1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/random.h> 58 #include <sys/refcount.h> 59 #include <sys/sdt.h> 60 #include <sys/socket.h> 61 #include <sys/sysctl.h> 62 #include <sys/taskqueue.h> 63 #include <sys/ucred.h> 64 65 #include <crypto/sha2/sha512.h> 66 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/if_private.h> 70 #include <net/if_types.h> 71 #include <net/if_vlan_var.h> 72 #include <net/route.h> 73 #include <net/route/nhop.h> 74 #include <net/vnet.h> 75 76 #include <net/pfil.h> 77 #include <net/pfvar.h> 78 #include <net/if_pflog.h> 79 #include <net/if_pfsync.h> 80 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_var.h> 83 #include <netinet/in_fib.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_fw.h> 86 #include <netinet/ip_icmp.h> 87 #include <netinet/icmp_var.h> 88 #include <netinet/ip_var.h> 89 #include <netinet/tcp.h> 90 #include <netinet/tcp_fsm.h> 91 #include <netinet/tcp_seq.h> 92 #include <netinet/tcp_timer.h> 93 #include <netinet/tcp_var.h> 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 97 /* dummynet */ 98 #include <netinet/ip_dummynet.h> 99 #include <netinet/ip_fw.h> 100 #include <netpfil/ipfw/dn_heap.h> 101 #include <netpfil/ipfw/ip_fw_private.h> 102 #include <netpfil/ipfw/ip_dn_private.h> 103 104 #ifdef INET6 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet6/in6_fib.h> 111 #include <netinet6/scope6_var.h> 112 #endif /* INET6 */ 113 114 #include <netinet/sctp_header.h> 115 #include <netinet/sctp_crc32.h> 116 117 #include <netipsec/ah.h> 118 119 #include <machine/in_cksum.h> 120 #include <security/mac/mac_framework.h> 121 122 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 SDT_PROBE_DEFINE2(pf, , log, log, "int", "const char *"); 163 164 /* 165 * Global variables 166 */ 167 168 /* state tables */ 169 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 170 VNET_DEFINE(struct pf_kpalist, pf_pabuf[3]); 171 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 172 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 173 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 174 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 175 VNET_DEFINE(struct pf_kstatus, pf_status); 176 177 VNET_DEFINE(u_int32_t, ticket_altqs_active); 178 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 179 VNET_DEFINE(int, altqs_inactive_open); 180 VNET_DEFINE(u_int32_t, ticket_pabuf); 181 182 static const int PF_HDR_LIMIT = 20; /* arbitrary limit */ 183 184 VNET_DEFINE(SHA512_CTX, pf_tcp_secret_ctx); 185 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 186 VNET_DEFINE(u_char, pf_tcp_secret[16]); 187 #define V_pf_tcp_secret VNET(pf_tcp_secret) 188 VNET_DEFINE(int, pf_tcp_secret_init); 189 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 190 VNET_DEFINE(int, pf_tcp_iss_off); 191 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 192 VNET_DECLARE(int, pf_vnet_active); 193 #define V_pf_vnet_active VNET(pf_vnet_active) 194 195 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 196 #define V_pf_purge_idx VNET(pf_purge_idx) 197 198 #ifdef PF_WANT_32_TO_64_COUNTER 199 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 200 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 201 202 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 203 VNET_DEFINE(size_t, pf_allrulecount); 204 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 205 #endif 206 207 #define PF_SCTP_MAX_ENDPOINTS 8 208 209 struct pf_sctp_endpoint; 210 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 211 struct pf_sctp_source { 212 sa_family_t af; 213 struct pf_addr addr; 214 TAILQ_ENTRY(pf_sctp_source) entry; 215 }; 216 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 217 struct pf_sctp_endpoint 218 { 219 uint32_t v_tag; 220 struct pf_sctp_sources sources; 221 RB_ENTRY(pf_sctp_endpoint) entry; 222 }; 223 static int 224 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 225 { 226 return (a->v_tag - b->v_tag); 227 } 228 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 229 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 230 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 231 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 232 static struct mtx_padalign pf_sctp_endpoints_mtx; 233 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 234 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 235 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 236 237 /* 238 * Queue for pf_intr() sends. 239 */ 240 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 241 struct pf_send_entry { 242 STAILQ_ENTRY(pf_send_entry) pfse_next; 243 struct mbuf *pfse_m; 244 enum { 245 PFSE_IP, 246 PFSE_IP6, 247 PFSE_ICMP, 248 PFSE_ICMP6, 249 } pfse_type; 250 struct { 251 int type; 252 int code; 253 int mtu; 254 } icmpopts; 255 }; 256 257 STAILQ_HEAD(pf_send_head, pf_send_entry); 258 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 259 #define V_pf_sendqueue VNET(pf_sendqueue) 260 261 static struct mtx_padalign pf_sendqueue_mtx; 262 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 263 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 264 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 265 266 /* 267 * Queue for pf_overload_task() tasks. 268 */ 269 struct pf_overload_entry { 270 SLIST_ENTRY(pf_overload_entry) next; 271 struct pf_addr addr; 272 sa_family_t af; 273 uint8_t dir; 274 struct pf_krule *rule; 275 }; 276 277 SLIST_HEAD(pf_overload_head, pf_overload_entry); 278 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 279 #define V_pf_overloadqueue VNET(pf_overloadqueue) 280 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 281 #define V_pf_overloadtask VNET(pf_overloadtask) 282 283 static struct mtx_padalign pf_overloadqueue_mtx; 284 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 285 "pf overload/flush queue", MTX_DEF); 286 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 287 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 288 289 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 290 struct mtx_padalign pf_unlnkdrules_mtx; 291 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 292 MTX_DEF); 293 294 struct sx pf_config_lock; 295 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 296 297 struct mtx_padalign pf_table_stats_lock; 298 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 299 MTX_DEF); 300 301 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 302 #define V_pf_sources_z VNET(pf_sources_z) 303 uma_zone_t pf_mtag_z; 304 VNET_DEFINE(uma_zone_t, pf_state_z); 305 VNET_DEFINE(uma_zone_t, pf_state_key_z); 306 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 307 308 VNET_DEFINE(struct unrhdr64, pf_stateid); 309 310 static void pf_src_tree_remove_state(struct pf_kstate *); 311 static int pf_check_threshold(struct pf_kthreshold *); 312 313 static void pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *, 314 struct pf_addr *, u_int16_t); 315 static int pf_modulate_sack(struct pf_pdesc *, 316 struct tcphdr *, struct pf_state_peer *); 317 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 318 u_int16_t *, u_int16_t *); 319 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 320 struct pf_addr *, struct pf_addr *, u_int16_t, 321 u_int16_t *, u_int16_t *, u_int16_t *, 322 u_int16_t *, u_int8_t, sa_family_t); 323 int pf_change_icmp_af(struct mbuf *, int, 324 struct pf_pdesc *, struct pf_pdesc *, 325 struct pf_addr *, struct pf_addr *, sa_family_t, 326 sa_family_t); 327 int pf_translate_icmp_af(int, void *); 328 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 329 int, sa_family_t, struct pf_krule *, int); 330 static void pf_detach_state(struct pf_kstate *); 331 static int pf_state_key_attach(struct pf_state_key *, 332 struct pf_state_key *, struct pf_kstate *); 333 static void pf_state_key_detach(struct pf_kstate *, int); 334 static int pf_state_key_ctor(void *, int, void *, int); 335 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 336 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 337 struct pf_mtag *pf_mtag); 338 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 339 struct pf_krule *, struct mbuf **); 340 static int pf_dummynet_route(struct pf_pdesc *, 341 struct pf_kstate *, struct pf_krule *, 342 struct ifnet *, const struct sockaddr *, struct mbuf **); 343 static int pf_test_eth_rule(int, struct pfi_kkif *, 344 struct mbuf **); 345 static enum pf_test_status pf_match_rule(struct pf_test_ctx *, struct pf_kruleset *); 346 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 347 struct pf_pdesc *, struct pf_krule **, 348 struct pf_kruleset **, u_short *, struct inpcb *, 349 struct pf_krule_slist *); 350 static int pf_create_state(struct pf_krule *, 351 struct pf_test_ctx *, 352 struct pf_kstate **, u_int16_t, u_int16_t); 353 static int pf_state_key_addr_setup(struct pf_pdesc *, 354 struct pf_state_key_cmp *, int); 355 static int pf_tcp_track_full(struct pf_kstate *, 356 struct pf_pdesc *, u_short *, int *, 357 struct pf_state_peer *, struct pf_state_peer *, 358 u_int8_t, u_int8_t); 359 static int pf_tcp_track_sloppy(struct pf_kstate *, 360 struct pf_pdesc *, u_short *, 361 struct pf_state_peer *, struct pf_state_peer *, 362 u_int8_t, u_int8_t); 363 static __inline int pf_synproxy_ack(struct pf_krule *, struct pf_pdesc *, 364 struct pf_kstate **, struct pf_rule_actions *); 365 static int pf_test_state(struct pf_kstate **, struct pf_pdesc *, 366 u_short *); 367 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 368 struct pf_pdesc *, struct pf_kstate **, 369 u_int16_t, u_int16_t, int, int *, int, int); 370 static int pf_test_state_icmp(struct pf_kstate **, 371 struct pf_pdesc *, u_short *); 372 static int pf_sctp_track(struct pf_kstate *, struct pf_pdesc *, 373 u_short *); 374 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 375 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 376 struct pfi_kkif *, struct pf_kstate *, int); 377 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 378 int, u_int16_t); 379 static int pf_check_proto_cksum(struct mbuf *, int, int, 380 u_int8_t, sa_family_t); 381 static int pf_walk_option(struct pf_pdesc *, struct ip *, 382 int, int, u_short *); 383 static int pf_walk_header(struct pf_pdesc *, struct ip *, u_short *); 384 #ifdef INET6 385 static int pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *, 386 int, int, u_short *); 387 static int pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *, 388 u_short *); 389 #endif 390 static void pf_print_state_parts(struct pf_kstate *, 391 struct pf_state_key *, struct pf_state_key *); 392 static int pf_patch_8(struct pf_pdesc *, u_int8_t *, u_int8_t, 393 bool); 394 static int pf_find_state(struct pf_pdesc *, 395 const struct pf_state_key_cmp *, struct pf_kstate **); 396 static bool pf_src_connlimit(struct pf_kstate *); 397 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 398 static void pf_counters_inc(int, struct pf_pdesc *, 399 struct pf_kstate *, struct pf_krule *, 400 struct pf_krule *, struct pf_krule_slist *); 401 static void pf_log_matches(struct pf_pdesc *, struct pf_krule *, 402 struct pf_krule *, struct pf_kruleset *, 403 struct pf_krule_slist *); 404 static void pf_overload_task(void *v, int pending); 405 static u_short pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX], 406 struct pf_srchash *[PF_SN_MAX], struct pf_krule *, 407 struct pf_addr *, sa_family_t, struct pf_addr *, 408 struct pfi_kkif *, sa_family_t, pf_sn_types_t); 409 static u_int pf_purge_expired_states(u_int, int); 410 static void pf_purge_unlinked_rules(void); 411 static int pf_mtag_uminit(void *, int, int); 412 static void pf_mtag_free(struct m_tag *); 413 static void pf_packet_rework_nat(struct pf_pdesc *, int, 414 struct pf_state_key *); 415 #ifdef INET 416 static int pf_route(struct pf_krule *, 417 struct ifnet *, struct pf_kstate *, 418 struct pf_pdesc *, struct inpcb *); 419 #endif /* INET */ 420 #ifdef INET6 421 static void pf_change_a6(struct pf_addr *, u_int16_t *, 422 struct pf_addr *, u_int8_t); 423 static int pf_route6(struct pf_krule *, 424 struct ifnet *, struct pf_kstate *, 425 struct pf_pdesc *, struct inpcb *); 426 #endif /* INET6 */ 427 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 428 429 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 430 431 static inline int 432 pf_statelim_id_cmp(const struct pf_statelim *a, const struct pf_statelim *b) 433 { 434 if (a->pfstlim_id > b->pfstlim_id) 435 return (1); 436 if (a->pfstlim_id < b->pfstlim_id) 437 return (-1); 438 439 return (0); 440 } 441 442 RB_GENERATE(pf_statelim_id_tree, pf_statelim, pfstlim_id_tree, 443 pf_statelim_id_cmp); 444 445 static inline int 446 pf_statelim_nm_cmp(const struct pf_statelim *a, const struct pf_statelim *b) 447 { 448 return (strncmp(a->pfstlim_nm, b->pfstlim_nm, sizeof(a->pfstlim_nm))); 449 } 450 451 RB_GENERATE(pf_statelim_nm_tree, pf_statelim, pfstlim_nm_tree, 452 pf_statelim_nm_cmp); 453 454 VNET_DEFINE(struct pf_statelim_id_tree, pf_statelim_id_tree_active); 455 VNET_DEFINE(struct pf_statelim_list, pf_statelim_list_active); 456 VNET_DEFINE(struct pf_statelim_id_tree, pf_statelim_id_tree_inactive); 457 VNET_DEFINE(struct pf_statelim_nm_tree, pf_statelim_nm_tree_inactive); 458 VNET_DEFINE(struct pf_statelim_list, pf_statelim_list_inactive); 459 460 static inline int 461 pf_sourcelim_id_cmp(const struct pf_sourcelim *a, const struct pf_sourcelim *b) 462 { 463 if (a->pfsrlim_id > b->pfsrlim_id) 464 return (1); 465 if (a->pfsrlim_id < b->pfsrlim_id) 466 return (-1); 467 468 return (0); 469 } 470 471 RB_GENERATE(pf_sourcelim_id_tree, pf_sourcelim, pfsrlim_id_tree, 472 pf_sourcelim_id_cmp); 473 474 static inline int 475 pf_sourcelim_nm_cmp(const struct pf_sourcelim *a, const struct pf_sourcelim *b) 476 { 477 return (strncmp(a->pfsrlim_nm, b->pfsrlim_nm, sizeof(a->pfsrlim_nm))); 478 } 479 480 RB_GENERATE(pf_sourcelim_nm_tree, pf_sourcelim, pfsrlim_nm_tree, 481 pf_sourcelim_nm_cmp); 482 483 static inline int 484 pf_source_cmp(const struct pf_source *a, const struct pf_source *b) 485 { 486 if (a->pfsr_af > b->pfsr_af) 487 return (1); 488 if (a->pfsr_af < b->pfsr_af) 489 return (-1); 490 if (a->pfsr_rdomain > b->pfsr_rdomain) 491 return (1); 492 if (a->pfsr_rdomain < b->pfsr_rdomain) 493 return (-1); 494 495 return (pf_addr_cmp(&a->pfsr_addr, &b->pfsr_addr, a->pfsr_af)); 496 } 497 498 RB_GENERATE(pf_source_tree, pf_source, pfsr_tree, pf_source_cmp); 499 500 static inline int 501 pf_source_ioc_cmp(const struct pf_source *a, const struct pf_source *b) 502 { 503 size_t i; 504 505 if (a->pfsr_af > b->pfsr_af) 506 return (1); 507 if (a->pfsr_af < b->pfsr_af) 508 return (-1); 509 if (a->pfsr_rdomain > b->pfsr_rdomain) 510 return (1); 511 if (a->pfsr_rdomain < b->pfsr_rdomain) 512 return (-1); 513 514 for (i = 0; i < nitems(a->pfsr_addr.addr32); i++) { 515 uint32_t wa = ntohl(a->pfsr_addr.addr32[i]); 516 uint32_t wb = ntohl(b->pfsr_addr.addr32[i]); 517 518 if (wa > wb) 519 return (1); 520 if (wa < wb) 521 return (-1); 522 } 523 524 return (0); 525 } 526 527 RB_GENERATE(pf_source_ioc_tree, pf_source, pfsr_ioc_tree, pf_source_ioc_cmp); 528 529 VNET_DEFINE(struct pf_sourcelim_id_tree, pf_sourcelim_id_tree_active); 530 VNET_DEFINE(struct pf_sourcelim_list, pf_sourcelim_list_active); 531 532 VNET_DEFINE(struct pf_sourcelim_id_tree, pf_sourcelim_id_tree_inactive); 533 VNET_DEFINE(struct pf_sourcelim_nm_tree, pf_sourcelim_nm_tree_inactive); 534 VNET_DEFINE(struct pf_sourcelim_list, pf_sourcelim_list_inactive); 535 536 static inline struct pf_statelim * 537 pf_statelim_find(uint32_t id) 538 { 539 struct pf_statelim key; 540 541 /* only the id is used in cmp, so don't have to zero all the things */ 542 key.pfstlim_id = id; 543 544 return (RB_FIND(pf_statelim_id_tree, 545 &V_pf_statelim_id_tree_active, &key)); 546 } 547 548 static inline struct pf_sourcelim * 549 pf_sourcelim_find(uint32_t id) 550 { 551 struct pf_sourcelim key; 552 553 /* only the id is used in cmp, so don't have to zero all the things */ 554 key.pfsrlim_id = id; 555 556 return (RB_FIND(pf_sourcelim_id_tree, 557 &V_pf_sourcelim_id_tree_active, &key)); 558 } 559 560 struct pf_source_list pf_source_gc = TAILQ_HEAD_INITIALIZER(pf_source_gc); 561 562 static void 563 pf_source_purge(void) 564 { 565 struct pf_source *sr, *nsr; 566 567 TAILQ_FOREACH_SAFE(sr, &pf_source_gc, pfsr_empty_gc, nsr) { 568 struct pf_sourcelim *srlim = sr->pfsr_parent; 569 570 if (time_uptime <= sr->pfsr_empty_ts + 571 srlim->pfsrlim_rate.seconds + 1) 572 continue; 573 574 TAILQ_REMOVE(&pf_source_gc, sr, pfsr_empty_gc); 575 576 RB_REMOVE(pf_source_tree, &srlim->pfsrlim_sources, sr); 577 RB_REMOVE(pf_source_ioc_tree, &srlim->pfsrlim_ioc_sources, sr); 578 srlim->pfsrlim_nsources--; 579 580 free(sr, M_PF_SOURCE_LIM); 581 } 582 } 583 584 static void 585 pf_source_pfr_addr(struct pfr_addr *p, const struct pf_source *sr) 586 { 587 struct pf_sourcelim *srlim = sr->pfsr_parent; 588 589 memset(p, 0, sizeof(*p)); 590 591 p->pfra_af = sr->pfsr_af; 592 switch (sr->pfsr_af) { 593 case AF_INET: 594 p->pfra_net = srlim->pfsrlim_ipv4_prefix; 595 p->pfra_ip4addr = sr->pfsr_addr.v4; 596 break; 597 #ifdef INET6 598 case AF_INET6: 599 p->pfra_net = srlim->pfsrlim_ipv6_prefix; 600 p->pfra_ip6addr = sr->pfsr_addr.v6; 601 break; 602 #endif /* INET6 */ 603 } 604 } 605 606 static void 607 pf_source_used(struct pf_source *sr) 608 { 609 struct pf_sourcelim *srlim = sr->pfsr_parent; 610 struct pfr_ktable *t; 611 unsigned int used; 612 613 used = sr->pfsr_inuse++; 614 sr->pfsr_rate_ts += srlim->pfsrlim_rate_token; 615 616 if (used == 0) 617 TAILQ_REMOVE(&pf_source_gc, sr, pfsr_empty_gc); 618 else if ((t = srlim->pfsrlim_overload.table) != NULL && 619 used >= srlim->pfsrlim_overload.hwm && !sr->pfsr_intable) { 620 struct pfr_addr p; 621 622 pf_source_pfr_addr(&p, sr); 623 624 pfr_insert_kentry(t, &p, time_second); 625 sr->pfsr_intable = 1; 626 } 627 } 628 629 static void 630 pf_source_rele(struct pf_source *sr) 631 { 632 struct pf_sourcelim *srlim = sr->pfsr_parent; 633 struct pfr_ktable *t; 634 unsigned int used; 635 636 used = --sr->pfsr_inuse; 637 638 t = srlim->pfsrlim_overload.table; 639 if (t != NULL && sr->pfsr_intable && 640 used < srlim->pfsrlim_overload.lwm) { 641 struct pfr_addr p; 642 643 pf_source_pfr_addr(&p, sr); 644 645 pfr_remove_kentry(t, &p); 646 sr->pfsr_intable = 0; 647 } 648 649 if (used == 0) { 650 TAILQ_INSERT_TAIL(&pf_source_gc, sr, pfsr_empty_gc); 651 sr->pfsr_empty_ts = time_uptime + srlim->pfsrlim_rate.seconds; 652 } 653 } 654 655 static inline void 656 pf_source_key(struct pf_sourcelim *srlim, struct pf_source *key, 657 sa_family_t af, const struct pf_addr *addr) 658 { 659 size_t i; 660 661 /* only af+addr is used for lookup. */ 662 key->pfsr_af = af; 663 key->pfsr_rdomain = 0; 664 switch (af) { 665 case AF_INET: 666 key->pfsr_addr.addr32[0] = 667 srlim->pfsrlim_ipv4_mask.v4.s_addr & 668 addr->v4.s_addr; 669 670 for (i = 1; i < nitems(key->pfsr_addr.addr32); i++) 671 key->pfsr_addr.addr32[i] = htonl(0); 672 break; 673 #ifdef INET6 674 case AF_INET6: 675 for (i = 0; i < nitems(key->pfsr_addr.addr32); i++) { 676 key->pfsr_addr.addr32[i] = 677 srlim->pfsrlim_ipv6_mask.addr32[i] & 678 addr->addr32[i]; 679 } 680 break; 681 #endif 682 default: 683 unhandled_af(af); 684 /* NOTREACHED */ 685 } 686 } 687 688 static inline struct pf_source * 689 pf_source_find(struct pf_sourcelim *srlim, struct pf_source *key) 690 { 691 return (RB_FIND(pf_source_tree, &srlim->pfsrlim_sources, key)); 692 } 693 694 extern int pf_end_threads; 695 extern struct proc *pf_purge_proc; 696 697 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 698 699 #define PACKET_UNDO_NAT(_pd, _off, _s) \ 700 do { \ 701 struct pf_state_key *nk; \ 702 if ((pd->dir) == PF_OUT) \ 703 nk = (_s)->key[PF_SK_STACK]; \ 704 else \ 705 nk = (_s)->key[PF_SK_WIRE]; \ 706 pf_packet_rework_nat(_pd, _off, nk); \ 707 } while (0) 708 709 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 710 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 711 712 static struct pfi_kkif * 713 BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd) 714 { 715 struct pfi_kkif *k = pd->kif; 716 717 SDT_PROBE2(pf, ip, , bound_iface, st, k); 718 719 /* Floating unless otherwise specified. */ 720 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 721 return (V_pfi_all); 722 723 /* 724 * Initially set to all, because we don't know what interface we'll be 725 * sending this out when we create the state. 726 */ 727 if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN)) 728 return (V_pfi_all); 729 730 /* 731 * If this state is created based on another state (e.g. SCTP 732 * multihome) always set it floating initially. We can't know for sure 733 * what interface the actual traffic for this state will come in on. 734 */ 735 if (pd->related_rule) 736 return (V_pfi_all); 737 738 /* Don't overrule the interface for states created on incoming packets. */ 739 if (st->direction == PF_IN) 740 return (k); 741 742 /* No route-to, so don't overrule. */ 743 if (st->act.rt != PF_ROUTETO) 744 return (k); 745 746 /* Bind to the route-to interface. */ 747 return (st->act.rt_kif); 748 } 749 750 #define STATE_INC_COUNTERS(s) \ 751 do { \ 752 struct pf_krule_item *mrm; \ 753 counter_u64_add(s->rule->states_cur, 1); \ 754 counter_u64_add(s->rule->states_tot, 1); \ 755 if (s->anchor != NULL) { \ 756 counter_u64_add(s->anchor->states_cur, 1); \ 757 counter_u64_add(s->anchor->states_tot, 1); \ 758 } \ 759 if (s->nat_rule != NULL && s->nat_rule != s->rule) { \ 760 counter_u64_add(s->nat_rule->states_cur, 1); \ 761 counter_u64_add(s->nat_rule->states_tot, 1); \ 762 } \ 763 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 764 if (s->nat_rule != mrm->r) { \ 765 counter_u64_add(mrm->r->states_cur, 1); \ 766 counter_u64_add(mrm->r->states_tot, 1); \ 767 } \ 768 } \ 769 } while (0) 770 771 #define STATE_DEC_COUNTERS(s) \ 772 do { \ 773 struct pf_krule_item *mrm; \ 774 counter_u64_add(s->rule->states_cur, -1); \ 775 if (s->anchor != NULL) \ 776 counter_u64_add(s->anchor->states_cur, -1); \ 777 if (s->nat_rule != NULL && s->nat_rule != s->rule) \ 778 counter_u64_add(s->nat_rule->states_cur, -1); \ 779 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 780 if (s->nat_rule != mrm->r) { \ 781 counter_u64_add(mrm->r->states_cur, -1);\ 782 } \ 783 } while (0) 784 785 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 786 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 787 MALLOC_DEFINE(M_PF_STATE_LINK, "pf_state_link", "pf(4) state links"); 788 MALLOC_DEFINE(M_PF_SOURCE_LIM, "pf_source_lim", "pf(4) source limiter"); 789 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 790 VNET_DEFINE(struct pf_idhash *, pf_idhash); 791 VNET_DEFINE(struct pf_srchash *, pf_srchash); 792 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 793 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 794 795 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 796 "pf(4)"); 797 798 VNET_DEFINE(u_long, pf_hashmask); 799 VNET_DEFINE(u_long, pf_srchashmask); 800 VNET_DEFINE(u_long, pf_udpendpointhashmask); 801 VNET_DEFINE_STATIC(u_long, pf_hashsize) = PF_HASHSIZ; 802 #define V_pf_hashsize VNET(pf_hashsize) 803 VNET_DEFINE_STATIC(u_long, pf_srchashsize) = PF_SRCHASHSIZ; 804 #define V_pf_srchashsize VNET(pf_srchashsize) 805 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize) = PF_UDPENDHASHSIZ; 806 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 807 u_long pf_ioctl_maxcount = 65535; 808 809 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 810 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 811 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 812 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 813 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 814 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 815 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 816 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 817 818 VNET_DEFINE(void *, pf_swi_cookie); 819 VNET_DEFINE(struct intr_event *, pf_swi_ie); 820 821 VNET_DEFINE(uint32_t, pf_hashseed); 822 #define V_pf_hashseed VNET(pf_hashseed) 823 824 static void 825 pf_sctp_checksum(struct mbuf *m, int off) 826 { 827 uint32_t sum = 0; 828 829 /* Zero out the checksum, to enable recalculation. */ 830 m_copyback(m, off + offsetof(struct sctphdr, checksum), 831 sizeof(sum), (caddr_t)&sum); 832 833 sum = sctp_calculate_cksum(m, off); 834 835 m_copyback(m, off + offsetof(struct sctphdr, checksum), 836 sizeof(sum), (caddr_t)&sum); 837 } 838 839 int 840 pf_addr_cmp(const struct pf_addr *a, const struct pf_addr *b, sa_family_t af) 841 { 842 843 switch (af) { 844 #ifdef INET 845 case AF_INET: 846 if (a->addr32[0] > b->addr32[0]) 847 return (1); 848 if (a->addr32[0] < b->addr32[0]) 849 return (-1); 850 break; 851 #endif /* INET */ 852 #ifdef INET6 853 case AF_INET6: 854 if (a->addr32[3] > b->addr32[3]) 855 return (1); 856 if (a->addr32[3] < b->addr32[3]) 857 return (-1); 858 if (a->addr32[2] > b->addr32[2]) 859 return (1); 860 if (a->addr32[2] < b->addr32[2]) 861 return (-1); 862 if (a->addr32[1] > b->addr32[1]) 863 return (1); 864 if (a->addr32[1] < b->addr32[1]) 865 return (-1); 866 if (a->addr32[0] > b->addr32[0]) 867 return (1); 868 if (a->addr32[0] < b->addr32[0]) 869 return (-1); 870 break; 871 #endif /* INET6 */ 872 default: 873 unhandled_af(af); 874 } 875 return (0); 876 } 877 878 static bool 879 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 880 { 881 switch (af) { 882 #ifdef INET 883 case AF_INET: 884 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 885 #endif /* INET */ 886 case AF_INET6: 887 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 888 default: 889 unhandled_af(af); 890 } 891 } 892 893 static void 894 pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk) 895 { 896 897 switch (pd->virtual_proto) { 898 case IPPROTO_TCP: { 899 struct tcphdr *th = &pd->hdr.tcp; 900 901 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 902 pf_change_ap(pd, pd->src, &th->th_sport, 903 &nk->addr[pd->sidx], nk->port[pd->sidx]); 904 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 905 pf_change_ap(pd, pd->dst, &th->th_dport, 906 &nk->addr[pd->didx], nk->port[pd->didx]); 907 m_copyback(pd->m, off, sizeof(*th), (caddr_t)th); 908 break; 909 } 910 case IPPROTO_UDP: { 911 struct udphdr *uh = &pd->hdr.udp; 912 913 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 914 pf_change_ap(pd, pd->src, &uh->uh_sport, 915 &nk->addr[pd->sidx], nk->port[pd->sidx]); 916 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 917 pf_change_ap(pd, pd->dst, &uh->uh_dport, 918 &nk->addr[pd->didx], nk->port[pd->didx]); 919 m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh); 920 break; 921 } 922 case IPPROTO_SCTP: { 923 struct sctphdr *sh = &pd->hdr.sctp; 924 925 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 926 pf_change_ap(pd, pd->src, &sh->src_port, 927 &nk->addr[pd->sidx], nk->port[pd->sidx]); 928 } 929 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 930 pf_change_ap(pd, pd->dst, &sh->dest_port, 931 &nk->addr[pd->didx], nk->port[pd->didx]); 932 } 933 934 break; 935 } 936 case IPPROTO_ICMP: { 937 struct icmp *ih = &pd->hdr.icmp; 938 939 if (nk->port[pd->sidx] != ih->icmp_id) { 940 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 941 ih->icmp_cksum, ih->icmp_id, 942 nk->port[pd->sidx], 0); 943 ih->icmp_id = nk->port[pd->sidx]; 944 pd->sport = &ih->icmp_id; 945 946 m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih); 947 } 948 /* FALLTHROUGH */ 949 } 950 default: 951 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 952 switch (pd->af) { 953 case AF_INET: 954 pf_change_a(&pd->src->v4.s_addr, 955 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 956 0); 957 break; 958 case AF_INET6: 959 pf_addrcpy(pd->src, &nk->addr[pd->sidx], 960 pd->af); 961 break; 962 default: 963 unhandled_af(pd->af); 964 } 965 } 966 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 967 switch (pd->af) { 968 case AF_INET: 969 pf_change_a(&pd->dst->v4.s_addr, 970 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 971 0); 972 break; 973 case AF_INET6: 974 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 975 pd->af); 976 break; 977 default: 978 unhandled_af(pd->af); 979 } 980 } 981 break; 982 } 983 } 984 985 static __inline uint32_t 986 pf_hashkey(const struct pf_state_key *sk) 987 { 988 uint32_t h; 989 990 h = murmur3_32_hash32((const uint32_t *)sk, 991 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 992 V_pf_hashseed); 993 994 return (h & V_pf_hashmask); 995 } 996 997 __inline uint32_t 998 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 999 { 1000 uint32_t h; 1001 1002 switch (af) { 1003 case AF_INET: 1004 h = murmur3_32_hash32((uint32_t *)&addr->v4, 1005 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 1006 break; 1007 case AF_INET6: 1008 h = murmur3_32_hash32((uint32_t *)&addr->v6, 1009 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 1010 break; 1011 default: 1012 unhandled_af(af); 1013 } 1014 1015 return (h & V_pf_srchashmask); 1016 } 1017 1018 static inline uint32_t 1019 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 1020 { 1021 uint32_t h; 1022 1023 h = murmur3_32_hash32((uint32_t *)endpoint, 1024 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 1025 V_pf_hashseed); 1026 return (h & V_pf_udpendpointhashmask); 1027 } 1028 1029 #ifdef ALTQ 1030 static int 1031 pf_state_hash(struct pf_kstate *s) 1032 { 1033 u_int32_t hv = (intptr_t)s / sizeof(*s); 1034 1035 hv ^= crc32(&s->src, sizeof(s->src)); 1036 hv ^= crc32(&s->dst, sizeof(s->dst)); 1037 if (hv == 0) 1038 hv = 1; 1039 return (hv); 1040 } 1041 #endif /* ALTQ */ 1042 1043 static __inline void 1044 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 1045 { 1046 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 1047 s->dst.state = newstate; 1048 if (which == PF_PEER_DST) 1049 return; 1050 if (s->src.state == newstate) 1051 return; 1052 if (s->creatorid == V_pf_status.hostid && 1053 s->key[PF_SK_STACK] != NULL && 1054 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 1055 !(TCPS_HAVEESTABLISHED(s->src.state) || 1056 s->src.state == TCPS_CLOSED) && 1057 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 1058 atomic_add_32(&V_pf_status.states_halfopen, -1); 1059 1060 s->src.state = newstate; 1061 } 1062 1063 bool 1064 pf_init_threshold(struct pf_kthreshold *threshold, 1065 u_int32_t limit, u_int32_t seconds) 1066 { 1067 threshold->limit = limit; 1068 threshold->seconds = seconds; 1069 threshold->cr = counter_rate_alloc(M_NOWAIT, seconds); 1070 1071 return (threshold->cr != NULL); 1072 } 1073 1074 static int 1075 pf_check_threshold(struct pf_kthreshold *threshold) 1076 { 1077 return (counter_ratecheck(threshold->cr, threshold->limit) < 0); 1078 } 1079 1080 static bool 1081 pf_src_connlimit(struct pf_kstate *state) 1082 { 1083 struct pf_overload_entry *pfoe; 1084 struct pf_ksrc_node *src_node = state->sns[PF_SN_LIMIT]; 1085 bool limited = false; 1086 1087 PF_STATE_LOCK_ASSERT(state); 1088 PF_SRC_NODE_LOCK(src_node); 1089 1090 src_node->conn++; 1091 state->src.tcp_est = 1; 1092 1093 if (state->rule->max_src_conn && 1094 state->rule->max_src_conn < 1095 src_node->conn) { 1096 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 1097 limited = true; 1098 } 1099 1100 if (state->rule->max_src_conn_rate.limit && 1101 pf_check_threshold(&src_node->conn_rate)) { 1102 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 1103 limited = true; 1104 } 1105 1106 if (!limited) 1107 goto done; 1108 1109 /* Kill this state. */ 1110 state->timeout = PFTM_PURGE; 1111 pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED); 1112 1113 if (state->rule->overload_tbl == NULL) 1114 goto done; 1115 1116 /* Schedule overloading and flushing task. */ 1117 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 1118 if (pfoe == NULL) 1119 goto done; /* too bad :( */ 1120 1121 bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 1122 pfoe->af = state->key[PF_SK_WIRE]->af; 1123 pfoe->rule = state->rule; 1124 pfoe->dir = state->direction; 1125 PF_OVERLOADQ_LOCK(); 1126 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 1127 PF_OVERLOADQ_UNLOCK(); 1128 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 1129 1130 done: 1131 PF_SRC_NODE_UNLOCK(src_node); 1132 return (limited); 1133 } 1134 1135 static void 1136 pf_overload_task(void *v, int pending) 1137 { 1138 struct pf_overload_head queue; 1139 struct pfr_addr p; 1140 struct pf_overload_entry *pfoe, *pfoe1; 1141 uint32_t killed = 0; 1142 1143 CURVNET_SET((struct vnet *)v); 1144 1145 PF_OVERLOADQ_LOCK(); 1146 queue = V_pf_overloadqueue; 1147 SLIST_INIT(&V_pf_overloadqueue); 1148 PF_OVERLOADQ_UNLOCK(); 1149 1150 bzero(&p, sizeof(p)); 1151 SLIST_FOREACH(pfoe, &queue, next) { 1152 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 1153 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1154 printf("%s: blocking address ", __func__); 1155 pf_print_host(&pfoe->addr, 0, pfoe->af); 1156 printf("\n"); 1157 } 1158 1159 p.pfra_af = pfoe->af; 1160 switch (pfoe->af) { 1161 #ifdef INET 1162 case AF_INET: 1163 p.pfra_net = 32; 1164 p.pfra_ip4addr = pfoe->addr.v4; 1165 break; 1166 #endif /* INET */ 1167 #ifdef INET6 1168 case AF_INET6: 1169 p.pfra_net = 128; 1170 p.pfra_ip6addr = pfoe->addr.v6; 1171 break; 1172 #endif /* INET6 */ 1173 default: 1174 unhandled_af(pfoe->af); 1175 } 1176 1177 PF_RULES_WLOCK(); 1178 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 1179 PF_RULES_WUNLOCK(); 1180 } 1181 1182 /* 1183 * Remove those entries, that don't need flushing. 1184 */ 1185 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 1186 if (pfoe->rule->flush == 0) { 1187 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 1188 free(pfoe, M_PFTEMP); 1189 } else 1190 counter_u64_add( 1191 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 1192 1193 /* If nothing to flush, return. */ 1194 if (SLIST_EMPTY(&queue)) { 1195 CURVNET_RESTORE(); 1196 return; 1197 } 1198 1199 for (int i = 0; i <= V_pf_hashmask; i++) { 1200 struct pf_idhash *ih = &V_pf_idhash[i]; 1201 struct pf_state_key *sk; 1202 struct pf_kstate *s; 1203 1204 PF_HASHROW_LOCK(ih); 1205 LIST_FOREACH(s, &ih->states, entry) { 1206 sk = s->key[PF_SK_WIRE]; 1207 SLIST_FOREACH(pfoe, &queue, next) 1208 if (sk->af == pfoe->af && 1209 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 1210 pfoe->rule == s->rule) && 1211 ((pfoe->dir == PF_OUT && 1212 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 1213 (pfoe->dir == PF_IN && 1214 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 1215 s->timeout = PFTM_PURGE; 1216 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 1217 killed++; 1218 } 1219 } 1220 PF_HASHROW_UNLOCK(ih); 1221 } 1222 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 1223 free(pfoe, M_PFTEMP); 1224 if (V_pf_status.debug >= PF_DEBUG_MISC) 1225 printf("%s: %u states killed", __func__, killed); 1226 1227 CURVNET_RESTORE(); 1228 } 1229 1230 /* 1231 * On node found always returns locked. On not found its configurable. 1232 */ 1233 struct pf_ksrc_node * 1234 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 1235 struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked) 1236 { 1237 struct pf_ksrc_node *n; 1238 1239 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1240 1241 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 1242 PF_HASHROW_LOCK(*sh); 1243 LIST_FOREACH(n, &(*sh)->nodes, entry) 1244 if (n->rule == rule && n->af == af && n->type == sn_type && 1245 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 1246 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 1247 break; 1248 1249 if (n == NULL && !returnlocked) 1250 PF_HASHROW_UNLOCK(*sh); 1251 1252 return (n); 1253 } 1254 1255 bool 1256 pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh) 1257 { 1258 struct pf_ksrc_node *cur; 1259 1260 if ((*sn) == NULL) 1261 return (false); 1262 1263 KASSERT(sh != NULL, ("%s: sh is NULL", __func__)); 1264 1265 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 1266 PF_HASHROW_LOCK(sh); 1267 LIST_FOREACH(cur, &(sh->nodes), entry) { 1268 if (cur == (*sn) && 1269 cur->expire != 1) /* Ignore nodes being killed */ 1270 return (true); 1271 } 1272 PF_HASHROW_UNLOCK(sh); 1273 (*sn) = NULL; 1274 return (false); 1275 } 1276 1277 void 1278 pf_free_src_node(struct pf_ksrc_node *sn) 1279 { 1280 1281 for (int i = 0; i < 2; i++) { 1282 counter_u64_free(sn->bytes[i]); 1283 counter_u64_free(sn->packets[i]); 1284 } 1285 counter_rate_free(sn->conn_rate.cr); 1286 uma_zfree(V_pf_sources_z, sn); 1287 } 1288 1289 static u_short 1290 pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX], 1291 struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule, 1292 struct pf_addr *src, sa_family_t af, struct pf_addr *raddr, 1293 struct pfi_kkif *rkif, sa_family_t raf, pf_sn_types_t sn_type) 1294 { 1295 u_short reason = 0; 1296 struct pf_krule *r_track = rule; 1297 struct pf_ksrc_node **sn = &(sns[sn_type]); 1298 struct pf_srchash **sh = &(snhs[sn_type]); 1299 1300 KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL), 1301 ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__)); 1302 1303 KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK), 1304 ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__)); 1305 1306 /* 1307 * XXX: There could be a KASSERT for 1308 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR) 1309 * but we'd need to pass pool *only* for this KASSERT. 1310 */ 1311 1312 if ( (rule->rule_flag & PFRULE_SRCTRACK) && 1313 !(rule->rule_flag & PFRULE_RULESRCTRACK)) 1314 r_track = &V_pf_default_rule; 1315 1316 /* 1317 * Request the sh to always be locked, as we might insert a new sn. 1318 */ 1319 if (*sn == NULL) 1320 *sn = pf_find_src_node(src, r_track, af, sh, sn_type, true); 1321 1322 if (*sn == NULL) { 1323 PF_HASHROW_ASSERT(*sh); 1324 1325 if (sn_type == PF_SN_LIMIT && rule->max_src_nodes && 1326 counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) { 1327 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1328 reason = PFRES_SRCLIMIT; 1329 goto done; 1330 } 1331 1332 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1333 if ((*sn) == NULL) { 1334 reason = PFRES_MEMORY; 1335 goto done; 1336 } 1337 1338 for (int i = 0; i < 2; i++) { 1339 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1340 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1341 1342 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1343 pf_free_src_node(*sn); 1344 reason = PFRES_MEMORY; 1345 goto done; 1346 } 1347 } 1348 1349 if (sn_type == PF_SN_LIMIT) 1350 if (! pf_init_threshold(&(*sn)->conn_rate, 1351 rule->max_src_conn_rate.limit, 1352 rule->max_src_conn_rate.seconds)) { 1353 pf_free_src_node(*sn); 1354 reason = PFRES_MEMORY; 1355 goto done; 1356 } 1357 1358 MPASS((*sn)->lock == NULL); 1359 (*sn)->lock = &(*sh)->lock; 1360 1361 (*sn)->af = af; 1362 (*sn)->rule = r_track; 1363 pf_addrcpy(&(*sn)->addr, src, af); 1364 if (raddr != NULL) 1365 pf_addrcpy(&(*sn)->raddr, raddr, raf); 1366 (*sn)->rkif = rkif; 1367 (*sn)->raf = raf; 1368 LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry); 1369 (*sn)->creation = time_uptime; 1370 (*sn)->ruletype = rule->action; 1371 (*sn)->type = sn_type; 1372 counter_u64_add(r_track->src_nodes[sn_type], 1); 1373 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1374 } else { 1375 if (sn_type == PF_SN_LIMIT && rule->max_src_states && 1376 (*sn)->states >= rule->max_src_states) { 1377 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1378 1); 1379 reason = PFRES_SRCLIMIT; 1380 goto done; 1381 } 1382 } 1383 done: 1384 if (reason == 0) 1385 (*sn)->states++; 1386 else 1387 (*sn) = NULL; 1388 1389 PF_HASHROW_UNLOCK(*sh); 1390 return (reason); 1391 } 1392 1393 void 1394 pf_unlink_src_node(struct pf_ksrc_node *src) 1395 { 1396 PF_SRC_NODE_LOCK_ASSERT(src); 1397 1398 LIST_REMOVE(src, entry); 1399 if (src->rule) 1400 counter_u64_add(src->rule->src_nodes[src->type], -1); 1401 } 1402 1403 u_int 1404 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1405 { 1406 struct pf_ksrc_node *sn, *tmp; 1407 u_int count = 0; 1408 1409 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1410 pf_free_src_node(sn); 1411 count++; 1412 } 1413 1414 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1415 1416 return (count); 1417 } 1418 1419 void 1420 pf_mtag_initialize(void) 1421 { 1422 1423 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1424 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1425 UMA_ALIGN_PTR, 0); 1426 } 1427 1428 /* Per-vnet data storage structures initialization. */ 1429 void 1430 pf_initialize(void) 1431 { 1432 struct hashalloc_args ha = { 1433 .mflags = M_NOWAIT, /* see bf56a3fe47ef4 and bug 209475 */ 1434 .mtype = M_PFHASH, 1435 .type = HASH_TYPE_POWER2, 1436 .head = HASH_HEAD_LIST, 1437 .lock = HASH_LOCK_MTX, 1438 }; 1439 1440 V_pf_hashseed = arc4random(); 1441 1442 /* States and state keys storage. */ 1443 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1444 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1445 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1446 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1447 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1448 V_pf_state_key_z = uma_zcreate("pf state keys", 1449 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1450 UMA_ALIGN_PTR, 0); 1451 retry_waitok: 1452 ha.size = V_pf_hashsize; 1453 ha.lname = "pf_keyhash"; 1454 ha.lopts = MTX_DEF | MTX_DUPOK; 1455 V_pf_keyhash = hashalloc(&ha); 1456 ha.lname = "pf_idhash"; 1457 ha.lopts = MTX_DEF; 1458 V_pf_idhash = hashalloc(&ha); 1459 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1460 printf("pf: Unable to allocate memory for " 1461 "state_hashsize %lu.\n", V_pf_hashsize); 1462 hashfree(V_pf_keyhash, &ha); 1463 hashfree(V_pf_idhash, &ha); 1464 V_pf_hashsize = PF_HASHSIZ; 1465 ha.mflags = M_WAITOK; 1466 goto retry_waitok; 1467 } 1468 V_pf_hashsize = ha.size; 1469 V_pf_hashmask = V_pf_hashsize - 1; 1470 1471 /* Source nodes. */ 1472 V_pf_sources_z = uma_zcreate("pf source nodes", 1473 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1474 0); 1475 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1476 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1477 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1478 ha.size = V_pf_srchashsize; 1479 ha.lname = "pf_srchash"; 1480 ha.lopts = MTX_DEF; 1481 ha.mflags = M_NOWAIT; 1482 retry_waitok2: 1483 V_pf_srchash = hashalloc(&ha); 1484 if (V_pf_srchash == NULL) { 1485 printf("pf: Unable to allocate memory for " 1486 "source_hashsize %lu.