFSE_ICMP; 4421 break; 4422 #endif /* INET */ 4423 #ifdef INET6 4424 case AF_INET6: 4425 pfse->pfse_type = PFSE_ICMP6; 4426 break; 4427 #endif /* INET6 */ 4428 } 4429 pfse->pfse_m = m0; 4430 pfse->icmpopts.type = type; 4431 pfse->icmpopts.code = code; 4432 pf_send(pfse); 4433 } 4434 4435 /* 4436 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 4437 * If n is 0, they match if they are equal. If n is != 0, they match if they 4438 * are different. 4439 */ 4440 int 4441 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4442 const struct pf_addr *b, sa_family_t af) 4443 { 4444 int match = 0; 4445 4446 switch (af) { 4447 #ifdef INET 4448 case AF_INET: 4449 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4450 match++; 4451 break; 4452 #endif /* INET */ 4453 #ifdef INET6 4454 case AF_INET6: 4455 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4456 match++; 4457 break; 4458 #endif /* INET6 */ 4459 } 4460 if (match) { 4461 if (n) 4462 return (0); 4463 else 4464 return (1); 4465 } else { 4466 if (n) 4467 return (1); 4468 else 4469 return (0); 4470 } 4471 } 4472 4473 /* 4474 * Return 1 if b <= a <= e, otherwise return 0. 4475 */ 4476 int 4477 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4478 const struct pf_addr *a, sa_family_t af) 4479 { 4480 switch (af) { 4481 #ifdef INET 4482 case AF_INET: 4483 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4484 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4485 return (0); 4486 break; 4487 #endif /* INET */ 4488 #ifdef INET6 4489 case AF_INET6: { 4490 int i; 4491 4492 /* check a >= b */ 4493 for (i = 0; i < 4; ++i) 4494 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4495 break; 4496 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4497 return (0); 4498 /* check a <= e */ 4499 for (i = 0; i < 4; ++i) 4500 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4501 break; 4502 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4503 return (0); 4504 break; 4505 } 4506 #endif /* INET6 */ 4507 } 4508 return (1); 4509 } 4510 4511 static int 4512 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4513 { 4514 switch (op) { 4515 case PF_OP_IRG: 4516 return ((p > a1) && (p < a2)); 4517 case PF_OP_XRG: 4518 return ((p < a1) || (p > a2)); 4519 case PF_OP_RRG: 4520 return ((p >= a1) && (p <= a2)); 4521 case PF_OP_EQ: 4522 return (p == a1); 4523 case PF_OP_NE: 4524 return (p != a1); 4525 case PF_OP_LT: 4526 return (p < a1); 4527 case PF_OP_LE: 4528 return (p <= a1); 4529 case PF_OP_GT: 4530 return (p > a1); 4531 case PF_OP_GE: 4532 return (p >= a1); 4533 } 4534 return (0); /* never reached */ 4535 } 4536 4537 int 4538 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4539 { 4540 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 4541 } 4542 4543 static int 4544 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 4545 { 4546 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4547 return (0); 4548 return (pf_match(op, a1, a2, u)); 4549 } 4550 4551 static int 4552 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 4553 { 4554 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 4555 return (0); 4556 return (pf_match(op, a1, a2, g)); 4557 } 4558 4559 int 4560 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 4561 { 4562 if (*tag == -1) 4563 *tag = mtag; 4564 4565 return ((!r->match_tag_not && r->match_tag == *tag) || 4566 (r->match_tag_not && r->match_tag != *tag)); 4567 } 4568 4569 static int 4570 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 4571 { 4572 struct ifnet *ifp = m->m_pkthdr.rcvif; 4573 struct pfi_kkif *kif; 4574 4575 if (ifp == NULL) 4576 return (0); 4577 4578 kif = (struct pfi_kkif *)ifp->if_pf_kif; 4579 4580 if (kif == NULL) { 4581 DPFPRINTF(PF_DEBUG_URGENT, 4582 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 4583 r->rcv_ifname)); 4584 return (0); 4585 } 4586 4587 return (pfi_kkif_match(r->rcv_kif, kif)); 4588 } 4589 4590 int 4591 pf_tag_packet(struct pf_pdesc *pd, int tag) 4592 { 4593 4594 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 4595 4596 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 4597 return (ENOMEM); 4598 4599 pd->pf_mtag->tag = tag; 4600 4601 return (0); 4602 } 4603 4604 #define PF_ANCHOR_STACKSIZE 32 4605 struct pf_kanchor_stackframe { 4606 struct pf_kruleset *rs; 4607 struct pf_krule *r; /* XXX: + match bit */ 4608 struct pf_kanchor *child; 4609 }; 4610 4611 /* 4612 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4613 */ 4614 #define PF_ANCHORSTACK_MATCH 0x00000001 4615 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4616 4617 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4618 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4619 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4620 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4621 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4622 } while (0) 4623 4624 void 4625 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4626 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a) 4627 { 4628 struct pf_kanchor_stackframe *f; 4629 4630 PF_RULES_RASSERT(); 4631 4632 if (*depth >= PF_ANCHOR_STACKSIZE) { 4633 printf("%s: anchor stack overflow on %s\n", 4634 __func__, (*r)->anchor->name); 4635 *r = TAILQ_NEXT(*r, entries); 4636 return; 4637 } else if (*depth == 0 && a != NULL) 4638 *a = *r; 4639 f = stack + (*depth)++; 4640 f->rs = *rs; 4641 f->r = *r; 4642 if ((*r)->anchor_wildcard) { 4643 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4644 4645 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4646 *r = NULL; 4647 return; 4648 } 4649 *rs = &f->child->ruleset; 4650 } else { 4651 f->child = NULL; 4652 *rs = &(*r)->anchor->ruleset; 4653 } 4654 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4655 } 4656 4657 int 4658 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4659 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4660 int *match) 4661 { 4662 struct pf_kanchor_stackframe *f; 4663 struct pf_krule *fr; 4664 int quick = 0; 4665 4666 PF_RULES_RASSERT(); 4667 4668 do { 4669 if (*depth <= 0) 4670 break; 4671 f = stack + *depth - 1; 4672 fr = PF_ANCHOR_RULE(f); 4673 if (f->child != NULL) { 4674 f->child = RB_NEXT(pf_kanchor_node, 4675 &fr->anchor->children, f->child); 4676 if (f->child != NULL) { 4677 *rs = &f->child->ruleset; 4678 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4679 if (*r == NULL) 4680 continue; 4681 else 4682 break; 4683 } 4684 } 4685 (*depth)--; 4686 if (*depth == 0 && a != NULL) 4687 *a = NULL; 4688 *rs = f->rs; 4689 if (match != NULL && *match > *depth) { 4690 *match = *depth; 4691 if (f->r->quick) 4692 quick = 1; 4693 } 4694 *r = TAILQ_NEXT(fr, entries); 4695 } while (*r == NULL); 4696 4697 return (quick); 4698 } 4699 4700 struct pf_keth_anchor_stackframe { 4701 struct pf_keth_ruleset *rs; 4702 struct pf_keth_rule *r; /* XXX: + match bit */ 4703 struct pf_keth_anchor *child; 4704 }; 4705 4706 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4707 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4708 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4709 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4710 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4711 } while (0) 4712 4713 void 4714 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4715 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4716 struct pf_keth_rule **a, int *match) 4717 { 4718 struct pf_keth_anchor_stackframe *f; 4719 4720 NET_EPOCH_ASSERT(); 4721 4722 if (match) 4723 *match = 0; 4724 if (*depth >= PF_ANCHOR_STACKSIZE) { 4725 printf("%s: anchor stack overflow on %s\n", 4726 __func__, (*r)->anchor->name); 4727 *r = TAILQ_NEXT(*r, entries); 4728 return; 4729 } else if (*depth == 0 && a != NULL) 4730 *a = *r; 4731 f = stack + (*depth)++; 4732 f->rs = *rs; 4733 f->r = *r; 4734 if ((*r)->anchor_wildcard) { 4735 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4736 4737 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4738 *r = NULL; 4739 return; 4740 } 4741 *rs = &f->child->ruleset; 4742 } else { 4743 f->child = NULL; 4744 *rs = &(*r)->anchor->ruleset; 4745 } 4746 *r = TAILQ_FIRST((*rs)->active.rules); 4747 } 4748 4749 int 4750 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4751 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4752 struct pf_keth_rule **a, int *match) 4753 { 4754 struct pf_keth_anchor_stackframe *f; 4755 struct pf_keth_rule *fr; 4756 int quick = 0; 4757 4758 NET_EPOCH_ASSERT(); 4759 4760 do { 4761 if (*depth <= 0) 4762 break; 4763 f = stack + *depth - 1; 4764 fr = PF_ETH_ANCHOR_RULE(f); 4765 if (f->child != NULL) { 4766 /* 4767 * This block traverses through 4768 * a wildcard anchor. 4769 */ 4770 if (match != NULL && *match) { 4771 /* 4772 * If any of "*" matched, then 4773 * "foo/ *" matched, mark frame 4774 * appropriately. 4775 */ 4776 PF_ETH_ANCHOR_SET_MATCH(f); 4777 *match = 0; 4778 } 4779 f->child = RB_NEXT(pf_keth_anchor_node, 4780 &fr->anchor->children, f->child); 4781 if (f->child != NULL) { 4782 *rs = &f->child->ruleset; 4783 *r = TAILQ_FIRST((*rs)->active.rules); 4784 if (*r == NULL) 4785 continue; 4786 else 4787 break; 4788 } 4789 } 4790 (*depth)--; 4791 if (*depth == 0 && a != NULL) 4792 *a = NULL; 4793 *rs = f->rs; 4794 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4795 quick = fr->quick; 4796 *r = TAILQ_NEXT(fr, entries); 4797 } while (*r == NULL); 4798 4799 return (quick); 4800 } 4801 4802 #ifdef INET6 4803 void 4804 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4805 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4806 { 4807 switch (af) { 4808 #ifdef INET 4809 case AF_INET: 4810 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4811 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4812 break; 4813 #endif /* INET */ 4814 case AF_INET6: 4815 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4816 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4817 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4818 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4819 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4820 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4821 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4822 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4823 break; 4824 } 4825 } 4826 4827 void 4828 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4829 { 4830 switch (af) { 4831 #ifdef INET 4832 case AF_INET: 4833 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4834 break; 4835 #endif /* INET */ 4836 case AF_INET6: 4837 if (addr->addr32[3] == 0xffffffff) { 4838 addr->addr32[3] = 0; 4839 if (addr->addr32[2] == 0xffffffff) { 4840 addr->addr32[2] = 0; 4841 if (addr->addr32[1] == 0xffffffff) { 4842 addr->addr32[1] = 0; 4843 addr->addr32[0] = 4844 htonl(ntohl(addr->addr32[0]) + 1); 4845 } else 4846 addr->addr32[1] = 4847 htonl(ntohl(addr->addr32[1]) + 1); 4848 } else 4849 addr->addr32[2] = 4850 htonl(ntohl(addr->addr32[2]) + 1); 4851 } else 4852 addr->addr32[3] = 4853 htonl(ntohl(addr->addr32[3]) + 1); 4854 break; 4855 } 4856 } 4857 #endif /* INET6 */ 4858 4859 void 4860 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4861 { 4862 /* 4863 * Modern rules use the same flags in rules as they do in states. 4864 */ 4865 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4866 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4867 4868 /* 4869 * Old-style scrub rules have different flags which need to be translated. 4870 */ 4871 if (r->rule_flag & PFRULE_RANDOMID) 4872 a->flags |= PFSTATE_RANDOMID; 4873 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4874 a->flags |= PFSTATE_SETTOS; 4875 a->set_tos = r->set_tos; 4876 } 4877 4878 if (r->qid) 4879 a->qid = r->qid; 4880 if (r->pqid) 4881 a->pqid = r->pqid; 4882 if (r->rtableid >= 0) 4883 a->rtableid = r->rtableid; 4884 a->log |= r->log; 4885 if (r->min_ttl) 4886 a->min_ttl = r->min_ttl; 4887 if (r->max_mss) 4888 a->max_mss = r->max_mss; 4889 if (r->dnpipe) 4890 a->dnpipe = r->dnpipe; 4891 if (r->dnrpipe) 4892 a->dnrpipe = r->dnrpipe; 4893 if (r->dnpipe || r->dnrpipe) { 4894 if (r->free_flags & PFRULE_DN_IS_PIPE) 4895 a->flags |= PFSTATE_DN_IS_PIPE; 4896 else 4897 a->flags &= ~PFSTATE_DN_IS_PIPE; 4898 } 4899 if (r->scrub_flags & PFSTATE_SETPRIO) { 4900 a->set_prio[0] = r->set_prio[0]; 4901 a->set_prio[1] = r->set_prio[1]; 4902 } 4903 if (r->allow_opts) 4904 a->allow_opts = r->allow_opts; 4905 } 4906 4907 int 4908 pf_socket_lookup(struct pf_pdesc *pd) 4909 { 4910 struct pf_addr *saddr, *daddr; 4911 u_int16_t sport, dport; 4912 struct inpcbinfo *pi; 4913 struct inpcb *inp; 4914 4915 pd->lookup.uid = UID_MAX; 4916 pd->lookup.gid = GID_MAX; 4917 4918 switch (pd->proto) { 4919 case IPPROTO_TCP: 4920 sport = pd->hdr.tcp.th_sport; 4921 dport = pd->hdr.tcp.th_dport; 4922 pi = &V_tcbinfo; 4923 break; 4924 case IPPROTO_UDP: 4925 sport = pd->hdr.udp.uh_sport; 4926 dport = pd->hdr.udp.uh_dport; 4927 pi = &V_udbinfo; 4928 break; 4929 default: 4930 return (-1); 4931 } 4932 if (pd->dir == PF_IN) { 4933 saddr = pd->src; 4934 daddr = pd->dst; 4935 } else { 4936 u_int16_t p; 4937 4938 p = sport; 4939 sport = dport; 4940 dport = p; 4941 saddr = pd->dst; 4942 daddr = pd->src; 4943 } 4944 switch (pd->af) { 4945 #ifdef INET 4946 case AF_INET: 4947 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4948 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4949 if (inp == NULL) { 4950 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4951 daddr->v4, dport, INPLOOKUP_WILDCARD | 4952 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4953 if (inp == NULL) 4954 return (-1); 4955 } 4956 break; 4957 #endif /* INET */ 4958 #ifdef INET6 4959 case AF_INET6: 4960 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4961 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4962 if (inp == NULL) { 4963 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4964 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4965 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4966 if (inp == NULL) 4967 return (-1); 4968 } 4969 break; 4970 #endif /* INET6 */ 4971 default: 4972 unhandled_af(pd->af); 4973 } 4974 INP_RLOCK_ASSERT(inp); 4975 pd->lookup.uid = inp->inp_cred->cr_uid; 4976 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4977 INP_RUNLOCK(inp); 4978 4979 return (1); 4980 } 4981 4982 u_int8_t 4983 pf_get_wscale(struct pf_pdesc *pd) 4984 { 4985 struct tcphdr *th = &pd->hdr.tcp; 4986 int hlen; 4987 u_int8_t hdr[60]; 4988 u_int8_t *opt, optlen; 4989 u_int8_t wscale = 0; 4990 4991 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4992 if (hlen <= sizeof(struct tcphdr)) 4993 return (0); 4994 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4995 return (0); 4996 opt = hdr + sizeof(struct tcphdr); 4997 hlen -= sizeof(struct tcphdr); 4998 while (hlen >= 3) { 4999 switch (*opt) { 5000 case TCPOPT_EOL: 5001 case TCPOPT_NOP: 5002 ++opt; 5003 --hlen; 5004 break; 5005 case TCPOPT_WINDOW: 5006 wscale = opt[2]; 5007 if (wscale > TCP_MAX_WINSHIFT) 5008 wscale = TCP_MAX_WINSHIFT; 5009 wscale |= PF_WSCALE_FLAG; 5010 /* FALLTHROUGH */ 5011 default: 5012 optlen = opt[1]; 5013 if (optlen < 2) 5014 optlen = 2; 5015 hlen -= optlen; 5016 opt += optlen; 5017 break; 5018 } 5019 } 5020 return (wscale); 5021 } 5022 5023 u_int16_t 5024 pf_get_mss(struct pf_pdesc *pd) 5025 { 5026 struct tcphdr *th = &pd->hdr.tcp; 5027 int hlen; 5028 u_int8_t hdr[60]; 5029 u_int8_t *opt, optlen; 5030 u_int16_t mss = V_tcp_mssdflt; 5031 5032 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 5033 if (hlen <= sizeof(struct tcphdr)) 5034 return (0); 5035 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 5036 return (0); 5037 opt = hdr + sizeof(struct tcphdr); 5038 hlen -= sizeof(struct tcphdr); 5039 while (hlen >= TCPOLEN_MAXSEG) { 5040 switch (*opt) { 5041 case TCPOPT_EOL: 5042 case TCPOPT_NOP: 5043 ++opt; 5044 --hlen; 5045 break; 5046 case TCPOPT_MAXSEG: 5047 memcpy(&mss, (opt + 2), 2); 5048 mss = ntohs(mss); 5049 /* FALLTHROUGH */ 5050 default: 5051 optlen = opt[1]; 5052 if (optlen < 2) 5053 optlen = 2; 5054 hlen -= optlen; 5055 opt += optlen; 5056 break; 5057 } 5058 } 5059 return (mss); 5060 } 5061 5062 static u_int16_t 5063 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5064 { 5065 struct nhop_object *nh; 5066 #ifdef INET6 5067 struct in6_addr dst6; 5068 uint32_t scopeid; 5069 #endif /* INET6 */ 5070 int hlen = 0; 5071 uint16_t mss = 0; 5072 5073 NET_EPOCH_ASSERT(); 5074 5075 switch (af) { 5076 #ifdef INET 5077 case AF_INET: 5078 hlen = sizeof(struct ip); 5079 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5080 if (nh != NULL) 5081 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5082 break; 5083 #endif /* INET */ 5084 #ifdef INET6 5085 case AF_INET6: 5086 hlen = sizeof(struct ip6_hdr); 5087 in6_splitscope(&addr->v6, &dst6, &scopeid); 5088 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5089 if (nh != NULL) 5090 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5091 break; 5092 #endif /* INET6 */ 5093 } 5094 5095 mss = max(V_tcp_mssdflt, mss); 5096 mss = min(mss, offer); 5097 mss = max(mss, 64); /* sanity - at least max opt space */ 5098 return (mss); 5099 } 5100 5101 static u_int32_t 5102 pf_tcp_iss(struct pf_pdesc *pd) 5103 { 5104 SHA512_CTX ctx; 5105 union { 5106 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5107 uint32_t words[1]; 5108 } digest; 5109 5110 if (V_pf_tcp_secret_init == 0) { 5111 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5112 SHA512_Init(&V_pf_tcp_secret_ctx); 5113 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5114 sizeof(V_pf_tcp_secret)); 5115 V_pf_tcp_secret_init = 1; 5116 } 5117 5118 ctx = V_pf_tcp_secret_ctx; 5119 5120 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5121 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5122 switch (pd->af) { 5123 case AF_INET6: 5124 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5125 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5126 break; 5127 case AF_INET: 5128 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5129 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5130 break; 5131 } 5132 SHA512_Final(digest.bytes, &ctx); 5133 V_pf_tcp_iss_off += 4096; 5134 #define ISN_RANDOM_INCREMENT (4096 - 1) 5135 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5136 V_pf_tcp_iss_off); 5137 #undef ISN_RANDOM_INCREMENT 5138 } 5139 5140 static bool 5141 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5142 { 5143 bool match = true; 5144 5145 /* Always matches if not set */ 5146 if (! r->isset) 5147 return (!r->neg); 5148 5149 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5150 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5151 match = false; 5152 break; 5153 } 5154 } 5155 5156 return (match ^ r->neg); 5157 } 5158 5159 static int 5160 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5161 { 5162 if (*tag == -1) 5163 *tag = mtag; 5164 5165 return ((!r->match_tag_not && r->match_tag == *tag) || 5166 (r->match_tag_not && r->match_tag != *tag)); 5167 } 5168 5169 static void 5170 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5171 { 5172 /* If we don't have the interface drop the packet. */ 5173 if (ifp == NULL) { 5174 m_freem(m); 5175 return; 5176 } 5177 5178 switch (ifp->if_type) { 5179 case IFT_ETHER: 5180 case IFT_XETHER: 5181 case IFT_L2VLAN: 5182 case IFT_BRIDGE: 5183 case IFT_IEEE8023ADLAG: 5184 break; 5185 default: 5186 m_freem(m); 5187 return; 5188 } 5189 5190 ifp->if_transmit(ifp, m); 5191 } 5192 5193 static int 5194 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5195 { 5196 #ifdef INET 5197 struct ip ip; 5198 #endif /* INET */ 5199 #ifdef INET6 5200 struct ip6_hdr ip6; 5201 #endif /* INET6 */ 5202 struct mbuf *m = *m0; 5203 struct ether_header *e; 5204 struct pf_keth_rule *r, *rm, *a = NULL; 5205 struct pf_keth_ruleset *ruleset = NULL; 5206 struct pf_mtag *mtag; 5207 struct pf_keth_ruleq *rules; 5208 struct pf_addr *src = NULL, *dst = NULL; 5209 struct pfi_kkif *bridge_to; 5210 sa_family_t af = 0; 5211 uint16_t proto; 5212 int asd = 0, match = 0; 5213 int tag = -1; 5214 uint8_t action; 5215 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5216 5217 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5218 NET_EPOCH_ASSERT(); 5219 5220 PF_RULES_RLOCK_TRACKER; 5221 5222 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5223 5224 mtag = pf_find_mtag(m); 5225 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5226 /* Dummynet re-injects packets after they've 5227 * completed their delay. We've already 5228 * processed them, so pass unconditionally. */ 5229 5230 /* But only once. We may see the packet multiple times (e.g. 5231 * PFIL_IN/PFIL_OUT). */ 5232 pf_dummynet_flag_remove(m, mtag); 5233 5234 return (PF_PASS); 5235 } 5236 5237 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5238 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5239 DPFPRINTF(PF_DEBUG_URGENT, 5240 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 5241 ", pullup failed\n")); 5242 return (PF_DROP); 5243 } 5244 e = mtod(m, struct ether_header *); 5245 proto = ntohs(e->ether_type); 5246 5247 switch (proto) { 5248 #ifdef INET 5249 case ETHERTYPE_IP: { 5250 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5251 sizeof(ip))) 5252 return (PF_DROP); 5253 5254 af = AF_INET; 5255 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5256 (caddr_t)&ip); 5257 src = (struct pf_addr *)&ip.ip_src; 5258 dst = (struct pf_addr *)&ip.ip_dst; 5259 break; 5260 } 5261 #endif /* INET */ 5262 #ifdef INET6 5263 case ETHERTYPE_IPV6: { 5264 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5265 sizeof(ip6))) 5266 return (PF_DROP); 5267 5268 af = AF_INET6; 5269 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5270 (caddr_t)&ip6); 5271 src = (struct pf_addr *)&ip6.ip6_src; 5272 dst = (struct pf_addr *)&ip6.ip6_dst; 5273 break; 5274 } 5275 #endif /* INET6 */ 5276 } 5277 5278 PF_RULES_RLOCK(); 5279 5280 ruleset = V_pf_keth; 5281 rules = atomic_load_ptr(&ruleset->active.rules); 5282 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5283 counter_u64_add(r->evaluations, 1); 5284 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5285 5286 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5287 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5288 "kif"); 5289 r = r->skip[PFE_SKIP_IFP].ptr; 5290 } 5291 else if (r->direction && r->direction != dir) { 5292 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5293 "dir"); 5294 r = r->skip[PFE_SKIP_DIR].ptr; 5295 } 5296 else if (r->proto && r->proto != proto) { 5297 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5298 "proto"); 5299 r = r->skip[PFE_SKIP_PROTO].ptr; 5300 } 5301 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5302 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5303 "src"); 5304 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5305 } 5306 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5307 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5308 "dst"); 5309 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5310 } 5311 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5312 r->ipsrc.neg, kif, M_GETFIB(m))) { 5313 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5314 "ip_src"); 5315 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5316 } 5317 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5318 r->ipdst.neg, kif, M_GETFIB(m))) { 5319 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5320 "ip_dst"); 5321 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5322 } 5323 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5324 mtag ? mtag->tag : 0)) { 5325 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5326 "match_tag"); 5327 r = TAILQ_NEXT(r, entries); 5328 } 5329 else { 5330 if (r->tag) 5331 tag = r->tag; 5332 if (r->anchor == NULL) { 5333 /* Rule matches */ 5334 rm = r; 5335 5336 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5337 5338 if (r->quick) 5339 break; 5340 5341 r = TAILQ_NEXT(r, entries); 5342 } else { 5343 pf_step_into_keth_anchor(anchor_stack, &asd, 5344 &ruleset, &r, &a, &match); 5345 } 5346 } 5347 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5348 &ruleset, &r, &a, &match)) 5349 break; 5350 } 5351 5352 r = rm; 5353 5354 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5355 5356 /* Default to pass. */ 5357 if (r == NULL) { 5358 PF_RULES_RUNLOCK(); 5359 return (PF_PASS); 5360 } 5361 5362 /* Execute action. */ 5363 counter_u64_add(r->packets[dir == PF_OUT], 1); 5364 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5365 pf_update_timestamp(r); 5366 5367 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5368 if (r->action == PF_DROP) { 5369 PF_RULES_RUNLOCK(); 5370 return (PF_DROP); 5371 } 5372 5373 if (tag > 0) { 5374 if (mtag == NULL) 5375 mtag = pf_get_mtag(m); 5376 if (mtag == NULL) { 5377 PF_RULES_RUNLOCK(); 5378 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5379 return (PF_DROP); 5380 } 5381 mtag->tag = tag; 5382 } 5383 5384 if (r->qid != 0) { 5385 if (mtag == NULL) 5386 mtag = pf_get_mtag(m); 5387 if (mtag == NULL) { 5388 PF_RULES_RUNLOCK(); 5389 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5390 return (PF_DROP); 5391 } 5392 mtag->qid = r->qid; 5393 } 5394 5395 action = r->action; 5396 bridge_to = r->bridge_to; 5397 5398 /* Dummynet */ 5399 if (r->dnpipe) { 5400 struct ip_fw_args dnflow; 5401 5402 /* Drop packet if dummynet is not loaded. */ 5403 if (ip_dn_io_ptr == NULL) { 5404 PF_RULES_RUNLOCK(); 5405 m_freem(m); 5406 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5407 return (PF_DROP); 5408 } 5409 if (mtag == NULL) 5410 mtag = pf_get_mtag(m); 5411 if (mtag == NULL) { 5412 PF_RULES_RUNLOCK(); 5413 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5414 return (PF_DROP); 5415 } 5416 5417 bzero(&dnflow, sizeof(dnflow)); 5418 5419 /* We don't have port numbers here, so we set 0. That means 5420 * that we'll be somewhat limited in distinguishing flows (i.e. 5421 * only based on IP addresses, not based on port numbers), but 5422 * it's better than nothing. */ 5423 dnflow.f_id.dst_port = 0; 5424 dnflow.f_id.src_port = 0; 5425 dnflow.f_id.proto = 0; 5426 5427 dnflow.rule.info = r->dnpipe; 5428 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5429 if (r->dnflags & PFRULE_DN_IS_PIPE) 5430 dnflow.rule.info |= IPFW_IS_PIPE; 5431 5432 dnflow.f_id.extra = dnflow.rule.info; 5433 5434 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5435 dnflow.flags |= IPFW_ARGS_ETHER; 5436 dnflow.ifp = kif->pfik_ifp; 5437 5438 switch (af) { 5439 case AF_INET: 5440 dnflow.f_id.addr_type = 4; 5441 dnflow.f_id.src_ip = src->v4.s_addr; 5442 dnflow.f_id.dst_ip = dst->v4.s_addr; 5443 break; 5444 case AF_INET6: 5445 dnflow.flags |= IPFW_ARGS_IP6; 5446 dnflow.f_id.addr_type = 6; 5447 dnflow.f_id.src_ip6 = src->v6; 5448 dnflow.f_id.dst_ip6 = dst->v6; 5449 break; 5450 } 5451 5452 PF_RULES_RUNLOCK(); 5453 5454 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5455 ip_dn_io_ptr(m0, &dnflow); 5456 if (*m0 != NULL) 5457 pf_dummynet_flag_remove(m, mtag); 5458 } else { 5459 PF_RULES_RUNLOCK(); 5460 } 5461 5462 if (action == PF_PASS && bridge_to) { 5463 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5464 *m0 = NULL; /* We've eaten the packet. */ 5465 } 5466 5467 return (action); 5468 } 5469 5470 #define PF_TEST_ATTRIB(t, a)\ 5471 do { \ 5472 if (t) { \ 5473 r = a; \ 5474 goto nextrule; \ 5475 } \ 5476 } while (0) 5477 5478 static int 5479 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 5480 struct pf_pdesc *pd, struct pf_krule **am, 5481 struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp) 5482 { 5483 struct pf_krule *nr = NULL; 5484 struct pf_krule *r, *a = NULL; 5485 struct pf_kruleset *ruleset = NULL; 5486 struct pf_krule_slist match_rules; 5487 struct pf_krule_item *ri; 5488 struct tcphdr *th = &pd->hdr.tcp; 5489 struct pf_state_key *sk = NULL, *nk = NULL; 5490 u_short transerror; 5491 int rewrite = 0; 5492 int tag = -1; 5493 int asd = 0; 5494 int match = 0; 5495 int state_icmp = 0, icmp_dir; 5496 int action = PF_PASS; 5497 u_int16_t virtual_type, virtual_id; 5498 u_int16_t bproto_sum = 0, bip_sum = 0; 5499 u_int8_t icmptype = 0, icmpcode = 0; 5500 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5501 struct pf_udp_mapping *udp_mapping = NULL; 5502 5503 PF_RULES_RASSERT(); 5504 5505 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5506 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5507 5508 SLIST_INIT(&match_rules); 5509 5510 if (inp != NULL) { 5511 INP_LOCK_ASSERT(inp); 5512 pd->lookup.uid = inp->inp_cred->cr_uid; 5513 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 5514 pd->lookup.done = 1; 5515 } 5516 5517 switch (pd->virtual_proto) { 5518 case IPPROTO_TCP: 5519 pd->nsport = th->th_sport; 5520 pd->ndport = th->th_dport; 5521 break; 5522 case IPPROTO_UDP: 5523 pd->nsport = pd->hdr.udp.uh_sport; 5524 pd->ndport = pd->hdr.udp.uh_dport; 5525 break; 5526 case IPPROTO_SCTP: 5527 pd->nsport = pd->hdr.sctp.src_port; 5528 pd->ndport = pd->hdr.sctp.dest_port; 5529 break; 5530 #ifdef INET 5531 case IPPROTO_ICMP: 5532 MPASS(pd->af == AF_INET); 5533 icmptype = pd->hdr.icmp.icmp_type; 5534 icmpcode = pd->hdr.icmp.icmp_code; 5535 state_icmp = pf_icmp_mapping(pd, icmptype, 5536 &icmp_dir, &virtual_id, &virtual_type); 5537 if (icmp_dir == PF_IN) { 5538 pd->nsport = virtual_id; 5539 pd->ndport = virtual_type; 5540 } else { 5541 pd->nsport = virtual_type; 5542 pd->ndport = virtual_id; 5543 } 5544 break; 5545 #endif /* INET */ 5546 #ifdef INET6 5547 case IPPROTO_ICMPV6: 5548 MPASS(pd->af == AF_INET6); 5549 icmptype = pd->hdr.icmp6.icmp6_type; 5550 icmpcode = pd->hdr.icmp6.icmp6_code; 5551 state_icmp = pf_icmp_mapping(pd, icmptype, 5552 &icmp_dir, &virtual_id, &virtual_type); 5553 if (icmp_dir == PF_IN) { 5554 pd->nsport = virtual_id; 5555 pd->ndport = virtual_type; 5556 } else { 5557 pd->nsport = virtual_type; 5558 pd->ndport = virtual_id; 5559 } 5560 5561 break; 5562 #endif /* INET6 */ 5563 default: 5564 pd->nsport = pd->ndport = 0; 5565 break; 5566 } 5567 pd->osport = pd->nsport; 5568 pd->odport = pd->ndport; 5569 5570 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5571 5572 /* check packet for BINAT/NAT/RDR */ 5573 transerror = pf_get_translation(pd, pd->off, &sk, &nk, anchor_stack, 5574 &nr, &udp_mapping); 5575 switch (transerror) { 5576 default: 5577 /* A translation error occurred. */ 5578 REASON_SET(reason, transerror); 5579 goto cleanup; 5580 case PFRES_MAX: 5581 /* No match. */ 5582 break; 5583 case PFRES_MATCH: 5584 KASSERT(sk != NULL, ("%s: null sk", __func__)); 5585 KASSERT(nk != NULL, ("%s: null nk", __func__)); 5586 5587 if (nr->log) { 5588 PFLOG_PACKET(nr->action, PFRES_MATCH, nr, a, 5589 ruleset, pd, 1, NULL); 5590 } 5591 5592 if (pd->ip_sum) 5593 bip_sum = *pd->ip_sum; 5594 5595 switch (pd->proto) { 5596 case IPPROTO_TCP: 5597 bproto_sum = th->th_sum; 5598 5599 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5600 nk->port[pd->sidx] != pd->nsport) { 5601 pf_change_ap(pd, pd->src, &th->th_sport, 5602 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5603 nk->port[pd->sidx], 0, pd->af, pd->naf); 5604 pd->sport = &th->th_sport; 5605 pd->nsport = th->th_sport; 5606 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5607 } 5608 5609 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5610 nk->port[pd->didx] != pd->ndport) { 5611 pf_change_ap(pd, pd->dst, &th->th_dport, 5612 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5613 nk->port[pd->didx], 0, pd->af, pd->naf); 5614 pd->dport = &th->th_dport; 5615 pd->ndport = th->th_dport; 5616 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5617 } 5618 rewrite++; 5619 break; 5620 case IPPROTO_UDP: 5621 bproto_sum = pd->hdr.udp.uh_sum; 5622 5623 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5624 nk->port[pd->sidx] != pd->nsport) { 5625 pf_change_ap(pd, pd->src, 5626 &pd->hdr.udp.uh_sport, 5627 pd->ip_sum, &pd->hdr.udp.uh_sum, 5628 &nk->addr[pd->sidx], 5629 nk->port[pd->sidx], 1, pd->af, pd->naf); 5630 pd->sport = &pd->hdr.udp.uh_sport; 5631 pd->nsport = pd->hdr.udp.uh_sport; 5632 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5633 } 5634 5635 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5636 nk->port[pd->didx] != pd->ndport) { 5637 pf_change_ap(pd, pd->dst, 5638 &pd->hdr.udp.uh_dport, 5639 pd->ip_sum, &pd->hdr.udp.uh_sum, 5640 &nk->addr[pd->didx], 5641 nk->port[pd->didx], 1, pd->af, pd->naf); 5642 pd->dport = &pd->hdr.udp.uh_dport; 5643 pd->ndport = pd->hdr.udp.uh_dport; 5644 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5645 } 5646 rewrite++; 5647 break; 5648 case IPPROTO_SCTP: { 5649 uint16_t checksum = 0; 5650 5651 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 5652 nk->port[pd->sidx] != pd->nsport) { 5653 pf_change_ap(pd, pd->src, 5654 &pd->hdr.sctp.src_port, pd->ip_sum, &checksum, 5655 &nk->addr[pd->sidx], 5656 nk->port[pd->sidx], 1, pd->af, pd->naf); 5657 pd->sport = &pd->hdr.sctp.src_port; 5658 pd->nsport = pd->hdr.sctp.src_port; 5659 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5660 } 5661 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 5662 nk->port[pd->didx] != pd->ndport) { 5663 pf_change_ap(pd, pd->dst, 5664 &pd->hdr.sctp.dest_port, pd->ip_sum, &checksum, 5665 &nk->addr[pd->didx], 5666 nk->port[pd->didx], 1, pd->af, pd->naf); 5667 pd->dport = &pd->hdr.sctp.dest_port; 5668 pd->ndport = pd->hdr.sctp.dest_port; 5669 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5670 } 5671 break; 5672 } 5673 #ifdef INET 5674 case IPPROTO_ICMP: 5675 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 5676 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 5677 nk->addr[pd->sidx].v4.s_addr, 0); 5678 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5679 } 5680 5681 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 5682 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 5683 nk->addr[pd->didx].v4.s_addr, 0); 5684 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5685 } 5686 5687 if (virtual_type == htons(ICMP_ECHO) && 5688 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5689 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5690 pd->hdr.icmp.icmp_cksum, pd->nsport, 5691 nk->port[pd->sidx], 0); 5692 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5693 pd->sport = &pd->hdr.icmp.icmp_id; 5694 } 5695 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5696 break; 5697 #endif /* INET */ 5698 #ifdef INET6 5699 case IPPROTO_ICMPV6: 5700 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 5701 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 5702 &nk->addr[pd->sidx], 0); 5703 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5704 } 5705 5706 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 5707 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 5708 &nk->addr[pd->didx], 0); 5709 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5710 } 5711 rewrite++; 5712 break; 5713 #endif /* INET */ 5714 default: 5715 switch (pd->af) { 5716 #ifdef INET 5717 case AF_INET: 5718 if (PF_ANEQ(&pd->nsaddr, 5719 &nk->addr[pd->sidx], AF_INET)) { 5720 pf_change_a(&pd->src->v4.s_addr, 5721 pd->ip_sum, 5722 nk->addr[pd->sidx].v4.s_addr, 0); 5723 PF_ACPY(&pd->nsaddr, pd->src, pd->af); 5724 } 5725 5726 if (PF_ANEQ(&pd->ndaddr, 5727 &nk->addr[pd->didx], AF_INET)) { 5728 pf_change_a(&pd->dst->v4.s_addr, 5729 pd->ip_sum, 5730 nk->addr[pd->didx].v4.s_addr, 0); 5731 PF_ACPY(&pd->ndaddr, pd->dst, pd->af); 5732 } 5733 break; 5734 #endif /* INET */ 5735 #ifdef INET6 5736 case AF_INET6: 5737 if (PF_ANEQ(&pd->nsaddr, 5738 &nk->addr[pd->sidx], AF_INET6)) { 5739 PF_ACPY(&pd->nsaddr, &nk->addr[pd->sidx], pd->af); 5740 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5741 } 5742 5743 if (PF_ANEQ(&pd->ndaddr, 5744 &nk->addr[pd->didx], AF_INET6)) { 5745 PF_ACPY(&pd->ndaddr, &nk->addr[pd->didx], pd->af); 5746 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5747 } 5748 break; 5749 #endif /* INET6 */ 5750 } 5751 break; 5752 } 5753 if (nr->natpass) 5754 r = NULL; 5755 } 5756 5757 while (r != NULL) { 5758 if (pd->related_rule) { 5759 *rm = pd->related_rule; 5760 break; 5761 } 5762 pf_counter_u64_add(&r->evaluations, 1); 5763 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5764 r->skip[PF_SKIP_IFP]); 5765 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5766 r->skip[PF_SKIP_DIR]); 5767 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5768 r->skip[PF_SKIP_AF]); 5769 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5770 r->skip[PF_SKIP_PROTO]); 5771 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5772 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5773 r->skip[PF_SKIP_SRC_ADDR]); 5774 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5775 r->dst.neg, NULL, M_GETFIB(pd->m)), 5776 r->skip[PF_SKIP_DST_ADDR]); 5777 switch (pd->virtual_proto) { 5778 case PF_VPROTO_FRAGMENT: 5779 /* tcp/udp only. port_op always 0 in other cases */ 5780 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5781 TAILQ_NEXT(r, entries)); 5782 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5783 TAILQ_NEXT(r, entries)); 5784 /* icmp only. type/code always 0 in other cases */ 5785 PF_TEST_ATTRIB((r->type || r->code), 5786 TAILQ_NEXT(r, entries)); 5787 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5788 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5789 TAILQ_NEXT(r, entries)); 5790 break; 5791 5792 case IPPROTO_TCP: 5793 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(th)) != r->flags, 5794 TAILQ_NEXT(r, entries)); 5795 /* FALLTHROUGH */ 5796 case IPPROTO_SCTP: 5797 case IPPROTO_UDP: 5798 /* tcp/udp only. port_op always 0 in other cases */ 5799 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5800 r->src.port[0], r->src.port[1], pd->nsport), 5801 r->skip[PF_SKIP_SRC_PORT]); 5802 /* tcp/udp only. port_op always 0 in other cases */ 5803 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5804 r->dst.port[0], r->dst.port[1], pd->ndport), 5805 r->skip[PF_SKIP_DST_PORT]); 5806 /* tcp/udp only. uid.op always 0 in other cases */ 5807 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5808 pf_socket_lookup(pd), 1)) && 5809 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5810 pd->lookup.uid), 5811 TAILQ_NEXT(r, entries)); 5812 /* tcp/udp only. gid.op always 0 in other cases */ 5813 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5814 pf_socket_lookup(pd), 1)) && 5815 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5816 pd->lookup.gid), 5817 TAILQ_NEXT(r, entries)); 5818 break; 5819 5820 case IPPROTO_ICMP: 5821 case IPPROTO_ICMPV6: 5822 /* icmp only. type always 0 in other cases */ 5823 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5824 TAILQ_NEXT(r, entries)); 5825 /* icmp only. type always 0 in other cases */ 5826 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5827 TAILQ_NEXT(r, entries)); 5828 break; 5829 5830 default: 5831 break; 5832 } 5833 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5834 TAILQ_NEXT(r, entries)); 5835 PF_TEST_ATTRIB(r->prio && 5836 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5837 TAILQ_NEXT(r, entries)); 5838 PF_TEST_ATTRIB(r->prob && 5839 r->prob <= arc4random(), 5840 TAILQ_NEXT(r, entries)); 5841 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5842 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5843 TAILQ_NEXT(r, entries)); 5844 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 5845 r->rcvifnot), 5846 TAILQ_NEXT(r, entries)); 5847 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5848 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5849 TAILQ_NEXT(r, entries)); 5850 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5851 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5852 pf_osfp_fingerprint(pd, th), 5853 r->os_fingerprint)), 5854 TAILQ_NEXT(r, entries)); 5855 /* FALLTHROUGH */ 5856 if (r->tag) 5857 tag = r->tag; 5858 if (r->anchor == NULL) { 5859 if (r->action == PF_MATCH) { 5860 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5861 if (ri == NULL) { 5862 REASON_SET(reason, PFRES_MEMORY); 5863 goto cleanup; 5864 } 5865 ri->r = r; 5866 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5867 pf_counter_u64_critical_enter(); 5868 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5869 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5870 pf_counter_u64_critical_exit(); 5871 pf_rule_to_actions(r, &pd->act); 5872 if (r->rule_flag & PFRULE_AFTO) 5873 pd->naf = r->naf; 5874 if (pd->af != pd->naf) { 5875 if (pf_get_transaddr_af(r, pd) == -1) { 5876 REASON_SET(reason, PFRES_TRANSLATE); 5877 goto cleanup; 5878 } 5879 } 5880 if (r->log) 5881 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5882 a, ruleset, pd, 1, NULL); 5883 } else { 5884 match = asd; 5885 *rm = r; 5886 *am = a; 5887 *rsm = ruleset; 5888 } 5889 if (pd->act.log & PF_LOG_MATCHES) 5890 pf_log_matches(pd, r, a, ruleset, &match_rules); 5891 if (r->quick) 5892 break; 5893 r = TAILQ_NEXT(r, entries); 5894 } else 5895 pf_step_into_anchor(anchor_stack, &asd, 5896 &ruleset, PF_RULESET_FILTER, &r, &a); 5897 nextrule: 5898 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5899 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5900 break; 5901 } 5902 r = *rm; 5903 a = *am; 5904 ruleset = *rsm; 5905 5906 REASON_SET(reason, PFRES_MATCH); 5907 5908 /* apply actions for last matching pass/block rule */ 5909 pf_rule_to_actions(r, &pd->act); 5910 if (r->rule_flag & PFRULE_AFTO) 5911 pd->naf = r->naf; 5912 if (pd->af != pd->naf) { 5913 if (pf_get_transaddr_af(r, pd) == -1) { 5914 REASON_SET(reason, PFRES_TRANSLATE); 5915 goto cleanup; 5916 } 5917 } 5918 5919 if (r->log) { 5920 if (rewrite) 5921 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5922 PFLOG_PACKET(r->action, *reason, r, a, ruleset, pd, 1, NULL); 5923 } 5924 if (pd->act.log & PF_LOG_MATCHES) 5925 pf_log_matches(pd, r, a, ruleset, &match_rules); 5926 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5927 (r->action == PF_DROP) && 5928 ((r->rule_flag & PFRULE_RETURNRST) || 5929 (r->rule_flag & PFRULE_RETURNICMP) || 5930 (r->rule_flag & PFRULE_RETURN))) { 5931 pf_return(r, nr, pd, th, bproto_sum, 5932 bip_sum, reason, r->rtableid); 5933 } 5934 5935 if (r->action == PF_DROP) 5936 goto cleanup; 5937 5938 if (tag > 0 && pf_tag_packet(pd, tag)) { 5939 REASON_SET(reason, PFRES_MEMORY); 5940 goto cleanup; 5941 } 5942 if (pd->act.rtableid >= 0) 5943 M_SETFIB(pd->m, pd->act.rtableid); 5944 5945 if (r->rt) { 5946 struct pf_ksrc_node *sn = NULL; 5947 struct pf_srchash *snh = NULL; 5948 struct pf_kpool *pool = &r->route; 5949 5950 /* Backwards compatibility. */ 5951 if (TAILQ_EMPTY(&pool->list)) 5952 pool = &r->rdr; 5953 5954 /* 5955 * Set act.rt here instead of in pf_rule_to_actions() because 5956 * it is applied only from the last pass rule. 5957 */ 5958 pd->act.rt = r->rt; 5959 /* Don't use REASON_SET, pf_map_addr increases the reason counters */ 5960 *reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, 5961 &pd->act.rt_kif, NULL, &sn, &snh, pool, PF_SN_ROUTE); 5962 if (*reason != 0) 5963 goto cleanup; 5964 } 5965 5966 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5967 (!state_icmp && (r->keep_state || nr != NULL || 5968 (pd->flags & PFDESC_TCP_NORM)))) { 5969 bool nat64; 5970 5971 action = pf_create_state(r, nr, a, pd, nk, sk, 5972 &rewrite, sm, tag, bproto_sum, bip_sum, 5973 &match_rules, udp_mapping); 5974 sk = nk = NULL; 5975 if (action != PF_PASS) { 5976 pf_udp_mapping_release(udp_mapping); 5977 pd->act.log |= PF_LOG_FORCE; 5978 if (action == PF_DROP && 5979 (r->rule_flag & PFRULE_RETURN)) 5980 pf_return(r, nr, pd, th, 5981 bproto_sum, bip_sum, reason, 5982 pd->act.rtableid); 5983 return (action); 5984 } 5985 5986 nat64 = pd->af != pd->naf; 5987 if (nat64) { 5988 int ret; 5989 5990 if (sk == NULL) 5991 sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 5992 if (nk == NULL) 5993 nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 5994 5995 if (pd->dir == PF_IN) { 5996 ret = pf_translate(pd, &sk->addr[pd->didx], 5997 sk->port[pd->didx], &sk->addr[pd->sidx], 5998 sk->port[pd->sidx], virtual_type, 5999 icmp_dir); 6000 } else { 6001 ret = pf_translate(pd, &sk->addr[pd->sidx], 6002 sk->port[pd->sidx], &sk->addr[pd->didx], 6003 sk->port[pd->didx], virtual_type, 6004 icmp_dir); 6005 } 6006 6007 if (ret < 0) 6008 goto cleanup; 6009 6010 rewrite += ret; 6011 6012 if (rewrite && sk->af != nk->af) 6013 action = PF_AFRT; 6014 } 6015 } else { 6016 while ((ri = SLIST_FIRST(&match_rules))) { 6017 SLIST_REMOVE_HEAD(&match_rules, entry); 6018 free(ri, M_PF_RULE_ITEM); 6019 } 6020 6021 uma_zfree(V_pf_state_key_z, sk); 6022 uma_zfree(V_pf_state_key_z, nk); 6023 sk = nk = NULL; 6024 pf_udp_mapping_release(udp_mapping); 6025 } 6026 6027 /* copy back packet headers if we performed NAT operations */ 6028 if (rewrite) 6029 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6030 6031 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 6032 pd->dir == PF_OUT && 6033 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 6034 /* 6035 * We want the state created, but we dont 6036 * want to send this in case a partner 6037 * firewall has to know about it to allow 6038 * replies through it. 6039 */ 6040 return (PF_DEFER); 6041 6042 return (action); 6043 6044 cleanup: 6045 while ((ri = SLIST_FIRST(&match_rules))) { 6046 SLIST_REMOVE_HEAD(&match_rules, entry); 6047 free(ri, M_PF_RULE_ITEM); 6048 } 6049 6050 uma_zfree(V_pf_state_key_z, sk); 6051 uma_zfree(V_pf_state_key_z, nk); 6052 pf_udp_mapping_release(udp_mapping); 6053 6054 return (PF_DROP); 6055 } 6056 6057 static int 6058 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 6059 struct pf_pdesc *pd, struct pf_state_key *nk, struct pf_state_key *sk, 6060 int *rewrite, struct pf_kstate **sm, int tag, u_int16_t bproto_sum, 6061 u_int16_t bip_sum, struct pf_krule_slist *match_rules, 6062 struct pf_udp_mapping *udp_mapping) 6063 { 6064 struct pf_kstate *s = NULL; 6065 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6066 /* 6067 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6068 * but for PF_SN_NAT it is different. Don't try optimizing it, 6069 * just store all 3 hashes. 