1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 125 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 126 "struct pf_kstate *"); 127 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 128 "struct pfi_kkif *"); 129 SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *", 130 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 131 SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int"); 132 SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int"); 133 SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *", 134 "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *"); 135 SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int"); 136 SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int"); 137 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 138 "struct pf_krule *", "struct mbuf *", "int"); 139 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 140 "struct pf_sctp_source *"); 141 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 142 "struct pf_kstate *", "struct pf_sctp_source *"); 143 144 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 145 "struct mbuf *"); 146 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 147 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 148 "int", "struct pf_keth_rule *", "char *"); 149 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 150 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 151 "int", "struct pf_keth_rule *"); 152 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 153 154 /* 155 * Global variables 156 */ 157 158 /* state tables */ 159 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 160 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 161 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 162 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 163 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 164 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 165 VNET_DEFINE(struct pf_kstatus, pf_status); 166 167 VNET_DEFINE(u_int32_t, ticket_altqs_active); 168 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 169 VNET_DEFINE(int, altqs_inactive_open); 170 VNET_DEFINE(u_int32_t, ticket_pabuf); 171 172 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 173 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 174 VNET_DEFINE(u_char, pf_tcp_secret[16]); 175 #define V_pf_tcp_secret VNET(pf_tcp_secret) 176 VNET_DEFINE(int, pf_tcp_secret_init); 177 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 178 VNET_DEFINE(int, pf_tcp_iss_off); 179 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 180 VNET_DECLARE(int, pf_vnet_active); 181 #define V_pf_vnet_active VNET(pf_vnet_active) 182 183 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 184 #define V_pf_purge_idx VNET(pf_purge_idx) 185 186 #ifdef PF_WANT_32_TO_64_COUNTER 187 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 188 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 189 190 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 191 VNET_DEFINE(size_t, pf_allrulecount); 192 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 193 #endif 194 195 struct pf_sctp_endpoint; 196 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 197 struct pf_sctp_source { 198 sa_family_t af; 199 struct pf_addr addr; 200 TAILQ_ENTRY(pf_sctp_source) entry; 201 }; 202 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 203 struct pf_sctp_endpoint 204 { 205 uint32_t v_tag; 206 struct pf_sctp_sources sources; 207 RB_ENTRY(pf_sctp_endpoint) entry; 208 }; 209 static int 210 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 211 { 212 return (a->v_tag - b->v_tag); 213 } 214 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 215 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 216 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 217 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 218 static struct mtx_padalign pf_sctp_endpoints_mtx; 219 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 220 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 221 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 222 223 /* 224 * Queue for pf_intr() sends. 225 */ 226 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 227 struct pf_send_entry { 228 STAILQ_ENTRY(pf_send_entry) pfse_next; 229 struct mbuf *pfse_m; 230 enum { 231 PFSE_IP, 232 PFSE_IP6, 233 PFSE_ICMP, 234 PFSE_ICMP6, 235 } pfse_type; 236 struct { 237 int type; 238 int code; 239 int mtu; 240 } icmpopts; 241 }; 242 243 STAILQ_HEAD(pf_send_head, pf_send_entry); 244 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 245 #define V_pf_sendqueue VNET(pf_sendqueue) 246 247 static struct mtx_padalign pf_sendqueue_mtx; 248 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 249 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 250 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 251 252 /* 253 * Queue for pf_overload_task() tasks. 254 */ 255 struct pf_overload_entry { 256 SLIST_ENTRY(pf_overload_entry) next; 257 struct pf_addr addr; 258 sa_family_t af; 259 uint8_t dir; 260 struct pf_krule *rule; 261 }; 262 263 SLIST_HEAD(pf_overload_head, pf_overload_entry); 264 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 265 #define V_pf_overloadqueue VNET(pf_overloadqueue) 266 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 267 #define V_pf_overloadtask VNET(pf_overloadtask) 268 269 static struct mtx_padalign pf_overloadqueue_mtx; 270 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 271 "pf overload/flush queue", MTX_DEF); 272 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 273 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 274 275 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 276 struct mtx_padalign pf_unlnkdrules_mtx; 277 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 278 MTX_DEF); 279 280 struct sx pf_config_lock; 281 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 282 283 struct mtx_padalign pf_table_stats_lock; 284 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 285 MTX_DEF); 286 287 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 288 #define V_pf_sources_z VNET(pf_sources_z) 289 uma_zone_t pf_mtag_z; 290 VNET_DEFINE(uma_zone_t, pf_state_z); 291 VNET_DEFINE(uma_zone_t, pf_state_key_z); 292 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 293 294 VNET_DEFINE(struct unrhdr64, pf_stateid); 295 296 static void pf_src_tree_remove_state(struct pf_kstate *); 297 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 298 u_int32_t); 299 static void pf_add_threshold(struct pf_threshold *); 300 static int pf_check_threshold(struct pf_threshold *); 301 302 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 303 u_int16_t *, u_int16_t *, struct pf_addr *, 304 u_int16_t, u_int8_t, sa_family_t); 305 static int pf_modulate_sack(struct pf_pdesc *, 306 struct tcphdr *, struct pf_state_peer *); 307 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 308 int *, u_int16_t *, u_int16_t *); 309 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 310 struct pf_addr *, struct pf_addr *, u_int16_t, 311 u_int16_t *, u_int16_t *, u_int16_t *, 312 u_int16_t *, u_int8_t, sa_family_t); 313 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 314 sa_family_t, struct pf_krule *, int); 315 static void pf_detach_state(struct pf_kstate *); 316 static int pf_state_key_attach(struct pf_state_key *, 317 struct pf_state_key *, struct pf_kstate *); 318 static void pf_state_key_detach(struct pf_kstate *, int); 319 static int pf_state_key_ctor(void *, int, void *, int); 320 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 321 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 322 struct pf_mtag *pf_mtag); 323 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 324 struct pf_krule *, struct mbuf **); 325 static int pf_dummynet_route(struct pf_pdesc *, 326 struct pf_kstate *, struct pf_krule *, 327 struct ifnet *, struct sockaddr *, struct mbuf **); 328 static int pf_test_eth_rule(int, struct pfi_kkif *, 329 struct mbuf **); 330 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 331 struct pf_pdesc *, struct pf_krule **, 332 struct pf_kruleset **, struct inpcb *); 333 static int pf_create_state(struct pf_krule *, struct pf_krule *, 334 struct pf_krule *, struct pf_pdesc *, 335 struct pf_ksrc_node *, struct pf_state_key *, 336 struct pf_state_key *, 337 u_int16_t, u_int16_t, int *, 338 struct pf_kstate **, int, u_int16_t, u_int16_t, 339 struct pf_krule_slist *, struct pf_udp_mapping *); 340 static int pf_state_key_addr_setup(struct pf_pdesc *, 341 struct pf_state_key_cmp *, int); 342 static int pf_tcp_track_full(struct pf_kstate **, 343 struct pf_pdesc *, u_short *, int *); 344 static int pf_tcp_track_sloppy(struct pf_kstate **, 345 struct pf_pdesc *, u_short *); 346 static int pf_test_state_tcp(struct pf_kstate **, 347 struct pf_pdesc *, u_short *); 348 static int pf_test_state_udp(struct pf_kstate **, 349 struct pf_pdesc *); 350 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 351 struct pf_pdesc *, struct pf_kstate **, 352 int, u_int16_t, u_int16_t, 353 int, int *, int, int); 354 static int pf_test_state_icmp(struct pf_kstate **, 355 struct pf_pdesc *, u_short *); 356 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 357 static void pf_sctp_multihome_delayed(struct pf_pdesc *, 358 struct pfi_kkif *, struct pf_kstate *, int); 359 static int pf_test_state_sctp(struct pf_kstate **, 360 struct pf_pdesc *, u_short *); 361 static int pf_test_state_other(struct pf_kstate **, 362 struct pf_pdesc *); 363 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 364 int, u_int16_t); 365 static int pf_check_proto_cksum(struct mbuf *, int, int, 366 u_int8_t, sa_family_t); 367 static int pf_walk_option6(struct mbuf *, int, int, uint32_t *, 368 u_short *); 369 static void pf_print_state_parts(struct pf_kstate *, 370 struct pf_state_key *, struct pf_state_key *); 371 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 372 bool, u_int8_t); 373 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 374 const struct pf_state_key_cmp *, u_int); 375 static int pf_src_connlimit(struct pf_kstate **); 376 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 377 static void pf_counters_inc(int, struct pf_pdesc *, 378 struct pf_kstate *, struct pf_krule *, 379 struct pf_krule *); 380 static void pf_overload_task(void *v, int pending); 381 static u_short pf_insert_src_node(struct pf_ksrc_node **, 382 struct pf_krule *, struct pf_addr *, sa_family_t); 383 static u_int pf_purge_expired_states(u_int, int); 384 static void pf_purge_unlinked_rules(void); 385 static int pf_mtag_uminit(void *, int, int); 386 static void pf_mtag_free(struct m_tag *); 387 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 388 int, struct pf_state_key *); 389 #ifdef INET 390 static void pf_route(struct mbuf **, struct pf_krule *, 391 struct ifnet *, struct pf_kstate *, 392 struct pf_pdesc *, struct inpcb *); 393 #endif /* INET */ 394 #ifdef INET6 395 static void pf_change_a6(struct pf_addr *, u_int16_t *, 396 struct pf_addr *, u_int8_t); 397 static void pf_route6(struct mbuf **, struct pf_krule *, 398 struct ifnet *, struct pf_kstate *, 399 struct pf_pdesc *, struct inpcb *); 400 #endif /* INET6 */ 401 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 402 403 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 404 405 extern int pf_end_threads; 406 extern struct proc *pf_purge_proc; 407 408 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 409 410 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 411 412 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 413 do { \ 414 struct pf_state_key *nk; \ 415 if ((pd->dir) == PF_OUT) \ 416 nk = (_s)->key[PF_SK_STACK]; \ 417 else \ 418 nk = (_s)->key[PF_SK_WIRE]; \ 419 pf_packet_rework_nat(_m, _pd, _off, nk); \ 420 } while (0) 421 422 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 423 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 424 425 #define STATE_LOOKUP(k, s, pd) \ 426 do { \ 427 (s) = pf_find_state((pd->kif), (k), (pd->dir)); \ 428 SDT_PROBE5(pf, ip, state, lookup, pd->kif, k, (pd->dir), pd, (s)); \ 429 if ((s) == NULL) \ 430 return (PF_DROP); \ 431 if (PACKET_LOOPED(pd)) \ 432 return (PF_PASS); \ 433 } while (0) 434 435 static struct pfi_kkif * 436 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 437 { 438 SDT_PROBE2(pf, ip, , bound_iface, st, k); 439 440 /* Floating unless otherwise specified. */ 441 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 442 return (V_pfi_all); 443 444 /* 445 * Initially set to all, because we don't know what interface we'll be 446 * sending this out when we create the state. 