\n", V_pf_srchashsize); 1487 ha.size = PF_SRCHASHSIZ; 1488 ha.mflags = M_WAITOK; 1489 goto retry_waitok2; 1490 } 1491 V_pf_srchashmask = ha.size; 1492 V_pf_srchashmask = V_pf_srchashsize - 1; 1493 1494 /* UDP endpoint mappings. */ 1495 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1496 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1497 UMA_ALIGN_PTR, 0); 1498 ha.size = V_pf_udpendpointhashsize; 1499 ha.lname = "pf_udpendpointhash"; 1500 ha.lopts = MTX_DEF | MTX_DUPOK; 1501 ha.mflags = M_NOWAIT; 1502 retry_waitok3: 1503 V_pf_udpendpointhash = hashalloc(&ha); 1504 if (V_pf_udpendpointhash == NULL) { 1505 printf("pf: Unable to allocate memory for " 1506 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1507 ha.size = PF_UDPENDHASHSIZ; 1508 ha.mflags = M_WAITOK; 1509 goto retry_waitok3; 1510 } 1511 V_pf_udpendpointhashsize = ha.size; 1512 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1513 1514 /* Anchors */ 1515 V_pf_anchor_z = uma_zcreate("pf anchors", 1516 sizeof(struct pf_kanchor), NULL, NULL, NULL, NULL, 1517 UMA_ALIGN_PTR, 0); 1518 V_pf_limits[PF_LIMIT_ANCHORS].zone = V_pf_anchor_z; 1519 uma_zone_set_max(V_pf_anchor_z, PF_ANCHOR_HIWAT); 1520 uma_zone_set_warning(V_pf_anchor_z, "PF anchor limit reached"); 1521 1522 V_pf_eth_anchor_z = uma_zcreate("pf Ethernet anchors", 1523 sizeof(struct pf_keth_anchor), NULL, NULL, NULL, NULL, 1524 UMA_ALIGN_PTR, 0); 1525 V_pf_limits[PF_LIMIT_ETH_ANCHORS].zone = V_pf_eth_anchor_z; 1526 uma_zone_set_max(V_pf_eth_anchor_z, PF_ANCHOR_HIWAT); 1527 uma_zone_set_warning(V_pf_eth_anchor_z, "PF Ethernet anchor limit reached"); 1528 1529 /* ALTQ */ 1530 TAILQ_INIT(&V_pf_altqs[0]); 1531 TAILQ_INIT(&V_pf_altqs[1]); 1532 TAILQ_INIT(&V_pf_altqs[2]); 1533 TAILQ_INIT(&V_pf_altqs[3]); 1534 TAILQ_INIT(&V_pf_pabuf[0]); 1535 TAILQ_INIT(&V_pf_pabuf[1]); 1536 TAILQ_INIT(&V_pf_pabuf[2]); 1537 V_pf_altqs_active = &V_pf_altqs[0]; 1538 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1539 V_pf_altqs_inactive = &V_pf_altqs[2]; 1540 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1541 1542 /* Send & overload+flush queues. */ 1543 STAILQ_INIT(&V_pf_sendqueue); 1544 SLIST_INIT(&V_pf_overloadqueue); 1545 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1546 1547 /* Unlinked, but may be referenced rules. */ 1548 TAILQ_INIT(&V_pf_unlinked_rules); 1549 1550 /* State limiters */ 1551 RB_INIT(&V_pf_statelim_id_tree_inactive); 1552 RB_INIT(&V_pf_statelim_nm_tree_inactive); 1553 TAILQ_INIT(&V_pf_statelim_list_inactive); 1554 1555 RB_INIT(&V_pf_statelim_id_tree_active); 1556 TAILQ_INIT(&V_pf_statelim_list_active); 1557 1558 /* Source limiters */ 1559 RB_INIT(&V_pf_sourcelim_id_tree_active); 1560 TAILQ_INIT(&V_pf_sourcelim_list_active); 1561 1562 RB_INIT(&V_pf_sourcelim_id_tree_inactive); 1563 RB_INIT(&V_pf_sourcelim_nm_tree_inactive); 1564 TAILQ_INIT(&V_pf_sourcelim_list_inactive); 1565 } 1566 1567 void 1568 pf_mtag_cleanup(void) 1569 { 1570 1571 uma_zdestroy(pf_mtag_z); 1572 } 1573 1574 void 1575 pf_cleanup(void) 1576 { 1577 struct hashalloc_args ha = { 1578 .size = V_pf_hashsize, 1579 .mtype = M_PFHASH, 1580 .head = HASH_HEAD_LIST, 1581 .lock = HASH_LOCK_MTX, 1582 }; 1583 struct pf_send_entry *pfse, *next; 1584 1585 hashfree(V_pf_keyhash, &ha); 1586 hashfree(V_pf_idhash, &ha); 1587 ha.size = V_pf_srchashsize; 1588 hashfree(V_pf_srchash, &ha); 1589 ha.size = V_pf_udpendpointhashsize; 1590 hashfree(V_pf_udpendpointhash, &ha); 1591 1592 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1593 m_freem(pfse->pfse_m); 1594 free(pfse, M_PFTEMP); 1595 } 1596 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1597 1598 uma_zdestroy(V_pf_sources_z); 1599 uma_zdestroy(V_pf_state_z); 1600 uma_zdestroy(V_pf_state_key_z); 1601 uma_zdestroy(V_pf_udp_mapping_z); 1602 uma_zdestroy(V_pf_anchor_z); 1603 uma_zdestroy(V_pf_eth_anchor_z); 1604 } 1605 1606 static int 1607 pf_mtag_uminit(void *mem, int size, int how) 1608 { 1609 struct m_tag *t; 1610 1611 t = (struct m_tag *)mem; 1612 t->m_tag_cookie = MTAG_ABI_COMPAT; 1613 t->m_tag_id = PACKET_TAG_PF; 1614 t->m_tag_len = sizeof(struct pf_mtag); 1615 t->m_tag_free = pf_mtag_free; 1616 1617 return (0); 1618 } 1619 1620 static void 1621 pf_mtag_free(struct m_tag *t) 1622 { 1623 1624 uma_zfree(pf_mtag_z, t); 1625 } 1626 1627 struct pf_mtag * 1628 pf_get_mtag(struct mbuf *m) 1629 { 1630 struct m_tag *mtag; 1631 1632 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1633 return ((struct pf_mtag *)(mtag + 1)); 1634 1635 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1636 if (mtag == NULL) 1637 return (NULL); 1638 bzero(mtag + 1, sizeof(struct pf_mtag)); 1639 m_tag_prepend(m, mtag); 1640 1641 return ((struct pf_mtag *)(mtag + 1)); 1642 } 1643 1644 static int 1645 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1646 struct pf_kstate *s) 1647 { 1648 struct pf_keyhash *khs, *khw, *kh; 1649 struct pf_state_key *sk, *cur; 1650 struct pf_kstate *si, *olds = NULL; 1651 int idx; 1652 1653 NET_EPOCH_ASSERT(); 1654 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1655 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1656 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1657 1658 /* 1659 * We need to lock hash slots of both keys. To avoid deadlock 1660 * we always lock the slot with lower address first. Unlock order 1661 * isn't important. 1662 * 1663 * We also need to lock ID hash slot before dropping key 1664 * locks. On success we return with ID hash slot locked. 1665 */ 1666 1667 if (skw == sks) { 1668 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1669 PF_HASHROW_LOCK(khs); 1670 } else { 1671 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1672 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1673 if (khs == khw) { 1674 PF_HASHROW_LOCK(khs); 1675 } else if (khs < khw) { 1676 PF_HASHROW_LOCK(khs); 1677 PF_HASHROW_LOCK(khw); 1678 } else { 1679 PF_HASHROW_LOCK(khw); 1680 PF_HASHROW_LOCK(khs); 1681 } 1682 } 1683 1684 #define KEYS_UNLOCK() do { \ 1685 if (khs != khw) { \ 1686 PF_HASHROW_UNLOCK(khs); \ 1687 PF_HASHROW_UNLOCK(khw); \ 1688 } else \ 1689 PF_HASHROW_UNLOCK(khs); \ 1690 } while (0) 1691 1692 /* 1693 * First run: start with wire key. 1694 */ 1695 sk = skw; 1696 kh = khw; 1697 idx = PF_SK_WIRE; 1698 1699 MPASS(s->lock == NULL); 1700 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1701 1702 keyattach: 1703 LIST_FOREACH(cur, &kh->keys, entry) 1704 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1705 break; 1706 1707 if (cur != NULL) { 1708 /* Key exists. Check for same kif, if none, add to key. */ 1709 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1710 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1711 1712 PF_HASHROW_LOCK(ih); 1713 if (si->kif == s->kif && 1714 ((si->key[PF_SK_WIRE]->af == sk->af && 1715 si->direction == s->direction) || 1716 (si->key[PF_SK_WIRE]->af != 1717 si->key[PF_SK_STACK]->af && 1718 sk->af == si->key[PF_SK_STACK]->af && 1719 si->direction != s->direction))) { 1720 bool reuse = false; 1721 1722 if (sk->proto == IPPROTO_TCP && 1723 si->src.state >= TCPS_FIN_WAIT_2 && 1724 si->dst.state >= TCPS_FIN_WAIT_2) 1725 reuse = true; 1726 1727 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1728 printf("pf: %s key attach " 1729 "%s on %s: ", 1730 (idx == PF_SK_WIRE) ? 1731 "wire" : "stack", 1732 reuse ? "reuse" : "failed", 1733 s->kif->pfik_name); 1734 pf_print_state_parts(s, 1735 (idx == PF_SK_WIRE) ? 1736 sk : NULL, 1737 (idx == PF_SK_STACK) ? 1738 sk : NULL); 1739 printf(", existing: "); 1740 pf_print_state_parts(si, 1741 (idx == PF_SK_WIRE) ? 1742 sk : NULL, 1743 (idx == PF_SK_STACK) ? 1744 sk : NULL); 1745 printf("\n"); 1746 } 1747 1748 if (reuse) { 1749 /* 1750 * New state matches an old >FIN_WAIT_2 1751 * state. We can't drop key hash locks, 1752 * thus we can't unlink it properly. 1753 * 1754 * As a workaround we drop it into 1755 * TCPS_CLOSED state, schedule purge 1756 * ASAP and push it into the very end 1757 * of the slot TAILQ, so that it won't 1758 * conflict with our new state. 1759 */ 1760 pf_set_protostate(si, PF_PEER_BOTH, 1761 TCPS_CLOSED); 1762 si->timeout = PFTM_PURGE; 1763 olds = si; 1764 } else { 1765 s->timeout = PFTM_UNLINKED; 1766 if (idx == PF_SK_STACK) 1767 /* 1768 * Remove the wire key from 1769 * the hash. Other threads 1770 * can't be referencing it 1771 * because we still hold the 1772 * hash lock. 1773 */ 1774 pf_state_key_detach(s, 1775 PF_SK_WIRE); 1776 PF_HASHROW_UNLOCK(ih); 1777 KEYS_UNLOCK(); 1778 if (idx == PF_SK_WIRE) 1779 /* 1780 * We've not inserted either key. 1781 * Free both. 1782 */ 1783 uma_zfree(V_pf_state_key_z, skw); 1784 if (skw != sks) 1785 uma_zfree( 1786 V_pf_state_key_z, 1787 sks); 1788 return (EEXIST); /* collision! */ 1789 } 1790 } 1791 PF_HASHROW_UNLOCK(ih); 1792 } 1793 uma_zfree(V_pf_state_key_z, sk); 1794 s->key[idx] = cur; 1795 } else { 1796 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1797 s->key[idx] = sk; 1798 } 1799 1800 stateattach: 1801 /* List is sorted, if-bound states before floating. */ 1802 if (s->kif == V_pfi_all) 1803 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1804 else 1805 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1806 1807 if (olds) { 1808 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1809 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1810 key_list[idx]); 1811 olds = NULL; 1812 } 1813 1814 /* 1815 * Attach done. See how should we (or should not?) 1816 * attach a second key. 1817 */ 1818 if (sks == skw) { 1819 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1820 idx = PF_SK_STACK; 1821 sks = NULL; 1822 goto stateattach; 1823 } else if (sks != NULL) { 1824 /* 1825 * Continue attaching with stack key. 1826 */ 1827 sk = sks; 1828 kh = khs; 1829 idx = PF_SK_STACK; 1830 sks = NULL; 1831 goto keyattach; 1832 } 1833 1834 PF_STATE_LOCK(s); 1835 KEYS_UNLOCK(); 1836 1837 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1838 ("%s failure", __func__)); 1839 1840 return (0); 1841 #undef KEYS_UNLOCK 1842 } 1843 1844 static void 1845 pf_detach_state(struct pf_kstate *s) 1846 { 1847 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1848 struct pf_keyhash *kh; 1849 1850 NET_EPOCH_ASSERT(); 1851 MPASS(s->timeout >= PFTM_MAX); 1852 1853 pf_sctp_multihome_detach_addr(s); 1854 1855 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1856 V_pflow_export_state_ptr(s); 1857 1858 if (sks != NULL) { 1859 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1860 PF_HASHROW_LOCK(kh); 1861 if (s->key[PF_SK_STACK] != NULL) 1862 pf_state_key_detach(s, PF_SK_STACK); 1863 /* 1864 * If both point to same key, then we are done. 1865 */ 1866 if (sks == s->key[PF_SK_WIRE]) { 1867 pf_state_key_detach(s, PF_SK_WIRE); 1868 PF_HASHROW_UNLOCK(kh); 1869 return; 1870 } 1871 PF_HASHROW_UNLOCK(kh); 1872 } 1873 1874 if (s->key[PF_SK_WIRE] != NULL) { 1875 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1876 PF_HASHROW_LOCK(kh); 1877 if (s->key[PF_SK_WIRE] != NULL) 1878 pf_state_key_detach(s, PF_SK_WIRE); 1879 PF_HASHROW_UNLOCK(kh); 1880 } 1881 } 1882 1883 static void 1884 pf_state_key_detach(struct pf_kstate *s, int idx) 1885 { 1886 struct pf_state_key *sk = s->key[idx]; 1887 #ifdef INVARIANTS 1888 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1889 1890 PF_HASHROW_ASSERT(kh); 1891 #endif /* INVARIANTS */ 1892 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1893 s->key[idx] = NULL; 1894 1895 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1896 LIST_REMOVE(sk, entry); 1897 uma_zfree(V_pf_state_key_z, sk); 1898 } 1899 } 1900 1901 static int 1902 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1903 { 1904 struct pf_state_key *sk = mem; 1905 1906 bzero(sk, sizeof(struct pf_state_key_cmp)); 1907 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1908 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1909 1910 return (0); 1911 } 1912 1913 static int 1914 pf_state_key_addr_setup(struct pf_pdesc *pd, 1915 struct pf_state_key_cmp *key, int multi) 1916 { 1917 struct pf_addr *saddr = pd->src; 1918 struct pf_addr *daddr = pd->dst; 1919 #ifdef INET6 1920 struct nd_neighbor_solicit nd; 1921 struct pf_addr *target; 1922 1923 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1924 goto copy; 1925 1926 switch (pd->hdr.icmp6.icmp6_type) { 1927 case ND_NEIGHBOR_SOLICIT: 1928 if (multi) 1929 return (-1); 1930 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL, 1931 pd->af)) 1932 return (-1); 1933 target = (struct pf_addr *)&nd.nd_ns_target; 1934 daddr = target; 1935 break; 1936 case ND_NEIGHBOR_ADVERT: 1937 if (multi) 1938 return (-1); 1939 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL, 1940 pd->af)) 1941 return (-1); 1942 target = (struct pf_addr *)&nd.nd_ns_target; 1943 saddr = target; 1944 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1945 key->addr[pd->didx].addr32[0] = 0; 1946 key->addr[pd->didx].addr32[1] = 0; 1947 key->addr[pd->didx].addr32[2] = 0; 1948 key->addr[pd->didx].addr32[3] = 0; 1949 daddr = NULL; /* overwritten */ 1950 } 1951 break; 1952 default: 1953 if (multi) { 1954 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1955 key->addr[pd->sidx].addr32[1] = 0; 1956 key->addr[pd->sidx].addr32[2] = 0; 1957 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1958 saddr = NULL; /* overwritten */ 1959 } 1960 } 1961 copy: 1962 #endif /* INET6 */ 1963 if (saddr) 1964 pf_addrcpy(&key->addr[pd->sidx], saddr, pd->af); 1965 if (daddr) 1966 pf_addrcpy(&key->addr[pd->didx], daddr, pd->af); 1967 1968 return (0); 1969 } 1970 1971 int 1972 pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport, 1973 struct pf_state_key **sk, struct pf_state_key **nk) 1974 { 1975 *sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1976 if (*sk == NULL) 1977 return (ENOMEM); 1978 1979 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk, 1980 0)) { 1981 uma_zfree(V_pf_state_key_z, *sk); 1982 *sk = NULL; 1983 return (ENOMEM); 1984 } 1985 1986 (*sk)->port[pd->sidx] = sport; 1987 (*sk)->port[pd->didx] = dport; 1988 (*sk)->proto = pd->proto; 1989 (*sk)->af = pd->af; 1990 1991 *nk = pf_state_key_clone(*sk); 1992 if (*nk == NULL) { 1993 uma_zfree(V_pf_state_key_z, *sk); 1994 *sk = NULL; 1995 return (ENOMEM); 1996 } 1997 1998 if (pd->af != pd->naf) { 1999 (*sk)->port[pd->sidx] = pd->osport; 2000 (*sk)->port[pd->didx] = pd->odport; 2001 2002 (*nk)->af = pd->naf; 2003 2004 /* 2005 * We're overwriting an address here, so potentially there's bits of an IPv6 2006 * address left in here. Clear that out first. 2007 */ 2008 bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0])); 2009 bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1])); 2010 if (pd->dir == PF_IN) { 2011 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->nsaddr, 2012 pd->naf); 2013 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->ndaddr, 2014 pd->naf); 2015 (*nk)->port[pd->didx] = pd->nsport; 2016 (*nk)->port[pd->sidx] = pd->ndport; 2017 } else { 2018 pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->nsaddr, 2019 pd->naf); 2020 pf_addrcpy(&(*nk)->addr[pd->didx], &pd->ndaddr, 2021 pd->naf); 2022 (*nk)->port[pd->sidx] = pd->nsport; 2023 (*nk)->port[pd->didx] = pd->ndport; 2024 } 2025 2026 switch (pd->proto) { 2027 case IPPROTO_ICMP: 2028 (*nk)->proto = IPPROTO_ICMPV6; 2029 break; 2030 case IPPROTO_ICMPV6: 2031 (*nk)->proto = IPPROTO_ICMP; 2032 break; 2033 default: 2034 (*nk)->proto = pd->proto; 2035 } 2036 } 2037 2038 return (0); 2039 } 2040 2041 struct pf_state_key * 2042 pf_state_key_clone(const struct pf_state_key *orig) 2043 { 2044 struct pf_state_key *sk; 2045 2046 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 2047 if (sk == NULL) 2048 return (NULL); 2049 2050 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 2051 2052 return (sk); 2053 } 2054 2055 int 2056 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 2057 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 2058 { 2059 struct pf_idhash *ih; 2060 struct pf_kstate *cur; 2061 int error; 2062 2063 NET_EPOCH_ASSERT(); 2064 2065 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 2066 ("%s: sks not pristine", __func__)); 2067 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 2068 ("%s: skw not pristine", __func__)); 2069 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 2070 2071 s->kif = kif; 2072 s->orig_kif = orig_kif; 2073 2074 if (s->id == 0 && s->creatorid == 0) { 2075 s->id = alloc_unr64(&V_pf_stateid); 2076 s->id = htobe64(s->id); 2077 s->creatorid = V_pf_status.hostid; 2078 } 2079 2080 /* Returns with ID locked on success. */ 2081 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 2082 return (error); 2083 skw = sks = NULL; 2084 2085 ih = &V_pf_idhash[PF_IDHASH(s)]; 2086 PF_HASHROW_ASSERT(ih); 2087 LIST_FOREACH(cur, &ih->states, entry) 2088 if (cur->id == s->id && cur->creatorid == s->creatorid) 2089 break; 2090 2091 if (cur != NULL) { 2092 s->timeout = PFTM_UNLINKED; 2093 PF_HASHROW_UNLOCK(ih); 2094 if (V_pf_status.debug >= PF_DEBUG_MISC) { 2095 printf("pf: state ID collision: " 2096 "id: %016llx creatorid: %08x\n", 2097 (unsigned long long)be64toh(s->id), 2098 ntohl(s->creatorid)); 2099 } 2100 pf_detach_state(s); 2101 return (EEXIST); 2102 } 2103 LIST_INSERT_HEAD(&ih->states, s, entry); 2104 /* One for keys, one for ID hash. */ 2105 refcount_init(&s->refs, 2); 2106 2107 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 2108 if (V_pfsync_insert_state_ptr != NULL) 2109 V_pfsync_insert_state_ptr(s); 2110 2111 /* Returns locked. */ 2112 return (0); 2113 } 2114 2115 /* 2116 * Find state by ID: returns with locked row on success. 2117 */ 2118 struct pf_kstate * 2119 pf_find_state_byid(uint64_t id, uint32_t creatorid) 2120 { 2121 struct pf_idhash *ih; 2122 struct pf_kstate *s; 2123 2124 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 2125 2126 ih = &V_pf_idhash[PF_IDHASHID(id)]; 2127 2128 PF_HASHROW_LOCK(ih); 2129 LIST_FOREACH(s, &ih->states, entry) 2130 if (s->id == id && s->creatorid == creatorid) 2131 break; 2132 2133 if (s == NULL) 2134 PF_HASHROW_UNLOCK(ih); 2135 2136 return (s); 2137 } 2138 2139 /* 2140 * Find state by key. 2141 * Returns with ID hash slot locked on success. 2142 */ 2143 static int 2144 pf_find_state(struct pf_pdesc *pd, const struct pf_state_key_cmp *key, 2145 struct pf_kstate **state) 2146 { 2147 struct pf_keyhash *kh; 2148 struct pf_state_key *sk; 2149 struct pf_kstate *s; 2150 int idx; 2151 2152 *state = NULL; 2153 2154 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 2155 2156 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 2157 2158 PF_HASHROW_LOCK(kh); 2159 LIST_FOREACH(sk, &kh->keys, entry) 2160 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 2161 break; 2162 if (sk == NULL) { 2163 PF_HASHROW_UNLOCK(kh); 2164 return (PF_DROP); 2165 } 2166 2167 idx = (pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 2168 2169 /* List is sorted, if-bound states before floating ones. */ 2170 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 2171 if (s->kif == V_pfi_all || s->kif == pd->kif || 2172 s->orig_kif == pd->kif) { 2173 PF_STATE_LOCK(s); 2174 PF_HASHROW_UNLOCK(kh); 2175 if (__predict_false(s->timeout >= PFTM_MAX)) { 2176 /* 2177 * State is either being processed by 2178 * pf_remove_state() in an other thread, or 2179 * is scheduled for immediate expiry. 2180 */ 2181 PF_STATE_UNLOCK(s); 2182 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 2183 key, (pd->dir), pd, *state); 2184 return (PF_DROP); 2185 } 2186 goto out; 2187 } 2188 2189 /* Look through the other list, in case of AF-TO */ 2190 idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE; 2191 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 2192 if (s->timeout < PFTM_MAX && 2193 s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af) 2194 continue; 2195 2196 if (s->kif == V_pfi_all || s->kif == pd->kif || 2197 s->orig_kif == pd->kif) { 2198 PF_STATE_LOCK(s); 2199 PF_HASHROW_UNLOCK(kh); 2200 if (__predict_false(s->timeout >= PFTM_MAX)) { 2201 /* 2202 * State is either being processed by 2203 * pf_remove_state() in an other thread, or 2204 * is scheduled for immediate expiry. 2205 */ 2206 PF_STATE_UNLOCK(s); 2207 SDT_PROBE5(pf, ip, state, lookup, pd->kif, 2208 key, (pd->dir), pd, NULL); 2209 return (PF_DROP); 2210 } 2211 goto out; 2212 } 2213 } 2214 2215 PF_HASHROW_UNLOCK(kh); 2216 2217 out: 2218 SDT_PROBE5(pf, ip, state, lookup, pd->kif, key, (pd->dir), pd, *state); 2219 2220 if (s == NULL || s->timeout == PFTM_PURGE) { 2221 if (s) 2222 PF_STATE_UNLOCK(s); 2223 return (PF_DROP); 2224 } 2225 2226 if ((s)->rule->pktrate.limit && pd->dir == (s)->direction) { 2227 if (pf_check_threshold(&(s)->rule->pktrate)) { 2228 PF_STATE_UNLOCK(s); 2229 return (PF_DROP); 2230 } 2231 } 2232 if (PACKET_LOOPED(pd)) { 2233 PF_STATE_UNLOCK(s); 2234 return (PF_PASS); 2235 } 2236 2237 *state = s; 2238 2239 return (PF_MATCH); 2240 } 2241 2242 /* 2243 * Returns with ID hash slot locked on success. 2244 */ 2245 struct pf_kstate * 2246 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 2247 { 2248 struct pf_keyhash *kh; 2249 struct pf_state_key *sk; 2250 struct pf_kstate *s, *ret = NULL; 2251 int idx, inout = 0; 2252 2253 if (more != NULL) 2254 *more = 0; 2255 2256 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 2257 2258 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 2259 2260 PF_HASHROW_LOCK(kh); 2261 LIST_FOREACH(sk, &kh->keys, entry) 2262 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 2263 break; 2264 if (sk == NULL) { 2265 PF_HASHROW_UNLOCK(kh); 2266 return (NULL); 2267 } 2268 switch (dir) { 2269 case PF_IN: 2270 idx = PF_SK_WIRE; 2271 break; 2272 case PF_OUT: 2273 idx = PF_SK_STACK; 2274 break; 2275 case PF_INOUT: 2276 idx = PF_SK_WIRE; 2277 inout = 1; 2278 break; 2279 default: 2280 panic("%s: dir %u", __func__, dir); 2281 } 2282 second_run: 2283 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 2284 if (more == NULL) { 2285 PF_STATE_LOCK(s); 2286 PF_HASHROW_UNLOCK(kh); 2287 return (s); 2288 } 2289 2290 if (ret) 2291 (*more)++; 2292 else { 2293 ret = s; 2294 PF_STATE_LOCK(s); 2295 } 2296 } 2297 if (inout == 1) { 2298 inout = 0; 2299 idx = PF_SK_STACK; 2300 goto second_run; 2301 } 2302 PF_HASHROW_UNLOCK(kh); 2303 2304 return (ret); 2305 } 2306 2307 /* 2308 * FIXME 2309 * This routine is inefficient -- locks the state only to unlock immediately on 2310 * return. 2311 * It is racy -- after the state is unlocked nothing stops other threads from 2312 * removing it. 2313 */ 2314 bool 2315 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 2316 { 2317 struct pf_kstate *s; 2318 2319 s = pf_find_state_all(key, dir, NULL); 2320 if (s != NULL) { 2321 PF_STATE_UNLOCK(s); 2322 return (true); 2323 } 2324 return (false); 2325 } 2326 2327 void 2328 pf_state_peer_hton(const struct pf_state_peer *s, struct pf_state_peer_export *d) 2329 { 2330 d->seqlo = htonl(s->seqlo); 2331 d->seqhi = htonl(s->seqhi); 2332 d->seqdiff = htonl(s->seqdiff); 2333 d->max_win = htons(s->max_win); 2334 d->mss = htons(s->mss); 2335 d->state = s->state; 2336 d->wscale = s->wscale; 2337 if (s->scrub) { 2338 d->scrub.pfss_flags = htons( 2339 s->scrub->pfss_flags & PFSS_TIMESTAMP); 2340 d->scrub.pfss_ttl = (s)->scrub->pfss_ttl; 2341 d->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod); 2342 d->scrub.scrub_flag = PF_SCRUB_FLAG_VALID; 2343 } 2344 } 2345 2346 void 2347 pf_state_peer_ntoh(const struct pf_state_peer_export *s, struct pf_state_peer *d) 2348 { 2349 d->seqlo = ntohl(s->seqlo); 2350 d->seqhi = ntohl(s->seqhi); 2351 d->seqdiff = ntohl(s->seqdiff); 2352 d->max_win = ntohs(s->max_win); 2353 d->mss = ntohs(s->mss); 2354 d->state = s->state; 2355 d->wscale = s->wscale; 2356 if (s->scrub.scrub_flag == PF_SCRUB_FLAG_VALID && 2357 d->scrub != NULL) { 2358 d->scrub->pfss_flags = ntohs(s->scrub.pfss_flags) & 2359 PFSS_TIMESTAMP; 2360 d->scrub->pfss_ttl = s->scrub.pfss_ttl; 2361 d->scrub->pfss_ts_mod = ntohl(s->scrub.pfss_ts_mod); 2362 } 2363 } 2364 2365 struct pf_udp_mapping * 2366 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 2367 struct pf_addr *nat_addr, uint16_t nat_port) 2368 { 2369 struct pf_udp_mapping *mapping; 2370 2371 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 2372 if (mapping == NULL) 2373 return (NULL); 2374 pf_addrcpy(&mapping->endpoints[0].addr, src_addr, af); 2375 mapping->endpoints[0].port = src_port; 2376 mapping->endpoints[0].af = af; 2377 mapping->endpoints[0].mapping = mapping; 2378 pf_addrcpy(&mapping->endpoints[1].addr, nat_addr, af); 2379 mapping->endpoints[1].port = nat_port; 2380 mapping->endpoints[1].af = af; 2381 mapping->endpoints[1].mapping = mapping; 2382 refcount_init(&mapping->refs, 1); 2383 return (mapping); 2384 } 2385 2386 int 2387 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 2388 { 2389 struct pf_udpendpointhash *h0, *h1; 2390 struct pf_udp_endpoint *endpoint; 2391 int ret = EEXIST; 2392 2393 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2394 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2395 if (h0 == h1) { 2396 PF_HASHROW_LOCK(h0); 2397 } else if (h0 < h1) { 2398 PF_HASHROW_LOCK(h0); 2399 PF_HASHROW_LOCK(h1); 2400 } else { 2401 PF_HASHROW_LOCK(h1); 2402 PF_HASHROW_LOCK(h0); 2403 } 2404 2405 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 2406 if (bcmp(endpoint, &mapping->endpoints[0], 2407 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2408 break; 2409 } 2410 if (endpoint != NULL) 2411 goto cleanup; 2412 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 2413 if (bcmp(endpoint, &mapping->endpoints[1], 2414 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2415 break; 2416 } 2417 if (endpoint != NULL) 2418 goto cleanup; 2419 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 2420 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 2421 ret = 0; 2422 2423 cleanup: 2424 if (h0 != h1) { 2425 PF_HASHROW_UNLOCK(h0); 2426 PF_HASHROW_UNLOCK(h1); 2427 } else { 2428 PF_HASHROW_UNLOCK(h0); 2429 } 2430 return (ret); 2431 } 2432 2433 void 2434 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 2435 { 2436 /* refcount is synchronized on the source endpoint's row lock */ 2437 struct pf_udpendpointhash *h0, *h1; 2438 2439 if (mapping == NULL) 2440 return; 2441 2442 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 2443 PF_HASHROW_LOCK(h0); 2444 if (refcount_release(&mapping->refs)) { 2445 LIST_REMOVE(&mapping->endpoints[0], entry); 2446 PF_HASHROW_UNLOCK(h0); 2447 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 2448 PF_HASHROW_LOCK(h1); 2449 LIST_REMOVE(&mapping->endpoints[1], entry); 2450 PF_HASHROW_UNLOCK(h1); 2451 2452 uma_zfree(V_pf_udp_mapping_z, mapping); 2453 } else { 2454 PF_HASHROW_UNLOCK(h0); 2455 } 2456 } 2457 2458 2459 struct pf_udp_mapping * 2460 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 2461 { 2462 struct pf_udpendpointhash *uh; 2463 struct pf_udp_endpoint *endpoint; 2464 2465 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 2466 2467 PF_HASHROW_LOCK(uh); 2468 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 2469 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 2470 bcmp(endpoint, &endpoint->mapping->endpoints[0], 2471 sizeof(struct pf_udp_endpoint_cmp)) == 0) 2472 break; 2473 } 2474 if (endpoint == NULL) { 2475 PF_HASHROW_UNLOCK(uh); 2476 return (NULL); 2477 } 2478 refcount_acquire(&endpoint->mapping->refs); 2479 PF_HASHROW_UNLOCK(uh); 2480 return (endpoint->mapping); 2481 } 2482 /* END state table stuff */ 2483 2484 static void 2485 pf_send(struct pf_send_entry *pfse) 2486 { 2487 2488 PF_SENDQ_LOCK(); 2489 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2490 PF_SENDQ_UNLOCK(); 2491 swi_sched(V_pf_swi_cookie, 0); 2492 } 2493 2494 static bool 2495 pf_isforlocal(struct mbuf *m, int af) 2496 { 2497 switch (af) { 2498 #ifdef INET 2499 case AF_INET: { 2500 struct ip *ip = mtod(m, struct ip *); 2501 2502 return (in_localip(ip->ip_dst)); 2503 } 2504 #endif /* INET */ 2505 #ifdef INET6 2506 case AF_INET6: { 2507 struct ip6_hdr *ip6; 2508 struct in6_ifaddr *ia; 2509 ip6 = mtod(m, struct ip6_hdr *); 2510 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2511 if (ia == NULL) 2512 return (false); 2513 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2514 } 2515 #endif /* INET6 */ 2516 default: 2517 unhandled_af(af); 2518 } 2519 2520 return (false); 2521 } 2522 2523 int 2524 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2525 int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type) 2526 { 2527 /* 2528 * ICMP types marked with PF_OUT are typically responses to 2529 * PF_IN, and will match states in the opposite direction. 2530 * PF_IN ICMP types need to match a state with that type. 2531 */ 2532 *icmp_dir = PF_OUT; 2533 2534 /* Queries (and responses) */ 2535 switch (pd->af) { 2536 #ifdef INET 2537 case AF_INET: 2538 switch (type) { 2539 case ICMP_ECHO: 2540 *icmp_dir = PF_IN; 2541 /* FALLTHROUGH */ 2542 case ICMP_ECHOREPLY: 2543 *virtual_type = ICMP_ECHO; 2544 *virtual_id = pd->hdr.icmp.icmp_id; 2545 break; 2546 2547 case ICMP_TSTAMP: 2548 *icmp_dir = PF_IN; 2549 /* FALLTHROUGH */ 2550 case ICMP_TSTAMPREPLY: 2551 *virtual_type = ICMP_TSTAMP; 2552 *virtual_id = pd->hdr.icmp.icmp_id; 2553 break; 2554 2555 case ICMP_IREQ: 2556 *icmp_dir = PF_IN; 2557 /* FALLTHROUGH */ 2558 case ICMP_IREQREPLY: 2559 *virtual_type = ICMP_IREQ; 2560 *virtual_id = pd->hdr.icmp.icmp_id; 2561 break; 2562 2563 case ICMP_MASKREQ: 2564 *icmp_dir = PF_IN; 2565 /* FALLTHROUGH */ 2566 case ICMP_MASKREPLY: 2567 *virtual_type = ICMP_MASKREQ; 2568 *virtual_id = pd->hdr.icmp.icmp_id; 2569 break; 2570 2571 case ICMP_IPV6_WHEREAREYOU: 2572 *icmp_dir = PF_IN; 2573 /* FALLTHROUGH */ 2574 case ICMP_IPV6_IAMHERE: 2575 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2576 *virtual_id = 0; /* Nothing sane to match on! */ 2577 break; 2578 2579 case ICMP_MOBILE_REGREQUEST: 2580 *icmp_dir = PF_IN; 2581 /* FALLTHROUGH */ 2582 case ICMP_MOBILE_REGREPLY: 2583 *virtual_type = ICMP_MOBILE_REGREQUEST; 2584 *virtual_id = 0; /* Nothing sane to match on! */ 2585 break; 2586 2587 case ICMP_ROUTERSOLICIT: 2588 *icmp_dir = PF_IN; 2589 /* FALLTHROUGH */ 2590 case ICMP_ROUTERADVERT: 2591 *virtual_type = ICMP_ROUTERSOLICIT; 2592 *virtual_id = 0; /* Nothing sane to match on! */ 2593 break; 2594 2595 /* These ICMP types map to other connections */ 2596 case ICMP_UNREACH: 2597 case ICMP_SOURCEQUENCH: 2598 case ICMP_REDIRECT: 2599 case ICMP_TIMXCEED: 2600 case ICMP_PARAMPROB: 2601 /* These will not be used, but set them anyway */ 2602 *icmp_dir = PF_IN; 2603 *virtual_type = type; 2604 *virtual_id = 0; 2605 *virtual_type = htons(*virtual_type); 2606 return (1); /* These types match to another state */ 2607 2608 /* 2609 * All remaining ICMP types get their own states, 2610 * and will only match in one direction. 2611 */ 2612 default: 2613 *icmp_dir = PF_IN; 2614 *virtual_type = type; 2615 *virtual_id = 0; 2616 break; 2617 } 2618 break; 2619 #endif /* INET */ 2620 #ifdef INET6 2621 case AF_INET6: 2622 switch (type) { 2623 case ICMP6_ECHO_REQUEST: 2624 *icmp_dir = PF_IN; 2625 /* FALLTHROUGH */ 2626 case ICMP6_ECHO_REPLY: 2627 *virtual_type = ICMP6_ECHO_REQUEST; 2628 *virtual_id = pd->hdr.icmp6.icmp6_id; 2629 break; 2630 2631 case MLD_LISTENER_QUERY: 2632 case MLD_LISTENER_REPORT: { 2633 /* 2634 * Listener Report can be sent by clients 2635 * without an associated Listener Query. 2636 * In addition to that, when Report is sent as a 2637 * reply to a Query its source and destination 2638 * address are different. 2639 */ 2640 *icmp_dir = PF_IN; 2641 *virtual_type = MLD_LISTENER_QUERY; 2642 *virtual_id = 0; 2643 break; 2644 } 2645 case MLD_MTRACE: 2646 *icmp_dir = PF_IN; 2647 /* FALLTHROUGH */ 2648 case MLD_MTRACE_RESP: 2649 *virtual_type = MLD_MTRACE; 2650 *virtual_id = 0; /* Nothing sane to match on! */ 2651 break; 2652 2653 case ND_NEIGHBOR_SOLICIT: 2654 *icmp_dir = PF_IN; 2655 /* FALLTHROUGH */ 2656 case ND_NEIGHBOR_ADVERT: { 2657 *virtual_type = ND_NEIGHBOR_SOLICIT; 2658 *virtual_id = 0; 2659 break; 2660 } 2661 2662 /* 2663 * These ICMP types map to other connections. 2664 * ND_REDIRECT can't be in this list because the triggering 2665 * packet header is optional. 2666 */ 2667 case ICMP6_DST_UNREACH: 2668 case ICMP6_PACKET_TOO_BIG: 2669 case ICMP6_TIME_EXCEEDED: 2670 case ICMP6_PARAM_PROB: 2671 /* These will not be used, but set them anyway */ 2672 *icmp_dir = PF_IN; 2673 *virtual_type = type; 2674 *virtual_id = 0; 2675 *virtual_type = htons(*virtual_type); 2676 return (1); /* These types match to another state */ 2677 /* 2678 * All remaining ICMP6 types get their own states, 2679 * and will only match in one direction. 2680 */ 2681 default: 2682 *icmp_dir = PF_IN; 2683 *virtual_type = type; 2684 *virtual_id = 0; 2685 break; 2686 } 2687 break; 2688 #endif /* INET6 */ 2689 default: 2690 unhandled_af(pd->af); 2691 } 2692 *virtual_type = htons(*virtual_type); 2693 return (0); /* These types match to their own state */ 2694 } 2695 2696 void 2697 pf_intr(void *v) 2698 { 2699 struct epoch_tracker et; 2700 struct pf_send_head queue; 2701 struct pf_send_entry *pfse, *next; 2702 2703 CURVNET_SET((struct vnet *)v); 2704 2705 PF_SENDQ_LOCK(); 2706 queue = V_pf_sendqueue; 2707 STAILQ_INIT(&V_pf_sendqueue); 2708 PF_SENDQ_UNLOCK(); 2709 2710 NET_EPOCH_ENTER(et); 2711 2712 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2713 switch (pfse->pfse_type) { 2714 #ifdef INET 2715 case PFSE_IP: { 2716 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2717 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2718 ("%s: rcvif != loif", __func__)); 2719 2720 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2721 pfse->pfse_m->m_pkthdr.csum_flags |= 2722 CSUM_IP_VALID | CSUM_IP_CHECKED | 2723 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2724 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2725 ip_input(pfse->pfse_m); 2726 } else { 2727 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2728 NULL); 2729 } 2730 break; 2731 } 2732 case PFSE_ICMP: 2733 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2734 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2735 break; 2736 #endif /* INET */ 2737 #ifdef INET6 2738 case PFSE_IP6: 2739 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2740 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2741 ("%s: rcvif != loif", __func__)); 2742 2743 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2744 M_LOOP; 2745 pfse->pfse_m->m_pkthdr.csum_flags |= 2746 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2747 pfse->pfse_m->m_pkthdr.csum_data = 0xffff; 2748 ip6_input(pfse->pfse_m); 2749 } else { 2750 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2751 NULL, NULL); 2752 } 2753 break; 2754 case PFSE_ICMP6: 2755 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2756 pfse->icmpopts.code, pfse->icmpopts.mtu); 2757 break; 2758 #endif /* INET6 */ 2759 default: 2760 panic("%s: unknown type", __func__); 2761 } 2762 free(pfse, M_PFTEMP); 2763 } 2764 NET_EPOCH_EXIT(et); 2765 CURVNET_RESTORE(); 2766 } 2767 2768 #define pf_purge_thread_period (hz / 10) 2769 2770 #ifdef PF_WANT_32_TO_64_COUNTER 2771 static void 2772 pf_status_counter_u64_periodic(void) 2773 { 2774 2775 PF_RULES_RASSERT(); 2776 2777 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2778 return; 2779 } 2780 2781 for (int i = 0; i < FCNT_MAX; i++) { 2782 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2783 } 2784 } 2785 2786 static void 2787 pf_kif_counter_u64_periodic(void) 2788 { 2789 struct pfi_kkif *kif; 2790 size_t r, run; 2791 2792 PF_RULES_RASSERT(); 2793 2794 if (__predict_false(V_pf_allkifcount == 0)) { 2795 return; 2796 } 2797 2798 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2799 return; 2800 } 2801 2802 run = V_pf_allkifcount / 10; 2803 if (run < 5) 2804 run = 5; 2805 2806 for (r = 0; r < run; r++) { 2807 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2808 if (kif == NULL) { 2809 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2810 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2811 break; 2812 } 2813 2814 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2815 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2816 2817 for (int i = 0; i < 2; i++) { 2818 for (int j = 0; j < 2; j++) { 2819 for (int k = 0; k < 2; k++) { 2820 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2821 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2822 } 2823 } 2824 } 2825 } 2826 } 2827 2828 static void 2829 pf_rule_counter_u64_periodic(void) 2830 { 2831 struct pf_krule *rule; 2832 size_t r, run; 2833 2834 PF_RULES_RASSERT(); 2835 2836 if (__predict_false(V_pf_allrulecount == 0)) { 2837 return; 2838 } 2839 2840 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2841 return; 2842 } 2843 2844 run = V_pf_allrulecount / 10; 2845 if (run < 5) 2846 run = 5; 2847 2848 for (r = 0; r < run; r++) { 2849 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2850 if (rule == NULL) { 2851 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2852 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2853 break; 2854 } 2855 2856 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2857 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2858 2859 pf_counter_u64_periodic(&rule->evaluations); 2860 for (int i = 0; i < 2; i++) { 2861 pf_counter_u64_periodic(&rule->packets[i]); 2862 pf_counter_u64_periodic(&rule->bytes[i]); 2863 } 2864 } 2865 } 2866 2867 static void 2868 pf_counter_u64_periodic_main(void) 2869 { 2870 PF_RULES_RLOCK_TRACKER; 2871 2872 V_pf_counter_periodic_iter++; 2873 2874 PF_RULES_RLOCK(); 2875 pf_counter_u64_critical_enter(); 2876 pf_status_counter_u64_periodic(); 2877 pf_kif_counter_u64_periodic(); 2878 pf_rule_counter_u64_periodic(); 2879 pf_counter_u64_critical_exit(); 2880 PF_RULES_RUNLOCK(); 2881 } 2882 #else 2883 #define pf_counter_u64_periodic_main() do { } while (0) 2884 #endif 2885 2886 void 2887 pf_purge_thread(void *unused __unused) 2888 { 2889 struct epoch_tracker et; 2890 2891 VNET_ITERATOR_DECL(vnet_iter); 2892 2893 sx_xlock(&pf_end_lock); 2894 while (pf_end_threads == 0) { 2895 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2896 2897 VNET_LIST_RLOCK(); 2898 NET_EPOCH_ENTER(et); 2899 VNET_FOREACH(vnet_iter) { 2900 CURVNET_SET(vnet_iter); 2901 2902 /* Wait until V_pf_default_rule is initialized. */ 2903 if (V_pf_vnet_active == 0) { 2904 CURVNET_RESTORE(); 2905 continue; 2906 } 2907 2908 pf_counter_u64_periodic_main(); 2909 2910 /* 2911 * Process 1/interval fraction of the state 2912 * table every run. 2913 */ 2914 V_pf_purge_idx = 2915 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2916 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2917 2918 /* 2919 * Purge other expired types every 2920 * PFTM_INTERVAL seconds. 2921 */ 2922 if (V_pf_purge_idx == 0) { 2923 /* 2924 * Order is important: 2925 * - states and src nodes reference rules 2926 * - states and rules reference kifs 2927 */ 2928 pf_purge_expired_fragments(); 2929 pf_purge_expired_src_nodes(); 2930 pf_purge_unlinked_rules(); 2931 pf_source_purge(); 2932 pfi_kkif_purge(); 2933 } 2934 CURVNET_RESTORE(); 2935 } 2936 NET_EPOCH_EXIT(et); 2937 VNET_LIST_RUNLOCK(); 2938 } 2939 2940 pf_end_threads++; 2941 sx_xunlock(&pf_end_lock); 2942 kproc_exit(0); 2943 } 2944 2945 void 2946 pf_unload_vnet_purge(void) 2947 { 2948 2949 /* 2950 * To cleanse up all kifs and rules we need 2951 * two runs: first one clears reference flags, 2952 * then pf_purge_expired_states() doesn't 2953 * raise them, and then second run frees. 2954 */ 2955 pf_purge_unlinked_rules(); 2956 pfi_kkif_purge(); 2957 2958 /* 2959 * Now purge everything. 2960 */ 2961 pf_purge_expired_states(0, V_pf_hashmask); 2962 pf_purge_fragments(UINT_MAX); 2963 pf_purge_expired_src_nodes(); 2964 pf_source_purge(); 2965 2966 /* 2967 * Now all kifs & rules should be unreferenced, 2968 * thus should be successfully freed. 2969 */ 2970 pf_purge_unlinked_rules(); 2971 pfi_kkif_purge(); 2972 } 2973 2974 u_int32_t 2975 pf_state_expires(const struct pf_kstate *state) 2976 { 2977 u_int32_t timeout; 2978 u_int32_t start; 2979 u_int32_t end; 2980 u_int32_t states; 2981 2982 /* handle all PFTM_* > PFTM_MAX here */ 2983 if (state->timeout == PFTM_PURGE) 2984 return (time_uptime); 2985 KASSERT(state->timeout != PFTM_UNLINKED, 2986 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2987 KASSERT((state->timeout < PFTM_MAX), 2988 ("pf_state_expires: timeout > PFTM_MAX")); 2989 timeout = state->rule->timeout[state->timeout]; 2990 if (!timeout) 2991 timeout = V_pf_default_rule.timeout[state->timeout]; 2992 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2993 if (start && state->rule != &V_pf_default_rule) { 2994 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2995 states = counter_u64_fetch(state->rule->states_cur); 2996 } else { 2997 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2998 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2999 states = V_pf_status.states; 3000 } 3001 if (end && states > start && start < end) { 3002 if (states < end) { 3003 timeout = (u_int64_t)timeout * (end - states) / 3004 (end - start); 3005 return ((state->expire / 1000) + timeout); 3006 } 3007 else 3008 return (time_uptime); 3009 } 3010 return ((state->expire / 1000) + timeout); 3011 } 3012 3013 void 3014 pf_purge_expired_src_nodes(void) 3015 { 3016 struct pf_ksrc_node_list freelist; 3017 struct pf_srchash *sh; 3018 struct pf_ksrc_node *cur, *next; 3019 int i; 3020 3021 LIST_INIT(&freelist); 3022 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 3023 PF_HASHROW_LOCK(sh); 3024 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 3025 if (cur->states == 0 && cur->expire <= time_uptime) { 3026 pf_unlink_src_node(cur); 3027 LIST_INSERT_HEAD(&freelist, cur, entry); 3028 } else if (cur->rule != NULL) 3029 cur->rule->rule_ref |= PFRULE_REFS; 3030 PF_HASHROW_UNLOCK(sh); 3031 } 3032 3033 pf_free_src_nodes(&freelist); 3034 3035 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 3036 } 3037 3038 static void 3039 pf_src_tree_remove_state(struct pf_kstate *s) 3040 { 3041 uint32_t timeout; 3042 3043 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 3044 s->rule->timeout[PFTM_SRC_NODE] : 3045 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 3046 3047 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 3048 if (s->sns[sn_type] == NULL) 3049 continue; 3050 PF_SRC_NODE_LOCK(s->sns[sn_type]); 3051 if (sn_type == PF_SN_LIMIT && s->src.tcp_est) 3052 --(s->sns[sn_type]->conn); 3053 if (--(s->sns[sn_type]->states) == 0) 3054 s->sns[sn_type]->expire = time_uptime + timeout; 3055 PF_SRC_NODE_UNLOCK(s->sns[sn_type]); 3056 s->sns[sn_type] = NULL; 3057 } 3058 3059 } 3060 3061 /* 3062 * Unlink and potentilly free a state. Function may be 3063 * called with ID hash row locked, but always returns 3064 * unlocked, since it needs to go through key hash locking. 3065 */ 3066 int 3067 pf_remove_state(struct pf_kstate *s) 3068 { 3069 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 3070 struct pf_state_link *pfl; 3071 3072 NET_EPOCH_ASSERT(); 3073 PF_HASHROW_ASSERT(ih); 3074 3075 if (s->timeout == PFTM_UNLINKED) { 3076 /* 3077 * State is being processed 3078 * by pf_remove_state() in 3079 * an other thread. 