6070 */ 6071 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6072 struct tcphdr *th = &pd->hdr.tcp; 6073 u_int16_t mss = V_tcp_mssdflt; 6074 u_short reason, sn_reason; 6075 struct pf_krule_item *ri; 6076 struct pf_kpool *pool_route = &r->route; 6077 6078 /* check maximums */ 6079 if (r->max_states && 6080 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6081 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6082 REASON_SET(&reason, PFRES_MAXSTATES); 6083 goto csfailed; 6084 } 6085 /* src node for limits */ 6086 if ((r->rule_flag & PFRULE_SRCTRACK) && 6087 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6088 NULL, NULL, PF_SN_LIMIT)) != 0) { 6089 REASON_SET(&reason, sn_reason); 6090 goto csfailed; 6091 } 6092 /* src node for route-to rule */ 6093 if (TAILQ_EMPTY(&pool_route->list)) /* Backwards compatibility. */ 6094 pool_route = &r->rdr; 6095 if ((pool_route->opts & PF_POOL_STICKYADDR) && 6096 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6097 &pd->act.rt_addr, pd->act.rt_kif, PF_SN_ROUTE)) != 0) { 6098 REASON_SET(&reason, sn_reason); 6099 goto csfailed; 6100 } 6101 /* src node for translation rule */ 6102 if (nr != NULL && (nr->rdr.opts & PF_POOL_STICKYADDR) && 6103 (sn_reason = pf_insert_src_node(sns, snhs, nr, &sk->addr[pd->sidx], 6104 pd->af, &nk->addr[1], NULL, PF_SN_NAT)) != 0 ) { 6105 REASON_SET(&reason, sn_reason); 6106 goto csfailed; 6107 } 6108 s = pf_alloc_state(M_NOWAIT); 6109 if (s == NULL) { 6110 REASON_SET(&reason, PFRES_MEMORY); 6111 goto csfailed; 6112 } 6113 s->rule = r; 6114 s->nat_rule = nr; 6115 s->anchor = a; 6116 memcpy(&s->match_rules, match_rules, sizeof(s->match_rules)); 6117 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6118 6119 if (pd->act.allow_opts) 6120 s->state_flags |= PFSTATE_ALLOWOPTS; 6121 if (r->rule_flag & PFRULE_STATESLOPPY) 6122 s->state_flags |= PFSTATE_SLOPPY; 6123 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6124 s->state_flags |= PFSTATE_SCRUB_TCP; 6125 if ((r->rule_flag & PFRULE_PFLOW) || 6126 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 6127 s->state_flags |= PFSTATE_PFLOW; 6128 6129 s->act.log = pd->act.log & PF_LOG_ALL; 6130 s->sync_state = PFSYNC_S_NONE; 6131 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6132 6133 if (nr != NULL) 6134 s->act.log |= nr->log & PF_LOG_ALL; 6135 switch (pd->proto) { 6136 case IPPROTO_TCP: 6137 s->src.seqlo = ntohl(th->th_seq); 6138 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6139 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6140 r->keep_state == PF_STATE_MODULATE) { 6141 /* Generate sequence number modulator */ 6142 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6143 0) 6144 s->src.seqdiff = 1; 6145 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6146 htonl(s->src.seqlo + s->src.seqdiff), 0); 6147 *rewrite = 1; 6148 } else 6149 s->src.seqdiff = 0; 6150 if (tcp_get_flags(th) & TH_SYN) { 6151 s->src.seqhi++; 6152 s->src.wscale = pf_get_wscale(pd); 6153 } 6154 s->src.max_win = MAX(ntohs(th->th_win), 1); 6155 if (s->src.wscale & PF_WSCALE_MASK) { 6156 /* Remove scale factor from initial window */ 6157 int win = s->src.max_win; 6158 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6159 s->src.max_win = (win - 1) >> 6160 (s->src.wscale & PF_WSCALE_MASK); 6161 } 6162 if (tcp_get_flags(th) & TH_FIN) 6163 s->src.seqhi++; 6164 s->dst.seqhi = 1; 6165 s->dst.max_win = 1; 6166 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6167 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6168 s->timeout = PFTM_TCP_FIRST_PACKET; 6169 atomic_add_32(&V_pf_status.states_halfopen, 1); 6170 break; 6171 case IPPROTO_UDP: 6172 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6173 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6174 s->timeout = PFTM_UDP_FIRST_PACKET; 6175 break; 6176 case IPPROTO_SCTP: 6177 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6178 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6179 s->timeout = PFTM_SCTP_FIRST_PACKET; 6180 break; 6181 case IPPROTO_ICMP: 6182 #ifdef INET6 6183 case IPPROTO_ICMPV6: 6184 #endif /* INET6 */ 6185 s->timeout = PFTM_ICMP_FIRST_PACKET; 6186 break; 6187 default: 6188 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6189 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6190 s->timeout = PFTM_OTHER_FIRST_PACKET; 6191 } 6192 6193 s->creation = s->expire = pf_get_uptime(); 6194 6195 if (pd->proto == IPPROTO_TCP) { 6196 if (s->state_flags & PFSTATE_SCRUB_TCP && 6197 pf_normalize_tcp_init(pd, th, &s->src)) { 6198 REASON_SET(&reason, PFRES_MEMORY); 6199 goto csfailed; 6200 } 6201 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6202 pf_normalize_tcp_stateful(pd, &reason, th, s, 6203 &s->src, &s->dst, rewrite)) { 6204 /* This really shouldn't happen!!! */ 6205 DPFPRINTF(PF_DEBUG_URGENT, 6206 ("pf_normalize_tcp_stateful failed on first " 6207 "pkt\n")); 6208 goto csfailed; 6209 } 6210 } else if (pd->proto == IPPROTO_SCTP) { 6211 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6212 goto csfailed; 6213 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6214 goto csfailed; 6215 } 6216 s->direction = pd->dir; 6217 6218 /* 6219 * sk/nk could already been setup by pf_get_translation(). 6220 */ 6221 if (nr == NULL) { 6222 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 6223 __func__, nr, sk, nk)); 6224 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6225 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6226 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, &sk, &nk)) { 6227 goto csfailed; 6228 } 6229 } else 6230 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 6231 __func__, nr, sk, nk)); 6232 6233 /* Swap sk/nk for PF_OUT. */ 6234 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6235 (pd->dir == PF_IN) ? sk : nk, 6236 (pd->dir == PF_IN) ? nk : sk, s)) { 6237 REASON_SET(&reason, PFRES_STATEINS); 6238 goto drop; 6239 } else 6240 *sm = s; 6241 sk = nk = NULL; 6242 6243 STATE_INC_COUNTERS(s); 6244 6245 /* 6246 * Lock order is important: first state, then source node. 6247 */ 6248 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6249 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6250 s->sns[sn_type] = sns[sn_type]; 6251 PF_HASHROW_UNLOCK(snhs[sn_type]); 6252 } 6253 } 6254 6255 if (tag > 0) 6256 s->tag = tag; 6257 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6258 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 6259 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6260 pf_undo_nat(nr, pd, bip_sum); 6261 s->src.seqhi = htonl(arc4random()); 6262 /* Find mss option */ 6263 int rtid = M_GETFIB(pd->m); 6264 mss = pf_get_mss(pd); 6265 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6266 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6267 s->src.mss = mss; 6268 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6269 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6270 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6271 pd->act.rtableid); 6272 REASON_SET(&reason, PFRES_SYNPROXY); 6273 return (PF_SYNPROXY_DROP); 6274 } 6275 6276 s->udp_mapping = udp_mapping; 6277 6278 return (PF_PASS); 6279 6280 csfailed: 6281 while ((ri = SLIST_FIRST(match_rules))) { 6282 SLIST_REMOVE_HEAD(match_rules, entry); 6283 free(ri, M_PF_RULE_ITEM); 6284 } 6285 6286 uma_zfree(V_pf_state_key_z, sk); 6287 uma_zfree(V_pf_state_key_z, nk); 6288 6289 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6290 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6291 if (--sns[sn_type]->states == 0 && 6292 sns[sn_type]->expire == 0) { 6293 pf_unlink_src_node(sns[sn_type]); 6294 pf_free_src_node(sns[sn_type]); 6295 counter_u64_add( 6296 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6297 } 6298 PF_HASHROW_UNLOCK(snhs[sn_type]); 6299 } 6300 } 6301 6302 drop: 6303 if (s != NULL) { 6304 pf_src_tree_remove_state(s); 6305 s->timeout = PFTM_UNLINKED; 6306 pf_free_state(s); 6307 } 6308 6309 return (PF_DROP); 6310 } 6311 6312 int 6313 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 6314 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 6315 int icmp_dir) 6316 { 6317 /* 6318 * pf_translate() implements OpenBSD's "new" NAT approach. 6319 * We don't follow it, because it involves a breaking syntax change 6320 * (removing nat/rdr rules, moving it into regular pf rules.) 6321 * It also moves NAT processing to be done after normal rules evaluation 6322 * whereas in FreeBSD that's done before rules processing. 6323 * 6324 * We adopt the function only for nat64, and keep other NAT processing 6325 * before rules processing. 6326 */ 6327 int rewrite = 0; 6328 int afto = pd->af != pd->naf; 6329 6330 MPASS(afto); 6331 6332 switch (pd->proto) { 6333 case IPPROTO_TCP: 6334 if (afto || *pd->sport != sport) { 6335 pf_change_ap(pd, pd->src, pd->sport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6336 saddr, sport, 0, pd->af, pd->naf); 6337 rewrite = 1; 6338 } 6339 if (afto || *pd->dport != dport) { 6340 pf_change_ap(pd, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.tcp.th_sum, 6341 daddr, dport, 0, pd->af, pd->naf); 6342 rewrite = 1; 6343 } 6344 break; 6345 6346 case IPPROTO_UDP: 6347 if (afto || *pd->sport != sport) { 6348 pf_change_ap(pd, pd->src, pd->sport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6349 saddr, sport, 1, pd->af, pd->naf); 6350 rewrite = 1; 6351 } 6352 if (afto || *pd->dport != dport) { 6353 pf_change_ap(pd, pd->dst, pd->dport, pd->ip_sum, &pd->hdr.udp.uh_sum, 6354 daddr, dport, 1, pd->af, pd->naf); 6355 rewrite = 1; 6356 } 6357 break; 6358 6359 case IPPROTO_SCTP: { 6360 uint16_t checksum = 0; 6361 if (afto || *pd->sport != sport) { 6362 pf_change_ap(pd, pd->src, pd->sport, pd->ip_sum, &checksum, 6363 saddr, sport, 1, pd->af, pd->naf); 6364 rewrite = 1; 6365 } 6366 if (afto || *pd->dport != dport) { 6367 pf_change_ap(pd, pd->dst, pd->dport, pd->ip_sum, &checksum, 6368 daddr, dport, 1, pd->af, pd->naf); 6369 rewrite = 1; 6370 } 6371 break; 6372 } 6373 6374 #ifdef INET 6375 case IPPROTO_ICMP: 6376 /* pf_translate() is also used when logging invalid packets */ 6377 if (pd->af != AF_INET) 6378 return (0); 6379 6380 if (afto) { 6381 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 6382 return (-1); 6383 pd->proto = IPPROTO_ICMPV6; 6384 rewrite = 1; 6385 } 6386 if (virtual_type == htons(ICMP_ECHO)) { 6387 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 6388 6389 if (icmpid != pd->hdr.icmp.icmp_id) { 6390 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 6391 pd->hdr.icmp.icmp_cksum, 6392 pd->hdr.icmp.icmp_id, icmpid, 0); 6393 pd->hdr.icmp.icmp_id = icmpid; 6394 /* XXX TODO copyback. */ 6395 rewrite = 1; 6396 } 6397 } 6398 break; 6399 #endif /* INET */ 6400 6401 #ifdef INET6 6402 case IPPROTO_ICMPV6: 6403 /* pf_translate() is also used when logging invalid packets */ 6404 if (pd->af != AF_INET6) 6405 return (0); 6406 6407 if (afto) { 6408 /* ip_sum will be recalculated in pf_translate_af */ 6409 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 6410 return (0); 6411 pd->proto = IPPROTO_ICMP; 6412 rewrite = 1; 6413 } 6414 break; 6415 #endif /* INET6 */ 6416 6417 default: 6418 break; 6419 } 6420 6421 return (rewrite); 6422 } 6423 6424 static int 6425 pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd, 6426 u_short *reason, int *copyback, struct pf_state_peer *src, 6427 struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst) 6428 { 6429 struct tcphdr *th = &pd->hdr.tcp; 6430 u_int16_t win = ntohs(th->th_win); 6431 u_int32_t ack, end, data_end, seq, orig_seq; 6432 u_int8_t sws, dws; 6433 int ackskew; 6434 6435 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 6436 sws = src->wscale & PF_WSCALE_MASK; 6437 dws = dst->wscale & PF_WSCALE_MASK; 6438 } else 6439 sws = dws = 0; 6440 6441 /* 6442 * Sequence tracking algorithm from Guido van Rooij's paper: 6443 * http://www.madison-gurkha.com/publications/tcp_filtering/ 6444 * tcp_filtering.ps 6445 */ 6446 6447 orig_seq = seq = ntohl(th->th_seq); 6448 if (src->seqlo == 0) { 6449 /* First packet from this end. Set its state */ 6450 6451 if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 6452 src->scrub == NULL) { 6453 if (pf_normalize_tcp_init(pd, th, src)) { 6454 REASON_SET(reason, PFRES_MEMORY); 6455 return (PF_DROP); 6456 } 6457 } 6458 6459 /* Deferred generation of sequence number modulator */ 6460 if (dst->seqdiff && !src->seqdiff) { 6461 /* use random iss for the TCP server */ 6462 while ((src->seqdiff = arc4random() - seq) == 0) 6463 ; 6464 ack = ntohl(th->th_ack) - dst->seqdiff; 6465 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6466 src->seqdiff), 0); 6467 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6468 *copyback = 1; 6469 } else { 6470 ack = ntohl(th->th_ack); 6471 } 6472 6473 end = seq + pd->p_len; 6474 if (tcp_get_flags(th) & TH_SYN) { 6475 end++; 6476 if (dst->wscale & PF_WSCALE_FLAG) { 6477 src->wscale = pf_get_wscale(pd); 6478 if (src->wscale & PF_WSCALE_FLAG) { 6479 /* Remove scale factor from initial 6480 * window */ 6481 sws = src->wscale & PF_WSCALE_MASK; 6482 win = ((u_int32_t)win + (1 << sws) - 1) 6483 >> sws; 6484 dws = dst->wscale & PF_WSCALE_MASK; 6485 } else { 6486 /* fixup other window */ 6487 dst->max_win = MIN(TCP_MAXWIN, 6488 (u_int32_t)dst->max_win << 6489 (dst->wscale & PF_WSCALE_MASK)); 6490 /* in case of a retrans SYN|ACK */ 6491 dst->wscale = 0; 6492 } 6493 } 6494 } 6495 data_end = end; 6496 if (tcp_get_flags(th) & TH_FIN) 6497 end++; 6498 6499 src->seqlo = seq; 6500 if (src->state < TCPS_SYN_SENT) 6501 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6502 6503 /* 6504 * May need to slide the window (seqhi may have been set by 6505 * the crappy stack check or if we picked up the connection 6506 * after establishment) 6507 */ 6508 if (src->seqhi == 1 || 6509 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 6510 src->seqhi = end + MAX(1, dst->max_win << dws); 6511 if (win > src->max_win) 6512 src->max_win = win; 6513 6514 } else { 6515 ack = ntohl(th->th_ack) - dst->seqdiff; 6516 if (src->seqdiff) { 6517 /* Modulate sequence numbers */ 6518 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 6519 src->seqdiff), 0); 6520 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 6521 *copyback = 1; 6522 } 6523 end = seq + pd->p_len; 6524 if (tcp_get_flags(th) & TH_SYN) 6525 end++; 6526 data_end = end; 6527 if (tcp_get_flags(th) & TH_FIN) 6528 end++; 6529 } 6530 6531 if ((tcp_get_flags(th) & TH_ACK) == 0) { 6532 /* Let it pass through the ack skew check */ 6533 ack = dst->seqlo; 6534 } else if ((ack == 0 && 6535 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 6536 /* broken tcp stacks do not set ack */ 6537 (dst->state < TCPS_SYN_SENT)) { 6538 /* 6539 * Many stacks (ours included) will set the ACK number in an 6540 * FIN|ACK if the SYN times out -- no sequence to ACK. 6541 */ 6542 ack = dst->seqlo; 6543 } 6544 6545 if (seq == end) { 6546 /* Ease sequencing restrictions on no data packets */ 6547 seq = src->seqlo; 6548 data_end = end = seq; 6549 } 6550 6551 ackskew = dst->seqlo - ack; 6552 6553 /* 6554 * Need to demodulate the sequence numbers in any TCP SACK options 6555 * (Selective ACK). We could optionally validate the SACK values 6556 * against the current ACK window, either forwards or backwards, but 6557 * I'm not confident that SACK has been implemented properly 6558 * everywhere. It wouldn't surprise me if several stacks accidentally 6559 * SACK too far backwards of previously ACKed data. There really aren't 6560 * any security implications of bad SACKing unless the target stack 6561 * doesn't validate the option length correctly. Someone trying to 6562 * spoof into a TCP connection won't bother blindly sending SACK 6563 * options anyway. 6564 */ 6565 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 6566 if (pf_modulate_sack(pd, th, dst)) 6567 *copyback = 1; 6568 } 6569 6570 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 6571 if (SEQ_GEQ(src->seqhi, data_end) && 6572 /* Last octet inside other's window space */ 6573 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 6574 /* Retrans: not more than one window back */ 6575 (ackskew >= -MAXACKWINDOW) && 6576 /* Acking not more than one reassembled fragment backwards */ 6577 (ackskew <= (MAXACKWINDOW << sws)) && 6578 /* Acking not more than one window forward */ 6579 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 6580 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 6581 /* Require an exact/+1 sequence match on resets when possible */ 6582 6583 if (dst->scrub || src->scrub) { 6584 if (pf_normalize_tcp_stateful(pd, reason, th, 6585 state, src, dst, copyback)) 6586 return (PF_DROP); 6587 } 6588 6589 /* update max window */ 6590 if (src->max_win < win) 6591 src->max_win = win; 6592 /* synchronize sequencing */ 6593 if (SEQ_GT(end, src->seqlo)) 6594 src->seqlo = end; 6595 /* slide the window of what the other end can send */ 6596 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6597 dst->seqhi = ack + MAX((win << sws), 1); 6598 6599 /* update states */ 6600 if (tcp_get_flags(th) & TH_SYN) 6601 if (src->state < TCPS_SYN_SENT) 6602 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6603 if (tcp_get_flags(th) & TH_FIN) 6604 if (src->state < TCPS_CLOSING) 6605 pf_set_protostate(state, psrc, TCPS_CLOSING); 6606 if (tcp_get_flags(th) & TH_ACK) { 6607 if (dst->state == TCPS_SYN_SENT) { 6608 pf_set_protostate(state, pdst, 6609 TCPS_ESTABLISHED); 6610 if (src->state == TCPS_ESTABLISHED && 6611 state->sns[PF_SN_LIMIT] != NULL && 6612 pf_src_connlimit(state)) { 6613 REASON_SET(reason, PFRES_SRCLIMIT); 6614 return (PF_DROP); 6615 } 6616 } else if (dst->state == TCPS_CLOSING) 6617 pf_set_protostate(state, pdst, 6618 TCPS_FIN_WAIT_2); 6619 } 6620 if (tcp_get_flags(th) & TH_RST) 6621 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6622 6623 /* update expire time */ 6624 state->expire = pf_get_uptime(); 6625 if (src->state >= TCPS_FIN_WAIT_2 && 6626 dst->state >= TCPS_FIN_WAIT_2) 6627 state->timeout = PFTM_TCP_CLOSED; 6628 else if (src->state >= TCPS_CLOSING && 6629 dst->state >= TCPS_CLOSING) 6630 state->timeout = PFTM_TCP_FIN_WAIT; 6631 else if (src->state < TCPS_ESTABLISHED || 6632 dst->state < TCPS_ESTABLISHED) 6633 state->timeout = PFTM_TCP_OPENING; 6634 else if (src->state >= TCPS_CLOSING || 6635 dst->state >= TCPS_CLOSING) 6636 state->timeout = PFTM_TCP_CLOSING; 6637 else 6638 state->timeout = PFTM_TCP_ESTABLISHED; 6639 6640 /* Fall through to PASS packet */ 6641 6642 } else if ((dst->state < TCPS_SYN_SENT || 6643 dst->state >= TCPS_FIN_WAIT_2 || 6644 src->state >= TCPS_FIN_WAIT_2) && 6645 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 6646 /* Within a window forward of the originating packet */ 6647 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 6648 /* Within a window backward of the originating packet */ 6649 6650 /* 6651 * This currently handles three situations: 6652 * 1) Stupid stacks will shotgun SYNs before their peer 6653 * replies. 6654 * 2) When PF catches an already established stream (the 6655 * firewall rebooted, the state table was flushed, routes 6656 * changed...) 6657 * 3) Packets get funky immediately after the connection 6658 * closes (this should catch Solaris spurious ACK|FINs 6659 * that web servers like to spew after a close) 6660 * 6661 * This must be a little more careful than the above code 6662 * since packet floods will also be caught here. We don't 6663 * update the TTL here to mitigate the damage of a packet 6664 * flood and so the same code can handle awkward establishment 6665 * and a loosened connection close. 6666 * In the establishment case, a correct peer response will 6667 * validate the connection, go through the normal state code 6668 * and keep updating the state TTL. 6669 */ 6670 6671 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6672 printf("pf: loose state match: "); 6673 pf_print_state(state); 6674 pf_print_flags(tcp_get_flags(th)); 6675 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6676 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 6677 pd->p_len, ackskew, (unsigned long long)state->packets[0], 6678 (unsigned long long)state->packets[1], 6679 pd->dir == PF_IN ? "in" : "out", 6680 pd->dir == state->direction ? "fwd" : "rev"); 6681 } 6682 6683 if (dst->scrub || src->scrub) { 6684 if (pf_normalize_tcp_stateful(pd, reason, th, 6685 state, src, dst, copyback)) 6686 return (PF_DROP); 6687 } 6688 6689 /* update max window */ 6690 if (src->max_win < win) 6691 src->max_win = win; 6692 /* synchronize sequencing */ 6693 if (SEQ_GT(end, src->seqlo)) 6694 src->seqlo = end; 6695 /* slide the window of what the other end can send */ 6696 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 6697 dst->seqhi = ack + MAX((win << sws), 1); 6698 6699 /* 6700 * Cannot set dst->seqhi here since this could be a shotgunned 6701 * SYN and not an already established connection. 6702 */ 6703 6704 if (tcp_get_flags(th) & TH_FIN) 6705 if (src->state < TCPS_CLOSING) 6706 pf_set_protostate(state, psrc, TCPS_CLOSING); 6707 if (tcp_get_flags(th) & TH_RST) 6708 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6709 6710 /* Fall through to PASS packet */ 6711 6712 } else { 6713 if (state->dst.state == TCPS_SYN_SENT && 6714 state->src.state == TCPS_SYN_SENT) { 6715 /* Send RST for state mismatches during handshake */ 6716 if (!