447 */ 448 if (st->rule->rt == PF_REPLYTO) 449 return (V_pfi_all); 450 451 /* Don't overrule the interface for states created on incoming packets. */ 452 if (st->direction == PF_IN) 453 return (k); 454 455 /* No route-to, so don't overrule. */ 456 if (st->rt != PF_ROUTETO) 457 return (k); 458 459 /* Bind to the route-to interface. */ 460 return (st->rt_kif); 461 } 462 463 #define STATE_INC_COUNTERS(s) \ 464 do { \ 465 struct pf_krule_item *mrm; \ 466 counter_u64_add(s->rule->states_cur, 1); \ 467 counter_u64_add(s->rule->states_tot, 1); \ 468 if (s->anchor != NULL) { \ 469 counter_u64_add(s->anchor->states_cur, 1); \ 470 counter_u64_add(s->anchor->states_tot, 1); \ 471 } \ 472 if (s->nat_rule != NULL) { \ 473 counter_u64_add(s->nat_rule->states_cur, 1);\ 474 counter_u64_add(s->nat_rule->states_tot, 1);\ 475 } \ 476 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 477 counter_u64_add(mrm->r->states_cur, 1); \ 478 counter_u64_add(mrm->r->states_tot, 1); \ 479 } \ 480 } while (0) 481 482 #define STATE_DEC_COUNTERS(s) \ 483 do { \ 484 struct pf_krule_item *mrm; \ 485 if (s->nat_rule != NULL) \ 486 counter_u64_add(s->nat_rule->states_cur, -1);\ 487 if (s->anchor != NULL) \ 488 counter_u64_add(s->anchor->states_cur, -1); \ 489 counter_u64_add(s->rule->states_cur, -1); \ 490 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 491 counter_u64_add(mrm->r->states_cur, -1); \ 492 } while (0) 493 494 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 495 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 496 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 497 VNET_DEFINE(struct pf_idhash *, pf_idhash); 498 VNET_DEFINE(struct pf_srchash *, pf_srchash); 499 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 500 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 501 502 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 503 "pf(4)"); 504 505 VNET_DEFINE(u_long, pf_hashmask); 506 VNET_DEFINE(u_long, pf_srchashmask); 507 VNET_DEFINE(u_long, pf_udpendpointhashmask); 508 VNET_DEFINE_STATIC(u_long, pf_hashsize); 509 #define V_pf_hashsize VNET(pf_hashsize) 510 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 511 #define V_pf_srchashsize VNET(pf_srchashsize) 512 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 513 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 514 u_long pf_ioctl_maxcount = 65535; 515 516 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 517 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 518 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 519 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 520 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 521 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 522 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 523 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 524 525 VNET_DEFINE(void *, pf_swi_cookie); 526 VNET_DEFINE(struct intr_event *, pf_swi_ie); 527 528 VNET_DEFINE(uint32_t, pf_hashseed); 529 #define V_pf_hashseed VNET(pf_hashseed) 530 531 static void 532 pf_sctp_checksum(struct mbuf *m, int off) 533 { 534 uint32_t sum = 0; 535 536 /* Zero out the checksum, to enable recalculation. */ 537 m_copyback(m, off + offsetof(struct sctphdr, checksum), 538 sizeof(sum), (caddr_t)&sum); 539 540 sum = sctp_calculate_cksum(m, off); 541 542 m_copyback(m, off + offsetof(struct sctphdr, checksum), 543 sizeof(sum), (caddr_t)&sum); 544 } 545 546 int 547 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 548 { 549 550 switch (af) { 551 #ifdef INET 552 case AF_INET: 553 if (a->addr32[0] > b->addr32[0]) 554 return (1); 555 if (a->addr32[0] < b->addr32[0]) 556 return (-1); 557 break; 558 #endif /* INET */ 559 #ifdef INET6 560 case AF_INET6: 561 if (a->addr32[3] > b->addr32[3]) 562 return (1); 563 if (a->addr32[3] < b->addr32[3]) 564 return (-1); 565 if (a->addr32[2] > b->addr32[2]) 566 return (1); 567 if (a->addr32[2] < b->addr32[2]) 568 return (-1); 569 if (a->addr32[1] > b->addr32[1]) 570 return (1); 571 if (a->addr32[1] < b->addr32[1]) 572 return (-1); 573 if (a->addr32[0] > b->addr32[0]) 574 return (1); 575 if (a->addr32[0] < b->addr32[0]) 576 return (-1); 577 break; 578 #endif /* INET6 */ 579 } 580 return (0); 581 } 582 583 static bool 584 pf_is_loopback(sa_family_t af, struct pf_addr *addr) 585 { 586 switch (af) { 587 #ifdef INET 588 case AF_INET: 589 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 590 #endif 591 case AF_INET6: 592 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 593 default: 594 panic("Unknown af %d", af); 595 } 596 } 597 598 static void 599 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 600 struct pf_state_key *nk) 601 { 602 603 switch (pd->proto) { 604 case IPPROTO_TCP: { 605 struct tcphdr *th = &pd->hdr.tcp; 606 607 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 608 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 609 &th->th_sum, &nk->addr[pd->sidx], 610 nk->port[pd->sidx], 0, pd->af); 611 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 612 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 613 &th->th_sum, &nk->addr[pd->didx], 614 nk->port[pd->didx], 0, pd->af); 615 m_copyback(m, off, sizeof(*th), (caddr_t)th); 616 break; 617 } 618 case IPPROTO_UDP: { 619 struct udphdr *uh = &pd->hdr.udp; 620 621 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 622 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 623 &uh->uh_sum, &nk->addr[pd->sidx], 624 nk->port[pd->sidx], 1, pd->af); 625 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 626 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 627 &uh->uh_sum, &nk->addr[pd->didx], 628 nk->port[pd->didx], 1, pd->af); 629 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 630 break; 631 } 632 case IPPROTO_SCTP: { 633 struct sctphdr *sh = &pd->hdr.sctp; 634 uint16_t checksum = 0; 635 636 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 637 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 638 &checksum, &nk->addr[pd->sidx], 639 nk->port[pd->sidx], 1, pd->af); 640 } 641 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 642 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 643 &checksum, &nk->addr[pd->didx], 644 nk->port[pd->didx], 1, pd->af); 645 } 646 647 break; 648 } 649 case IPPROTO_ICMP: { 650 struct icmp *ih = &pd->hdr.icmp; 651 652 if (nk->port[pd->sidx] != ih->icmp_id) { 653 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 654 ih->icmp_cksum, ih->icmp_id, 655 nk->port[pd->sidx], 0); 656 ih->icmp_id = nk->port[pd->sidx]; 657 pd->sport = &ih->icmp_id; 658 659 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 660 } 661 /* FALLTHROUGH */ 662 } 663 default: 664 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 665 switch (pd->af) { 666 case AF_INET: 667 pf_change_a(&pd->src->v4.s_addr, 668 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 669 0); 670 break; 671 case AF_INET6: 672 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 673 break; 674 } 675 } 676 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 677 switch (pd->af) { 678 case AF_INET: 679 pf_change_a(&pd->dst->v4.s_addr, 680 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 681 0); 682 break; 683 case AF_INET6: 684 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 685 break; 686 } 687 } 688 break; 689 } 690 } 691 692 static __inline uint32_t 693 pf_hashkey(const struct pf_state_key *sk) 694 { 695 uint32_t h; 696 697 h = murmur3_32_hash32((const uint32_t *)sk, 698 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 699 V_pf_hashseed); 700 701 return (h & V_pf_hashmask); 702 } 703 704 static __inline uint32_t 705 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 706 { 707 uint32_t h; 708 709 switch (af) { 710 case AF_INET: 711 h = murmur3_32_hash32((uint32_t *)&addr->v4, 712 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 713 break; 714 case AF_INET6: 715 h = murmur3_32_hash32((uint32_t *)&addr->v6, 716 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 717 break; 718 } 719 720 return (h & V_pf_srchashmask); 721 } 722 723 static inline uint32_t 724 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 725 { 726 uint32_t h; 727 728 h = murmur3_32_hash32((uint32_t *)endpoint, 729 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 730 V_pf_hashseed); 731 return (h & V_pf_udpendpointhashmask); 732 } 733 734 #ifdef ALTQ 735 static int 736 pf_state_hash(struct pf_kstate *s) 737 { 738 u_int32_t hv = (intptr_t)s / sizeof(*s); 739 740 hv ^= crc32(&s->src, sizeof(s->src)); 741 hv ^= crc32(&s->dst, sizeof(s->dst)); 742 if (hv == 0) 743 hv = 1; 744 return (hv); 745 } 746 #endif 747 748 static __inline void 749 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 750 { 751 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 752 s->dst.state = newstate; 753 if (which == PF_PEER_DST) 754 return; 755 if (s->src.state == newstate) 756 return; 757 if (s->creatorid == V_pf_status.hostid && 758 s->key[PF_SK_STACK] != NULL && 759 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 760 !(TCPS_HAVEESTABLISHED(s->src.state) || 761 s->src.state == TCPS_CLOSED) && 762 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 763 atomic_add_32(&V_pf_status.states_halfopen, -1); 764 765 s->src.state = newstate; 766 } 767 768 #ifdef INET6 769 void 770 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 771 { 772 switch (af) { 773 #ifdef INET 774 case AF_INET: 775 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 776 break; 777 #endif /* INET */ 778 case AF_INET6: 779 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 780 break; 781 } 782 } 783 #endif /* INET6 */ 784 785 static void 786 pf_init_threshold(struct pf_threshold *threshold, 787 u_int32_t limit, u_int32_t seconds) 788 { 789 threshold->limit = limit * PF_THRESHOLD_MULT; 790 threshold->seconds = seconds; 791 threshold->count = 0; 792 threshold->last = time_uptime; 793 } 794 795 static void 796 pf_add_threshold(struct pf_threshold *threshold) 797 { 798 u_int32_t t = time_uptime, diff = t - threshold->last; 799 800 if (diff >= threshold->seconds) 801 threshold->count = 0; 802 else 803 threshold->count -= threshold->count * diff / 804 threshold->seconds; 805 threshold->count += PF_THRESHOLD_MULT; 806 threshold->last = t; 807 } 808 809 static int 810 pf_check_threshold(struct pf_threshold *threshold) 811 { 812 return (threshold->count > threshold->limit); 813 } 814 815 static int 816 pf_src_connlimit(struct pf_kstate **state) 817 { 818 struct pf_overload_entry *pfoe; 819 int bad = 0; 820 821 PF_STATE_LOCK_ASSERT(*state); 822 /* 823 * XXXKS: The src node is accessed unlocked! 824 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 825 */ 826 827 (*state)->src_node->conn++; 828 (*state)->src.tcp_est = 1; 829 pf_add_threshold(&(*state)->src_node->conn_rate); 830 831 if ((*state)->rule->max_src_conn && 832 (*state)->rule->max_src_conn < 833 (*state)->src_node->conn) { 834 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 835 bad++; 836 } 837 838 if ((*state)->rule->max_src_conn_rate.limit && 839 pf_check_threshold(&(*state)->src_node->conn_rate)) { 840 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 841 bad++; 842 } 843 844 if (!bad) 845 return (0); 846 847 /* Kill this state. */ 848 (*state)->timeout = PFTM_PURGE; 849 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 850 851 if ((*state)->rule->overload_tbl == NULL) 852 return (1); 853 854 /* Schedule overloading and flushing task. */ 855 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 856 if (pfoe == NULL) 857 return (1); /* too bad :( */ 858 859 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 860 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 861 pfoe->rule = (*state)->rule; 862 pfoe->dir = (*state)->direction; 863 PF_OVERLOADQ_LOCK(); 864 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 865 PF_OVERLOADQ_UNLOCK(); 866 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 867 868 return (1); 869 } 870 871 static void 872 pf_overload_task(void *v, int pending) 873 { 874 struct pf_overload_head queue; 875 struct pfr_addr p; 876 struct pf_overload_entry *pfoe, *pfoe1; 877 uint32_t killed = 0; 878 879 CURVNET_SET((struct vnet *)v); 880 881 PF_OVERLOADQ_LOCK(); 882 queue = V_pf_overloadqueue; 883 SLIST_INIT(&V_pf_overloadqueue); 884 PF_OVERLOADQ_UNLOCK(); 885 886 bzero(&p, sizeof(p)); 887 SLIST_FOREACH(pfoe, &queue, next) { 888 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 889 if (V_pf_status.debug >= PF_DEBUG_MISC) { 890 printf("%s: blocking address ", __func__); 891 pf_print_host(&pfoe->addr, 0, pfoe->af); 892 printf("\n"); 893 } 894 895 p.pfra_af = pfoe->af; 896 switch (pfoe->af) { 897 #ifdef INET 898 case AF_INET: 899 p.pfra_net = 32; 900 p.pfra_ip4addr = pfoe->addr.v4; 901 break; 902 #endif 903 #ifdef INET6 904 case AF_INET6: 905 p.pfra_net = 128; 906 p.pfra_ip6addr = pfoe->addr.v6; 907 break; 908 #endif 909 } 910 911 PF_RULES_WLOCK(); 912 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 913 PF_RULES_WUNLOCK(); 914 } 915 916 /* 917 * Remove those entries, that don't need flushing. 918 */ 919 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 920 if (pfoe->rule->flush == 0) { 921 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 922 free(pfoe, M_PFTEMP); 923 } else 924 counter_u64_add( 925 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 926 927 /* If nothing to flush, return. */ 928 if (SLIST_EMPTY(&queue)) { 929 CURVNET_RESTORE(); 930 return; 931 } 932 933 for (int i = 0; i <= V_pf_hashmask; i++) { 934 struct pf_idhash *ih = &V_pf_idhash[i]; 935 struct pf_state_key *sk; 936 struct pf_kstate *s; 937 938 PF_HASHROW_LOCK(ih); 939 LIST_FOREACH(s, &ih->states, entry) { 940 sk = s->key[PF_SK_WIRE]; 941 SLIST_FOREACH(pfoe, &queue, next) 942 if (sk->af == pfoe->af && 943 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 944 pfoe->rule == s->rule) && 945 ((pfoe->dir == PF_OUT && 946 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 947 (pfoe->dir == PF_IN && 948 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 949 s->timeout = PFTM_PURGE; 950 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 951 killed++; 952 } 953 } 954 PF_HASHROW_UNLOCK(ih); 955 } 956 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 957 free(pfoe, M_PFTEMP); 958 if (V_pf_status.debug >= PF_DEBUG_MISC) 959 printf("%s: %u states killed", __func__, killed); 960 961 CURVNET_RESTORE(); 962 } 963 964 /* 965 * Can return locked on failure, so that we can consistently 966 * allocate and insert a new one. 967 */ 968 struct pf_ksrc_node * 969 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 970 struct pf_srchash **sh, bool returnlocked) 971 { 972 struct pf_ksrc_node *n; 973 974 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 975 976 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 977 PF_HASHROW_LOCK(*sh); 978 LIST_FOREACH(n, &(*sh)->nodes, entry) 979 if (n->rule == rule && n->af == af && 980 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 981 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 982 break; 983 984 if (n != NULL) { 985 n->states++; 986 PF_HASHROW_UNLOCK(*sh); 987 } else if (returnlocked == false) 988 PF_HASHROW_UNLOCK(*sh); 989 990 return (n); 991 } 992 993 static void 994 pf_free_src_node(struct pf_ksrc_node *sn) 995 { 996 997 for (int i = 0; i < 2; i++) { 998 counter_u64_free(sn->bytes[i]); 999 counter_u64_free(sn->packets[i]); 1000 } 1001 uma_zfree(V_pf_sources_z, sn); 1002 } 1003 1004 static u_short 1005 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 1006 struct pf_addr *src, sa_family_t af) 1007 { 1008 u_short reason = 0; 1009 struct pf_srchash *sh = NULL; 1010 1011 