3080 */ 3081 PF_HASHROW_UNLOCK(ih); 3082 return (0); /* XXXGL: undefined actually */ 3083 } 3084 3085 if (s->src.state == PF_TCPS_PROXY_DST) { 3086 /* XXX wire key the right one? */ 3087 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 3088 &s->key[PF_SK_WIRE]->addr[1], 3089 &s->key[PF_SK_WIRE]->addr[0], 3090 s->key[PF_SK_WIRE]->port[1], 3091 s->key[PF_SK_WIRE]->port[0], 3092 s->src.seqhi, s->src.seqlo + 1, 3093 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 3094 s->act.rtableid, NULL); 3095 } 3096 3097 LIST_REMOVE(s, entry); 3098 pf_src_tree_remove_state(s); 3099 3100 if (V_pfsync_delete_state_ptr != NULL) 3101 V_pfsync_delete_state_ptr(s); 3102 3103 STATE_DEC_COUNTERS(s); 3104 3105 s->timeout = PFTM_UNLINKED; 3106 3107 /* Ensure we remove it from the list of halfopen states, if needed. */ 3108 if (s->key[PF_SK_STACK] != NULL && 3109 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 3110 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 3111 3112 while ((pfl = SLIST_FIRST(&s->linkage)) != NULL) { 3113 struct pf_state_link_list *list; 3114 unsigned int gen; 3115 3116 SLIST_REMOVE_HEAD(&s->linkage, pfl_linkage); 3117 3118 switch (pfl->pfl_type) { 3119 case PF_STATE_LINK_TYPE_STATELIM: { 3120 struct pf_statelim *stlim; 3121 3122 stlim = pf_statelim_find(s->statelim); 3123 KASSERT(stlim != NULL, 3124 ("pf_state %p pfl %p cannot find statelim %u", s, 3125 pfl, s->statelim)); 3126 3127 gen = pf_statelim_enter(stlim); 3128 stlim->pfstlim_inuse--; 3129 pf_statelim_leave(stlim, gen); 3130 3131 list = &stlim->pfstlim_states; 3132 break; 3133 } 3134 case PF_STATE_LINK_TYPE_SOURCELIM: { 3135 struct pf_sourcelim *srlim; 3136 struct pf_source key, *sr; 3137 3138 srlim = pf_sourcelim_find(s->sourcelim); 3139 KASSERT(srlim != NULL, 3140 ("pf_state %p pfl %p cannot find sourcelim %u", s, 3141 pfl, s->sourcelim)); 3142 3143 pf_source_key(srlim, &key, s->key[PF_SK_WIRE]->af, 3144 &s->key[PF_SK_WIRE]->addr[0 /* XXX or 1? */]); 3145 3146 sr = pf_source_find(srlim, &key); 3147 KASSERT(sr != NULL, 3148 ("pf_state %p pfl %p cannot find source in %u", s, 3149 pfl, s->sourcelim)); 3150 3151 gen = pf_sourcelim_enter(srlim); 3152 srlim->pfsrlim_counters.inuse--; 3153 pf_sourcelim_leave(srlim, gen); 3154 pf_source_rele(sr); 3155 3156 list = &sr->pfsr_states; 3157 break; 3158 } 3159 default: 3160 panic("%s: unexpected link type on pfl %p", __func__, 3161 pfl); 3162 } 3163 3164 PF_STATE_LOCK_ASSERT(s); 3165 TAILQ_REMOVE(list, pfl, pfl_link); 3166 free(pfl, M_PF_STATE_LINK); 3167 } 3168 3169 PF_HASHROW_UNLOCK(ih); 3170 3171 pf_detach_state(s); 3172 3173 pf_udp_mapping_release(s->udp_mapping); 3174 3175 /* pf_state_insert() initialises refs to 2 */ 3176 return (pf_release_staten(s, 2)); 3177 } 3178 3179 struct pf_kstate * 3180 pf_alloc_state(int flags) 3181 { 3182 3183 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 3184 } 3185 3186 static __inline void 3187 pf_free_match_rules(struct pf_krule_slist *match_rules) { 3188 struct pf_krule_item *ri; 3189 3190 while ((ri = SLIST_FIRST(match_rules))) { 3191 SLIST_REMOVE_HEAD(match_rules, entry); 3192 free(ri, M_PF_RULE_ITEM); 3193 } 3194 } 3195 3196 void 3197 pf_free_state(struct pf_kstate *cur) 3198 { 3199 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 3200 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 3201 cur->timeout)); 3202 3203 pf_free_match_rules(&(cur->match_rules)); 3204 pf_normalize_tcp_cleanup(cur); 3205 uma_zfree(V_pf_state_z, cur); 3206 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 3207 } 3208 3209 /* 3210 * Called only from pf_purge_thread(), thus serialized. 3211 */ 3212 static u_int 3213 pf_purge_expired_states(u_int i, int maxcheck) 3214 { 3215 struct pf_idhash *ih; 3216 struct pf_kstate *s; 3217 struct pf_krule_item *mrm; 3218 size_t count __unused; 3219 3220 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 3221 3222 /* 3223 * Go through hash and unlink states that expire now. 3224 */ 3225 while (maxcheck > 0) { 3226 count = 0; 3227 ih = &V_pf_idhash[i]; 3228 3229 /* only take the lock if we expect to do work */ 3230 if (!LIST_EMPTY(&ih->states)) { 3231 relock: 3232 PF_HASHROW_LOCK(ih); 3233 LIST_FOREACH(s, &ih->states, entry) { 3234 if (pf_state_expires(s) <= time_uptime) { 3235 V_pf_status.states -= 3236 pf_remove_state(s); 3237 goto relock; 3238 } 3239 s->rule->rule_ref |= PFRULE_REFS; 3240 if (s->nat_rule != NULL) 3241 s->nat_rule->rule_ref |= PFRULE_REFS; 3242 if (s->anchor != NULL) 3243 s->anchor->rule_ref |= PFRULE_REFS; 3244 s->kif->pfik_flags |= PFI_IFLAG_REFS; 3245 SLIST_FOREACH(mrm, &s->match_rules, entry) 3246 mrm->r->rule_ref |= PFRULE_REFS; 3247 if (s->act.rt_kif) 3248 s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS; 3249 count++; 3250 } 3251 PF_HASHROW_UNLOCK(ih); 3252 } 3253 3254 SDT_PROBE2(pf, purge, state, rowcount, i, count); 3255 3256 /* Return when we hit end of hash. */ 3257 if (++i > V_pf_hashmask) { 3258 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 3259 return (0); 3260 } 3261 3262 maxcheck--; 3263 } 3264 3265 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 3266 3267 return (i); 3268 } 3269 3270 static void 3271 pf_purge_unlinked_rules(void) 3272 { 3273 struct pf_krulequeue tmpq; 3274 struct pf_krule *r, *r1; 3275 3276 /* 3277 * If we have overloading task pending, then we'd 3278 * better skip purging this time. There is a tiny 3279 * probability that overloading task references 3280 * an already unlinked rule. 3281 */ 3282 PF_OVERLOADQ_LOCK(); 3283 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 3284 PF_OVERLOADQ_UNLOCK(); 3285 return; 3286 } 3287 PF_OVERLOADQ_UNLOCK(); 3288 3289 /* 3290 * Do naive mark-and-sweep garbage collecting of old rules. 3291 * Reference flag is raised by pf_purge_expired_states() 3292 * and pf_purge_expired_src_nodes(). 3293 * 3294 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 3295 * use a temporary queue. 3296 */ 3297 TAILQ_INIT(&tmpq); 3298 PF_UNLNKDRULES_LOCK(); 3299 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 3300 if (!(r->rule_ref & PFRULE_REFS)) { 3301 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 3302 TAILQ_INSERT_TAIL(&tmpq, r, entries); 3303 } else 3304 r->rule_ref &= ~PFRULE_REFS; 3305 } 3306 PF_UNLNKDRULES_UNLOCK(); 3307 3308 if (!TAILQ_EMPTY(&tmpq)) { 3309 PF_CONFIG_LOCK(); 3310 PF_RULES_WLOCK(); 3311 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 3312 TAILQ_REMOVE(&tmpq, r, entries); 3313 pf_free_rule(r); 3314 } 3315 PF_RULES_WUNLOCK(); 3316 PF_CONFIG_UNLOCK(); 3317 } 3318 } 3319 3320 void 3321 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 3322 { 3323 switch (af) { 3324 #ifdef INET 3325 case AF_INET: { 3326 u_int32_t a = ntohl(addr->addr32[0]); 3327 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 3328 (a>>8)&255, a&255); 3329 if (p) { 3330 p = ntohs(p); 3331 printf(":%u", p); 3332 } 3333 break; 3334 } 3335 #endif /* INET */ 3336 #ifdef INET6 3337 case AF_INET6: { 3338 u_int16_t b; 3339 u_int8_t i, curstart, curend, maxstart, maxend; 3340 curstart = curend = maxstart = maxend = 255; 3341 for (i = 0; i < 8; i++) { 3342 if (!addr->addr16[i]) { 3343 if (curstart == 255) 3344 curstart = i; 3345 curend = i; 3346 } else { 3347 if ((curend - curstart) > 3348 (maxend - maxstart)) { 3349 maxstart = curstart; 3350 maxend = curend; 3351 } 3352 curstart = curend = 255; 3353 } 3354 } 3355 if ((curend - curstart) > 3356 (maxend - maxstart)) { 3357 maxstart = curstart; 3358 maxend = curend; 3359 } 3360 for (i = 0; i < 8; i++) { 3361 if (i >= maxstart && i <= maxend) { 3362 if (i == 0) 3363 printf(":"); 3364 if (i == maxend) 3365 printf(":"); 3366 } else { 3367 b = ntohs(addr->addr16[i]); 3368 printf("%x", b); 3369 if (i < 7) 3370 printf(":"); 3371 } 3372 } 3373 if (p) { 3374 p = ntohs(p); 3375 printf("[%u]", p); 3376 } 3377 break; 3378 } 3379 #endif /* INET6 */ 3380 default: 3381 unhandled_af(af); 3382 } 3383 } 3384 3385 void 3386 pf_print_state(struct pf_kstate *s) 3387 { 3388 pf_print_state_parts(s, NULL, NULL); 3389 } 3390 3391 static void 3392 pf_print_state_parts(struct pf_kstate *s, 3393 struct pf_state_key *skwp, struct pf_state_key *sksp) 3394 { 3395 struct pf_state_key *skw, *sks; 3396 u_int8_t proto, dir; 3397 3398 /* Do our best to fill these, but they're skipped if NULL */ 3399 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 3400 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 3401 proto = skw ? skw->proto : (sks ? sks->proto : 0); 3402 dir = s ? s->direction : 0; 3403 3404 switch (proto) { 3405 case IPPROTO_IPV4: 3406 printf("IPv4"); 3407 break; 3408 case IPPROTO_IPV6: 3409 printf("IPv6"); 3410 break; 3411 case IPPROTO_TCP: 3412 printf("TCP"); 3413 break; 3414 case IPPROTO_UDP: 3415 printf("UDP"); 3416 break; 3417 case IPPROTO_ICMP: 3418 printf("ICMP"); 3419 break; 3420 case IPPROTO_ICMPV6: 3421 printf("ICMPv6"); 3422 break; 3423 default: 3424 printf("%u", proto); 3425 break; 3426 } 3427 switch (dir) { 3428 case PF_IN: 3429 printf(" in"); 3430 break; 3431 case PF_OUT: 3432 printf(" out"); 3433 break; 3434 } 3435 if (skw) { 3436 printf(" wire: "); 3437 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 3438 printf(" "); 3439 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 3440 } 3441 if (sks) { 3442 printf(" stack: "); 3443 if (sks != skw) { 3444 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 3445 printf(" "); 3446 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 3447 } else 3448 printf("-"); 3449 } 3450 if (s) { 3451 if (proto == IPPROTO_TCP) { 3452 printf(" [lo=%u high=%u win=%u modulator=%u", 3453 s->src.seqlo, s->src.seqhi, 3454 s->src.max_win, s->src.seqdiff); 3455 if (s->src.wscale && s->dst.wscale) 3456 printf(" wscale=%u", 3457 s->src.wscale & PF_WSCALE_MASK); 3458 printf("]"); 3459 printf(" [lo=%u high=%u win=%u modulator=%u", 3460 s->dst.seqlo, s->dst.seqhi, 3461 s->dst.max_win, s->dst.seqdiff); 3462 if (s->src.wscale && s->dst.wscale) 3463 printf(" wscale=%u", 3464 s->dst.wscale & PF_WSCALE_MASK); 3465 printf("]"); 3466 } 3467 printf(" %u:%u", s->src.state, s->dst.state); 3468 if (s->rule) 3469 printf(" @%d", s->rule->nr); 3470 } 3471 } 3472 3473 void 3474 pf_print_flags(uint16_t f) 3475 { 3476 if (f) 3477 printf(" "); 3478 if (f & TH_FIN) 3479 printf("F"); 3480 if (f & TH_SYN) 3481 printf("S"); 3482 if (f & TH_RST) 3483 printf("R"); 3484 if (f & TH_PUSH) 3485 printf("P"); 3486 if (f & TH_ACK) 3487 printf("A"); 3488 if (f & TH_URG) 3489 printf("U"); 3490 if (f & TH_ECE) 3491 printf("E"); 3492 if (f & TH_CWR) 3493 printf("W"); 3494 if (f & TH_AE) 3495 printf("e"); 3496 } 3497 3498 #define PF_SET_SKIP_STEPS(i) \ 3499 do { \ 3500 while (head[i] != cur) { \ 3501 head[i]->skip[i] = cur; \ 3502 head[i] = TAILQ_NEXT(head[i], entries); \ 3503 } \ 3504 } while (0) 3505 3506 void 3507 pf_calc_skip_steps(struct pf_krulequeue *rules) 3508 { 3509 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 3510 int i; 3511 3512 cur = TAILQ_FIRST(rules); 3513 prev = cur; 3514 for (i = 0; i < PF_SKIP_COUNT; ++i) 3515 head[i] = cur; 3516 while (cur != NULL) { 3517 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 3518 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 3519 if (cur->direction != prev->direction) 3520 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 3521 if (cur->af != prev->af) 3522 PF_SET_SKIP_STEPS(PF_SKIP_AF); 3523 if (cur->proto != prev->proto) 3524 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 3525 if (cur->src.neg != prev->src.neg || 3526 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 3527 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 3528 if (cur->dst.neg != prev->dst.neg || 3529 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 3530 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 3531 if (cur->src.port[0] != prev->src.port[0] || 3532 cur->src.port[1] != prev->src.port[1] || 3533 cur->src.port_op != prev->src.port_op) 3534 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 3535 if (cur->dst.port[0] != prev->dst.port[0] || 3536 cur->dst.port[1] != prev->dst.port[1] || 3537 cur->dst.port_op != prev->dst.port_op) 3538 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 3539 3540 prev = cur; 3541 cur = TAILQ_NEXT(cur, entries); 3542 } 3543 for (i = 0; i < PF_SKIP_COUNT; ++i) 3544 PF_SET_SKIP_STEPS(i); 3545 } 3546 3547 int 3548 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 3549 { 3550 if (aw1->type != aw2->type) 3551 return (1); 3552 switch (aw1->type) { 3553 case PF_ADDR_ADDRMASK: 3554 case PF_ADDR_RANGE: 3555 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 3556 return (1); 3557 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 3558 return (1); 3559 return (0); 3560 case PF_ADDR_DYNIFTL: 3561 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3562 case PF_ADDR_NONE: 3563 case PF_ADDR_NOROUTE: 3564 case PF_ADDR_URPFFAILED: 3565 return (0); 3566 case PF_ADDR_TABLE: 3567 return (aw1->p.tbl != aw2->p.tbl); 3568 default: 3569 printf("invalid address type: %d\n", aw1->type); 3570 return (1); 3571 } 3572 } 3573 3574 /** 3575 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3576 * header isn't always a full checksum. In some cases (i.e. output) it's a 3577 * pseudo-header checksum, which is a partial checksum over src/dst IP 3578 * addresses, protocol number and length. 3579 * 3580 * That means we have the following cases: 3581 * * Input or forwarding: we don't have TSO, the checksum fields are full 3582 * checksums, we need to update the checksum whenever we change anything. 3583 * * Output (i.e. the checksum is a pseudo-header checksum): 3584 * x The field being updated is src/dst address or affects the length of 3585 * the packet. We need to update the pseudo-header checksum (note that this 3586 * checksum is not ones' complement). 3587 * x Some other field is being modified (e.g. src/dst port numbers): We 3588 * don't have to update anything. 3589 **/ 3590 u_int16_t 3591 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3592 { 3593 u_int32_t x; 3594 3595 x = cksum + old - new; 3596 x = (x + (x >> 16)) & 0xffff; 3597 3598 /* optimise: eliminate a branch when not udp */ 3599 if (udp && cksum == 0x0000) 3600 return cksum; 3601 if (udp && x == 0x0000) 3602 x = 0xffff; 3603 3604 return (u_int16_t)(x); 3605 } 3606 3607 static int 3608 pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi) 3609 { 3610 int rewrite = 0; 3611 3612 if (*f != v) { 3613 uint16_t old = htons(hi ? (*f << 8) : *f); 3614 uint16_t new = htons(hi ? ( v << 8) : v); 3615 3616 *f = v; 3617 3618 if (! (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3619 CSUM_DELAY_DATA_IPV6))) 3620 *pd->pcksum = pf_cksum_fixup(*pd->pcksum, old, new, 3621 pd->proto == IPPROTO_UDP); 3622 3623 rewrite = 1; 3624 } 3625 3626 return (rewrite); 3627 } 3628 3629 int 3630 pf_patch_16(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi) 3631 { 3632 int rewrite = 0; 3633 u_int8_t *fb = (u_int8_t *)f; 3634 u_int8_t *vb = (u_int8_t *)&v; 3635 3636 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3637 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3638 3639 return (rewrite); 3640 } 3641 3642 int 3643 pf_patch_32(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi) 3644 { 3645 int rewrite = 0; 3646 u_int8_t *fb = (u_int8_t *)f; 3647 u_int8_t *vb = (u_int8_t *)&v; 3648 3649 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3650 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3651 rewrite += pf_patch_8(pd, fb++, *vb++, hi); 3652 rewrite += pf_patch_8(pd, fb++, *vb++, !hi); 3653 3654 return (rewrite); 3655 } 3656 3657 u_int16_t 3658 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3659 u_int16_t new, u_int8_t udp) 3660 { 3661 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3662 return (cksum); 3663 3664 return (pf_cksum_fixup(cksum, old, new, udp)); 3665 } 3666 3667 static void 3668 pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p, 3669 struct pf_addr *an, u_int16_t pn) 3670 { 3671 struct pf_addr ao; 3672 u_int16_t po; 3673 uint8_t u = pd->virtual_proto == IPPROTO_UDP; 3674 3675 MPASS(pd->pcksum != NULL); 3676 if (pd->af == AF_INET) { 3677 MPASS(pd->ip_sum); 3678 } 3679 3680 pf_addrcpy(&ao, a, pd->af); 3681 if (pd->af == pd->naf) 3682 pf_addrcpy(a, an, pd->af); 3683 3684 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3685 *pd->pcksum = ~*pd->pcksum; 3686 3687 if (p == NULL) /* no port -> done. no cksum to worry about. */ 3688 return; 3689 po = *p; 3690 *p = pn; 3691 3692 switch (pd->af) { 3693 #ifdef INET 3694 case AF_INET: 3695 switch (pd->naf) { 3696 case AF_INET: 3697 *pd->ip_sum = pf_cksum_fixup(pf_cksum_fixup(*pd->ip_sum, 3698 ao.addr16[0], an->addr16[0], 0), 3699 ao.addr16[1], an->addr16[1], 0); 3700 *p = pn; 3701 3702 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3703 ao.addr16[0], an->addr16[0], u), 3704 ao.addr16[1], an->addr16[1], u); 3705 3706 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3707 break; 3708 #ifdef INET6 3709 case AF_INET6: 3710 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3711 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3712 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3713 ao.addr16[0], an->addr16[0], u), 3714 ao.addr16[1], an->addr16[1], u), 3715 0, an->addr16[2], u), 3716 0, an->addr16[3], u), 3717 0, an->addr16[4], u), 3718 0, an->addr16[5], u), 3719 0, an->addr16[6], u), 3720 0, an->addr16[7], u), 3721 po, pn, u); 3722 break; 3723 #endif /* INET6 */ 3724 default: 3725 unhandled_af(pd->naf); 3726 } 3727 break; 3728 #endif /* INET */ 3729 #ifdef INET6 3730 case AF_INET6: 3731 switch (pd->naf) { 3732 #ifdef INET 3733 case AF_INET: 3734 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3735 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3736 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3737 ao.addr16[0], an->addr16[0], u), 3738 ao.addr16[1], an->addr16[1], u), 3739 ao.addr16[2], 0, u), 3740 ao.addr16[3], 0, u), 3741 ao.addr16[4], 0, u), 3742 ao.addr16[5], 0, u), 3743 ao.addr16[6], 0, u), 3744 ao.addr16[7], 0, u), 3745 po, pn, u); 3746 break; 3747 #endif /* INET */ 3748 case AF_INET6: 3749 *pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3750 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3751 pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum, 3752 ao.addr16[0], an->addr16[0], u), 3753 ao.addr16[1], an->addr16[1], u), 3754 ao.addr16[2], an->addr16[2], u), 3755 ao.addr16[3], an->addr16[3], u), 3756 ao.addr16[4], an->addr16[4], u), 3757 ao.addr16[5], an->addr16[5], u), 3758 ao.addr16[6], an->addr16[6], u), 3759 ao.addr16[7], an->addr16[7], u); 3760 3761 *pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u); 3762 break; 3763 default: 3764 unhandled_af(pd->naf); 3765 } 3766 break; 3767 #endif /* INET6 */ 3768 default: 3769 unhandled_af(pd->af); 3770 } 3771 3772 if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3773 CSUM_DELAY_DATA_IPV6)) { 3774 *pd->pcksum = ~*pd->pcksum; 3775 if (! *pd->pcksum) 3776 *pd->pcksum = 0xffff; 3777 } 3778 } 3779 3780 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3781 void 3782 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3783 { 3784 u_int32_t ao; 3785 3786 memcpy(&ao, a, sizeof(ao)); 3787 memcpy(a, &an, sizeof(u_int32_t)); 3788 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3789 ao % 65536, an % 65536, u); 3790 } 3791 3792 void 3793 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3794 { 3795 u_int32_t ao; 3796 3797 memcpy(&ao, a, sizeof(ao)); 3798 memcpy(a, &an, sizeof(u_int32_t)); 3799 3800 *c = pf_proto_cksum_fixup(m, 3801 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3802 ao % 65536, an % 65536, udp); 3803 } 3804 3805 #ifdef INET6 3806 static void 3807 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3808 { 3809 struct pf_addr ao; 3810 3811 pf_addrcpy(&ao, a, AF_INET6); 3812 pf_addrcpy(a, an, AF_INET6); 3813 3814 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3815 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3816 pf_cksum_fixup(pf_cksum_fixup(*c, 3817 ao.addr16[0], an->addr16[0], u), 3818 ao.addr16[1], an->addr16[1], u), 3819 ao.addr16[2], an->addr16[2], u), 3820 ao.addr16[3], an->addr16[3], u), 3821 ao.addr16[4], an->addr16[4], u), 3822 ao.addr16[5], an->addr16[5], u), 3823 ao.addr16[6], an->addr16[6], u), 3824 ao.addr16[7], an->addr16[7], u); 3825 } 3826 #endif /* INET6 */ 3827 3828 static void 3829 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3830 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3831 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3832 { 3833 struct pf_addr oia, ooa; 3834 3835 pf_addrcpy(&oia, ia, af); 3836 if (oa) 3837 pf_addrcpy(&ooa, oa, af); 3838 3839 /* Change inner protocol port, fix inner protocol checksum. */ 3840 if (ip != NULL) { 3841 u_int16_t oip = *ip; 3842 u_int16_t opc; 3843 3844 if (pc != NULL) 3845 opc = *pc; 3846 *ip = np; 3847 if (pc != NULL) 3848 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3849 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3850 if (pc != NULL) 3851 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3852 } 3853 /* Change inner ip address, fix inner ip and icmp checksums. */ 3854 pf_addrcpy(ia, na, af); 3855 switch (af) { 3856 #ifdef INET 3857 case AF_INET: { 3858 u_int16_t oh2c = *h2c; 3859 3860 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3861 oia.addr16[0], ia->addr16[0], 0), 3862 oia.addr16[1], ia->addr16[1], 0); 3863 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3864 oia.addr16[0], ia->addr16[0], 0), 3865 oia.addr16[1], ia->addr16[1], 0); 3866 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3867 break; 3868 } 3869 #endif /* INET */ 3870 #ifdef INET6 3871 case AF_INET6: 3872 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3873 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3874 pf_cksum_fixup(pf_cksum_fixup(*ic, 3875 oia.addr16[0], ia->addr16[0], u), 3876 oia.addr16[1], ia->addr16[1], u), 3877 oia.addr16[2], ia->addr16[2], u), 3878 oia.addr16[3], ia->addr16[3], u), 3879 oia.addr16[4], ia->addr16[4], u), 3880 oia.addr16[5], ia->addr16[5], u), 3881 oia.addr16[6], ia->addr16[6], u), 3882 oia.addr16[7], ia->addr16[7], u); 3883 break; 3884 #endif /* INET6 */ 3885 } 3886 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3887 if (oa) { 3888 pf_addrcpy(oa, na, af); 3889 switch (af) { 3890 #ifdef INET 3891 case AF_INET: 3892 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3893 ooa.addr16[0], oa->addr16[0], 0), 3894 ooa.addr16[1], oa->addr16[1], 0); 3895 break; 3896 #endif /* INET */ 3897 #ifdef INET6 3898 case AF_INET6: 3899 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3900 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3901 pf_cksum_fixup(pf_cksum_fixup(*ic, 3902 ooa.addr16[0], oa->addr16[0], u), 3903 ooa.addr16[1], oa->addr16[1], u), 3904 ooa.addr16[2], oa->addr16[2], u), 3905 ooa.addr16[3], oa->addr16[3], u), 3906 ooa.addr16[4], oa->addr16[4], u), 3907 ooa.addr16[5], oa->addr16[5], u), 3908 ooa.addr16[6], oa->addr16[6], u), 3909 ooa.addr16[7], oa->addr16[7], u); 3910 break; 3911 #endif /* INET6 */ 3912 } 3913 } 3914 } 3915 3916 static int 3917 pf_translate_af(struct pf_pdesc *pd, struct pf_krule *r) 3918 { 3919 #if defined(INET) && defined(INET6) 3920 struct mbuf *mp; 3921 struct ip *ip4; 3922 struct ip6_hdr *ip6; 3923 struct icmp6_hdr *icmp; 3924 struct m_tag *mtag; 3925 struct pf_fragment_tag *ftag; 3926 int hlen; 3927 3928 if (pd->ttl == 1) { 3929 /* We'd generate an ICMP error. Do so now rather than after af translation. */ 3930 if (pd->af == AF_INET) { 3931 pf_send_icmp(pd->m, ICMP_TIMXCEED, 3932 ICMP_TIMXCEED_INTRANS, 0, pd->af, r, 3933 pd->act.rtableid); 3934 } else { 3935 pf_send_icmp(pd->m, ICMP6_TIME_EXCEEDED, 3936 ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r, 3937 pd->act.rtableid); 3938 } 3939 3940 return (-1); 3941 } 3942 3943 hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 3944 3945 /* trim the old header */ 3946 m_adj(pd->m, pd->off); 3947 3948 /* prepend a new one */ 3949 M_PREPEND(pd->m, hlen, M_NOWAIT); 3950 if (pd->m == NULL) 3951 return (-1); 3952 3953 switch (pd->naf) { 3954 case AF_INET: 3955 ip4 = mtod(pd->m, struct ip *); 3956 bzero(ip4, hlen); 3957 ip4->ip_v = IPVERSION; 3958 ip4->ip_hl = hlen >> 2; 3959 ip4->ip_tos = pd->tos; 3960 ip4->ip_len = htons(hlen + (pd->tot_len - pd->off)); 3961 ip_fillid(ip4, V_ip_random_id); 3962 ip4->ip_ttl = pd->ttl; 3963 ip4->ip_p = pd->proto; 3964 ip4->ip_src = pd->nsaddr.v4; 3965 ip4->ip_dst = pd->ndaddr.v4; 3966 pd->src = (struct pf_addr *)&ip4->ip_src; 3967 pd->dst = (struct pf_addr *)&ip4->ip_dst; 3968 pd->off = sizeof(struct ip); 3969 if (pd->m->m_pkthdr.csum_flags & CSUM_TCP_IPV6) { 3970 pd->m->m_pkthdr.csum_flags &= ~CSUM_TCP_IPV6; 3971 pd->m->m_pkthdr.csum_flags |= CSUM_TCP; 3972 } 3973 if (pd->m->m_pkthdr.csum_flags & CSUM_UDP_IPV6) { 3974 pd->m->m_pkthdr.csum_flags &= ~CSUM_UDP_IPV6; 3975 pd->m->m_pkthdr.csum_flags |= CSUM_UDP; 3976 } 3977 if (pd->m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) { 3978 pd->m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; 3979 pd->m->m_pkthdr.csum_flags |= CSUM_SCTP; 3980 } 3981 break; 3982 case AF_INET6: 3983 ip6 = mtod(pd->m, struct ip6_hdr *); 3984 bzero(ip6, hlen); 3985 ip6->ip6_vfc = IPV6_VERSION; 3986 ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20); 3987 ip6->ip6_plen = htons(pd->tot_len - pd->off); 3988 ip6->ip6_nxt = pd->proto; 3989 if (!pd->ttl || pd->ttl > IPV6_DEFHLIM) 3990 ip6->ip6_hlim = IPV6_DEFHLIM; 3991 else 3992 ip6->ip6_hlim = pd->ttl; 3993 ip6->ip6_src = pd->nsaddr.v6; 3994 ip6->ip6_dst = pd->ndaddr.v6; 3995 pd->src = (struct pf_addr *)&ip6->ip6_src; 3996 pd->dst = (struct pf_addr *)&ip6->ip6_dst; 3997 pd->off = sizeof(struct ip6_hdr); 3998 if (pd->m->m_pkthdr.csum_flags & CSUM_TCP) { 3999 pd->m->m_pkthdr.csum_flags &= ~CSUM_TCP; 4000 pd->m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4001 } 4002 if (pd->m->m_pkthdr.csum_flags & CSUM_UDP) { 4003 pd->m->m_pkthdr.csum_flags &= ~CSUM_UDP; 4004 pd->m->m_pkthdr.csum_flags |= CSUM_UDP_IPV6; 4005 } 4006 if (pd->m->m_pkthdr.csum_flags & CSUM_SCTP) { 4007 pd->m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 4008 pd->m->m_pkthdr.csum_flags |= CSUM_SCTP_IPV6; 4009 } 4010 4011 /* 4012 * If we're dealing with a reassembled packet we need to adjust 4013 * the header length from the IPv4 header size to IPv6 header 4014 * size. 4015 */ 4016 mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL); 4017 if (mtag) { 4018 ftag = (struct pf_fragment_tag *)(mtag + 1); 4019 ftag->ft_hdrlen = sizeof(*ip6); 4020 ftag->ft_maxlen -= sizeof(struct ip6_hdr) - 4021 sizeof(struct ip) + sizeof(struct ip6_frag); 4022 } 4023 break; 4024 default: 4025 return (-1); 4026 } 4027 4028 /* recalculate icmp/icmp6 checksums */ 4029 if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) { 4030 int off; 4031 if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) == 4032 NULL) { 4033 pd->m = NULL; 4034 return (-1); 4035 } 4036 icmp = (struct icmp6_hdr *)(mp->m_data + off); 4037 icmp->icmp6_cksum = 0; 4038 icmp->icmp6_cksum = pd->naf == AF_INET ? 4039 in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) : 4040 in6_cksum(pd->m, IPPROTO_ICMPV6, hlen, 4041 ntohs(ip6->ip6_plen)); 4042 } 4043 #endif /* INET && INET6 */ 4044 4045 return (0); 4046 } 4047 4048 int 4049 pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd, 4050 struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst, 4051 sa_family_t af, sa_family_t naf) 4052 { 4053 #if defined(INET) && defined(INET6) 4054 struct mbuf *n = NULL; 4055 struct ip *ip4; 4056 struct ip6_hdr *ip6; 4057 int hlen, olen, mlen; 4058 4059 if (af == naf || (af != AF_INET && af != AF_INET6) || 4060 (naf != AF_INET && naf != AF_INET6)) 4061 return (-1); 4062 4063 /* split the mbuf chain on the inner ip/ip6 header boundary */ 4064 if ((n = m_split(m, off, M_NOWAIT)) == NULL) 4065 return (-1); 4066 4067 /* old header */ 4068 olen = pd2->off - off; 4069 /* new header */ 4070 hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6); 4071 4072 /* trim old header */ 4073 m_adj(n, olen); 4074 4075 /* prepend a new one */ 4076 M_PREPEND(n, hlen, M_NOWAIT); 4077 if (n == NULL) 4078 return (-1); 4079 4080 /* translate inner ip/ip6 header */ 4081 switch (naf) { 4082 case AF_INET: 4083 ip4 = mtod(n, struct ip *); 4084 bzero(ip4, sizeof(*ip4)); 4085 ip4->ip_v = IPVERSION; 4086 ip4->ip_hl = sizeof(*ip4) >> 2; 4087 ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen); 4088 ip_fillid(ip4, V_ip_random_id); 4089 ip4->ip_off = htons(IP_DF); 4090 ip4->ip_ttl = pd2->ttl; 4091 if (pd2->proto == IPPROTO_ICMPV6) 4092 ip4->ip_p = IPPROTO_ICMP; 4093 else 4094 ip4->ip_p = pd2->proto; 4095 ip4->ip_src = src->v4; 4096 ip4->ip_dst = dst->v4; 4097 ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2); 4098 break; 4099 case AF_INET6: 4100 ip6 = mtod(n, struct ip6_hdr *); 4101 bzero(ip6, sizeof(*ip6)); 4102 ip6->ip6_vfc = IPV6_VERSION; 4103 ip6->ip6_plen = htons(pd2->tot_len - olen); 4104 if (pd2->proto == IPPROTO_ICMP) 4105 ip6->ip6_nxt = IPPROTO_ICMPV6; 4106 else 4107 ip6->ip6_nxt = pd2->proto; 4108 if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM) 4109 ip6->ip6_hlim = IPV6_DEFHLIM; 4110 else 4111 ip6->ip6_hlim = pd2->ttl; 4112 ip6->ip6_src = src->v6; 4113 ip6->ip6_dst = dst->v6; 4114 break; 4115 default: 4116 unhandled_af(naf); 4117 } 4118 4119 /* adjust payload offset and total packet length */ 4120 pd2->off += hlen - olen; 4121 pd->tot_len += hlen - olen; 4122 4123 /* merge modified inner packet with the original header */ 4124 mlen = n->m_pkthdr.len; 4125 m_cat(m, n); 4126 m->m_pkthdr.len += mlen; 4127 #endif /* INET && INET6 */ 4128 4129 return (0); 4130 } 4131 4132 #define PTR_IP(field) (offsetof(struct ip, field)) 4133 #define PTR_IP6(field) (offsetof(struct ip6_hdr, field)) 4134 4135 int 4136 pf_translate_icmp_af(int af, void *arg) 4137 { 4138 #if defined(INET) && defined(INET6) 4139 struct icmp *icmp4; 4140 struct icmp6_hdr *icmp6; 4141 u_int32_t mtu; 4142 int32_t ptr = -1; 4143 u_int8_t type; 4144 u_int8_t code; 4145 4146 switch (af) { 4147 case AF_INET: 4148 icmp6 = arg; 4149 type = icmp6->icmp6_type; 4150 code = icmp6->icmp6_code; 4151 mtu = ntohl(icmp6->icmp6_mtu); 4152 4153 switch (type) { 4154 case ICMP6_ECHO_REQUEST: 4155 type = ICMP_ECHO; 4156 break; 4157 case ICMP6_ECHO_REPLY: 4158 type = ICMP_ECHOREPLY; 4159 break; 4160 case ICMP6_DST_UNREACH: 4161 type = ICMP_UNREACH; 4162 switch (code) { 4163 case ICMP6_DST_UNREACH_NOROUTE: 4164 case ICMP6_DST_UNREACH_BEYONDSCOPE: 4165 case ICMP6_DST_UNREACH_ADDR: 4166 code = ICMP_UNREACH_HOST; 4167 break; 4168 case ICMP6_DST_UNREACH_ADMIN: 4169 code = ICMP_UNREACH_HOST_PROHIB; 4170 break; 4171 case ICMP6_DST_UNREACH_NOPORT: 4172 code = ICMP_UNREACH_PORT; 4173 break; 4174 default: 4175 return (-1); 4176 } 4177 break; 4178 case ICMP6_PACKET_TOO_BIG: 4179 type = ICMP_UNREACH; 4180 code = ICMP_UNREACH_NEEDFRAG; 4181 mtu -= 20; 4182 break; 4183 case ICMP6_TIME_EXCEEDED: 4184 type = ICMP_TIMXCEED; 4185 break; 4186 case ICMP6_PARAM_PROB: 4187 switch (code) { 4188 case ICMP6_PARAMPROB_HEADER: 4189 type = ICMP_PARAMPROB; 4190 code = ICMP_PARAMPROB_ERRATPTR; 4191 ptr = ntohl(icmp6->icmp6_pptr); 4192 4193 if (ptr == PTR_IP6(ip6_vfc)) 4194 ; /* preserve */ 4195 else if (ptr == PTR_IP6(ip6_vfc) + 1) 4196 ptr = PTR_IP(ip_tos); 4197 else if (ptr == PTR_IP6(ip6_plen) || 4198 ptr == PTR_IP6(ip6_plen) + 1) 4199 ptr = PTR_IP(ip_len); 4200 else if (ptr == PTR_IP6(ip6_nxt)) 4201 ptr = PTR_IP(ip_p); 4202 else if (ptr == PTR_IP6(ip6_hlim)) 4203 ptr = PTR_IP(ip_ttl); 4204 else if (ptr >= PTR_IP6(ip6_src) && 4205 ptr < PTR_IP6(ip6_dst)) 4206 ptr = PTR_IP(ip_src); 4207 else if (ptr >= PTR_IP6(ip6_dst) && 4208 ptr < sizeof(struct ip6_hdr)) 4209 ptr = PTR_IP(ip_dst); 4210 else { 4211 return (-1); 4212 } 4213 break; 4214 case ICMP6_PARAMPROB_NEXTHEADER: 4215 type = ICMP_UNREACH; 4216 code = ICMP_UNREACH_PROTOCOL; 4217 break; 4218 default: 4219 return (-1); 4220 } 4221 break; 4222 default: 4223 return (-1); 4224 } 4225 if (icmp6->icmp6_type != type) { 4226 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 4227 icmp6->icmp6_type, type, 0); 4228 icmp6->icmp6_type = type; 4229 } 4230 if (icmp6->icmp6_code != code) { 4231 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 4232 icmp6->icmp6_code, code, 0); 4233 icmp6->icmp6_code = code; 4234 } 4235 if (icmp6->icmp6_mtu != htonl(mtu)) { 4236 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 4237 htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0); 4238 /* aligns well with a icmpv4 nextmtu */ 4239 icmp6->icmp6_mtu = htonl(mtu); 4240 } 4241 if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) { 4242 icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum, 4243 htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0); 4244 /* icmpv4 pptr is a one most significant byte */ 4245 icmp6->icmp6_pptr = htonl(ptr << 24); 4246 } 4247 break; 4248 case AF_INET6: 4249 icmp4 = arg; 4250 type = icmp4->icmp_type; 4251 code = icmp4->icmp_code; 4252 mtu = ntohs(icmp4->icmp_nextmtu); 4253 4254 switch (type) { 4255 case ICMP_ECHO: 4256 type = ICMP6_ECHO_REQUEST; 4257 break; 4258 case ICMP_ECHOREPLY: 4259 type = ICMP6_ECHO_REPLY; 4260 break; 4261 case ICMP_UNREACH: 4262 type = ICMP6_DST_UNREACH; 4263 switch (code) { 4264 case ICMP_UNREACH_NET: 4265 case ICMP_UNREACH_HOST: 4266 case ICMP_UNREACH_NET_UNKNOWN: 4267 case ICMP_UNREACH_HOST_UNKNOWN: 4268 case ICMP_UNREACH_ISOLATED: 4269 case ICMP_UNREACH_TOSNET: 4270 case ICMP_UNREACH_TOSHOST: 4271 code = ICMP6_DST_UNREACH_NOROUTE; 4272 break; 4273 case ICMP_UNREACH_PORT: 4274 code = ICMP6_DST_UNREACH_NOPORT; 4275 break; 4276 case ICMP_UNREACH_NET_PROHIB: 4277 case ICMP_UNREACH_HOST_PROHIB: 4278 case ICMP_UNREACH_FILTER_PROHIB: 4279 case ICMP_UNREACH_PRECEDENCE_CUTOFF: 4280 code = ICMP6_DST_UNREACH_ADMIN; 4281 break; 4282 case ICMP_UNREACH_PROTOCOL: 4283 type = ICMP6_PARAM_PROB; 4284 code = ICMP6_PARAMPROB_NEXTHEADER; 4285 ptr = offsetof(struct ip6_hdr, ip6_nxt); 4286 break; 4287 case ICMP_UNREACH_NEEDFRAG: 4288 type = ICMP6_PACKET_TOO_BIG; 4289 code = 0; 4290 mtu += 20; 4291 break; 4292 default: 4293 return (-1); 4294 } 4295 break; 4296 case ICMP_TIMXCEED: 4297 type = ICMP6_TIME_EXCEEDED; 4298 break; 4299 case ICMP_PARAMPROB: 4300 type = ICMP6_PARAM_PROB; 4301 switch (code) { 4302 case ICMP_PARAMPROB_ERRATPTR: 4303 code = ICMP6_PARAMPROB_HEADER; 4304 break; 4305 case ICMP_PARAMPROB_LENGTH: 4306 code = ICMP6_PARAMPROB_HEADER; 4307 break; 4308 default: 4309 return (-1); 4310 } 4311 4312 ptr = icmp4->icmp_pptr; 4313 if (ptr == 0 || ptr == PTR_IP(ip_tos)) 4314 ; /* preserve */ 4315 else if (ptr == PTR_IP(ip_len) || 4316 ptr == PTR_IP(ip_len) + 1) 4317 ptr = PTR_IP6(ip6_plen); 4318 else if (ptr == PTR_IP(ip_ttl)) 4319 ptr = PTR_IP6(ip6_hlim); 4320 else if (ptr == PTR_IP(ip_p)) 4321 ptr = PTR_IP6(ip6_nxt); 4322 else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst)) 4323 ptr = PTR_IP6(ip6_src); 4324 else if (ptr >= PTR_IP(ip_dst) && 4325 ptr < sizeof(struct ip)) 4326 ptr = PTR_IP6(ip6_dst); 4327 else { 4328 return (-1); 4329 } 4330 break; 4331 default: 4332 return (-1); 4333 } 4334 if (icmp4->icmp_type != type) { 4335 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4336 icmp4->icmp_type, type, 0); 4337 icmp4->icmp_type = type; 4338 } 4339 if (icmp4->icmp_code != code) { 4340 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4341 icmp4->icmp_code, code, 0); 4342 icmp4->icmp_code = code; 4343 } 4344 if (icmp4->icmp_nextmtu != htons(mtu)) { 4345 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4346 icmp4->icmp_nextmtu, htons(mtu), 0); 4347 icmp4->icmp_nextmtu = htons(mtu); 4348 } 4349 if (ptr >= 0 && icmp4->icmp_void != ptr) { 4350 icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum, 4351 htons(icmp4->icmp_pptr), htons(ptr), 0); 4352 icmp4->icmp_void = htonl(ptr); 4353 } 4354 break; 4355 default: 4356 unhandled_af(af); 4357 } 4358 #endif /* INET && INET6 */ 4359 4360 return (0); 4361 } 4362 4363 /* 4364 * Need to modulate the sequence numbers in the TCP SACK option 4365 * (credits to Krzysztof Pfaff for report and patch) 4366 */ 4367 static int 4368 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 4369 struct pf_state_peer *dst) 4370 { 4371 struct sackblk sack; 4372 int copyback = 0, i; 4373 int olen, optsoff; 4374 uint8_t opts[MAX_TCPOPTLEN], *opt, *eoh; 4375 4376 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 4377 optsoff = pd->off + sizeof(struct tcphdr); 4378 #define TCPOLEN_MINSACK (TCPOLEN_SACK + 2) 4379 if (olen < TCPOLEN_MINSACK || 4380 !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, pd->af)) 4381 return (0); 4382 4383 eoh = opts + olen; 4384 opt = opts; 4385 while ((opt = pf_find_tcpopt(opt, opts, olen, 4386 TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL) 4387 { 4388 size_t safelen = MIN(opt[1], (eoh - opt)); 4389 for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) { 4390 size_t startoff = (opt + i) - opts; 4391 memcpy(&sack, &opt[i], sizeof(sack)); 4392 pf_patch_32(pd, &sack.start, 4393 htonl(ntohl(sack.start) - dst->seqdiff), 4394 PF_ALGNMNT(startoff)); 4395 pf_patch_32(pd, &sack.end, 4396 htonl(ntohl(sack.end) - dst->seqdiff), 4397 PF_ALGNMNT(startoff + sizeof(sack.start))); 4398 memcpy(&opt[i], &sack, sizeof(sack)); 4399 } 4400 copyback = 1; 4401 opt += opt[1]; 4402 } 4403 4404 if (copyback) 4405 m_copyback(pd->m, optsoff, olen, (caddr_t)opts); 4406 4407 return (copyback); 4408 } 4409 4410 struct mbuf * 4411 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 4412 const struct pf_addr *saddr, const struct pf_addr *daddr, 4413 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4414 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4415 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, u_int sack, 4416 int rtableid, u_short *reason) 4417 { 4418 struct mbuf *m; 4419 int len, tlen; 4420 #ifdef INET 4421 struct ip *h = NULL; 4422 #endif /* INET */ 4423 #ifdef INET6 4424 struct ip6_hdr *h6 = NULL; 4425 #endif /* INET6 */ 4426 struct tcphdr *th; 4427 char *opt; 4428 struct pf_mtag *pf_mtag; 4429 4430 len = 0; 4431 th = NULL; 4432 4433 /* maximum segment size tcp option */ 4434 tlen = sizeof(struct tcphdr); 4435 if (mss) 4436 tlen += 4; 4437 if (sack) 4438 tlen += 2; 4439 4440 switch (af) { 4441 #ifdef INET 4442 case AF_INET: 4443 len = sizeof(struct ip) + tlen; 4444 break; 4445 #endif /* INET */ 4446 #ifdef INET6 4447 case AF_INET6: 4448 len = sizeof(struct ip6_hdr) + tlen; 4449 break; 4450 #endif /* INET6 */ 4451 default: 4452 unhandled_af(af); 4453 } 4454 4455 m = m_gethdr(M_NOWAIT, MT_DATA); 4456 if (m == NULL) { 4457 REASON_SET(reason, PFRES_MEMORY); 4458 return (NULL); 4459 } 4460 4461 #ifdef MAC 4462 mac_netinet_firewall_send(m); 4463 #endif 4464 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 4465 REASON_SET(reason, PFRES_MEMORY); 4466 m_freem(m); 4467 return (NULL); 4468 } 4469 m->m_flags |= mbuf_flags; 4470 pf_mtag->tag = mtag_tag; 4471 pf_mtag->flags = mtag_flags; 4472 4473 if (rtableid >= 0) 4474 M_SETFIB(m, rtableid); 4475 4476 #ifdef ALTQ 4477 if (r != NULL && r->qid) { 4478 pf_mtag->qid = r->qid; 4479 4480 /* add hints for ecn */ 4481 pf_mtag->hdr = mtod(m, struct ip *); 4482 } 4483 #endif /* ALTQ */ 4484 m->m_data += max_linkhdr; 4485 m->m_pkthdr.len = m->m_len = len; 4486 /* The rest of the stack assumes a rcvif, so provide one. 4487 * This is a locally generated packet, so .. close enough. */ 4488 m->m_pkthdr.rcvif = V_loif; 4489 bzero(m->m_data, len); 4490 switch (af) { 4491 #ifdef INET 4492 case AF_INET: 4493 m->m_pkthdr.csum_flags |= CSUM_TCP; 4494 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4495 4496 h = mtod(m, struct ip *); 4497 4498 h->ip_p = IPPROTO_TCP; 4499 h->ip_len = htons(tlen); 4500 h->ip_v = 4; 4501 h->ip_hl = sizeof(*h) >> 2; 4502 h->ip_tos = IPTOS_LOWDELAY; 4503 h->ip_len = htons(len); 4504 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 4505 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4506 h->ip_sum = 0; 4507 h->ip_src.s_addr = saddr->v4.s_addr; 4508 h->ip_dst.s_addr = daddr->v4.s_addr; 4509 4510 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 4511 th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr, 4512 htons(len - sizeof(struct ip) + IPPROTO_TCP)); 4513 break; 4514 #endif /* INET */ 4515 #ifdef INET6 4516 case AF_INET6: 4517 m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6; 4518 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 4519 4520 h6 = mtod(m, struct ip6_hdr *); 4521 4522 /* IP header fields included in the TCP checksum */ 4523 h6->ip6_nxt = IPPROTO_TCP; 4524 h6->ip6_plen = htons(tlen); 4525 h6->ip6_vfc |= IPV6_VERSION; 4526 h6->ip6_hlim = V_ip6_defhlim; 4527 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 4528 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 4529 4530 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 4531 th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr), 4532 IPPROTO_TCP, 0); 4533 break; 4534 #endif /* INET6 */ 4535 } 4536 4537 /* TCP header */ 4538 th->th_sport = sport; 4539 th->th_dport = dport; 4540 th->th_seq = htonl(seq); 4541 th->th_ack = htonl(ack); 4542 th->th_off = tlen >> 2; 4543 tcp_set_flags(th, tcp_flags); 4544 th->th_win = htons(win); 4545 4546 opt = (char *)(th + 1); 4547 if (mss) { 4548 opt = (char *)(th + 1); 4549 opt[0] = TCPOPT_MAXSEG; 4550 opt[1] = 4; 4551 mss = htons(mss); 4552 memcpy((opt + 2), &mss, 2); 4553 opt += 4; 4554 } 4555 if (sack) { 4556 opt[0] = TCPOPT_SACK_PERMITTED; 4557 opt[1] = 2; 4558 opt += 2; 4559 } 4560 4561 return (m); 4562 } 4563 4564 static void 4565 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 4566 uint8_t ttl, int rtableid) 4567 { 4568 struct mbuf *m; 4569 #ifdef INET 4570 struct ip *h = NULL; 4571 #endif /* INET */ 4572 #ifdef INET6 4573 struct ip6_hdr *h6 = NULL; 4574 #endif /* INET6 */ 4575 struct sctphdr *hdr; 4576 struct sctp_chunkhdr *chunk; 4577 struct pf_send_entry *pfse; 4578 int off = 0; 4579 4580 MPASS(af == pd->af); 4581 4582 m = m_gethdr(M_NOWAIT, MT_DATA); 4583 if (m == NULL) 4584 return; 4585 4586 m->m_data += max_linkhdr; 4587 m->m_flags |= M_SKIP_FIREWALL; 4588 /* The rest of the stack assumes a rcvif, so provide one. 4589 * This is a locally generated packet, so .. close enough. */ 4590 m->m_pkthdr.rcvif = V_loif; 4591 4592 /* IPv4|6 header */ 4593 switch (af) { 4594 #ifdef INET 4595 case AF_INET: 4596 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 4597 4598 h = mtod(m, struct ip *); 4599 4600 /* IP header fields included in the TCP checksum */ 4601 4602 h->ip_p = IPPROTO_SCTP; 4603 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 4604 h->ip_ttl = ttl ? ttl : V_ip_defttl; 4605 h->ip_src = pd->dst->v4; 4606 h->ip_dst = pd->src->v4; 4607 4608 off += sizeof(struct ip); 4609 break; 4610 #endif /* INET */ 4611 #ifdef INET6 4612 case AF_INET6: 4613 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 4614 4615 h6 = mtod(m, struct ip6_hdr *); 4616 4617 /* IP header fields included in the TCP checksum */ 4618 h6->ip6_vfc |= IPV6_VERSION; 4619 h6->ip6_nxt = IPPROTO_SCTP; 4620 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 4621 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 4622 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 4623 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 4624 4625 off += sizeof(struct ip6_hdr); 4626 break; 4627 #endif /* INET6 */ 4628 default: 4629 unhandled_af(af); 4630 } 4631 4632 /* SCTP header */ 4633 hdr = mtodo(m, off); 4634 4635 hdr->src_port = pd->hdr.sctp.dest_port; 4636 hdr->dest_port = pd->hdr.sctp.src_port; 4637 hdr->v_tag = pd->sctp_initiate_tag; 4638 hdr->checksum = 0; 4639 4640 /* Abort chunk. */ 4641 off += sizeof(struct sctphdr); 4642 chunk = mtodo(m, off); 4643 4644 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 4645 chunk->chunk_length = htons(sizeof(*chunk)); 4646 4647 /* SCTP checksum */ 4648 off += sizeof(*chunk); 4649 m->m_pkthdr.len = m->m_len = off; 4650 4651 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 4652 4653 if (rtableid >= 0) 4654 M_SETFIB(m, rtableid); 4655 4656 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4657 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4658 if (pfse == NULL) { 4659 m_freem(m); 4660 return; 4661 } 4662 4663 switch (af) { 4664 #ifdef INET 4665 case AF_INET: 4666 pfse->pfse_type = PFSE_IP; 4667 break; 4668 #endif /* INET */ 4669 #ifdef INET6 4670 case AF_INET6: 4671 pfse->pfse_type = PFSE_IP6; 4672 break; 4673 #endif /* INET6 */ 4674 } 4675 4676 pfse->pfse_m = m; 4677 pf_send(pfse); 4678 } 4679 4680 void 4681 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 4682 const struct pf_addr *saddr, const struct pf_addr *daddr, 4683 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 4684 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 4685 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid, 4686 u_short *reason) 4687 { 4688 struct pf_send_entry *pfse; 4689 struct mbuf *m; 4690 4691 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 4692 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, 0, rtableid, reason); 4693 if (m == NULL) 4694 return; 4695 4696 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4697 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4698 if (pfse == NULL) { 4699 m_freem(m); 4700 REASON_SET(reason, PFRES_MEMORY); 4701 return; 4702 } 4703 4704 switch (af) { 4705 #ifdef INET 4706 case AF_INET: 4707 pfse->pfse_type = PFSE_IP; 4708 break; 4709 #endif /* INET */ 4710 #ifdef INET6 4711 case AF_INET6: 4712 pfse->pfse_type = PFSE_IP6; 4713 break; 4714 #endif /* INET6 */ 4715 default: 4716 unhandled_af(af); 4717 } 4718 4719 pfse->pfse_m = m; 4720 pf_send(pfse); 4721 } 4722 4723 static void 4724 pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum) 4725 { 4726 /* undo NAT changes, if they have taken place */ 4727 if (nr != NULL) { 4728 pf_addrcpy(pd->src, &pd->osrc, pd->af); 4729 pf_addrcpy(pd->dst, &pd->odst, pd->af); 4730 if (pd->sport) 4731 *pd->sport = pd->osport; 4732 if (pd->dport) 4733 *pd->dport = pd->odport; 4734 if (pd->ip_sum) 4735 *pd->ip_sum = bip_sum; 4736 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 4737 } 4738 } 4739 4740 static void 4741 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 4742 struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum, 4743 u_short *reason, int rtableid) 4744 { 4745 pf_undo_nat(nr, pd, bip_sum); 4746 4747 if (pd->proto == IPPROTO_TCP && 4748 ((r->rule_flag & PFRULE_RETURNRST) || 4749 (r->rule_flag & PFRULE_RETURN)) && 4750 !