(tcp_get_flags(th) & TH_RST)) 6717 pf_send_tcp(state->rule, pd->af, 6718 pd->dst, pd->src, th->th_dport, 6719 th->th_sport, ntohl(th->th_ack), 0, 6720 TH_RST, 0, 0, 6721 state->rule->return_ttl, M_SKIP_FIREWALL, 6722 0, 0, state->act.rtableid); 6723 src->seqlo = 0; 6724 src->seqhi = 1; 6725 src->max_win = 1; 6726 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6727 printf("pf: BAD state: "); 6728 pf_print_state(state); 6729 pf_print_flags(tcp_get_flags(th)); 6730 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6731 "pkts=%llu:%llu dir=%s,%s\n", 6732 seq, orig_seq, ack, pd->p_len, ackskew, 6733 (unsigned long long)state->packets[0], 6734 (unsigned long long)state->packets[1], 6735 pd->dir == PF_IN ? "in" : "out", 6736 pd->dir == state->direction ? "fwd" : "rev"); 6737 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6738 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6739 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6740 ' ': '2', 6741 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6742 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6743 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6744 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6745 } 6746 REASON_SET(reason, PFRES_BADSTATE); 6747 return (PF_DROP); 6748 } 6749 6750 return (PF_PASS); 6751 } 6752 6753 static int 6754 pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd, 6755 u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst, 6756 u_int8_t psrc, u_int8_t pdst) 6757 { 6758 struct tcphdr *th = &pd->hdr.tcp; 6759 6760 if (tcp_get_flags(th) & TH_SYN) 6761 if (src->state < TCPS_SYN_SENT) 6762 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 6763 if (tcp_get_flags(th) & TH_FIN) 6764 if (src->state < TCPS_CLOSING) 6765 pf_set_protostate(state, psrc, TCPS_CLOSING); 6766 if (tcp_get_flags(th) & TH_ACK) { 6767 if (dst->state == TCPS_SYN_SENT) { 6768 pf_set_protostate(state, pdst, TCPS_ESTABLISHED); 6769 if (src->state == TCPS_ESTABLISHED && 6770 state->sns[PF_SN_LIMIT] != NULL && 6771 pf_src_connlimit(state)) { 6772 REASON_SET(reason, PFRES_SRCLIMIT); 6773 return (PF_DROP); 6774 } 6775 } else if (dst->state == TCPS_CLOSING) { 6776 pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2); 6777 } else if (src->state == TCPS_SYN_SENT && 6778 dst->state < TCPS_SYN_SENT) { 6779 /* 6780 * Handle a special sloppy case where we only see one 6781 * half of the connection. If there is a ACK after 6782 * the initial SYN without ever seeing a packet from 6783 * the destination, set the connection to established. 6784 */ 6785 pf_set_protostate(state, PF_PEER_BOTH, 6786 TCPS_ESTABLISHED); 6787 dst->state = src->state = TCPS_ESTABLISHED; 6788 if (state->sns[PF_SN_LIMIT] != NULL && 6789 pf_src_connlimit(state)) { 6790 REASON_SET(reason, PFRES_SRCLIMIT); 6791 return (PF_DROP); 6792 } 6793 } else if (src->state == TCPS_CLOSING && 6794 dst->state == TCPS_ESTABLISHED && 6795 dst->seqlo == 0) { 6796 /* 6797 * Handle the closing of half connections where we 6798 * don't see the full bidirectional FIN/ACK+ACK 6799 * handshake. 6800 */ 6801 pf_set_protostate(state, pdst, TCPS_CLOSING); 6802 } 6803 } 6804 if (tcp_get_flags(th) & TH_RST) 6805 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6806 6807 /* update expire time */ 6808 state->expire = pf_get_uptime(); 6809 if (src->state >= TCPS_FIN_WAIT_2 && 6810 dst->state >= TCPS_FIN_WAIT_2) 6811 state->timeout = PFTM_TCP_CLOSED; 6812 else if (src->state >= TCPS_CLOSING && 6813 dst->state >= TCPS_CLOSING) 6814 state->timeout = PFTM_TCP_FIN_WAIT; 6815 else if (src->state < TCPS_ESTABLISHED || 6816 dst->state < TCPS_ESTABLISHED) 6817 state->timeout = PFTM_TCP_OPENING; 6818 else if (src->state >= TCPS_CLOSING || 6819 dst->state >= TCPS_CLOSING) 6820 state->timeout = PFTM_TCP_CLOSING; 6821 else 6822 state->timeout = PFTM_TCP_ESTABLISHED; 6823 6824 return (PF_PASS); 6825 } 6826 6827 static int 6828 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason) 6829 { 6830 struct pf_state_key *sk = state->key[pd->didx]; 6831 struct tcphdr *th = &pd->hdr.tcp; 6832 6833 if (state->src.state == PF_TCPS_PROXY_SRC) { 6834 if (pd->dir != state->direction) { 6835 REASON_SET(reason, PFRES_SYNPROXY); 6836 return (PF_SYNPROXY_DROP); 6837 } 6838 if (tcp_get_flags(th) & TH_SYN) { 6839 if (ntohl(th->th_seq) != state->src.seqlo) { 6840 REASON_SET(reason, PFRES_SYNPROXY); 6841 return (PF_DROP); 6842 } 6843 pf_send_tcp(state->rule, pd->af, pd->dst, 6844 pd->src, th->th_dport, th->th_sport, 6845 state->src.seqhi, ntohl(th->th_seq) + 1, 6846 TH_SYN|TH_ACK, 0, state->src.mss, 0, 6847 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6848 REASON_SET(reason, PFRES_SYNPROXY); 6849 return (PF_SYNPROXY_DROP); 6850 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6851 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6852 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6853 REASON_SET(reason, PFRES_SYNPROXY); 6854 return (PF_DROP); 6855 } else if (state->sns[PF_SN_LIMIT] != NULL && 6856 pf_src_connlimit(state)) { 6857 REASON_SET(reason, PFRES_SRCLIMIT); 6858 return (PF_DROP); 6859 } else 6860 pf_set_protostate(state, PF_PEER_SRC, 6861 PF_TCPS_PROXY_DST); 6862 } 6863 if (state->src.state == PF_TCPS_PROXY_DST) { 6864 if (pd->dir == state->direction) { 6865 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 6866 (ntohl(th->th_ack) != state->src.seqhi + 1) || 6867 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 6868 REASON_SET(reason, PFRES_SYNPROXY); 6869 return (PF_DROP); 6870 } 6871 state->src.max_win = MAX(ntohs(th->th_win), 1); 6872 if (state->dst.seqhi == 1) 6873 state->dst.seqhi = htonl(arc4random()); 6874 pf_send_tcp(state->rule, pd->af, 6875 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6876 sk->port[pd->sidx], sk->port[pd->didx], 6877 state->dst.seqhi, 0, TH_SYN, 0, 6878 state->src.mss, 0, 6879 state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6880 state->tag, 0, state->act.rtableid); 6881 REASON_SET(reason, PFRES_SYNPROXY); 6882 return (PF_SYNPROXY_DROP); 6883 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 6884 (TH_SYN|TH_ACK)) || 6885 (ntohl(th->th_ack) != state->dst.seqhi + 1)) { 6886 REASON_SET(reason, PFRES_SYNPROXY); 6887 return (PF_DROP); 6888 } else { 6889 state->dst.max_win = MAX(ntohs(th->th_win), 1); 6890 state->dst.seqlo = ntohl(th->th_seq); 6891 pf_send_tcp(state->rule, pd->af, pd->dst, 6892 pd->src, th->th_dport, th->th_sport, 6893 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6894 TH_ACK, state->src.max_win, 0, 0, 0, 6895 state->tag, 0, state->act.rtableid); 6896 pf_send_tcp(state->rule, pd->af, 6897 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6898 sk->port[pd->sidx], sk->port[pd->didx], 6899 state->src.seqhi + 1, state->src.seqlo + 1, 6900 TH_ACK, state->dst.max_win, 0, 0, 6901 M_SKIP_FIREWALL, 0, 0, state->act.rtableid); 6902 state->src.seqdiff = state->dst.seqhi - 6903 state->src.seqlo; 6904 state->dst.seqdiff = state->src.seqhi - 6905 state->dst.seqlo; 6906 state->src.seqhi = state->src.seqlo + 6907 state->dst.max_win; 6908 state->dst.seqhi = state->dst.seqlo + 6909 state->src.max_win; 6910 state->src.wscale = state->dst.wscale = 0; 6911 pf_set_protostate(state, PF_PEER_BOTH, 6912 TCPS_ESTABLISHED); 6913 REASON_SET(reason, PFRES_SYNPROXY); 6914 return (PF_SYNPROXY_DROP); 6915 } 6916 } 6917 6918 return (PF_PASS); 6919 } 6920 6921 static int 6922 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6923 { 6924 struct pf_state_key_cmp key; 6925 int copyback = 0; 6926 struct pf_state_peer *src, *dst; 6927 uint8_t psrc, pdst; 6928 int action = PF_PASS; 6929 6930 bzero(&key, sizeof(key)); 6931 key.af = pd->af; 6932 key.proto = pd->virtual_proto; 6933 PF_ACPY(&key.addr[pd->sidx], pd->src, key.af); 6934 PF_ACPY(&key.addr[pd->didx], pd->dst, key.af); 6935 key.port[pd->sidx] = pd->osport; 6936 key.port[pd->didx] = pd->odport; 6937 6938 STATE_LOOKUP(&key, *state, pd); 6939 6940 if (pd->dir == (*state)->direction) { 6941 if (PF_REVERSED_KEY(*state, pd->af)) { 6942 src = &(*state)->dst; 6943 dst = &(*state)->src; 6944 psrc = PF_PEER_DST; 6945 pdst = PF_PEER_SRC; 6946 } else { 6947 src = &(*state)->src; 6948 dst = &(*state)->dst; 6949 psrc = PF_PEER_SRC; 6950 pdst = PF_PEER_DST; 6951 } 6952 } else { 6953 if (PF_REVERSED_KEY(*state, pd->af)) { 6954 src = &(*state)->src; 6955 dst = &(*state)->dst; 6956 psrc = PF_PEER_SRC; 6957 pdst = PF_PEER_DST; 6958 } else { 6959 src = &(*state)->dst; 6960 dst = &(*state)->src; 6961 psrc = PF_PEER_DST; 6962 pdst = PF_PEER_SRC; 6963 } 6964 } 6965 6966 switch (pd->virtual_proto) { 6967 case IPPROTO_TCP: { 6968 struct tcphdr *th = &pd->hdr.tcp; 6969 6970 if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS) 6971 return (action); 6972 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 6973 (*state)->dst.state >= TCPS_FIN_WAIT_2 && 6974 (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN) || 6975 ((tcp_get_flags(th) & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6976 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6977 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6978 printf("pf: state reuse "); 6979 pf_print_state(*state); 6980 pf_print_flags(tcp_get_flags(th)); 6981 printf("\n"); 6982 } 6983 /* XXX make sure it's the same direction ?? */ 6984 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6985 pf_remove_state(*state); 6986 *state = NULL; 6987 return (PF_DROP); 6988 } 6989 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6990 if (pf_tcp_track_sloppy(*state, pd, reason, src, dst, 6991 psrc, pdst) == PF_DROP) 6992 return (PF_DROP); 6993 } else { 6994 int ret; 6995 6996 ret = pf_tcp_track_full(*state, pd, reason, 6997 ©back, src, dst, psrc, pdst); 6998 if (ret == PF_DROP) 6999 return (PF_DROP); 7000 } 7001 break; 7002 } 7003 case IPPROTO_UDP: 7004 /* update states */ 7005 if (src->state < PFUDPS_SINGLE) 7006 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 7007 if (dst->state == PFUDPS_SINGLE) 7008 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 7009 7010 /* update expire time */ 7011 (*state)->expire = pf_get_uptime(); 7012 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 7013 (*state)->timeout = PFTM_UDP_MULTIPLE; 7014 else 7015 (*state)->timeout = PFTM_UDP_SINGLE; 7016 break; 7017 case IPPROTO_SCTP: 7018 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 7019 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 7020 pd->sctp_flags & PFDESC_SCTP_INIT) { 7021 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7022 pf_remove_state(*state); 7023 *state = NULL; 7024 return (PF_DROP); 7025 } 7026 7027 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7028 return (PF_DROP); 7029 7030 /* Track state. */ 7031 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7032 if (src->state < SCTP_COOKIE_WAIT) { 7033 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7034 (*state)->timeout = PFTM_SCTP_OPENING; 7035 } 7036 } 7037 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7038 MPASS(dst->scrub != NULL); 7039 if (dst->scrub->pfss_v_tag == 0) 7040 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7041 } 7042 7043 /* 7044 * Bind to the correct interface if we're if-bound. For multihomed 7045 * extra associations we don't know which interface that will be until 7046 * here, so we've inserted the state on V_pf_all. Fix that now. 7047 */ 7048 if ((*state)->kif == V_pfi_all && 7049 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7050 (*state)->kif = pd->kif; 7051 7052 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7053 if (src->state < SCTP_ESTABLISHED) { 7054 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7055 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7056 } 7057 } 7058 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7059 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7060 if (src->state < SCTP_SHUTDOWN_PENDING) { 7061 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7062 (*state)->timeout = PFTM_SCTP_CLOSING; 7063 } 7064 } 7065 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7066 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7067 (*state)->timeout = PFTM_SCTP_CLOSED; 7068 } 7069 7070 (*state)->expire = pf_get_uptime(); 7071 break; 7072 default: 7073 /* update states */ 7074 if (src->state < PFOTHERS_SINGLE) 7075 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7076 if (dst->state == PFOTHERS_SINGLE) 7077 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7078 7079 /* update expire time */ 7080 (*state)->expire = pf_get_uptime(); 7081 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7082 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7083 else 7084 (*state)->timeout = PFTM_OTHER_SINGLE; 7085 break; 7086 } 7087 7088 /* translate source/destination address, if necessary */ 7089 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7090 struct pf_state_key *nk; 7091 int afto, sidx, didx; 7092 7093 if (PF_REVERSED_KEY(*state, pd->af)) 7094 nk = (*state)->key[pd->sidx]; 7095 else 7096 nk = (*state)->key[pd->didx]; 7097 7098 afto = pd->af != nk->af; 7099 7100 if (afto && (*state)->direction == PF_IN) { 7101 sidx = pd->didx; 7102 didx = pd->sidx; 7103 } else { 7104 sidx = pd->sidx; 7105 didx = pd->didx; 7106 } 7107 7108 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 7109 nk->port[sidx] != pd->osport) 7110 pf_change_ap(pd, pd->src, pd->sport, pd->ip_sum, 7111 pd->pcksum, &nk->addr[sidx], 7112 nk->port[sidx], pd->virtual_proto == IPPROTO_UDP, 7113 pd->af, nk->af); 7114 7115 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 7116 nk->port[didx] != pd->odport) 7117 pf_change_ap(pd, pd->dst, pd->dport, pd->ip_sum, 7118 pd->pcksum, &nk->addr[didx], 7119 nk->port[didx], pd->virtual_proto == IPPROTO_UDP, 7120 pd->af, nk->af); 7121 7122 if (afto) { 7123 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7124 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7125 pd->naf = nk->af; 7126 action = PF_AFRT; 7127 } 7128 7129 copyback = 1; 7130 } 7131 7132 if (copyback && pd->hdrlen > 0) 7133 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 7134 7135 return (action); 7136 } 7137 7138 static int 7139 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 7140 u_short *reason) 7141 { 7142 struct pf_state_peer *src; 7143 if (pd->dir == state->direction) { 7144 if (PF_REVERSED_KEY(state, pd->af)) 7145 src = &state->dst; 7146 else 7147 src = &state->src; 7148 } else { 7149 if (PF_REVERSED_KEY(state, pd->af)) 7150 src = &state->src; 7151 else 7152 src = &state->dst; 7153 } 7154 7155 if (src->scrub != NULL) { 7156 if (src->scrub->pfss_v_tag == 0) 7157 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 7158 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 7159 return (PF_DROP); 7160 } 7161 7162 return (PF_PASS); 7163 } 7164 7165 static void 7166 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 7167 { 7168 struct pf_sctp_endpoint key; 7169 struct pf_sctp_endpoint *ep; 7170 struct pf_state_key *sks = s->key[PF_SK_STACK]; 7171 struct pf_sctp_source *i, *tmp; 7172 7173 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 7174 return; 7175 7176 PF_SCTP_ENDPOINTS_LOCK(); 7177 7178 key.v_tag = s->dst.scrub->pfss_v_tag; 7179 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7180 if (ep != NULL) { 7181 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7182 if (pf_addr_cmp(&i->addr, 7183 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 7184 s->key[PF_SK_WIRE]->af) == 0) { 7185 SDT_PROBE3(pf, sctp, multihome, remove, 7186 key.v_tag, s, i); 7187 TAILQ_REMOVE(&ep->sources, i, entry); 7188 free(i, M_PFTEMP); 7189 break; 7190 } 7191 } 7192 7193 if (TAILQ_EMPTY(&ep->sources)) { 7194 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7195 free(ep, M_PFTEMP); 7196 } 7197 } 7198 7199 /* Other direction. */ 7200 key.v_tag = s->src.scrub->pfss_v_tag; 7201 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7202 if (ep != NULL) { 7203 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 7204 if (pf_addr_cmp(&i->addr, 7205 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 7206 s->key[PF_SK_WIRE]->af) == 0) { 7207 SDT_PROBE3(pf, sctp, multihome, remove, 7208 key.v_tag, s, i); 7209 TAILQ_REMOVE(&ep->sources, i, entry); 7210 free(i, M_PFTEMP); 7211 break; 7212 } 7213 } 7214 7215 if (TAILQ_EMPTY(&ep->sources)) { 7216 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7217 free(ep, M_PFTEMP); 7218 } 7219 } 7220 7221 PF_SCTP_ENDPOINTS_UNLOCK(); 7222 } 7223 7224 static void 7225 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 7226 { 7227 struct pf_sctp_endpoint key = { 7228 .v_tag = v_tag, 7229 }; 7230 struct pf_sctp_source *i; 7231 struct pf_sctp_endpoint *ep; 7232 7233 PF_SCTP_ENDPOINTS_LOCK(); 7234 7235 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7236 if (ep == NULL) { 7237 ep = malloc(sizeof(struct pf_sctp_endpoint), 7238 M_PFTEMP, M_NOWAIT); 7239 if (ep == NULL) { 7240 PF_SCTP_ENDPOINTS_UNLOCK(); 7241 return; 7242 } 7243 7244 ep->v_tag = v_tag; 7245 TAILQ_INIT(&ep->sources); 7246 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 7247 } 7248 7249 /* Avoid inserting duplicates. */ 7250 TAILQ_FOREACH(i, &ep->sources, entry) { 7251 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 7252 PF_SCTP_ENDPOINTS_UNLOCK(); 7253 return; 7254 } 7255 } 7256 7257 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 7258 if (i == NULL) { 7259 PF_SCTP_ENDPOINTS_UNLOCK(); 7260 return; 7261 } 7262 7263 i->af = pd->af; 7264 memcpy(&i->addr, a, sizeof(*a)); 7265 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 7266 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 7267 7268 PF_SCTP_ENDPOINTS_UNLOCK(); 7269 } 7270 7271 static void 7272 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 7273 struct pf_kstate *s, int action) 7274 { 7275 struct pf_sctp_multihome_job *j, *tmp; 7276 struct pf_sctp_source *i; 7277 int ret __unused; 7278 struct pf_kstate *sm = NULL; 7279 struct pf_krule *ra = NULL; 7280 struct pf_krule *r = &V_pf_default_rule; 7281 struct pf_kruleset *rs = NULL; 7282 u_short reason; 7283 bool do_extra = true; 7284 7285 PF_RULES_RLOCK_TRACKER; 7286 7287 again: 7288 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 7289 if (s == NULL || action != PF_PASS) 7290 goto free; 7291 7292 /* Confirm we don't recurse here. */ 7293 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 7294 7295 switch (j->op) { 7296 case SCTP_ADD_IP_ADDRESS: { 7297 uint32_t v_tag = pd->sctp_initiate_tag; 7298 7299 if (v_tag == 0) { 7300 if (s->direction == pd->dir) 7301 v_tag = s->src.scrub->pfss_v_tag; 7302 else 7303 v_tag = s->dst.scrub->pfss_v_tag; 7304 } 7305 7306 /* 7307 * Avoid duplicating states. We'll already have 7308 * created a state based on the source address of 7309 * the packet, but SCTP endpoints may also list this 7310 * address again in the INIT(_ACK) parameters. 7311 */ 7312 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 7313 break; 7314 } 7315 7316 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 7317 PF_RULES_RLOCK(); 7318 sm = NULL; 7319 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 7320 j->pd.related_rule = s->rule; 7321 } 7322 ret = pf_test_rule(&r, &sm, 7323 &j->pd, &ra, &rs, &reason, NULL); 7324 PF_RULES_RUNLOCK(); 7325 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 7326 if (ret != PF_DROP && sm != NULL) { 7327 /* Inherit v_tag values. */ 7328 if (sm->direction == s->direction) { 7329 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7330 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7331 } else { 7332 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 7333 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 7334 } 7335 PF_STATE_UNLOCK(sm); 7336 } else { 7337 /* If we try duplicate inserts? */ 7338 break; 7339 } 7340 7341 /* Only add the address if we've actually allowed the state. */ 7342 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 7343 7344 if (! do_extra) { 7345 break; 7346 } 7347 /* 7348 * We need to do this for each of our source addresses. 7349 * Find those based on the verification tag. 7350 */ 7351 struct pf_sctp_endpoint key = { 7352 .v_tag = pd->hdr.sctp.v_tag, 7353 }; 7354 struct pf_sctp_endpoint *ep; 7355 7356 PF_SCTP_ENDPOINTS_LOCK(); 7357 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 7358 if (ep == NULL) { 7359 PF_SCTP_ENDPOINTS_UNLOCK(); 7360 break; 7361 } 7362 MPASS(ep != NULL); 7363 7364 TAILQ_FOREACH(i, &ep->sources, entry) { 7365 struct pf_sctp_multihome_job *nj; 7366 7367 /* SCTP can intermingle IPv4 and IPv6. */ 7368 if (i->af != pd->af) 7369 continue; 7370 7371 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 7372 if (! nj) { 7373 continue; 7374 } 7375 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 7376 memcpy(&nj->src, &j->src, sizeof(nj->src)); 7377 nj->pd.src = &nj->src; 7378 // New destination address! 7379 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 7380 nj->pd.dst = &nj->dst; 7381 nj->pd.m = j->pd.m; 7382 nj->op = j->op; 7383 7384 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 7385 } 7386 PF_SCTP_ENDPOINTS_UNLOCK(); 7387 7388 break; 7389 } 7390 case SCTP_DEL_IP_ADDRESS: { 7391 struct pf_state_key_cmp key; 7392 uint8_t psrc; 7393 7394 bzero(&key, sizeof(key)); 7395 key.af = j->pd.af; 7396 key.proto = IPPROTO_SCTP; 7397 if (j->pd.dir == PF_IN) { /* wire side, straight */ 7398 PF_ACPY(&key.addr[0], j->pd.src, key.af); 7399 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 7400 key.port[0] = j->pd.hdr.sctp.src_port; 7401 key.port[1] = j->pd.hdr.sctp.dest_port; 7402 } else { /* stack side, reverse */ 7403 PF_ACPY(&key.addr[1], j->pd.src, key.af); 7404 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 7405 key.port[1] = j->pd.hdr.sctp.src_port; 7406 key.port[0] = j->pd.hdr.sctp.dest_port; 7407 } 7408 7409 sm = pf_find_state(kif, &key, j->pd.dir); 7410 if (sm != NULL) { 7411 PF_STATE_LOCK_ASSERT(sm); 7412 if (j->pd.dir == sm->direction) { 7413 psrc = PF_PEER_SRC; 7414 } else { 7415 psrc = PF_PEER_DST; 7416 } 7417 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 7418 sm->timeout = PFTM_SCTP_CLOSING; 7419 PF_STATE_UNLOCK(sm); 7420 } 7421 break; 7422 default: 7423 panic("Unknown op %#x", j->op); 7424 } 7425 } 7426 7427 free: 7428 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 7429 free(j, M_PFTEMP); 7430 } 7431 7432 /* We may have inserted extra work while processing the list. */ 7433 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 7434 do_extra = false; 7435 goto again; 7436 } 7437 } 7438 7439 static int 7440 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 7441 { 7442 int off = 0; 7443 struct pf_sctp_multihome_job *job; 7444 7445 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 7446 7447 while (off < len) { 7448 struct sctp_paramhdr h; 7449 7450 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 7451 pd->af)) 7452 return (PF_DROP); 7453 7454 /* Parameters are at least 4 bytes. */ 7455 if (ntohs(h.param_length) < 4) 7456 return (PF_DROP); 7457 7458 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 7459 ntohs(h.param_length)); 7460 7461 switch (ntohs(h.param_type)) { 7462 case SCTP_IPV4_ADDRESS: { 7463 struct in_addr t; 7464 7465 if (ntohs(h.param_length) != 7466 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7467 return (PF_DROP); 7468 7469 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7470 NULL, NULL, pd->af)) 7471 return (PF_DROP); 7472 7473 if (in_nullhost(t)) 7474 t.s_addr = pd->src->v4.s_addr; 7475 7476 /* 7477 * We hold the state lock (idhash) here, which means 7478 * that we can't acquire the keyhash, or we'll get a 7479 * LOR (and potentially double-lock things too). We also 7480 * can't release the state lock here, so instead we'll 7481 * enqueue this for async handling. 