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1012 rule->rpool.opts & PF_POOL_STICKYADDR), 1013 ("%s for non-tracking rule %p", __func__, rule)); 1014 1015 if (*sn == NULL) 1016 *sn = pf_find_src_node(src, rule, af, &sh, true); 1017 1018 if (*sn == NULL) { 1019 PF_HASHROW_ASSERT(sh); 1020 1021 if (rule->max_src_nodes && 1022 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1023 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1024 PF_HASHROW_UNLOCK(sh); 1025 reason = PFRES_SRCLIMIT; 1026 goto done; 1027 } 1028 1029 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1030 if ((*sn) == NULL) { 1031 PF_HASHROW_UNLOCK(sh); 1032 reason = PFRES_MEMORY; 1033 goto done; 1034 } 1035 1036 for (int i = 0; i < 2; i++) { 1037 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1038 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1039 1040 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1041 pf_free_src_node(*sn); 1042 PF_HASHROW_UNLOCK(sh); 1043 reason = PFRES_MEMORY; 1044 goto done; 1045 } 1046 } 1047 1048 pf_init_threshold(&(*sn)->conn_rate, 1049 rule->max_src_conn_rate.limit, 1050 rule->max_src_conn_rate.seconds); 1051 1052 MPASS((*sn)->lock == NULL); 1053 (*sn)->lock = &sh->lock; 1054 1055 (*sn)->af = af; 1056 (*sn)->rule = rule; 1057 PF_ACPY(&(*sn)->addr, src, af); 1058 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1059 (*sn)->creation = time_uptime; 1060 (*sn)->ruletype = rule->action; 1061 (*sn)->states = 1; 1062 if ((*sn)->rule != NULL) 1063 counter_u64_add((*sn)->rule->src_nodes, 1); 1064 PF_HASHROW_UNLOCK(sh); 1065 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1066 } else { 1067 if (rule->max_src_states && 1068 (*sn)->states >= rule->max_src_states) { 1069 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1070 1); 1071 reason = PFRES_SRCLIMIT; 1072 goto done; 1073 } 1074 } 1075 done: 1076 return (reason); 1077 } 1078 1079 void 1080 pf_unlink_src_node(struct pf_ksrc_node *src) 1081 { 1082 PF_SRC_NODE_LOCK_ASSERT(src); 1083 1084 LIST_REMOVE(src, entry); 1085 if (src->rule) 1086 counter_u64_add(src->rule->src_nodes, -1); 1087 } 1088 1089 u_int 1090 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1091 { 1092 struct pf_ksrc_node *sn, *tmp; 1093 u_int count = 0; 1094 1095 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1096 pf_free_src_node(sn); 1097 count++; 1098 } 1099 1100 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1101 1102 return (count); 1103 } 1104 1105 void 1106 pf_mtag_initialize(void) 1107 { 1108 1109 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1110 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1111 UMA_ALIGN_PTR, 0); 1112 } 1113 1114 /* Per-vnet data storage structures initialization. */ 1115 void 1116 pf_initialize(void) 1117 { 1118 struct pf_keyhash *kh; 1119 struct pf_idhash *ih; 1120 struct pf_srchash *sh; 1121 struct pf_udpendpointhash *uh; 1122 u_int i; 1123 1124 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1125 V_pf_hashsize = PF_HASHSIZ; 1126 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1127 V_pf_srchashsize = PF_SRCHASHSIZ; 1128 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1129 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1130 1131 V_pf_hashseed = arc4random(); 1132 1133 /* States and state keys storage. */ 1134 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1135 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1136 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1137 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1138 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1139 1140 V_pf_state_key_z = uma_zcreate("pf state keys", 1141 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1142 UMA_ALIGN_PTR, 0); 1143 1144 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1145 M_PFHASH, M_NOWAIT | M_ZERO); 1146 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1147 M_PFHASH, M_NOWAIT | M_ZERO); 1148 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1149 printf("pf: Unable to allocate memory for " 1150 "state_hashsize %lu.\n", V_pf_hashsize); 1151 1152 free(V_pf_keyhash, M_PFHASH); 1153 free(V_pf_idhash, M_PFHASH); 1154 1155 V_pf_hashsize = PF_HASHSIZ; 1156 V_pf_keyhash = mallocarray(V_pf_hashsize, 1157 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1158 V_pf_idhash = mallocarray(V_pf_hashsize, 1159 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1160 } 1161 1162 V_pf_hashmask = V_pf_hashsize - 1; 1163 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1164 i++, kh++, ih++) { 1165 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1166 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1167 } 1168 1169 /* Source nodes. */ 1170 V_pf_sources_z = uma_zcreate("pf source nodes", 1171 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1172 0); 1173 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1174 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1175 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1176 1177 V_pf_srchash = mallocarray(V_pf_srchashsize, 1178 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1179 if (V_pf_srchash == NULL) { 1180 printf("pf: Unable to allocate memory for " 1181 "source_hashsize %lu.\n", V_pf_srchashsize); 1182 1183 V_pf_srchashsize = PF_SRCHASHSIZ; 1184 V_pf_srchash = mallocarray(V_pf_srchashsize, 1185 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1186 } 1187 1188 V_pf_srchashmask = V_pf_srchashsize - 1; 1189 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1190 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1191 1192 1193 /* UDP endpoint mappings. */ 1194 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1195 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1196 UMA_ALIGN_PTR, 0); 1197 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1198 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1199 if (V_pf_udpendpointhash == NULL) { 1200 printf("pf: Unable to allocate memory for " 1201 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1202 1203 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1204 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1205 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1206 } 1207 1208 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1209 for (i = 0, uh = V_pf_udpendpointhash; 1210 i <= V_pf_udpendpointhashmask; 1211 i++, uh++) { 1212 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1213 MTX_DEF | MTX_DUPOK); 1214 } 1215 1216 /* ALTQ */ 1217 TAILQ_INIT(&V_pf_altqs[0]); 1218 TAILQ_INIT(&V_pf_altqs[1]); 1219 TAILQ_INIT(&V_pf_altqs[2]); 1220 TAILQ_INIT(&V_pf_altqs[3]); 1221 TAILQ_INIT(&V_pf_pabuf); 1222 V_pf_altqs_active = &V_pf_altqs[0]; 1223 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1224 V_pf_altqs_inactive = &V_pf_altqs[2]; 1225 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1226 1227 /* Send & overload+flush queues. */ 1228 STAILQ_INIT(&V_pf_sendqueue); 1229 SLIST_INIT(&V_pf_overloadqueue); 1230 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1231 1232 /* Unlinked, but may be referenced rules. */ 1233 TAILQ_INIT(&V_pf_unlinked_rules); 1234 } 1235 1236 void 1237 pf_mtag_cleanup(void) 1238 { 1239 1240 uma_zdestroy(pf_mtag_z); 1241 } 1242 1243 void 1244 pf_cleanup(void) 1245 { 1246 struct pf_keyhash *kh; 1247 struct pf_idhash *ih; 1248 struct pf_srchash *sh; 1249 struct pf_udpendpointhash *uh; 1250 struct pf_send_entry *pfse, *next; 1251 u_int i; 1252 1253 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1254 i <= V_pf_hashmask; 1255 i++, kh++, ih++) { 1256 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1257 __func__)); 1258 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1259 __func__)); 1260 mtx_destroy(&kh->lock); 1261 mtx_destroy(&ih->lock); 1262 } 1263 free(V_pf_keyhash, M_PFHASH); 1264 free(V_pf_idhash, M_PFHASH); 1265 1266 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1267 KASSERT(LIST_EMPTY(&sh->nodes), 1268 ("%s: source node hash not empty", __func__)); 1269 mtx_destroy(&sh->lock); 1270 } 1271 free(V_pf_srchash, M_PFHASH); 1272 1273 for (i = 0, uh = V_pf_udpendpointhash; 1274 i <= V_pf_udpendpointhashmask; 1275 i++, uh++) { 1276 KASSERT(LIST_EMPTY(&uh->endpoints), 1277 ("%s: udp endpoint hash not empty", __func__)); 1278 mtx_destroy(&uh->lock); 1279 } 1280 free(V_pf_udpendpointhash, M_PFHASH); 1281 1282 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1283 m_freem(pfse->pfse_m); 1284 free(pfse, M_PFTEMP); 1285 } 1286 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1287 1288 uma_zdestroy(V_pf_sources_z); 1289 uma_zdestroy(V_pf_state_z); 1290 uma_zdestroy(V_pf_state_key_z); 1291 uma_zdestroy(V_pf_udp_mapping_z); 1292 } 1293 1294 static int 1295 pf_mtag_uminit(void *mem, int size, int how) 1296 { 1297 struct m_tag *t; 1298 1299 t = (struct m_tag *)mem; 1300 t->m_tag_cookie = MTAG_ABI_COMPAT; 1301 t->m_tag_id = PACKET_TAG_PF; 1302 t->m_tag_len = sizeof(struct pf_mtag); 1303 t->m_tag_free = pf_mtag_free; 1304 1305 return (0); 1306 } 1307 1308 static void 1309 pf_mtag_free(struct m_tag *t) 1310 { 1311 1312 uma_zfree(pf_mtag_z, t); 1313 } 1314 1315 struct pf_mtag * 1316 pf_get_mtag(struct mbuf *m) 1317 { 1318 struct m_tag *mtag; 1319 1320 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1321 return ((struct pf_mtag *)(mtag + 1)); 1322 1323 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1324 if (mtag == NULL) 1325 return (NULL); 1326 bzero(mtag + 1, sizeof(struct pf_mtag)); 1327 m_tag_prepend(m, mtag); 1328 1329 return ((struct pf_mtag *)(mtag + 1)); 1330 } 1331 1332 static int 1333 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1334 struct pf_kstate *s) 1335 { 1336 struct pf_keyhash *khs, *khw, *kh; 1337 struct pf_state_key *sk, *cur; 1338 struct pf_kstate *si, *olds = NULL; 1339 int idx; 1340 1341 NET_EPOCH_ASSERT(); 1342 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1343 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1344 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1345 1346 /* 1347 * We need to lock hash slots of both keys. To avoid deadlock 1348 * we always lock the slot with lower address first. Unlock order 1349 * isn't important. 1350 * 1351 * We also need to lock ID hash slot before dropping key 1352 * locks. On success we return with ID hash slot locked. 1353 */ 1354 1355 if (skw == sks) { 1356 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1357 PF_HASHROW_LOCK(khs); 1358 } else { 1359 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1360 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1361 if (khs == khw) { 1362 PF_HASHROW_LOCK(khs); 1363 } else if (khs < khw) { 1364 PF_HASHROW_LOCK(khs); 1365 PF_HASHROW_LOCK(khw); 1366 } else { 1367 PF_HASHROW_LOCK(khw); 1368 PF_HASHROW_LOCK(khs); 1369 } 1370 } 1371 1372 #define KEYS_UNLOCK() do { \ 1373 if (khs != khw) { \ 1374 PF_HASHROW_UNLOCK(khs); \ 1375 PF_HASHROW_UNLOCK(khw); \ 1376 } else \ 1377 PF_HASHROW_UNLOCK(khs); \ 1378 } while (0) 1379 1380 /* 1381 * First run: start with wire key. 1382 */ 1383 sk = skw; 1384 kh = khw; 1385 idx = PF_SK_WIRE; 1386 1387 MPASS(s->lock == NULL); 1388 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1389 1390 keyattach: 1391 LIST_FOREACH(cur, &kh->keys, entry) 1392 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1393 break; 1394 1395 if (cur != NULL) { 1396 /* Key exists. Check for same kif, if none, add to key. */ 1397 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1398 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1399 1400 PF_HASHROW_LOCK(ih); 1401 if (si->kif == s->kif && 1402 si->direction == s->direction) { 1403 if (sk->proto == IPPROTO_TCP && 1404 si->src.state >= TCPS_FIN_WAIT_2 && 1405 si->dst.state >= TCPS_FIN_WAIT_2) { 1406 /* 1407 * New state matches an old >FIN_WAIT_2 1408 * state. We can't drop key hash locks, 1409 * thus we can't unlink it properly. 1410 * 1411 * As a workaround we drop it into 1412 * TCPS_CLOSED state, schedule purge 1413 * ASAP and push it into the very end 1414 * of the slot TAILQ, so that it won't 1415 * conflict with our new state. 1416 */ 1417 pf_set_protostate(si, PF_PEER_BOTH, 1418 TCPS_CLOSED); 1419 si->timeout = PFTM_PURGE; 1420 olds = si; 1421 } else { 1422 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1423 printf("pf: %s key attach " 1424 "failed on %s: ", 1425 (idx == PF_SK_WIRE) ? 1426 "wire" : "stack", 1427 s->kif->pfik_name); 1428 pf_print_state_parts(s, 1429 (idx == PF_SK_WIRE) ? 1430 sk : NULL, 1431 (idx == PF_SK_STACK) ? 1432 sk : NULL); 1433 printf(", existing: "); 1434 pf_print_state_parts(si, 1435 (idx == PF_SK_WIRE) ? 1436 sk : NULL, 1437 (idx == PF_SK_STACK) ? 