(tcp_get_flags(th) & TH_RST)) { 4751 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 4752 4753 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 4754 IPPROTO_TCP, pd->af)) 4755 REASON_SET(reason, PFRES_PROTCKSUM); 4756 else { 4757 if (tcp_get_flags(th) & TH_SYN) 4758 ack++; 4759 if (tcp_get_flags(th) & TH_FIN) 4760 ack++; 4761 pf_send_tcp(r, pd->af, pd->dst, 4762 pd->src, th->th_dport, th->th_sport, 4763 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 4764 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid, 4765 reason); 4766 } 4767 } else if (pd->proto == IPPROTO_SCTP && 4768 (r->rule_flag & PFRULE_RETURN)) { 4769 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 4770 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 4771 r->return_icmp) 4772 pf_send_icmp(pd->m, r->return_icmp >> 8, 4773 r->return_icmp & 255, 0, pd->af, r, rtableid); 4774 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 4775 r->return_icmp6) 4776 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 4777 r->return_icmp6 & 255, 0, pd->af, r, rtableid); 4778 } 4779 4780 static int 4781 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 4782 { 4783 struct m_tag *mtag; 4784 u_int8_t mpcp; 4785 4786 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 4787 if (mtag == NULL) 4788 return (0); 4789 4790 if (prio == PF_PRIO_ZERO) 4791 prio = 0; 4792 4793 mpcp = *(uint8_t *)(mtag + 1); 4794 4795 return (mpcp == prio); 4796 } 4797 4798 static int 4799 pf_icmp_to_bandlim(uint8_t type) 4800 { 4801 switch (type) { 4802 case ICMP_ECHO: 4803 case ICMP_ECHOREPLY: 4804 return (BANDLIM_ICMP_ECHO); 4805 case ICMP_TSTAMP: 4806 case ICMP_TSTAMPREPLY: 4807 return (BANDLIM_ICMP_TSTAMP); 4808 case ICMP_UNREACH: 4809 default: 4810 return (BANDLIM_ICMP_UNREACH); 4811 } 4812 } 4813 4814 static void 4815 pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_kstate *s, 4816 struct pf_state_peer *src, struct pf_state_peer *dst, 4817 u_short *reason) 4818 { 4819 /* 4820 * We are sending challenge ACK as a response to SYN packet, which 4821 * matches existing state (modulo TCP window check). Therefore packet 4822 * must be sent on behalf of destination. 4823 * 4824 * We expect sender to remain either silent, or send RST packet 4825 * so both, firewall and remote peer, can purge dead state from 4826 * memory. 4827 */ 4828 pf_send_tcp(s->rule, pd->af, pd->dst, pd->src, 4829 pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo, 4830 src->seqlo, TH_ACK, 0, 0, s->rule->return_ttl, 0, 0, 0, 4831 s->rule->rtableid, reason); 4832 } 4833 4834 static void 4835 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int mtu, 4836 sa_family_t af, struct pf_krule *r, int rtableid) 4837 { 4838 struct pf_send_entry *pfse; 4839 struct mbuf *m0; 4840 struct pf_mtag *pf_mtag; 4841 4842 /* ICMP packet rate limitation. */ 4843 switch (af) { 4844 #ifdef INET6 4845 case AF_INET6: 4846 if (icmp6_ratelimit(NULL, type, code)) 4847 return; 4848 break; 4849 #endif /* INET6 */ 4850 #ifdef INET 4851 case AF_INET: 4852 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 4853 return; 4854 break; 4855 #endif /* INET */ 4856 } 4857 4858 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 4859 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 4860 if (pfse == NULL) 4861 return; 4862 4863 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 4864 free(pfse, M_PFTEMP); 4865 return; 4866 } 4867 4868 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 4869 free(pfse, M_PFTEMP); 4870 return; 4871 } 4872 /* XXX: revisit */ 4873 m0->m_flags |= M_SKIP_FIREWALL; 4874 4875 if (rtableid >= 0) 4876 M_SETFIB(m0, rtableid); 4877 4878 #ifdef ALTQ 4879 if (r->qid) { 4880 pf_mtag->qid = r->qid; 4881 /* add hints for ecn */ 4882 pf_mtag->hdr = mtod(m0, struct ip *); 4883 } 4884 #endif /* ALTQ */ 4885 4886 switch (af) { 4887 #ifdef INET 4888 case AF_INET: 4889 pfse->pfse_type = PFSE_ICMP; 4890 break; 4891 #endif /* INET */ 4892 #ifdef INET6 4893 case AF_INET6: 4894 pfse->pfse_type = PFSE_ICMP6; 4895 break; 4896 #endif /* INET6 */ 4897 } 4898 pfse->pfse_m = m0; 4899 pfse->icmpopts.type = type; 4900 pfse->icmpopts.code = code; 4901 pfse->icmpopts.mtu = mtu; 4902 pf_send(pfse); 4903 } 4904 4905 /* 4906 * Return ((n = 0) == (a = b [with mask m])) 4907 * Note: n != 0 => returns (a != b [with mask m]) 4908 */ 4909 int 4910 pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m, 4911 const struct pf_addr *b, sa_family_t af) 4912 { 4913 switch (af) { 4914 #ifdef INET 4915 case AF_INET: 4916 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 4917 return (n == 0); 4918 break; 4919 #endif /* INET */ 4920 #ifdef INET6 4921 case AF_INET6: 4922 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 4923 return (n == 0); 4924 break; 4925 #endif /* INET6 */ 4926 } 4927 4928 return (n != 0); 4929 } 4930 4931 /* 4932 * Return 1 if b <= a <= e, otherwise return 0. 4933 */ 4934 int 4935 pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e, 4936 const struct pf_addr *a, sa_family_t af) 4937 { 4938 switch (af) { 4939 #ifdef INET 4940 case AF_INET: 4941 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 4942 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 4943 return (0); 4944 break; 4945 #endif /* INET */ 4946 #ifdef INET6 4947 case AF_INET6: { 4948 int i; 4949 4950 /* check a >= b */ 4951 for (i = 0; i < 4; ++i) 4952 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 4953 break; 4954 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 4955 return (0); 4956 /* check a <= e */ 4957 for (i = 0; i < 4; ++i) 4958 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 4959 break; 4960 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 4961 return (0); 4962 break; 4963 } 4964 #endif /* INET6 */ 4965 } 4966 return (1); 4967 } 4968 4969 static int 4970 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 4971 { 4972 switch (op) { 4973 case PF_OP_IRG: 4974 return ((p > a1) && (p < a2)); 4975 case PF_OP_XRG: 4976 return ((p < a1) || (p > a2)); 4977 case PF_OP_RRG: 4978 return ((p >= a1) && (p <= a2)); 4979 case PF_OP_EQ: 4980 return (p == a1); 4981 case PF_OP_NE: 4982 return (p != a1); 4983 case PF_OP_LT: 4984 return (p < a1); 4985 case PF_OP_LE: 4986 return (p <= a1); 4987 case PF_OP_GT: 4988 return (p > a1); 4989 case PF_OP_GE: 4990 return (p >= a1); 4991 } 4992 return (0); /* never reached */ 4993 } 4994 4995 int 4996 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 4997 { 4998 return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p))); 4999 } 5000 5001 static int 5002 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 5003 { 5004 if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE) 5005 return (0); 5006 return (pf_match(op, a1, a2, u)); 5007 } 5008 5009 static int 5010 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 5011 { 5012 if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE) 5013 return (0); 5014 return (pf_match(op, a1, a2, g)); 5015 } 5016 5017 int 5018 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 5019 { 5020 if (*tag == -1) 5021 *tag = mtag; 5022 5023 return ((!r->match_tag_not && r->match_tag == *tag) || 5024 (r->match_tag_not && r->match_tag != *tag)); 5025 } 5026 5027 static int 5028 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 5029 { 5030 struct ifnet *ifp = m->m_pkthdr.rcvif; 5031 struct pfi_kkif *kif; 5032 5033 if (ifp == NULL) 5034 return (0); 5035 5036 kif = (struct pfi_kkif *)ifp->if_pf_kif; 5037 5038 if (kif == NULL) { 5039 DPFPRINTF(PF_DEBUG_URGENT, 5040 "%s: kif == NULL, @%d via %s", __func__, r->nr, 5041 r->rcv_ifname); 5042 return (0); 5043 } 5044 5045 return (pfi_kkif_match(r->rcv_kif, kif)); 5046 } 5047 5048 int 5049 pf_tag_packet(struct pf_pdesc *pd, int tag) 5050 { 5051 5052 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 5053 5054 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 5055 return (ENOMEM); 5056 5057 pd->pf_mtag->tag = tag; 5058 5059 return (0); 5060 } 5061 5062 /* 5063 * XXX: We rely on malloc(9) returning pointer aligned addresses. 5064 */ 5065 #define PF_ANCHORSTACK_MATCH 0x00000001 5066 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 5067 5068 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 5069 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 5070 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 5071 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 5072 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 5073 } while (0) 5074 5075 enum pf_test_status 5076 pf_step_into_anchor(struct pf_test_ctx *ctx, struct pf_krule *r) 5077 { 5078 enum pf_test_status rv; 5079 5080 PF_RULES_RASSERT(); 5081 5082 if (ctx->depth >= PF_ANCHOR_STACK_MAX) { 5083 printf("%s: anchor stack overflow on %s\n", 5084 __func__, r->anchor->name); 5085 return (PF_TEST_FAIL); 5086 } 5087 5088 ctx->depth++; 5089 5090 if (r->anchor_wildcard) { 5091 struct pf_kanchor *child; 5092 rv = PF_TEST_OK; 5093 RB_FOREACH(child, pf_kanchor_node, &r->anchor->children) { 5094 rv = pf_match_rule(ctx, &child->ruleset); 5095 if ((rv == PF_TEST_QUICK) || (rv == PF_TEST_FAIL)) { 5096 /* 5097 * we either hit a rule with quick action 5098 * (more likely), or hit some runtime 5099 * error (e.g. pool_get() failure). 5100 */ 5101 break; 5102 } 5103 } 5104 } else { 5105 rv = pf_match_rule(ctx, &r->anchor->ruleset); 5106 /* 5107 * Unless errors occured, stop iff any rule matched 5108 * within quick anchors. 5109 */ 5110 if (rv != PF_TEST_FAIL && r->quick == PF_TEST_QUICK && 5111 *ctx->am == r) 5112 rv = PF_TEST_QUICK; 5113 } 5114 5115 ctx->depth--; 5116 5117 return (rv); 5118 } 5119 5120 struct pf_keth_anchor_stackframe { 5121 struct pf_keth_ruleset *rs; 5122 struct pf_keth_rule *r; /* XXX: + match bit */ 5123 struct pf_keth_anchor *child; 5124 }; 5125 5126 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 5127 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 5128 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 5129 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 5130 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 5131 } while (0) 5132 5133 void 5134 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 5135 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 5136 struct pf_keth_rule **a, int *match) 5137 { 5138 struct pf_keth_anchor_stackframe *f; 5139 5140 NET_EPOCH_ASSERT(); 5141 5142 if (match) 5143 *match = 0; 5144 if (*depth >= PF_ANCHOR_STACK_MAX) { 5145 printf("%s: anchor stack overflow on %s\n", 5146 __func__, (*r)->anchor->name); 5147 *r = TAILQ_NEXT(*r, entries); 5148 return; 5149 } else if (*depth == 0 && a != NULL) 5150 *a = *r; 5151 f = stack + (*depth)++; 5152 f->rs = *rs; 5153 f->r = *r; 5154 if ((*r)->anchor_wildcard) { 5155 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 5156 5157 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 5158 *r = NULL; 5159 return; 5160 } 5161 *rs = &f->child->ruleset; 5162 } else { 5163 f->child = NULL; 5164 *rs = &(*r)->anchor->ruleset; 5165 } 5166 *r = TAILQ_FIRST((*rs)->active.rules); 5167 } 5168 5169 int 5170 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 5171 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 5172 struct pf_keth_rule **a, int *match) 5173 { 5174 struct pf_keth_anchor_stackframe *f; 5175 struct pf_keth_rule *fr; 5176 int quick = 0; 5177 5178 NET_EPOCH_ASSERT(); 5179 5180 do { 5181 if (*depth <= 0) 5182 break; 5183 f = stack + *depth - 1; 5184 fr = PF_ETH_ANCHOR_RULE(f); 5185 if (f->child != NULL) { 5186 /* 5187 * This block traverses through 5188 * a wildcard anchor. 5189 */ 5190 if (match != NULL && *match) { 5191 /* 5192 * If any of "*" matched, then 5193 * "foo/ *" matched, mark frame 5194 * appropriately. 5195 */ 5196 PF_ETH_ANCHOR_SET_MATCH(f); 5197 *match = 0; 5198 } 5199 f->child = RB_NEXT(pf_keth_anchor_node, 5200 &fr->anchor->children, f->child); 5201 if (f->child != NULL) { 5202 *rs = &f->child->ruleset; 5203 *r = TAILQ_FIRST((*rs)->active.rules); 5204 if (*r == NULL) 5205 continue; 5206 else 5207 break; 5208 } 5209 } 5210 (*depth)--; 5211 if (*depth == 0 && a != NULL) 5212 *a = NULL; 5213 *rs = f->rs; 5214 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 5215 quick = fr->quick; 5216 *r = TAILQ_NEXT(fr, entries); 5217 } while (*r == NULL); 5218 5219 return (quick); 5220 } 5221 5222 void 5223 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 5224 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 5225 { 5226 switch (af) { 5227 #ifdef INET 5228 case AF_INET: 5229 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 5230 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 5231 break; 5232 #endif /* INET */ 5233 #ifdef INET6 5234 case AF_INET6: 5235 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 5236 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 5237 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 5238 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 5239 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 5240 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 5241 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 5242 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 5243 break; 5244 #endif /* INET6 */ 5245 } 5246 } 5247 5248 void 5249 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 5250 { 5251 switch (af) { 5252 #ifdef INET 5253 case AF_INET: 5254 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 5255 break; 5256 #endif /* INET */ 5257 #ifdef INET6 5258 case AF_INET6: 5259 if (addr->addr32[3] == 0xffffffff) { 5260 addr->addr32[3] = 0; 5261 if (addr->addr32[2] == 0xffffffff) { 5262 addr->addr32[2] = 0; 5263 if (addr->addr32[1] == 0xffffffff) { 5264 addr->addr32[1] = 0; 5265 addr->addr32[0] = 5266 htonl(ntohl(addr->addr32[0]) + 1); 5267 } else 5268 addr->addr32[1] = 5269 htonl(ntohl(addr->addr32[1]) + 1); 5270 } else 5271 addr->addr32[2] = 5272 htonl(ntohl(addr->addr32[2]) + 1); 5273 } else 5274 addr->addr32[3] = 5275 htonl(ntohl(addr->addr32[3]) + 1); 5276 break; 5277 #endif /* INET6 */ 5278 } 5279 } 5280 5281 void 5282 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 5283 { 5284 /* 5285 * Modern rules use the same flags in rules as they do in states. 5286 */ 5287 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 5288 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 5289 5290 /* 5291 * Old-style scrub rules have different flags which need to be translated. 5292 */ 5293 if (r->rule_flag & PFRULE_RANDOMID) 5294 a->flags |= PFSTATE_RANDOMID; 5295 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 5296 a->flags |= PFSTATE_SETTOS; 5297 a->set_tos = r->set_tos; 5298 } 5299 5300 if (r->qid) 5301 a->qid = r->qid; 5302 if (r->pqid) 5303 a->pqid = r->pqid; 5304 if (r->rtableid >= 0) 5305 a->rtableid = r->rtableid; 5306 a->log |= r->log; 5307 if (r->min_ttl) 5308 a->min_ttl = r->min_ttl; 5309 if (r->max_mss) 5310 a->max_mss = r->max_mss; 5311 if (r->dnpipe) 5312 a->dnpipe = r->dnpipe; 5313 if (r->dnrpipe) 5314 a->dnrpipe = r->dnrpipe; 5315 if (r->dnpipe || r->dnrpipe) { 5316 if (r->free_flags & PFRULE_DN_IS_PIPE) 5317 a->flags |= PFSTATE_DN_IS_PIPE; 5318 else 5319 a->flags &= ~PFSTATE_DN_IS_PIPE; 5320 } 5321 if (r->scrub_flags & PFSTATE_SETPRIO) { 5322 a->set_prio[0] = r->set_prio[0]; 5323 a->set_prio[1] = r->set_prio[1]; 5324 } 5325 if (r->allow_opts) 5326 a->allow_opts = r->allow_opts; 5327 if (r->max_pkt_size) 5328 a->max_pkt_size = r->max_pkt_size; 5329 } 5330 5331 int 5332 pf_socket_lookup(struct pf_pdesc *pd) 5333 { 5334 struct pf_addr *saddr, *daddr; 5335 u_int16_t sport, dport; 5336 struct inpcbinfo *pi; 5337 struct inpcb *inp; 5338 5339 pd->lookup.uid = -1; 5340 pd->lookup.gid = -1; 5341 5342 switch (pd->proto) { 5343 case IPPROTO_TCP: 5344 sport = pd->hdr.tcp.th_sport; 5345 dport = pd->hdr.tcp.th_dport; 5346 pi = &V_tcbinfo; 5347 break; 5348 case IPPROTO_UDP: 5349 sport = pd->hdr.udp.uh_sport; 5350 dport = pd->hdr.udp.uh_dport; 5351 pi = &V_udbinfo; 5352 break; 5353 default: 5354 return (-1); 5355 } 5356 if (pd->dir == PF_IN) { 5357 saddr = pd->src; 5358 daddr = pd->dst; 5359 } else { 5360 u_int16_t p; 5361 5362 p = sport; 5363 sport = dport; 5364 dport = p; 5365 saddr = pd->dst; 5366 daddr = pd->src; 5367 } 5368 switch (pd->af) { 5369 #ifdef INET 5370 case AF_INET: 5371 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 5372 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 5373 if (inp == NULL) { 5374 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 5375 daddr->v4, dport, INPLOOKUP_WILDCARD | 5376 INPLOOKUP_RLOCKPCB, NULL, pd->m); 5377 if (inp == NULL) 5378 return (-1); 5379 } 5380 break; 5381 #endif /* INET */ 5382 #ifdef INET6 5383 case AF_INET6: 5384 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 5385 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 5386 if (inp == NULL) { 5387 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 5388 &daddr->v6, dport, INPLOOKUP_WILDCARD | 5389 INPLOOKUP_RLOCKPCB, NULL, pd->m); 5390 if (inp == NULL) 5391 return (-1); 5392 } 5393 break; 5394 #endif /* INET6 */ 5395 default: 5396 unhandled_af(pd->af); 5397 } 5398 INP_RLOCK_ASSERT(inp); 5399 pd->lookup.uid = inp->inp_cred->cr_uid; 5400 pd->lookup.gid = inp->inp_cred->cr_gid; 5401 INP_RUNLOCK(inp); 5402 5403 return (1); 5404 } 5405 5406 /* post: r => (r[0] == type /\ r[1] >= min_typelen >= 2 "validity" 5407 * /\ (eoh - r) >= min_typelen >= 2 "safety" ) 5408 * 5409 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen 5410 */ 5411 uint8_t* 5412 pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type, 5413 u_int8_t min_typelen) 5414 { 5415 uint8_t *eoh = opts + hlen; 5416 5417 if (min_typelen < 2) 5418 return (NULL); 5419 5420 while ((eoh - opt) >= min_typelen) { 5421 switch (*opt) { 5422 case TCPOPT_EOL: 5423 /* FALLTHROUGH - Workaround the failure of some 5424 systems to NOP-pad their bzero'd option buffers, 5425 producing spurious EOLs */ 5426 case TCPOPT_NOP: 5427 opt++; 5428 continue; 5429 default: 5430 if (opt[0] == type && 5431 opt[1] >= min_typelen) 5432 return (opt); 5433 } 5434 5435 opt += MAX(opt[1], 2); /* evade infinite loops */ 5436 } 5437 5438 return (NULL); 5439 } 5440 5441 u_int8_t 5442 pf_get_wscale(struct pf_pdesc *pd) 5443 { 5444 int olen; 5445 uint8_t opts[MAX_TCPOPTLEN], *opt; 5446 uint8_t wscale = 0; 5447 5448 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5449 if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m, 5450 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 5451 return (0); 5452 5453 opt = opts; 5454 while ((opt = pf_find_tcpopt(opt, opts, olen, 5455 TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) { 5456 wscale = opt[2]; 5457 wscale = MIN(wscale, TCP_MAX_WINSHIFT); 5458 wscale |= PF_WSCALE_FLAG; 5459 5460 opt += opt[1]; 5461 } 5462 5463 return (wscale); 5464 } 5465 5466 u_int16_t 5467 pf_get_mss(struct pf_pdesc *pd) 5468 { 5469 int olen; 5470 uint8_t opts[MAX_TCPOPTLEN], *opt; 5471 u_int16_t mss = V_tcp_mssdflt; 5472 5473 olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr); 5474 if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m, 5475 pd->off + sizeof(struct tcphdr), opts, olen, NULL, pd->af)) 5476 return (0); 5477 5478 opt = opts; 5479 while ((opt = pf_find_tcpopt(opt, opts, olen, 5480 TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) { 5481 memcpy(&mss, (opt + 2), 2); 5482 mss = ntohs(mss); 5483 opt += opt[1]; 5484 } 5485 5486 return (mss); 5487 } 5488 5489 static u_int16_t 5490 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 5491 { 5492 struct nhop_object *nh; 5493 #ifdef INET6 5494 struct in6_addr dst6; 5495 uint32_t scopeid; 5496 #endif /* INET6 */ 5497 int hlen = 0; 5498 uint16_t mss = 0; 5499 5500 NET_EPOCH_ASSERT(); 5501 5502 switch (af) { 5503 #ifdef INET 5504 case AF_INET: 5505 hlen = sizeof(struct ip); 5506 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 5507 if (nh != NULL) 5508 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5509 break; 5510 #endif /* INET */ 5511 #ifdef INET6 5512 case AF_INET6: 5513 hlen = sizeof(struct ip6_hdr); 5514 in6_splitscope(&addr->v6, &dst6, &scopeid); 5515 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 5516 if (nh != NULL) 5517 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 5518 break; 5519 #endif /* INET6 */ 5520 } 5521 5522 mss = max(V_tcp_mssdflt, mss); 5523 mss = min(mss, offer); 5524 mss = max(mss, 64); /* sanity - at least max opt space */ 5525 return (mss); 5526 } 5527 5528 static u_int32_t 5529 pf_tcp_iss(struct pf_pdesc *pd) 5530 { 5531 SHA512_CTX ctx; 5532 union { 5533 uint8_t bytes[SHA512_DIGEST_LENGTH]; 5534 uint32_t words[1]; 5535 } digest; 5536 5537 if (V_pf_tcp_secret_init == 0) { 5538 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 5539 SHA512_Init(&V_pf_tcp_secret_ctx); 5540 SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 5541 sizeof(V_pf_tcp_secret)); 5542 V_pf_tcp_secret_init = 1; 5543 } 5544 5545 ctx = V_pf_tcp_secret_ctx; 5546 5547 SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short)); 5548 SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short)); 5549 switch (pd->af) { 5550 case AF_INET6: 5551 SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr)); 5552 SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr)); 5553 break; 5554 case AF_INET: 5555 SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr)); 5556 SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr)); 5557 break; 5558 } 5559 SHA512_Final(digest.bytes, &ctx); 5560 V_pf_tcp_iss_off += 4096; 5561 #define ISN_RANDOM_INCREMENT (4096 - 1) 5562 return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 5563 V_pf_tcp_iss_off); 5564 #undef ISN_RANDOM_INCREMENT 5565 } 5566 5567 static bool 5568 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 5569 { 5570 bool match = true; 5571 5572 /* Always matches if not set */ 5573 if (! r->isset) 5574 return (!r->neg); 5575 5576 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 5577 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 5578 match = false; 5579 break; 5580 } 5581 } 5582 5583 return (match ^ r->neg); 5584 } 5585 5586 static int 5587 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 5588 { 5589 if (*tag == -1) 5590 *tag = mtag; 5591 5592 return ((!r->match_tag_not && r->match_tag == *tag) || 5593 (r->match_tag_not && r->match_tag != *tag)); 5594 } 5595 5596 static void 5597 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 5598 { 5599 /* If we don't have the interface drop the packet. */ 5600 if (ifp == NULL) { 5601 m_freem(m); 5602 return; 5603 } 5604 5605 switch (ifp->if_type) { 5606 case IFT_ETHER: 5607 case IFT_XETHER: 5608 case IFT_L2VLAN: 5609 case IFT_BRIDGE: 5610 case IFT_IEEE8023ADLAG: 5611 break; 5612 default: 5613 m_freem(m); 5614 return; 5615 } 5616 5617 ifp->if_transmit(ifp, m); 5618 } 5619 5620 static int 5621 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 5622 { 5623 #ifdef INET 5624 struct ip ip; 5625 #endif /* INET */ 5626 #ifdef INET6 5627 struct ip6_hdr ip6; 5628 #endif /* INET6 */ 5629 struct mbuf *m = *m0; 5630 struct ether_header *e; 5631 struct pf_keth_rule *r, *rm, *a = NULL; 5632 struct pf_keth_ruleset *ruleset = NULL; 5633 struct pf_mtag *mtag; 5634 struct pf_keth_ruleq *rules; 5635 struct pf_addr *src = NULL, *dst = NULL; 5636 struct pfi_kkif *bridge_to; 5637 sa_family_t af = 0; 5638 uint16_t proto; 5639 int asd = 0, match = 0; 5640 int tag = -1; 5641 uint8_t action; 5642 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACK_MAX]; 5643 5644 MPASS(kif->pfik_ifp->if_vnet == curvnet); 5645 NET_EPOCH_ASSERT(); 5646 5647 PF_RULES_RLOCK_TRACKER; 5648 5649 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 5650 5651 mtag = pf_find_mtag(m); 5652 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 5653 /* Dummynet re-injects packets after they've 5654 * completed their delay. We've already 5655 * processed them, so pass unconditionally. */ 5656 5657 /* But only once. We may see the packet multiple times (e.g. 5658 * PFIL_IN/PFIL_OUT). */ 5659 pf_dummynet_flag_remove(m, mtag); 5660 5661 return (PF_PASS); 5662 } 5663 5664 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 5665 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 5666 DPFPRINTF(PF_DEBUG_URGENT, 5667 "%s: m_len < sizeof(struct ether_header)" 5668 ", pullup failed", __func__); 5669 return (PF_DROP); 5670 } 5671 e = mtod(m, struct ether_header *); 5672 proto = ntohs(e->ether_type); 5673 5674 switch (proto) { 5675 #ifdef INET 5676 case ETHERTYPE_IP: { 5677 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5678 sizeof(ip))) 5679 return (PF_DROP); 5680 5681 af = AF_INET; 5682 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 5683 (caddr_t)&ip); 5684 src = (struct pf_addr *)&ip.ip_src; 5685 dst = (struct pf_addr *)&ip.ip_dst; 5686 break; 5687 } 5688 #endif /* INET */ 5689 #ifdef INET6 5690 case ETHERTYPE_IPV6: { 5691 if (m_length(m, NULL) < (sizeof(struct ether_header) + 5692 sizeof(ip6))) 5693 return (PF_DROP); 5694 5695 af = AF_INET6; 5696 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 5697 (caddr_t)&ip6); 5698 src = (struct pf_addr *)&ip6.ip6_src; 5699 dst = (struct pf_addr *)&ip6.ip6_dst; 5700 break; 5701 } 5702 #endif /* INET6 */ 5703 } 5704 5705 PF_RULES_RLOCK(); 5706 5707 ruleset = V_pf_keth; 5708 rules = atomic_load_ptr(&ruleset->active.rules); 5709 for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) { 5710 counter_u64_add(r->evaluations, 1); 5711 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 5712 5713 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 5714 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5715 "kif"); 5716 r = r->skip[PFE_SKIP_IFP].ptr; 5717 } 5718 else if (r->direction && r->direction != dir) { 5719 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5720 "dir"); 5721 r = r->skip[PFE_SKIP_DIR].ptr; 5722 } 5723 else if (r->proto && r->proto != proto) { 5724 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5725 "proto"); 5726 r = r->skip[PFE_SKIP_PROTO].ptr; 5727 } 5728 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 5729 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5730 "src"); 5731 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 5732 } 5733 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 5734 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5735 "dst"); 5736 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 5737 } 5738 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 5739 r->ipsrc.neg, kif, M_GETFIB(m))) { 5740 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5741 "ip_src"); 5742 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 5743 } 5744 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 5745 r->ipdst.neg, kif, M_GETFIB(m))) { 5746 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5747 "ip_dst"); 5748 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 5749 } 5750 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 5751 mtag ? mtag->tag : 0)) { 5752 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 5753 "match_tag"); 5754 r = TAILQ_NEXT(r, entries); 5755 } 5756 else { 5757 if (r->tag) 5758 tag = r->tag; 5759 if (r->anchor == NULL) { 5760 /* Rule matches */ 5761 rm = r; 5762 5763 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 5764 5765 if (r->quick) 5766 break; 5767 5768 r = TAILQ_NEXT(r, entries); 5769 } else { 5770 pf_step_into_keth_anchor(anchor_stack, &asd, 5771 &ruleset, &r, &a, &match); 5772 } 5773 } 5774 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 5775 &ruleset, &r, &a, &match)) 5776 break; 5777 } 5778 5779 r = rm; 5780 5781 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 5782 5783 /* Default to pass. */ 5784 if (r == NULL) { 5785 PF_RULES_RUNLOCK(); 5786 return (PF_PASS); 5787 } 5788 5789 /* Execute action. */ 5790 counter_u64_add(r->packets[dir == PF_OUT], 1); 5791 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 5792 pf_update_timestamp(r); 5793 5794 /* Shortcut. Don't tag if we're just going to drop anyway. */ 5795 if (r->action == PF_DROP) { 5796 PF_RULES_RUNLOCK(); 5797 return (PF_DROP); 5798 } 5799 5800 if (tag > 0) { 5801 if (mtag == NULL) 5802 mtag = pf_get_mtag(m); 5803 if (mtag == NULL) { 5804 PF_RULES_RUNLOCK(); 5805 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5806 return (PF_DROP); 5807 } 5808 mtag->tag = tag; 5809 } 5810 5811 if (r->qid != 0) { 5812 if (mtag == NULL) 5813 mtag = pf_get_mtag(m); 5814 if (mtag == NULL) { 5815 PF_RULES_RUNLOCK(); 5816 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5817 return (PF_DROP); 5818 } 5819 mtag->qid = r->qid; 5820 } 5821 5822 action = r->action; 5823 bridge_to = r->bridge_to; 5824 5825 /* Dummynet */ 5826 if (r->dnpipe) { 5827 struct ip_fw_args dnflow; 5828 5829 /* Drop packet if dummynet is not loaded. */ 5830 if (ip_dn_io_ptr == NULL) { 5831 PF_RULES_RUNLOCK(); 5832 m_freem(m); 5833 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5834 return (PF_DROP); 5835 } 5836 if (mtag == NULL) 5837 mtag = pf_get_mtag(m); 5838 if (mtag == NULL) { 5839 PF_RULES_RUNLOCK(); 5840 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 5841 return (PF_DROP); 5842 } 5843 5844 bzero(&dnflow, sizeof(dnflow)); 5845 5846 /* We don't have port numbers here, so we set 0. That means 5847 * that we'll be somewhat limited in distinguishing flows (i.e. 5848 * only based on IP addresses, not based on port numbers), but 5849 * it's better than nothing. */ 5850 dnflow.f_id.dst_port = 0; 5851 dnflow.f_id.src_port = 0; 5852 dnflow.f_id.proto = 0; 5853 5854 dnflow.rule.info = r->dnpipe; 5855 dnflow.rule.info |= IPFW_IS_DUMMYNET; 5856 if (r->dnflags & PFRULE_DN_IS_PIPE) 5857 dnflow.rule.info |= IPFW_IS_PIPE; 5858 5859 dnflow.f_id.extra = dnflow.rule.info; 5860 5861 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 5862 dnflow.flags |= IPFW_ARGS_ETHER; 5863 dnflow.ifp = kif->pfik_ifp; 5864 5865 switch (af) { 5866 case AF_INET: 5867 dnflow.f_id.addr_type = 4; 5868 dnflow.f_id.src_ip = src->v4.s_addr; 5869 dnflow.f_id.dst_ip = dst->v4.s_addr; 5870 break; 5871 case AF_INET6: 5872 dnflow.flags |= IPFW_ARGS_IP6; 5873 dnflow.f_id.addr_type = 6; 5874 dnflow.f_id.src_ip6 = src->v6; 5875 dnflow.f_id.dst_ip6 = dst->v6; 5876 break; 5877 } 5878 5879 PF_RULES_RUNLOCK(); 5880 5881 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 5882 ip_dn_io_ptr(m0, &dnflow); 5883 if (*m0 != NULL) 5884 pf_dummynet_flag_remove(m, mtag); 5885 } else { 5886 PF_RULES_RUNLOCK(); 5887 } 5888 5889 if (action == PF_PASS && bridge_to) { 5890 pf_bridge_to(bridge_to->pfik_ifp, *m0); 5891 *m0 = NULL; /* We've eaten the packet. */ 5892 } 5893 5894 return (action); 5895 } 5896 5897 #define PF_TEST_ATTRIB(t, a) \ 5898 if (t) { \ 5899 r = a; \ 5900 continue; \ 5901 } else do { \ 5902 } while (0) 5903 5904 static __inline u_short 5905 pf_rule_apply_nat(struct pf_test_ctx *ctx, struct pf_krule *r) 5906 { 5907 struct pf_pdesc *pd = ctx->pd; 5908 u_short transerror; 5909 u_int8_t nat_action; 5910 5911 if (r->rule_flag & PFRULE_AFTO) { 5912 /* Don't translate if there was an old style NAT rule */ 5913 if (ctx->nr != NULL) 5914 return (PFRES_TRANSLATE); 5915 5916 /* pass af-to rules, unsupported on match rules */ 5917 KASSERT(r->action != PF_MATCH, ("%s: af-to on match rule", __func__)); 5918 /* XXX I can imagine scenarios where we have both NAT and RDR source tracking */ 5919 ctx->nat_pool = &(r->nat); 5920 ctx->nr = r; 5921 pd->naf = r->naf; 5922 if (pf_get_transaddr_af(ctx->nr, pd) == -1) { 5923 return (PFRES_TRANSLATE); 5924 } 5925 return (PFRES_MATCH); 5926 } else if (r->rdr.cur || r->nat.cur) { 5927 /* Don't translate if there was an old style NAT rule */ 5928 if (ctx->nr != NULL) 5929 return (PFRES_TRANSLATE); 5930 5931 /* match/pass nat-to/rdr-to rules */ 5932 ctx->nr = r; 5933 if (r->nat.cur) { 5934 nat_action = PF_NAT; 5935 ctx->nat_pool = &(r->nat); 5936 } else { 5937 nat_action = PF_RDR; 5938 ctx->nat_pool = &(r->rdr); 5939 } 5940 5941 transerror = pf_get_transaddr(ctx, ctx->nr, 5942 nat_action, ctx->nat_pool); 5943 if (transerror == PFRES_MATCH) { 5944 ctx->rewrite += pf_translate_compat(ctx); 5945 return(PFRES_MATCH); 5946 } 5947 return (transerror); 5948 } 5949 5950 return (PFRES_MAX); 5951 } 5952 5953 enum pf_test_status 5954 pf_match_rule(struct pf_test_ctx *ctx, struct pf_kruleset *ruleset) 5955 { 5956 struct pf_krule_item *ri; 5957 struct pf_krule *r; 5958 struct pf_krule *save_a; 5959 struct pf_kruleset *save_aruleset; 5960 struct pf_pdesc *pd = ctx->pd; 5961 u_short transerror; 5962 5963 r = TAILQ_FIRST(ruleset->rules[PF_RULESET_FILTER].active.ptr); 5964 while (r != NULL) { 5965 struct pf_statelim *stlim = NULL; 5966 struct pf_sourcelim *srlim = NULL; 5967 struct pf_source *sr = NULL; 5968 unsigned int gen; 5969 5970 if (ctx->pd->related_rule) { 5971 *ctx->rm = ctx->pd->related_rule; 5972 break; 5973 } 5974 PF_TEST_ATTRIB(r->rule_flag & PFRULE_EXPIRED, 5975 TAILQ_NEXT(r, entries)); 5976 /* Don't count expired rule evaluations. */ 5977 pf_counter_u64_add(&r->evaluations, 1); 5978 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5979 r->skip[PF_SKIP_IFP]); 5980 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5981 r->skip[PF_SKIP_DIR]); 5982 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5983 r->skip[PF_SKIP_AF]); 5984 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5985 r->skip[PF_SKIP_PROTO]); 5986 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf, 5987 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5988 r->skip[PF_SKIP_SRC_ADDR]); 5989 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af, 5990 r->dst.neg, NULL, M_GETFIB(pd->m)), 5991 r->skip[PF_SKIP_DST_ADDR]); 5992 switch (pd->virtual_proto) { 5993 case PF_VPROTO_FRAGMENT: 5994 /* tcp/udp only. port_op always 0 in other cases */ 5995 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5996 TAILQ_NEXT(r, entries)); 5997 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5998 TAILQ_NEXT(r, entries)); 5999 /* icmp only. type/code always 0 in other cases */ 6000 PF_TEST_ATTRIB((r->type || r->code), 6001 TAILQ_NEXT(r, entries)); 6002 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 6003 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 6004 TAILQ_NEXT(r, entries)); 6005 break; 6006 6007 case IPPROTO_TCP: 6008 PF_TEST_ATTRIB((r->flagset & tcp_get_flags(ctx->th)) 6009 != r->flags, 6010 TAILQ_NEXT(r, entries)); 6011 /* FALLTHROUGH */ 6012 case IPPROTO_SCTP: 6013 case IPPROTO_UDP: 6014 /* tcp/udp only. port_op always 0 in other cases */ 6015 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 6016 r->src.port[0], r->src.port[1], pd->nsport), 6017 r->skip[PF_SKIP_SRC_PORT]); 6018 /* tcp/udp only. port_op always 0 in other cases */ 6019 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 6020 r->dst.port[0], r->dst.port[1], pd->ndport), 6021 r->skip[PF_SKIP_DST_PORT]); 6022 /* tcp/udp only. uid.op always 0 in other cases */ 6023 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 6024 pf_socket_lookup(pd), 1)) && 6025 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 6026 pd->lookup.uid), 6027 TAILQ_NEXT(r, entries)); 6028 /* tcp/udp only. gid.op always 0 in other cases */ 6029 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 6030 pf_socket_lookup(pd), 1)) && 6031 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 6032 pd->lookup.gid), 6033 TAILQ_NEXT(r, entries)); 6034 break; 6035 6036 case IPPROTO_ICMP: 6037 case IPPROTO_ICMPV6: 6038 /* icmp only. type always 0 in other cases */ 6039 PF_TEST_ATTRIB(r->type && r->type != ctx->icmptype + 1, 6040 TAILQ_NEXT(r, entries)); 6041 /* icmp only. type always 0 in other cases */ 6042 PF_TEST_ATTRIB(r->code && r->code != ctx->icmpcode + 1, 6043 TAILQ_NEXT(r, entries)); 6044 break; 6045 6046 default: 6047 break; 6048 } 6049 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 6050 TAILQ_NEXT(r, entries)); 6051 PF_TEST_ATTRIB(r->prio && 6052 !pf_match_ieee8021q_pcp(r->prio, pd->m), 6053 TAILQ_NEXT(r, entries)); 6054 PF_TEST_ATTRIB(r->prob && 6055 r->prob <= arc4random(), 6056 TAILQ_NEXT(r, entries)); 6057 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, 6058 &ctx->tag, pd->pf_mtag ? pd->pf_mtag->tag : 0), 6059 TAILQ_NEXT(r, entries)); 6060 PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) == 6061 r->rcvifnot), 6062 TAILQ_NEXT(r, entries)); 6063 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 6064 pd->virtual_proto != PF_VPROTO_FRAGMENT), 6065 TAILQ_NEXT(r, entries)); 6066 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 6067 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 6068 pf_osfp_fingerprint(pd, ctx->th), 6069 r->os_fingerprint)), 6070 TAILQ_NEXT(r, entries)); 6071 if (r->statelim.id != PF_STATELIM_ID_NONE) { 6072 stlim = pf_statelim_find(r->statelim.id); 6073 6074 /* 6075 * Treat a missing limiter like an exhausted limiter. 6076 * There is no "backend" to get a resource out of 6077 * so the rule can't create state. 6078 */ 6079 PF_TEST_ATTRIB(stlim == NULL, TAILQ_NEXT(r, entries)); 6080 6081 /* 6082 * An overcommitted pool means this rule 6083 * can't create state. 6084 */ 6085 if (stlim->pfstlim_inuse >= stlim->pfstlim_limit) { 6086 gen = pf_statelim_enter(stlim); 6087 stlim->pfstlim_counters.hardlimited++; 6088 pf_statelim_leave(stlim, gen); 6089 if (r->statelim.limiter_action == PF_LIMITER_BLOCK) { 6090 ctx->limiter_drop = 1; 6091 REASON_SET(&ctx->reason, PFRES_MAXSTATES); 6092 break; /* stop rule processing */ 6093 } 6094 r = TAILQ_NEXT(r, entries); 6095 continue; 6096 } 6097 6098 /* 6099 * Is access to the pool rate limited? 6100 */ 6101 if (stlim->pfstlim_rate.limit != 0) { 6102 struct timespec ts; 6103 getnanouptime(&ts); 6104 uint64_t diff = SEC_TO_NSEC(ts.tv_sec) + 6105 ts.tv_nsec - stlim->pfstlim_rate_ts; 6106 6107 if (diff < stlim->pfstlim_rate_token) { 6108 gen = pf_statelim_enter(stlim); 6109 stlim->pfstlim_counters.ratelimited++; 6110 pf_statelim_leave(stlim, gen); 6111 if (r->statelim.limiter_action == 6112 PF_LIMITER_BLOCK) { 6113 ctx->limiter_drop = 1; 6114 REASON_SET(&ctx->reason, 6115 PFRES_MAXSTATES); 6116 /* stop rule processing */ 6117 break; 6118 } 6119 r = TAILQ_NEXT(r, entries); 6120 continue; 6121 } 6122 6123 if (diff > stlim->pfstlim_rate_bucket) { 6124 stlim->pfstlim_rate_ts = 6125 SEC_TO_NSEC(ts.tv_sec) + ts.tv_nsec - 6126 stlim->pfstlim_rate_bucket; 6127 } 6128 } 6129 } 6130 6131 if (r->sourcelim.id != PF_SOURCELIM_ID_NONE) { 6132 struct pf_source key; 6133 6134 srlim = pf_sourcelim_find(r->sourcelim.id); 6135 6136 /* 6137 * Treat a missing pool like an overcommitted pool. 6138 * There is no "backend" to get a resource out of 6139 * so the rule can't create state. 6140 */ 6141 PF_TEST_ATTRIB(srlim == NULL, TAILQ_NEXT(r, entries)); 6142 6143 pf_source_key(srlim, &key, ctx->pd->af, 6144 ctx->pd->src); 6145 sr = pf_source_find(srlim, &key); 6146 if (sr != NULL) { 6147 /* 6148 * An overcommitted limiter means this rule 6149 * can't create state. 6150 */ 6151 if (sr->pfsr_inuse >= srlim->pfsrlim_limit) { 6152 sr->pfsr_counters.hardlimited++; 6153 gen = pf_sourcelim_enter(srlim); 6154 srlim->pfsrlim_counters.hardlimited++; 6155 pf_sourcelim_leave(srlim, gen); 6156 if (r->sourcelim.limiter_action == 6157 PF_LIMITER_BLOCK) { 6158 ctx->limiter_drop = 1; 6159 REASON_SET(&ctx->reason, 6160 PFRES_SRCLIMIT); 6161 /* stop rule processing */ 6162 break; 6163 } 6164 r = TAILQ_NEXT(r, entries); 6165 continue; 6166 } 6167 6168 /* 6169 * Is access to the pool rate limited? 6170 */ 6171 if (srlim->pfsrlim_rate.limit != 0) { 6172 struct timespec ts; 6173 getnanouptime(&ts); 6174 uint64_t diff = SEC_TO_NSEC(ts.tv_sec) + 6175 ts.tv_nsec - sr->pfsr_rate_ts; 6176 6177 if (diff < srlim->pfsrlim_rate_token) { 6178 sr->pfsr_counters.ratelimited++; 6179 gen = pf_sourcelim_enter(srlim); 6180 srlim->pfsrlim_counters 6181 .ratelimited++; 6182 pf_sourcelim_leave(srlim, gen); 6183 if (r->sourcelim.limiter_action == 6184 PF_LIMITER_BLOCK) { 6185 ctx->limiter_drop = 1; 6186 REASON_SET(&ctx->reason, 6187 PFRES_SRCLIMIT); 6188 /* stop rules */ 6189 break; 6190 } 6191 r = TAILQ_NEXT(r, entries); 6192 continue; 6193 } 6194 6195 if (diff > srlim->pfsrlim_rate_bucket) { 6196 sr->pfsr_rate_ts = 6197 SEC_TO_NSEC(ts.tv_sec) + ts.tv_nsec - 6198 srlim->pfsrlim_rate_bucket; 6199 } 6200 } 6201 } else { 6202 /* 6203 * a new source entry will (should) 6204 * admit a state. 6205 */ 6206 6207 if (srlim->pfsrlim_nsources >= 6208 srlim->pfsrlim_entries) { 6209 gen = pf_sourcelim_enter(srlim); 6210 srlim->pfsrlim_counters.addrlimited++; 6211 pf_sourcelim_leave(srlim, gen); 6212 r = TAILQ_NEXT(r, entries); 6213 continue; 6214 } 6215 } 6216 } 6217 6218 /* must be last! */ 6219 if (r->pktrate.limit) { 6220 PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)), 6221 TAILQ_NEXT(r, entries)); 6222 } 6223 /* FALLTHROUGH */ 6224 if (r->tag) 6225 ctx->tag = r->tag; 6226 if (r->anchor == NULL) { 6227 6228 if (r->rule_flag & PFRULE_ONCE) { 6229 uint32_t rule_flag; 6230 6231 rule_flag = r->rule_flag; 6232 if ((rule_flag & PFRULE_EXPIRED) == 0 && 6233 atomic_cmpset_int(&r->rule_flag, rule_flag, 6234 rule_flag | PFRULE_EXPIRED)) { 6235 r->exptime = time_uptime; 6236 } else { 6237 r = TAILQ_NEXT(r, entries); 6238 continue; 6239 } 6240 } 6241 6242 if (r->action == PF_MATCH) { 6243 /* 6244 * Apply translations before increasing counters, 6245 * in case it fails. 