7482 * There's a relatively small race here, in that a 7483 * packet using the new addresses could arrive already, 7484 * but that's just though luck for it. 7485 */ 7486 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7487 if (! job) 7488 return (PF_DROP); 7489 7490 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 7491 7492 memcpy(&job->pd, pd, sizeof(*pd)); 7493 7494 // New source address! 7495 memcpy(&job->src, &t, sizeof(t)); 7496 job->pd.src = &job->src; 7497 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7498 job->pd.dst = &job->dst; 7499 job->pd.m = pd->m; 7500 job->op = op; 7501 7502 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7503 break; 7504 } 7505 #ifdef INET6 7506 case SCTP_IPV6_ADDRESS: { 7507 struct in6_addr t; 7508 7509 if (ntohs(h.param_length) != 7510 (sizeof(struct sctp_paramhdr) + sizeof(t))) 7511 return (PF_DROP); 7512 7513 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 7514 NULL, NULL, pd->af)) 7515 return (PF_DROP); 7516 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 7517 break; 7518 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 7519 memcpy(&t, &pd->src->v6, sizeof(t)); 7520 7521 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 7522 if (! job) 7523 return (PF_DROP); 7524 7525 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 7526 7527 memcpy(&job->pd, pd, sizeof(*pd)); 7528 memcpy(&job->src, &t, sizeof(t)); 7529 job->pd.src = &job->src; 7530 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 7531 job->pd.dst = &job->dst; 7532 job->pd.m = pd->m; 7533 job->op = op; 7534 7535 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 7536 break; 7537 } 7538 #endif /* INET6 */ 7539 case SCTP_ADD_IP_ADDRESS: { 7540 int ret; 7541 struct sctp_asconf_paramhdr ah; 7542 7543 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7544 NULL, NULL, pd->af)) 7545 return (PF_DROP); 7546 7547 ret = pf_multihome_scan(start + off + sizeof(ah), 7548 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7549 SCTP_ADD_IP_ADDRESS); 7550 if (ret != PF_PASS) 7551 return (ret); 7552 break; 7553 } 7554 case SCTP_DEL_IP_ADDRESS: { 7555 int ret; 7556 struct sctp_asconf_paramhdr ah; 7557 7558 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 7559 NULL, NULL, pd->af)) 7560 return (PF_DROP); 7561 ret = pf_multihome_scan(start + off + sizeof(ah), 7562 ntohs(ah.ph.param_length) - sizeof(ah), pd, 7563 SCTP_DEL_IP_ADDRESS); 7564 if (ret != PF_PASS) 7565 return (ret); 7566 break; 7567 } 7568 default: 7569 break; 7570 } 7571 7572 off += roundup(ntohs(h.param_length), 4); 7573 } 7574 7575 return (PF_PASS); 7576 } 7577 7578 int 7579 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 7580 { 7581 start += sizeof(struct sctp_init_chunk); 7582 len -= sizeof(struct sctp_init_chunk); 7583 7584 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7585 } 7586 7587 int 7588 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 7589 { 7590 start += sizeof(struct sctp_asconf_chunk); 7591 len -= sizeof(struct sctp_asconf_chunk); 7592 7593 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 7594 } 7595 7596 int 7597 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 7598 struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir, 7599 int *iidx, int multi, int inner) 7600 { 7601 int direction = pd->dir; 7602 7603 key->af = pd->af; 7604 key->proto = pd->proto; 7605 if (icmp_dir == PF_IN) { 7606 *iidx = pd->sidx; 7607 key->port[pd->sidx] = icmpid; 7608 key->port[pd->didx] = type; 7609 } else { 7610 *iidx = pd->didx; 7611 key->port[pd->sidx] = type; 7612 key->port[pd->didx] = icmpid; 7613 } 7614 if (pf_state_key_addr_setup(pd, key, multi)) 7615 return (PF_DROP); 7616 7617 STATE_LOOKUP(key, *state, pd); 7618 7619 if ((*state)->state_flags & PFSTATE_SLOPPY) 7620 return (-1); 7621 7622 /* Is this ICMP message flowing in right direction? */ 7623 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 7624 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 7625 PF_IN : PF_OUT; 7626 else 7627 direction = (*state)->direction; 7628 if ((*state)->rule->type && 7629 (((!inner && direction == pd->dir) || 7630 (inner && direction != pd->dir)) ? 7631 PF_IN : PF_OUT) != icmp_dir) { 7632 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7633 printf("pf: icmp type %d in wrong direction (%d): ", 7634 ntohs(type), icmp_dir); 7635 pf_print_state(*state); 7636 printf("\n"); 7637 } 7638 PF_STATE_UNLOCK(*state); 7639 *state = NULL; 7640 return (PF_DROP); 7641 } 7642 return (-1); 7643 } 7644 7645 static int 7646 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 7647 u_short *reason) 7648 { 7649 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 7650 u_int16_t *icmpsum, virtual_id, virtual_type; 7651 u_int8_t icmptype, icmpcode; 7652 int icmp_dir, iidx, ret; 7653 struct pf_state_key_cmp key; 7654 #ifdef INET 7655 u_int16_t icmpid; 7656 #endif /* INET*/ 7657 7658 MPASS(*state == NULL); 7659 7660 bzero(&key, sizeof(key)); 7661 switch (pd->proto) { 7662 #ifdef INET 7663 case IPPROTO_ICMP: 7664 icmptype = pd->hdr.icmp.icmp_type; 7665 icmpcode = pd->hdr.icmp.icmp_code; 7666 icmpid = pd->hdr.icmp.icmp_id; 7667 icmpsum = &pd->hdr.icmp.icmp_cksum; 7668 break; 7669 #endif /* INET */ 7670 #ifdef INET6 7671 case IPPROTO_ICMPV6: 7672 icmptype = pd->hdr.icmp6.icmp6_type; 7673 icmpcode = pd->hdr.icmp6.icmp6_code; 7674 #ifdef INET 7675 icmpid = pd->hdr.icmp6.icmp6_id; 7676 #endif /* INET */ 7677 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7678 break; 7679 #endif /* INET6 */ 7680 default: 7681 panic("unhandled proto %d", pd->proto); 7682 } 7683 7684 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 7685 &virtual_type) == 0) { 7686 /* 7687 * ICMP query/reply message not related to a TCP/UDP/SCTP 7688 * packet. Search for an ICMP state. 7689 */ 7690 ret = pf_icmp_state_lookup(&key, pd, state, virtual_id, 7691 virtual_type, icmp_dir, &iidx, 0, 0); 7692 /* IPv6? try matching a multicast address */ 7693 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) { 7694 MPASS(*state == NULL); 7695 ret = pf_icmp_state_lookup(&key, pd, state, 7696 virtual_id, virtual_type, 7697 icmp_dir, &iidx, 1, 0); 7698 } 7699 if (ret >= 0) { 7700 MPASS(*state == NULL); 7701 return (ret); 7702 } 7703 7704 (*state)->expire = pf_get_uptime(); 7705 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7706 7707 /* translate source/destination address, if necessary */ 7708 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7709 struct pf_state_key *nk; 7710 int afto, sidx, didx; 7711 7712 if (PF_REVERSED_KEY(*state, pd->af)) 7713 nk = (*state)->key[pd->sidx]; 7714 else 7715 nk = (*state)->key[pd->didx]; 7716 7717 afto = pd->af != nk->af; 7718 7719 if (afto && (*state)->direction == PF_IN) { 7720 sidx = pd->didx; 7721 didx = pd->sidx; 7722 iidx = !iidx; 7723 } else { 7724 sidx = pd->sidx; 7725 didx = pd->didx; 7726 } 7727 7728 switch (pd->af) { 7729 #ifdef INET 7730 case AF_INET: 7731 #ifdef INET6 7732 if (afto) { 7733 if (pf_translate_icmp_af(AF_INET6, 7734 &pd->hdr.icmp)) 7735 return (PF_DROP); 7736 pd->proto = IPPROTO_ICMPV6; 7737 } 7738 #endif /* INET6 */ 7739 if (!afto && 7740 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 7741 pf_change_a(&saddr->v4.s_addr, 7742 pd->ip_sum, 7743 nk->addr[sidx].v4.s_addr, 7744 0); 7745 7746 if (!afto && PF_ANEQ(pd->dst, 7747 &nk->addr[didx], AF_INET)) 7748 pf_change_a(&daddr->v4.s_addr, 7749 pd->ip_sum, 7750 nk->addr[didx].v4.s_addr, 0); 7751 7752 if (nk->port[iidx] != 7753 pd->hdr.icmp.icmp_id) { 7754 pd->hdr.icmp.icmp_cksum = 7755 pf_cksum_fixup( 7756 pd->hdr.icmp.icmp_cksum, icmpid, 7757 nk->port[iidx], 0); 7758 pd->hdr.icmp.icmp_id = 7759 nk->port[iidx]; 7760 } 7761 7762 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7763 (caddr_t )&pd->hdr.icmp); 7764 break; 7765 #endif /* INET */ 7766 #ifdef INET6 7767 case AF_INET6: 7768 #ifdef INET 7769 if (afto) { 7770 if (pf_translate_icmp_af(AF_INET, 7771 &pd->hdr.icmp6)) 7772 return (PF_DROP); 7773 pd->proto = IPPROTO_ICMP; 7774 } 7775 #endif /* INET */ 7776 if (!afto && 7777 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 7778 pf_change_a6(saddr, 7779 &pd->hdr.icmp6.icmp6_cksum, 7780 &nk->addr[sidx], 0); 7781 7782 if (!afto && PF_ANEQ(pd->dst, 7783 &nk->addr[didx], AF_INET6)) 7784 pf_change_a6(daddr, 7785 &pd->hdr.icmp6.icmp6_cksum, 7786 &nk->addr[didx], 0); 7787 7788 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 7789 pd->hdr.icmp6.icmp6_id = 7790 nk->port[iidx]; 7791 7792 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7793 (caddr_t )&pd->hdr.icmp6); 7794 break; 7795 #endif /* INET6 */ 7796 } 7797 if (afto) { 7798 PF_ACPY(&pd->nsaddr, &nk->addr[sidx], nk->af); 7799 PF_ACPY(&pd->ndaddr, &nk->addr[didx], nk->af); 7800 pd->naf = nk->af; 7801 return (PF_AFRT); 7802 } 7803 } 7804 return (PF_PASS); 7805 7806 } else { 7807 /* 7808 * ICMP error message in response to a TCP/UDP packet. 7809 * Extract the inner TCP/UDP header and search for that state. 7810 */ 7811 7812 struct pf_pdesc pd2; 7813 bzero(&pd2, sizeof pd2); 7814 #ifdef INET 7815 struct ip h2; 7816 #endif /* INET */ 7817 #ifdef INET6 7818 struct ip6_hdr h2_6; 7819 #endif /* INET6 */ 7820 int ipoff2 = 0; 7821 7822 pd2.af = pd->af; 7823 pd2.dir = pd->dir; 7824 /* Payload packet is from the opposite direction. */ 7825 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7826 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7827 pd2.m = pd->m; 7828 pd2.kif = pd->kif; 7829 switch (pd->af) { 7830 #ifdef INET 7831 case AF_INET: 7832 /* offset of h2 in mbuf chain */ 7833 ipoff2 = pd->off + ICMP_MINLEN; 7834 7835 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7836 NULL, reason, pd2.af)) { 7837 DPFPRINTF(PF_DEBUG_MISC, 7838 ("pf: ICMP error message too short " 7839 "(ip)\n")); 7840 return (PF_DROP); 7841 } 7842 /* 7843 * ICMP error messages don't refer to non-first 7844 * fragments 7845 */ 7846 if (h2.ip_off & htons(IP_OFFMASK)) { 7847 REASON_SET(reason, PFRES_FRAG); 7848 return (PF_DROP); 7849 } 7850 7851 /* offset of protocol header that follows h2 */ 7852 pd2.off = ipoff2 + (h2.ip_hl << 2); 7853 7854 pd2.proto = h2.ip_p; 7855 pd2.tot_len = ntohs(h2.ip_len); 7856 pd2.src = (struct pf_addr *)&h2.ip_src; 7857 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7858 pd2.ip_sum = &h2.ip_sum; 7859 break; 7860 #endif /* INET */ 7861 #ifdef INET6 7862 case AF_INET6: 7863 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7864 7865 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7866 NULL, reason, pd2.af)) { 7867 DPFPRINTF(PF_DEBUG_MISC, 7868 ("pf: ICMP error message too short " 7869 "(ip6)\n")); 7870 return (PF_DROP); 7871 } 7872 pd2.off = ipoff2; 7873 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 7874 return (PF_DROP); 7875 7876 pd2.tot_len = ntohs(h2_6.ip6_plen) + 7877 sizeof(struct ip6_hdr); 7878 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7879 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7880 pd2.ip_sum = NULL; 7881 break; 7882 #endif /* INET6 */ 7883 default: 7884 unhandled_af(pd->af); 7885 } 7886 7887 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7888 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7889 printf("pf: BAD ICMP %d:%d outer dst: ", 7890 icmptype, icmpcode); 7891 pf_print_host(pd->src, 0, pd->af); 7892 printf(" -> "); 7893 pf_print_host(pd->dst, 0, pd->af); 7894 printf(" inner src: "); 7895 pf_print_host(pd2.src, 0, pd2.af); 7896 printf(" -> "); 7897 pf_print_host(pd2.dst, 0, pd2.af); 7898 printf("\n"); 7899 } 7900 REASON_SET(reason, PFRES_BADSTATE); 7901 return (PF_DROP); 7902 } 7903 7904 switch (pd2.proto) { 7905 case IPPROTO_TCP: { 7906 struct tcphdr th; 7907 u_int32_t seq; 7908 struct pf_state_peer *src, *dst; 7909 u_int8_t dws; 7910 int copyback = 0; 7911 7912 /* 7913 * Only the first 8 bytes of the TCP header can be 7914 * expected. Don't access any TCP header fields after 7915 * th_seq, an ackskew test is not possible. 7916 */ 7917 if (!pf_pull_hdr(pd->m, pd2.off, &th, 8, NULL, reason, 7918 pd2.af)) { 7919 DPFPRINTF(PF_DEBUG_MISC, 7920 ("pf: ICMP error message too short " 7921 "(tcp)\n")); 7922 return (PF_DROP); 7923 } 7924 7925 key.af = pd2.af; 7926 key.proto = IPPROTO_TCP; 7927 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7928 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7929 key.port[pd2.sidx] = th.th_sport; 7930 key.port[pd2.didx] = th.th_dport; 7931 7932 STATE_LOOKUP(&key, *state, pd); 7933 7934 if (pd->dir == (*state)->direction) { 7935 if (PF_REVERSED_KEY(*state, pd->af)) { 7936 src = &(*state)->src; 7937 dst = &(*state)->dst; 7938 } else { 7939 src = &(*state)->dst; 7940 dst = &(*state)->src; 7941 } 7942 } else { 7943 if (PF_REVERSED_KEY(*state, pd->af)) { 7944 src = &(*state)->dst; 7945 dst = &(*state)->src; 7946 } else { 7947 src = &(*state)->src; 7948 dst = &(*state)->dst; 7949 } 7950 } 7951 7952 if (src->wscale && dst->wscale) 7953 dws = dst->wscale & PF_WSCALE_MASK; 7954 else 7955 dws = 0; 7956 7957 /* Demodulate sequence number */ 7958 seq = ntohl(th.th_seq) - src->seqdiff; 7959 if (src->seqdiff) { 7960 pf_change_a(&th.th_seq, icmpsum, 7961 htonl(seq), 0); 7962 copyback = 1; 7963 } 7964 7965 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7966 (!SEQ_GEQ(src->seqhi, seq) || 7967 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7968 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7969 printf("pf: BAD ICMP %d:%d ", 7970 icmptype, icmpcode); 7971 pf_print_host(pd->src, 0, pd->af); 7972 printf(" -> "); 7973 pf_print_host(pd->dst, 0, pd->af); 7974 printf(" state: "); 7975 pf_print_state(*state); 7976 printf(" seq=%u\n", seq); 7977 } 7978 REASON_SET(reason, PFRES_BADSTATE); 7979 return (PF_DROP); 7980 } else { 7981 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7982 printf("pf: OK ICMP %d:%d ", 7983 icmptype, icmpcode); 7984 pf_print_host(pd->src, 0, pd->af); 7985 printf(" -> "); 7986 pf_print_host(pd->dst, 0, pd->af); 7987 printf(" state: "); 7988 pf_print_state(*state); 7989 printf(" seq=%u\n", seq); 7990 } 7991 } 7992 7993 /* translate source/destination address, if necessary */ 7994 if ((*state)->key[PF_SK_WIRE] != 7995 (*state)->key[PF_SK_STACK]) { 7996 7997 struct pf_state_key *nk; 7998 7999 if (PF_REVERSED_KEY(*state, pd->af)) 8000 nk = (*state)->key[pd->sidx]; 8001 else 8002 nk = (*state)->key[pd->didx]; 8003 8004 #if defined(INET) && defined(INET6) 8005 int afto, sidx, didx; 8006 u_int16_t dummy_cksum = 0; 8007 8008 afto = pd->af != nk->af; 8009 8010 if (afto && (*state)->direction == PF_IN) { 8011 sidx = pd2.didx; 8012 didx = pd2.sidx; 8013 } else { 8014 sidx = pd2.sidx; 8015 didx = pd2.didx; 8016 } 8017 8018 if (afto) { 8019 if (pf_translate_icmp_af(nk->af, 8020 &pd->hdr.icmp)) 8021 return (PF_DROP); 8022 m_copyback(pd->m, pd->off, 8023 sizeof(struct icmp6_hdr), 8024 (c_caddr_t)&pd->hdr.icmp6); 8025 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8026 &pd2, &nk->addr[sidx], 8027 &nk->addr[didx], pd->af, 8028 nk->af)) 8029 return (PF_DROP); 8030 pf_change_ap(pd, pd2.src, &th.th_sport, 8031 pd->ip_sum, &dummy_cksum, &nk->addr[pd2.sidx], 8032 nk->port[sidx], 1, pd->af, nk->af); 8033 pf_change_ap(pd, pd2.dst, &th.th_dport, 8034 pd->ip_sum, &dummy_cksum, &nk->addr[pd2.didx], 8035 nk->port[didx], 1, pd->af, nk->af); 8036 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)&th); 8037 PF_ACPY(&pd->nsaddr, &nk->addr[pd2.sidx], 8038 nk->af); 8039 PF_ACPY(&pd->ndaddr, 8040 &nk->addr[pd2.didx], nk->af); 8041 if (nk->af == AF_INET) { 8042 pd->proto = IPPROTO_ICMP; 8043 } else { 8044 pd->proto = IPPROTO_ICMPV6; 8045 /* 8046 * IPv4 becomes IPv6 so we must 8047 * copy IPv4 src addr to least 8048 * 32bits in IPv6 address to 8049 * keep traceroute/icmp 8050 * working. 8051 */ 8052 pd->nsaddr.addr32[3] = 8053 pd->src->addr32[0]; 8054 } 8055 pd->naf = nk->af; 8056 return (PF_AFRT); 8057 } 8058 #endif /* INET && INET6 */ 8059 8060 if (PF_ANEQ(pd2.src, 8061 &nk->addr[pd2.sidx], pd2.af) || 8062 nk->port[pd2.sidx] != th.th_sport) 8063 pf_change_icmp(pd2.src, &th.th_sport, 8064 daddr, &nk->addr[pd2.sidx], 8065 nk->port[pd2.sidx], NULL, 8066 pd2.ip_sum, icmpsum, 8067 pd->ip_sum, 0, pd2.af); 8068 8069 if (PF_ANEQ(pd2.dst, 8070 &nk->addr[pd2.didx], pd2.af) || 8071 nk->port[pd2.didx] != th.th_dport) 8072 pf_change_icmp(pd2.dst, &th.th_dport, 8073 saddr, &nk->addr[pd2.didx], 8074 nk->port[pd2.didx], NULL, 8075 pd2.ip_sum, icmpsum, 8076 pd->ip_sum, 0, pd2.af); 8077 copyback = 1; 8078 } 8079 8080 if (copyback) { 8081 switch (pd2.af) { 8082 #ifdef INET 8083 case AF_INET: 8084 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8085 (caddr_t )&pd->hdr.icmp); 8086 m_copyback(pd->m, ipoff2, sizeof(h2), 8087 (caddr_t )&h2); 8088 break; 8089 #endif /* INET */ 8090 #ifdef INET6 8091 case AF_INET6: 8092 m_copyback(pd->m, pd->off, 8093 sizeof(struct icmp6_hdr), 8094 (caddr_t )&pd->hdr.icmp6); 8095 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8096 (caddr_t )&h2_6); 8097 break; 8098 #endif /* INET6 */ 8099 default: 8100 unhandled_af(pd->af); 8101 } 8102 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 8103 } 8104 8105 return (PF_PASS); 8106 break; 8107 } 8108 case IPPROTO_UDP: { 8109 struct udphdr uh; 8110 8111 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 8112 NULL, reason, pd2.af)) { 8113 DPFPRINTF(PF_DEBUG_MISC, 8114 ("pf: ICMP error message too short " 8115 "(udp)\n")); 8116 return (PF_DROP); 8117 } 8118 8119 key.af = pd2.af; 8120 key.proto = IPPROTO_UDP; 8121 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8122 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8123 key.port[pd2.sidx] = uh.uh_sport; 8124 key.port[pd2.didx] = uh.uh_dport; 8125 8126 STATE_LOOKUP(&key, *state, pd); 8127 8128 /* translate source/destination address, if necessary */ 8129 if ((*state)->key[PF_SK_WIRE] != 8130 (*state)->key[PF_SK_STACK]) { 8131 struct pf_state_key *nk; 8132 8133 if (PF_REVERSED_KEY(*state, pd->af)) 8134 nk = (*state)->key[pd->sidx]; 8135 else 8136 nk = (*state)->key[pd->didx]; 8137 8138 #if defined(INET) && defined(INET6) 8139 int afto, sidx, didx; 8140 8141 afto = pd->af != nk->af; 8142 8143 if (afto && (*state)->direction == PF_IN) { 8144 sidx = pd2.didx; 8145 didx = pd2.sidx; 8146 } else { 8147 sidx = pd2.sidx; 8148 didx = pd2.didx; 8149 } 8150 8151 if (afto) { 8152 if (pf_translate_icmp_af(nk->af, 8153 &pd->hdr.icmp)) 8154 return (PF_DROP); 8155 m_copyback(pd->m, pd->off, 8156 sizeof(struct icmp6_hdr), 8157 (c_caddr_t)&pd->hdr.icmp6); 8158 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8159 &pd2, &nk->addr[sidx], 8160 &nk->addr[didx], pd->af, 8161 nk->af)) 8162 return (PF_DROP); 8163 pf_change_ap(pd, pd2.src, &uh.uh_sport, 8164 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.sidx], 8165 nk->port[sidx], 1, pd->af, nk->af); 8166 pf_change_ap(pd, pd2.dst, &uh.uh_dport, 8167 pd->ip_sum, &uh.uh_sum, &nk->addr[pd2.didx], 8168 nk->port[didx], 1, pd->af, nk->af); 8169 m_copyback(pd2.m, pd2.off, sizeof(uh), 8170 (c_caddr_t)&uh); 8171 PF_ACPY(&pd->nsaddr, 8172 &nk->addr[pd2.sidx], nk->af); 8173 PF_ACPY(&pd->ndaddr, 8174 &nk->addr[pd2.didx], nk->af); 8175 if (nk->af == AF_INET) { 8176 pd->proto = IPPROTO_ICMP; 8177 } else { 8178 pd->proto = IPPROTO_ICMPV6; 8179 /* 8180 * IPv4 becomes IPv6 so we must 8181 * copy IPv4 src addr to least 8182 * 32bits in IPv6 address to 8183 * keep traceroute/icmp 8184 * working. 8185 */ 8186 pd->nsaddr.addr32[3] = 8187 pd->src->addr32[0]; 8188 } 8189 pd->naf = nk->af; 8190 return (PF_AFRT); 8191 } 8192 #endif /* INET && INET6 */ 8193 8194 if (PF_ANEQ(pd2.src, 8195 &nk->addr[pd2.sidx], pd2.af) || 8196 nk->port[pd2.sidx] != uh.uh_sport) 8197 pf_change_icmp(pd2.src, &uh.uh_sport, 8198 daddr, &nk->addr[pd2.sidx], 8199 nk->port[pd2.sidx], &uh.uh_sum, 8200 pd2.ip_sum, icmpsum, 8201 pd->ip_sum, 1, pd2.af); 8202 8203 if (PF_ANEQ(pd2.dst, 8204 &nk->addr[pd2.didx], pd2.af) || 8205 nk->port[pd2.didx] != uh.uh_dport) 8206 pf_change_icmp(pd2.dst, &uh.uh_dport, 8207 saddr, &nk->addr[pd2.didx], 8208 nk->port[pd2.didx], &uh.uh_sum, 8209 pd2.ip_sum, icmpsum, 8210 pd->ip_sum, 1, pd2.af); 8211 8212 switch (pd2.af) { 8213 #ifdef INET 8214 case AF_INET: 8215 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8216 (caddr_t )&pd->hdr.icmp); 8217 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8218 break; 8219 #endif /* INET */ 8220 #ifdef INET6 8221 case AF_INET6: 8222 m_copyback(pd->m, pd->off, 8223 sizeof(struct icmp6_hdr), 8224 (caddr_t )&pd->hdr.icmp6); 8225 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8226 (caddr_t )&h2_6); 8227 break; 8228 #endif /* INET6 */ 8229 } 8230 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 8231 } 8232 return (PF_PASS); 8233 break; 8234 } 8235 #ifdef INET 8236 case IPPROTO_SCTP: { 8237 struct sctphdr sh; 8238 struct pf_state_peer *src; 8239 int copyback = 0; 8240 8241 if (! pf_pull_hdr(pd->m, pd2.off, &sh, sizeof(sh), NULL, reason, 8242 pd2.af)) { 8243 DPFPRINTF(PF_DEBUG_MISC, 8244 ("pf: ICMP error message too short " 8245 "(sctp)\n")); 8246 return (PF_DROP); 8247 } 8248 8249 key.af = pd2.af; 8250 key.proto = IPPROTO_SCTP; 8251 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8252 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8253 key.port[pd2.sidx] = sh.src_port; 8254 key.port[pd2.didx] = sh.dest_port; 8255 8256 STATE_LOOKUP(&key, *state, pd); 8257 8258 if (pd->dir == (*state)->direction) { 8259 if (PF_REVERSED_KEY(*state, pd->af)) 8260 src = &(*state)->src; 8261 else 8262 src = &(*state)->dst; 8263 } else { 8264 if (PF_REVERSED_KEY(*state, pd->af)) 8265 src = &(*state)->dst; 8266 else 8267 src = &(*state)->src; 8268 } 8269 8270 if (src->scrub->pfss_v_tag != sh.v_tag) { 8271 DPFPRINTF(PF_DEBUG_MISC, 8272 ("pf: ICMP error message has incorrect " 8273 "SCTP v_tag\n")); 8274 return (PF_DROP); 8275 } 8276 8277 /* translate source/destination address, if necessary */ 8278 if ((*state)->key[PF_SK_WIRE] != 8279 (*state)->key[PF_SK_STACK]) { 8280 8281 struct pf_state_key *nk; 8282 8283 if (PF_REVERSED_KEY(*state, pd->af)) 8284 nk = (*state)->key[pd->sidx]; 8285 else 8286 nk = (*state)->key[pd->didx]; 8287 8288 #if defined(INET) && defined(INET6) 8289 int afto, sidx, didx; 8290 8291 afto = pd->af != nk->af; 8292 8293 if (afto && (*state)->direction == PF_IN) { 8294 sidx = pd2.didx; 8295 didx = pd2.sidx; 8296 } else { 8297 sidx = pd2.sidx; 8298 didx = pd2.didx; 8299 } 8300 8301 if (afto) { 8302 if (pf_translate_icmp_af(nk->af, 8303 &pd->hdr.icmp)) 8304 return (PF_DROP); 8305 m_copyback(pd->m, pd->off, 8306 sizeof(struct icmp6_hdr), 8307 (c_caddr_t)&pd->hdr.icmp6); 8308 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8309 &pd2, &nk->addr[sidx], 8310 &nk->addr[didx], pd->af, 8311 nk->af)) 8312 return (PF_DROP); 8313 sh.src_port = nk->port[sidx]; 8314 sh.dest_port = nk->port[didx]; 8315 m_copyback(pd2.m, pd2.off, sizeof(sh), (c_caddr_t)&sh); 8316 PF_ACPY(&pd->nsaddr, 8317 &nk->addr[pd2.sidx], nk->af); 8318 PF_ACPY(&pd->ndaddr, 8319 &nk->addr[pd2.didx], nk->af); 8320 if (nk->af == AF_INET) { 8321 pd->proto = IPPROTO_ICMP; 8322 } else { 8323 pd->proto = IPPROTO_ICMPV6; 8324 /* 8325 * IPv4 becomes IPv6 so we must 8326 * copy IPv4 src addr to least 8327 * 32bits in IPv6 address to 8328 * keep traceroute/icmp 8329 * working. 