1438 sk : NULL); 1439 printf("\n"); 1440 } 1441 s->timeout = PFTM_UNLINKED; 1442 PF_HASHROW_UNLOCK(ih); 1443 KEYS_UNLOCK(); 1444 uma_zfree(V_pf_state_key_z, skw); 1445 if (skw != sks) 1446 uma_zfree(V_pf_state_key_z, sks); 1447 if (idx == PF_SK_STACK) 1448 pf_detach_state(s); 1449 return (EEXIST); /* collision! */ 1450 } 1451 } 1452 PF_HASHROW_UNLOCK(ih); 1453 } 1454 uma_zfree(V_pf_state_key_z, sk); 1455 s->key[idx] = cur; 1456 } else { 1457 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1458 s->key[idx] = sk; 1459 } 1460 1461 stateattach: 1462 /* List is sorted, if-bound states before floating. */ 1463 if (s->kif == V_pfi_all) 1464 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1465 else 1466 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1467 1468 if (olds) { 1469 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1470 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1471 key_list[idx]); 1472 olds = NULL; 1473 } 1474 1475 /* 1476 * Attach done. See how should we (or should not?) 1477 * attach a second key. 1478 */ 1479 if (sks == skw) { 1480 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1481 idx = PF_SK_STACK; 1482 sks = NULL; 1483 goto stateattach; 1484 } else if (sks != NULL) { 1485 /* 1486 * Continue attaching with stack key. 1487 */ 1488 sk = sks; 1489 kh = khs; 1490 idx = PF_SK_STACK; 1491 sks = NULL; 1492 goto keyattach; 1493 } 1494 1495 PF_STATE_LOCK(s); 1496 KEYS_UNLOCK(); 1497 1498 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1499 ("%s failure", __func__)); 1500 1501 return (0); 1502 #undef KEYS_UNLOCK 1503 } 1504 1505 static void 1506 pf_detach_state(struct pf_kstate *s) 1507 { 1508 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1509 struct pf_keyhash *kh; 1510 1511 NET_EPOCH_ASSERT(); 1512 MPASS(s->timeout >= PFTM_MAX); 1513 1514 pf_sctp_multihome_detach_addr(s); 1515 1516 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1517 V_pflow_export_state_ptr(s); 1518 1519 if (sks != NULL) { 1520 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1521 PF_HASHROW_LOCK(kh); 1522 if (s->key[PF_SK_STACK] != NULL) 1523 pf_state_key_detach(s, PF_SK_STACK); 1524 /* 1525 * If both point to same key, then we are done. 1526 */ 1527 if (sks == s->key[PF_SK_WIRE]) { 1528 pf_state_key_detach(s, PF_SK_WIRE); 1529 PF_HASHROW_UNLOCK(kh); 1530 return; 1531 } 1532 PF_HASHROW_UNLOCK(kh); 1533 } 1534 1535 if (s->key[PF_SK_WIRE] != NULL) { 1536 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1537 PF_HASHROW_LOCK(kh); 1538 if (s->key[PF_SK_WIRE] != NULL) 1539 pf_state_key_detach(s, PF_SK_WIRE); 1540 PF_HASHROW_UNLOCK(kh); 1541 } 1542 } 1543 1544 static void 1545 pf_state_key_detach(struct pf_kstate *s, int idx) 1546 { 1547 struct pf_state_key *sk = s->key[idx]; 1548 #ifdef INVARIANTS 1549 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1550 1551 PF_HASHROW_ASSERT(kh); 1552 #endif 1553 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1554 s->key[idx] = NULL; 1555 1556 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1557 LIST_REMOVE(sk, entry); 1558 uma_zfree(V_pf_state_key_z, sk); 1559 } 1560 } 1561 1562 static int 1563 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1564 { 1565 struct pf_state_key *sk = mem; 1566 1567 bzero(sk, sizeof(struct pf_state_key_cmp)); 1568 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1569 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1570 1571 return (0); 1572 } 1573 1574 static int 1575 pf_state_key_addr_setup(struct pf_pdesc *pd, 1576 struct pf_state_key_cmp *key, int multi) 1577 { 1578 struct pf_addr *saddr = pd->src; 1579 struct pf_addr *daddr = pd->dst; 1580 #ifdef INET6 1581 struct nd_neighbor_solicit nd; 1582 struct pf_addr *target; 1583 u_short action, reason; 1584 1585 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1586 goto copy; 1587 1588 switch (pd->hdr.icmp6.icmp6_type) { 1589 case ND_NEIGHBOR_SOLICIT: 1590 if (multi) 1591 return (-1); 1592 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1593 return (-1); 1594 target = (struct pf_addr *)&nd.nd_ns_target; 1595 daddr = target; 1596 break; 1597 case ND_NEIGHBOR_ADVERT: 1598 if (multi) 1599 return (-1); 1600 if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), &action, &reason, pd->af)) 1601 return (-1); 1602 target = (struct pf_addr *)&nd.nd_ns_target; 1603 saddr = target; 1604 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1605 key->addr[pd->didx].addr32[0] = 0; 1606 key->addr[pd->didx].addr32[1] = 0; 1607 key->addr[pd->didx].addr32[2] = 0; 1608 key->addr[pd->didx].addr32[3] = 0; 1609 daddr = NULL; /* overwritten */ 1610 } 1611 break; 1612 default: 1613 if (multi == PF_ICMP_MULTI_LINK) { 1614 key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1615 key->addr[pd->sidx].addr32[1] = 0; 1616 key->addr[pd->sidx].addr32[2] = 0; 1617 key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1618 saddr = NULL; /* overwritten */ 1619 } 1620 } 1621 copy: 1622 #endif 1623 if (saddr) 1624 PF_ACPY(&key->addr[pd->sidx], saddr, pd->af); 1625 if (daddr) 1626 PF_ACPY(&key->addr[pd->didx], daddr, pd->af); 1627 1628 return (0); 1629 } 1630 1631 struct pf_state_key * 1632 pf_state_key_setup(struct pf_pdesc *pd, 1633 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1634 u_int16_t dport) 1635 { 1636 struct pf_state_key *sk; 1637 1638 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1639 if (sk == NULL) 1640 return (NULL); 1641 1642 if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)sk, 1643 0)) { 1644 uma_zfree(V_pf_state_key_z, sk); 1645 return (NULL); 1646 } 1647 1648 sk->port[pd->sidx] = sport; 1649 sk->port[pd->didx] = dport; 1650 sk->proto = pd->proto; 1651 sk->af = pd->af; 1652 1653 return (sk); 1654 } 1655 1656 struct pf_state_key * 1657 pf_state_key_clone(const struct pf_state_key *orig) 1658 { 1659 struct pf_state_key *sk; 1660 1661 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1662 if (sk == NULL) 1663 return (NULL); 1664 1665 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1666 1667 return (sk); 1668 } 1669 1670 int 1671 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1672 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1673 { 1674 struct pf_idhash *ih; 1675 struct pf_kstate *cur; 1676 int error; 1677 1678 NET_EPOCH_ASSERT(); 1679 1680 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1681 ("%s: sks not pristine", __func__)); 1682 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1683 ("%s: skw not pristine", __func__)); 1684 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1685 1686 s->kif = kif; 1687 s->orig_kif = orig_kif; 1688 1689 if (s->id == 0 && s->creatorid == 0) { 1690 s->id = alloc_unr64(&V_pf_stateid); 1691 s->id = htobe64(s->id); 1692 s->creatorid = V_pf_status.hostid; 1693 } 1694 1695 /* Returns with ID locked on success. */ 1696 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1697 return (error); 1698 1699 ih = &V_pf_idhash[PF_IDHASH(s)]; 1700 PF_HASHROW_ASSERT(ih); 1701 LIST_FOREACH(cur, &ih->states, entry) 1702 if (cur->id == s->id && cur->creatorid == s->creatorid) 1703 break; 1704 1705 if (cur != NULL) { 1706 s->timeout = PFTM_UNLINKED; 1707 PF_HASHROW_UNLOCK(ih); 1708 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1709 printf("pf: state ID collision: " 1710 "id: %016llx creatorid: %08x\n", 1711 (unsigned long long)be64toh(s->id), 1712 ntohl(s->creatorid)); 1713 } 1714 pf_detach_state(s); 1715 return (EEXIST); 1716 } 1717 LIST_INSERT_HEAD(&ih->states, s, entry); 1718 /* One for keys, one for ID hash. */ 1719 refcount_init(&s->refs, 2); 1720 1721 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1722 if (V_pfsync_insert_state_ptr != NULL) 1723 V_pfsync_insert_state_ptr(s); 1724 1725 /* Returns locked. */ 1726 return (0); 1727 } 1728 1729 /* 1730 * Find state by ID: returns with locked row on success. 1731 */ 1732 struct pf_kstate * 1733 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1734 { 1735 struct pf_idhash *ih; 1736 struct pf_kstate *s; 1737 1738 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1739 1740 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1741 1742 PF_HASHROW_LOCK(ih); 1743 LIST_FOREACH(s, &ih->states, entry) 1744 if (s->id == id && s->creatorid == creatorid) 1745 break; 1746 1747 if (s == NULL) 1748 PF_HASHROW_UNLOCK(ih); 1749 1750 return (s); 1751 } 1752 1753 /* 1754 * Find state by key. 1755 * Returns with ID hash slot locked on success. 1756 */ 1757 static struct pf_kstate * 1758 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1759 u_int dir) 1760 { 1761 struct pf_keyhash *kh; 1762 struct pf_state_key *sk; 1763 struct pf_kstate *s; 1764 int idx; 1765 1766 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1767 1768 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1769 1770 PF_HASHROW_LOCK(kh); 1771 LIST_FOREACH(sk, &kh->keys, entry) 1772 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1773 break; 1774 if (sk == NULL) { 1775 PF_HASHROW_UNLOCK(kh); 1776 return (NULL); 1777 } 1778 1779 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1780 1781 /* List is sorted, if-bound states before floating ones. */ 1782 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1783 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1784 PF_STATE_LOCK(s); 1785 PF_HASHROW_UNLOCK(kh); 1786 if (__predict_false(s->timeout >= PFTM_MAX)) { 1787 /* 1788 * State is either being processed by 1789 * pf_unlink_state() in an other thread, or 1790 * is scheduled for immediate expiry. 1791 */ 1792 PF_STATE_UNLOCK(s); 1793 return (NULL); 1794 } 1795 return (s); 1796 } 1797 PF_HASHROW_UNLOCK(kh); 1798 1799 return (NULL); 1800 } 1801 1802 /* 1803 * Returns with ID hash slot locked on success. 1804 */ 1805 struct pf_kstate * 1806 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1807 { 1808 struct pf_keyhash *kh; 1809 struct pf_state_key *sk; 1810 struct pf_kstate *s, *ret = NULL; 1811 int idx, inout = 0; 1812 1813 if (more != NULL) 1814 *more = 0; 1815 1816 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1817 1818 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1819 1820 PF_HASHROW_LOCK(kh); 1821 LIST_FOREACH(sk, &kh->keys, entry) 1822 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1823 break; 1824 if (sk == NULL) { 1825 PF_HASHROW_UNLOCK(kh); 1826 return (NULL); 1827 } 1828 switch (dir) { 1829 case PF_IN: 1830 idx = PF_SK_WIRE; 1831 break; 1832 case PF_OUT: 1833 idx = PF_SK_STACK; 1834 break; 1835 case PF_INOUT: 1836 idx = PF_SK_WIRE; 1837 inout = 1; 1838 break; 1839 default: 1840 panic("%s: dir %u", __func__, dir); 1841 } 1842 second_run: 1843 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1844 if (more == NULL) { 1845 PF_STATE_LOCK(s); 1846 PF_HASHROW_UNLOCK(kh); 1847 return (s); 1848 } 1849 1850 if (ret) 1851 (*more)++; 1852 else { 1853 ret = s; 1854 PF_STATE_LOCK(s); 1855 } 1856 } 1857 if (inout == 1) { 1858 inout = 0; 1859 idx = PF_SK_STACK; 1860 goto second_run; 1861 } 1862 PF_HASHROW_UNLOCK(kh); 1863 1864 return (ret); 1865 } 1866 1867 /* 1868 * FIXME 1869 * This routine is inefficient -- locks the state only to unlock immediately on 1870 * return. 1871 * It is racy -- after the state is unlocked nothing stops other threads from 1872 * removing it. 1873 */ 1874 bool 1875 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1876 { 1877 struct pf_kstate *s; 1878 1879 s = pf_find_state_all(key, dir, NULL); 1880 if (s != NULL) { 1881 PF_STATE_UNLOCK(s); 1882 return (true); 1883 } 1884 return (false); 1885 } 1886 1887 struct pf_udp_mapping * 1888 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1889 struct pf_addr *nat_addr, uint16_t nat_port) 1890 { 1891 struct pf_udp_mapping *mapping; 1892 1893 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1894 if (mapping == NULL) 1895 return (NULL); 1896 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1897 mapping->endpoints[0].port = src_port; 1898 mapping->endpoints[0].af = af; 1899 mapping->endpoints[0].mapping = mapping; 1900 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1901 mapping->endpoints[1].port = nat_port; 1902 mapping->endpoints[1].af = af; 1903 mapping->endpoints[1].mapping = mapping; 1904 refcount_init(&mapping->refs, 1); 1905 return (mapping); 1906 } 1907 1908 int 1909 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1910 { 1911 struct pf_udpendpointhash *h0, *h1; 1912 struct pf_udp_endpoint *endpoint; 1913 int ret = EEXIST; 1914 1915 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1916 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1917 if (h0 == h1) { 1918 PF_HASHROW_LOCK(h0); 1919 } else if (h0 < h1) { 1920 PF_HASHROW_LOCK(h0); 1921 PF_HASHROW_LOCK(h1); 1922 } else { 1923 PF_HASHROW_LOCK(h1); 1924 PF_HASHROW_LOCK(h0); 1925 } 1926 1927 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1928 if (bcmp(endpoint, &mapping->endpoints[0], 1929 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1930 break; 1931 } 1932 if (endpoint != NULL) 1933 goto cleanup; 1934 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1935 if (bcmp(endpoint, &mapping->endpoints[1], 1936 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1937 break; 1938 } 1939 if (endpoint != NULL) 1940 goto cleanup; 1941 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1942 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1943 ret = 0; 1944 1945 cleanup: 1946 if (h0 != h1) { 1947 PF_HASHROW_UNLOCK(h0); 1948 PF_HASHROW_UNLOCK(h1); 1949 } else { 1950 PF_HASHROW_UNLOCK(h0); 1951 } 1952 return (ret); 1953 } 1954 1955 void 1956 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1957 { 1958 /* refcount is synchronized on the source endpoint's row lock */ 1959 struct pf_udpendpointhash *h0, *h1; 1960 1961 if (mapping == NULL) 1962 return; 1963 1964 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1965 PF_HASHROW_LOCK(h0); 1966 if (refcount_release(&mapping->refs)) { 1967 LIST_REMOVE(&mapping->endpoints[0], entry); 1968 PF_HASHROW_UNLOCK(h0); 1969 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1970 PF_HASHROW_LOCK(h1); 1971 LIST_REMOVE(&mapping->endpoints[1], entry); 1972 PF_HASHROW_UNLOCK(h1); 1973 1974 uma_zfree(V_pf_udp_mapping_z, mapping); 1975 } else { 1976 PF_HASHROW_UNLOCK(h0); 1977 } 1978 } 1979 1980 1981 struct pf_udp_mapping * 1982 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1983 { 1984 struct pf_udpendpointhash *uh; 1985 struct pf_udp_endpoint *endpoint; 1986 1987 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1988 1989 PF_HASHROW_LOCK(uh); 1990 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1991 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1992 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1993 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1994 break; 1995 } 1996 if (endpoint == NULL) { 1997 PF_HASHROW_UNLOCK(uh); 1998 return (NULL); 1999 } 2000 refcount_acquire(&endpoint->mapping->refs); 2001 PF_HASHROW_UNLOCK(uh); 2002 return (endpoint->mapping); 2003 } 2004 /* END state table stuff */ 2005 2006 static void 2007 pf_send(struct pf_send_entry *pfse) 2008 { 2009 2010 PF_SENDQ_LOCK(); 2011 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2012 PF_SENDQ_UNLOCK(); 2013 swi_sched(V_pf_swi_cookie, 0); 2014 } 2015 2016 static bool 2017 pf_isforlocal(struct mbuf *m, int af) 2018 { 2019 switch (af) { 2020 #ifdef INET 2021 case AF_INET: { 2022 struct ip *ip = mtod(m, struct ip *); 2023 2024 return (in_localip(ip->ip_dst)); 2025 } 2026 #endif 2027 #ifdef INET6 2028 case AF_INET6: { 2029 struct ip6_hdr *ip6; 2030 struct in6_ifaddr *ia; 2031 ip6 = mtod(m, struct ip6_hdr *); 2032 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2033 if (ia == NULL) 2034 return (false); 2035 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2036 } 2037 #endif 2038 } 2039 2040 return (false); 2041 } 2042 2043 int 2044 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2045 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2046 { 2047 /* 2048 * ICMP types marked with PF_OUT are typically responses to 2049 * PF_IN, and will match states in the opposite direction. 