6246 */ 6247 transerror = pf_rule_apply_nat(ctx, r); 6248 switch (transerror) { 6249 case PFRES_MATCH: 6250 /* Translation action found in rule and applied successfully */ 6251 case PFRES_MAX: 6252 /* No translation action found in rule */ 6253 break; 6254 default: 6255 /* Translation action found in rule but failed to apply */ 6256 REASON_SET(&ctx->reason, transerror); 6257 return (PF_TEST_FAIL); 6258 } 6259 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 6260 if (ri == NULL) { 6261 REASON_SET(&ctx->reason, PFRES_MEMORY); 6262 return (PF_TEST_FAIL); 6263 } 6264 ri->r = r; 6265 6266 if (SLIST_EMPTY(ctx->match_rules)) { 6267 SLIST_INSERT_HEAD(ctx->match_rules, ri, entry); 6268 } else { 6269 SLIST_INSERT_AFTER(ctx->last_match_rule, ri, entry); 6270 } 6271 ctx->last_match_rule = ri; 6272 6273 pf_rule_to_actions(r, &pd->act); 6274 if (r->log) 6275 PFLOG_PACKET(r->action, PFRES_MATCH, r, 6276 ctx->a, ruleset, pd, 1, NULL); 6277 } else { 6278 /* 6279 * found matching r 6280 */ 6281 *ctx->rm = r; 6282 /* 6283 * anchor, with ruleset, where r belongs to 6284 */ 6285 *ctx->am = ctx->a; 6286 /* 6287 * ruleset where r belongs to 6288 */ 6289 *ctx->rsm = ruleset; 6290 /* 6291 * ruleset, where anchor belongs to. 6292 */ 6293 ctx->arsm = ctx->aruleset; 6294 /* 6295 * state/source pools 6296 */ 6297 6298 ctx->statelim = stlim; 6299 ctx->sourcelim = srlim; 6300 ctx->source = sr; 6301 } 6302 if (pd->act.log & PF_LOG_MATCHES) 6303 pf_log_matches(pd, r, ctx->a, ruleset, ctx->match_rules); 6304 if (r->quick) { 6305 ctx->test_status = PF_TEST_QUICK; 6306 break; 6307 } 6308 } else { 6309 save_a = ctx->a; 6310 save_aruleset = ctx->aruleset; 6311 6312 ctx->a = r; /* remember anchor */ 6313 ctx->aruleset = ruleset; /* and its ruleset */ 6314 if (ctx->a->quick) 6315 ctx->test_status = PF_TEST_QUICK; 6316 /* 6317 * Note: we don't need to restore if we are not going 6318 * to continue with ruleset evaluation. 6319 */ 6320 if (pf_step_into_anchor(ctx, r) != PF_TEST_OK) { 6321 break; 6322 } 6323 ctx->a = save_a; 6324 ctx->aruleset = save_aruleset; 6325 } 6326 r = TAILQ_NEXT(r, entries); 6327 } 6328 6329 6330 return (ctx->test_status); 6331 } 6332 6333 static int 6334 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 6335 struct pf_pdesc *pd, struct pf_krule **am, 6336 struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp, 6337 struct pf_krule_slist *match_rules) 6338 { 6339 struct pf_krule *r = NULL; 6340 struct pf_kruleset *ruleset = NULL; 6341 struct pf_test_ctx ctx; 6342 u_short transerror; 6343 int action = PF_PASS; 6344 u_int16_t bproto_sum = 0, bip_sum = 0; 6345 enum pf_test_status rv; 6346 6347 PF_RULES_RASSERT(); 6348 6349 bzero(&ctx, sizeof(ctx)); 6350 ctx.tag = -1; 6351 ctx.pd = pd; 6352 ctx.rm = rm; 6353 ctx.am = am; 6354 ctx.rsm = rsm; 6355 ctx.th = &pd->hdr.tcp; 6356 ctx.reason = *reason; 6357 ctx.match_rules = match_rules; 6358 6359 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 6360 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 6361 6362 if (inp != NULL) { 6363 INP_LOCK_ASSERT(inp); 6364 pd->lookup.uid = inp->inp_cred->cr_uid; 6365 pd->lookup.gid = inp->inp_cred->cr_gid; 6366 pd->lookup.done = 1; 6367 } 6368 6369 if (pd->ip_sum) 6370 bip_sum = *pd->ip_sum; 6371 6372 switch (pd->virtual_proto) { 6373 case IPPROTO_TCP: 6374 bproto_sum = ctx.th->th_sum; 6375 pd->nsport = ctx.th->th_sport; 6376 pd->ndport = ctx.th->th_dport; 6377 break; 6378 case IPPROTO_UDP: 6379 bproto_sum = pd->hdr.udp.uh_sum; 6380 pd->nsport = pd->hdr.udp.uh_sport; 6381 pd->ndport = pd->hdr.udp.uh_dport; 6382 break; 6383 case IPPROTO_SCTP: 6384 pd->nsport = pd->hdr.sctp.src_port; 6385 pd->ndport = pd->hdr.sctp.dest_port; 6386 break; 6387 #ifdef INET 6388 case IPPROTO_ICMP: 6389 MPASS(pd->af == AF_INET); 6390 ctx.icmptype = pd->hdr.icmp.icmp_type; 6391 ctx.icmpcode = pd->hdr.icmp.icmp_code; 6392 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 6393 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 6394 if (ctx.icmp_dir == PF_IN) { 6395 pd->nsport = ctx.virtual_id; 6396 pd->ndport = ctx.virtual_type; 6397 } else { 6398 pd->nsport = ctx.virtual_type; 6399 pd->ndport = ctx.virtual_id; 6400 } 6401 break; 6402 #endif /* INET */ 6403 #ifdef INET6 6404 case IPPROTO_ICMPV6: 6405 MPASS(pd->af == AF_INET6); 6406 ctx.icmptype = pd->hdr.icmp6.icmp6_type; 6407 ctx.icmpcode = pd->hdr.icmp6.icmp6_code; 6408 ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype, 6409 &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type); 6410 if (ctx.icmp_dir == PF_IN) { 6411 pd->nsport = ctx.virtual_id; 6412 pd->ndport = ctx.virtual_type; 6413 } else { 6414 pd->nsport = ctx.virtual_type; 6415 pd->ndport = ctx.virtual_id; 6416 } 6417 6418 break; 6419 #endif /* INET6 */ 6420 default: 6421 pd->nsport = pd->ndport = 0; 6422 break; 6423 } 6424 pd->osport = pd->nsport; 6425 pd->odport = pd->ndport; 6426 6427 /* check packet for BINAT/NAT/RDR */ 6428 transerror = pf_get_translation(&ctx); 6429 switch (transerror) { 6430 default: 6431 /* A translation error occurred. */ 6432 REASON_SET(&ctx.reason, transerror); 6433 goto cleanup; 6434 case PFRES_MAX: 6435 /* No match. */ 6436 break; 6437 case PFRES_MATCH: 6438 KASSERT(ctx.sk != NULL, ("%s: null sk", __func__)); 6439 KASSERT(ctx.nk != NULL, ("%s: null nk", __func__)); 6440 if (ctx.nr->log) { 6441 PFLOG_PACKET(ctx.nr->action, PFRES_MATCH, ctx.nr, ctx.a, 6442 ruleset, pd, 1, NULL); 6443 } 6444 6445 ctx.rewrite += pf_translate_compat(&ctx); 6446 ctx.nat_pool = &(ctx.nr->rdr); 6447 } 6448 6449 *ctx.rm = &V_pf_default_rule; 6450 if (ctx.nr && ctx.nr->natpass) { 6451 r = ctx.nr; 6452 ruleset = *ctx.rsm; 6453 } else { 6454 ruleset = &pf_main_ruleset; 6455 rv = pf_match_rule(&ctx, ruleset); 6456 if (rv == PF_TEST_FAIL || ctx.limiter_drop == 1) { 6457 REASON_SET(reason, ctx.reason); 6458 goto cleanup; 6459 } 6460 6461 r = *ctx.rm; /* matching rule */ 6462 ctx.a = *ctx.am; /* rule that defines an anchor containing 'r' */ 6463 ruleset = *ctx.rsm; /* ruleset of the anchor defined by the rule 'a' */ 6464 ctx.aruleset = ctx.arsm; /* ruleset of the 'a' rule itself */ 6465 6466 /* apply actions for last matching pass/block rule */ 6467 pf_rule_to_actions(r, &pd->act); 6468 transerror = pf_rule_apply_nat(&ctx, r); 6469 switch (transerror) { 6470 case PFRES_MATCH: 6471 /* Translation action found in rule and applied successfully */ 6472 case PFRES_MAX: 6473 /* No translation action found in rule */ 6474 break; 6475 default: 6476 /* Translation action found in rule but failed to apply */ 6477 REASON_SET(&ctx.reason, transerror); 6478 goto cleanup; 6479 } 6480 } 6481 6482 REASON_SET(&ctx.reason, PFRES_MATCH); 6483 6484 if (r->log) { 6485 if (ctx.rewrite) 6486 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6487 PFLOG_PACKET(r->action, ctx.reason, r, ctx.a, ruleset, pd, 1, NULL); 6488 } 6489 if (pd->act.log & PF_LOG_MATCHES) 6490 pf_log_matches(pd, r, ctx.a, ruleset, ctx.match_rules); 6491 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 6492 (r->action == PF_DROP) && 6493 ((r->rule_flag & PFRULE_RETURNRST) || 6494 (r->rule_flag & PFRULE_RETURNICMP) || 6495 (r->rule_flag & PFRULE_RETURN))) { 6496 pf_return(r, ctx.nr, pd, ctx.th, bproto_sum, 6497 bip_sum, &ctx.reason, r->rtableid); 6498 } 6499 6500 if (r->action == PF_DROP) 6501 goto cleanup; 6502 6503 if (ctx.tag > 0 && pf_tag_packet(pd, ctx.tag)) { 6504 REASON_SET(&ctx.reason, PFRES_MEMORY); 6505 goto cleanup; 6506 } 6507 if (pd->act.rtableid >= 0) 6508 M_SETFIB(pd->m, pd->act.rtableid); 6509 6510 if (r->rt) { 6511 /* 6512 * Set act.rt here instead of in pf_rule_to_actions() because 6513 * it is applied only from the last pass rule. For rules 6514 * with the prefer-ipv6-nexthop option act.rt_af is a hint 6515 * about AF of the forwarded packet and might be changed. 6516 */ 6517 pd->act.rt = r->rt; 6518 if (r->rt == PF_REPLYTO) 6519 pd->act.rt_af = pd->af; 6520 else 6521 pd->act.rt_af = pd->naf; 6522 if ((transerror = pf_map_addr_sn(pd->af, r, pd->src, 6523 &pd->act.rt_addr, &pd->act.rt_af, &pd->act.rt_kif, NULL, 6524 &(r->route), PF_SN_ROUTE)) != PFRES_MATCH) { 6525 REASON_SET(&ctx.reason, transerror); 6526 goto cleanup; 6527 } 6528 } 6529 6530 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 6531 (!ctx.state_icmp && (r->keep_state || ctx.nr != NULL || 6532 (pd->flags & PFDESC_TCP_NORM)))) { 6533 bool nat64; 6534 6535 action = pf_create_state(r, &ctx, sm, bproto_sum, bip_sum); 6536 ctx.sk = ctx.nk = NULL; 6537 if (action != PF_PASS) { 6538 pf_udp_mapping_release(ctx.udp_mapping); 6539 if (r->log || (ctx.nr != NULL && ctx.nr->log) || 6540 ctx.reason == PFRES_MEMORY) 6541 pd->act.log |= PF_LOG_FORCE; 6542 if (action == PF_DROP && 6543 (r->rule_flag & PFRULE_RETURN)) 6544 pf_return(r, ctx.nr, pd, ctx.th, 6545 bproto_sum, bip_sum, &ctx.reason, 6546 pd->act.rtableid); 6547 *reason = ctx.reason; 6548 return (action); 6549 } 6550 6551 if (pd->proto == IPPROTO_TCP && 6552 r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 6553 action = pf_synproxy_ack(r, pd, sm, &ctx.act); 6554 if (action != PF_PASS) 6555 goto cleanup; /* PF_SYNPROXY_DROP */ 6556 } 6557 6558 nat64 = pd->af != pd->naf; 6559 if (nat64) { 6560 int ret; 6561 6562 if (ctx.sk == NULL) 6563 ctx.sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE]; 6564 if (ctx.nk == NULL) 6565 ctx.nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK]; 6566 6567 if (pd->dir == PF_IN) { 6568 ret = pf_translate(pd, &ctx.sk->addr[pd->didx], 6569 ctx.sk->port[pd->didx], &ctx.sk->addr[pd->sidx], 6570 ctx.sk->port[pd->sidx], ctx.virtual_type, 6571 ctx.icmp_dir); 6572 } else { 6573 ret = pf_translate(pd, &ctx.sk->addr[pd->sidx], 6574 ctx.sk->port[pd->sidx], &ctx.sk->addr[pd->didx], 6575 ctx.sk->port[pd->didx], ctx.virtual_type, 6576 ctx.icmp_dir); 6577 } 6578 6579 if (ret < 0) 6580 goto cleanup; 6581 6582 ctx.rewrite += ret; 6583 6584 if (ctx.rewrite && ctx.sk->af != ctx.nk->af) 6585 action = PF_AFRT; 6586 } 6587 } else { 6588 uma_zfree(V_pf_state_key_z, ctx.sk); 6589 uma_zfree(V_pf_state_key_z, ctx.nk); 6590 ctx.sk = ctx.nk = NULL; 6591 pf_udp_mapping_release(ctx.udp_mapping); 6592 } 6593 6594 /* copy back packet headers if we performed NAT operations */ 6595 if (ctx.rewrite) 6596 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 6597 6598 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 6599 pd->dir == PF_OUT && 6600 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) { 6601 /* 6602 * We want the state created, but we dont 6603 * want to send this in case a partner 6604 * firewall has to know about it to allow 6605 * replies through it. 6606 */ 6607 *reason = ctx.reason; 6608 return (PF_DEFER); 6609 } 6610 6611 *reason = ctx.reason; 6612 return (action); 6613 6614 cleanup: 6615 uma_zfree(V_pf_state_key_z, ctx.sk); 6616 uma_zfree(V_pf_state_key_z, ctx.nk); 6617 pf_udp_mapping_release(ctx.udp_mapping); 6618 *reason = ctx.reason; 6619 6620 return (PF_DROP); 6621 } 6622 6623 static int 6624 pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx, 6625 struct pf_kstate **sm, u_int16_t bproto_sum, u_int16_t bip_sum) 6626 { 6627 struct pf_pdesc *pd = ctx->pd; 6628 struct pf_kstate *s = NULL; 6629 struct pf_statelim *stlim = NULL; 6630 struct pf_sourcelim *srlim = NULL; 6631 struct pf_source *sr = NULL; 6632 struct pf_state_link *pfl; 6633 struct pf_ksrc_node *sns[PF_SN_MAX] = { NULL }; 6634 /* 6635 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same 6636 * but for PF_SN_NAT it is different. Don't try optimizing it, 6637 * just store all 3 hashes. 6638 */ 6639 struct pf_srchash *snhs[PF_SN_MAX] = { NULL }; 6640 struct tcphdr *th = &pd->hdr.tcp; 6641 u_int16_t mss = V_tcp_mssdflt; 6642 u_short sn_reason; 6643 6644 /* check maximums */ 6645 if (r->max_states && 6646 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 6647 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 6648 REASON_SET(&ctx->reason, PFRES_MAXSTATES); 6649 goto csfailed; 6650 } 6651 /* src node for limits */ 6652 if ((r->rule_flag & PFRULE_SRCTRACK) && 6653 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af, 6654 NULL, NULL, pd->af, PF_SN_LIMIT)) != 0) { 6655 REASON_SET(&ctx->reason, sn_reason); 6656 goto csfailed; 6657 } 6658 /* src node for route-to rule */ 6659 if (r->rt) { 6660 if ((r->route.opts & PF_POOL_STICKYADDR) && 6661 (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, 6662 pd->af, &pd->act.rt_addr, pd->act.rt_kif, pd->act.rt_af, 6663 PF_SN_ROUTE)) != 0) { 6664 REASON_SET(&ctx->reason, sn_reason); 6665 goto csfailed; 6666 } 6667 } 6668 /* src node for translation rule */ 6669 if (ctx->nr != NULL) { 6670 KASSERT(ctx->nat_pool != NULL, ("%s: nat_pool is NULL", __func__)); 6671 /* 6672 * The NAT addresses are chosen during ruleset parsing. 6673 * The new afto code stores post-nat addresses in nsaddr. 6674 * The old nat code (also used for new nat-to rules) creates 6675 * state keys and stores addresses in them. 6676 */ 6677 if ((ctx->nat_pool->opts & PF_POOL_STICKYADDR) && 6678 (sn_reason = pf_insert_src_node(sns, snhs, ctx->nr, 6679 ctx->sk ? &(ctx->sk->addr[pd->sidx]) : pd->src, pd->af, 6680 ctx->nk ? &(ctx->nk->addr[1]) : &(pd->nsaddr), NULL, 6681 pd->naf, PF_SN_NAT)) != 0 ) { 6682 REASON_SET(&ctx->reason, sn_reason); 6683 goto csfailed; 6684 } 6685 } 6686 s = pf_alloc_state(M_NOWAIT); 6687 if (s == NULL) { 6688 REASON_SET(&ctx->reason, PFRES_MEMORY); 6689 goto csfailed; 6690 } 6691 s->rule = r; 6692 s->nat_rule = ctx->nr; 6693 s->anchor = ctx->a; 6694 s->match_rules = *ctx->match_rules; 6695 SLIST_INIT(&s->linkage); 6696 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 6697 6698 if (pd->act.allow_opts) 6699 s->state_flags |= PFSTATE_ALLOWOPTS; 6700 if (r->rule_flag & PFRULE_STATESLOPPY) 6701 s->state_flags |= PFSTATE_SLOPPY; 6702 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 6703 s->state_flags |= PFSTATE_SCRUB_TCP; 6704 if ((r->rule_flag & PFRULE_PFLOW) || 6705 (ctx->nr != NULL && ctx->nr->rule_flag & PFRULE_PFLOW)) 6706 s->state_flags |= PFSTATE_PFLOW; 6707 6708 s->act.log = pd->act.log & PF_LOG_ALL; 6709 s->sync_state = PFSYNC_S_NONE; 6710 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 6711 6712 if (ctx->nr != NULL) 6713 s->act.log |= ctx->nr->log & PF_LOG_ALL; 6714 switch (pd->proto) { 6715 case IPPROTO_TCP: 6716 s->src.seqlo = ntohl(th->th_seq); 6717 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 6718 if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN && 6719 r->keep_state == PF_STATE_MODULATE) { 6720 /* Generate sequence number modulator */ 6721 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 6722 0) 6723 s->src.seqdiff = 1; 6724 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 6725 htonl(s->src.seqlo + s->src.seqdiff), 0); 6726 ctx->rewrite = 1; 6727 } else 6728 s->src.seqdiff = 0; 6729 if (tcp_get_flags(th) & TH_SYN) { 6730 s->src.seqhi++; 6731 s->src.wscale = pf_get_wscale(pd); 6732 } 6733 s->src.max_win = MAX(ntohs(th->th_win), 1); 6734 if (s->src.wscale & PF_WSCALE_MASK) { 6735 /* Remove scale factor from initial window */ 6736 int win = s->src.max_win; 6737 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 6738 s->src.max_win = (win - 1) >> 6739 (s->src.wscale & PF_WSCALE_MASK); 6740 } 6741 if (tcp_get_flags(th) & TH_FIN) 6742 s->src.seqhi++; 6743 s->dst.seqhi = 1; 6744 s->dst.max_win = 1; 6745 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 6746 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 6747 s->timeout = PFTM_TCP_FIRST_PACKET; 6748 atomic_add_32(&V_pf_status.states_halfopen, 1); 6749 break; 6750 case IPPROTO_UDP: 6751 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 6752 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 6753 s->timeout = PFTM_UDP_FIRST_PACKET; 6754 break; 6755 case IPPROTO_SCTP: 6756 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 6757 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 6758 s->timeout = PFTM_SCTP_FIRST_PACKET; 6759 break; 6760 case IPPROTO_ICMP: 6761 #ifdef INET6 6762 case IPPROTO_ICMPV6: 6763 #endif /* INET6 */ 6764 s->timeout = PFTM_ICMP_FIRST_PACKET; 6765 break; 6766 default: 6767 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 6768 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 6769 s->timeout = PFTM_OTHER_FIRST_PACKET; 6770 } 6771 6772 s->creation = s->expire = pf_get_uptime(); 6773 6774 if (pd->proto == IPPROTO_TCP) { 6775 if (s->state_flags & PFSTATE_SCRUB_TCP && 6776 pf_normalize_tcp_init(pd, th, &s->src)) { 6777 REASON_SET(&ctx->reason, PFRES_MEMORY); 6778 goto csfailed; 6779 } 6780 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 6781 pf_normalize_tcp_stateful(pd, &ctx->reason, th, s, 6782 &s->src, &s->dst, &ctx->rewrite)) { 6783 /* This really shouldn't happen!!! */ 6784 DPFPRINTF(PF_DEBUG_URGENT, 6785 "%s: tcp normalize failed on first " 6786 "pkt", __func__); 6787 goto csfailed; 6788 } 6789 } else if (pd->proto == IPPROTO_SCTP) { 6790 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 6791 goto csfailed; 6792 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 6793 goto csfailed; 6794 } 6795 s->direction = pd->dir; 6796 6797 /* 6798 * sk/nk could already been setup by pf_get_translation(). 6799 */ 6800 if (ctx->sk == NULL && ctx->nk == NULL) { 6801 MPASS(pd->sport == NULL || (pd->osport == *pd->sport)); 6802 MPASS(pd->dport == NULL || (pd->odport == *pd->dport)); 6803 if (pf_state_key_setup(pd, pd->nsport, pd->ndport, 6804 &ctx->sk, &ctx->nk)) { 6805 goto csfailed; 6806 } 6807 } else 6808 KASSERT((ctx->sk != NULL && ctx->nk != NULL), ("%s: nr %p sk %p, nk %p", 6809 __func__, ctx->nr, ctx->sk, ctx->nk)); 6810 6811 stlim = ctx->statelim; 6812 if (stlim != NULL) { 6813 unsigned int gen; 6814 6815 pfl = malloc(sizeof(*pfl), M_PF_STATE_LINK, M_NOWAIT); 6816 if (pfl == NULL) { 6817 REASON_SET(&ctx->reason, PFRES_MEMORY); 6818 goto csfailed; 6819 } 6820 6821 gen = pf_statelim_enter(stlim); 6822 stlim->pfstlim_counters.admitted++; 6823 stlim->pfstlim_inuse++; 6824 pf_statelim_leave(stlim, gen); 6825 6826 stlim->pfstlim_rate_ts += stlim->pfstlim_rate_token; 6827 6828 s->statelim = stlim->pfstlim_id; 6829 pfl->pfl_state = s; 6830 pfl->pfl_type = PF_STATE_LINK_TYPE_STATELIM; 6831 6832 TAILQ_INSERT_TAIL(&stlim->pfstlim_states, pfl, pfl_link); 6833 SLIST_INSERT_HEAD(&s->linkage, pfl, pfl_linkage); 6834 } 6835 6836 srlim = ctx->sourcelim; 6837 if (srlim != NULL) { 6838 unsigned int gen; 6839 6840 sr = ctx->source; 6841 if (sr == NULL) { 6842 sr = malloc(sizeof(*sr), M_PF_SOURCE_LIM, M_NOWAIT | M_ZERO); 6843 if (sr == NULL) { 6844 gen = pf_sourcelim_enter(srlim); 6845 srlim->pfsrlim_counters.addrnomem++; 6846 pf_sourcelim_leave(srlim, gen); 6847 REASON_SET(&ctx->reason, PFRES_MEMORY); 6848 goto csfailed; 6849 } 6850 6851 sr->pfsr_parent = srlim; 6852 pf_source_key(srlim, sr, ctx->pd->af, ctx->pd->src); 6853 TAILQ_INIT(&sr->pfsr_states); 6854 6855 if (RB_INSERT(pf_source_tree, &srlim->pfsrlim_sources, 6856 sr) != NULL) { 6857 panic("%s: source pool %u (%p) " 6858 "insert collision %p?!", 6859 __func__, srlim->pfsrlim_id, srlim, sr); 6860 } 6861 6862 if (RB_INSERT(pf_source_ioc_tree, 6863 &srlim->pfsrlim_ioc_sources, sr) != NULL) { 6864 panic("%s: source pool %u (%p) ioc " 6865 "insert collision (%p)?!", 6866 __func__, srlim->pfsrlim_id, srlim, sr); 6867 } 6868 6869 sr->pfsr_empty_ts = time_uptime; 6870 TAILQ_INSERT_TAIL(&pf_source_gc, sr, pfsr_empty_gc); 6871 6872 gen = pf_sourcelim_enter(srlim); 6873 srlim->pfsrlim_nsources++; 6874 srlim->pfsrlim_counters.addrallocs++; 6875 pf_sourcelim_leave(srlim, gen); 6876 } else { 6877 MPASS(sr->pfsr_parent == srlim); 6878 } 6879 6880 pfl = malloc(sizeof(*pfl), M_PF_STATE_LINK, M_NOWAIT); 6881 if (pfl == NULL) { 6882 REASON_SET(&ctx->reason, PFRES_MEMORY); 6883 goto csfailed; 6884 } 6885 6886 pf_source_used(sr); 6887 6888 sr->pfsr_counters.admitted++; 6889 6890 gen = pf_sourcelim_enter(srlim); 6891 srlim->pfsrlim_counters.inuse++; 6892 srlim->pfsrlim_counters.admitted++; 6893 pf_sourcelim_leave(srlim, gen); 6894 6895 s->sourcelim = srlim->pfsrlim_id; 6896 pfl->pfl_state = s; 6897 pfl->pfl_type = PF_STATE_LINK_TYPE_SOURCELIM; 6898 6899 TAILQ_INSERT_TAIL(&sr->pfsr_states, pfl, pfl_link); 6900 SLIST_INSERT_HEAD(&s->linkage, pfl, pfl_linkage); 6901 } 6902 6903 /* Swap sk/nk for PF_OUT. */ 6904 if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif, 6905 (pd->dir == PF_IN) ? ctx->sk : ctx->nk, 6906 (pd->dir == PF_IN) ? ctx->nk : ctx->sk, s)) { 6907 REASON_SET(&ctx->reason, PFRES_STATEINS); 6908 goto drop; 6909 } else 6910 *sm = s; 6911 ctx->sk = ctx->nk = NULL; 6912 6913 STATE_INC_COUNTERS(s); 6914 6915 /* 6916 * Lock order is important: first state, then source node. 6917 */ 6918 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6919 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6920 s->sns[sn_type] = sns[sn_type]; 6921 PF_HASHROW_UNLOCK(snhs[sn_type]); 6922 } 6923 } 6924 6925 if (ctx->tag > 0) 6926 s->tag = ctx->tag; 6927 if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == 6928 TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { 6929 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 6930 pf_undo_nat(ctx->nr, pd, bip_sum); 6931 s->src.seqhi = arc4random(); 6932 /* Find mss option */ 6933 int rtid = M_GETFIB(pd->m); 6934 mss = pf_get_mss(pd); 6935 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 6936 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 6937 s->src.mss = mss; 6938 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 6939 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 6940 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 6941 pd->act.rtableid, &ctx->reason); 6942 REASON_SET(&ctx->reason, PFRES_SYNPROXY); 6943 return (PF_SYNPROXY_DROP); 6944 } 6945 6946 s->udp_mapping = ctx->udp_mapping; 6947 6948 return (PF_PASS); 6949 6950 csfailed: 6951 uma_zfree(V_pf_state_key_z, ctx->sk); 6952 uma_zfree(V_pf_state_key_z, ctx->nk); 6953 6954 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 6955 if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) { 6956 if (--sns[sn_type]->states == 0 && 6957 sns[sn_type]->expire == 0) { 6958 pf_unlink_src_node(sns[sn_type]); 6959 pf_free_src_node(sns[sn_type]); 6960 counter_u64_add( 6961 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 6962 } 6963 PF_HASHROW_UNLOCK(snhs[sn_type]); 6964 } 6965 } 6966 6967 drop: 6968 if (s != NULL) { 6969 struct pf_state_link *npfl; 6970 6971 SLIST_FOREACH_SAFE(pfl, &s->linkage, pfl_linkage, npfl) { 6972 struct pf_state_link_list *list; 6973 unsigned int gen; 6974 6975 /* who needs KASSERTS when we have NULL derefs */ 6976 6977 switch (pfl->pfl_type) { 6978 case PF_STATE_LINK_TYPE_STATELIM: 6979 gen = pf_statelim_enter(stlim); 6980 stlim->pfstlim_inuse--; 6981 pf_statelim_leave(stlim, gen); 6982 6983 stlim->pfstlim_rate_ts -= 6984 stlim->pfstlim_rate_token; 6985 list = &stlim->pfstlim_states; 6986 break; 6987 case PF_STATE_LINK_TYPE_SOURCELIM: 6988 gen = pf_sourcelim_enter(srlim); 6989 srlim->pfsrlim_counters.inuse--; 6990 pf_sourcelim_leave(srlim, gen); 6991 6992 sr->pfsr_rate_ts -= srlim->pfsrlim_rate_token; 6993 pf_source_rele(sr); 6994 6995 list = &sr->pfsr_states; 6996 break; 6997 default: 6998 panic("%s: unexpected link type on pfl %p", 6999 __func__, pfl); 7000 } 7001 7002 TAILQ_REMOVE(list, pfl, pfl_link); 7003 PF_STATE_LOCK_ASSERT(s); 7004 free(pfl, M_PF_STATE_LINK); 7005 } 7006 7007 pf_src_tree_remove_state(s); 7008 s->timeout = PFTM_UNLINKED; 7009 pf_free_state(s); 7010 } 7011 7012 return (PF_DROP); 7013 } 7014 7015 int 7016 pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport, 7017 struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type, 7018 int icmp_dir) 7019 { 7020 /* 7021 * pf_translate() implements OpenBSD's "new" NAT approach. 7022 * We don't follow it, because it involves a breaking syntax change 7023 * (removing nat/rdr rules, moving it into regular pf rules.) 7024 * It also moves NAT processing to be done after normal rules evaluation 7025 * whereas in FreeBSD that's done before rules processing. 7026 * 7027 * We adopt the function only for nat64, and keep other NAT processing 7028 * before rules processing. 7029 */ 7030 int rewrite = 0; 7031 int afto = pd->af != pd->naf; 7032 7033 MPASS(afto); 7034 7035 switch (pd->proto) { 7036 case IPPROTO_TCP: 7037 case IPPROTO_UDP: 7038 case IPPROTO_SCTP: 7039 if (afto || *pd->sport != sport) { 7040 pf_change_ap(pd, pd->src, pd->sport, 7041 saddr, sport); 7042 rewrite = 1; 7043 } 7044 if (afto || *pd->dport != dport) { 7045 pf_change_ap(pd, pd->dst, pd->dport, 7046 daddr, dport); 7047 rewrite = 1; 7048 } 7049 break; 7050 7051 #ifdef INET 7052 case IPPROTO_ICMP: 7053 /* pf_translate() is also used when logging invalid packets */ 7054 if (pd->af != AF_INET) 7055 return (0); 7056 7057 if (afto) { 7058 if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp)) 7059 return (-1); 7060 pd->proto = IPPROTO_ICMPV6; 7061 rewrite = 1; 7062 } 7063 if (virtual_type == htons(ICMP_ECHO)) { 7064 u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport; 7065 7066 if (icmpid != pd->hdr.icmp.icmp_id) { 7067 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 7068 pd->hdr.icmp.icmp_cksum, 7069 pd->hdr.icmp.icmp_id, icmpid, 0); 7070 pd->hdr.icmp.icmp_id = icmpid; 7071 /* XXX TODO copyback. */ 7072 rewrite = 1; 7073 } 7074 } 7075 break; 7076 #endif /* INET */ 7077 7078 #ifdef INET6 7079 case IPPROTO_ICMPV6: 7080 /* pf_translate() is also used when logging invalid packets */ 7081 if (pd->af != AF_INET6) 7082 return (0); 7083 7084 if (afto) { 7085 /* ip_sum will be recalculated in pf_translate_af */ 7086 if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6)) 7087 return (0); 7088 pd->proto = IPPROTO_ICMP; 7089 rewrite = 1; 7090 } 7091 break; 7092 #endif /* INET6 */ 7093 7094 default: 7095 break; 7096 } 7097 7098 return (rewrite); 7099 } 7100 7101 int 7102 pf_translate_compat(struct pf_test_ctx *ctx) 7103 { 7104 struct pf_pdesc *pd = ctx->pd; 7105 struct pf_state_key *nk = ctx->nk; 7106 struct tcphdr *th = &pd->hdr.tcp; 7107 int rewrite = 0; 7108 7109 KASSERT(ctx->sk != NULL, ("%s: null sk", __func__)); 7110 KASSERT(ctx->nk != NULL, ("%s: null nk", __func__)); 7111 7112 switch (pd->virtual_proto) { 7113 case IPPROTO_TCP: 7114 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 7115 nk->port[pd->sidx] != pd->nsport) { 7116 pf_change_ap(pd, pd->src, &th->th_sport, 7117 &nk->addr[pd->sidx], nk->port[pd->sidx]); 7118 pd->sport = &th->th_sport; 7119 pd->nsport = th->th_sport; 7120 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7121 } 7122 7123 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 7124 nk->port[pd->didx] != pd->ndport) { 7125 pf_change_ap(pd, pd->dst, &th->th_dport, 7126 &nk->addr[pd->didx], nk->port[pd->didx]); 7127 pd->dport = &th->th_dport; 7128 pd->ndport = th->th_dport; 7129 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7130 } 7131 rewrite++; 7132 break; 7133 case IPPROTO_UDP: 7134 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 7135 nk->port[pd->sidx] != pd->nsport) { 7136 pf_change_ap(pd, pd->src, 7137 &pd->hdr.udp.uh_sport, 7138 &nk->addr[pd->sidx], 7139 nk->port[pd->sidx]); 7140 pd->sport = &pd->hdr.udp.uh_sport; 7141 pd->nsport = pd->hdr.udp.uh_sport; 7142 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7143 } 7144 7145 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 7146 nk->port[pd->didx] != pd->ndport) { 7147 pf_change_ap(pd, pd->dst, 7148 &pd->hdr.udp.uh_dport, 7149 &nk->addr[pd->didx], 7150 nk->port[pd->didx]); 7151 pd->dport = &pd->hdr.udp.uh_dport; 7152 pd->ndport = pd->hdr.udp.uh_dport; 7153 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7154 } 7155 rewrite++; 7156 break; 7157 case IPPROTO_SCTP: { 7158 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) || 7159 nk->port[pd->sidx] != pd->nsport) { 7160 pf_change_ap(pd, pd->src, 7161 &pd->hdr.sctp.src_port, 7162 &nk->addr[pd->sidx], 7163 nk->port[pd->sidx]); 7164 pd->sport = &pd->hdr.sctp.src_port; 7165 pd->nsport = pd->hdr.sctp.src_port; 7166 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7167 } 7168 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) || 7169 nk->port[pd->didx] != pd->ndport) { 7170 pf_change_ap(pd, pd->dst, 7171 &pd->hdr.sctp.dest_port, 7172 &nk->addr[pd->didx], 7173 nk->port[pd->didx]); 7174 pd->dport = &pd->hdr.sctp.dest_port; 7175 pd->ndport = pd->hdr.sctp.dest_port; 7176 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7177 } 7178 break; 7179 } 7180 #ifdef INET 7181 case IPPROTO_ICMP: 7182 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) { 7183 pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, 7184 nk->addr[pd->sidx].v4.s_addr, 0); 7185 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7186 } 7187 7188 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) { 7189 pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, 7190 nk->addr[pd->didx].v4.s_addr, 0); 7191 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7192 } 7193 7194 if (ctx->virtual_type == htons(ICMP_ECHO) && 7195 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 7196 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 7197 pd->hdr.icmp.icmp_cksum, pd->nsport, 7198 nk->port[pd->sidx], 0); 7199 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 7200 pd->sport = &pd->hdr.icmp.icmp_id; 7201 } 7202 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7203 break; 7204 #endif /* INET */ 7205 #ifdef INET6 7206 case IPPROTO_ICMPV6: 7207 if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) { 7208 pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum, 7209 &nk->addr[pd->sidx], 0); 7210 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7211 } 7212 7213 if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) { 7214 pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum, 7215 &nk->addr[pd->didx], 0); 7216 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7217 } 7218 rewrite++; 7219 break; 7220 #endif /* INET */ 7221 default: 7222 switch (pd->af) { 7223 #ifdef INET 7224 case AF_INET: 7225 if (PF_ANEQ(&pd->nsaddr, 7226 &nk->addr[pd->sidx], AF_INET)) { 7227 pf_change_a(&pd->src->v4.s_addr, 7228 pd->ip_sum, 7229 nk->addr[pd->sidx].v4.s_addr, 0); 7230 pf_addrcpy(&pd->nsaddr, pd->src, pd->af); 7231 } 7232 7233 if (PF_ANEQ(&pd->ndaddr, 7234 &nk->addr[pd->didx], AF_INET)) { 7235 pf_change_a(&pd->dst->v4.s_addr, 7236 pd->ip_sum, 7237 nk->addr[pd->didx].v4.s_addr, 0); 7238 pf_addrcpy(&pd->ndaddr, pd->dst, pd->af); 7239 } 7240 break; 7241 #endif /* INET */ 7242 #ifdef INET6 7243 case AF_INET6: 7244 if (PF_ANEQ(&pd->nsaddr, 7245 &nk->addr[pd->sidx], AF_INET6)) { 7246 pf_addrcpy(&pd->nsaddr, &nk->addr[pd->sidx], 7247 pd->af); 7248 pf_addrcpy(pd->src, &nk->addr[pd->sidx], pd->af); 7249 } 7250 7251 if (PF_ANEQ(&pd->ndaddr, 7252 &nk->addr[pd->didx], AF_INET6)) { 7253 pf_addrcpy(&pd->ndaddr, &nk->addr[pd->didx], 7254 pd->af); 7255 pf_addrcpy(pd->dst, &nk->addr[pd->didx], 7256 pd->af); 7257 } 7258 break; 7259 #endif /* INET6 */ 7260 } 7261 break; 7262 } 7263 return (rewrite); 7264 } 7265 7266 static int 7267 pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd, 7268 u_short *reason, int *copyback, struct pf_state_peer *src, 7269 struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst) 7270 { 7271 struct tcphdr *th = &pd->hdr.tcp; 7272 u_int16_t win = ntohs(th->th_win); 7273 u_int32_t ack, end, data_end, seq, orig_seq; 7274 u_int8_t sws, dws; 7275 int ackskew; 7276 7277 if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) { 7278 sws = src->wscale & PF_WSCALE_MASK; 7279 dws = dst->wscale & PF_WSCALE_MASK; 7280 } else 7281 sws = dws = 0; 7282 7283 /* 7284 * Sequence tracking algorithm from Guido van Rooij's paper: 7285 * http://www.madison-gurkha.com/publications/tcp_filtering/ 7286 * tcp_filtering.ps 7287 */ 7288 7289 orig_seq = seq = ntohl(th->th_seq); 7290 if (src->seqlo == 0) { 7291 /* First packet from this end. Set its state */ 7292 7293 if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 7294 src->scrub == NULL) { 7295 if (pf_normalize_tcp_init(pd, th, src)) { 7296 REASON_SET(reason, PFRES_MEMORY); 7297 return (PF_DROP); 7298 } 7299 } 7300 7301 /* Deferred generation of sequence number modulator */ 7302 if (dst->seqdiff && !src->seqdiff) { 7303 /* use random iss for the TCP server */ 7304 while ((src->seqdiff = arc4random() - seq) == 0) 7305 ; 7306 ack = ntohl(th->th_ack) - dst->seqdiff; 7307 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 7308 src->seqdiff), 0); 7309 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 7310 *copyback = 1; 7311 } else { 7312 ack = ntohl(th->th_ack); 7313 } 7314 7315 end = seq + pd->p_len; 7316 if (tcp_get_flags(th) & TH_SYN) { 7317 end++; 7318 if (dst->wscale & PF_WSCALE_FLAG) { 7319 src->wscale = pf_get_wscale(pd); 7320 if (src->wscale & PF_WSCALE_FLAG) { 7321 /* Remove scale factor from initial 7322 * window */ 7323 sws = src->wscale & PF_WSCALE_MASK; 7324 win = ((u_int32_t)win + (1 << sws) - 1) 7325 >> sws; 7326 dws = dst->wscale & PF_WSCALE_MASK; 7327 } else { 7328 /* fixup other window */ 7329 dst->max_win = MIN(TCP_MAXWIN, 7330 (u_int32_t)dst->max_win << 7331 (dst->wscale & PF_WSCALE_MASK)); 7332 /* in case of a retrans SYN|ACK */ 7333 dst->wscale = 0; 7334 } 7335 } 7336 } 7337 data_end = end; 7338 if (tcp_get_flags(th) & TH_FIN) 7339 end++; 7340 7341 src->seqlo = seq; 7342 if (src->state < TCPS_SYN_SENT) 7343 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 7344 7345 /* 7346 * May need to slide the window (seqhi may have been set by 7347 * the crappy stack check or if we picked up the connection 7348 * after establishment) 7349 */ 7350 if (src->seqhi == 1 || 7351 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 7352 src->seqhi = end + MAX(1, dst->max_win << dws); 7353 if (win > src->max_win) 7354 src->max_win = win; 7355 7356 } else { 7357 ack = ntohl(th->th_ack) - dst->seqdiff; 7358 if (src->seqdiff) { 7359 /* Modulate sequence numbers */ 7360 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 7361 src->seqdiff), 0); 7362 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 7363 *copyback = 1; 7364 } 7365 end = seq + pd->p_len; 7366 if (tcp_get_flags(th) & TH_SYN) 7367 end++; 7368 data_end = end; 7369 if (tcp_get_flags(th) & TH_FIN) 7370 end++; 7371 } 7372 7373 if ((tcp_get_flags(th) & TH_ACK) == 0) { 7374 /* Let it pass through the ack skew check */ 7375 ack = dst->seqlo; 7376 } else if ((ack == 0 && 7377 (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 7378 /* broken tcp stacks do not set ack */ 7379 (dst->state < TCPS_SYN_SENT)) { 7380 /* 7381 * Many stacks (ours included) will set the ACK number in an 7382 * FIN|ACK if the SYN times out -- no sequence to ACK. 7383 */ 7384 ack = dst->seqlo; 7385 } 7386 7387 if (seq == end) { 7388 /* Ease sequencing restrictions on no data packets */ 7389 seq = src->seqlo; 7390 data_end = end = seq; 7391 } 7392 7393 ackskew = dst->seqlo - ack; 7394 7395 /* 7396 * Need to demodulate the sequence numbers in any TCP SACK options 7397 * (Selective ACK). We could optionally validate the SACK values 7398 * against the current ACK window, either forwards or backwards, but 7399 * I'm not confident that SACK has been implemented properly 7400 * everywhere. It wouldn't surprise me if several stacks accidentally 7401 * SACK too far backwards of previously ACKed data. There really aren't 7402 * any security implications of bad SACKing unless the target stack 7403 * doesn't validate the option length correctly. Someone trying to 7404 * spoof into a TCP connection won't bother blindly sending SACK 7405 * options anyway. 7406 */ 7407 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 7408 if (pf_modulate_sack(pd, th, dst)) 7409 *copyback = 1; 7410 } 7411 7412 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 7413 if (SEQ_GEQ(src->seqhi, data_end) && 7414 /* Last octet inside other's window space */ 7415 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 7416 /* Retrans: not more than one window back */ 7417 (ackskew >= -MAXACKWINDOW) && 7418 /* Acking not more than one reassembled fragment backwards */ 7419 (ackskew <= (MAXACKWINDOW << sws)) && 7420 /* Acking not more than one window forward */ 7421 ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo || 7422 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 7423 /* Require an exact/+1 sequence match on resets when possible */ 7424 (SEQ_GEQ(orig_seq, src->seqlo - (dst->max_win << dws)) && 7425 SEQ_LEQ(orig_seq, src->seqlo + 1) && ackskew == 0 && 7426 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)))) { 7427 /* Allow resets to match sequence window if ack is perfect match */ 7428 7429 if (dst->scrub || src->scrub) { 7430 if (pf_normalize_tcp_stateful(pd, reason, th, 7431 state, src, dst, copyback)) 7432 return (PF_DROP); 7433 } 7434 7435 /* update max window */ 7436 if (src->max_win < win) 7437 src->max_win = win; 7438 /* synchronize sequencing */ 7439 if (SEQ_GT(end, src->seqlo)) 7440 src->seqlo = end; 7441 /* slide the window of what the other end can send */ 7442 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 7443 dst->seqhi = ack + MAX((win << sws), 1); 7444 7445 /* update states */ 7446 if (tcp_get_flags(th) & TH_SYN) 7447 if (src->state < TCPS_SYN_SENT) 7448 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 7449 if (tcp_get_flags(th) & TH_FIN) 7450 if (src->state < TCPS_CLOSING) 7451 pf_set_protostate(state, psrc, TCPS_CLOSING); 7452 if (tcp_get_flags(th) & TH_ACK) { 7453 if (dst->state == TCPS_SYN_SENT) { 7454 pf_set_protostate(state, pdst, 7455 TCPS_ESTABLISHED); 7456 if (src->state == TCPS_ESTABLISHED && 7457 state->sns[PF_SN_LIMIT] != NULL && 7458 pf_src_connlimit(state)) { 7459 REASON_SET(reason, PFRES_SRCLIMIT); 7460 return (PF_DROP); 7461 } 7462 } else if (dst->state == TCPS_CLOSING) 7463 pf_set_protostate(state, pdst, 7464 TCPS_FIN_WAIT_2); 7465 } 7466 if (tcp_get_flags(th) & TH_RST) 7467 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 7468 7469 /* update expire time */ 7470 state->expire = pf_get_uptime(); 7471 if (src->state >= TCPS_FIN_WAIT_2 && 7472 dst->state >= TCPS_FIN_WAIT_2) 7473 state->timeout = PFTM_TCP_CLOSED; 7474 else if (src->state >= TCPS_CLOSING && 7475 dst->state >= TCPS_CLOSING) 7476 state->timeout = PFTM_TCP_FIN_WAIT; 7477 else if (src->state < TCPS_ESTABLISHED || 7478 dst->state < TCPS_ESTABLISHED) 7479 state->timeout = PFTM_TCP_OPENING; 7480 else if (src->state >= TCPS_CLOSING || 7481 dst->state >= TCPS_CLOSING) 7482 state->timeout = PFTM_TCP_CLOSING; 7483 else 7484 state->timeout = PFTM_TCP_ESTABLISHED; 7485 7486 /* Fall through to PASS packet */ 7487 7488 } else if ((dst->state < TCPS_SYN_SENT || 7489 dst->state >= TCPS_FIN_WAIT_2 || 7490 src->state >= TCPS_FIN_WAIT_2) && 7491 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 7492 /* Within a window forward of the originating packet */ 7493 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 7494 /* Within a window backward of the originating packet */ 7495 7496 /* 7497 * This currently handles three situations: 7498 * 1) Stupid stacks will shotgun SYNs before their peer 7499 * replies. 7500 * 2) When PF catches an already established stream (the 7501 * firewall rebooted, the state table was flushed, routes 7502 * changed...) 7503 * 3) Packets get funky immediately after the connection 7504 * closes (this should catch Solaris spurious ACK|FINs 7505 * that web servers like to spew after a close) 7506 * 7507 * This must be a little more careful than the above code 7508 * since packet floods will also be caught here. We don't 7509 * update the TTL here to mitigate the damage of a packet 7510 * flood and so the same code can handle awkward establishment 7511 * and a loosened connection close. 7512 * In the establishment case, a correct peer response will 7513 * validate the connection, go through the normal state code 7514 * and keep updating the state TTL. 7515 */ 7516 7517 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7518 printf("pf: loose state match: "); 7519 pf_print_state(state); 7520 pf_print_flags(tcp_get_flags(th)); 7521 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 7522 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 7523 pd->p_len, ackskew, (unsigned long long)state->packets[0], 7524 (unsigned long long)state->packets[1], 7525 pd->dir == PF_IN ? "in" : "out", 7526 pd->dir == state->direction ? "fwd" : "rev"); 7527 } 7528 7529 if (dst->scrub || src->scrub) { 7530 if (pf_normalize_tcp_stateful(pd, reason, th, 7531 state, src, dst, copyback)) 7532 return (PF_DROP); 7533 } 7534 7535 /* update max window */ 7536 if (src->max_win < win) 7537 src->max_win = win; 7538 /* synchronize sequencing */ 7539 if (SEQ_GT(end, src->seqlo)) 7540 src->seqlo = end; 7541 /* slide the window of what the other end can send */ 7542 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 7543 dst->seqhi = ack + MAX((win << sws), 1); 7544 7545 /* 7546 * Cannot set dst->seqhi here since this could be a shotgunned 7547 * SYN and not an already established connection. 7548 */ 7549 7550 if (tcp_get_flags(th) & TH_FIN) 7551 if (src->state < TCPS_CLOSING) 7552 pf_set_protostate(state, psrc, TCPS_CLOSING); 7553 if (tcp_get_flags(th) & TH_RST) 7554 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 7555 7556 /* Fall through to PASS packet */ 7557 7558 } else { 7559 if (state->dst.state == TCPS_SYN_SENT && 7560 state->src.state == TCPS_SYN_SENT) { 7561 /* Send RST for state mismatches during handshake */ 7562 if (!(tcp_get_flags(th) & TH_RST)) 7563 pf_send_tcp(state->rule, pd->af, 7564 pd->dst, pd->src, th->th_dport, 7565 th->th_sport, ntohl(th->th_ack), 0, 7566 TH_RST, 0, 0, 7567 state->rule->return_ttl, M_SKIP_FIREWALL, 7568 0, 0, state->act.rtableid, reason); 7569 src->seqlo = 0; 7570 src->seqhi = 1; 7571 src->max_win = 1; 7572 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 7573 printf("pf: BAD state: "); 7574 pf_print_state(state); 7575 pf_print_flags(tcp_get_flags(th)); 7576 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 7577 "pkts=%llu:%llu dir=%s,%s\n", 7578 seq, orig_seq, ack, pd->p_len, ackskew, 7579 (unsigned long long)state->packets[0], 7580 (unsigned long long)state->packets[1], 7581 pd->dir == PF_IN ? "in" : "out", 7582 pd->dir == state->direction ? "fwd" : "rev"); 7583 printf("pf: State failure on: %c %c %c %c | %c %c\n", 7584 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 7585 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 7586 ' ': '2', 7587 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 7588 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 7589 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 7590 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 7591 } 7592 REASON_SET(reason, PFRES_BADSTATE); 7593 return (PF_DROP); 7594 } 7595 7596 return (PF_PASS); 7597 } 7598 7599 static int 7600 pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd, 7601 u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst, 7602 u_int8_t psrc, u_int8_t pdst) 7603 { 7604 struct tcphdr *th = &pd->hdr.tcp; 7605 7606 if (tcp_get_flags(th) & TH_SYN) 7607 if (src->state < TCPS_SYN_SENT) 7608 pf_set_protostate(state, psrc, TCPS_SYN_SENT); 7609 if (tcp_get_flags(th) & TH_FIN) 7610 if (src->state < TCPS_CLOSING) 7611 pf_set_protostate(state, psrc, TCPS_CLOSING); 7612 if (tcp_get_flags(th) & TH_ACK) { 7613 if (dst->state == TCPS_SYN_SENT) { 7614 pf_set_protostate(state, pdst, TCPS_ESTABLISHED); 7615 if (src->state == TCPS_ESTABLISHED && 7616 state->sns[PF_SN_LIMIT] != NULL && 7617 pf_src_connlimit(state)) { 7618 REASON_SET(reason, PFRES_SRCLIMIT); 7619 return (PF_DROP); 7620 } 7621 } else if (dst->state == TCPS_CLOSING) { 7622 pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2); 7623 } else if (src->state == TCPS_SYN_SENT && 7624 dst->state < TCPS_SYN_SENT) { 7625 /* 7626 * Handle a special sloppy case where we only see one 7627 * half of the connection. If there is a ACK after 7628 * the initial SYN without ever seeing a packet from 7629 * the destination, set the connection to established. 