8330 */ 8331 pd->nsaddr.addr32[3] = 8332 pd->src->addr32[0]; 8333 } 8334 pd->naf = nk->af; 8335 return (PF_AFRT); 8336 } 8337 #endif /* INET && INET6 */ 8338 8339 if (PF_ANEQ(pd2.src, 8340 &nk->addr[pd2.sidx], pd2.af) || 8341 nk->port[pd2.sidx] != sh.src_port) 8342 pf_change_icmp(pd2.src, &sh.src_port, 8343 daddr, &nk->addr[pd2.sidx], 8344 nk->port[pd2.sidx], NULL, 8345 pd2.ip_sum, icmpsum, 8346 pd->ip_sum, 0, pd2.af); 8347 8348 if (PF_ANEQ(pd2.dst, 8349 &nk->addr[pd2.didx], pd2.af) || 8350 nk->port[pd2.didx] != sh.dest_port) 8351 pf_change_icmp(pd2.dst, &sh.dest_port, 8352 saddr, &nk->addr[pd2.didx], 8353 nk->port[pd2.didx], NULL, 8354 pd2.ip_sum, icmpsum, 8355 pd->ip_sum, 0, pd2.af); 8356 copyback = 1; 8357 } 8358 8359 if (copyback) { 8360 switch (pd2.af) { 8361 #ifdef INET 8362 case AF_INET: 8363 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8364 (caddr_t )&pd->hdr.icmp); 8365 m_copyback(pd->m, ipoff2, sizeof(h2), 8366 (caddr_t )&h2); 8367 break; 8368 #endif /* INET */ 8369 #ifdef INET6 8370 case AF_INET6: 8371 m_copyback(pd->m, pd->off, 8372 sizeof(struct icmp6_hdr), 8373 (caddr_t )&pd->hdr.icmp6); 8374 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8375 (caddr_t )&h2_6); 8376 break; 8377 #endif /* INET6 */ 8378 } 8379 m_copyback(pd->m, pd2.off, sizeof(sh), (caddr_t)&sh); 8380 } 8381 8382 return (PF_PASS); 8383 break; 8384 } 8385 case IPPROTO_ICMP: { 8386 struct icmp *iih = &pd2.hdr.icmp; 8387 8388 if (pd2.af != AF_INET) { 8389 REASON_SET(reason, PFRES_NORM); 8390 return (PF_DROP); 8391 } 8392 8393 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 8394 NULL, reason, pd2.af)) { 8395 DPFPRINTF(PF_DEBUG_MISC, 8396 ("pf: ICMP error message too short i" 8397 "(icmp)\n")); 8398 return (PF_DROP); 8399 } 8400 8401 icmpid = iih->icmp_id; 8402 pf_icmp_mapping(&pd2, iih->icmp_type, 8403 &icmp_dir, &virtual_id, &virtual_type); 8404 8405 ret = pf_icmp_state_lookup(&key, &pd2, state, 8406 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8407 if (ret >= 0) { 8408 MPASS(*state == NULL); 8409 return (ret); 8410 } 8411 8412 /* translate source/destination address, if necessary */ 8413 if ((*state)->key[PF_SK_WIRE] != 8414 (*state)->key[PF_SK_STACK]) { 8415 struct pf_state_key *nk; 8416 8417 if (PF_REVERSED_KEY(*state, pd->af)) 8418 nk = (*state)->key[pd->sidx]; 8419 else 8420 nk = (*state)->key[pd->didx]; 8421 8422 #if defined(INET) && defined(INET6) 8423 int afto, sidx, didx; 8424 8425 afto = pd->af != nk->af; 8426 8427 if (afto && (*state)->direction == PF_IN) { 8428 sidx = pd2.didx; 8429 didx = pd2.sidx; 8430 iidx = !iidx; 8431 } else { 8432 sidx = pd2.sidx; 8433 didx = pd2.didx; 8434 } 8435 8436 if (afto) { 8437 if (nk->af != AF_INET6) 8438 return (PF_DROP); 8439 if (pf_translate_icmp_af(nk->af, 8440 &pd->hdr.icmp)) 8441 return (PF_DROP); 8442 m_copyback(pd->m, pd->off, 8443 sizeof(struct icmp6_hdr), 8444 (c_caddr_t)&pd->hdr.icmp6); 8445 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8446 &pd2, &nk->addr[sidx], 8447 &nk->addr[didx], pd->af, 8448 nk->af)) 8449 return (PF_DROP); 8450 pd->proto = IPPROTO_ICMPV6; 8451 if (pf_translate_icmp_af(nk->af, iih)) 8452 return (PF_DROP); 8453 if (virtual_type == htons(ICMP_ECHO) && 8454 nk->port[iidx] != iih->icmp_id) 8455 iih->icmp_id = nk->port[iidx]; 8456 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 8457 (c_caddr_t)iih); 8458 PF_ACPY(&pd->nsaddr, 8459 &nk->addr[pd2.sidx], nk->af); 8460 PF_ACPY(&pd->ndaddr, 8461 &nk->addr[pd2.didx], nk->af); 8462 /* 8463 * IPv4 becomes IPv6 so we must copy 8464 * IPv4 src addr to least 32bits in 8465 * IPv6 address to keep traceroute 8466 * working. 8467 */ 8468 pd->nsaddr.addr32[3] = 8469 pd->src->addr32[0]; 8470 pd->naf = nk->af; 8471 return (PF_AFRT); 8472 } 8473 #endif /* INET && INET6 */ 8474 8475 if (PF_ANEQ(pd2.src, 8476 &nk->addr[pd2.sidx], pd2.af) || 8477 (virtual_type == htons(ICMP_ECHO) && 8478 nk->port[iidx] != iih->icmp_id)) 8479 pf_change_icmp(pd2.src, 8480 (virtual_type == htons(ICMP_ECHO)) ? 8481 &iih->icmp_id : NULL, 8482 daddr, &nk->addr[pd2.sidx], 8483 (virtual_type == htons(ICMP_ECHO)) ? 8484 nk->port[iidx] : 0, NULL, 8485 pd2.ip_sum, icmpsum, 8486 pd->ip_sum, 0, AF_INET); 8487 8488 if (PF_ANEQ(pd2.dst, 8489 &nk->addr[pd2.didx], pd2.af)) 8490 pf_change_icmp(pd2.dst, NULL, NULL, 8491 &nk->addr[pd2.didx], 0, NULL, 8492 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 8493 AF_INET); 8494 8495 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 8496 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8497 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 8498 } 8499 return (PF_PASS); 8500 break; 8501 } 8502 #endif /* INET */ 8503 #ifdef INET6 8504 case IPPROTO_ICMPV6: { 8505 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 8506 8507 if (pd2.af != AF_INET6) { 8508 REASON_SET(reason, PFRES_NORM); 8509 return (PF_DROP); 8510 } 8511 8512 if (!pf_pull_hdr(pd->m, pd2.off, iih, 8513 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 8514 DPFPRINTF(PF_DEBUG_MISC, 8515 ("pf: ICMP error message too short " 8516 "(icmp6)\n")); 8517 return (PF_DROP); 8518 } 8519 8520 pf_icmp_mapping(&pd2, iih->icmp6_type, 8521 &icmp_dir, &virtual_id, &virtual_type); 8522 8523 ret = pf_icmp_state_lookup(&key, &pd2, state, 8524 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 8525 /* IPv6? try matching a multicast address */ 8526 if (ret == PF_DROP && pd2.af == AF_INET6 && 8527 icmp_dir == PF_OUT) { 8528 MPASS(*state == NULL); 8529 ret = pf_icmp_state_lookup(&key, &pd2, 8530 state, virtual_id, virtual_type, 8531 icmp_dir, &iidx, 1, 1); 8532 } 8533 if (ret >= 0) { 8534 MPASS(*state == NULL); 8535 return (ret); 8536 } 8537 8538 /* translate source/destination address, if necessary */ 8539 if ((*state)->key[PF_SK_WIRE] != 8540 (*state)->key[PF_SK_STACK]) { 8541 struct pf_state_key *nk; 8542 8543 if (PF_REVERSED_KEY(*state, pd->af)) 8544 nk = (*state)->key[pd->sidx]; 8545 else 8546 nk = (*state)->key[pd->didx]; 8547 8548 #if defined(INET) && defined(INET6) 8549 int afto, sidx, didx; 8550 8551 afto = pd->af != nk->af; 8552 8553 if (afto && (*state)->direction == PF_IN) { 8554 sidx = pd2.didx; 8555 didx = pd2.sidx; 8556 iidx = !iidx; 8557 } else { 8558 sidx = pd2.sidx; 8559 didx = pd2.didx; 8560 } 8561 8562 if (afto) { 8563 if (nk->af != AF_INET) 8564 return (PF_DROP); 8565 if (pf_translate_icmp_af(nk->af, 8566 &pd->hdr.icmp)) 8567 return (PF_DROP); 8568 m_copyback(pd->m, pd->off, 8569 sizeof(struct icmp6_hdr), 8570 (c_caddr_t)&pd->hdr.icmp6); 8571 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8572 &pd2, &nk->addr[sidx], 8573 &nk->addr[didx], pd->af, 8574 nk->af)) 8575 return (PF_DROP); 8576 pd->proto = IPPROTO_ICMP; 8577 if (pf_translate_icmp_af(nk->af, iih)) 8578 return (PF_DROP); 8579 if (virtual_type == 8580 htons(ICMP6_ECHO_REQUEST) && 8581 nk->port[iidx] != iih->icmp6_id) 8582 iih->icmp6_id = nk->port[iidx]; 8583 m_copyback(pd2.m, pd2.off, 8584 sizeof(struct icmp6_hdr), (c_caddr_t)iih); 8585 PF_ACPY(&pd->nsaddr, 8586 &nk->addr[pd2.sidx], nk->af); 8587 PF_ACPY(&pd->ndaddr, 8588 &nk->addr[pd2.didx], nk->af); 8589 pd->naf = nk->af; 8590 return (PF_AFRT); 8591 } 8592 #endif /* INET && INET6 */ 8593 8594 if (PF_ANEQ(pd2.src, 8595 &nk->addr[pd2.sidx], pd2.af) || 8596 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 8597 nk->port[pd2.sidx] != iih->icmp6_id)) 8598 pf_change_icmp(pd2.src, 8599 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8600 ? &iih->icmp6_id : NULL, 8601 daddr, &nk->addr[pd2.sidx], 8602 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 8603 ? nk->port[iidx] : 0, NULL, 8604 pd2.ip_sum, icmpsum, 8605 pd->ip_sum, 0, AF_INET6); 8606 8607 if (PF_ANEQ(pd2.dst, 8608 &nk->addr[pd2.didx], pd2.af)) 8609 pf_change_icmp(pd2.dst, NULL, NULL, 8610 &nk->addr[pd2.didx], 0, NULL, 8611 pd2.ip_sum, icmpsum, 8612 pd->ip_sum, 0, AF_INET6); 8613 8614 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8615 (caddr_t)&pd->hdr.icmp6); 8616 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 8617 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 8618 (caddr_t)iih); 8619 } 8620 return (PF_PASS); 8621 break; 8622 } 8623 #endif /* INET6 */ 8624 default: { 8625 key.af = pd2.af; 8626 key.proto = pd2.proto; 8627 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 8628 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 8629 key.port[0] = key.port[1] = 0; 8630 8631 STATE_LOOKUP(&key, *state, pd); 8632 8633 /* translate source/destination address, if necessary */ 8634 if ((*state)->key[PF_SK_WIRE] != 8635 (*state)->key[PF_SK_STACK]) { 8636 struct pf_state_key *nk = 8637 (*state)->key[pd->didx]; 8638 8639 if (PF_ANEQ(pd2.src, 8640 &nk->addr[pd2.sidx], pd2.af)) 8641 pf_change_icmp(pd2.src, NULL, daddr, 8642 &nk->addr[pd2.sidx], 0, NULL, 8643 pd2.ip_sum, icmpsum, 8644 pd->ip_sum, 0, pd2.af); 8645 8646 if (PF_ANEQ(pd2.dst, 8647 &nk->addr[pd2.didx], pd2.af)) 8648 pf_change_icmp(pd2.dst, NULL, saddr, 8649 &nk->addr[pd2.didx], 0, NULL, 8650 pd2.ip_sum, icmpsum, 8651 pd->ip_sum, 0, pd2.af); 8652 8653 switch (pd2.af) { 8654 #ifdef INET 8655 case AF_INET: 8656 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8657 (caddr_t)&pd->hdr.icmp); 8658 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 8659 break; 8660 #endif /* INET */ 8661 #ifdef INET6 8662 case AF_INET6: 8663 m_copyback(pd->m, pd->off, 8664 sizeof(struct icmp6_hdr), 8665 (caddr_t )&pd->hdr.icmp6); 8666 m_copyback(pd->m, ipoff2, sizeof(h2_6), 8667 (caddr_t )&h2_6); 8668 break; 8669 #endif /* INET6 */ 8670 } 8671 } 8672 return (PF_PASS); 8673 break; 8674 } 8675 } 8676 } 8677 } 8678 8679 /* 8680 * ipoff and off are measured from the start of the mbuf chain. 8681 * h must be at "ipoff" on the mbuf chain. 8682 */ 8683 void * 8684 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 8685 u_short *actionp, u_short *reasonp, sa_family_t af) 8686 { 8687 switch (af) { 8688 #ifdef INET 8689 case AF_INET: { 8690 const struct ip *h = mtod(m, struct ip *); 8691 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 8692 8693 if (fragoff) { 8694 if (fragoff >= len) 8695 ACTION_SET(actionp, PF_PASS); 8696 else { 8697 ACTION_SET(actionp, PF_DROP); 8698 REASON_SET(reasonp, PFRES_FRAG); 8699 } 8700 return (NULL); 8701 } 8702 if (m->m_pkthdr.len < off + len || 8703 ntohs(h->ip_len) < off + len) { 8704 ACTION_SET(actionp, PF_DROP); 8705 REASON_SET(reasonp, PFRES_SHORT); 8706 return (NULL); 8707 } 8708 break; 8709 } 8710 #endif /* INET */ 8711 #ifdef INET6 8712 case AF_INET6: { 8713 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8714 8715 if (m->m_pkthdr.len < off + len || 8716 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 8717 (unsigned)(off + len)) { 8718 ACTION_SET(actionp, PF_DROP); 8719 REASON_SET(reasonp, PFRES_SHORT); 8720 return (NULL); 8721 } 8722 break; 8723 } 8724 #endif /* INET6 */ 8725 } 8726 m_copydata(m, off, len, p); 8727 return (p); 8728 } 8729 8730 int 8731 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 8732 int rtableid) 8733 { 8734 struct ifnet *ifp; 8735 8736 /* 8737 * Skip check for addresses with embedded interface scope, 8738 * as they would always match anyway. 8739 */ 8740 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 8741 return (1); 8742 8743 if (af != AF_INET && af != AF_INET6) 8744 return (0); 8745 8746 if (kif == V_pfi_all) 8747 return (1); 8748 8749 /* Skip checks for ipsec interfaces */ 8750 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 8751 return (1); 8752 8753 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 8754 8755 switch (af) { 8756 #ifdef INET6 8757 case AF_INET6: 8758 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 8759 ifp)); 8760 #endif /* INET6 */ 8761 #ifdef INET 8762 case AF_INET: 8763 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 8764 ifp)); 8765 #endif /* INET */ 8766 } 8767 8768 return (0); 8769 } 8770 8771 #ifdef INET 8772 static void 8773 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8774 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8775 { 8776 struct mbuf *m0, *m1, *md; 8777 struct route ro; 8778 const struct sockaddr *gw = &ro.ro_dst; 8779 struct sockaddr_in *dst; 8780 struct ip *ip; 8781 struct ifnet *ifp = NULL; 8782 int error = 0; 8783 uint16_t ip_len, ip_off; 8784 uint16_t tmp; 8785 int r_dir; 8786 bool skip_test = false; 8787 8788 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8789 8790 SDT_PROBE4(pf, ip, route_to, entry, *m, pd, s, oifp); 8791 8792 if (s) { 8793 r_dir = s->direction; 8794 } else { 8795 r_dir = r->direction; 8796 } 8797 8798 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8799 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8800 __func__)); 8801 8802 if ((pd->pf_mtag == NULL && 8803 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8804 pd->pf_mtag->routed++ > 3) { 8805 m0 = *m; 8806 *m = NULL; 8807 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8808 goto bad_locked; 8809 } 8810 8811 if (pd->act.rt_kif != NULL) 8812 ifp = pd->act.rt_kif->pfik_ifp; 8813 8814 if (pd->act.rt == PF_DUPTO) { 8815 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8816 if (s != NULL) { 8817 PF_STATE_UNLOCK(s); 8818 } 8819 if (ifp == oifp) { 8820 /* When the 2nd interface is not skipped */ 8821 return; 8822 } else { 8823 m0 = *m; 8824 *m = NULL; 8825 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8826 goto bad; 8827 } 8828 } else { 8829 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8830 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8831 if (s) 8832 PF_STATE_UNLOCK(s); 8833 return; 8834 } 8835 } 8836 } else { 8837 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8838 if (pd->af == pd->naf) { 8839 pf_dummynet(pd, s, r, m); 8840 if (s) 8841 PF_STATE_UNLOCK(s); 8842 return; 8843 } else { 8844 if (r_dir == PF_IN) { 8845 skip_test = true; 8846 } 8847 } 8848 } 8849 8850 /* 8851 * If we're actually doing route-to and af-to and are in the 8852 * reply direction. 8853 */ 8854 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 8855 pd->af != pd->naf) { 8856 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 8857 /* Un-set ifp so we do a plain route lookup. */ 8858 ifp = NULL; 8859 } 8860 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 8861 /* Un-set ifp so we do a plain route lookup. */ 8862 ifp = NULL; 8863 } 8864 } 8865 m0 = *m; 8866 } 8867 8868 ip = mtod(m0, struct ip *); 8869 8870 bzero(&ro, sizeof(ro)); 8871 dst = (struct sockaddr_in *)&ro.ro_dst; 8872 dst->sin_family = AF_INET; 8873 dst->sin_len = sizeof(struct sockaddr_in); 8874 dst->sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 8875 8876 if (s != NULL){ 8877 if (ifp == NULL && (pd->af != pd->naf)) { 8878 /* We're in the AFTO case. Do a route lookup. */ 8879 const struct nhop_object *nh; 8880 nh = fib4_lookup(M_GETFIB(*m), ip->ip_dst, 0, NHR_NONE, 0); 8881 if (nh) { 8882 ifp = nh->nh_ifp; 8883 8884 /* Use the gateway if needed. */ 8885 if (nh->nh_flags & NHF_GATEWAY) { 8886 gw = &nh->gw_sa; 8887 ro.ro_flags |= RT_HAS_GW; 8888 } else { 8889 dst->sin_addr = ip->ip_dst; 8890 } 8891 8892 /* 8893 * Bind to the correct interface if we're 8894 * if-bound. We don't know which interface 8895 * that will be until here, so we've inserted 8896 * the state on V_pf_all. Fix that now. 8897 */ 8898 if (s->kif == V_pfi_all && ifp != NULL && 8899 r->rule_flag & PFRULE_IFBOUND) 8900 s->kif = ifp->if_pf_kif; 8901 } 8902 } 8903 8904 if (r->rule_flag & PFRULE_IFBOUND && 8905 pd->act.rt == PF_REPLYTO && 8906 s->kif == V_pfi_all) { 8907 s->kif = pd->act.rt_kif; 8908 s->orig_kif = oifp->if_pf_kif; 8909 } 8910 8911 PF_STATE_UNLOCK(s); 8912 } 8913 8914 if (ifp == NULL) { 8915 m0 = *m; 8916 *m = NULL; 8917 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8918 goto bad; 8919 } 8920 8921 if (pd->dir == PF_IN && !skip_test) { 8922 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8923 &pd->act) != PF_PASS) { 8924 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8925 goto bad; 8926 } else if (m0 == NULL) { 8927 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8928 goto done; 8929 } 8930 if (m0->m_len < sizeof(struct ip)) { 8931 DPFPRINTF(PF_DEBUG_URGENT, 8932 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 8933 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 8934 goto bad; 8935 } 8936 ip = mtod(m0, struct ip *); 8937 } 8938 8939 if (ifp->if_flags & IFF_LOOPBACK) 8940 m0->m_flags |= M_SKIP_FIREWALL; 8941 8942 ip_len = ntohs(ip->ip_len); 8943 ip_off = ntohs(ip->ip_off); 8944 8945 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 8946 m0->m_pkthdr.csum_flags |= CSUM_IP; 8947 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 8948 in_delayed_cksum(m0); 8949 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 8950 } 8951 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 8952 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 8953 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 8954 } 8955 8956 if (pd->dir == PF_IN) { 8957 /* 8958 * Make sure dummynet gets the correct direction, in case it needs to 8959 * re-inject later. 8960 */ 8961 pd->dir = PF_OUT; 8962 8963 /* 8964 * The following processing is actually the rest of the inbound processing, even 8965 * though we've marked it as outbound (so we don't look through dummynet) and it 8966 * happens after the outbound processing (pf_test(PF_OUT) above). 8967 * Swap the dummynet pipe numbers, because it's going to come to the wrong 8968 * conclusion about what direction it's processing, and we can't fix it or it 8969 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 8970 * decision will pick the right pipe, and everything will mostly work as expected. 8971 */ 8972 tmp = pd->act.dnrpipe; 8973 pd->act.dnrpipe = pd->act.dnpipe; 8974 pd->act.dnpipe = tmp; 8975 } 8976 8977 /* 8978 * If small enough for interface, or the interface will take 8979 * care of the fragmentation for us, we can just send directly. 8980 */ 8981 if (ip_len <= ifp->if_mtu || 8982 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 8983 ip->ip_sum = 0; 8984 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 8985 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 8986 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 8987 } 8988 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 8989 8990 md = m0; 8991 error = pf_dummynet_route(pd, s, r, ifp, gw, &md); 8992 if (md != NULL) { 8993 error = (*ifp->if_output)(ifp, md, gw, &ro); 8994 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 8995 } 8996 goto done; 8997 } 8998 8999 /* Balk when DF bit is set or the interface didn't support TSO. */ 9000 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 9001 error = EMSGSIZE; 9002 KMOD_IPSTAT_INC(ips_cantfrag); 9003 if (pd->act.rt != PF_DUPTO) { 9004 if (s && s->nat_rule != NULL) 9005 MPASS(m0 == pd->m); 9006 PACKET_UNDO_NAT(pd, 9007 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 9008 s); 9009 9010 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 9011 ifp->if_mtu); 9012 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9013 goto done; 9014 } else { 9015 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9016 goto bad; 9017 } 9018 } 9019 9020 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 9021 if (error) { 9022 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9023 goto bad; 9024 } 9025 9026 for (; m0; m0 = m1) { 9027 m1 = m0->m_nextpkt; 9028 m0->m_nextpkt = NULL; 9029 if (error == 0) { 9030 m_clrprotoflags(m0); 9031 md = m0; 9032 pd->pf_mtag = pf_find_mtag(md); 9033 error = pf_dummynet_route(pd, s, r, ifp, 9034 gw, &md); 9035 if (md != NULL) { 9036 error = (*ifp->if_output)(ifp, md, gw, &ro); 9037 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 9038 } 9039 } else 9040 m_freem(m0); 9041 } 9042 9043 if (error == 0) 9044 KMOD_IPSTAT_INC(ips_fragmented); 9045 9046 done: 9047 if (pd->act.rt != PF_DUPTO) 9048 *m = NULL; 9049 return; 9050 9051 bad_locked: 9052 if (s) 9053 PF_STATE_UNLOCK(s); 9054 bad: 9055 m_freem(m0); 9056 goto done; 9057 } 9058 #endif /* INET */ 9059 9060 #ifdef INET6 9061 static void 9062 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 9063 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9064 { 9065 struct mbuf *m0, *md; 9066 struct m_tag *mtag; 9067 struct sockaddr_in6 dst; 9068 struct ip6_hdr *ip6; 9069 struct ifnet *ifp = NULL; 9070 int r_dir; 9071 bool skip_test = false; 9072 9073 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 9074 9075 SDT_PROBE4(pf, ip6, route_to, entry, *m, pd, s, oifp); 9076 9077 if (s) { 9078 r_dir = s->direction; 9079 } else { 9080 r_dir = r->direction; 9081 } 9082 9083 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9084 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9085 __func__)); 9086 9087 if ((pd->pf_mtag == NULL && 9088 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 9089 pd->pf_mtag->routed++ > 3) { 9090 m0 = *m; 9091 *m = NULL; 9092 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9093 goto bad_locked; 9094 } 9095 9096 if (pd->act.rt_kif != NULL) 9097 ifp = pd->act.rt_kif->pfik_ifp; 9098 9099 if (pd->act.rt == PF_DUPTO) { 9100 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9101 if (s != NULL) { 9102 PF_STATE_UNLOCK(s); 9103 } 9104 if (ifp == oifp) { 9105 /* When the 2nd interface is not skipped */ 9106 return; 9107 } else { 9108 m0 = *m; 9109 *m = NULL; 9110 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9111 goto bad; 9112 } 9113 } else { 9114 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9115 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 9116 if (s) 9117 PF_STATE_UNLOCK(s); 9118 return; 9119 } 9120 } 9121 } else { 9122 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9123 if (pd->af == pd->naf) { 9124 pf_dummynet(pd, s, r, m); 9125 if (s) 9126 PF_STATE_UNLOCK(s); 9127 return; 9128 } else { 9129 if (r_dir == PF_IN) { 9130 skip_test = true; 9131 } 9132 } 9133 } 9134 9135 /* 9136 * If we're actually doing route-to and af-to and are in the 9137 * reply direction. 