2050 * PF_IN ICMP types need to match a state with that type. 2051 */ 2052 *icmp_dir = PF_OUT; 2053 *multi = PF_ICMP_MULTI_LINK; 2054 /* Queries (and responses) */ 2055 switch (pd->af) { 2056 #ifdef INET 2057 case AF_INET: 2058 switch (type) { 2059 case ICMP_ECHO: 2060 *icmp_dir = PF_IN; 2061 case ICMP_ECHOREPLY: 2062 *virtual_type = ICMP_ECHO; 2063 *virtual_id = pd->hdr.icmp.icmp_id; 2064 break; 2065 2066 case ICMP_TSTAMP: 2067 *icmp_dir = PF_IN; 2068 case ICMP_TSTAMPREPLY: 2069 *virtual_type = ICMP_TSTAMP; 2070 *virtual_id = pd->hdr.icmp.icmp_id; 2071 break; 2072 2073 case ICMP_IREQ: 2074 *icmp_dir = PF_IN; 2075 case ICMP_IREQREPLY: 2076 *virtual_type = ICMP_IREQ; 2077 *virtual_id = pd->hdr.icmp.icmp_id; 2078 break; 2079 2080 case ICMP_MASKREQ: 2081 *icmp_dir = PF_IN; 2082 case ICMP_MASKREPLY: 2083 *virtual_type = ICMP_MASKREQ; 2084 *virtual_id = pd->hdr.icmp.icmp_id; 2085 break; 2086 2087 case ICMP_IPV6_WHEREAREYOU: 2088 *icmp_dir = PF_IN; 2089 case ICMP_IPV6_IAMHERE: 2090 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2091 *virtual_id = 0; /* Nothing sane to match on! */ 2092 break; 2093 2094 case ICMP_MOBILE_REGREQUEST: 2095 *icmp_dir = PF_IN; 2096 case ICMP_MOBILE_REGREPLY: 2097 *virtual_type = ICMP_MOBILE_REGREQUEST; 2098 *virtual_id = 0; /* Nothing sane to match on! */ 2099 break; 2100 2101 case ICMP_ROUTERSOLICIT: 2102 *icmp_dir = PF_IN; 2103 case ICMP_ROUTERADVERT: 2104 *virtual_type = ICMP_ROUTERSOLICIT; 2105 *virtual_id = 0; /* Nothing sane to match on! */ 2106 break; 2107 2108 /* These ICMP types map to other connections */ 2109 case ICMP_UNREACH: 2110 case ICMP_SOURCEQUENCH: 2111 case ICMP_REDIRECT: 2112 case ICMP_TIMXCEED: 2113 case ICMP_PARAMPROB: 2114 /* These will not be used, but set them anyway */ 2115 *icmp_dir = PF_IN; 2116 *virtual_type = type; 2117 *virtual_id = 0; 2118 HTONS(*virtual_type); 2119 return (1); /* These types match to another state */ 2120 2121 /* 2122 * All remaining ICMP types get their own states, 2123 * and will only match in one direction. 2124 */ 2125 default: 2126 *icmp_dir = PF_IN; 2127 *virtual_type = type; 2128 *virtual_id = 0; 2129 break; 2130 } 2131 break; 2132 #endif /* INET */ 2133 #ifdef INET6 2134 case AF_INET6: 2135 switch (type) { 2136 case ICMP6_ECHO_REQUEST: 2137 *icmp_dir = PF_IN; 2138 case ICMP6_ECHO_REPLY: 2139 *virtual_type = ICMP6_ECHO_REQUEST; 2140 *virtual_id = pd->hdr.icmp6.icmp6_id; 2141 break; 2142 2143 case MLD_LISTENER_QUERY: 2144 case MLD_LISTENER_REPORT: { 2145 /* 2146 * Listener Report can be sent by clients 2147 * without an associated Listener Query. 2148 * In addition to that, when Report is sent as a 2149 * reply to a Query its source and destination 2150 * address are different. 2151 */ 2152 *icmp_dir = PF_IN; 2153 *virtual_type = MLD_LISTENER_QUERY; 2154 *virtual_id = 0; 2155 break; 2156 } 2157 case MLD_MTRACE: 2158 *icmp_dir = PF_IN; 2159 case MLD_MTRACE_RESP: 2160 *virtual_type = MLD_MTRACE; 2161 *virtual_id = 0; /* Nothing sane to match on! */ 2162 break; 2163 2164 case ND_NEIGHBOR_SOLICIT: 2165 *icmp_dir = PF_IN; 2166 case ND_NEIGHBOR_ADVERT: { 2167 *virtual_type = ND_NEIGHBOR_SOLICIT; 2168 *virtual_id = 0; 2169 break; 2170 } 2171 2172 /* 2173 * These ICMP types map to other connections. 2174 * ND_REDIRECT can't be in this list because the triggering 2175 * packet header is optional. 2176 */ 2177 case ICMP6_DST_UNREACH: 2178 case ICMP6_PACKET_TOO_BIG: 2179 case ICMP6_TIME_EXCEEDED: 2180 case ICMP6_PARAM_PROB: 2181 /* These will not be used, but set them anyway */ 2182 *icmp_dir = PF_IN; 2183 *virtual_type = type; 2184 *virtual_id = 0; 2185 HTONS(*virtual_type); 2186 return (1); /* These types match to another state */ 2187 /* 2188 * All remaining ICMP6 types get their own states, 2189 * and will only match in one direction. 2190 */ 2191 default: 2192 *icmp_dir = PF_IN; 2193 *virtual_type = type; 2194 *virtual_id = 0; 2195 break; 2196 } 2197 break; 2198 #endif /* INET6 */ 2199 } 2200 HTONS(*virtual_type); 2201 return (0); /* These types match to their own state */ 2202 } 2203 2204 void 2205 pf_intr(void *v) 2206 { 2207 struct epoch_tracker et; 2208 struct pf_send_head queue; 2209 struct pf_send_entry *pfse, *next; 2210 2211 CURVNET_SET((struct vnet *)v); 2212 2213 PF_SENDQ_LOCK(); 2214 queue = V_pf_sendqueue; 2215 STAILQ_INIT(&V_pf_sendqueue); 2216 PF_SENDQ_UNLOCK(); 2217 2218 NET_EPOCH_ENTER(et); 2219 2220 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2221 switch (pfse->pfse_type) { 2222 #ifdef INET 2223 case PFSE_IP: { 2224 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2225 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2226 ("%s: rcvif != loif", __func__)); 2227 2228 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2229 pfse->pfse_m->m_pkthdr.csum_flags |= 2230 CSUM_IP_VALID | CSUM_IP_CHECKED; 2231 ip_input(pfse->pfse_m); 2232 } else { 2233 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2234 NULL); 2235 } 2236 break; 2237 } 2238 case PFSE_ICMP: 2239 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2240 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2241 break; 2242 #endif /* INET */ 2243 #ifdef INET6 2244 case PFSE_IP6: 2245 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2246 KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif, 2247 ("%s: rcvif != loif", __func__)); 2248 2249 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL | 2250 M_LOOP; 2251 ip6_input(pfse->pfse_m); 2252 } else { 2253 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2254 NULL, NULL); 2255 } 2256 break; 2257 case PFSE_ICMP6: 2258 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2259 pfse->icmpopts.code, pfse->icmpopts.mtu); 2260 break; 2261 #endif /* INET6 */ 2262 default: 2263 panic("%s: unknown type", __func__); 2264 } 2265 free(pfse, M_PFTEMP); 2266 } 2267 NET_EPOCH_EXIT(et); 2268 CURVNET_RESTORE(); 2269 } 2270 2271 #define pf_purge_thread_period (hz / 10) 2272 2273 #ifdef PF_WANT_32_TO_64_COUNTER 2274 static void 2275 pf_status_counter_u64_periodic(void) 2276 { 2277 2278 PF_RULES_RASSERT(); 2279 2280 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2281 return; 2282 } 2283 2284 for (int i = 0; i < FCNT_MAX; i++) { 2285 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2286 } 2287 } 2288 2289 static void 2290 pf_kif_counter_u64_periodic(void) 2291 { 2292 struct pfi_kkif *kif; 2293 size_t r, run; 2294 2295 PF_RULES_RASSERT(); 2296 2297 if (__predict_false(V_pf_allkifcount == 0)) { 2298 return; 2299 } 2300 2301 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2302 return; 2303 } 2304 2305 run = V_pf_allkifcount / 10; 2306 if (run < 5) 2307 run = 5; 2308 2309 for (r = 0; r < run; r++) { 2310 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2311 if (kif == NULL) { 2312 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2313 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2314 break; 2315 } 2316 2317 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2318 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2319 2320 for (int i = 0; i < 2; i++) { 2321 for (int j = 0; j < 2; j++) { 2322 for (int k = 0; k < 2; k++) { 2323 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2324 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2325 } 2326 } 2327 } 2328 } 2329 } 2330 2331 static void 2332 pf_rule_counter_u64_periodic(void) 2333 { 2334 struct pf_krule *rule; 2335 size_t r, run; 2336 2337 PF_RULES_RASSERT(); 2338 2339 if (__predict_false(V_pf_allrulecount == 0)) { 2340 return; 2341 } 2342 2343 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2344 return; 2345 } 2346 2347 run = V_pf_allrulecount / 10; 2348 if (run < 5) 2349 run = 5; 2350 2351 for (r = 0; r < run; r++) { 2352 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2353 if (rule == NULL) { 2354 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2355 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2356 break; 2357 } 2358 2359 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2360 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2361 2362 pf_counter_u64_periodic(&rule->evaluations); 2363 for (int i = 0; i < 2; i++) { 2364 pf_counter_u64_periodic(&rule->packets[i]); 2365 pf_counter_u64_periodic(&rule->bytes[i]); 2366 } 2367 } 2368 } 2369 2370 static void 2371 pf_counter_u64_periodic_main(void) 2372 { 2373 PF_RULES_RLOCK_TRACKER; 2374 2375 V_pf_counter_periodic_iter++; 2376 2377 PF_RULES_RLOCK(); 2378 pf_counter_u64_critical_enter(); 2379 pf_status_counter_u64_periodic(); 2380 pf_kif_counter_u64_periodic(); 2381 pf_rule_counter_u64_periodic(); 2382 pf_counter_u64_critical_exit(); 2383 PF_RULES_RUNLOCK(); 2384 } 2385 #else 2386 #define pf_counter_u64_periodic_main() do { } while (0) 2387 #endif 2388 2389 void 2390 pf_purge_thread(void *unused __unused) 2391 { 2392 struct epoch_tracker et; 2393 2394 VNET_ITERATOR_DECL(vnet_iter); 2395 2396 sx_xlock(&pf_end_lock); 2397 while (pf_end_threads == 0) { 2398 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2399 2400 VNET_LIST_RLOCK(); 2401 NET_EPOCH_ENTER(et); 2402 VNET_FOREACH(vnet_iter) { 2403 CURVNET_SET(vnet_iter); 2404 2405 /* Wait until V_pf_default_rule is initialized. */ 2406 if (V_pf_vnet_active == 0) { 2407 CURVNET_RESTORE(); 2408 continue; 2409 } 2410 2411 pf_counter_u64_periodic_main(); 2412 2413 /* 2414 * Process 1/interval fraction of the state 2415 * table every run. 2416 */ 2417 V_pf_purge_idx = 2418 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2419 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2420 2421 /* 2422 * Purge other expired types every 2423 * PFTM_INTERVAL seconds. 2424 */ 2425 if (V_pf_purge_idx == 0) { 2426 /* 2427 * Order is important: 2428 * - states and src nodes reference rules 2429 * - states and rules reference kifs 2430 */ 2431 pf_purge_expired_fragments(); 2432 pf_purge_expired_src_nodes(); 2433 pf_purge_unlinked_rules(); 2434 pfi_kkif_purge(); 2435 } 2436 CURVNET_RESTORE(); 2437 } 2438 NET_EPOCH_EXIT(et); 2439 VNET_LIST_RUNLOCK(); 2440 } 2441 2442 pf_end_threads++; 2443 sx_xunlock(&pf_end_lock); 2444 kproc_exit(0); 2445 } 2446 2447 void 2448 pf_unload_vnet_purge(void) 2449 { 2450 2451 /* 2452 * To cleanse up all kifs and rules we need 2453 * two runs: first one clears reference flags, 2454 * then pf_purge_expired_states() doesn't 2455 * raise them, and then second run frees. 2456 */ 2457 pf_purge_unlinked_rules(); 2458 pfi_kkif_purge(); 2459 2460 /* 2461 * Now purge everything. 2462 */ 2463 pf_purge_expired_states(0, V_pf_hashmask); 2464 pf_purge_fragments(UINT_MAX); 2465 pf_purge_expired_src_nodes(); 2466 2467 /* 2468 * Now all kifs & rules should be unreferenced, 2469 * thus should be successfully freed. 2470 */ 2471 pf_purge_unlinked_rules(); 2472 pfi_kkif_purge(); 2473 } 2474 2475 u_int32_t 2476 pf_state_expires(const struct pf_kstate *state) 2477 { 2478 u_int32_t timeout; 2479 u_int32_t start; 2480 u_int32_t end; 2481 u_int32_t states; 2482 2483 /* handle all PFTM_* > PFTM_MAX here */ 2484 if (state->timeout == PFTM_PURGE) 2485 return (time_uptime); 2486 KASSERT(state->timeout != PFTM_UNLINKED, 2487 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2488 KASSERT((state->timeout < PFTM_MAX), 2489 ("pf_state_expires: timeout > PFTM_MAX")); 2490 timeout = state->rule->timeout[state->timeout]; 2491 if (!timeout) 2492 timeout = V_pf_default_rule.timeout[state->timeout]; 2493 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2494 if (start && state->rule != &V_pf_default_rule) { 2495 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2496 states = counter_u64_fetch(state->rule->states_cur); 2497 } else { 2498 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2499 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2500 states = V_pf_status.states; 2501 } 2502 if (end && states > start && start < end) { 2503 if (states < end) { 2504 timeout = (u_int64_t)timeout * (end - states) / 2505 (end - start); 2506 return ((state->expire / 1000) + timeout); 2507 } 2508 else 2509 return (time_uptime); 2510 } 2511 return ((state->expire / 1000) + timeout); 2512 } 2513 2514 void 2515 pf_purge_expired_src_nodes(void) 2516 { 2517 struct pf_ksrc_node_list freelist; 2518 struct pf_srchash *sh; 2519 struct pf_ksrc_node *cur, *next; 2520 int i; 2521 2522 LIST_INIT(&freelist); 2523 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2524 PF_HASHROW_LOCK(sh); 2525 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2526 if (cur->states == 0 && cur->expire <= time_uptime) { 2527 pf_unlink_src_node(cur); 2528 LIST_INSERT_HEAD(&freelist, cur, entry); 2529 } else if (cur->rule != NULL) 2530 cur->rule->rule_ref |= PFRULE_REFS; 2531 PF_HASHROW_UNLOCK(sh); 2532 } 2533 2534 pf_free_src_nodes(&freelist); 2535 2536 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2537 } 2538 2539 static void 2540 pf_src_tree_remove_state(struct pf_kstate *s) 2541 { 2542 struct pf_ksrc_node *sn; 2543 uint32_t timeout; 2544 2545 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2546 s->rule->timeout[PFTM_SRC_NODE] : 2547 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2548 2549 if (s->src_node != NULL) { 2550 sn = s->src_node; 2551 PF_SRC_NODE_LOCK(sn); 2552 if (s->src.tcp_est) 2553 --sn->conn; 2554 if (--sn->states == 0) 2555 sn->expire = time_uptime + timeout; 2556 PF_SRC_NODE_UNLOCK(sn); 2557 } 2558 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2559 sn = s->nat_src_node; 2560 PF_SRC_NODE_LOCK(sn); 2561 if (--sn->states == 0) 2562 sn->expire = time_uptime + timeout; 2563 PF_SRC_NODE_UNLOCK(sn); 2564 } 2565 s->src_node = s->nat_src_node = NULL; 2566 } 2567 2568 /* 2569 * Unlink and potentilly free a state. Function may be 2570 * called with ID hash row locked, but always returns 2571 * unlocked, since it needs to go through key hash locking. 