7630 */ 7631 pf_set_protostate(state, PF_PEER_BOTH, 7632 TCPS_ESTABLISHED); 7633 dst->state = src->state = TCPS_ESTABLISHED; 7634 if (state->sns[PF_SN_LIMIT] != NULL && 7635 pf_src_connlimit(state)) { 7636 REASON_SET(reason, PFRES_SRCLIMIT); 7637 return (PF_DROP); 7638 } 7639 } else if (src->state == TCPS_CLOSING && 7640 dst->state == TCPS_ESTABLISHED && 7641 dst->seqlo == 0) { 7642 /* 7643 * Handle the closing of half connections where we 7644 * don't see the full bidirectional FIN/ACK+ACK 7645 * handshake. 7646 */ 7647 pf_set_protostate(state, pdst, TCPS_CLOSING); 7648 } 7649 } 7650 if (tcp_get_flags(th) & TH_RST) 7651 pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT); 7652 7653 /* update expire time */ 7654 state->expire = pf_get_uptime(); 7655 if (src->state >= TCPS_FIN_WAIT_2 && 7656 dst->state >= TCPS_FIN_WAIT_2) 7657 state->timeout = PFTM_TCP_CLOSED; 7658 else if (src->state >= TCPS_CLOSING && 7659 dst->state >= TCPS_CLOSING) 7660 state->timeout = PFTM_TCP_FIN_WAIT; 7661 else if (src->state < TCPS_ESTABLISHED || 7662 dst->state < TCPS_ESTABLISHED) 7663 state->timeout = PFTM_TCP_OPENING; 7664 else if (src->state >= TCPS_CLOSING || 7665 dst->state >= TCPS_CLOSING) 7666 state->timeout = PFTM_TCP_CLOSING; 7667 else 7668 state->timeout = PFTM_TCP_ESTABLISHED; 7669 7670 return (PF_PASS); 7671 } 7672 7673 static int 7674 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason) 7675 { 7676 struct pf_state_key *sk = state->key[pd->didx]; 7677 struct tcphdr *th = &pd->hdr.tcp; 7678 7679 if (state->src.state == PF_TCPS_PROXY_SRC) { 7680 if (pd->dir != state->direction) { 7681 REASON_SET(reason, PFRES_SYNPROXY); 7682 return (PF_SYNPROXY_DROP); 7683 } 7684 if (tcp_get_flags(th) & TH_SYN) { 7685 if (ntohl(th->th_seq) != state->src.seqlo) { 7686 REASON_SET(reason, PFRES_SYNPROXY); 7687 return (PF_DROP); 7688 } 7689 pf_send_tcp(state->rule, pd->af, pd->dst, 7690 pd->src, th->th_dport, th->th_sport, 7691 state->src.seqhi, ntohl(th->th_seq) + 1, 7692 TH_SYN|TH_ACK, 0, state->src.mss, 0, 7693 M_SKIP_FIREWALL, 0, 0, state->act.rtableid, 7694 reason); 7695 REASON_SET(reason, PFRES_SYNPROXY); 7696 return (PF_SYNPROXY_DROP); 7697 } else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 7698 (ntohl(th->th_ack) != state->src.seqhi + 1) || 7699 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 7700 REASON_SET(reason, PFRES_SYNPROXY); 7701 return (PF_DROP); 7702 } else if (state->sns[PF_SN_LIMIT] != NULL && 7703 pf_src_connlimit(state)) { 7704 REASON_SET(reason, PFRES_SRCLIMIT); 7705 return (PF_DROP); 7706 } else 7707 pf_set_protostate(state, PF_PEER_SRC, 7708 PF_TCPS_PROXY_DST); 7709 } 7710 if (state->src.state == PF_TCPS_PROXY_DST) { 7711 if (pd->dir == state->direction) { 7712 if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) || 7713 (ntohl(th->th_ack) != state->src.seqhi + 1) || 7714 (ntohl(th->th_seq) != state->src.seqlo + 1)) { 7715 REASON_SET(reason, PFRES_SYNPROXY); 7716 return (PF_DROP); 7717 } 7718 state->src.max_win = MAX(ntohs(th->th_win), 1); 7719 if (state->dst.seqhi == 1) 7720 state->dst.seqhi = arc4random(); 7721 pf_send_tcp(state->rule, pd->af, 7722 &sk->addr[pd->sidx], &sk->addr[pd->didx], 7723 sk->port[pd->sidx], sk->port[pd->didx], 7724 state->dst.seqhi, 0, TH_SYN, 0, 7725 state->src.mss, 0, 7726 state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 7727 state->tag, 0, state->act.rtableid, 7728 reason); 7729 REASON_SET(reason, PFRES_SYNPROXY); 7730 return (PF_SYNPROXY_DROP); 7731 } else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != 7732 (TH_SYN|TH_ACK)) || 7733 (ntohl(th->th_ack) != state->dst.seqhi + 1)) { 7734 REASON_SET(reason, PFRES_SYNPROXY); 7735 return (PF_DROP); 7736 } else { 7737 state->dst.max_win = MAX(ntohs(th->th_win), 1); 7738 state->dst.seqlo = ntohl(th->th_seq); 7739 pf_send_tcp(state->rule, pd->af, pd->dst, 7740 pd->src, th->th_dport, th->th_sport, 7741 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 7742 TH_ACK, state->src.max_win, 0, 0, 0, 7743 state->tag, 0, state->act.rtableid, 7744 reason); 7745 pf_send_tcp(state->rule, pd->af, 7746 &sk->addr[pd->sidx], &sk->addr[pd->didx], 7747 sk->port[pd->sidx], sk->port[pd->didx], 7748 state->src.seqhi + 1, state->src.seqlo + 1, 7749 TH_ACK, state->dst.max_win, 0, 0, 7750 M_SKIP_FIREWALL, 0, 0, state->act.rtableid, 7751 reason); 7752 state->src.seqdiff = state->dst.seqhi - 7753 state->src.seqlo; 7754 state->dst.seqdiff = state->src.seqhi - 7755 state->dst.seqlo; 7756 state->src.seqhi = state->src.seqlo + 7757 state->dst.max_win; 7758 state->dst.seqhi = state->dst.seqlo + 7759 state->src.max_win; 7760 state->src.wscale = state->dst.wscale = 0; 7761 pf_set_protostate(state, PF_PEER_BOTH, 7762 TCPS_ESTABLISHED); 7763 REASON_SET(reason, PFRES_SYNPROXY); 7764 return (PF_SYNPROXY_DROP); 7765 } 7766 } 7767 7768 return (PF_PASS); 7769 } 7770 7771 static __inline int 7772 pf_synproxy_ack(struct pf_krule *r, struct pf_pdesc *pd, struct pf_kstate **sm, 7773 struct pf_rule_actions *act) 7774 { 7775 struct tcphdr *th = &pd->hdr.tcp; 7776 struct pf_kstate *s; 7777 u_int16_t mss; 7778 int rtid; 7779 u_short reason; 7780 7781 if ((th->th_flags & (TH_SYN | TH_ACK)) != TH_SYN) 7782 return (PF_PASS); 7783 7784 s = *sm; 7785 rtid = act->rtableid; 7786 7787 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 7788 s->src.seqhi = arc4random(); 7789 /* Find mss option */ 7790 mss = pf_get_mss(pd); 7791 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 7792 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 7793 s->src.mss = mss; 7794 7795 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 7796 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 7797 TH_SYN | TH_ACK, 0, s->src.mss, 0, 1, 0, 0, r->rtableid, NULL); 7798 7799 REASON_SET(&reason, PFRES_SYNPROXY); 7800 return (PF_SYNPROXY_DROP); 7801 } 7802 7803 static int 7804 pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 7805 { 7806 struct pf_state_key_cmp key; 7807 int copyback = 0; 7808 struct pf_state_peer *src, *dst; 7809 uint8_t psrc, pdst; 7810 int action; 7811 7812 bzero(&key, sizeof(key)); 7813 key.af = pd->af; 7814 key.proto = pd->virtual_proto; 7815 pf_addrcpy(&key.addr[pd->sidx], pd->src, key.af); 7816 pf_addrcpy(&key.addr[pd->didx], pd->dst, key.af); 7817 key.port[pd->sidx] = pd->osport; 7818 key.port[pd->didx] = pd->odport; 7819 7820 action = pf_find_state(pd, &key, state); 7821 if (action != PF_MATCH) 7822 return (action); 7823 7824 action = PF_PASS; 7825 if (pd->dir == (*state)->direction) { 7826 if (PF_REVERSED_KEY(*state, pd->af)) { 7827 src = &(*state)->dst; 7828 dst = &(*state)->src; 7829 psrc = PF_PEER_DST; 7830 pdst = PF_PEER_SRC; 7831 } else { 7832 src = &(*state)->src; 7833 dst = &(*state)->dst; 7834 psrc = PF_PEER_SRC; 7835 pdst = PF_PEER_DST; 7836 } 7837 } else { 7838 if (PF_REVERSED_KEY(*state, pd->af)) { 7839 src = &(*state)->src; 7840 dst = &(*state)->dst; 7841 psrc = PF_PEER_SRC; 7842 pdst = PF_PEER_DST; 7843 } else { 7844 src = &(*state)->dst; 7845 dst = &(*state)->src; 7846 psrc = PF_PEER_DST; 7847 pdst = PF_PEER_SRC; 7848 } 7849 } 7850 7851 switch (pd->virtual_proto) { 7852 case IPPROTO_TCP: { 7853 struct tcphdr *th = &pd->hdr.tcp; 7854 7855 if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS) 7856 return (action); 7857 if (((tcp_get_flags(th) & (TH_SYN | TH_ACK)) == TH_SYN) || 7858 ((th->th_flags & (TH_SYN | TH_ACK | TH_RST)) == TH_ACK && 7859 pf_syncookie_check(pd) && pd->dir == PF_IN)) { 7860 if ((*state)->src.state >= TCPS_FIN_WAIT_2 && 7861 (*state)->dst.state >= TCPS_FIN_WAIT_2) { 7862 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7863 printf("pf: state reuse "); 7864 pf_print_state(*state); 7865 pf_print_flags(tcp_get_flags(th)); 7866 printf("\n"); 7867 } 7868 /* XXX make sure it's the same direction ?? */ 7869 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 7870 pf_remove_state(*state); 7871 *state = NULL; 7872 return (PF_DROP); 7873 } else if ((*state)->src.state >= TCPS_ESTABLISHED && 7874 (*state)->dst.state >= TCPS_ESTABLISHED) { 7875 /* 7876 * SYN matches existing state??? 7877 * Typically happens when sender boots up after 7878 * sudden panic. Certain protocols (NFSv3) are 7879 * always using same port numbers. Challenge 7880 * ACK enables all parties (firewall and peers) 7881 * to get in sync again. 7882 */ 7883 pf_send_challenge_ack(pd, *state, src, dst, reason); 7884 return (PF_DROP); 7885 } 7886 } 7887 if ((*state)->state_flags & PFSTATE_SLOPPY) { 7888 if (pf_tcp_track_sloppy(*state, pd, reason, src, dst, 7889 psrc, pdst) == PF_DROP) 7890 return (PF_DROP); 7891 } else { 7892 int ret; 7893 7894 ret = pf_tcp_track_full(*state, pd, reason, 7895 ©back, src, dst, psrc, pdst); 7896 if (ret == PF_DROP) 7897 return (PF_DROP); 7898 } 7899 break; 7900 } 7901 case IPPROTO_UDP: 7902 /* update states */ 7903 if (src->state < PFUDPS_SINGLE) 7904 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 7905 if (dst->state == PFUDPS_SINGLE) 7906 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 7907 7908 /* update expire time */ 7909 (*state)->expire = pf_get_uptime(); 7910 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 7911 (*state)->timeout = PFTM_UDP_MULTIPLE; 7912 else 7913 (*state)->timeout = PFTM_UDP_SINGLE; 7914 break; 7915 case IPPROTO_SCTP: 7916 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 7917 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 7918 pd->sctp_flags & PFDESC_SCTP_INIT) { 7919 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 7920 pf_remove_state(*state); 7921 *state = NULL; 7922 return (PF_DROP); 7923 } 7924 7925 if (pf_sctp_track(*state, pd, reason) != PF_PASS) 7926 return (PF_DROP); 7927 7928 /* Track state. */ 7929 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 7930 if (src->state < SCTP_COOKIE_WAIT) { 7931 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 7932 (*state)->timeout = PFTM_SCTP_OPENING; 7933 } 7934 } 7935 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 7936 MPASS(dst->scrub != NULL); 7937 if (dst->scrub->pfss_v_tag == 0) 7938 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 7939 } 7940 7941 /* 7942 * Bind to the correct interface if we're if-bound. For multihomed 7943 * extra associations we don't know which interface that will be until 7944 * here, so we've inserted the state on V_pf_all. Fix that now. 7945 */ 7946 if ((*state)->kif == V_pfi_all && 7947 (*state)->rule->rule_flag & PFRULE_IFBOUND) 7948 (*state)->kif = pd->kif; 7949 7950 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 7951 if (src->state < SCTP_ESTABLISHED) { 7952 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 7953 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 7954 } 7955 } 7956 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | 7957 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 7958 if (src->state < SCTP_SHUTDOWN_PENDING) { 7959 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 7960 (*state)->timeout = PFTM_SCTP_CLOSING; 7961 } 7962 } 7963 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) { 7964 pf_set_protostate(*state, psrc, SCTP_CLOSED); 7965 (*state)->timeout = PFTM_SCTP_CLOSED; 7966 } 7967 7968 (*state)->expire = pf_get_uptime(); 7969 break; 7970 default: 7971 /* update states */ 7972 if (src->state < PFOTHERS_SINGLE) 7973 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7974 if (dst->state == PFOTHERS_SINGLE) 7975 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7976 7977 /* update expire time */ 7978 (*state)->expire = pf_get_uptime(); 7979 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7980 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7981 else 7982 (*state)->timeout = PFTM_OTHER_SINGLE; 7983 break; 7984 } 7985 7986 /* translate source/destination address, if necessary */ 7987 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7988 struct pf_state_key *nk; 7989 int afto, sidx, didx; 7990 7991 if (PF_REVERSED_KEY(*state, pd->af)) 7992 nk = (*state)->key[pd->sidx]; 7993 else 7994 nk = (*state)->key[pd->didx]; 7995 7996 afto = pd->af != nk->af; 7997 7998 if (afto && (*state)->direction == PF_IN) { 7999 sidx = pd->didx; 8000 didx = pd->sidx; 8001 } else { 8002 sidx = pd->sidx; 8003 didx = pd->didx; 8004 } 8005 8006 if (afto) { 8007 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af); 8008 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af); 8009 pd->naf = nk->af; 8010 action = PF_AFRT; 8011 } 8012 8013 if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) || 8014 nk->port[sidx] != pd->osport) 8015 pf_change_ap(pd, pd->src, pd->sport, 8016 &nk->addr[sidx], nk->port[sidx]); 8017 8018 if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) || 8019 nk->port[didx] != pd->odport) 8020 pf_change_ap(pd, pd->dst, pd->dport, 8021 &nk->addr[didx], nk->port[didx]); 8022 8023 copyback = 1; 8024 } 8025 8026 if (copyback && pd->hdrlen > 0) 8027 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 8028 8029 return (action); 8030 } 8031 8032 static int 8033 pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd, 8034 u_short *reason) 8035 { 8036 struct pf_state_peer *src; 8037 if (pd->dir == state->direction) { 8038 if (PF_REVERSED_KEY(state, pd->af)) 8039 src = &state->dst; 8040 else 8041 src = &state->src; 8042 } else { 8043 if (PF_REVERSED_KEY(state, pd->af)) 8044 src = &state->src; 8045 else 8046 src = &state->dst; 8047 } 8048 8049 if (src->scrub != NULL) { 8050 /* 8051 * Allow tags to be updated, in case of retransmission of 8052 * INIT/INIT_ACK chunks. 8053 **/ 8054 if (src->state <= SCTP_COOKIE_WAIT) 8055 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 8056 else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 8057 return (PF_DROP); 8058 } 8059 8060 return (PF_PASS); 8061 } 8062 8063 static void 8064 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 8065 { 8066 struct pf_sctp_endpoint key; 8067 struct pf_sctp_endpoint *ep; 8068 struct pf_state_key *sks = s->key[PF_SK_STACK]; 8069 struct pf_sctp_source *i, *tmp; 8070 8071 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 8072 return; 8073 8074 PF_SCTP_ENDPOINTS_LOCK(); 8075 8076 key.v_tag = s->dst.scrub->pfss_v_tag; 8077 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 8078 if (ep != NULL) { 8079 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 8080 if (pf_addr_cmp(&i->addr, 8081 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 8082 s->key[PF_SK_WIRE]->af) == 0) { 8083 SDT_PROBE3(pf, sctp, multihome, remove, 8084 key.v_tag, s, i); 8085 TAILQ_REMOVE(&ep->sources, i, entry); 8086 free(i, M_PFTEMP); 8087 break; 8088 } 8089 } 8090 8091 if (TAILQ_EMPTY(&ep->sources)) { 8092 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 8093 free(ep, M_PFTEMP); 8094 } 8095 } 8096 8097 /* Other direction. */ 8098 key.v_tag = s->src.scrub->pfss_v_tag; 8099 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 8100 if (ep != NULL) { 8101 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 8102 if (pf_addr_cmp(&i->addr, 8103 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 8104 s->key[PF_SK_WIRE]->af) == 0) { 8105 SDT_PROBE3(pf, sctp, multihome, remove, 8106 key.v_tag, s, i); 8107 TAILQ_REMOVE(&ep->sources, i, entry); 8108 free(i, M_PFTEMP); 8109 break; 8110 } 8111 } 8112 8113 if (TAILQ_EMPTY(&ep->sources)) { 8114 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 8115 free(ep, M_PFTEMP); 8116 } 8117 } 8118 8119 PF_SCTP_ENDPOINTS_UNLOCK(); 8120 } 8121 8122 static void 8123 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 8124 { 8125 struct pf_sctp_endpoint key = { 8126 .v_tag = v_tag, 8127 }; 8128 struct pf_sctp_source *i; 8129 struct pf_sctp_endpoint *ep; 8130 int count; 8131 8132 PF_SCTP_ENDPOINTS_LOCK(); 8133 8134 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 8135 if (ep == NULL) { 8136 ep = malloc(sizeof(struct pf_sctp_endpoint), 8137 M_PFTEMP, M_NOWAIT); 8138 if (ep == NULL) { 8139 PF_SCTP_ENDPOINTS_UNLOCK(); 8140 return; 8141 } 8142 8143 ep->v_tag = v_tag; 8144 TAILQ_INIT(&ep->sources); 8145 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 8146 } 8147 8148 /* Avoid inserting duplicates. */ 8149 count = 0; 8150 TAILQ_FOREACH(i, &ep->sources, entry) { 8151 count++; 8152 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 8153 PF_SCTP_ENDPOINTS_UNLOCK(); 8154 return; 8155 } 8156 } 8157 8158 /* Limit the number of addresses per endpoint. */ 8159 if (count >= PF_SCTP_MAX_ENDPOINTS) { 8160 PF_SCTP_ENDPOINTS_UNLOCK(); 8161 return; 8162 } 8163 8164 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 8165 if (i == NULL) { 8166 PF_SCTP_ENDPOINTS_UNLOCK(); 8167 return; 8168 } 8169 8170 i->af = pd->af; 8171 memcpy(&i->addr, a, sizeof(*a)); 8172 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 8173 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 8174 8175 PF_SCTP_ENDPOINTS_UNLOCK(); 8176 } 8177 8178 static void 8179 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 8180 struct pf_kstate *s, int action) 8181 { 8182 struct pf_krule_slist match_rules; 8183 struct pf_sctp_multihome_job *j, *tmp; 8184 struct pf_sctp_source *i; 8185 int ret; 8186 struct pf_kstate *sm = NULL; 8187 struct pf_krule *ra = NULL; 8188 struct pf_krule *r = &V_pf_default_rule; 8189 struct pf_kruleset *rs = NULL; 8190 u_short reason; 8191 bool do_extra = true; 8192 8193 PF_RULES_RLOCK_TRACKER; 8194 8195 again: 8196 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 8197 if (s == NULL || action != PF_PASS) 8198 goto free; 8199 8200 /* Confirm we don't recurse here. */ 8201 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 8202 8203 switch (j->op) { 8204 case SCTP_ADD_IP_ADDRESS: { 8205 uint32_t v_tag = pd->sctp_initiate_tag; 8206 8207 if (v_tag == 0) { 8208 if (s->direction == pd->dir) 8209 v_tag = s->src.scrub->pfss_v_tag; 8210 else 8211 v_tag = s->dst.scrub->pfss_v_tag; 8212 } 8213 8214 /* 8215 * Avoid duplicating states. We'll already have 8216 * created a state based on the source address of 8217 * the packet, but SCTP endpoints may also list this 8218 * address again in the INIT(_ACK) parameters. 8219 */ 8220 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 8221 break; 8222 } 8223 8224 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 8225 PF_RULES_RLOCK(); 8226 sm = NULL; 8227 if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) { 8228 j->pd.related_rule = s->rule; 8229 } 8230 SLIST_INIT(&match_rules); 8231 ret = pf_test_rule(&r, &sm, 8232 &j->pd, &ra, &rs, &reason, NULL, &match_rules); 8233 /* 8234 * Nothing to do about match rules, the processed 8235 * packet has already increased the counters. 8236 */ 8237 pf_free_match_rules(&match_rules); 8238 PF_RULES_RUNLOCK(); 8239 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 8240 if (ret != PF_DROP && sm != NULL) { 8241 /* Inherit v_tag values. */ 8242 if (sm->direction == s->direction) { 8243 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 8244 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 8245 } else { 8246 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 8247 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 8248 } 8249 PF_STATE_UNLOCK(sm); 8250 } else { 8251 /* If we try duplicate inserts? */ 8252 break; 8253 } 8254 8255 /* Only add the address if we've actually allowed the state. */ 8256 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 8257 8258 if (! do_extra) { 8259 break; 8260 } 8261 /* 8262 * We need to do this for each of our source addresses. 8263 * Find those based on the verification tag. 8264 */ 8265 struct pf_sctp_endpoint key = { 8266 .v_tag = pd->hdr.sctp.v_tag, 8267 }; 8268 struct pf_sctp_endpoint *ep; 8269 8270 PF_SCTP_ENDPOINTS_LOCK(); 8271 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 8272 if (ep == NULL) { 8273 PF_SCTP_ENDPOINTS_UNLOCK(); 8274 break; 8275 } 8276 MPASS(ep != NULL); 8277 8278 TAILQ_FOREACH(i, &ep->sources, entry) { 8279 struct pf_sctp_multihome_job *nj; 8280 8281 /* SCTP can intermingle IPv4 and IPv6. */ 8282 if (i->af != pd->af) 8283 continue; 8284 8285 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 8286 if (! nj) { 8287 continue; 8288 } 8289 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 8290 memcpy(&nj->src, &j->src, sizeof(nj->src)); 8291 nj->pd.src = &nj->src; 8292 // New destination address! 8293 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 8294 nj->pd.dst = &nj->dst; 8295 nj->pd.m = j->pd.m; 8296 nj->op = j->op; 8297 8298 MPASS(nj->pd.pcksum); 8299 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 8300 } 8301 PF_SCTP_ENDPOINTS_UNLOCK(); 8302 8303 break; 8304 } 8305 case SCTP_DEL_IP_ADDRESS: { 8306 struct pf_state_key_cmp key; 8307 uint8_t psrc; 8308 int action; 8309 8310 bzero(&key, sizeof(key)); 8311 key.af = j->pd.af; 8312 key.proto = IPPROTO_SCTP; 8313 if (j->pd.dir == PF_IN) { /* wire side, straight */ 8314 pf_addrcpy(&key.addr[0], j->pd.src, key.af); 8315 pf_addrcpy(&key.addr[1], j->pd.dst, key.af); 8316 key.port[0] = j->pd.hdr.sctp.src_port; 8317 key.port[1] = j->pd.hdr.sctp.dest_port; 8318 } else { /* stack side, reverse */ 8319 pf_addrcpy(&key.addr[1], j->pd.src, key.af); 8320 pf_addrcpy(&key.addr[0], j->pd.dst, key.af); 8321 key.port[1] = j->pd.hdr.sctp.src_port; 8322 key.port[0] = j->pd.hdr.sctp.dest_port; 8323 } 8324 8325 action = pf_find_state(&j->pd, &key, &sm); 8326 if (action == PF_MATCH) { 8327 PF_STATE_LOCK_ASSERT(sm); 8328 if (j->pd.dir == sm->direction) { 8329 psrc = PF_PEER_SRC; 8330 } else { 8331 psrc = PF_PEER_DST; 8332 } 8333 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 8334 sm->timeout = PFTM_SCTP_CLOSING; 8335 PF_STATE_UNLOCK(sm); 8336 } 8337 break; 8338 default: 8339 panic("Unknown op %#x", j->op); 8340 } 8341 } 8342 8343 free: 8344 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 8345 free(j, M_PFTEMP); 8346 } 8347 8348 /* We may have inserted extra work while processing the list. */ 8349 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 8350 do_extra = false; 8351 goto again; 8352 } 8353 } 8354 8355 static int 8356 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op, bool asconf) 8357 { 8358 int off = 0; 8359 struct pf_sctp_multihome_job *job; 8360 8361 SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op); 8362 8363 while (off < len) { 8364 struct sctp_paramhdr h; 8365 8366 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, 8367 pd->af)) 8368 return (PF_DROP); 8369 8370 /* Parameters are at least 4 bytes. */ 8371 if (ntohs(h.param_length) < 4) 8372 return (PF_DROP); 8373 8374 SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type), 8375 ntohs(h.param_length)); 8376 8377 switch (ntohs(h.param_type)) { 8378 case SCTP_IPV4_ADDRESS: { 8379 struct in_addr t; 8380 8381 if (ntohs(h.param_length) != 8382 (sizeof(struct sctp_paramhdr) + sizeof(t))) 8383 return (PF_DROP); 8384 8385 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 8386 NULL, pd->af)) 8387 return (PF_DROP); 8388 8389 if (in_nullhost(t)) 8390 t.s_addr = pd->src->v4.s_addr; 8391 8392 /* 8393 * We hold the state lock (idhash) here, which means 8394 * that we can't acquire the keyhash, or we'll get a 8395 * LOR (and potentially double-lock things too). We also 8396 * can't release the state lock here, so instead we'll 8397 * enqueue this for async handling. 8398 * There's a relatively small race here, in that a 8399 * packet using the new addresses could arrive already, 8400 * but that's just though luck for it. 8401 */ 8402 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 8403 if (! job) 8404 return (PF_DROP); 8405 8406 SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op); 8407 8408 memcpy(&job->pd, pd, sizeof(*pd)); 8409 8410 // New source address! 8411 memcpy(&job->src, &t, sizeof(t)); 8412 job->pd.src = &job->src; 8413 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 8414 job->pd.dst = &job->dst; 8415 job->pd.m = pd->m; 8416 job->op = op; 8417 8418 MPASS(job->pd.pcksum); 8419 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 8420 break; 8421 } 8422 #ifdef INET6 8423 case SCTP_IPV6_ADDRESS: { 8424 struct in6_addr t; 8425 8426 if (ntohs(h.param_length) != 8427 (sizeof(struct sctp_paramhdr) + sizeof(t))) 8428 return (PF_DROP); 8429 8430 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 8431 NULL, pd->af)) 8432 return (PF_DROP); 8433 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 8434 break; 8435 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 8436 memcpy(&t, &pd->src->v6, sizeof(t)); 8437 8438 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 8439 if (! job) 8440 return (PF_DROP); 8441 8442 SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op); 8443 8444 memcpy(&job->pd, pd, sizeof(*pd)); 8445 memcpy(&job->src, &t, sizeof(t)); 8446 job->pd.src = &job->src; 8447 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 8448 job->pd.dst = &job->dst; 8449 job->pd.m = pd->m; 8450 job->op = op; 8451 8452 MPASS(job->pd.pcksum); 8453 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 8454 break; 8455 } 8456 #endif /* INET6 */ 8457 case SCTP_ADD_IP_ADDRESS: { 8458 int ret; 8459 struct sctp_asconf_paramhdr ah; 8460 8461 if (asconf) 8462 return (PF_DROP); 8463 8464 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 8465 NULL, pd->af)) 8466 return (PF_DROP); 8467 8468 if (ntohs(ah.ph.param_length) < sizeof(ah)) 8469 return (PF_DROP); 8470 8471 ret = pf_multihome_scan(start + off + sizeof(ah), 8472 ntohs(ah.ph.param_length) - sizeof(ah), pd, 8473 SCTP_ADD_IP_ADDRESS, true); 8474 if (ret != PF_PASS) 8475 return (ret); 8476 break; 8477 } 8478 case SCTP_DEL_IP_ADDRESS: { 8479 int ret; 8480 struct sctp_asconf_paramhdr ah; 8481 8482 if (asconf) 8483 return (PF_DROP); 8484 8485 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 8486 NULL, pd->af)) 8487 return (PF_DROP); 8488 8489 if (ntohs(ah.ph.param_length) < sizeof(ah)) 8490 return (PF_DROP); 8491 8492 ret = pf_multihome_scan(start + off + sizeof(ah), 8493 ntohs(ah.ph.param_length) - sizeof(ah), pd, 8494 SCTP_DEL_IP_ADDRESS, true); 8495 if (ret != PF_PASS) 8496 return (ret); 8497 break; 8498 } 8499 default: 8500 break; 8501 } 8502 8503 off += roundup(ntohs(h.param_length), 4); 8504 } 8505 8506 return (PF_PASS); 8507 } 8508 8509 int 8510 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 8511 { 8512 start += sizeof(struct sctp_init_chunk); 8513 len -= sizeof(struct sctp_init_chunk); 8514 8515 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS, false)); 8516 } 8517 8518 int 8519 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 8520 { 8521 start += sizeof(struct sctp_asconf_chunk); 8522 len -= sizeof(struct sctp_asconf_chunk); 8523 8524 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS, false)); 8525 } 8526 8527 int 8528 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 8529 struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir, 8530 int *iidx, int multi, int inner) 8531 { 8532 int action, direction = pd->dir; 8533 8534 key->af = pd->af; 8535 key->proto = pd->proto; 8536 if (icmp_dir == PF_IN) { 8537 *iidx = pd->sidx; 8538 key->port[pd->sidx] = icmpid; 8539 key->port[pd->didx] = type; 8540 } else { 8541 *iidx = pd->didx; 8542 key->port[pd->sidx] = type; 8543 key->port[pd->didx] = icmpid; 8544 } 8545 if (pf_state_key_addr_setup(pd, key, multi)) 8546 return (PF_DROP); 8547 8548 action = pf_find_state(pd, key, state); 8549 if (action != PF_MATCH) 8550 return (action); 8551 8552 if ((*state)->state_flags & PFSTATE_SLOPPY) 8553 return (-1); 8554 8555 /* Is this ICMP message flowing in right direction? */ 8556 if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af) 8557 direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ? 8558 PF_IN : PF_OUT; 8559 else 8560 direction = (*state)->direction; 8561 if ((*state)->rule->type && 8562 (((!inner && direction == pd->dir) || 8563 (inner && direction != pd->dir)) ? 8564 PF_IN : PF_OUT) != icmp_dir) { 8565 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8566 printf("pf: icmp type %d in wrong direction (%d): ", 8567 ntohs(type), icmp_dir); 8568 pf_print_state(*state); 8569 printf("\n"); 8570 } 8571 PF_STATE_UNLOCK(*state); 8572 *state = NULL; 8573 return (PF_DROP); 8574 } 8575 return (-1); 8576 } 8577 8578 static int 8579 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 8580 u_short *reason) 8581 { 8582 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 8583 u_int16_t *icmpsum, virtual_id, virtual_type; 8584 u_int8_t icmptype, icmpcode; 8585 int icmp_dir, iidx, ret; 8586 struct pf_state_key_cmp key; 8587 #ifdef INET 8588 u_int16_t icmpid; 8589 #endif /* INET*/ 8590 8591 MPASS(*state == NULL); 8592 8593 bzero(&key, sizeof(key)); 8594 switch (pd->proto) { 8595 #ifdef INET 8596 case IPPROTO_ICMP: 8597 icmptype = pd->hdr.icmp.icmp_type; 8598 icmpcode = pd->hdr.icmp.icmp_code; 8599 icmpid = pd->hdr.icmp.icmp_id; 8600 icmpsum = &pd->hdr.icmp.icmp_cksum; 8601 break; 8602 #endif /* INET */ 8603 #ifdef INET6 8604 case IPPROTO_ICMPV6: 8605 icmptype = pd->hdr.icmp6.icmp6_type; 8606 icmpcode = pd->hdr.icmp6.icmp6_code; 8607 #ifdef INET 8608 icmpid = pd->hdr.icmp6.icmp6_id; 8609 #endif /* INET */ 8610 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 8611 break; 8612 #endif /* INET6 */ 8613 default: 8614 panic("unhandled proto %d", pd->proto); 8615 } 8616 8617 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id, 8618 &virtual_type) == 0) { 8619 /* 8620 * ICMP query/reply message not related to a TCP/UDP/SCTP 8621 * packet. Search for an ICMP state. 8622 */ 8623 ret = pf_icmp_state_lookup(&key, pd, state, virtual_id, 8624 virtual_type, icmp_dir, &iidx, 0, 0); 8625 /* IPv6? try matching a multicast address */ 8626 if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) { 8627 MPASS(*state == NULL); 8628 ret = pf_icmp_state_lookup(&key, pd, state, 8629 virtual_id, virtual_type, 8630 icmp_dir, &iidx, 1, 0); 8631 } 8632 if (ret >= 0) { 8633 MPASS(*state == NULL); 8634 return (ret); 8635 } 8636 8637 (*state)->expire = pf_get_uptime(); 8638 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 8639 8640 /* translate source/destination address, if necessary */ 8641 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 8642 struct pf_state_key *nk; 8643 int afto, sidx, didx; 8644 8645 if (PF_REVERSED_KEY(*state, pd->af)) 8646 nk = (*state)->key[pd->sidx]; 8647 else 8648 nk = (*state)->key[pd->didx]; 8649 8650 afto = pd->af != nk->af; 8651 8652 if (afto && (*state)->direction == PF_IN) { 8653 sidx = pd->didx; 8654 didx = pd->sidx; 8655 iidx = !iidx; 8656 } else { 8657 sidx = pd->sidx; 8658 didx = pd->didx; 8659 } 8660 8661 switch (pd->af) { 8662 #ifdef INET 8663 case AF_INET: 8664 #ifdef INET6 8665 if (afto) { 8666 if (pf_translate_icmp_af(AF_INET6, 8667 &pd->hdr.icmp)) 8668 return (PF_DROP); 8669 pd->proto = IPPROTO_ICMPV6; 8670 } 8671 #endif /* INET6 */ 8672 if (!afto && 8673 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET)) 8674 pf_change_a(&saddr->v4.s_addr, 8675 pd->ip_sum, 8676 nk->addr[sidx].v4.s_addr, 8677 0); 8678 8679 if (!afto && PF_ANEQ(pd->dst, 8680 &nk->addr[didx], AF_INET)) 8681 pf_change_a(&daddr->v4.s_addr, 8682 pd->ip_sum, 8683 nk->addr[didx].v4.s_addr, 0); 8684 8685 if (nk->port[iidx] != 8686 pd->hdr.icmp.icmp_id) { 8687 pd->hdr.icmp.icmp_cksum = 8688 pf_cksum_fixup( 8689 pd->hdr.icmp.icmp_cksum, icmpid, 8690 nk->port[iidx], 0); 8691 pd->hdr.icmp.icmp_id = 8692 nk->port[iidx]; 8693 } 8694 8695 m_copyback(pd->m, pd->off, ICMP_MINLEN, 8696 (caddr_t )&pd->hdr.icmp); 8697 break; 8698 #endif /* INET */ 8699 #ifdef INET6 8700 case AF_INET6: 8701 #ifdef INET 8702 if (afto) { 8703 if (pf_translate_icmp_af(AF_INET, 8704 &pd->hdr.icmp6)) 8705 return (PF_DROP); 8706 pd->proto = IPPROTO_ICMP; 8707 } 8708 #endif /* INET */ 8709 if (!afto && 8710 PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6)) 8711 pf_change_a6(saddr, 8712 &pd->hdr.icmp6.icmp6_cksum, 8713 &nk->addr[sidx], 0); 8714 8715 if (!afto && PF_ANEQ(pd->dst, 8716 &nk->addr[didx], AF_INET6)) 8717 pf_change_a6(daddr, 8718 &pd->hdr.icmp6.icmp6_cksum, 8719 &nk->addr[didx], 0); 8720 8721 if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id) 8722 pd->hdr.icmp6.icmp6_id = 8723 nk->port[iidx]; 8724 8725 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 8726 (caddr_t )&pd->hdr.icmp6); 8727 break; 8728 #endif /* INET6 */ 8729 } 8730 if (afto) { 8731 pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], 8732 nk->af); 8733 pf_addrcpy(&pd->ndaddr, &nk->addr[didx], 8734 nk->af); 8735 pd->naf = nk->af; 8736 return (PF_AFRT); 8737 } 8738 } 8739 return (PF_PASS); 8740 8741 } else { 8742 /* 8743 * ICMP error message in response to a TCP/UDP packet. 8744 * Extract the inner TCP/UDP header and search for that state. 8745 */ 8746 8747 struct pf_pdesc pd2; 8748 bzero(&pd2, sizeof pd2); 8749 #ifdef INET 8750 struct ip h2; 8751 #endif /* INET */ 8752 #ifdef INET6 8753 struct ip6_hdr h2_6; 8754 #endif /* INET6 */ 8755 int ipoff2 = 0; 8756 8757 pd2.af = pd->af; 8758 pd2.dir = pd->dir; 8759 /* Payload packet is from the opposite direction. */ 8760 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 8761 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 8762 pd2.m = pd->m; 8763 pd2.pf_mtag = pd->pf_mtag; 8764 pd2.kif = pd->kif; 8765 switch (pd->af) { 8766 #ifdef INET 8767 case AF_INET: 8768 /* offset of h2 in mbuf chain */ 8769 ipoff2 = pd->off + ICMP_MINLEN; 8770 8771 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 8772 reason, pd2.af)) { 8773 DPFPRINTF(PF_DEBUG_MISC, 8774 "pf: ICMP error message too short " 8775 "(ip)"); 8776 return (PF_DROP); 8777 } 8778 /* 8779 * ICMP error messages don't refer to non-first 8780 * fragments 8781 */ 8782 if (h2.ip_off & htons(IP_OFFMASK)) { 8783 REASON_SET(reason, PFRES_FRAG); 8784 return (PF_DROP); 8785 } 8786 8787 /* offset of protocol header that follows h2 */ 8788 pd2.off = ipoff2; 8789 if (pf_walk_header(&pd2, &h2, reason) != PF_PASS) 8790 return (PF_DROP); 8791 8792 pd2.tot_len = ntohs(h2.ip_len); 8793 pd2.ttl = h2.ip_ttl; 8794 pd2.src = (struct pf_addr *)&h2.ip_src; 8795 pd2.dst = (struct pf_addr *)&h2.ip_dst; 8796 pd2.ip_sum = &h2.ip_sum; 8797 break; 8798 #endif /* INET */ 8799 #ifdef INET6 8800 case AF_INET6: 8801 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 8802 8803 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 8804 reason, pd2.af)) { 8805 DPFPRINTF(PF_DEBUG_MISC, 8806 "pf: ICMP error message too short " 8807 "(ip6)"); 8808 return (PF_DROP); 8809 } 8810 pd2.off = ipoff2; 8811 if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS) 8812 return (PF_DROP); 8813 8814 pd2.tot_len = ntohs(h2_6.ip6_plen) + 8815 sizeof(struct ip6_hdr); 8816 pd2.ttl = h2_6.ip6_hlim; 8817 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 8818 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 8819 pd2.ip_sum = NULL; 8820 break; 8821 #endif /* INET6 */ 8822 default: 8823 unhandled_af(pd->af); 8824 } 8825 8826 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 8827 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8828 printf("pf: BAD ICMP %d:%d outer dst: ", 8829 icmptype, icmpcode); 8830 pf_print_host(pd->src, 0, pd->af); 8831 printf(" -> "); 8832 pf_print_host(pd->dst, 0, pd->af); 8833 printf(" inner src: "); 8834 pf_print_host(pd2.src, 0, pd2.af); 8835 printf(" -> "); 8836 pf_print_host(pd2.dst, 0, pd2.af); 8837 printf("\n"); 8838 } 8839 REASON_SET(reason, PFRES_BADSTATE); 8840 return (PF_DROP); 8841 } 8842 8843 switch (pd2.proto) { 8844 case IPPROTO_TCP: { 8845 struct tcphdr *th = &pd2.hdr.tcp; 8846 u_int32_t seq; 8847 struct pf_state_peer *src, *dst; 8848 u_int8_t dws; 8849 int copyback = 0; 8850 int action; 8851 8852 /* 8853 * Only the first 8 bytes of the TCP header can be 8854 * expected. Don't access any TCP header fields after 8855 * th_seq, an ackskew test is not possible. 8856 */ 8857 if (!pf_pull_hdr(pd->m, pd2.off, th, 8, reason, 8858 pd2.af)) { 8859 DPFPRINTF(PF_DEBUG_MISC, 8860 "pf: ICMP error message too short " 8861 "(tcp)"); 8862 return (PF_DROP); 8863 } 8864 pd2.pcksum = &pd2.hdr.tcp.th_sum; 8865 8866 key.af = pd2.af; 8867 key.proto = IPPROTO_TCP; 8868 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 8869 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 8870 key.port[pd2.sidx] = th->th_sport; 8871 key.port[pd2.didx] = th->th_dport; 8872 8873 action = pf_find_state(&pd2, &key, state); 8874 if (action != PF_MATCH) 8875 return (action); 8876 8877 if (pd->dir == (*state)->direction) { 8878 if (PF_REVERSED_KEY(*state, pd->af)) { 8879 src = &(*state)->src; 8880 dst = &(*state)->dst; 8881 } else { 8882 src = &(*state)->dst; 8883 dst = &(*state)->src; 8884 } 8885 } else { 8886 if (PF_REVERSED_KEY(*state, pd->af)) { 8887 src = &(*state)->dst; 8888 dst = &(*state)->src; 8889 } else { 8890 src = &(*state)->src; 8891 dst = &(*state)->dst; 8892 } 8893 } 8894 8895 if (src->wscale && dst->wscale) 8896 dws = dst->wscale & PF_WSCALE_MASK; 8897 else 8898 dws = 0; 8899 8900 /* Demodulate sequence number */ 8901 seq = ntohl(th->th_seq) - src->seqdiff; 8902 if (src->seqdiff) { 8903 pf_change_a(&th->th_seq, icmpsum, 8904 htonl(seq), 0); 8905 copyback = 1; 8906 } 8907 8908 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 8909 (!SEQ_GEQ(src->seqhi, seq) || 8910 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 8911 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8912 printf("pf: BAD ICMP %d:%d ", 8913 icmptype, icmpcode); 8914 pf_print_host(pd->src, 0, pd->af); 8915 printf(" -> "); 8916 pf_print_host(pd->dst, 0, pd->af); 8917 printf(" state: "); 8918 pf_print_state(*state); 8919 printf(" seq=%u\n", seq); 8920 } 8921 REASON_SET(reason, PFRES_BADSTATE); 8922 return (PF_DROP); 8923 } else { 8924 if (V_pf_status.debug >= PF_DEBUG_MISC) { 8925 printf("pf: OK ICMP %d:%d ", 8926 icmptype, icmpcode); 8927 pf_print_host(pd->src, 0, pd->af); 8928 printf(" -> "); 8929 pf_print_host(pd->dst, 0, pd->af); 8930 printf(" state: "); 8931 pf_print_state(*state); 8932 printf(" seq=%u\n", seq); 8933 } 8934 } 8935 8936 /* translate source/destination address, if necessary */ 8937 if ((*state)->key[PF_SK_WIRE] != 8938 (*state)->key[PF_SK_STACK]) { 8939 8940 struct pf_state_key *nk; 8941 8942 if (PF_REVERSED_KEY(*state, pd->af)) 8943 nk = (*state)->key[pd->sidx]; 8944 else 8945 nk = (*state)->key[pd->didx]; 8946 8947 #if defined(INET) && defined(INET6) 8948 int afto, sidx, didx; 8949 8950 afto = pd->af != nk->af; 8951 8952 if (afto && (*state)->direction == PF_IN) { 8953 sidx = pd2.didx; 8954 didx = pd2.sidx; 8955 } else { 8956 sidx = pd2.sidx; 8957 didx = pd2.didx; 8958 } 8959 8960 if (afto) { 8961 if (pf_translate_icmp_af(nk->af, 8962 &pd->hdr.icmp)) 8963 return (PF_DROP); 8964 m_copyback(pd->m, pd->off, 8965 sizeof(struct icmp6_hdr), 8966 (c_caddr_t)&pd->hdr.icmp6); 8967 if (pf_change_icmp_af(pd->m, ipoff2, pd, 8968 &pd2, &nk->addr[sidx], 8969 &nk->addr[didx], pd->af, 8970 nk->af)) 8971 return (PF_DROP); 8972 pf_addrcpy(&pd->nsaddr, 8973 &nk->addr[pd2.sidx], nk->af); 8974 pf_addrcpy(&pd->ndaddr, 8975 &nk->addr[pd2.didx], nk->af); 8976 if (nk->af == AF_INET) { 8977 pd->proto = IPPROTO_ICMP; 8978 } else { 8979 pd->proto = IPPROTO_ICMPV6; 8980 /* 8981 * IPv4 becomes IPv6 so we must 8982 * copy IPv4 src addr to least 8983 * 32bits in IPv6 address to 8984 * keep traceroute/icmp 8985 * working. 8986 */ 8987 pd->nsaddr.addr32[3] = 8988 pd->src->addr32[0]; 8989 } 8990 pd->naf = pd2.naf = nk->af; 8991 pf_change_ap(&pd2, pd2.src, &th->th_sport, 8992 &nk->addr[pd2.sidx], nk->port[sidx]); 8993 pf_change_ap(&pd2, pd2.dst, &th->th_dport, 8994 &nk->addr[pd2.didx], nk->port[didx]); 8995 m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)th); 8996 return (PF_AFRT); 8997 } 8998 #endif /* INET && INET6 */ 8999 9000 if (PF_ANEQ(pd2.src, 9001 &nk->addr[pd2.sidx], pd2.af) || 9002 nk->port[pd2.sidx] != th->th_sport) 9003 pf_change_icmp(pd2.src, &th->th_sport, 9004 daddr, &nk->addr[pd2.sidx], 9005 nk->port[pd2.sidx], NULL, 9006 pd2.ip_sum, icmpsum, 9007 pd->ip_sum, 0, pd2.af); 9008 9009 if (PF_ANEQ(pd2.dst, 9010 &nk->addr[pd2.didx], pd2.af) || 9011 nk->port[pd2.didx] != th->th_dport) 9012 pf_change_icmp(pd2.dst, &th->th_dport, 9013 saddr, &nk->addr[pd2.didx], 9014 nk->port[pd2.didx], NULL, 9015 pd2.ip_sum, icmpsum, 9016 pd->ip_sum, 0, pd2.af); 9017 copyback = 1; 9018 } 9019 9020 if (copyback) { 9021 switch (pd2.af) { 9022 #ifdef INET 9023 case AF_INET: 9024 m_copyback(pd->m, pd->off, ICMP_MINLEN, 9025 (caddr_t )&pd->hdr.icmp); 9026 m_copyback(pd->m, ipoff2, sizeof(h2), 9027 (caddr_t )&h2); 9028 break; 9029 #endif /* INET */ 9030 #ifdef INET6 9031 case AF_INET6: 9032 m_copyback(pd->m, pd->off, 9033 sizeof(struct icmp6_hdr), 9034 (caddr_t )&pd->hdr.icmp6); 9035 m_copyback(pd->m, ipoff2, sizeof(h2_6), 9036 (caddr_t )&h2_6); 9037 break; 9038 #endif /* INET6 */ 9039 default: 9040 unhandled_af(pd->af); 9041 } 9042 m_copyback(pd->m, pd2.off, 8, (caddr_t)th); 9043 } 9044 9045 return (PF_PASS); 9046 break; 9047 } 9048 case IPPROTO_UDP: { 9049 struct udphdr *uh = &pd2.hdr.udp; 9050 int action; 9051 9052 if (!pf_pull_hdr(pd->m, pd2.off, uh, sizeof(*uh), 9053 reason, pd2.af)) { 9054 DPFPRINTF(PF_DEBUG_MISC, 9055 "pf: ICMP error message too short " 9056 "(udp)"); 9057 return (PF_DROP); 9058 } 9059 pd2.pcksum = &pd2.hdr.udp.uh_sum; 9060 9061 key.af = pd2.af; 9062 key.proto = IPPROTO_UDP; 9063 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 9064 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 9065 key.port[pd2.sidx] = uh->uh_sport; 9066 key.port[pd2.didx] = uh->uh_dport; 9067 9068 action = pf_find_state(&pd2, &key, state); 9069 if (action != PF_MATCH) 9070 return (action); 9071 9072 /* translate source/destination address, if necessary */ 9073 if ((*state)->key[PF_SK_WIRE] != 9074 (*state)->key[PF_SK_STACK]) { 9075 struct pf_state_key *nk; 9076 9077 if (PF_REVERSED_KEY(*state, pd->af)) 9078 nk = (*state)->key[pd->sidx]; 9079 else 9080 nk = (*state)->key[pd->didx]; 9081 9082 #if defined(INET) && defined(INET6) 9083 int afto, sidx, didx; 9084 9085 afto = pd->af != nk->af; 9086 9087 if (afto && (*state)->direction == PF_IN) { 9088 sidx = pd2.didx; 9089 didx = pd2.sidx; 9090 } else { 9091 sidx = pd2.sidx; 9092 didx = pd2.didx; 9093 } 9094 9095 if (afto) { 9096 if (pf_translate_icmp_af(nk->af, 9097 &pd->hdr.icmp)) 9098 return (PF_DROP); 9099 m_copyback(pd->m, pd->off, 9100 sizeof(struct icmp6_hdr), 9101 (c_caddr_t)&pd->hdr.icmp6); 9102 if (pf_change_icmp_af(pd->m, ipoff2, pd, 9103 &pd2, &nk->addr[sidx], 9104 &nk->addr[didx], pd->af, 9105 nk->af)) 9106 return (PF_DROP); 9107 pf_addrcpy(&pd->nsaddr, 9108 &nk->addr[pd2.sidx], nk->af); 9109 pf_addrcpy(&pd->ndaddr, 9110 &nk->addr[pd2.didx], nk->af); 9111 if (nk->af == AF_INET) { 9112 pd->proto = IPPROTO_ICMP; 9113 } else { 9114 pd->proto = IPPROTO_ICMPV6; 9115 /* 9116 * IPv4 becomes IPv6 so we must 9117 * copy IPv4 src addr to least 9118 * 32bits in IPv6 address to 9119 * keep traceroute/icmp 9120 * working. 