9138 */ 9139 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9140 pd->af != pd->naf) { 9141 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 9142 /* Un-set ifp so we do a plain route lookup. */ 9143 ifp = NULL; 9144 } 9145 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 9146 /* Un-set ifp so we do a plain route lookup. */ 9147 ifp = NULL; 9148 } 9149 } 9150 m0 = *m; 9151 } 9152 9153 ip6 = mtod(m0, struct ip6_hdr *); 9154 9155 bzero(&dst, sizeof(dst)); 9156 dst.sin6_family = AF_INET6; 9157 dst.sin6_len = sizeof(dst); 9158 PF_ACPY((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, AF_INET6); 9159 9160 if (s != NULL) { 9161 if (ifp == NULL && (pd->af != pd->naf)) { 9162 const struct nhop_object *nh; 9163 nh = fib6_lookup(M_GETFIB(*m), &ip6->ip6_dst, 0, NHR_NONE, 0); 9164 if (nh) { 9165 ifp = nh->nh_ifp; 9166 9167 /* Use the gateway if needed. */ 9168 if (nh->nh_flags & NHF_GATEWAY) 9169 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 9170 sizeof(dst.sin6_addr)); 9171 else 9172 dst.sin6_addr = ip6->ip6_dst; 9173 9174 /* 9175 * Bind to the correct interface if we're 9176 * if-bound. We don't know which interface 9177 * that will be until here, so we've inserted 9178 * the state on V_pf_all. Fix that now. 9179 */ 9180 if (s->kif == V_pfi_all && ifp != NULL && 9181 r->rule_flag & PFRULE_IFBOUND) 9182 s->kif = ifp->if_pf_kif; 9183 } 9184 } 9185 9186 if (r->rule_flag & PFRULE_IFBOUND && 9187 pd->act.rt == PF_REPLYTO && 9188 s->kif == V_pfi_all) { 9189 s->kif = pd->act.rt_kif; 9190 s->orig_kif = oifp->if_pf_kif; 9191 } 9192 9193 PF_STATE_UNLOCK(s); 9194 } 9195 9196 if (pd->af != pd->naf) { 9197 struct udphdr *uh = &pd->hdr.udp; 9198 9199 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 9200 uh->uh_sum = in6_cksum_pseudo(ip6, 9201 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 9202 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 9203 } 9204 } 9205 9206 if (ifp == NULL) { 9207 m0 = *m; 9208 *m = NULL; 9209 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9210 goto bad; 9211 } 9212 9213 if (pd->dir == PF_IN && !skip_test) { 9214 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 9215 ifp, &m0, inp, &pd->act) != PF_PASS) { 9216 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9217 goto bad; 9218 } else if (m0 == NULL) { 9219 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9220 goto done; 9221 } 9222 if (m0->m_len < sizeof(struct ip6_hdr)) { 9223 DPFPRINTF(PF_DEBUG_URGENT, 9224 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 9225 __func__)); 9226 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9227 goto bad; 9228 } 9229 ip6 = mtod(m0, struct ip6_hdr *); 9230 } 9231 9232 if (ifp->if_flags & IFF_LOOPBACK) 9233 m0->m_flags |= M_SKIP_FIREWALL; 9234 9235 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 9236 ~ifp->if_hwassist) { 9237 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 9238 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 9239 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 9240 } 9241 9242 if (pd->dir == PF_IN) { 9243 uint16_t tmp; 9244 /* 9245 * Make sure dummynet gets the correct direction, in case it needs to 9246 * re-inject later. 9247 */ 9248 pd->dir = PF_OUT; 9249 9250 /* 9251 * The following processing is actually the rest of the inbound processing, even 9252 * though we've marked it as outbound (so we don't look through dummynet) and it 9253 * happens after the outbound processing (pf_test(PF_OUT) above). 9254 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9255 * conclusion about what direction it's processing, and we can't fix it or it 9256 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9257 * decision will pick the right pipe, and everything will mostly work as expected. 9258 */ 9259 tmp = pd->act.dnrpipe; 9260 pd->act.dnrpipe = pd->act.dnpipe; 9261 pd->act.dnpipe = tmp; 9262 } 9263 9264 /* 9265 * If the packet is too large for the outgoing interface, 9266 * send back an icmp6 error. 9267 */ 9268 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 9269 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 9270 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 9271 if (mtag != NULL) { 9272 int ret __sdt_used; 9273 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 9274 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9275 goto done; 9276 } 9277 9278 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 9279 md = m0; 9280 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 9281 if (md != NULL) { 9282 int ret __sdt_used; 9283 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 9284 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 9285 } 9286 } 9287 else { 9288 in6_ifstat_inc(ifp, ifs6_in_toobig); 9289 if (pd->act.rt != PF_DUPTO) { 9290 if (s && s->nat_rule != NULL) 9291 MPASS(m0 == pd->m); 9292 PACKET_UNDO_NAT(pd, 9293 ((caddr_t)ip6 - m0->m_data) + 9294 sizeof(struct ip6_hdr), s); 9295 9296 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 9297 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9298 } else { 9299 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 9300 goto bad; 9301 } 9302 } 9303 9304 done: 9305 if (pd->act.rt != PF_DUPTO) 9306 *m = NULL; 9307 return; 9308 9309 bad_locked: 9310 if (s) 9311 PF_STATE_UNLOCK(s); 9312 bad: 9313 m_freem(m0); 9314 goto done; 9315 } 9316 #endif /* INET6 */ 9317 9318 /* 9319 * FreeBSD supports cksum offloads for the following drivers. 9320 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 9321 * 9322 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 9323 * network driver performed cksum including pseudo header, need to verify 9324 * csum_data 9325 * CSUM_DATA_VALID : 9326 * network driver performed cksum, needs to additional pseudo header 9327 * cksum computation with partial csum_data(i.e. lack of H/W support for 9328 * pseudo header, for instance sk(4) and possibly gem(4)) 9329 * 9330 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 9331 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 9332 * TCP/UDP layer. 9333 * Also, set csum_data to 0xffff to force cksum validation. 9334 */ 9335 static int 9336 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 9337 { 9338 u_int16_t sum = 0; 9339 int hw_assist = 0; 9340 struct ip *ip; 9341 9342 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 9343 return (1); 9344 if (m->m_pkthdr.len < off + len) 9345 return (1); 9346 9347 switch (p) { 9348 case IPPROTO_TCP: 9349 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9350 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9351 sum = m->m_pkthdr.csum_data; 9352 } else { 9353 ip = mtod(m, struct ip *); 9354 sum = in_pseudo(ip->ip_src.s_addr, 9355 ip->ip_dst.s_addr, htonl((u_short)len + 9356 m->m_pkthdr.csum_data + IPPROTO_TCP)); 9357 } 9358 sum ^= 0xffff; 9359 ++hw_assist; 9360 } 9361 break; 9362 case IPPROTO_UDP: 9363 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 9364 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 9365 sum = m->m_pkthdr.csum_data; 9366 } else { 9367 ip = mtod(m, struct ip *); 9368 sum = in_pseudo(ip->ip_src.s_addr, 9369 ip->ip_dst.s_addr, htonl((u_short)len + 9370 m->m_pkthdr.csum_data + IPPROTO_UDP)); 9371 } 9372 sum ^= 0xffff; 9373 ++hw_assist; 9374 } 9375 break; 9376 case IPPROTO_ICMP: 9377 #ifdef INET6 9378 case IPPROTO_ICMPV6: 9379 #endif /* INET6 */ 9380 break; 9381 default: 9382 return (1); 9383 } 9384 9385 if (!hw_assist) { 9386 switch (af) { 9387 case AF_INET: 9388 if (m->m_len < sizeof(struct ip)) 9389 return (1); 9390 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 9391 break; 9392 #ifdef INET6 9393 case AF_INET6: 9394 if (m->m_len < sizeof(struct ip6_hdr)) 9395 return (1); 9396 sum = in6_cksum(m, p, off, len); 9397 break; 9398 #endif /* INET6 */ 9399 } 9400 } 9401 if (sum) { 9402 switch (p) { 9403 case IPPROTO_TCP: 9404 { 9405 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 9406 break; 9407 } 9408 case IPPROTO_UDP: 9409 { 9410 KMOD_UDPSTAT_INC(udps_badsum); 9411 break; 9412 } 9413 #ifdef INET 9414 case IPPROTO_ICMP: 9415 { 9416 KMOD_ICMPSTAT_INC(icps_checksum); 9417 break; 9418 } 9419 #endif 9420 #ifdef INET6 9421 case IPPROTO_ICMPV6: 9422 { 9423 KMOD_ICMP6STAT_INC(icp6s_checksum); 9424 break; 9425 } 9426 #endif /* INET6 */ 9427 } 9428 return (1); 9429 } else { 9430 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 9431 m->m_pkthdr.csum_flags |= 9432 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 9433 m->m_pkthdr.csum_data = 0xffff; 9434 } 9435 } 9436 return (0); 9437 } 9438 9439 static bool 9440 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 9441 const struct pf_kstate *s, struct ip_fw_args *dnflow) 9442 { 9443 int dndir = r->direction; 9444 9445 if (s && dndir == PF_INOUT) { 9446 dndir = s->direction; 9447 } else if (dndir == PF_INOUT) { 9448 /* Assume primary direction. Happens when we've set dnpipe in 9449 * the ethernet level code. */ 9450 dndir = pd->dir; 9451 } 9452 9453 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 9454 return (false); 9455 9456 memset(dnflow, 0, sizeof(*dnflow)); 9457 9458 if (pd->dport != NULL) 9459 dnflow->f_id.dst_port = ntohs(*pd->dport); 9460 if (pd->sport != NULL) 9461 dnflow->f_id.src_port = ntohs(*pd->sport); 9462 9463 if (pd->dir == PF_IN) 9464 dnflow->flags |= IPFW_ARGS_IN; 9465 else 9466 dnflow->flags |= IPFW_ARGS_OUT; 9467 9468 if (pd->dir != dndir && pd->act.dnrpipe) { 9469 dnflow->rule.info = pd->act.dnrpipe; 9470 } 9471 else if (pd->dir == dndir && pd->act.dnpipe) { 9472 dnflow->rule.info = pd->act.dnpipe; 9473 } 9474 else { 9475 return (false); 9476 } 9477 9478 dnflow->rule.info |= IPFW_IS_DUMMYNET; 9479 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 9480 dnflow->rule.info |= IPFW_IS_PIPE; 9481 9482 dnflow->f_id.proto = pd->proto; 9483 dnflow->f_id.extra = dnflow->rule.info; 9484 switch (pd->naf) { 9485 case AF_INET: 9486 dnflow->f_id.addr_type = 4; 9487 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 9488 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 9489 break; 9490 case AF_INET6: 9491 dnflow->flags |= IPFW_ARGS_IP6; 9492 dnflow->f_id.addr_type = 6; 9493 dnflow->f_id.src_ip6 = pd->src->v6; 9494 dnflow->f_id.dst_ip6 = pd->dst->v6; 9495 break; 9496 } 9497 9498 return (true); 9499 } 9500 9501 int 9502 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 9503 struct inpcb *inp) 9504 { 9505 struct pfi_kkif *kif; 9506 struct mbuf *m = *m0; 9507 9508 M_ASSERTPKTHDR(m); 9509 MPASS(ifp->if_vnet == curvnet); 9510 NET_EPOCH_ASSERT(); 9511 9512 if (!V_pf_status.running) 9513 return (PF_PASS); 9514 9515 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9516 9517 if (kif == NULL) { 9518 DPFPRINTF(PF_DEBUG_URGENT, 9519 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 9520 return (PF_DROP); 9521 } 9522 if (kif->pfik_flags & PFI_IFLAG_SKIP) 9523 return (PF_PASS); 9524 9525 if (m->m_flags & M_SKIP_FIREWALL) 9526 return (PF_PASS); 9527 9528 if (__predict_false(! M_WRITABLE(*m0))) { 9529 m = *m0 = m_unshare(*m0, M_NOWAIT); 9530 if (*m0 == NULL) 9531 return (PF_DROP); 9532 } 9533 9534 /* Stateless! */ 9535 return (pf_test_eth_rule(dir, kif, m0)); 9536 } 9537 9538 static __inline void 9539 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 9540 { 9541 struct m_tag *mtag; 9542 9543 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 9544 9545 /* dummynet adds this tag, but pf does not need it, 9546 * and keeping it creates unexpected behavior, 9547 * e.g. in case of divert(4) usage right after dummynet. */ 9548 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9549 if (mtag != NULL) 9550 m_tag_delete(m, mtag); 9551 } 9552 9553 static int 9554 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 9555 struct pf_krule *r, struct mbuf **m0) 9556 { 9557 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 9558 } 9559 9560 static int 9561 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 9562 struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa, 9563 struct mbuf **m0) 9564 { 9565 struct ip_fw_args dnflow; 9566 9567 NET_EPOCH_ASSERT(); 9568 9569 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 9570 return (0); 9571 9572 if (ip_dn_io_ptr == NULL) { 9573 m_freem(*m0); 9574 *m0 = NULL; 9575 return (ENOMEM); 9576 } 9577 9578 if (pd->pf_mtag == NULL && 9579 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 9580 m_freem(*m0); 9581 *m0 = NULL; 9582 return (ENOMEM); 9583 } 9584 9585 if (ifp != NULL) { 9586 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 9587 9588 pd->pf_mtag->if_index = ifp->if_index; 9589 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 9590 9591 MPASS(sa != NULL); 9592 9593 switch (sa->sa_family) { 9594 case AF_INET: 9595 memcpy(&pd->pf_mtag->dst, sa, 9596 sizeof(struct sockaddr_in)); 9597 break; 9598 case AF_INET6: 9599 memcpy(&pd->pf_mtag->dst, sa, 9600 sizeof(struct sockaddr_in6)); 9601 break; 9602 } 9603 } 9604 9605 if (s != NULL && s->nat_rule != NULL && 9606 s->nat_rule->action == PF_RDR && 9607 ( 9608 #ifdef INET 9609 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 9610 #endif /* INET */ 9611 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 9612 /* 9613 * If we're redirecting to loopback mark this packet 9614 * as being local. Otherwise it might get dropped 9615 * if dummynet re-injects. 9616 */ 9617 (*m0)->m_pkthdr.rcvif = V_loif; 9618 } 9619 9620 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 9621 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 9622 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 9623 ip_dn_io_ptr(m0, &dnflow); 9624 if (*m0 != NULL) { 9625 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9626 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 9627 } 9628 } 9629 9630 return (0); 9631 } 9632 9633 #ifdef INET6 9634 static int 9635 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 9636 u_short *reason) 9637 { 9638 struct ip6_opt opt; 9639 struct ip6_opt_jumbo jumbo; 9640 9641 while (off < end) { 9642 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 9643 sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) { 9644 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 9645 return (PF_DROP); 9646 } 9647 if (opt.ip6o_type == IP6OPT_PAD1) { 9648 off++; 9649 continue; 9650 } 9651 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL, 9652 reason, AF_INET6)) { 9653 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 9654 return (PF_DROP); 9655 } 9656 if (off + sizeof(opt) + opt.ip6o_len > end) { 9657 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 9658 REASON_SET(reason, PFRES_IPOPTIONS); 9659 return (PF_DROP); 9660 } 9661 switch (opt.ip6o_type) { 9662 case IP6OPT_JUMBO: 9663 if (pd->jumbolen != 0) { 9664 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 9665 REASON_SET(reason, PFRES_IPOPTIONS); 9666 return (PF_DROP); 9667 } 9668 if (ntohs(h->ip6_plen) != 0) { 9669 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 9670 REASON_SET(reason, PFRES_IPOPTIONS); 9671 return (PF_DROP); 9672 } 9673 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL, 9674 reason, AF_INET6)) { 9675 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 9676 return (PF_DROP); 9677 } 9678 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 9679 sizeof(pd->jumbolen)); 9680 pd->jumbolen = ntohl(pd->jumbolen); 9681 if (pd->jumbolen < IPV6_MAXPACKET) { 9682 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 9683 REASON_SET(reason, PFRES_IPOPTIONS); 9684 return (PF_DROP); 9685 } 9686 break; 9687 default: 9688 break; 9689 } 9690 off += sizeof(opt) + opt.ip6o_len; 9691 } 9692 9693 return (PF_PASS); 9694 } 9695 9696 int 9697 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 9698 { 9699 struct ip6_frag frag; 9700 struct ip6_ext ext; 9701 struct ip6_rthdr rthdr; 9702 uint32_t end; 9703 int fraghdr_cnt = 0, rthdr_cnt = 0; 9704 9705 pd->off += sizeof(struct ip6_hdr); 9706 end = pd->off + ntohs(h->ip6_plen); 9707 pd->fragoff = pd->extoff = pd->jumbolen = 0; 9708 pd->proto = h->ip6_nxt; 9709 for (;;) { 9710 switch (pd->proto) { 9711 case IPPROTO_FRAGMENT: 9712 if (fraghdr_cnt++) { 9713 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 9714 REASON_SET(reason, PFRES_FRAG); 9715 return (PF_DROP); 9716 } 9717 /* jumbo payload packets cannot be fragmented */ 9718 if (pd->jumbolen != 0) { 9719 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 9720 REASON_SET(reason, PFRES_FRAG); 9721 return (PF_DROP); 9722 } 9723 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 9724 NULL, reason, AF_INET6)) { 9725 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 9726 return (PF_DROP); 9727 } 9728 /* stop walking over non initial fragments */ 9729 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 9730 pd->fragoff = pd->off; 9731 return (PF_PASS); 9732 } 9733 /* RFC6946: reassemble only non atomic fragments */ 9734 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 9735 pd->fragoff = pd->off; 9736 pd->off += sizeof(frag); 9737 pd->proto = frag.ip6f_nxt; 9738 break; 9739 case IPPROTO_ROUTING: 9740 if (rthdr_cnt++) { 9741 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 9742 REASON_SET(reason, PFRES_IPOPTIONS); 9743 return (PF_DROP); 9744 } 9745 /* fragments may be short */ 9746 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 9747 pd->off = pd->fragoff; 9748 pd->proto = IPPROTO_FRAGMENT; 9749 return (PF_PASS); 9750 } 9751 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 9752 NULL, reason, AF_INET6)) { 9753 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 9754 return (PF_DROP); 9755 } 9756 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 9757 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 9758 REASON_SET(reason, PFRES_IPOPTIONS); 9759 return (PF_DROP); 9760 } 9761 /* FALLTHROUGH */ 9762 case IPPROTO_AH: 9763 case IPPROTO_HOPOPTS: 9764 case IPPROTO_DSTOPTS: 9765 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 9766 NULL, reason, AF_INET6)) { 9767 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 9768 return (PF_DROP); 9769 } 9770 /* fragments may be short */ 9771 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 9772 pd->off = pd->fragoff; 9773 pd->proto = IPPROTO_FRAGMENT; 9774 return (PF_PASS); 9775 } 9776 /* reassembly needs the ext header before the frag */ 9777 if (pd->fragoff == 0) 9778 pd->extoff = pd->off; 9779 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0) { 9780 if (pf_walk_option6(pd, h, 9781 pd->off + sizeof(ext), 9782 pd->off + (ext.ip6e_len + 1) * 8, reason) 9783 != PF_PASS) 9784 return (PF_DROP); 9785 if (ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 9786 DPFPRINTF(PF_DEBUG_MISC, 9787 ("IPv6 missing jumbo")); 9788 REASON_SET(reason, PFRES_IPOPTIONS); 9789 return (PF_DROP); 9790 } 9791 } 9792 if (pd->proto == IPPROTO_AH) 9793 pd->off += (ext.ip6e_len + 2) * 4; 9794 else 9795 pd->off += (ext.ip6e_len + 1) * 8; 9796 pd->proto = ext.ip6e_nxt; 9797 break; 9798 case IPPROTO_TCP: 9799 case IPPROTO_UDP: 9800 case IPPROTO_SCTP: 9801 case IPPROTO_ICMPV6: 9802 /* fragments may be short, ignore inner header then */ 9803 if (pd->fragoff != 0 && end < pd->off + 9804 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 9805 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 9806 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 9807 sizeof(struct icmp6_hdr))) { 9808 pd->off = pd->fragoff; 9809 pd->proto = IPPROTO_FRAGMENT; 9810 } 9811 /* FALLTHROUGH */ 9812 default: 9813 return (PF_PASS); 9814 } 9815 } 9816 } 9817 #endif /* INET6 */ 9818 9819 static void 9820 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 9821 { 9822 memset(pd, 0, sizeof(*pd)); 9823 pd->pf_mtag = pf_find_mtag(m); 9824 pd->m = m; 9825 } 9826 9827 static int 9828 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 9829 u_short *action, u_short *reason, struct pfi_kkif *kif, 9830 struct pf_rule_actions *default_actions) 9831 { 9832 pd->dir = dir; 9833 pd->kif = kif; 9834 pd->m = *m0; 9835 pd->sidx = (dir == PF_IN) ? 0 : 1; 9836 pd->didx = (dir == PF_IN) ? 1 : 0; 9837 pd->af = pd->naf = af; 9838 9839 TAILQ_INIT(&pd->sctp_multihome_jobs); 9840 if (default_actions != NULL) 9841 memcpy(&pd->act, default_actions, sizeof(pd->act)); 9842 9843 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 9844 pd->act.dnpipe = pd->pf_mtag->dnpipe; 9845 pd->act.flags = pd->pf_mtag->dnflags; 9846 } 9847 9848 switch (af) { 9849 #ifdef INET 9850 case AF_INET: { 9851 struct ip *h; 9852 9853 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 9854 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 9855 DPFPRINTF(PF_DEBUG_URGENT, 9856 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 9857 *action = PF_DROP; 9858 REASON_SET(reason, PFRES_SHORT); 9859 return (-1); 9860 } 9861 9862 if (pf_normalize_ip(reason, pd) != PF_PASS) { 9863 /* We do IP header normalization and packet reassembly here */ 9864 *m0 = pd->m; 9865 *action = PF_DROP; 9866 return (-1); 9867 } 9868 *m0 = pd->m; 9869 9870 h = mtod(pd->m, struct ip *); 9871 pd->off = h->ip_hl << 2; 9872 if (pd->off < (int)sizeof(*h)) { 9873 *action = PF_DROP; 9874 REASON_SET(reason, PFRES_SHORT); 9875 return (-1); 9876 } 9877 pd->src = (struct pf_addr *)&h->ip_src; 9878 pd->dst = (struct pf_addr *)&h->ip_dst; 9879 PF_ACPY(&pd->osrc, pd->src, af); 9880 PF_ACPY(&pd->odst, pd->dst, af); 9881 pd->ip_sum = &h->ip_sum; 9882 pd->virtual_proto = pd->proto = h->ip_p; 9883 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 9884 pd->ttl = h->ip_ttl; 9885 pd->tot_len = ntohs(h->ip_len); 9886 pd->act.rtableid = -1; 9887 pd->df = h->ip_off & htons(IP_DF); 9888 9889 if (h->ip_hl > 5) /* has options */ 9890 pd->badopts++; 9891 9892 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 9893 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9894 9895 break; 9896 } 9897 #endif /* INET */ 9898 #ifdef INET6 9899 case AF_INET6: { 9900 struct ip6_hdr *h; 9901 9902 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 9903 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 9904 DPFPRINTF(PF_DEBUG_URGENT, 9905 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 9906 ", pullup failed\n")); 9907 *action = PF_DROP; 9908 REASON_SET(reason, PFRES_SHORT); 9909 return (-1); 9910 } 9911 9912 h = mtod(pd->m, struct ip6_hdr *); 9913 pd->off = 0; 9914 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9915 *action = PF_DROP; 9916 return (-1); 9917 } 9918 9919 h = mtod(pd->m, struct ip6_hdr *); 9920 pd->src = (struct pf_addr *)&h->ip6_src; 9921 pd->dst = (struct pf_addr *)&h->ip6_dst; 9922 PF_ACPY(&pd->osrc, pd->src, af); 9923 PF_ACPY(&pd->odst, pd->dst, af); 9924 pd->ip_sum = NULL; 9925 pd->tos = IPV6_DSCP(h); 9926 pd->ttl = h->ip6_hlim; 9927 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 9928 pd->virtual_proto = pd->proto = h->ip6_nxt; 9929 pd->act.rtableid = -1; 9930 9931 if (pd->fragoff != 0) 9932 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9933 9934 /* 9935 * we do not support jumbogram. if we keep going, zero ip6_plen 9936 * will do something bad, so drop the packet for now. 9937 */ 9938 if (htons(h->ip6_plen) == 0) { 9939 *action = PF_DROP; 9940 return (-1); 9941 } 9942 9943 /* We do IP header normalization and packet reassembly here */ 9944 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 9945 PF_PASS) { 9946 *m0 = pd->m; 9947 *action = PF_DROP; 9948 return (-1); 9949 } 9950 *m0 = pd->m; 9951 if (pd->m == NULL) { 9952 /* packet sits in reassembly queue, no error */ 9953 *action = PF_PASS; 9954 return (-1); 9955 } 9956 9957 /* Update pointers into the packet. */ 9958 h = mtod(pd->m, struct ip6_hdr *); 9959 pd->src = (struct pf_addr *)&h->ip6_src; 9960 pd->dst = (struct pf_addr *)&h->ip6_dst; 9961 9962 pd->off = 0; 9963 9964 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 9965 *action = PF_DROP; 9966 return (-1); 9967 } 9968 9969 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 9970 /* 9971 * Reassembly may have changed the next protocol from 9972 * fragment to something else, so update. 