2572 */ 2573 int 2574 pf_unlink_state(struct pf_kstate *s) 2575 { 2576 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2577 2578 NET_EPOCH_ASSERT(); 2579 PF_HASHROW_ASSERT(ih); 2580 2581 if (s->timeout == PFTM_UNLINKED) { 2582 /* 2583 * State is being processed 2584 * by pf_unlink_state() in 2585 * an other thread. 2586 */ 2587 PF_HASHROW_UNLOCK(ih); 2588 return (0); /* XXXGL: undefined actually */ 2589 } 2590 2591 if (s->src.state == PF_TCPS_PROXY_DST) { 2592 /* XXX wire key the right one? */ 2593 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2594 &s->key[PF_SK_WIRE]->addr[1], 2595 &s->key[PF_SK_WIRE]->addr[0], 2596 s->key[PF_SK_WIRE]->port[1], 2597 s->key[PF_SK_WIRE]->port[0], 2598 s->src.seqhi, s->src.seqlo + 1, 2599 TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0, 2600 s->act.rtableid); 2601 } 2602 2603 LIST_REMOVE(s, entry); 2604 pf_src_tree_remove_state(s); 2605 2606 if (V_pfsync_delete_state_ptr != NULL) 2607 V_pfsync_delete_state_ptr(s); 2608 2609 STATE_DEC_COUNTERS(s); 2610 2611 s->timeout = PFTM_UNLINKED; 2612 2613 /* Ensure we remove it from the list of halfopen states, if needed. */ 2614 if (s->key[PF_SK_STACK] != NULL && 2615 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2616 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2617 2618 PF_HASHROW_UNLOCK(ih); 2619 2620 pf_detach_state(s); 2621 2622 pf_udp_mapping_release(s->udp_mapping); 2623 2624 /* pf_state_insert() initialises refs to 2 */ 2625 return (pf_release_staten(s, 2)); 2626 } 2627 2628 struct pf_kstate * 2629 pf_alloc_state(int flags) 2630 { 2631 2632 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2633 } 2634 2635 void 2636 pf_free_state(struct pf_kstate *cur) 2637 { 2638 struct pf_krule_item *ri; 2639 2640 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2641 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2642 cur->timeout)); 2643 2644 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2645 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2646 free(ri, M_PF_RULE_ITEM); 2647 } 2648 2649 pf_normalize_tcp_cleanup(cur); 2650 uma_zfree(V_pf_state_z, cur); 2651 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2652 } 2653 2654 /* 2655 * Called only from pf_purge_thread(), thus serialized. 2656 */ 2657 static u_int 2658 pf_purge_expired_states(u_int i, int maxcheck) 2659 { 2660 struct pf_idhash *ih; 2661 struct pf_kstate *s; 2662 struct pf_krule_item *mrm; 2663 size_t count __unused; 2664 2665 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2666 2667 /* 2668 * Go through hash and unlink states that expire now. 2669 */ 2670 while (maxcheck > 0) { 2671 count = 0; 2672 ih = &V_pf_idhash[i]; 2673 2674 /* only take the lock if we expect to do work */ 2675 if (!LIST_EMPTY(&ih->states)) { 2676 relock: 2677 PF_HASHROW_LOCK(ih); 2678 LIST_FOREACH(s, &ih->states, entry) { 2679 if (pf_state_expires(s) <= time_uptime) { 2680 V_pf_status.states -= 2681 pf_unlink_state(s); 2682 goto relock; 2683 } 2684 s->rule->rule_ref |= PFRULE_REFS; 2685 if (s->nat_rule != NULL) 2686 s->nat_rule->rule_ref |= PFRULE_REFS; 2687 if (s->anchor != NULL) 2688 s->anchor->rule_ref |= PFRULE_REFS; 2689 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2690 SLIST_FOREACH(mrm, &s->match_rules, entry) 2691 mrm->r->rule_ref |= PFRULE_REFS; 2692 if (s->rt_kif) 2693 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2694 count++; 2695 } 2696 PF_HASHROW_UNLOCK(ih); 2697 } 2698 2699 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2700 2701 /* Return when we hit end of hash. */ 2702 if (++i > V_pf_hashmask) { 2703 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2704 return (0); 2705 } 2706 2707 maxcheck--; 2708 } 2709 2710 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2711 2712 return (i); 2713 } 2714 2715 static void 2716 pf_purge_unlinked_rules(void) 2717 { 2718 struct pf_krulequeue tmpq; 2719 struct pf_krule *r, *r1; 2720 2721 /* 2722 * If we have overloading task pending, then we'd 2723 * better skip purging this time. There is a tiny 2724 * probability that overloading task references 2725 * an already unlinked rule. 2726 */ 2727 PF_OVERLOADQ_LOCK(); 2728 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2729 PF_OVERLOADQ_UNLOCK(); 2730 return; 2731 } 2732 PF_OVERLOADQ_UNLOCK(); 2733 2734 /* 2735 * Do naive mark-and-sweep garbage collecting of old rules. 2736 * Reference flag is raised by pf_purge_expired_states() 2737 * and pf_purge_expired_src_nodes(). 2738 * 2739 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2740 * use a temporary queue. 2741 */ 2742 TAILQ_INIT(&tmpq); 2743 PF_UNLNKDRULES_LOCK(); 2744 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2745 if (!(r->rule_ref & PFRULE_REFS)) { 2746 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2747 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2748 } else 2749 r->rule_ref &= ~PFRULE_REFS; 2750 } 2751 PF_UNLNKDRULES_UNLOCK(); 2752 2753 if (!TAILQ_EMPTY(&tmpq)) { 2754 PF_CONFIG_LOCK(); 2755 PF_RULES_WLOCK(); 2756 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2757 TAILQ_REMOVE(&tmpq, r, entries); 2758 pf_free_rule(r); 2759 } 2760 PF_RULES_WUNLOCK(); 2761 PF_CONFIG_UNLOCK(); 2762 } 2763 } 2764 2765 void 2766 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2767 { 2768 switch (af) { 2769 #ifdef INET 2770 case AF_INET: { 2771 u_int32_t a = ntohl(addr->addr32[0]); 2772 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2773 (a>>8)&255, a&255); 2774 if (p) { 2775 p = ntohs(p); 2776 printf(":%u", p); 2777 } 2778 break; 2779 } 2780 #endif /* INET */ 2781 #ifdef INET6 2782 case AF_INET6: { 2783 u_int16_t b; 2784 u_int8_t i, curstart, curend, maxstart, maxend; 2785 curstart = curend = maxstart = maxend = 255; 2786 for (i = 0; i < 8; i++) { 2787 if (!addr->addr16[i]) { 2788 if (curstart == 255) 2789 curstart = i; 2790 curend = i; 2791 } else { 2792 if ((curend - curstart) > 2793 (maxend - maxstart)) { 2794 maxstart = curstart; 2795 maxend = curend; 2796 } 2797 curstart = curend = 255; 2798 } 2799 } 2800 if ((curend - curstart) > 2801 (maxend - maxstart)) { 2802 maxstart = curstart; 2803 maxend = curend; 2804 } 2805 for (i = 0; i < 8; i++) { 2806 if (i >= maxstart && i <= maxend) { 2807 if (i == 0) 2808 printf(":"); 2809 if (i == maxend) 2810 printf(":"); 2811 } else { 2812 b = ntohs(addr->addr16[i]); 2813 printf("%x", b); 2814 if (i < 7) 2815 printf(":"); 2816 } 2817 } 2818 if (p) { 2819 p = ntohs(p); 2820 printf("[%u]", p); 2821 } 2822 break; 2823 } 2824 #endif /* INET6 */ 2825 } 2826 } 2827 2828 void 2829 pf_print_state(struct pf_kstate *s) 2830 { 2831 pf_print_state_parts(s, NULL, NULL); 2832 } 2833 2834 static void 2835 pf_print_state_parts(struct pf_kstate *s, 2836 struct pf_state_key *skwp, struct pf_state_key *sksp) 2837 { 2838 struct pf_state_key *skw, *sks; 2839 u_int8_t proto, dir; 2840 2841 /* Do our best to fill these, but they're skipped if NULL */ 2842 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2843 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2844 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2845 dir = s ? s->direction : 0; 2846 2847 switch (proto) { 2848 case IPPROTO_IPV4: 2849 printf("IPv4"); 2850 break; 2851 case IPPROTO_IPV6: 2852 printf("IPv6"); 2853 break; 2854 case IPPROTO_TCP: 2855 printf("TCP"); 2856 break; 2857 case IPPROTO_UDP: 2858 printf("UDP"); 2859 break; 2860 case IPPROTO_ICMP: 2861 printf("ICMP"); 2862 break; 2863 case IPPROTO_ICMPV6: 2864 printf("ICMPv6"); 2865 break; 2866 default: 2867 printf("%u", proto); 2868 break; 2869 } 2870 switch (dir) { 2871 case PF_IN: 2872 printf(" in"); 2873 break; 2874 case PF_OUT: 2875 printf(" out"); 2876 break; 2877 } 2878 if (skw) { 2879 printf(" wire: "); 2880 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2881 printf(" "); 2882 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2883 } 2884 if (sks) { 2885 printf(" stack: "); 2886 if (sks != skw) { 2887 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2888 printf(" "); 2889 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2890 } else 2891 printf("-"); 2892 } 2893 if (s) { 2894 if (proto == IPPROTO_TCP) { 2895 printf(" [lo=%u high=%u win=%u modulator=%u", 2896 s->src.seqlo, s->src.seqhi, 2897 s->src.max_win, s->src.seqdiff); 2898 if (s->src.wscale && s->dst.wscale) 2899 printf(" wscale=%u", 2900 s->src.wscale & PF_WSCALE_MASK); 2901 printf("]"); 2902 printf(" [lo=%u high=%u win=%u modulator=%u", 2903 s->dst.seqlo, s->dst.seqhi, 2904 s->dst.max_win, s->dst.seqdiff); 2905 if (s->src.wscale && s->dst.wscale) 2906 printf(" wscale=%u", 2907 s->dst.wscale & PF_WSCALE_MASK); 2908 printf("]"); 2909 } 2910 printf(" %u:%u", s->src.state, s->dst.state); 2911 if (s->rule) 2912 printf(" @%d", s->rule->nr); 2913 } 2914 } 2915 2916 void 2917 pf_print_flags(u_int8_t f) 2918 { 2919 if (f) 2920 printf(" "); 2921 if (f & TH_FIN) 2922 printf("F"); 2923 if (f & TH_SYN) 2924 printf("S"); 2925 if (f & TH_RST) 2926 printf("R"); 2927 if (f & TH_PUSH) 2928 printf("P"); 2929 if (f & TH_ACK) 2930 printf("A"); 2931 if (f & TH_URG) 2932 printf("U"); 2933 if (f & TH_ECE) 2934 printf("E"); 2935 if (f & TH_CWR) 2936 printf("W"); 2937 } 2938 2939 #define PF_SET_SKIP_STEPS(i) \ 2940 do { \ 2941 while (head[i] != cur) { \ 2942 head[i]->skip[i] = cur; \ 2943 head[i] = TAILQ_NEXT(head[i], entries); \ 2944 } \ 2945 } while (0) 2946 2947 void 2948 pf_calc_skip_steps(struct pf_krulequeue *rules) 2949 { 2950 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2951 int i; 2952 2953 cur = TAILQ_FIRST(rules); 2954 prev = cur; 2955 for (i = 0; i < PF_SKIP_COUNT; ++i) 2956 head[i] = cur; 2957 while (cur != NULL) { 2958 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2959 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2960 if (cur->direction != prev->direction) 2961 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2962 if (cur->af != prev->af) 2963 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2964 if (cur->proto != prev->proto) 2965 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2966 if (cur->src.neg != prev->src.neg || 2967 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2968 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2969 if (cur->dst.neg != prev->dst.neg || 2970 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2971 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2972 if (cur->src.port[0] != prev->src.port[0] || 2973 cur->src.port[1] != prev->src.port[1] || 2974 cur->src.port_op != prev->src.port_op) 2975 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2976 if (cur->dst.port[0] != prev->dst.port[0] || 2977 cur->dst.port[1] != prev->dst.port[1] || 2978 cur->dst.port_op != prev->dst.port_op) 2979 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2980 2981 prev = cur; 2982 cur = TAILQ_NEXT(cur, entries); 2983 } 2984 for (i = 0; i < PF_SKIP_COUNT; ++i) 2985 PF_SET_SKIP_STEPS(i); 2986 } 2987 2988 int 2989 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2990 { 2991 if (aw1->type != aw2->type) 2992 return (1); 2993 switch (aw1->type) { 2994 case PF_ADDR_ADDRMASK: 2995 case PF_ADDR_RANGE: 2996 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2997 return (1); 2998 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2999 return (1); 3000 return (0); 3001 case PF_ADDR_DYNIFTL: 3002 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 3003 case PF_ADDR_NOROUTE: 3004 case PF_ADDR_URPFFAILED: 3005 return (0); 3006 case PF_ADDR_TABLE: 3007 return (aw1->p.tbl != aw2->p.tbl); 3008 default: 3009 printf("invalid address type: %d\n", aw1->type); 3010 return (1); 3011 } 3012 } 3013 3014 /** 3015 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3016 * header isn't always a full checksum. In some cases (i.e. output) it's a 3017 * pseudo-header checksum, which is a partial checksum over src/dst IP 3018 * addresses, protocol number and length. 3019 * 3020 * That means we have the following cases: 3021 * * Input or forwarding: we don't have TSO, the checksum fields are full 3022 * checksums, we need to update the checksum whenever we change anything. 3023 * * Output (i.e. the checksum is a pseudo-header checksum): 3024 * x The field being updated is src/dst address or affects the length of 3025 * the packet. We need to update the pseudo-header checksum (note that this 3026 * checksum is not ones' complement). 3027 * x Some other field is being modified (e.g. src/dst port numbers): We 3028 * don't have to update anything. 