9121 */ 9122 pd->nsaddr.addr32[3] = 9123 pd->src->addr32[0]; 9124 } 9125 pd->naf = pd2.naf = nk->af; 9126 pf_change_ap(&pd2, pd2.src, &uh->uh_sport, 9127 &nk->addr[pd2.sidx], nk->port[sidx]); 9128 pf_change_ap(&pd2, pd2.dst, &uh->uh_dport, 9129 &nk->addr[pd2.didx], nk->port[didx]); 9130 m_copyback(pd2.m, pd2.off, sizeof(*uh), 9131 (c_caddr_t)uh); 9132 return (PF_AFRT); 9133 } 9134 #endif /* INET && INET6 */ 9135 9136 if (PF_ANEQ(pd2.src, 9137 &nk->addr[pd2.sidx], pd2.af) || 9138 nk->port[pd2.sidx] != uh->uh_sport) 9139 pf_change_icmp(pd2.src, &uh->uh_sport, 9140 daddr, &nk->addr[pd2.sidx], 9141 nk->port[pd2.sidx], &uh->uh_sum, 9142 pd2.ip_sum, icmpsum, 9143 pd->ip_sum, 1, pd2.af); 9144 9145 if (PF_ANEQ(pd2.dst, 9146 &nk->addr[pd2.didx], pd2.af) || 9147 nk->port[pd2.didx] != uh->uh_dport) 9148 pf_change_icmp(pd2.dst, &uh->uh_dport, 9149 saddr, &nk->addr[pd2.didx], 9150 nk->port[pd2.didx], &uh->uh_sum, 9151 pd2.ip_sum, icmpsum, 9152 pd->ip_sum, 1, pd2.af); 9153 9154 switch (pd2.af) { 9155 #ifdef INET 9156 case AF_INET: 9157 m_copyback(pd->m, pd->off, ICMP_MINLEN, 9158 (caddr_t )&pd->hdr.icmp); 9159 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 9160 break; 9161 #endif /* INET */ 9162 #ifdef INET6 9163 case AF_INET6: 9164 m_copyback(pd->m, pd->off, 9165 sizeof(struct icmp6_hdr), 9166 (caddr_t )&pd->hdr.icmp6); 9167 m_copyback(pd->m, ipoff2, sizeof(h2_6), 9168 (caddr_t )&h2_6); 9169 break; 9170 #endif /* INET6 */ 9171 } 9172 m_copyback(pd->m, pd2.off, sizeof(*uh), (caddr_t)uh); 9173 } 9174 return (PF_PASS); 9175 break; 9176 } 9177 #ifdef INET 9178 case IPPROTO_SCTP: { 9179 struct sctphdr *sh = &pd2.hdr.sctp; 9180 struct pf_state_peer *src; 9181 int copyback = 0; 9182 int action; 9183 9184 if (! pf_pull_hdr(pd->m, pd2.off, sh, sizeof(*sh), reason, 9185 pd2.af)) { 9186 DPFPRINTF(PF_DEBUG_MISC, 9187 "pf: ICMP error message too short " 9188 "(sctp)"); 9189 return (PF_DROP); 9190 } 9191 pd2.pcksum = &pd2.sctp_dummy_sum; 9192 9193 key.af = pd2.af; 9194 key.proto = IPPROTO_SCTP; 9195 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 9196 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 9197 key.port[pd2.sidx] = sh->src_port; 9198 key.port[pd2.didx] = sh->dest_port; 9199 9200 action = pf_find_state(&pd2, &key, state); 9201 if (action != PF_MATCH) 9202 return (action); 9203 9204 if (pd->dir == (*state)->direction) { 9205 if (PF_REVERSED_KEY(*state, pd->af)) 9206 src = &(*state)->src; 9207 else 9208 src = &(*state)->dst; 9209 } else { 9210 if (PF_REVERSED_KEY(*state, pd->af)) 9211 src = &(*state)->dst; 9212 else 9213 src = &(*state)->src; 9214 } 9215 9216 if (src->scrub->pfss_v_tag != sh->v_tag) { 9217 DPFPRINTF(PF_DEBUG_MISC, 9218 "pf: ICMP error message has incorrect " 9219 "SCTP v_tag"); 9220 return (PF_DROP); 9221 } 9222 9223 /* translate source/destination address, if necessary */ 9224 if ((*state)->key[PF_SK_WIRE] != 9225 (*state)->key[PF_SK_STACK]) { 9226 9227 struct pf_state_key *nk; 9228 9229 if (PF_REVERSED_KEY(*state, pd->af)) 9230 nk = (*state)->key[pd->sidx]; 9231 else 9232 nk = (*state)->key[pd->didx]; 9233 9234 #if defined(INET) && defined(INET6) 9235 int afto, sidx, didx; 9236 9237 afto = pd->af != nk->af; 9238 9239 if (afto && (*state)->direction == PF_IN) { 9240 sidx = pd2.didx; 9241 didx = pd2.sidx; 9242 } else { 9243 sidx = pd2.sidx; 9244 didx = pd2.didx; 9245 } 9246 9247 if (afto) { 9248 if (pf_translate_icmp_af(nk->af, 9249 &pd->hdr.icmp)) 9250 return (PF_DROP); 9251 m_copyback(pd->m, pd->off, 9252 sizeof(struct icmp6_hdr), 9253 (c_caddr_t)&pd->hdr.icmp6); 9254 if (pf_change_icmp_af(pd->m, ipoff2, pd, 9255 &pd2, &nk->addr[sidx], 9256 &nk->addr[didx], pd->af, 9257 nk->af)) 9258 return (PF_DROP); 9259 sh->src_port = nk->port[sidx]; 9260 sh->dest_port = nk->port[didx]; 9261 m_copyback(pd2.m, pd2.off, sizeof(*sh), (c_caddr_t)sh); 9262 pf_addrcpy(&pd->nsaddr, 9263 &nk->addr[pd2.sidx], nk->af); 9264 pf_addrcpy(&pd->ndaddr, 9265 &nk->addr[pd2.didx], nk->af); 9266 if (nk->af == AF_INET) { 9267 pd->proto = IPPROTO_ICMP; 9268 } else { 9269 pd->proto = IPPROTO_ICMPV6; 9270 /* 9271 * IPv4 becomes IPv6 so we must 9272 * copy IPv4 src addr to least 9273 * 32bits in IPv6 address to 9274 * keep traceroute/icmp 9275 * working. 9276 */ 9277 pd->nsaddr.addr32[3] = 9278 pd->src->addr32[0]; 9279 } 9280 pd->naf = nk->af; 9281 return (PF_AFRT); 9282 } 9283 #endif /* INET && INET6 */ 9284 9285 if (PF_ANEQ(pd2.src, 9286 &nk->addr[pd2.sidx], pd2.af) || 9287 nk->port[pd2.sidx] != sh->src_port) 9288 pf_change_icmp(pd2.src, &sh->src_port, 9289 daddr, &nk->addr[pd2.sidx], 9290 nk->port[pd2.sidx], NULL, 9291 pd2.ip_sum, icmpsum, 9292 pd->ip_sum, 0, pd2.af); 9293 9294 if (PF_ANEQ(pd2.dst, 9295 &nk->addr[pd2.didx], pd2.af) || 9296 nk->port[pd2.didx] != sh->dest_port) 9297 pf_change_icmp(pd2.dst, &sh->dest_port, 9298 saddr, &nk->addr[pd2.didx], 9299 nk->port[pd2.didx], NULL, 9300 pd2.ip_sum, icmpsum, 9301 pd->ip_sum, 0, pd2.af); 9302 copyback = 1; 9303 } 9304 9305 if (copyback) { 9306 switch (pd2.af) { 9307 #ifdef INET 9308 case AF_INET: 9309 m_copyback(pd->m, pd->off, ICMP_MINLEN, 9310 (caddr_t )&pd->hdr.icmp); 9311 m_copyback(pd->m, ipoff2, sizeof(h2), 9312 (caddr_t )&h2); 9313 break; 9314 #endif /* INET */ 9315 #ifdef INET6 9316 case AF_INET6: 9317 m_copyback(pd->m, pd->off, 9318 sizeof(struct icmp6_hdr), 9319 (caddr_t )&pd->hdr.icmp6); 9320 m_copyback(pd->m, ipoff2, sizeof(h2_6), 9321 (caddr_t )&h2_6); 9322 break; 9323 #endif /* INET6 */ 9324 } 9325 m_copyback(pd->m, pd2.off, sizeof(*sh), (caddr_t)sh); 9326 } 9327 9328 return (PF_PASS); 9329 break; 9330 } 9331 case IPPROTO_ICMP: { 9332 struct icmp *iih = &pd2.hdr.icmp; 9333 9334 if (pd2.af != AF_INET) { 9335 REASON_SET(reason, PFRES_NORM); 9336 return (PF_DROP); 9337 } 9338 9339 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 9340 reason, pd2.af)) { 9341 DPFPRINTF(PF_DEBUG_MISC, 9342 "pf: ICMP error message too short i" 9343 "(icmp)"); 9344 return (PF_DROP); 9345 } 9346 pd2.pcksum = &pd2.hdr.icmp.icmp_cksum; 9347 9348 icmpid = iih->icmp_id; 9349 pf_icmp_mapping(&pd2, iih->icmp_type, 9350 &icmp_dir, &virtual_id, &virtual_type); 9351 9352 ret = pf_icmp_state_lookup(&key, &pd2, state, 9353 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 9354 if (ret >= 0) { 9355 MPASS(*state == NULL); 9356 return (ret); 9357 } 9358 9359 /* translate source/destination address, if necessary */ 9360 if ((*state)->key[PF_SK_WIRE] != 9361 (*state)->key[PF_SK_STACK]) { 9362 struct pf_state_key *nk; 9363 9364 if (PF_REVERSED_KEY(*state, pd->af)) 9365 nk = (*state)->key[pd->sidx]; 9366 else 9367 nk = (*state)->key[pd->didx]; 9368 9369 #if defined(INET) && defined(INET6) 9370 int afto, sidx, didx; 9371 9372 afto = pd->af != nk->af; 9373 9374 if (afto && (*state)->direction == PF_IN) { 9375 sidx = pd2.didx; 9376 didx = pd2.sidx; 9377 iidx = !iidx; 9378 } else { 9379 sidx = pd2.sidx; 9380 didx = pd2.didx; 9381 } 9382 9383 if (afto) { 9384 if (nk->af != AF_INET6) 9385 return (PF_DROP); 9386 if (pf_translate_icmp_af(nk->af, 9387 &pd->hdr.icmp)) 9388 return (PF_DROP); 9389 m_copyback(pd->m, pd->off, 9390 sizeof(struct icmp6_hdr), 9391 (c_caddr_t)&pd->hdr.icmp6); 9392 if (pf_change_icmp_af(pd->m, ipoff2, pd, 9393 &pd2, &nk->addr[sidx], 9394 &nk->addr[didx], pd->af, 9395 nk->af)) 9396 return (PF_DROP); 9397 pd->proto = IPPROTO_ICMPV6; 9398 if (pf_translate_icmp_af(nk->af, iih)) 9399 return (PF_DROP); 9400 if (virtual_type == htons(ICMP_ECHO) && 9401 nk->port[iidx] != iih->icmp_id) 9402 iih->icmp_id = nk->port[iidx]; 9403 m_copyback(pd2.m, pd2.off, ICMP_MINLEN, 9404 (c_caddr_t)iih); 9405 pf_addrcpy(&pd->nsaddr, 9406 &nk->addr[pd2.sidx], nk->af); 9407 pf_addrcpy(&pd->ndaddr, 9408 &nk->addr[pd2.didx], nk->af); 9409 /* 9410 * IPv4 becomes IPv6 so we must copy 9411 * IPv4 src addr to least 32bits in 9412 * IPv6 address to keep traceroute 9413 * working. 9414 */ 9415 pd->nsaddr.addr32[3] = 9416 pd->src->addr32[0]; 9417 pd->naf = nk->af; 9418 return (PF_AFRT); 9419 } 9420 #endif /* INET && INET6 */ 9421 9422 if (PF_ANEQ(pd2.src, 9423 &nk->addr[pd2.sidx], pd2.af) || 9424 (virtual_type == htons(ICMP_ECHO) && 9425 nk->port[iidx] != iih->icmp_id)) 9426 pf_change_icmp(pd2.src, 9427 (virtual_type == htons(ICMP_ECHO)) ? 9428 &iih->icmp_id : NULL, 9429 daddr, &nk->addr[pd2.sidx], 9430 (virtual_type == htons(ICMP_ECHO)) ? 9431 nk->port[iidx] : 0, NULL, 9432 pd2.ip_sum, icmpsum, 9433 pd->ip_sum, 0, AF_INET); 9434 9435 if (PF_ANEQ(pd2.dst, 9436 &nk->addr[pd2.didx], pd2.af)) 9437 pf_change_icmp(pd2.dst, NULL, NULL, 9438 &nk->addr[pd2.didx], 0, NULL, 9439 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 9440 AF_INET); 9441 9442 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 9443 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 9444 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 9445 } 9446 return (PF_PASS); 9447 break; 9448 } 9449 #endif /* INET */ 9450 #ifdef INET6 9451 case IPPROTO_ICMPV6: { 9452 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 9453 9454 if (pd2.af != AF_INET6) { 9455 REASON_SET(reason, PFRES_NORM); 9456 return (PF_DROP); 9457 } 9458 9459 if (!pf_pull_hdr(pd->m, pd2.off, iih, 9460 sizeof(struct icmp6_hdr), reason, pd2.af)) { 9461 DPFPRINTF(PF_DEBUG_MISC, 9462 "pf: ICMP error message too short " 9463 "(icmp6)"); 9464 return (PF_DROP); 9465 } 9466 pd2.pcksum = &pd2.hdr.icmp6.icmp6_cksum; 9467 9468 pf_icmp_mapping(&pd2, iih->icmp6_type, 9469 &icmp_dir, &virtual_id, &virtual_type); 9470 9471 ret = pf_icmp_state_lookup(&key, &pd2, state, 9472 virtual_id, virtual_type, icmp_dir, &iidx, 0, 1); 9473 /* IPv6? try matching a multicast address */ 9474 if (ret == PF_DROP && pd2.af == AF_INET6 && 9475 icmp_dir == PF_OUT) { 9476 MPASS(*state == NULL); 9477 ret = pf_icmp_state_lookup(&key, &pd2, 9478 state, virtual_id, virtual_type, 9479 icmp_dir, &iidx, 1, 1); 9480 } 9481 if (ret >= 0) { 9482 MPASS(*state == NULL); 9483 return (ret); 9484 } 9485 9486 /* translate source/destination address, if necessary */ 9487 if ((*state)->key[PF_SK_WIRE] != 9488 (*state)->key[PF_SK_STACK]) { 9489 struct pf_state_key *nk; 9490 9491 if (PF_REVERSED_KEY(*state, pd->af)) 9492 nk = (*state)->key[pd->sidx]; 9493 else 9494 nk = (*state)->key[pd->didx]; 9495 9496 #if defined(INET) && defined(INET6) 9497 int afto, sidx, didx; 9498 9499 afto = pd->af != nk->af; 9500 9501 if (afto && (*state)->direction == PF_IN) { 9502 sidx = pd2.didx; 9503 didx = pd2.sidx; 9504 iidx = !iidx; 9505 } else { 9506 sidx = pd2.sidx; 9507 didx = pd2.didx; 9508 } 9509 9510 if (afto) { 9511 if (nk->af != AF_INET) 9512 return (PF_DROP); 9513 if (pf_translate_icmp_af(nk->af, 9514 &pd->hdr.icmp)) 9515 return (PF_DROP); 9516 m_copyback(pd->m, pd->off, 9517 sizeof(struct icmp6_hdr), 9518 (c_caddr_t)&pd->hdr.icmp6); 9519 if (pf_change_icmp_af(pd->m, ipoff2, pd, 9520 &pd2, &nk->addr[sidx], 9521 &nk->addr[didx], pd->af, 9522 nk->af)) 9523 return (PF_DROP); 9524 pd->proto = IPPROTO_ICMP; 9525 if (pf_translate_icmp_af(nk->af, iih)) 9526 return (PF_DROP); 9527 if (virtual_type == 9528 htons(ICMP6_ECHO_REQUEST) && 9529 nk->port[iidx] != iih->icmp6_id) 9530 iih->icmp6_id = nk->port[iidx]; 9531 m_copyback(pd2.m, pd2.off, 9532 sizeof(struct icmp6_hdr), (c_caddr_t)iih); 9533 pf_addrcpy(&pd->nsaddr, 9534 &nk->addr[pd2.sidx], nk->af); 9535 pf_addrcpy(&pd->ndaddr, 9536 &nk->addr[pd2.didx], nk->af); 9537 pd->naf = nk->af; 9538 return (PF_AFRT); 9539 } 9540 #endif /* INET && INET6 */ 9541 9542 if (PF_ANEQ(pd2.src, 9543 &nk->addr[pd2.sidx], pd2.af) || 9544 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 9545 nk->port[pd2.sidx] != iih->icmp6_id)) 9546 pf_change_icmp(pd2.src, 9547 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 9548 ? &iih->icmp6_id : NULL, 9549 daddr, &nk->addr[pd2.sidx], 9550 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 9551 ? nk->port[iidx] : 0, NULL, 9552 pd2.ip_sum, icmpsum, 9553 pd->ip_sum, 0, AF_INET6); 9554 9555 if (PF_ANEQ(pd2.dst, 9556 &nk->addr[pd2.didx], pd2.af)) 9557 pf_change_icmp(pd2.dst, NULL, NULL, 9558 &nk->addr[pd2.didx], 0, NULL, 9559 pd2.ip_sum, icmpsum, 9560 pd->ip_sum, 0, AF_INET6); 9561 9562 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 9563 (caddr_t)&pd->hdr.icmp6); 9564 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 9565 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 9566 (caddr_t)iih); 9567 } 9568 return (PF_PASS); 9569 break; 9570 } 9571 #endif /* INET6 */ 9572 default: { 9573 int action; 9574 9575 /* 9576 * Placeholder value, so future calls to pf_change_ap() 9577 * don't try to update a NULL checksum pointer. 9578 */ 9579 pd->pcksum = &pd->sctp_dummy_sum; 9580 key.af = pd2.af; 9581 key.proto = pd2.proto; 9582 pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af); 9583 pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af); 9584 key.port[0] = key.port[1] = 0; 9585 9586 action = pf_find_state(&pd2, &key, state); 9587 if (action != PF_MATCH) 9588 return (action); 9589 9590 /* translate source/destination address, if necessary */ 9591 if ((*state)->key[PF_SK_WIRE] != 9592 (*state)->key[PF_SK_STACK]) { 9593 struct pf_state_key *nk = 9594 (*state)->key[pd->didx]; 9595 9596 if (PF_ANEQ(pd2.src, 9597 &nk->addr[pd2.sidx], pd2.af)) 9598 pf_change_icmp(pd2.src, NULL, daddr, 9599 &nk->addr[pd2.sidx], 0, NULL, 9600 pd2.ip_sum, icmpsum, 9601 pd->ip_sum, 0, pd2.af); 9602 9603 if (PF_ANEQ(pd2.dst, 9604 &nk->addr[pd2.didx], pd2.af)) 9605 pf_change_icmp(pd2.dst, NULL, saddr, 9606 &nk->addr[pd2.didx], 0, NULL, 9607 pd2.ip_sum, icmpsum, 9608 pd->ip_sum, 0, pd2.af); 9609 9610 switch (pd2.af) { 9611 #ifdef INET 9612 case AF_INET: 9613 m_copyback(pd->m, pd->off, ICMP_MINLEN, 9614 (caddr_t)&pd->hdr.icmp); 9615 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 9616 break; 9617 #endif /* INET */ 9618 #ifdef INET6 9619 case AF_INET6: 9620 m_copyback(pd->m, pd->off, 9621 sizeof(struct icmp6_hdr), 9622 (caddr_t )&pd->hdr.icmp6); 9623 m_copyback(pd->m, ipoff2, sizeof(h2_6), 9624 (caddr_t )&h2_6); 9625 break; 9626 #endif /* INET6 */ 9627 } 9628 } 9629 return (PF_PASS); 9630 break; 9631 } 9632 } 9633 } 9634 } 9635 9636 /* 9637 * ipoff and off are measured from the start of the mbuf chain. 9638 * h must be at "ipoff" on the mbuf chain. 9639 */ 9640 void * 9641 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 9642 u_short *reasonp, sa_family_t af) 9643 { 9644 int iplen = 0; 9645 switch (af) { 9646 #ifdef INET 9647 case AF_INET: { 9648 const struct ip *h = mtod(m, struct ip *); 9649 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 9650 9651 if (fragoff) { 9652 REASON_SET(reasonp, PFRES_FRAG); 9653 return (NULL); 9654 } 9655 iplen = ntohs(h->ip_len); 9656 break; 9657 } 9658 #endif /* INET */ 9659 #ifdef INET6 9660 case AF_INET6: { 9661 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 9662 9663 iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 9664 break; 9665 } 9666 #endif /* INET6 */ 9667 } 9668 if (m->m_pkthdr.len < off + len || iplen < off + len) { 9669 REASON_SET(reasonp, PFRES_SHORT); 9670 return (NULL); 9671 } 9672 m_copydata(m, off, len, p); 9673 return (p); 9674 } 9675 9676 int 9677 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 9678 int rtableid) 9679 { 9680 struct ifnet *ifp; 9681 9682 /* 9683 * Skip check for addresses with embedded interface scope, 9684 * as they would always match anyway. 9685 */ 9686 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 9687 return (1); 9688 9689 if (af != AF_INET && af != AF_INET6) 9690 return (0); 9691 9692 if (kif == V_pfi_all) 9693 return (1); 9694 9695 /* Skip checks for ipsec interfaces */ 9696 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 9697 return (1); 9698 9699 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 9700 9701 switch (af) { 9702 #ifdef INET6 9703 case AF_INET6: 9704 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 9705 ifp)); 9706 #endif /* INET6 */ 9707 #ifdef INET 9708 case AF_INET: 9709 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 9710 ifp)); 9711 #endif /* INET */ 9712 } 9713 9714 return (0); 9715 } 9716 9717 #ifdef INET 9718 static int 9719 pf_route(struct pf_krule *r, struct ifnet *oifp, 9720 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 9721 { 9722 struct mbuf *m0, *m1, *md; 9723 struct route_in6 ro; 9724 union sockaddr_union rt_gw; 9725 const union sockaddr_union *gw = (const union sockaddr_union *)&ro.ro_dst; 9726 union sockaddr_union *dst; 9727 struct ip *ip; 9728 struct ifnet *ifp = NULL; 9729 int error = 0; 9730 uint16_t ip_len, ip_off; 9731 uint16_t tmp; 9732 int r_dir; 9733 bool skip_test = false; 9734 int action = PF_PASS; 9735 9736 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 9737 9738 SDT_PROBE4(pf, ip, route_to, entry, pd->m, pd, s, oifp); 9739 9740 if (s) { 9741 r_dir = s->direction; 9742 } else { 9743 r_dir = r->direction; 9744 } 9745 9746 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 9747 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 9748 __func__)); 9749 9750 if ((pd->pf_mtag == NULL && 9751 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 9752 pd->pf_mtag->routed++ > 3) { 9753 m0 = pd->m; 9754 pd->m = NULL; 9755 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9756 action = PF_DROP; 9757 goto bad_locked; 9758 } 9759 9760 if (pd->act.rt_kif != NULL) 9761 ifp = pd->act.rt_kif->pfik_ifp; 9762 9763 if (pd->act.rt == PF_DUPTO) { 9764 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 9765 if (s != NULL) { 9766 PF_STATE_UNLOCK(s); 9767 } 9768 if (ifp == oifp) { 9769 /* When the 2nd interface is not skipped */ 9770 return (action); 9771 } else { 9772 m0 = pd->m; 9773 pd->m = NULL; 9774 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9775 action = PF_DROP; 9776 goto bad; 9777 } 9778 } else { 9779 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 9780 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 9781 if (s) 9782 PF_STATE_UNLOCK(s); 9783 return (action); 9784 } 9785 } 9786 } else { 9787 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 9788 if (pd->af == pd->naf) { 9789 pf_dummynet(pd, s, r, &pd->m); 9790 if (s) 9791 PF_STATE_UNLOCK(s); 9792 return (action); 9793 } else { 9794 if (r_dir == PF_IN) { 9795 skip_test = true; 9796 } 9797 } 9798 } 9799 9800 /* 9801 * If we're actually doing route-to and af-to and are in the 9802 * reply direction. 9803 */ 9804 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 9805 pd->af != pd->naf) { 9806 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) { 9807 /* Un-set ifp so we do a plain route lookup. */ 9808 ifp = NULL; 9809 } 9810 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) { 9811 /* Un-set ifp so we do a plain route lookup. */ 9812 ifp = NULL; 9813 } 9814 } 9815 m0 = pd->m; 9816 } 9817 9818 ip = mtod(m0, struct ip *); 9819 9820 bzero(&ro, sizeof(ro)); 9821 dst = (union sockaddr_union *)&ro.ro_dst; 9822 dst->sin.sin_family = AF_INET; 9823 dst->sin.sin_len = sizeof(struct sockaddr_in); 9824 dst->sin.sin_addr = ip->ip_dst; 9825 if (ifp) { /* Only needed in forward direction and route-to */ 9826 bzero(&rt_gw, sizeof(rt_gw)); 9827 ro.ro_flags |= RT_HAS_GW; 9828 gw = &rt_gw; 9829 switch (pd->act.rt_af) { 9830 #ifdef INET 9831 case AF_INET: 9832 rt_gw.sin.sin_family = AF_INET; 9833 rt_gw.sin.sin_len = sizeof(struct sockaddr_in); 9834 rt_gw.sin.sin_addr.s_addr = pd->act.rt_addr.v4.s_addr; 9835 break; 9836 #endif /* INET */ 9837 #ifdef INET6 9838 case AF_INET6: 9839 rt_gw.sin6.sin6_family = AF_INET6; 9840 rt_gw.sin6.sin6_len = sizeof(struct sockaddr_in6); 9841 pf_addrcpy((struct pf_addr *)&rt_gw.sin6.sin6_addr, 9842 &pd->act.rt_addr, AF_INET6); 9843 break; 9844 #endif /* INET6 */ 9845 default: 9846 /* Normal af-to without route-to */ 9847 break; 9848 } 9849 } 9850 9851 if (pd->dir == PF_IN) { 9852 if (ip->ip_ttl <= IPTTLDEC) { 9853 if (r->rt != PF_DUPTO && pd->naf == pd->af) 9854 pf_send_icmp(m0, ICMP_TIMXCEED, 9855 ICMP_TIMXCEED_INTRANS, 0, pd->af, r, 9856 pd->act.rtableid); 9857 action = PF_DROP; 9858 goto bad_locked; 9859 } 9860 ip->ip_ttl -= IPTTLDEC; 9861 } 9862 9863 if (s != NULL) { 9864 if (ifp == NULL && (pd->af != pd->naf)) { 9865 /* We're in the AFTO case. Do a route lookup. */ 9866 const struct nhop_object *nh; 9867 nh = fib4_lookup(M_GETFIB(m0), ip->ip_dst, 0, NHR_NONE, 0); 9868 if (nh) { 9869 ifp = nh->nh_ifp; 9870 9871 /* Use the gateway if needed. */ 9872 if (nh->nh_flags & NHF_GATEWAY) { 9873 gw = (const union sockaddr_union *)&nh->gw_sa; 9874 ro.ro_flags |= RT_HAS_GW; 9875 } else { 9876 dst->sin.sin_addr = ip->ip_dst; 9877 } 9878 } 9879 } 9880 PF_STATE_UNLOCK(s); 9881 } 9882 9883 /* It must have been either set from rt_af or from fib4_lookup */ 9884 KASSERT(gw->sin.sin_family != 0, ("%s: gw address family undetermined", __func__)); 9885 9886 if (ifp == NULL) { 9887 m0 = pd->m; 9888 pd->m = NULL; 9889 action = PF_DROP; 9890 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9891 goto bad; 9892 } 9893 9894 /* 9895 * Bind to the correct interface if we're if-bound. We don't know which 9896 * interface that will be until here, so we've inserted the state 9897 * on V_pf_all. Fix that now. 9898 */ 9899 if (s != NULL && s->kif == V_pfi_all && r->rule_flag & PFRULE_IFBOUND) { 9900 /* Verify that we're here because of BOUND_IFACE */ 9901 MPASS(r->rt == PF_REPLYTO || (pd->af != pd->naf && s->direction == PF_IN)); 9902 s->kif = ifp->if_pf_kif; 9903 if (pd->act.rt == PF_REPLYTO) { 9904 s->orig_kif = oifp->if_pf_kif; 9905 } 9906 } 9907 9908 if (r->rt == PF_DUPTO || (pd->af != pd->naf && s->direction == PF_IN)) 9909 skip_test = true; 9910 9911 if (pd->dir == PF_IN) { 9912 if (skip_test) { 9913 struct pfi_kkif *out_kif = (struct pfi_kkif *)ifp->if_pf_kif; 9914 MPASS(s != NULL); 9915 pf_counter_u64_critical_enter(); 9916 pf_counter_u64_add_protected( 9917 &out_kif->pfik_bytes[pd->naf == AF_INET6][1] 9918 [action != PF_PASS && action != PF_AFRT], pd->tot_len); 9919 pf_counter_u64_add_protected( 9920 &out_kif->pfik_packets[pd->naf == AF_INET6][1] 9921 [action != PF_PASS && action != PF_AFRT], 1); 9922 pf_counter_u64_critical_exit(); 9923 } else { 9924 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 9925 &pd->act) != PF_PASS) { 9926 action = PF_DROP; 9927 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9928 goto bad; 9929 } else if (m0 == NULL) { 9930 action = PF_DROP; 9931 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9932 goto done; 9933 } 9934 if (m0->m_len < sizeof(struct ip)) { 9935 DPFPRINTF(PF_DEBUG_URGENT, 9936 "%s: m0->m_len < sizeof(struct ip)", __func__); 9937 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 9938 action = PF_DROP; 9939 goto bad; 9940 } 9941 ip = mtod(m0, struct ip *); 9942 } 9943 } 9944 9945 if (ifp->if_flags & IFF_LOOPBACK) 9946 m0->m_flags |= M_SKIP_FIREWALL; 9947 9948 ip_len = ntohs(ip->ip_len); 9949 ip_off = ntohs(ip->ip_off); 9950 9951 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 9952 m0->m_pkthdr.csum_flags |= CSUM_IP; 9953 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 9954 in_delayed_cksum(m0); 9955 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 9956 } 9957 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 9958 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 9959 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 9960 } 9961 9962 if (pd->dir == PF_IN) { 9963 /* 9964 * Make sure dummynet gets the correct direction, in case it needs to 9965 * re-inject later. 9966 */ 9967 pd->dir = PF_OUT; 9968 9969 /* 9970 * The following processing is actually the rest of the inbound processing, even 9971 * though we've marked it as outbound (so we don't look through dummynet) and it 9972 * happens after the outbound processing (pf_test(PF_OUT) above). 9973 * Swap the dummynet pipe numbers, because it's going to come to the wrong 9974 * conclusion about what direction it's processing, and we can't fix it or it 9975 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 9976 * decision will pick the right pipe, and everything will mostly work as expected. 9977 */ 9978 tmp = pd->act.dnrpipe; 9979 pd->act.dnrpipe = pd->act.dnpipe; 9980 pd->act.dnpipe = tmp; 9981 } 9982 9983 /* 9984 * If small enough for interface, or the interface will take 9985 * care of the fragmentation for us, we can just send directly. 9986 */ 9987 if (ip_len <= ifp->if_mtu || 9988 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 9989 ip->ip_sum = 0; 9990 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 9991 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 9992 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 9993 } 9994 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 9995 9996 md = m0; 9997 error = pf_dummynet_route(pd, s, r, ifp, 9998 (const struct sockaddr *)gw, &md); 9999 if (md != NULL) { 10000 error = (*ifp->if_output)(ifp, md, 10001 (const struct sockaddr *)gw, (struct route *)&ro); 10002 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 10003 } 10004 goto done; 10005 } 10006 10007 /* Balk when DF bit is set or the interface didn't support TSO. */ 10008 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 10009 error = EMSGSIZE; 10010 KMOD_IPSTAT_INC(ips_cantfrag); 10011 if (pd->act.rt != PF_DUPTO) { 10012 if (s && s->nat_rule != NULL) { 10013 MPASS(m0 == pd->m); 10014 PACKET_UNDO_NAT(pd, 10015 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 10016 s); 10017 } 10018 10019 pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 10020 ifp->if_mtu, pd->af, r, pd->act.rtableid); 10021 } 10022 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 10023 /* Return pass, so we return PFIL_CONSUMED to the stack. */ 10024 action = PF_PASS; 10025 goto bad; 10026 } 10027 10028 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 10029 if (error) { 10030 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 10031 action = PF_DROP; 10032 goto bad; 10033 } 10034 10035 for (; m0; m0 = m1) { 10036 m1 = m0->m_nextpkt; 10037 m0->m_nextpkt = NULL; 10038 if (error == 0) { 10039 m_clrprotoflags(m0); 10040 md = m0; 10041 pd->pf_mtag = pf_find_mtag(md); 10042 error = pf_dummynet_route(pd, s, r, ifp, 10043 (const struct sockaddr *)gw, &md); 10044 if (md != NULL) { 10045 error = (*ifp->if_output)(ifp, md, 10046 (const struct sockaddr *)gw, 10047 (struct route *)&ro); 10048 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 10049 } 10050 } else 10051 m_freem(m0); 10052 } 10053 10054 if (error == 0) 10055 KMOD_IPSTAT_INC(ips_fragmented); 10056 10057 done: 10058 if (pd->act.rt != PF_DUPTO) 10059 pd->m = NULL; 10060 else 10061 action = PF_PASS; 10062 return (action); 10063 10064 bad_locked: 10065 if (s) 10066 PF_STATE_UNLOCK(s); 10067 bad: 10068 m_freem(m0); 10069 goto done; 10070 } 10071 #endif /* INET */ 10072 10073 #ifdef INET6 10074 static int 10075 pf_route6(struct pf_krule *r, struct ifnet *oifp, 10076 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 10077 { 10078 struct mbuf *m0, *md; 10079 struct m_tag *mtag; 10080 struct sockaddr_in6 dst; 10081 struct ip6_hdr *ip6; 10082 struct ifnet *ifp = NULL; 10083 int r_dir; 10084 bool skip_test = false; 10085 int action = PF_PASS; 10086 10087 KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__)); 10088 10089 SDT_PROBE4(pf, ip6, route_to, entry, pd->m, pd, s, oifp); 10090 10091 if (s) { 10092 r_dir = s->direction; 10093 } else { 10094 r_dir = r->direction; 10095 } 10096 10097 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 10098 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 10099 __func__)); 10100 10101 if ((pd->pf_mtag == NULL && 10102 ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) || 10103 pd->pf_mtag->routed++ > 3) { 10104 m0 = pd->m; 10105 pd->m = NULL; 10106 action = PF_DROP; 10107 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10108 goto bad_locked; 10109 } 10110 10111 if (pd->act.rt_kif != NULL) 10112 ifp = pd->act.rt_kif->pfik_ifp; 10113 10114 if (pd->act.rt == PF_DUPTO) { 10115 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 10116 if (s != NULL) { 10117 PF_STATE_UNLOCK(s); 10118 } 10119 if (ifp == oifp) { 10120 /* When the 2nd interface is not skipped */ 10121 return (action); 10122 } else { 10123 m0 = pd->m; 10124 pd->m = NULL; 10125 action = PF_DROP; 10126 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10127 goto bad; 10128 } 10129 } else { 10130 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 10131 if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) { 10132 if (s) 10133 PF_STATE_UNLOCK(s); 10134 return (action); 10135 } 10136 } 10137 } else { 10138 if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) { 10139 if (pd->af == pd->naf) { 10140 pf_dummynet(pd, s, r, &pd->m); 10141 if (s) 10142 PF_STATE_UNLOCK(s); 10143 return (action); 10144 } else { 10145 if (r_dir == PF_IN) { 10146 skip_test = true; 10147 } 10148 } 10149 } 10150 10151 /* 10152 * If we're actually doing route-to and af-to and are in the 10153 * reply direction. 10154 */ 10155 if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp && 10156 pd->af != pd->naf) { 10157 if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) { 10158 /* Un-set ifp so we do a plain route lookup. */ 10159 ifp = NULL; 10160 } 10161 if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) { 10162 /* Un-set ifp so we do a plain route lookup. */ 10163 ifp = NULL; 10164 } 10165 } 10166 m0 = pd->m; 10167 } 10168 10169 ip6 = mtod(m0, struct ip6_hdr *); 10170 10171 bzero(&dst, sizeof(dst)); 10172 dst.sin6_family = AF_INET6; 10173 dst.sin6_len = sizeof(dst); 10174 pf_addrcpy((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr, 10175 AF_INET6); 10176 10177 if (pd->dir == PF_IN) { 10178 if (ip6->ip6_hlim <= IPV6_HLIMDEC) { 10179 if (r->rt != PF_DUPTO && pd->naf == pd->af) 10180 pf_send_icmp(m0, ICMP6_TIME_EXCEEDED, 10181 ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r, 10182 pd->act.rtableid); 10183 action = PF_DROP; 10184 goto bad_locked; 10185 } 10186 ip6->ip6_hlim -= IPV6_HLIMDEC; 10187 } 10188 10189 if (s != NULL) { 10190 if (ifp == NULL && (pd->af != pd->naf)) { 10191 const struct nhop_object *nh; 10192 nh = fib6_lookup(M_GETFIB(m0), &ip6->ip6_dst, 0, NHR_NONE, 0); 10193 if (nh) { 10194 ifp = nh->nh_ifp; 10195 10196 /* Use the gateway if needed. */ 10197 if (nh->nh_flags & NHF_GATEWAY) 10198 bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr, 10199 sizeof(dst.sin6_addr)); 10200 else 10201 dst.sin6_addr = ip6->ip6_dst; 10202 } 10203 } 10204 PF_STATE_UNLOCK(s); 10205 } 10206 10207 if (pd->af != pd->naf) { 10208 struct udphdr *uh = &pd->hdr.udp; 10209 10210 if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) { 10211 uh->uh_sum = in6_cksum_pseudo(ip6, 10212 ntohs(uh->uh_ulen), IPPROTO_UDP, 0); 10213 m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any); 10214 } 10215 } 10216 10217 if (ifp == NULL) { 10218 m0 = pd->m; 10219 pd->m = NULL; 10220 action = PF_DROP; 10221 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10222 goto bad; 10223 } 10224 10225 /* 10226 * Bind to the correct interface if we're if-bound. We don't know which 10227 * interface that will be until here, so we've inserted the state 10228 * on V_pf_all. Fix that now. 10229 */ 10230 if (s != NULL && s->kif == V_pfi_all && r->rule_flag & PFRULE_IFBOUND) { 10231 /* Verify that we're here because of BOUND_IFACE */ 10232 MPASS(r->rt == PF_REPLYTO || (pd->af != pd->naf && s->direction == PF_IN)); 10233 s->kif = ifp->if_pf_kif; 10234 if (pd->act.rt == PF_REPLYTO) { 10235 s->orig_kif = oifp->if_pf_kif; 10236 } 10237 } 10238 10239 if (r->rt == PF_DUPTO || (pd->af != pd->naf && s->direction == PF_IN)) 10240 skip_test = true; 10241 10242 if (pd->dir == PF_IN) { 10243 if (skip_test) { 10244 struct pfi_kkif *out_kif = (struct pfi_kkif *)ifp->if_pf_kif; 10245 MPASS(s != NULL); 10246 pf_counter_u64_critical_enter(); 10247 pf_counter_u64_add_protected( 10248 &out_kif->pfik_bytes[pd->naf == AF_INET6][1] 10249 [action != PF_PASS && action != PF_AFRT], pd->tot_len); 10250 pf_counter_u64_add_protected( 10251 &out_kif->pfik_packets[pd->naf == AF_INET6][1] 10252 [action != PF_PASS && action != PF_AFRT], 1); 10253 pf_counter_u64_critical_exit(); 10254 } else { 10255 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, 10256 ifp, &m0, inp, &pd->act) != PF_PASS) { 10257 action = PF_DROP; 10258 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10259 goto bad; 10260 } else if (m0 == NULL) { 10261 action = PF_DROP; 10262 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10263 goto done; 10264 } 10265 if (m0->m_len < sizeof(struct ip6_hdr)) { 10266 DPFPRINTF(PF_DEBUG_URGENT, 10267 "%s: m0->m_len < sizeof(struct ip6_hdr)", 10268 __func__); 10269 action = PF_DROP; 10270 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10271 goto bad; 10272 } 10273 ip6 = mtod(m0, struct ip6_hdr *); 10274 } 10275 } 10276 10277 if (ifp->if_flags & IFF_LOOPBACK) 10278 m0->m_flags |= M_SKIP_FIREWALL; 10279 10280 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 10281 ~ifp->if_hwassist) { 10282 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 10283 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 10284 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 10285 } 10286 10287 if (pd->dir == PF_IN) { 10288 uint16_t tmp; 10289 /* 10290 * Make sure dummynet gets the correct direction, in case it needs to 10291 * re-inject later. 10292 */ 10293 pd->dir = PF_OUT; 10294 10295 /* 10296 * The following processing is actually the rest of the inbound processing, even 10297 * though we've marked it as outbound (so we don't look through dummynet) and it 10298 * happens after the outbound processing (pf_test(PF_OUT) above). 10299 * Swap the dummynet pipe numbers, because it's going to come to the wrong 10300 * conclusion about what direction it's processing, and we can't fix it or it 10301 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 10302 * decision will pick the right pipe, and everything will mostly work as expected. 10303 */ 10304 tmp = pd->act.dnrpipe; 10305 pd->act.dnrpipe = pd->act.dnpipe; 10306 pd->act.dnpipe = tmp; 10307 } 10308 10309 /* 10310 * If the packet is too large for the outgoing interface, 10311 * send back an icmp6 error. 10312 */ 10313 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 10314 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 10315 mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL); 10316 if (mtag != NULL) { 10317 int ret __sdt_used; 10318 ret = pf_refragment6(ifp, &m0, mtag, ifp, true); 10319 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 10320 goto done; 10321 } 10322 10323 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 10324 md = m0; 10325 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 10326 if (md != NULL) { 10327 int ret __sdt_used; 10328 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 10329 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 10330 } 10331 } 10332 else { 10333 in6_ifstat_inc(ifp, ifs6_in_toobig); 10334 if (pd->act.rt != PF_DUPTO) { 10335 if (s && s->nat_rule != NULL) { 10336 MPASS(m0 == pd->m); 10337 PACKET_UNDO_NAT(pd, 10338 ((caddr_t)ip6 - m0->m_data) + 10339 sizeof(struct ip6_hdr), s); 10340 } 10341 10342 if (r->rt != PF_DUPTO) 10343 pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0, 10344 ifp->if_mtu, pd->af, r, pd->act.rtableid); 10345 } 10346 /* Return pass, so we return PFIL_CONSUMED to the stack. */ 10347 action = PF_PASS; 10348 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 10349 goto bad; 10350 } 10351 10352 done: 10353 if (pd->act.rt != PF_DUPTO) 10354 pd->m = NULL; 10355 else 10356 action = PF_PASS; 10357 return (action); 10358 10359 bad_locked: 10360 if (s) 10361 PF_STATE_UNLOCK(s); 10362 bad: 10363 m_freem(m0); 10364 goto done; 10365 } 10366 #endif /* INET6 */ 10367 10368 /* 10369 * FreeBSD supports cksum offloads for the following drivers. 10370 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 10371 * 10372 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 10373 * network driver performed cksum including pseudo header, need to verify 10374 * csum_data 10375 * CSUM_DATA_VALID : 10376 * network driver performed cksum, needs to additional pseudo header 10377 * cksum computation with partial csum_data(i.e. lack of H/W support for 10378 * pseudo header, for instance sk(4) and possibly gem(4)) 10379 * 10380 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 10381 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 10382 * TCP/UDP layer. 10383 * Also, set csum_data to 0xffff to force cksum validation. 10384 */ 10385 static int 10386 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 10387 { 10388 u_int16_t sum = 0; 10389 int hw_assist = 0; 10390 struct ip *ip; 10391 10392 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 10393 return (1); 10394 if (m->m_pkthdr.len < off + len) 10395 return (1); 10396 10397 switch (p) { 10398 case IPPROTO_TCP: 10399 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 10400 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 10401 sum = m->m_pkthdr.csum_data; 10402 } else { 10403 ip = mtod(m, struct ip *); 10404 sum = in_pseudo(ip->ip_src.s_addr, 10405 ip->ip_dst.s_addr, htonl((u_short)len + 10406 m->m_pkthdr.csum_data + IPPROTO_TCP)); 10407 } 10408 sum ^= 0xffff; 10409 ++hw_assist; 10410 } 10411 break; 10412 case IPPROTO_UDP: 10413 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 10414 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 10415 sum = m->m_pkthdr.csum_data; 10416 } else { 10417 ip = mtod(m, struct ip *); 10418 sum = in_pseudo(ip->ip_src.s_addr, 10419 ip->ip_dst.s_addr, htonl((u_short)len + 10420 m->m_pkthdr.csum_data + IPPROTO_UDP)); 10421 } 10422 sum ^= 0xffff; 10423 ++hw_assist; 10424 } 10425 break; 10426 case IPPROTO_ICMP: 10427 #ifdef INET6 10428 case IPPROTO_ICMPV6: 10429 #endif /* INET6 */ 10430 break; 10431 default: 10432 return (1); 10433 } 10434 10435 if (!hw_assist) { 10436 switch (af) { 10437 case AF_INET: 10438 if (m->m_len < sizeof(struct ip)) 10439 return (1); 10440 sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len); 10441 break; 10442 #ifdef INET6 10443 case AF_INET6: 10444 if (m->m_len < sizeof(struct ip6_hdr)) 10445 return (1); 10446 sum = in6_cksum(m, p, off, len); 10447 break; 10448 #endif /* INET6 */ 10449 } 10450 } 10451 if (sum) { 10452 switch (p) { 10453 case IPPROTO_TCP: 10454 { 10455 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 10456 break; 10457 } 10458 case IPPROTO_UDP: 10459 { 10460 KMOD_UDPSTAT_INC(udps_badsum); 10461 break; 10462 } 10463 #ifdef INET 10464 case IPPROTO_ICMP: 10465 { 10466 KMOD_ICMPSTAT_INC(icps_checksum); 10467 break; 10468 } 10469 #endif 10470 #ifdef INET6 10471 case IPPROTO_ICMPV6: 10472 { 10473 KMOD_ICMP6STAT_INC(icp6s_checksum); 10474 break; 10475 } 10476 #endif /* INET6 */ 10477 } 10478 return (1); 10479 } else { 10480 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 10481 m->m_pkthdr.csum_flags |= 10482 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 10483 m->m_pkthdr.csum_data = 0xffff; 10484 } 10485 } 10486 return (0); 10487 } 10488 10489 static bool 10490 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 10491 const struct pf_kstate *s, struct ip_fw_args *dnflow) 10492 { 10493 int dndir = r->direction; 10494 sa_family_t af = pd->naf; 10495 10496 if (s && dndir == PF_INOUT) { 10497 dndir = s->direction; 10498 } else if (dndir == PF_INOUT) { 10499 /* Assume primary direction. Happens when we've set dnpipe in 10500 * the ethernet level code. */ 10501 dndir = pd->dir; 10502 } 10503 10504 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 10505 return (false); 10506 10507 memset(dnflow, 0, sizeof(*dnflow)); 10508 10509 if (pd->dport != NULL) 10510 dnflow->f_id.dst_port = ntohs(*pd->dport); 10511 if (pd->sport != NULL) 10512 dnflow->f_id.src_port = ntohs(*pd->sport); 10513 10514 if (pd->dir == PF_IN) 10515 dnflow->flags |= IPFW_ARGS_IN; 10516 else 10517 dnflow->flags |= IPFW_ARGS_OUT; 10518 10519 if (pd->dir != dndir && pd->act.dnrpipe) { 10520 dnflow->rule.info = pd->act.dnrpipe; 10521 } 10522 else if (pd->dir == dndir && pd->act.dnpipe) { 10523 dnflow->rule.info = pd->act.dnpipe; 10524 } 10525 else { 10526 return (false); 10527 } 10528 10529 dnflow->rule.info |= IPFW_IS_DUMMYNET; 10530 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 10531 dnflow->rule.info |= IPFW_IS_PIPE; 10532 10533 dnflow->f_id.proto = pd->proto; 10534 dnflow->f_id.extra = dnflow->rule.info; 10535 if (s) 10536 af = s->key[PF_SK_STACK]->af; 10537 10538 switch (af) { 10539 case AF_INET: 10540 dnflow->f_id.addr_type = 4; 10541 if (s) { 10542 dnflow->f_id.src_ip = htonl( 10543 s->key[PF_SK_STACK]->addr[pd->sidx].v4.s_addr); 10544 dnflow->f_id.dst_ip = htonl( 10545 s->key[PF_SK_STACK]->addr[pd->didx].v4.s_addr); 10546 } else { 10547 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 10548 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 10549 } 10550 break; 10551 case AF_INET6: 10552 dnflow->f_id.addr_type = 6; 10553 10554 if (s) { 10555 dnflow->f_id.src_ip6 = 10556 s->key[PF_SK_STACK]->addr[pd->sidx].v6; 10557 dnflow->f_id.dst_ip6 = 10558 s->key[PF_SK_STACK]->addr[pd->didx].v6; 10559 } else { 10560 dnflow->f_id.src_ip6 = pd->src->v6; 10561 dnflow->f_id.dst_ip6 = pd->dst->v6; 10562 } 10563 break; 10564 } 10565 10566 /* 10567 * Separate this out, because while we pass the pre-NAT addresses to 10568 * dummynet we want the post-nat address family in case of nat64. 10569 * Dummynet may call ip_output/ip6_output itself, and we need it to 10570 * call the correct one. 10571 */ 10572 if (pd->naf == AF_INET6) 10573 dnflow->flags |= IPFW_ARGS_IP6; 10574 10575 return (true); 10576 } 10577 10578 int 10579 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 10580 struct inpcb *inp) 10581 { 10582 struct pfi_kkif *kif; 10583 struct mbuf *m = *m0; 10584 10585 M_ASSERTPKTHDR(m); 10586 MPASS(ifp->if_vnet == curvnet); 10587 NET_EPOCH_ASSERT(); 10588 10589 if (!V_pf_status.running) 10590 return (PF_PASS); 10591 10592 kif = (struct pfi_kkif *)ifp->if_pf_kif; 10593 10594 if (kif == NULL) { 10595 DPFPRINTF(PF_DEBUG_URGENT, 10596 "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname); 10597 return (PF_DROP); 10598 } 10599 if (kif->pfik_flags & PFI_IFLAG_SKIP) 10600 return (PF_PASS); 10601 10602 if (m->m_flags & M_SKIP_FIREWALL) 10603 return (PF_PASS); 10604 10605 if (__predict_false(! M_WRITABLE(*m0))) { 10606 m = *m0 = m_unshare(*m0, M_NOWAIT); 10607 if (*m0 == NULL) 10608 return (PF_DROP); 10609 } 10610 10611 /* Stateless! */ 10612 return (pf_test_eth_rule(dir, kif, m0)); 10613 } 10614 10615 static __inline void 10616 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 10617 { 10618 struct m_tag *mtag; 10619 10620 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 10621 10622 /* dummynet adds this tag, but pf does not need it, 10623 * and keeping it creates unexpected behavior, 10624 * e.g. in case of divert(4) usage right after dummynet. */ 10625 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 10626 if (mtag != NULL) 10627 m_tag_delete(m, mtag); 10628 } 10629 10630 static int 10631 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 10632 struct pf_krule *r, struct mbuf **m0) 10633 { 10634 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 10635 } 10636 10637 static int 10638 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 10639 struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa, 10640 struct mbuf **m0) 10641 { 10642 struct ip_fw_args dnflow; 10643 10644 NET_EPOCH_ASSERT(); 10645 10646 if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0) 10647 return (0); 10648 10649 if (ip_dn_io_ptr == NULL) { 10650 m_freem(*m0); 10651 *m0 = NULL; 10652 return (ENOMEM); 10653 } 10654 10655 if (pd->pf_mtag == NULL && 10656 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 10657 m_freem(*m0); 10658 *m0 = NULL; 10659 return (ENOMEM); 10660 } 10661 10662 if (ifp != NULL) { 10663 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 10664 10665 pd->pf_mtag->if_index = ifp->if_index; 10666 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 10667 10668 MPASS(sa != NULL); 10669 10670 switch (sa->sa_family) { 10671 case AF_INET: 10672 memcpy(&pd->pf_mtag->dst, sa, 10673 sizeof(struct sockaddr_in)); 10674 break; 10675 case AF_INET6: 10676 memcpy(&pd->pf_mtag->dst, sa, 10677 sizeof(struct sockaddr_in6)); 10678 break; 10679 } 10680 } 10681 10682 if (s != NULL && s->nat_rule != NULL && 10683 s->nat_rule->action == PF_RDR && 10684 ( 10685 #ifdef INET 10686 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 10687 #endif /* INET */ 10688 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 10689 /* 10690 * If we're redirecting to loopback mark this packet 10691 * as being local. Otherwise it might get dropped 10692 * if dummynet re-injects. 