9973 */ 9974 pd->virtual_proto = pd->proto; 9975 MPASS(pd->fragoff == 0); 9976 } 9977 9978 if (pd->fragoff != 0) 9979 pd->virtual_proto = PF_VPROTO_FRAGMENT; 9980 9981 break; 9982 } 9983 #endif /* INET6 */ 9984 default: 9985 panic("pf_setup_pdesc called with illegal af %u", af); 9986 } 9987 9988 switch (pd->virtual_proto) { 9989 case IPPROTO_TCP: { 9990 struct tcphdr *th = &pd->hdr.tcp; 9991 9992 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 9993 reason, af)) { 9994 *action = PF_DROP; 9995 REASON_SET(reason, PFRES_SHORT); 9996 return (-1); 9997 } 9998 pd->hdrlen = sizeof(*th); 9999 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 10000 pd->sport = &th->th_sport; 10001 pd->dport = &th->th_dport; 10002 pd->pcksum = &th->th_sum; 10003 break; 10004 } 10005 case IPPROTO_UDP: { 10006 struct udphdr *uh = &pd->hdr.udp; 10007 10008 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 10009 reason, af)) { 10010 *action = PF_DROP; 10011 REASON_SET(reason, PFRES_SHORT); 10012 return (-1); 10013 } 10014 pd->hdrlen = sizeof(*uh); 10015 if (uh->uh_dport == 0 || 10016 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 10017 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 10018 *action = PF_DROP; 10019 REASON_SET(reason, PFRES_SHORT); 10020 return (-1); 10021 } 10022 pd->sport = &uh->uh_sport; 10023 pd->dport = &uh->uh_dport; 10024 pd->pcksum = &uh->uh_sum; 10025 break; 10026 } 10027 case IPPROTO_SCTP: { 10028 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 10029 action, reason, af)) { 10030 *action = PF_DROP; 10031 REASON_SET(reason, PFRES_SHORT); 10032 return (-1); 10033 } 10034 pd->hdrlen = sizeof(pd->hdr.sctp); 10035 pd->p_len = pd->tot_len - pd->off; 10036 10037 pd->sport = &pd->hdr.sctp.src_port; 10038 pd->dport = &pd->hdr.sctp.dest_port; 10039 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 10040 *action = PF_DROP; 10041 REASON_SET(reason, PFRES_SHORT); 10042 return (-1); 10043 } 10044 if (pf_scan_sctp(pd) != PF_PASS) { 10045 *action = PF_DROP; 10046 REASON_SET(reason, PFRES_SHORT); 10047 return (-1); 10048 } 10049 /* 10050 * Placeholder. The SCTP checksum is 32-bits, but 10051 * pf_test_state() expects to update a 16-bit checksum. 10052 * Provide a dummy value which we'll subsequently ignore. 10053 */ 10054 pd->pcksum = &pd->sctp_dummy_sum; 10055 break; 10056 } 10057 case IPPROTO_ICMP: { 10058 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 10059 action, reason, af)) { 10060 *action = PF_DROP; 10061 REASON_SET(reason, PFRES_SHORT); 10062 return (-1); 10063 } 10064 pd->hdrlen = ICMP_MINLEN; 10065 break; 10066 } 10067 #ifdef INET6 10068 case IPPROTO_ICMPV6: { 10069 size_t icmp_hlen = sizeof(struct icmp6_hdr); 10070 10071 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10072 action, reason, af)) { 10073 *action = PF_DROP; 10074 REASON_SET(reason, PFRES_SHORT); 10075 return (-1); 10076 } 10077 /* ICMP headers we look further into to match state */ 10078 switch (pd->hdr.icmp6.icmp6_type) { 10079 case MLD_LISTENER_QUERY: 10080 case MLD_LISTENER_REPORT: 10081 icmp_hlen = sizeof(struct mld_hdr); 10082 break; 10083 case ND_NEIGHBOR_SOLICIT: 10084 case ND_NEIGHBOR_ADVERT: 10085 icmp_hlen = sizeof(struct nd_neighbor_solicit); 10086 break; 10087 } 10088 if (icmp_hlen > sizeof(struct icmp6_hdr) && 10089 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 10090 action, reason, af)) { 10091 *action = PF_DROP; 10092 REASON_SET(reason, PFRES_SHORT); 10093 return (-1); 10094 } 10095 pd->hdrlen = icmp_hlen; 10096 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 10097 break; 10098 } 10099 #endif /* INET6 */ 10100 } 10101 10102 if (pd->sport) 10103 pd->osport = pd->nsport = *pd->sport; 10104 if (pd->dport) 10105 pd->odport = pd->ndport = *pd->dport; 10106 10107 return (0); 10108 } 10109 10110 static void 10111 pf_counters_inc(int action, struct pf_pdesc *pd, 10112 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 10113 { 10114 struct pf_krule *tr; 10115 int dir = pd->dir; 10116 int dirndx; 10117 10118 pf_counter_u64_critical_enter(); 10119 pf_counter_u64_add_protected( 10120 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10121 pd->tot_len); 10122 pf_counter_u64_add_protected( 10123 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 10124 1); 10125 10126 if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) { 10127 dirndx = (dir == PF_OUT); 10128 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 10129 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 10130 pf_update_timestamp(r); 10131 10132 if (a != NULL) { 10133 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 10134 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 10135 } 10136 if (s != NULL) { 10137 struct pf_krule_item *ri; 10138 10139 if (s->nat_rule != NULL) { 10140 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 10141 1); 10142 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 10143 pd->tot_len); 10144 } 10145 /* 10146 * Source nodes are accessed unlocked here. 10147 * But since we are operating with stateful tracking 10148 * and the state is locked, those SNs could not have 10149 * been freed. 10150 */ 10151 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 10152 if (s->sns[sn_type] != NULL) { 10153 counter_u64_add( 10154 s->sns[sn_type]->packets[dirndx], 10155 1); 10156 counter_u64_add( 10157 s->sns[sn_type]->bytes[dirndx], 10158 pd->tot_len); 10159 } 10160 } 10161 dirndx = (dir == s->direction) ? 0 : 1; 10162 s->packets[dirndx]++; 10163 s->bytes[dirndx] += pd->tot_len; 10164 10165 SLIST_FOREACH(ri, &s->match_rules, entry) { 10166 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 10167 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 10168 10169 if (ri->r->src.addr.type == PF_ADDR_TABLE) 10170 pfr_update_stats(ri->r->src.addr.p.tbl, 10171 (s == NULL) ? pd->src : 10172 &s->key[(s->direction == PF_IN)]-> 10173 addr[(s->direction == PF_OUT)], 10174 pd->af, pd->tot_len, dir == PF_OUT, 10175 r->action == PF_PASS, ri->r->src.neg); 10176 if (ri->r->dst.addr.type == PF_ADDR_TABLE) 10177 pfr_update_stats(ri->r->dst.addr.p.tbl, 10178 (s == NULL) ? pd->dst : 10179 &s->key[(s->direction == PF_IN)]-> 10180 addr[(s->direction == PF_IN)], 10181 pd->af, pd->tot_len, dir == PF_OUT, 10182 r->action == PF_PASS, ri->r->dst.neg); 10183 } 10184 } 10185 10186 tr = r; 10187 if (s != NULL && s->nat_rule != NULL && 10188 r == &V_pf_default_rule) 10189 tr = s->nat_rule; 10190 10191 if (tr->src.addr.type == PF_ADDR_TABLE) 10192 pfr_update_stats(tr->src.addr.p.tbl, 10193 (s == NULL) ? pd->src : 10194 &s->key[(s->direction == PF_IN)]-> 10195 addr[(s->direction == PF_OUT)], 10196 pd->af, pd->tot_len, dir == PF_OUT, 10197 r->action == PF_PASS, tr->src.neg); 10198 if (tr->dst.addr.type == PF_ADDR_TABLE) 10199 pfr_update_stats(tr->dst.addr.p.tbl, 10200 (s == NULL) ? pd->dst : 10201 &s->key[(s->direction == PF_IN)]-> 10202 addr[(s->direction == PF_IN)], 10203 pd->af, pd->tot_len, dir == PF_OUT, 10204 r->action == PF_PASS, tr->dst.neg); 10205 } 10206 pf_counter_u64_critical_exit(); 10207 } 10208 static void 10209 pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm, 10210 struct pf_krule *am, struct pf_kruleset *ruleset, 10211 struct pf_krule_slist *matchrules) 10212 { 10213 struct pf_krule_item *ri; 10214 10215 /* if this is the log(matches) rule, packet has been logged already */ 10216 if (rm->log & PF_LOG_MATCHES) 10217 return; 10218 10219 SLIST_FOREACH(ri, matchrules, entry) 10220 if (ri->r->log & PF_LOG_MATCHES) 10221 PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am, 10222 ruleset, pd, 1, ri->r); 10223 } 10224 10225 #if defined(INET) || defined(INET6) 10226 int 10227 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10228 struct inpcb *inp, struct pf_rule_actions *default_actions) 10229 { 10230 struct pfi_kkif *kif; 10231 u_short action, reason = 0; 10232 struct m_tag *mtag; 10233 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 10234 struct pf_kstate *s = NULL; 10235 struct pf_kruleset *ruleset = NULL; 10236 struct pf_pdesc pd; 10237 int use_2nd_queue = 0; 10238 uint16_t tag; 10239 10240 PF_RULES_RLOCK_TRACKER; 10241 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 10242 M_ASSERTPKTHDR(*m0); 10243 10244 if (!V_pf_status.running) 10245 return (PF_PASS); 10246 10247 PF_RULES_RLOCK(); 10248 10249 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10250 10251 if (__predict_false(kif == NULL)) { 10252 DPFPRINTF(PF_DEBUG_URGENT, 10253 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 10254 PF_RULES_RUNLOCK(); 10255 return (PF_DROP); 10256 } 10257 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 10258 PF_RULES_RUNLOCK(); 10259 return (PF_PASS); 10260 } 10261 10262 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 10263 PF_RULES_RUNLOCK(); 10264 return (PF_PASS); 10265 } 10266 10267 if (__predict_false(! M_WRITABLE(*m0))) { 10268 *m0 = m_unshare(*m0, M_NOWAIT); 10269 if (*m0 == NULL) { 10270 PF_RULES_RUNLOCK(); 10271 return (PF_DROP); 10272 } 10273 } 10274 10275 pf_init_pdesc(&pd, *m0); 10276 10277 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 10278 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10279 10280 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 10281 pd.pf_mtag->if_idxgen); 10282 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 10283 PF_RULES_RUNLOCK(); 10284 m_freem(*m0); 10285 *m0 = NULL; 10286 return (PF_PASS); 10287 } 10288 PF_RULES_RUNLOCK(); 10289 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 10290 *m0 = NULL; 10291 return (PF_PASS); 10292 } 10293 10294 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 10295 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 10296 /* Dummynet re-injects packets after they've 10297 * completed their delay. We've already 10298 * processed them, so pass unconditionally. */ 10299 10300 /* But only once. We may see the packet multiple times (e.g. 10301 * PFIL_IN/PFIL_OUT). */ 10302 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 10303 PF_RULES_RUNLOCK(); 10304 10305 return (PF_PASS); 10306 } 10307 10308 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 10309 kif, default_actions) == -1) { 10310 if (action != PF_PASS) 10311 pd.act.log |= PF_LOG_FORCE; 10312 goto done; 10313 } 10314 10315 #ifdef INET 10316 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 10317 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 10318 PF_RULES_RUNLOCK(); 10319 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 10320 0, ifp->if_mtu); 10321 *m0 = NULL; 10322 return (PF_DROP); 10323 } 10324 #endif /* INET */ 10325 #ifdef INET6 10326 /* 10327 * If we end up changing IP addresses (e.g. binat) the stack may get 10328 * confused and fail to send the icmp6 packet too big error. Just send 10329 * it here, before we do any NAT. 10330 */ 10331 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 10332 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 10333 PF_RULES_RUNLOCK(); 10334 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 10335 *m0 = NULL; 10336 return (PF_DROP); 10337 } 10338 #endif /* INET6 */ 10339 10340 if (__predict_false(ip_divert_ptr != NULL) && 10341 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 10342 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 10343 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 10344 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 10345 if (pd.pf_mtag == NULL && 10346 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10347 action = PF_DROP; 10348 goto done; 10349 } 10350 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 10351 } 10352 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 10353 pd.m->m_flags |= M_FASTFWD_OURS; 10354 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10355 } 10356 m_tag_delete(pd.m, mtag); 10357 10358 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 10359 if (mtag != NULL) 10360 m_tag_delete(pd.m, mtag); 10361 } 10362 10363 switch (pd.virtual_proto) { 10364 case PF_VPROTO_FRAGMENT: 10365 /* 10366 * handle fragments that aren't reassembled by 10367 * normalization 10368 */ 10369 if (kif == NULL || r == NULL) /* pflog */ 10370 action = PF_DROP; 10371 else 10372 action = pf_test_rule(&r, &s, &pd, &a, 10373 &ruleset, &reason, inp); 10374 if (action != PF_PASS) 10375 REASON_SET(&reason, PFRES_FRAG); 10376 break; 10377 10378 case IPPROTO_TCP: { 10379 /* Respond to SYN with a syncookie. */ 10380 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 10381 pd.dir == PF_IN && pf_synflood_check(&pd)) { 10382 pf_syncookie_send(&pd); 10383 action = PF_DROP; 10384 break; 10385 } 10386 10387 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 10388 use_2nd_queue = 1; 10389 action = pf_normalize_tcp(&pd); 10390 if (action == PF_DROP) 10391 goto done; 10392 action = pf_test_state(&s, &pd, &reason); 10393 if (action == PF_PASS || action == PF_AFRT) { 10394 if (V_pfsync_update_state_ptr != NULL) 10395 V_pfsync_update_state_ptr(s); 10396 r = s->rule; 10397 a = s->anchor; 10398 } else if (s == NULL) { 10399 /* Validate remote SYN|ACK, re-create original SYN if 10400 * valid. */ 10401 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 10402 TH_ACK && pf_syncookie_validate(&pd) && 10403 pd.dir == PF_IN) { 10404 struct mbuf *msyn; 10405 10406 msyn = pf_syncookie_recreate_syn(&pd); 10407 if (msyn == NULL) { 10408 action = PF_DROP; 10409 break; 10410 } 10411 10412 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 10413 &pd.act); 10414 m_freem(msyn); 10415 if (action != PF_PASS) 10416 break; 10417 10418 action = pf_test_state(&s, &pd, &reason); 10419 if (action != PF_PASS || s == NULL) { 10420 action = PF_DROP; 10421 break; 10422 } 10423 10424 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 10425 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 10426 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 10427 action = pf_synproxy(&pd, s, &reason); 10428 break; 10429 } else { 10430 action = pf_test_rule(&r, &s, &pd, 10431 &a, &ruleset, &reason, inp); 10432 } 10433 } 10434 break; 10435 } 10436 10437 case IPPROTO_SCTP: 10438 action = pf_normalize_sctp(&pd); 10439 if (action == PF_DROP) 10440 goto done; 10441 /* fallthrough */ 10442 case IPPROTO_UDP: 10443 default: 10444 action = pf_test_state(&s, &pd, &reason); 10445 if (action == PF_PASS || action == PF_AFRT) { 10446 if (V_pfsync_update_state_ptr != NULL) 10447 V_pfsync_update_state_ptr(s); 10448 r = s->rule; 10449 a = s->anchor; 10450 } else if (s == NULL) { 10451 action = pf_test_rule(&r, &s, 10452 &pd, &a, &ruleset, &reason, inp); 10453 } 10454 break; 10455 10456 case IPPROTO_ICMP: 10457 case IPPROTO_ICMPV6: { 10458 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 10459 action = PF_DROP; 10460 REASON_SET(&reason, PFRES_NORM); 10461 DPFPRINTF(PF_DEBUG_MISC, 10462 ("dropping IPv6 packet with ICMPv4 payload")); 10463 goto done; 10464 } 10465 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 10466 action = PF_DROP; 10467 REASON_SET(&reason, PFRES_NORM); 10468 DPFPRINTF(PF_DEBUG_MISC, 10469 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 10470 goto done; 10471 } 10472 action = pf_test_state_icmp(&s, &pd, &reason); 10473 if (action == PF_PASS || action == PF_AFRT) { 10474 if (V_pfsync_update_state_ptr != NULL) 10475 V_pfsync_update_state_ptr(s); 10476 r = s->rule; 10477 a = s->anchor; 10478 } else if (s == NULL) 10479 action = pf_test_rule(&r, &s, &pd, 10480 &a, &ruleset, &reason, inp); 10481 break; 10482 } 10483 10484 } 10485 10486 done: 10487 PF_RULES_RUNLOCK(); 10488 10489 if (pd.m == NULL) 10490 goto eat_pkt; 10491 10492 if (action == PF_PASS && pd.badopts && 10493 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || pd.act.allow_opts)) { 10494 action = PF_DROP; 10495 REASON_SET(&reason, PFRES_IPOPTIONS); 10496 pd.act.log = PF_LOG_FORCE; 10497 DPFPRINTF(PF_DEBUG_MISC, 10498 ("pf: dropping packet with dangerous headers\n")); 10499 } 10500 10501 if (s) { 10502 uint8_t log = pd.act.log; 10503 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 10504 pd.act.log |= log; 10505 tag = s->tag; 10506 } else { 10507 tag = r->tag; 10508 } 10509 10510 if (tag > 0 && pf_tag_packet(&pd, tag)) { 10511 action = PF_DROP; 10512 REASON_SET(&reason, PFRES_MEMORY); 10513 } 10514 10515 pf_scrub(&pd); 10516 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 10517 pf_normalize_mss(&pd); 10518 10519 if (pd.act.rtableid >= 0) 10520 M_SETFIB(pd.m, pd.act.rtableid); 10521 10522 if (pd.act.flags & PFSTATE_SETPRIO) { 10523 if (pd.tos & IPTOS_LOWDELAY) 10524 use_2nd_queue = 1; 10525 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 10526 action = PF_DROP; 10527 REASON_SET(&reason, PFRES_MEMORY); 10528 pd.act.log = PF_LOG_FORCE; 10529 DPFPRINTF(PF_DEBUG_MISC, 10530 ("pf: failed to allocate 802.1q mtag\n")); 10531 } 10532 } 10533 10534 #ifdef ALTQ 10535 if (action == PF_PASS && pd.act.qid) { 10536 if (pd.pf_mtag == NULL && 10537 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10538 action = PF_DROP; 10539 REASON_SET(&reason, PFRES_MEMORY); 10540 } else { 10541 if (s != NULL) 10542 pd.pf_mtag->qid_hash = pf_state_hash(s); 10543 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 10544 pd.pf_mtag->qid = pd.act.pqid; 10545 else 10546 pd.pf_mtag->qid = pd.act.qid; 10547 /* Add hints for ecn. */ 10548 pd.pf_mtag->hdr = mtod(pd.m, void *); 10549 } 10550 } 10551 #endif /* ALTQ */ 10552 10553 /* 10554 * connections redirected to loopback should not match sockets 10555 * bound specifically to loopback due to security implications, 10556 * see tcp_input() and in_pcblookup_listen(). 10557 */ 10558 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 10559 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 10560 (s->nat_rule->action == PF_RDR || 10561 s->nat_rule->action == PF_BINAT) && 10562 pf_is_loopback(af, pd.dst)) 10563 pd.m->m_flags |= M_SKIP_FIREWALL; 10564 10565 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 10566 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 10567 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 10568 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 10569 if (mtag != NULL) { 10570 ((struct pf_divert_mtag *)(mtag+1))->port = 10571 ntohs(r->divert.port); 10572 ((struct pf_divert_mtag *)(mtag+1))->idir = 10573 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 10574 PF_DIVERT_MTAG_DIR_OUT; 10575 10576 if (s) 10577 PF_STATE_UNLOCK(s); 10578 10579 m_tag_prepend(pd.m, mtag); 10580 if (pd.m->m_flags & M_FASTFWD_OURS) { 10581 if (pd.pf_mtag == NULL && 10582 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 10583 action = PF_DROP; 10584 REASON_SET(&reason, PFRES_MEMORY); 10585 pd.act.log = PF_LOG_FORCE; 10586 DPFPRINTF(PF_DEBUG_MISC, 10587 ("pf: failed to allocate tag\n")); 10588 } else { 10589 pd.pf_mtag->flags |= 10590 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 10591 pd.m->m_flags &= ~M_FASTFWD_OURS; 10592 } 10593 } 10594 ip_divert_ptr(*m0, dir == PF_IN); 10595 *m0 = NULL; 10596 10597 return (action); 10598 } else { 10599 /* XXX: ipfw has the same behaviour! */ 10600 action = PF_DROP; 10601 REASON_SET(&reason, PFRES_MEMORY); 10602 pd.act.log = PF_LOG_FORCE; 10603 DPFPRINTF(PF_DEBUG_MISC, 10604 ("pf: failed to allocate divert tag\n")); 10605 } 10606 } 10607 /* XXX: Anybody working on it?! */ 10608 if (af == AF_INET6 && r->divert.port) 10609 printf("pf: divert(9) is not supported for IPv6\n"); 10610 10611 /* this flag will need revising if the pkt is forwarded */ 10612 if (pd.pf_mtag) 10613 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 10614 10615 if (pd.act.log) { 10616 struct pf_krule *lr; 10617 struct pf_krule_item *ri; 10618 10619 if (s != NULL && s->nat_rule != NULL && 10620 s->nat_rule->log & PF_LOG_ALL) 10621 lr = s->nat_rule; 10622 else 10623 lr = r; 10624 10625 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 10626 PFLOG_PACKET(action, reason, lr, a, 10627 ruleset, &pd, (s == NULL), NULL); 10628 if (s) { 10629 SLIST_FOREACH(ri, &s->match_rules, entry) 10630 if (ri->r->log & PF_LOG_ALL) 10631 PFLOG_PACKET(action, 10632 reason, ri->r, a, ruleset, &pd, 0, NULL); 10633 } 10634 } 10635 10636 pf_counters_inc(action, &pd, s, r, a); 10637 10638 switch (action) { 10639 case PF_SYNPROXY_DROP: 10640 m_freem(*m0); 10641 case PF_DEFER: 10642 *m0 = NULL; 10643 action = PF_PASS; 10644 break; 10645 case PF_DROP: 10646 m_freem(*m0); 10647 *m0 = NULL; 10648 break; 10649 case PF_AFRT: 10650 if (pf_translate_af(&pd)) { 10651 if (!pd.m) 10652 *m0 = NULL; 10653 action = PF_DROP; 10654 break; 10655 } 10656 *m0 = pd.m; /* pf_translate_af may change pd.m */ 10657 #ifdef INET 10658 if (pd.naf == AF_INET) 10659 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10660 #endif /* INET */ 10661 #ifdef INET6 10662 if (pd.naf == AF_INET6) 10663 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10664 #endif /* INET6 */ 10665 *m0 = NULL; 10666 action = PF_PASS; 10667 goto out; 10668 break; 10669 default: 10670 if (pd.act.rt) { 10671 switch (af) { 10672 #ifdef INET 10673 case AF_INET: 10674 /* pf_route() returns unlocked. */ 10675 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 10676 break; 10677 #endif /* INET */ 10678 #ifdef INET6 10679 case AF_INET6: 10680 /* pf_route6() returns unlocked. */ 10681 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 10682 break; 10683 #endif /* INET6 */ 10684 } 10685 goto out; 10686 } 10687 if (pf_dummynet(&pd, s, r, m0) != 0) { 10688 action = PF_DROP; 10689 REASON_SET(&reason, PFRES_MEMORY); 10690 } 10691 break; 10692 } 10693 10694 eat_pkt: 10695 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 10696 10697 if (s && action != PF_DROP) { 10698 if (!s->if_index_in && dir == PF_IN) 10699 s->if_index_in = ifp->if_index; 10700 else if (!s->if_index_out && dir == PF_OUT) 10701 s->if_index_out = ifp->if_index; 10702 } 10703 10704 if (s) 10705 PF_STATE_UNLOCK(s); 10706 10707 out: 10708 #ifdef INET6 10709 /* If reassembled packet passed, create new fragments. */ 10710 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 10711 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 10712 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 10713 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 10714 #endif /* INET6 */ 10715 10716 pf_sctp_multihome_delayed(&pd, kif, s, action); 10717 10718 return (action); 10719 } 10720 #endif /* INET || INET6 */ 10721