3029 **/ 3030 u_int16_t 3031 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3032 { 3033 u_int32_t x; 3034 3035 x = cksum + old - new; 3036 x = (x + (x >> 16)) & 0xffff; 3037 3038 /* optimise: eliminate a branch when not udp */ 3039 if (udp && cksum == 0x0000) 3040 return cksum; 3041 if (udp && x == 0x0000) 3042 x = 0xffff; 3043 3044 return (u_int16_t)(x); 3045 } 3046 3047 static void 3048 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3049 u_int8_t udp) 3050 { 3051 u_int16_t old = htons(hi ? (*f << 8) : *f); 3052 u_int16_t new = htons(hi ? ( v << 8) : v); 3053 3054 if (*f == v) 3055 return; 3056 3057 *f = v; 3058 3059 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3060 return; 3061 3062 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3063 } 3064 3065 void 3066 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3067 bool hi, u_int8_t udp) 3068 { 3069 u_int8_t *fb = (u_int8_t *)f; 3070 u_int8_t *vb = (u_int8_t *)&v; 3071 3072 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3073 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3074 } 3075 3076 void 3077 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3078 bool hi, u_int8_t udp) 3079 { 3080 u_int8_t *fb = (u_int8_t *)f; 3081 u_int8_t *vb = (u_int8_t *)&v; 3082 3083 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3084 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3085 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3086 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3087 } 3088 3089 u_int16_t 3090 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3091 u_int16_t new, u_int8_t udp) 3092 { 3093 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3094 return (cksum); 3095 3096 return (pf_cksum_fixup(cksum, old, new, udp)); 3097 } 3098 3099 static void 3100 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3101 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3102 sa_family_t af) 3103 { 3104 struct pf_addr ao; 3105 u_int16_t po = *p; 3106 3107 PF_ACPY(&ao, a, af); 3108 PF_ACPY(a, an, af); 3109 3110 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3111 *pc = ~*pc; 3112 3113 *p = pn; 3114 3115 switch (af) { 3116 #ifdef INET 3117 case AF_INET: 3118 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3119 ao.addr16[0], an->addr16[0], 0), 3120 ao.addr16[1], an->addr16[1], 0); 3121 *p = pn; 3122 3123 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3124 ao.addr16[0], an->addr16[0], u), 3125 ao.addr16[1], an->addr16[1], u); 3126 3127 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3128 break; 3129 #endif /* INET */ 3130 #ifdef INET6 3131 case AF_INET6: 3132 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3133 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3134 pf_cksum_fixup(pf_cksum_fixup(*pc, 3135 ao.addr16[0], an->addr16[0], u), 3136 ao.addr16[1], an->addr16[1], u), 3137 ao.addr16[2], an->addr16[2], u), 3138 ao.addr16[3], an->addr16[3], u), 3139 ao.addr16[4], an->addr16[4], u), 3140 ao.addr16[5], an->addr16[5], u), 3141 ao.addr16[6], an->addr16[6], u), 3142 ao.addr16[7], an->addr16[7], u); 3143 3144 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3145 break; 3146 #endif /* INET6 */ 3147 } 3148 3149 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3150 CSUM_DELAY_DATA_IPV6)) { 3151 *pc = ~*pc; 3152 if (! *pc) 3153 *pc = 0xffff; 3154 } 3155 } 3156 3157 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3158 void 3159 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3160 { 3161 u_int32_t ao; 3162 3163 memcpy(&ao, a, sizeof(ao)); 3164 memcpy(a, &an, sizeof(u_int32_t)); 3165 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3166 ao % 65536, an % 65536, u); 3167 } 3168 3169 void 3170 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3171 { 3172 u_int32_t ao; 3173 3174 memcpy(&ao, a, sizeof(ao)); 3175 memcpy(a, &an, sizeof(u_int32_t)); 3176 3177 *c = pf_proto_cksum_fixup(m, 3178 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3179 ao % 65536, an % 65536, udp); 3180 } 3181 3182 #ifdef INET6 3183 static void 3184 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3185 { 3186 struct pf_addr ao; 3187 3188 PF_ACPY(&ao, a, AF_INET6); 3189 PF_ACPY(a, an, AF_INET6); 3190 3191 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3192 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3193 pf_cksum_fixup(pf_cksum_fixup(*c, 3194 ao.addr16[0], an->addr16[0], u), 3195 ao.addr16[1], an->addr16[1], u), 3196 ao.addr16[2], an->addr16[2], u), 3197 ao.addr16[3], an->addr16[3], u), 3198 ao.addr16[4], an->addr16[4], u), 3199 ao.addr16[5], an->addr16[5], u), 3200 ao.addr16[6], an->addr16[6], u), 3201 ao.addr16[7], an->addr16[7], u); 3202 } 3203 #endif /* INET6 */ 3204 3205 static void 3206 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3207 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3208 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3209 { 3210 struct pf_addr oia, ooa; 3211 3212 PF_ACPY(&oia, ia, af); 3213 if (oa) 3214 PF_ACPY(&ooa, oa, af); 3215 3216 /* Change inner protocol port, fix inner protocol checksum. */ 3217 if (ip != NULL) { 3218 u_int16_t oip = *ip; 3219 u_int32_t opc; 3220 3221 if (pc != NULL) 3222 opc = *pc; 3223 *ip = np; 3224 if (pc != NULL) 3225 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3226 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3227 if (pc != NULL) 3228 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3229 } 3230 /* Change inner ip address, fix inner ip and icmp checksums. */ 3231 PF_ACPY(ia, na, af); 3232 switch (af) { 3233 #ifdef INET 3234 case AF_INET: { 3235 u_int32_t oh2c = *h2c; 3236 3237 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3238 oia.addr16[0], ia->addr16[0], 0), 3239 oia.addr16[1], ia->addr16[1], 0); 3240 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3241 oia.addr16[0], ia->addr16[0], 0), 3242 oia.addr16[1], ia->addr16[1], 0); 3243 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3244 break; 3245 } 3246 #endif /* INET */ 3247 #ifdef INET6 3248 case AF_INET6: 3249 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3250 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3251 pf_cksum_fixup(pf_cksum_fixup(*ic, 3252 oia.addr16[0], ia->addr16[0], u), 3253 oia.addr16[1], ia->addr16[1], u), 3254 oia.addr16[2], ia->addr16[2], u), 3255 oia.addr16[3], ia->addr16[3], u), 3256 oia.addr16[4], ia->addr16[4], u), 3257 oia.addr16[5], ia->addr16[5], u), 3258 oia.addr16[6], ia->addr16[6], u), 3259 oia.addr16[7], ia->addr16[7], u); 3260 break; 3261 #endif /* INET6 */ 3262 } 3263 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3264 if (oa) { 3265 PF_ACPY(oa, na, af); 3266 switch (af) { 3267 #ifdef INET 3268 case AF_INET: 3269 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3270 ooa.addr16[0], oa->addr16[0], 0), 3271 ooa.addr16[1], oa->addr16[1], 0); 3272 break; 3273 #endif /* INET */ 3274 #ifdef INET6 3275 case AF_INET6: 3276 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3277 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3278 pf_cksum_fixup(pf_cksum_fixup(*ic, 3279 ooa.addr16[0], oa->addr16[0], u), 3280 ooa.addr16[1], oa->addr16[1], u), 3281 ooa.addr16[2], oa->addr16[2], u), 3282 ooa.addr16[3], oa->addr16[3], u), 3283 ooa.addr16[4], oa->addr16[4], u), 3284 ooa.addr16[5], oa->addr16[5], u), 3285 ooa.addr16[6], oa->addr16[6], u), 3286 ooa.addr16[7], oa->addr16[7], u); 3287 break; 3288 #endif /* INET6 */ 3289 } 3290 } 3291 } 3292 3293 /* 3294 * Need to modulate the sequence numbers in the TCP SACK option 3295 * (credits to Krzysztof Pfaff for report and patch) 3296 */ 3297 static int 3298 pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th, 3299 struct pf_state_peer *dst) 3300 { 3301 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3302 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3303 int copyback = 0, i, olen; 3304 struct sackblk sack; 3305 3306 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3307 if (hlen < TCPOLEN_SACKLEN || 3308 !pf_pull_hdr(pd->m, pd->off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3309 return 0; 3310 3311 while (hlen >= TCPOLEN_SACKLEN) { 3312 size_t startoff = opt - opts; 3313 olen = opt[1]; 3314 switch (*opt) { 3315 case TCPOPT_EOL: /* FALLTHROUGH */ 3316 case TCPOPT_NOP: 3317 opt++; 3318 hlen--; 3319 break; 3320 case TCPOPT_SACK: 3321 if (olen > hlen) 3322 olen = hlen; 3323 if (olen >= TCPOLEN_SACKLEN) { 3324 for (i = 2; i + TCPOLEN_SACK <= olen; 3325 i += TCPOLEN_SACK) { 3326 memcpy(&sack, &opt[i], sizeof(sack)); 3327 pf_patch_32_unaligned(pd->m, 3328 &th->th_sum, &sack.start, 3329 htonl(ntohl(sack.start) - dst->seqdiff), 3330 PF_ALGNMNT(startoff), 3331 0); 3332 pf_patch_32_unaligned(pd->m, &th->th_sum, 3333 &sack.end, 3334 htonl(ntohl(sack.end) - dst->seqdiff), 3335 PF_ALGNMNT(startoff), 3336 0); 3337 memcpy(&opt[i], &sack, sizeof(sack)); 3338 } 3339 copyback = 1; 3340 } 3341 /* FALLTHROUGH */ 3342 default: 3343 if (olen < 2) 3344 olen = 2; 3345 hlen -= olen; 3346 opt += olen; 3347 } 3348 } 3349 3350 if (copyback) 3351 m_copyback(pd->m, pd->off + sizeof(*th), thoptlen, (caddr_t)opts); 3352 return (copyback); 3353 } 3354 3355 struct mbuf * 3356 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3357 const struct pf_addr *saddr, const struct pf_addr *daddr, 3358 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3359 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3360 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3361 { 3362 struct mbuf *m; 3363 int len, tlen; 3364 #ifdef INET 3365 struct ip *h = NULL; 3366 #endif /* INET */ 3367 #ifdef INET6 3368 struct ip6_hdr *h6 = NULL; 3369 #endif /* INET6 */ 3370 struct tcphdr *th; 3371 char *opt; 3372 struct pf_mtag *pf_mtag; 3373 3374 len = 0; 3375 th = NULL; 3376 3377 /* maximum segment size tcp option */ 3378 tlen = sizeof(struct tcphdr); 3379 if (mss) 3380 tlen += 4; 3381 3382 switch (af) { 3383 #ifdef INET 3384 case AF_INET: 3385 len = sizeof(struct ip) + tlen; 3386 break; 3387 #endif /* INET */ 3388 #ifdef INET6 3389 case AF_INET6: 3390 len = sizeof(struct ip6_hdr) + tlen; 3391 break; 3392 #endif /* INET6 */ 3393 } 3394 3395 m = m_gethdr(M_NOWAIT, MT_DATA); 3396 if (m == NULL) 3397 return (NULL); 3398 3399 #ifdef MAC 3400 mac_netinet_firewall_send(m); 3401 #endif 3402 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3403 m_freem(m); 3404 return (NULL); 3405 } 3406 m->m_flags |= mbuf_flags; 3407 pf_mtag->tag = mtag_tag; 3408 pf_mtag->flags = mtag_flags; 3409 3410 if (rtableid >= 0) 3411 M_SETFIB(m, rtableid); 3412 3413 #ifdef ALTQ 3414 if (r != NULL && r->qid) { 3415 pf_mtag->qid = r->qid; 3416 3417 /* add hints for ecn */ 3418 pf_mtag->hdr = mtod(m, struct ip *); 3419 } 3420 #endif /* ALTQ */ 3421 m->m_data += max_linkhdr; 3422 m->m_pkthdr.len = m->m_len = len; 3423 /* The rest of the stack assumes a rcvif, so provide one. 3424 * This is a locally generated packet, so .. close enough. */ 3425 m->m_pkthdr.rcvif = V_loif; 3426 bzero(m->m_data, len); 3427 switch (af) { 3428 #ifdef INET 3429 case AF_INET: 3430 h = mtod(m, struct ip *); 3431 3432 /* IP header fields included in the TCP checksum */ 3433 h->ip_p = IPPROTO_TCP; 3434 h->ip_len = htons(tlen); 3435 h->ip_src.s_addr = saddr->v4.s_addr; 3436 h->ip_dst.s_addr = daddr->v4.s_addr; 3437 3438 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3439 break; 3440 #endif /* INET */ 3441 #ifdef INET6 3442 case AF_INET6: 3443 h6 = mtod(m, struct ip6_hdr *); 3444 3445 /* IP header fields included in the TCP checksum */ 3446 h6->ip6_nxt = IPPROTO_TCP; 3447 h6->ip6_plen = htons(tlen); 3448 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3449 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3450 3451 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3452 break; 3453 #endif /* INET6 */ 3454 } 3455 3456 /* TCP header */ 3457 th->th_sport = sport; 3458 th->th_dport = dport; 3459 th->th_seq = htonl(seq); 3460 th->th_ack = htonl(ack); 3461 th->th_off = tlen >> 2; 3462 th->th_flags = tcp_flags; 3463 th->th_win = htons(win); 3464 3465 if (mss) { 3466 opt = (char *)(th + 1); 3467 opt[0] = TCPOPT_MAXSEG; 3468 opt[1] = 4; 3469 HTONS(mss); 3470 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3471 } 3472 3473 switch (af) { 3474 #ifdef INET 3475 case AF_INET: 3476 /* TCP checksum */ 3477 th->th_sum = in_cksum(m, len); 3478 3479 /* Finish the IP header */ 3480 h->ip_v = 4; 3481 h->ip_hl = sizeof(*h) >> 2; 3482 h->ip_tos = IPTOS_LOWDELAY; 3483 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3484 h->ip_len = htons(len); 3485 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3486 h->ip_sum = 0; 3487 break; 3488 #endif /* INET */ 3489 #ifdef INET6 3490 case AF_INET6: 3491 /* TCP checksum */ 3492 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3493 sizeof(struct ip6_hdr), tlen); 3494 3495 h6->ip6_vfc |= IPV6_VERSION; 3496 h6->ip6_hlim = IPV6_DEFHLIM; 3497 break; 3498 #endif /* INET6 */ 3499 } 3500 3501 return (m); 3502 } 3503 3504 static void 3505 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3506 uint8_t ttl, int rtableid) 3507 { 3508 struct mbuf *m; 3509 #ifdef INET 3510 struct ip *h = NULL; 3511 #endif /* INET */ 3512 #ifdef INET6 3513 struct ip6_hdr *h6 = NULL; 3514 #endif /* INET6 */ 3515 struct sctphdr *hdr; 3516 struct sctp_chunkhdr *chunk; 3517 struct pf_send_entry *pfse; 3518 int off = 0; 3519 3520 MPASS(af == pd->af); 3521 3522 m = m_gethdr(M_NOWAIT, MT_DATA); 3523 if (m == NULL) 3524 return; 3525 3526 m->m_data += max_linkhdr; 3527 m->m_flags |= M_SKIP_FIREWALL; 3528 /* The rest of the stack assumes a rcvif, so provide one. 3529 * This is a locally generated packet, so .. close enough. */ 3530 m->m_pkthdr.rcvif = V_loif; 3531 3532 /* IPv4|6 header */ 3533 switch (af) { 3534 #ifdef INET 3535 case AF_INET: 3536 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3537 3538 h = mtod(m, struct ip *); 3539 3540 /* IP header fields included in the TCP checksum */ 3541 3542 h->ip_p = IPPROTO_SCTP; 3543 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3544 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3545 h->ip_src = pd->dst->v4; 3546 h->ip_dst = pd->src->v4; 3547 3548 off += sizeof(struct ip); 3549 break; 3550 #endif /* INET */ 3551 #ifdef INET6 3552 case AF_INET6: 3553 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3554 3555 h6 = mtod(m, struct ip6_hdr *); 3556 3557 /* IP header fields included in the TCP checksum */ 3558 h6->ip6_vfc |= IPV6_VERSION; 3559 h6->ip6_nxt = IPPROTO_SCTP; 3560 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3561 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3562 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3563 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3564 3565 off += sizeof(struct ip6_hdr); 3566 break; 3567 #endif /* INET6 */ 3568 } 3569 3570 /* SCTP header */ 3571 hdr = mtodo(m, off); 3572 3573 hdr->src_port = pd->hdr.sctp.dest_port; 3574 hdr->dest_port = pd->hdr.sctp.src_port; 3575 hdr->v_tag = pd->sctp_initiate_tag; 3576 hdr->checksum = 0; 3577 3578 /* Abort chunk. */ 3579 off += sizeof(struct sctphdr); 3580 chunk = mtodo(m, off); 3581 3582 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3583 chunk->chunk_length = htons(sizeof(*chunk)); 3584 3585 /* SCTP checksum */ 3586 off += sizeof(*chunk); 3587 m->m_pkthdr.len = m->m_len = off; 3588 3589 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3590 3591 if (rtableid >= 0) 3592 M_SETFIB(m, rtableid); 3593 3594 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3595 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3596 if (pfse == NULL) { 3597 m_freem(m); 3598 return; 3599 } 3600 3601 switch (af) { 3602 #ifdef INET 3603 case AF_INET: 3604 pfse->pfse_type = PFSE_IP; 3605 break; 3606 #endif /* INET */ 3607 #ifdef INET6 3608 case AF_INET6: 3609 pfse->pfse_type = PFSE_IP6; 3610 break; 3611 #endif /* INET6 */ 3612 } 3613 3614 pfse->pfse_m = m; 3615 pf_send(pfse); 3616 } 3617 3618 void 3619 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3620 const struct pf_addr *saddr, const struct pf_addr *daddr, 3621 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3622 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3623 int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3624 { 3625 struct pf_send_entry *pfse; 3626 struct mbuf *m; 3627 3628 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3629 win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, rtableid); 3630 if (m == NULL) 3631 return; 3632 3633 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3634 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3635 if (pfse == NULL) { 3636 m_freem(m); 3637 return; 3638 } 3639 3640 switch (af) { 3641 #ifdef INET 3642 case AF_INET: 3643 pfse->pfse_type = PFSE_IP; 3644 break; 3645 #endif /* INET */ 3646 #ifdef INET6 3647 case AF_INET6: 3648 pfse->pfse_type = PFSE_IP6; 3649 break; 3650 #endif /* INET6 */ 3651 } 3652 3653 pfse->pfse_m = m; 3654 pf_send(pfse); 3655 } 3656 3657 static void 3658 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3659 struct pf_state_key *sk, struct tcphdr *th, 3660 u_int16_t bproto_sum, u_int16_t bip_sum, 3661 u_short *reason, int rtableid) 3662 { 3663 struct pf_addr * const saddr = pd->src; 3664 struct pf_addr * const daddr = pd->dst; 3665 3666 /* undo NAT changes, if they have taken place */ 3667 if (nr != NULL) { 3668 PF_ACPY(saddr, &sk->addr[pd->sidx], pd->af); 3669 PF_ACPY(daddr, &sk->addr[pd->didx], pd->af); 3670 if (pd->sport) 3671 *pd->sport = sk->port[pd->sidx]; 3672 if (pd->dport) 3673 *pd->dport = sk->port[pd->didx]; 3674 if (pd->proto_sum) 3675 *pd->proto_sum = bproto_sum; 3676 if (pd->ip_sum) 3677 *pd->ip_sum = bip_sum; 3678 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 3679 } 3680 if (pd->proto == IPPROTO_TCP && 3681 ((r->rule_flag & PFRULE_RETURNRST) || 3682 (r->rule_flag & PFRULE_RETURN)) && 3683 !