10693 */ 10694 (*m0)->m_pkthdr.rcvif = V_loif; 10695 } 10696 10697 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 10698 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 10699 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 10700 ip_dn_io_ptr(m0, &dnflow); 10701 if (*m0 != NULL) { 10702 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 10703 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 10704 } 10705 } 10706 10707 return (0); 10708 } 10709 10710 static int 10711 pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end, 10712 u_short *reason) 10713 { 10714 uint8_t type, length, opts[15 * 4 - sizeof(struct ip)]; 10715 10716 /* IP header in payload of ICMP packet may be too short */ 10717 if (pd->m->m_pkthdr.len < end) { 10718 DPFPRINTF(PF_DEBUG_MISC, "IP option too short"); 10719 REASON_SET(reason, PFRES_SHORT); 10720 return (PF_DROP); 10721 } 10722 10723 MPASS(end - off <= sizeof(opts)); 10724 m_copydata(pd->m, off, end - off, opts); 10725 end -= off; 10726 off = 0; 10727 10728 while (off < end) { 10729 type = opts[off]; 10730 if (type == IPOPT_EOL) 10731 break; 10732 if (type == IPOPT_NOP) { 10733 off++; 10734 continue; 10735 } 10736 if (off + 2 > end) { 10737 DPFPRINTF(PF_DEBUG_MISC, "IP length opt"); 10738 REASON_SET(reason, PFRES_IPOPTIONS); 10739 return (PF_DROP); 10740 } 10741 length = opts[off + 1]; 10742 if (length < 2) { 10743 DPFPRINTF(PF_DEBUG_MISC, "IP short opt"); 10744 REASON_SET(reason, PFRES_IPOPTIONS); 10745 return (PF_DROP); 10746 } 10747 if (off + length > end) { 10748 DPFPRINTF(PF_DEBUG_MISC, "IP long opt"); 10749 REASON_SET(reason, PFRES_IPOPTIONS); 10750 return (PF_DROP); 10751 } 10752 switch (type) { 10753 case IPOPT_RA: 10754 pd->badopts |= PF_OPT_ROUTER_ALERT; 10755 break; 10756 default: 10757 pd->badopts |= PF_OPT_OTHER; 10758 break; 10759 } 10760 off += length; 10761 } 10762 10763 return (PF_PASS); 10764 } 10765 10766 static int 10767 pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason) 10768 { 10769 struct ah ext; 10770 u_int32_t hlen, end; 10771 int hdr_cnt; 10772 10773 hlen = h->ip_hl << 2; 10774 if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) { 10775 REASON_SET(reason, PFRES_SHORT); 10776 return (PF_DROP); 10777 } 10778 if (hlen != sizeof(struct ip)) { 10779 if (pf_walk_option(pd, h, pd->off + sizeof(struct ip), 10780 pd->off + hlen, reason) != PF_PASS) 10781 return (PF_DROP); 10782 /* header options which contain only padding is fishy */ 10783 if (pd->badopts == 0) 10784 pd->badopts |= PF_OPT_OTHER; 10785 } 10786 end = pd->off + ntohs(h->ip_len); 10787 pd->off += hlen; 10788 pd->proto = h->ip_p; 10789 /* IGMP packets have router alert options, allow them */ 10790 if (pd->proto == IPPROTO_IGMP) { 10791 /* 10792 * According to RFC 1112 ttl must be set to 1 in all IGMP 10793 * packets sent to 224.0.0.1 10794 */ 10795 if ((h->ip_ttl != 1) && 10796 (h->ip_dst.s_addr == INADDR_ALLHOSTS_GROUP)) { 10797 DPFPRINTF(PF_DEBUG_MISC, "Invalid IGMP"); 10798 REASON_SET(reason, PFRES_IPOPTIONS); 10799 return (PF_DROP); 10800 } 10801 pd->badopts &= ~PF_OPT_ROUTER_ALERT; 10802 } 10803 /* stop walking over non initial fragments */ 10804 if ((h->ip_off & htons(IP_OFFMASK)) != 0) 10805 return (PF_PASS); 10806 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 10807 switch (pd->proto) { 10808 case IPPROTO_AH: 10809 /* fragments may be short */ 10810 if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 && 10811 end < pd->off + sizeof(ext)) 10812 return (PF_PASS); 10813 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10814 reason, AF_INET)) { 10815 DPFPRINTF(PF_DEBUG_MISC, "IP short exthdr"); 10816 return (PF_DROP); 10817 } 10818 pd->off += (ext.ah_len + 2) * 4; 10819 pd->proto = ext.ah_nxt; 10820 break; 10821 default: 10822 return (PF_PASS); 10823 } 10824 } 10825 DPFPRINTF(PF_DEBUG_MISC, "IPv4 nested authentication header limit"); 10826 REASON_SET(reason, PFRES_IPOPTIONS); 10827 return (PF_DROP); 10828 } 10829 10830 #ifdef INET6 10831 static int 10832 pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end, 10833 u_short *reason) 10834 { 10835 struct ip6_opt opt; 10836 struct ip6_opt_jumbo jumbo; 10837 10838 while (off < end) { 10839 if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type, 10840 sizeof(opt.ip6o_type), reason, AF_INET6)) { 10841 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt type"); 10842 return (PF_DROP); 10843 } 10844 if (opt.ip6o_type == IP6OPT_PAD1) { 10845 off++; 10846 continue; 10847 } 10848 if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), 10849 reason, AF_INET6)) { 10850 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt"); 10851 return (PF_DROP); 10852 } 10853 if (off + sizeof(opt) + opt.ip6o_len > end) { 10854 DPFPRINTF(PF_DEBUG_MISC, "IPv6 long opt"); 10855 REASON_SET(reason, PFRES_IPOPTIONS); 10856 return (PF_DROP); 10857 } 10858 switch (opt.ip6o_type) { 10859 case IP6OPT_PADN: 10860 break; 10861 case IP6OPT_JUMBO: 10862 pd->badopts |= PF_OPT_JUMBO; 10863 if (pd->jumbolen != 0) { 10864 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple jumbo"); 10865 REASON_SET(reason, PFRES_IPOPTIONS); 10866 return (PF_DROP); 10867 } 10868 if (ntohs(h->ip6_plen) != 0) { 10869 DPFPRINTF(PF_DEBUG_MISC, "IPv6 bad jumbo plen"); 10870 REASON_SET(reason, PFRES_IPOPTIONS); 10871 return (PF_DROP); 10872 } 10873 if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), 10874 reason, AF_INET6)) { 10875 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbo"); 10876 return (PF_DROP); 10877 } 10878 memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len, 10879 sizeof(pd->jumbolen)); 10880 pd->jumbolen = ntohl(pd->jumbolen); 10881 if (pd->jumbolen < IPV6_MAXPACKET) { 10882 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbolen"); 10883 REASON_SET(reason, PFRES_IPOPTIONS); 10884 return (PF_DROP); 10885 } 10886 break; 10887 case IP6OPT_ROUTER_ALERT: 10888 pd->badopts |= PF_OPT_ROUTER_ALERT; 10889 break; 10890 default: 10891 pd->badopts |= PF_OPT_OTHER; 10892 break; 10893 } 10894 off += sizeof(opt) + opt.ip6o_len; 10895 } 10896 10897 return (PF_PASS); 10898 } 10899 10900 int 10901 pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason) 10902 { 10903 struct ip6_frag frag; 10904 struct ip6_ext ext; 10905 struct icmp6_hdr icmp6; 10906 struct ip6_rthdr rthdr; 10907 uint32_t end; 10908 int hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0; 10909 10910 pd->off += sizeof(struct ip6_hdr); 10911 end = pd->off + ntohs(h->ip6_plen); 10912 pd->fragoff = pd->extoff = pd->jumbolen = 0; 10913 pd->proto = h->ip6_nxt; 10914 for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) { 10915 switch (pd->proto) { 10916 case IPPROTO_ROUTING: 10917 case IPPROTO_DSTOPTS: 10918 pd->badopts |= PF_OPT_OTHER; 10919 break; 10920 case IPPROTO_HOPOPTS: 10921 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10922 reason, AF_INET6)) { 10923 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr"); 10924 return (PF_DROP); 10925 } 10926 if (pf_walk_option6(pd, h, pd->off + sizeof(ext), 10927 pd->off + (ext.ip6e_len + 1) * 8, 10928 reason) != PF_PASS) 10929 return (PF_DROP); 10930 /* option header which contains only padding is fishy */ 10931 if (pd->badopts == 0) 10932 pd->badopts |= PF_OPT_OTHER; 10933 break; 10934 } 10935 switch (pd->proto) { 10936 case IPPROTO_FRAGMENT: 10937 if (fraghdr_cnt++) { 10938 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple fragment"); 10939 REASON_SET(reason, PFRES_FRAG); 10940 return (PF_DROP); 10941 } 10942 /* jumbo payload packets cannot be fragmented */ 10943 if (pd->jumbolen != 0) { 10944 DPFPRINTF(PF_DEBUG_MISC, "IPv6 fragmented jumbo"); 10945 REASON_SET(reason, PFRES_FRAG); 10946 return (PF_DROP); 10947 } 10948 if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag), 10949 reason, AF_INET6)) { 10950 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short fragment"); 10951 return (PF_DROP); 10952 } 10953 /* stop walking over non initial fragments */ 10954 if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) { 10955 pd->fragoff = pd->off; 10956 return (PF_PASS); 10957 } 10958 /* RFC6946: reassemble only non atomic fragments */ 10959 if (frag.ip6f_offlg & IP6F_MORE_FRAG) 10960 pd->fragoff = pd->off; 10961 pd->off += sizeof(frag); 10962 pd->proto = frag.ip6f_nxt; 10963 break; 10964 case IPPROTO_ROUTING: 10965 if (rthdr_cnt++) { 10966 DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple rthdr"); 10967 REASON_SET(reason, PFRES_IPOPTIONS); 10968 return (PF_DROP); 10969 } 10970 /* fragments may be short */ 10971 if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) { 10972 pd->off = pd->fragoff; 10973 pd->proto = IPPROTO_FRAGMENT; 10974 return (PF_PASS); 10975 } 10976 if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr), 10977 reason, AF_INET6)) { 10978 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short rthdr"); 10979 return (PF_DROP); 10980 } 10981 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 10982 DPFPRINTF(PF_DEBUG_MISC, "IPv6 rthdr0"); 10983 REASON_SET(reason, PFRES_IPOPTIONS); 10984 return (PF_DROP); 10985 } 10986 /* FALLTHROUGH */ 10987 case IPPROTO_HOPOPTS: 10988 /* RFC2460 4.1: Hop-by-Hop only after IPv6 header */ 10989 if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) { 10990 DPFPRINTF(PF_DEBUG_MISC, "IPv6 hopopts not first"); 10991 REASON_SET(reason, PFRES_IPOPTIONS); 10992 return (PF_DROP); 10993 } 10994 /* FALLTHROUGH */ 10995 case IPPROTO_AH: 10996 case IPPROTO_DSTOPTS: 10997 if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext), 10998 reason, AF_INET6)) { 10999 DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr"); 11000 return (PF_DROP); 11001 } 11002 /* fragments may be short */ 11003 if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) { 11004 pd->off = pd->fragoff; 11005 pd->proto = IPPROTO_FRAGMENT; 11006 return (PF_PASS); 11007 } 11008 /* reassembly needs the ext header before the frag */ 11009 if (pd->fragoff == 0) 11010 pd->extoff = pd->off; 11011 if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 && 11012 ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) { 11013 DPFPRINTF(PF_DEBUG_MISC, "IPv6 missing jumbo"); 11014 REASON_SET(reason, PFRES_IPOPTIONS); 11015 return (PF_DROP); 11016 } 11017 if (pd->proto == IPPROTO_AH) 11018 pd->off += (ext.ip6e_len + 2) * 4; 11019 else 11020 pd->off += (ext.ip6e_len + 1) * 8; 11021 pd->proto = ext.ip6e_nxt; 11022 break; 11023 case IPPROTO_ICMPV6: 11024 /* fragments may be short, ignore inner header then */ 11025 if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) { 11026 pd->off = pd->fragoff; 11027 pd->proto = IPPROTO_FRAGMENT; 11028 return (PF_PASS); 11029 } 11030 if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6), 11031 reason, AF_INET6)) { 11032 DPFPRINTF(PF_DEBUG_MISC, 11033 "IPv6 short icmp6hdr"); 11034 return (PF_DROP); 11035 } 11036 /* ICMP multicast packets have router alert options */ 11037 switch (icmp6.icmp6_type) { 11038 case MLD_LISTENER_QUERY: 11039 case MLD_LISTENER_REPORT: 11040 case MLD_LISTENER_DONE: 11041 case MLDV2_LISTENER_REPORT: 11042 /* 11043 * According to RFC 2710 all MLD messages are 11044 * sent with hop-limit (ttl) set to 1, and link 11045 * local source address. If either one is 11046 * missing then MLD message is invalid and 11047 * should be discarded. 11048 */ 11049 if ((h->ip6_hlim != 1) || 11050 !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) { 11051 DPFPRINTF(PF_DEBUG_MISC, "Invalid MLD"); 11052 REASON_SET(reason, PFRES_IPOPTIONS); 11053 return (PF_DROP); 11054 } 11055 pd->badopts &= ~PF_OPT_ROUTER_ALERT; 11056 break; 11057 } 11058 return (PF_PASS); 11059 case IPPROTO_TCP: 11060 case IPPROTO_UDP: 11061 case IPPROTO_SCTP: 11062 /* fragments may be short, ignore inner header then */ 11063 if (pd->fragoff != 0 && end < pd->off + 11064 (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) : 11065 pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) : 11066 pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) : 11067 sizeof(struct icmp6_hdr))) { 11068 pd->off = pd->fragoff; 11069 pd->proto = IPPROTO_FRAGMENT; 11070 } 11071 /* FALLTHROUGH */ 11072 default: 11073 return (PF_PASS); 11074 } 11075 } 11076 DPFPRINTF(PF_DEBUG_MISC, "IPv6 nested extension header limit"); 11077 REASON_SET(reason, PFRES_IPOPTIONS); 11078 return (PF_DROP); 11079 } 11080 #endif /* INET6 */ 11081 11082 static void 11083 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 11084 { 11085 memset(pd, 0, sizeof(*pd)); 11086 pd->pf_mtag = pf_find_mtag(m); 11087 pd->m = m; 11088 } 11089 11090 static int 11091 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 11092 u_short *action, u_short *reason, struct pfi_kkif *kif, 11093 struct pf_rule_actions *default_actions) 11094 { 11095 pd->dir = dir; 11096 pd->kif = kif; 11097 pd->m = *m0; 11098 pd->sidx = (dir == PF_IN) ? 0 : 1; 11099 pd->didx = (dir == PF_IN) ? 1 : 0; 11100 pd->af = pd->naf = af; 11101 11102 PF_RULES_ASSERT(); 11103 11104 TAILQ_INIT(&pd->sctp_multihome_jobs); 11105 if (default_actions != NULL) 11106 memcpy(&pd->act, default_actions, sizeof(pd->act)); 11107 11108 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 11109 pd->act.dnpipe = pd->pf_mtag->dnpipe; 11110 pd->act.flags = pd->pf_mtag->dnflags; 11111 } 11112 11113 switch (af) { 11114 #ifdef INET 11115 case AF_INET: { 11116 struct ip *h; 11117 11118 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 11119 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 11120 DPFPRINTF(PF_DEBUG_URGENT, 11121 "%s: m_len < sizeof(struct ip), pullup failed", 11122 __func__); 11123 *action = PF_DROP; 11124 REASON_SET(reason, PFRES_SHORT); 11125 return (PF_DROP); 11126 } 11127 11128 h = mtod(pd->m, struct ip *); 11129 if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) { 11130 *action = PF_DROP; 11131 REASON_SET(reason, PFRES_SHORT); 11132 return (PF_DROP); 11133 } 11134 11135 if (pf_normalize_ip(reason, pd) != PF_PASS) { 11136 /* We do IP header normalization and packet reassembly here */ 11137 *m0 = pd->m; 11138 *action = PF_DROP; 11139 return (PF_DROP); 11140 } 11141 *m0 = pd->m; 11142 h = mtod(pd->m, struct ip *); 11143 11144 if (pf_walk_header(pd, h, reason) != PF_PASS) { 11145 *action = PF_DROP; 11146 return (PF_DROP); 11147 } 11148 11149 pd->src = (struct pf_addr *)&h->ip_src; 11150 pd->dst = (struct pf_addr *)&h->ip_dst; 11151 pf_addrcpy(&pd->osrc, pd->src, af); 11152 pf_addrcpy(&pd->odst, pd->dst, af); 11153 pd->ip_sum = &h->ip_sum; 11154 pd->tos = h->ip_tos & ~IPTOS_ECN_MASK; 11155 pd->ttl = h->ip_ttl; 11156 pd->tot_len = ntohs(h->ip_len); 11157 pd->act.rtableid = -1; 11158 pd->df = h->ip_off & htons(IP_DF); 11159 pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ? 11160 PF_VPROTO_FRAGMENT : pd->proto; 11161 11162 break; 11163 } 11164 #endif /* INET */ 11165 #ifdef INET6 11166 case AF_INET6: { 11167 struct ip6_hdr *h; 11168 11169 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 11170 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 11171 DPFPRINTF(PF_DEBUG_URGENT, 11172 "%s: m_len < sizeof(struct ip6_hdr)" 11173 ", pullup failed", __func__); 11174 *action = PF_DROP; 11175 REASON_SET(reason, PFRES_SHORT); 11176 return (PF_DROP); 11177 } 11178 11179 h = mtod(pd->m, struct ip6_hdr *); 11180 if (pd->m->m_pkthdr.len < 11181 sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) { 11182 *action = PF_DROP; 11183 REASON_SET(reason, PFRES_SHORT); 11184 return (PF_DROP); 11185 } 11186 11187 /* 11188 * we do not support jumbogram. if we keep going, zero ip6_plen 11189 * will do something bad, so drop the packet for now. 11190 */ 11191 if (htons(h->ip6_plen) == 0) { 11192 *action = PF_DROP; 11193 return (PF_DROP); 11194 } 11195 11196 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 11197 *action = PF_DROP; 11198 return (PF_DROP); 11199 } 11200 11201 h = mtod(pd->m, struct ip6_hdr *); 11202 pd->src = (struct pf_addr *)&h->ip6_src; 11203 pd->dst = (struct pf_addr *)&h->ip6_dst; 11204 pf_addrcpy(&pd->osrc, pd->src, af); 11205 pf_addrcpy(&pd->odst, pd->dst, af); 11206 pd->ip_sum = NULL; 11207 pd->tos = IPV6_DSCP(h); 11208 pd->ttl = h->ip6_hlim; 11209 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 11210 pd->act.rtableid = -1; 11211 11212 pd->virtual_proto = (pd->fragoff != 0) ? 11213 PF_VPROTO_FRAGMENT : pd->proto; 11214 11215 /* We do IP header normalization and packet reassembly here */ 11216 if (pf_normalize_ip6(pd->fragoff, reason, pd) != 11217 PF_PASS) { 11218 *m0 = pd->m; 11219 *action = PF_DROP; 11220 return (PF_DROP); 11221 } 11222 *m0 = pd->m; 11223 if (pd->m == NULL) { 11224 /* packet sits in reassembly queue, no error */ 11225 *action = PF_PASS; 11226 return (PF_DROP); 11227 } 11228 11229 /* Update pointers into the packet. */ 11230 h = mtod(pd->m, struct ip6_hdr *); 11231 pd->src = (struct pf_addr *)&h->ip6_src; 11232 pd->dst = (struct pf_addr *)&h->ip6_dst; 11233 11234 pd->off = 0; 11235 11236 if (pf_walk_header6(pd, h, reason) != PF_PASS) { 11237 *action = PF_DROP; 11238 return (PF_DROP); 11239 } 11240 11241 if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) { 11242 /* 11243 * Reassembly may have changed the next protocol from 11244 * fragment to something else, so update. 11245 */ 11246 pd->virtual_proto = pd->proto; 11247 MPASS(pd->fragoff == 0); 11248 } 11249 11250 if (pd->fragoff != 0) 11251 pd->virtual_proto = PF_VPROTO_FRAGMENT; 11252 11253 break; 11254 } 11255 #endif /* INET6 */ 11256 default: 11257 panic("pf_setup_pdesc called with illegal af %u", af); 11258 } 11259 11260 switch (pd->virtual_proto) { 11261 case IPPROTO_TCP: { 11262 struct tcphdr *th = &pd->hdr.tcp; 11263 11264 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), 11265 reason, af)) { 11266 *action = PF_DROP; 11267 REASON_SET(reason, PFRES_SHORT); 11268 return (PF_DROP); 11269 } 11270 pd->hdrlen = sizeof(*th); 11271 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 11272 pd->sport = &th->th_sport; 11273 pd->dport = &th->th_dport; 11274 pd->pcksum = &th->th_sum; 11275 break; 11276 } 11277 case IPPROTO_UDP: { 11278 struct udphdr *uh = &pd->hdr.udp; 11279 11280 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), 11281 reason, af)) { 11282 *action = PF_DROP; 11283 REASON_SET(reason, PFRES_SHORT); 11284 return (PF_DROP); 11285 } 11286 pd->hdrlen = sizeof(*uh); 11287 if (uh->uh_dport == 0 || 11288 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 11289 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 11290 *action = PF_DROP; 11291 REASON_SET(reason, PFRES_SHORT); 11292 return (PF_DROP); 11293 } 11294 pd->sport = &uh->uh_sport; 11295 pd->dport = &uh->uh_dport; 11296 pd->pcksum = &uh->uh_sum; 11297 break; 11298 } 11299 case IPPROTO_SCTP: { 11300 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 11301 reason, af)) { 11302 *action = PF_DROP; 11303 REASON_SET(reason, PFRES_SHORT); 11304 return (PF_DROP); 11305 } 11306 pd->hdrlen = sizeof(pd->hdr.sctp); 11307 pd->p_len = pd->tot_len - pd->off; 11308 11309 pd->sport = &pd->hdr.sctp.src_port; 11310 pd->dport = &pd->hdr.sctp.dest_port; 11311 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 11312 *action = PF_DROP; 11313 REASON_SET(reason, PFRES_SHORT); 11314 return (PF_DROP); 11315 } 11316 11317 /* 11318 * Placeholder. The SCTP checksum is 32-bits, but 11319 * pf_test_state() expects to update a 16-bit checksum. 11320 * Provide a dummy value which we'll subsequently ignore. 11321 * Do this before pf_scan_sctp() so any jobs we enqueue 11322 * have a pcksum set. 11323 */ 11324 pd->pcksum = &pd->sctp_dummy_sum; 11325 11326 if (pf_scan_sctp(pd) != PF_PASS) { 11327 *action = PF_DROP; 11328 REASON_SET(reason, PFRES_SHORT); 11329 return (PF_DROP); 11330 } 11331 break; 11332 } 11333 case IPPROTO_ICMP: { 11334 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 11335 reason, af)) { 11336 *action = PF_DROP; 11337 REASON_SET(reason, PFRES_SHORT); 11338 return (PF_DROP); 11339 } 11340 pd->pcksum = &pd->hdr.icmp.icmp_cksum; 11341 pd->hdrlen = ICMP_MINLEN; 11342 break; 11343 } 11344 #ifdef INET6 11345 case IPPROTO_ICMPV6: { 11346 size_t icmp_hlen = sizeof(struct icmp6_hdr); 11347 11348 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 11349 reason, af)) { 11350 *action = PF_DROP; 11351 REASON_SET(reason, PFRES_SHORT); 11352 return (PF_DROP); 11353 } 11354 /* ICMP headers we look further into to match state */ 11355 switch (pd->hdr.icmp6.icmp6_type) { 11356 case MLD_LISTENER_QUERY: 11357 case MLD_LISTENER_REPORT: 11358 icmp_hlen = sizeof(struct mld_hdr); 11359 break; 11360 case ND_NEIGHBOR_SOLICIT: 11361 case ND_NEIGHBOR_ADVERT: 11362 icmp_hlen = sizeof(struct nd_neighbor_solicit); 11363 /* FALLTHROUGH */ 11364 case ND_ROUTER_SOLICIT: 11365 case ND_ROUTER_ADVERT: 11366 case ND_REDIRECT: 11367 if (pd->ttl != 255) { 11368 REASON_SET(reason, PFRES_NORM); 11369 return (PF_DROP); 11370 } 11371 break; 11372 } 11373 if (icmp_hlen > sizeof(struct icmp6_hdr) && 11374 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 11375 reason, af)) { 11376 *action = PF_DROP; 11377 REASON_SET(reason, PFRES_SHORT); 11378 return (PF_DROP); 11379 } 11380 pd->hdrlen = icmp_hlen; 11381 pd->pcksum = &pd->hdr.icmp6.icmp6_cksum; 11382 break; 11383 } 11384 #endif /* INET6 */ 11385 default: 11386 /* 11387 * Placeholder value, so future calls to pf_change_ap() don't 11388 * try to update a NULL checksum pointer. 11389 */ 11390 pd->pcksum = &pd->sctp_dummy_sum; 11391 break; 11392 } 11393 11394 if (pd->sport) 11395 pd->osport = pd->nsport = *pd->sport; 11396 if (pd->dport) 11397 pd->odport = pd->ndport = *pd->dport; 11398 11399 MPASS(pd->pcksum != NULL); 11400 11401 return (PF_PASS); 11402 } 11403 11404 static __inline void 11405 pf_rule_counters_inc(struct pf_pdesc *pd, struct pf_krule *r, int dir_out, 11406 int op_pass, sa_family_t af, struct pf_addr *src_host, 11407 struct pf_addr *dst_host) 11408 { 11409 pf_counter_u64_add_protected(&(r->packets[dir_out]), 1); 11410 pf_counter_u64_add_protected(&(r->bytes[dir_out]), pd->tot_len); 11411 pf_update_timestamp(r); 11412 11413 if (r->src.addr.type == PF_ADDR_TABLE) 11414 pfr_update_stats(r->src.addr.p.tbl, src_host, af, 11415 pd->tot_len, dir_out, op_pass, r->src.neg); 11416 if (r->dst.addr.type == PF_ADDR_TABLE) 11417 pfr_update_stats(r->dst.addr.p.tbl, dst_host, af, 11418 pd->tot_len, dir_out, op_pass, r->dst.neg); 11419 } 11420 11421 static void 11422 pf_counters_inc(int action, struct pf_pdesc *pd, struct pf_kstate *s, 11423 struct pf_krule *r, struct pf_krule *a, struct pf_krule_slist *match_rules) 11424 { 11425 struct pf_krule_slist *mr = match_rules; 11426 struct pf_krule_item *ri; 11427 struct pf_krule *nr = NULL; 11428 struct pf_addr *src_host = pd->src; 11429 struct pf_addr *dst_host = pd->dst; 11430 struct pf_state_key *key; 11431 int dir_out = (pd->dir == PF_OUT); 11432 int op_r_pass = (r->action == PF_PASS); 11433 int op_pass = (action == PF_PASS || action == PF_AFRT); 11434 int s_dir_in, s_dir_out, s_dir_rev; 11435 sa_family_t af = pd->af; 11436 11437 pf_counter_u64_critical_enter(); 11438 11439 /* 11440 * Set AF for interface counters, it will be later overwritten for 11441 * rule and state counters with value from proper state key. 11442 */ 11443 if (action == PF_AFRT) { 11444 MPASS(s != NULL); 11445 if (s->direction == PF_OUT && dir_out) 11446 af = pd->naf; 11447 } 11448 11449 pf_counter_u64_add_protected( 11450 &pd->kif->pfik_bytes[af == AF_INET6][dir_out][!op_pass], 11451 pd->tot_len); 11452 pf_counter_u64_add_protected( 11453 &pd->kif->pfik_packets[af == AF_INET6][dir_out][!op_pass], 11454 1); 11455 11456 /* If the rule has failed to apply, don't increase its counters */ 11457 if (!(op_pass || r->action == PF_DROP)) { 11458 pf_counter_u64_critical_exit(); 11459 return; 11460 } 11461 11462 if (s != NULL) { 11463 PF_STATE_LOCK_ASSERT(s); 11464 mr = &(s->match_rules); 11465 11466 /* 11467 * For af-to on the inbound direction we can determine 11468 * the direction of passing packet only by checking direction 11469 * of AF translation. The af-to in "in" direction covers both 11470 * the inbound and the outbound side of state tracking, 11471 * so pd->dir is always PF_IN. We set dir_out and s_dir_rev 11472 * in a way to count packets as if the state was outbound, 11473 * because pfctl -ss shows the state with "->", as if it was 11474 * oubound. 11475 */ 11476 if (action == PF_AFRT && s->direction == PF_IN) { 11477 dir_out = (pd->naf == s->rule->naf); 11478 s_dir_in = 1; 11479 s_dir_out = 0; 11480 s_dir_rev = (pd->naf == s->rule->af); 11481 } else { 11482 dir_out = (pd->dir == PF_OUT); 11483 s_dir_in = (s->direction == PF_IN); 11484 s_dir_out = (s->direction == PF_OUT); 11485 s_dir_rev = (pd->dir != s->direction); 11486 } 11487 11488 /* pd->tot_len is a problematic with af-to rules. Sure, we can 11489 * agree that it's the post-af-to packet length that was 11490 * forwarded through a state, but what about tables which match 11491 * on pre-af-to addresses? We don't have access the the original 11492 * packet length anymore. 11493 */ 11494 s->packets[s_dir_rev]++; 11495 s->bytes[s_dir_rev] += pd->tot_len; 11496 11497 /* 11498 * Source nodes are accessed unlocked here. But since we are 11499 * operating with stateful tracking and the state is locked, 11500 * those SNs could not have been freed. 11501 */ 11502 for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) { 11503 if (s->sns[sn_type] != NULL) { 11504 counter_u64_add( 11505 s->sns[sn_type]->packets[dir_out], 11506 1); 11507 counter_u64_add( 11508 s->sns[sn_type]->bytes[dir_out], 11509 pd->tot_len); 11510 } 11511 } 11512 11513 /* Start with pre-NAT addresses */ 11514 key = s->key[(s->direction == PF_OUT)]; 11515 src_host = &(key->addr[s_dir_out]); 11516 dst_host = &(key->addr[s_dir_in]); 11517 af = key->af; 11518 if (s->nat_rule) { 11519 /* Old-style NAT rules */ 11520 if (s->nat_rule->action == PF_NAT || 11521 s->nat_rule->action == PF_RDR || 11522 s->nat_rule->action == PF_BINAT) { 11523 nr = s->nat_rule; 11524 pf_rule_counters_inc(pd, s->nat_rule, dir_out, 11525 op_r_pass, af, src_host, dst_host); 11526 /* Use post-NAT addresses from now on */ 11527 key = s->key[s_dir_in]; 11528 src_host = &(key->addr[s_dir_out]); 11529 dst_host = &(key->addr[s_dir_in]); 11530 af = key->af; 11531 } 11532 } 11533 } 11534 11535 SLIST_FOREACH(ri, mr, entry) { 11536 pf_rule_counters_inc(pd, ri->r, dir_out, op_r_pass, af, 11537 src_host, dst_host); 11538 if (s && s->nat_rule == ri->r) { 11539 /* Use post-NAT addresses after a match NAT rule */ 11540 key = s->key[s_dir_in]; 11541 src_host = &(key->addr[s_dir_out]); 11542 dst_host = &(key->addr[s_dir_in]); 11543 af = key->af; 11544 } 11545 } 11546 11547 if (a != NULL) { 11548 pf_rule_counters_inc(pd, a, dir_out, op_r_pass, af, 11549 src_host, dst_host); 11550 } 11551 11552 if (r != nr) { 11553 pf_rule_counters_inc(pd, r, dir_out, op_r_pass, af, 11554 src_host, dst_host); 11555 } 11556 11557 pf_counter_u64_critical_exit(); 11558 11559 if (s == NULL) { 11560 pf_free_match_rules(mr); 11561 } 11562 } 11563 11564 static void 11565 pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm, 11566 struct pf_krule *am, struct pf_kruleset *ruleset, 11567 struct pf_krule_slist *match_rules) 11568 { 11569 struct pf_krule_item *ri; 11570 11571 /* if this is the log(matches) rule, packet has been logged already */ 11572 if (rm->log & PF_LOG_MATCHES) 11573 return; 11574 11575 SLIST_FOREACH(ri, match_rules, entry) 11576 if (ri->r->log & PF_LOG_MATCHES) 11577 PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am, 11578 ruleset, pd, 1, ri->r); 11579 } 11580 11581 #if defined(INET) || defined(INET6) 11582 int 11583 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 11584 struct inpcb *inp, struct pf_rule_actions *default_actions) 11585 { 11586 struct pfi_kkif *kif; 11587 u_short action, reason = 0; 11588 struct m_tag *mtag; 11589 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 11590 struct pf_kstate *s = NULL; 11591 struct pf_kruleset *ruleset = NULL; 11592 struct pf_krule_item *ri; 11593 struct pf_krule_slist match_rules; 11594 struct pf_pdesc pd; 11595 int use_2nd_queue = 0; 11596 uint16_t tag; 11597 11598 PF_RULES_RLOCK_TRACKER; 11599 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 11600 M_ASSERTPKTHDR(*m0); 11601 NET_EPOCH_ASSERT(); 11602 11603 if (!V_pf_status.running) 11604 return (PF_PASS); 11605 11606 kif = (struct pfi_kkif *)ifp->if_pf_kif; 11607 11608 if (__predict_false(kif == NULL)) { 11609 DPFPRINTF(PF_DEBUG_URGENT, 11610 "%s: kif == NULL, if_xname %s", 11611 __func__, ifp->if_xname); 11612 return (PF_DROP); 11613 } 11614 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 11615 return (PF_PASS); 11616 } 11617 11618 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 11619 return (PF_PASS); 11620 } 11621 11622 if (__predict_false(! M_WRITABLE(*m0))) { 11623 *m0 = m_unshare(*m0, M_NOWAIT); 11624 if (*m0 == NULL) { 11625 return (PF_DROP); 11626 } 11627 } 11628 11629 pf_init_pdesc(&pd, *m0); 11630 SLIST_INIT(&match_rules); 11631 11632 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 11633 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 11634 11635 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 11636 pd.pf_mtag->if_idxgen); 11637 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 11638 m_freem(*m0); 11639 *m0 = NULL; 11640 return (PF_PASS); 11641 } 11642 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 11643 *m0 = NULL; 11644 return (PF_PASS); 11645 } 11646 11647 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 11648 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 11649 /* Dummynet re-injects packets after they've 11650 * completed their delay. We've already 11651 * processed them, so pass unconditionally. */ 11652 11653 /* But only once. We may see the packet multiple times (e.g. 11654 * PFIL_IN/PFIL_OUT). */ 11655 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 11656 11657 return (PF_PASS); 11658 } 11659 11660 PF_RULES_RLOCK(); 11661 11662 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 11663 kif, default_actions) != PF_PASS) { 11664 if (action != PF_PASS) 11665 pd.act.log |= PF_LOG_FORCE; 11666 goto done; 11667 } 11668 11669 #ifdef INET 11670 if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD && 11671 pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) { 11672 PF_RULES_RUNLOCK(); 11673 icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 11674 0, ifp->if_mtu); 11675 *m0 = NULL; 11676 return (PF_DROP); 11677 } 11678 #endif /* INET */ 11679 #ifdef INET6 11680 /* 11681 * If we end up changing IP addresses (e.g. binat) the stack may get 11682 * confused and fail to send the icmp6 packet too big error. Just send 11683 * it here, before we do any NAT. 11684 */ 11685 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 11686 in6_ifmtu(ifp) < pf_max_frag_size(*m0)) { 11687 PF_RULES_RUNLOCK(); 11688 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, in6_ifmtu(ifp)); 11689 *m0 = NULL; 11690 return (PF_DROP); 11691 } 11692 #endif /* INET6 */ 11693 11694 if (__predict_false(ip_divert_ptr != NULL) && 11695 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 11696 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 11697 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 11698 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 11699 if (pd.pf_mtag == NULL && 11700 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11701 action = PF_DROP; 11702 goto done; 11703 } 11704 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 11705 } 11706 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 11707 pd.m->m_flags |= M_FASTFWD_OURS; 11708 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 11709 } 11710 m_tag_delete(pd.m, mtag); 11711 11712 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 11713 if (mtag != NULL) 11714 m_tag_delete(pd.m, mtag); 11715 } 11716 11717 switch (pd.virtual_proto) { 11718 case PF_VPROTO_FRAGMENT: 11719 /* 11720 * handle fragments that aren't reassembled by 11721 * normalization 11722 */ 11723 if (kif == NULL || r == NULL) /* pflog */ 11724 action = PF_DROP; 11725 else 11726 action = pf_test_rule(&r, &s, &pd, &a, 11727 &ruleset, &reason, inp, &match_rules); 11728 if (action != PF_PASS) 11729 REASON_SET(&reason, PFRES_FRAG); 11730 break; 11731 11732 case IPPROTO_TCP: { 11733 /* Respond to SYN with a syncookie. */ 11734 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 11735 pd.dir == PF_IN && pf_synflood_check(&pd)) { 11736 pf_syncookie_send(&pd, &reason); 11737 action = PF_DROP; 11738 break; 11739 } 11740 11741 if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0) 11742 use_2nd_queue = 1; 11743 action = pf_normalize_tcp(&pd); 11744 if (action == PF_DROP) 11745 break; 11746 action = pf_test_state(&s, &pd, &reason); 11747 if (action == PF_PASS || action == PF_AFRT) { 11748 if (s != NULL) { 11749 if (V_pfsync_update_state_ptr != NULL) 11750 V_pfsync_update_state_ptr(s); 11751 r = s->rule; 11752 a = s->anchor; 11753 } 11754 } else if (s == NULL) { 11755 /* Validate remote SYN|ACK, re-create original SYN if 11756 * valid. */ 11757 if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == 11758 TH_ACK && pf_syncookie_validate(&pd) && 11759 pd.dir == PF_IN) { 11760 struct mbuf *msyn; 11761 11762 msyn = pf_syncookie_recreate_syn(&pd, &reason); 11763 if (msyn == NULL) { 11764 action = PF_DROP; 11765 break; 11766 } 11767 11768 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 11769 &pd.act); 11770 m_freem(msyn); 11771 if (action != PF_PASS) 11772 break; 11773 11774 action = pf_test_state(&s, &pd, &reason); 11775 if (action != PF_PASS || s == NULL) { 11776 action = PF_DROP; 11777 break; 11778 } 11779 11780 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 11781 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 11782 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 11783 action = pf_synproxy(&pd, s, &reason); 11784 break; 11785 } else { 11786 action = pf_test_rule(&r, &s, &pd, 11787 &a, &ruleset, &reason, inp, &match_rules); 11788 } 11789 } 11790 break; 11791 } 11792 11793 case IPPROTO_SCTP: 11794 action = pf_normalize_sctp(&pd); 11795 if (action == PF_DROP) 11796 break; 11797 /* fallthrough */ 11798 case IPPROTO_UDP: 11799 default: 11800 action = pf_test_state(&s, &pd, &reason); 11801 if (action == PF_PASS || action == PF_AFRT) { 11802 if (s != NULL) { 11803 if (V_pfsync_update_state_ptr != NULL) 11804 V_pfsync_update_state_ptr(s); 11805 r = s->rule; 11806 a = s->anchor; 11807 } 11808 } else if (s == NULL) { 11809 action = pf_test_rule(&r, &s, 11810 &pd, &a, &ruleset, &reason, inp, &match_rules); 11811 } 11812 break; 11813 11814 case IPPROTO_ICMP: 11815 case IPPROTO_ICMPV6: { 11816 if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) { 11817 action = PF_DROP; 11818 REASON_SET(&reason, PFRES_NORM); 11819 DPFPRINTF(PF_DEBUG_MISC, 11820 "dropping IPv6 packet with ICMPv4 payload"); 11821 break; 11822 } 11823 if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) { 11824 action = PF_DROP; 11825 REASON_SET(&reason, PFRES_NORM); 11826 DPFPRINTF(PF_DEBUG_MISC, 11827 "pf: dropping IPv4 packet with ICMPv6 payload"); 11828 break; 11829 } 11830 action = pf_test_state_icmp(&s, &pd, &reason); 11831 if (action == PF_PASS || action == PF_AFRT) { 11832 if (s != NULL) { 11833 if (V_pfsync_update_state_ptr != NULL) 11834 V_pfsync_update_state_ptr(s); 11835 r = s->rule; 11836 a = s->anchor; 11837 } 11838 } else if (s == NULL) 11839 action = pf_test_rule(&r, &s, &pd, 11840 &a, &ruleset, &reason, inp, &match_rules); 11841 break; 11842 } 11843 11844 } 11845 11846 done: 11847 PF_RULES_RUNLOCK(); 11848 11849 /* if packet sits in reassembly queue, return without error */ 11850 if (pd.m == NULL) { 11851 pf_free_match_rules(&match_rules); 11852 goto eat_pkt; 11853 } 11854 11855 if (s) 11856 memcpy(&pd.act, &s->act, sizeof(s->act)); 11857 11858 if (action == PF_PASS && pd.badopts != 0 && !pd.act.allow_opts) { 11859 action = PF_DROP; 11860 REASON_SET(&reason, PFRES_IPOPTIONS); 11861 pd.act.log = PF_LOG_FORCE; 11862 DPFPRINTF(PF_DEBUG_MISC, 11863 "pf: dropping packet with dangerous headers"); 11864 } 11865 11866 if (pd.act.max_pkt_size && pd.act.max_pkt_size && 11867 pd.tot_len > pd.act.max_pkt_size) { 11868 action = PF_DROP; 11869 REASON_SET(&reason, PFRES_NORM); 11870 pd.act.log = PF_LOG_FORCE; 11871 DPFPRINTF(PF_DEBUG_MISC, 11872 "pf: dropping overly long packet"); 11873 } 11874 11875 if (s) { 11876 uint8_t log = pd.act.log; 11877 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 11878 pd.act.log |= log; 11879 tag = s->tag; 11880 } else { 11881 tag = r->tag; 11882 } 11883 11884 if (tag > 0 && pf_tag_packet(&pd, tag)) { 11885 action = PF_DROP; 11886 REASON_SET(&reason, PFRES_MEMORY); 11887 } 11888 11889 pf_scrub(&pd); 11890 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 11891 pf_normalize_mss(&pd); 11892 11893 if (pd.act.rtableid >= 0) 11894 M_SETFIB(pd.m, pd.act.rtableid); 11895 11896 if (pd.act.flags & PFSTATE_SETPRIO) { 11897 if (pd.tos & IPTOS_LOWDELAY) 11898 use_2nd_queue = 1; 11899 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 11900 action = PF_DROP; 11901 REASON_SET(&reason, PFRES_MEMORY); 11902 pd.act.log = PF_LOG_FORCE; 11903 DPFPRINTF(PF_DEBUG_MISC, 11904 "pf: failed to allocate 802.1q mtag"); 11905 } 11906 } 11907 11908 #ifdef ALTQ 11909 if (action == PF_PASS && pd.act.qid) { 11910 if (pd.pf_mtag == NULL && 11911 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11912 action = PF_DROP; 11913 REASON_SET(&reason, PFRES_MEMORY); 11914 } else { 11915 if (s != NULL) 11916 pd.pf_mtag->qid_hash = pf_state_hash(s); 11917 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 11918 pd.pf_mtag->qid = pd.act.pqid; 11919 else 11920 pd.pf_mtag->qid = pd.act.qid; 11921 /* Add hints for ecn. */ 11922 pd.pf_mtag->hdr = mtod(pd.m, void *); 11923 } 11924 } 11925 #endif /* ALTQ */ 11926 11927 /* 11928 * connections redirected to loopback should not match sockets 11929 * bound specifically to loopback due to security implications, 11930 * see tcp_input() and in_pcblookup_listen(). 11931 */ 11932 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 11933 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 11934 (s->nat_rule->action == PF_RDR || 11935 s->nat_rule->action == PF_BINAT) && 11936 pf_is_loopback(af, pd.dst)) 11937 pd.m->m_flags |= M_SKIP_FIREWALL; 11938 11939 if (action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 11940 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 11941 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 11942 if (__predict_true(mtag != NULL && ip_divert_ptr != NULL)) { 11943 ((struct pf_divert_mtag *)(mtag+1))->port = 11944 ntohs(r->divert.port); 11945 ((struct pf_divert_mtag *)(mtag+1))->idir = 11946 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 11947 PF_DIVERT_MTAG_DIR_OUT; 11948 11949 pf_counters_inc(action, &pd, s, r, a, &match_rules); 11950 11951 if (s) 11952 PF_STATE_UNLOCK(s); 11953 11954 m_tag_prepend(pd.m, mtag); 11955 if (pd.m->m_flags & M_FASTFWD_OURS) { 11956 if (pd.pf_mtag == NULL && 11957 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 11958 action = PF_DROP; 11959 REASON_SET(&reason, PFRES_MEMORY); 11960 pd.act.log = PF_LOG_FORCE; 11961 DPFPRINTF(PF_DEBUG_MISC, 11962 "pf: failed to allocate tag"); 11963 } else { 11964 pd.pf_mtag->flags |= 11965 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 11966 pd.m->m_flags &= ~M_FASTFWD_OURS; 11967 } 11968 } 11969 ip_divert_ptr(*m0, s != NULL ? s->id : 0, dir == PF_IN); 11970 *m0 = NULL; 11971 return (action); 11972 } else if (mtag == NULL) { 11973 /* XXX: ipfw has the same behaviour! */ 11974 action = PF_DROP; 11975 REASON_SET(&reason, PFRES_MEMORY); 11976 pd.act.log = PF_LOG_FORCE; 11977 DPFPRINTF(PF_DEBUG_MISC, 11978 "pf: failed to allocate divert tag"); 11979 } else { 11980 action = PF_DROP; 11981 REASON_SET(&reason, PFRES_MATCH); 11982 pd.act.log = PF_LOG_FORCE; 11983 DPFPRINTF(PF_DEBUG_MISC, 11984 "pf: divert(4) is not loaded"); 11985 } 11986 } 11987 11988 /* this flag will need revising if the pkt is forwarded */ 11989 if (pd.pf_mtag) 11990 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 11991 11992 if (pd.act.log) { 11993 struct pf_krule *lr; 11994 11995 if (s != NULL && s->nat_rule != NULL && 11996 s->nat_rule->log & PF_LOG_ALL) 11997 lr = s->nat_rule; 11998 else 11999 lr = r; 12000 12001 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 12002 PFLOG_PACKET(action, reason, lr, a, 12003 ruleset, &pd, (s == NULL), NULL); 12004 if (s) { 12005 SLIST_FOREACH(ri, &s->match_rules, entry) 12006 if (ri->r->log & PF_LOG_ALL) 12007 PFLOG_PACKET(action, 12008 reason, ri->r, a, ruleset, &pd, 0, NULL); 12009 } 12010 } 12011 12012 pf_counters_inc(action, &pd, s, r, a, &match_rules); 12013 12014 switch (action) { 12015 case PF_SYNPROXY_DROP: 12016 m_freem(*m0); 12017 case PF_DEFER: 12018 *m0 = NULL; 12019 action = PF_PASS; 12020 break; 12021 case PF_DROP: 12022 m_freem(*m0); 12023 *m0 = NULL; 12024 break; 12025 case PF_AFRT: 12026 if (pf_translate_af(&pd, r)) { 12027 *m0 = pd.m; 12028 action = PF_DROP; 12029 break; 12030 } 12031 #ifdef INET 12032 if (pd.naf == AF_INET) { 12033 action = pf_route(r, kif->pfik_ifp, s, &pd, 12034 inp); 12035 } 12036 #endif /* INET */ 12037 #ifdef INET6 12038 if (pd.naf == AF_INET6) { 12039 action = pf_route6(r, kif->pfik_ifp, s, &pd, 12040 inp); 12041 } 12042 #endif /* INET6 */ 12043 *m0 = pd.m; 12044 goto out; 12045 break; 12046 default: 12047 if (pd.act.rt) { 12048 switch (af) { 12049 #ifdef INET 12050 case AF_INET: 12051 /* pf_route() returns unlocked. */ 12052 action = pf_route(r, kif->pfik_ifp, s, &pd, 12053 inp); 12054 break; 12055 #endif /* INET */ 12056 #ifdef INET6 12057 case AF_INET6: 12058 /* pf_route6() returns unlocked. */ 12059 action = pf_route6(r, kif->pfik_ifp, s, &pd, 12060 inp); 12061 break; 12062 #endif /* INET6 */ 12063 } 12064 *m0 = pd.m; 12065 goto out; 12066 } 12067 if (pf_dummynet(&pd, s, r, m0) != 0) { 12068 action = PF_DROP; 12069 REASON_SET(&reason, PFRES_MEMORY); 12070 } 12071 break; 12072 } 12073 12074 eat_pkt: 12075 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 12076 12077 if (s && action != PF_DROP) { 12078 if (!s->if_index_in && dir == PF_IN) 12079 s->if_index_in = ifp->if_index; 12080 else if (!s->if_index_out && dir == PF_OUT) 12081 s->if_index_out = ifp->if_index; 12082 } 12083 12084 if (s) 12085 PF_STATE_UNLOCK(s); 12086 12087 out: 12088 #ifdef INET6 12089 /* If reassembled packet passed, create new fragments. */ 12090 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 12091 (! (pflags & PF_PFIL_NOREFRAGMENT)) && 12092 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 12093 action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD); 12094 #endif /* INET6 */ 12095 12096 pf_sctp_multihome_delayed(&pd, kif, s, action); 12097 12098 return (action); 12099 } 12100 #endif /* INET || INET6 */ 12101