(th->th_flags & TH_RST)) { 3684 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3685 3686 if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off, 3687 IPPROTO_TCP, pd->af)) 3688 REASON_SET(reason, PFRES_PROTCKSUM); 3689 else { 3690 if (th->th_flags & TH_SYN) 3691 ack++; 3692 if (th->th_flags & TH_FIN) 3693 ack++; 3694 pf_send_tcp(r, pd->af, pd->dst, 3695 pd->src, th->th_dport, th->th_sport, 3696 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3697 r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid); 3698 } 3699 } else if (pd->proto == IPPROTO_SCTP && 3700 (r->rule_flag & PFRULE_RETURN)) { 3701 pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid); 3702 } else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET && 3703 r->return_icmp) 3704 pf_send_icmp(pd->m, r->return_icmp >> 8, 3705 r->return_icmp & 255, pd->af, r, rtableid); 3706 else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 && 3707 r->return_icmp6) 3708 pf_send_icmp(pd->m, r->return_icmp6 >> 8, 3709 r->return_icmp6 & 255, pd->af, r, rtableid); 3710 } 3711 3712 static int 3713 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3714 { 3715 struct m_tag *mtag; 3716 u_int8_t mpcp; 3717 3718 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3719 if (mtag == NULL) 3720 return (0); 3721 3722 if (prio == PF_PRIO_ZERO) 3723 prio = 0; 3724 3725 mpcp = *(uint8_t *)(mtag + 1); 3726 3727 return (mpcp == prio); 3728 } 3729 3730 static int 3731 pf_icmp_to_bandlim(uint8_t type) 3732 { 3733 switch (type) { 3734 case ICMP_ECHO: 3735 case ICMP_ECHOREPLY: 3736 return (BANDLIM_ICMP_ECHO); 3737 case ICMP_TSTAMP: 3738 case ICMP_TSTAMPREPLY: 3739 return (BANDLIM_ICMP_TSTAMP); 3740 case ICMP_UNREACH: 3741 default: 3742 return (BANDLIM_ICMP_UNREACH); 3743 } 3744 } 3745 3746 static void 3747 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3748 struct pf_krule *r, int rtableid) 3749 { 3750 struct pf_send_entry *pfse; 3751 struct mbuf *m0; 3752 struct pf_mtag *pf_mtag; 3753 3754 /* ICMP packet rate limitation. */ 3755 switch (af) { 3756 #ifdef INET6 3757 case AF_INET6: 3758 if (icmp6_ratelimit(NULL, type, code)) 3759 return; 3760 break; 3761 #endif 3762 #ifdef INET 3763 case AF_INET: 3764 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3765 return; 3766 break; 3767 #endif 3768 } 3769 3770 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3771 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3772 if (pfse == NULL) 3773 return; 3774 3775 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3776 free(pfse, M_PFTEMP); 3777 return; 3778 } 3779 3780 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3781 free(pfse, M_PFTEMP); 3782 return; 3783 } 3784 /* XXX: revisit */ 3785 m0->m_flags |= M_SKIP_FIREWALL; 3786 3787 if (rtableid >= 0) 3788 M_SETFIB(m0, rtableid); 3789 3790 #ifdef ALTQ 3791 if (r->qid) { 3792 pf_mtag->qid = r->qid; 3793 /* add hints for ecn */ 3794 pf_mtag->hdr = mtod(m0, struct ip *); 3795 } 3796 #endif /* ALTQ */ 3797 3798 switch (af) { 3799 #ifdef INET 3800 case AF_INET: 3801 pfse->pfse_type = PFSE_ICMP; 3802 break; 3803 #endif /* INET */ 3804 #ifdef INET6 3805 case AF_INET6: 3806 pfse->pfse_type = PFSE_ICMP6; 3807 break; 3808 #endif /* INET6 */ 3809 } 3810 pfse->pfse_m = m0; 3811 pfse->icmpopts.type = type; 3812 pfse->icmpopts.code = code; 3813 pf_send(pfse); 3814 } 3815 3816 /* 3817 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3818 * If n is 0, they match if they are equal. If n is != 0, they match if they 3819 * are different. 3820 */ 3821 int 3822 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3823 struct pf_addr *b, sa_family_t af) 3824 { 3825 int match = 0; 3826 3827 switch (af) { 3828 #ifdef INET 3829 case AF_INET: 3830 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3831 match++; 3832 break; 3833 #endif /* INET */ 3834 #ifdef INET6 3835 case AF_INET6: 3836 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3837 match++; 3838 break; 3839 #endif /* INET6 */ 3840 } 3841 if (match) { 3842 if (n) 3843 return (0); 3844 else 3845 return (1); 3846 } else { 3847 if (n) 3848 return (1); 3849 else 3850 return (0); 3851 } 3852 } 3853 3854 /* 3855 * Return 1 if b <= a <= e, otherwise return 0. 3856 */ 3857 int 3858 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3859 struct pf_addr *a, sa_family_t af) 3860 { 3861 switch (af) { 3862 #ifdef INET 3863 case AF_INET: 3864 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3865 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3866 return (0); 3867 break; 3868 #endif /* INET */ 3869 #ifdef INET6 3870 case AF_INET6: { 3871 int i; 3872 3873 /* check a >= b */ 3874 for (i = 0; i < 4; ++i) 3875 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3876 break; 3877 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3878 return (0); 3879 /* check a <= e */ 3880 for (i = 0; i < 4; ++i) 3881 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3882 break; 3883 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3884 return (0); 3885 break; 3886 } 3887 #endif /* INET6 */ 3888 } 3889 return (1); 3890 } 3891 3892 static int 3893 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3894 { 3895 switch (op) { 3896 case PF_OP_IRG: 3897 return ((p > a1) && (p < a2)); 3898 case PF_OP_XRG: 3899 return ((p < a1) || (p > a2)); 3900 case PF_OP_RRG: 3901 return ((p >= a1) && (p <= a2)); 3902 case PF_OP_EQ: 3903 return (p == a1); 3904 case PF_OP_NE: 3905 return (p != a1); 3906 case PF_OP_LT: 3907 return (p < a1); 3908 case PF_OP_LE: 3909 return (p <= a1); 3910 case PF_OP_GT: 3911 return (p > a1); 3912 case PF_OP_GE: 3913 return (p >= a1); 3914 } 3915 return (0); /* never reached */ 3916 } 3917 3918 int 3919 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3920 { 3921 NTOHS(a1); 3922 NTOHS(a2); 3923 NTOHS(p); 3924 return (pf_match(op, a1, a2, p)); 3925 } 3926 3927 static int 3928 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3929 { 3930 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3931 return (0); 3932 return (pf_match(op, a1, a2, u)); 3933 } 3934 3935 static int 3936 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3937 { 3938 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3939 return (0); 3940 return (pf_match(op, a1, a2, g)); 3941 } 3942 3943 int 3944 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3945 { 3946 if (*tag == -1) 3947 *tag = mtag; 3948 3949 return ((!r->match_tag_not && r->match_tag == *tag) || 3950 (r->match_tag_not && r->match_tag != *tag)); 3951 } 3952 3953 static int 3954 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3955 { 3956 struct ifnet *ifp = m->m_pkthdr.rcvif; 3957 struct pfi_kkif *kif; 3958 3959 if (ifp == NULL) 3960 return (0); 3961 3962 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3963 3964 if (kif == NULL) { 3965 DPFPRINTF(PF_DEBUG_URGENT, 3966 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3967 r->rcv_ifname)); 3968 return (0); 3969 } 3970 3971 return (pfi_kkif_match(r->rcv_kif, kif)); 3972 } 3973 3974 int 3975 pf_tag_packet(struct pf_pdesc *pd, int tag) 3976 { 3977 3978 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3979 3980 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) 3981 return (ENOMEM); 3982 3983 pd->pf_mtag->tag = tag; 3984 3985 return (0); 3986 } 3987 3988 #define PF_ANCHOR_STACKSIZE 32 3989 struct pf_kanchor_stackframe { 3990 struct pf_kruleset *rs; 3991 struct pf_krule *r; /* XXX: + match bit */ 3992 struct pf_kanchor *child; 3993 }; 3994 3995 /* 3996 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3997 */ 3998 #define PF_ANCHORSTACK_MATCH 0x00000001 3999 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4000 4001 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4002 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4003 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4004 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4005 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4006 } while (0) 4007 4008 void 4009 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4010 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4011 int *match) 4012 { 4013 struct pf_kanchor_stackframe *f; 4014 4015 PF_RULES_RASSERT(); 4016 4017 if (match) 4018 *match = 0; 4019 if (*depth >= PF_ANCHOR_STACKSIZE) { 4020 printf("%s: anchor stack overflow on %s\n", 4021 __func__, (*r)->anchor->name); 4022 *r = TAILQ_NEXT(*r, entries); 4023 return; 4024 } else if (*depth == 0 && a != NULL) 4025 *a = *r; 4026 f = stack + (*depth)++; 4027 f->rs = *rs; 4028 f->r = *r; 4029 if ((*r)->anchor_wildcard) { 4030 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4031 4032 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4033 *r = NULL; 4034 return; 4035 } 4036 *rs = &f->child->ruleset; 4037 } else { 4038 f->child = NULL; 4039 *rs = &(*r)->anchor->ruleset; 4040 } 4041 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4042 } 4043 4044 int 4045 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4046 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4047 int *match) 4048 { 4049 struct pf_kanchor_stackframe *f; 4050 struct pf_krule *fr; 4051 int quick = 0; 4052 4053 PF_RULES_RASSERT(); 4054 4055 do { 4056 if (*depth <= 0) 4057 break; 4058 f = stack + *depth - 1; 4059 fr = PF_ANCHOR_RULE(f); 4060 if (f->child != NULL) { 4061 /* 4062 * This block traverses through 4063 * a wildcard anchor. 4064 */ 4065 if (match != NULL && *match) { 4066 /* 4067 * If any of "*" matched, then 4068 * "foo/ *" matched, mark frame 4069 * appropriately. 4070 */ 4071 PF_ANCHOR_SET_MATCH(f); 4072 *match = 0; 4073 } 4074 f->child = RB_NEXT(pf_kanchor_node, 4075 &fr->anchor->children, f->child); 4076 if (f->child != NULL) { 4077 *rs = &f->child->ruleset; 4078 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4079 if (*r == NULL) 4080 continue; 4081 else 4082 break; 4083 } 4084 } 4085 (*depth)--; 4086 if (*depth == 0 && a != NULL) 4087 *a = NULL; 4088 *rs = f->rs; 4089 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4090 quick = fr->quick; 4091 *r = TAILQ_NEXT(fr, entries); 4092 } while (*r == NULL); 4093 4094 return (quick); 4095 } 4096 4097 struct pf_keth_anchor_stackframe { 4098 struct pf_keth_ruleset *rs; 4099 struct pf_keth_rule *r; /* XXX: + match bit */ 4100 struct pf_keth_anchor *child; 4101 }; 4102 4103 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4104 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4105 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4106 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4107 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4108 } while (0) 4109 4110 void 4111 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4112 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4113 struct pf_keth_rule **a, int *match) 4114 { 4115 struct pf_keth_anchor_stackframe *f; 4116 4117 NET_EPOCH_ASSERT(); 4118 4119 if (match) 4120 *match = 0; 4121 if (*depth >= PF_ANCHOR_STACKSIZE) { 4122 printf("%s: anchor stack overflow on %s\n", 4123 __func__, (*r)->anchor->name); 4124 *r = TAILQ_NEXT(*r, entries); 4125 return; 4126 } else if (*depth == 0 && a != NULL) 4127 *a = *r; 4128 f = stack + (*depth)++; 4129 f->rs = *rs; 4130 f->r = *r; 4131 if ((*r)->anchor_wildcard) { 4132 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4133 4134 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4135 *r = NULL; 4136 return; 4137 } 4138 *rs = &f->child->ruleset; 4139 } else { 4140 f->child = NULL; 4141 *rs = &(*r)->anchor->ruleset; 4142 } 4143 *r = TAILQ_FIRST((*rs)->active.rules); 4144 } 4145 4146 int 4147 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4148 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4149 struct pf_keth_rule **a, int *match) 4150 { 4151 struct pf_keth_anchor_stackframe *f; 4152 struct pf_keth_rule *fr; 4153 int quick = 0; 4154 4155 NET_EPOCH_ASSERT(); 4156 4157 do { 4158 if (*depth <= 0) 4159 break; 4160 f = stack + *depth - 1; 4161 fr = PF_ETH_ANCHOR_RULE(f); 4162 if (f->child != NULL) { 4163 /* 4164 * This block traverses through 4165 * a wildcard anchor. 4166 */ 4167 if (match != NULL && *match) { 4168 /* 4169 * If any of "*" matched, then 4170 * "foo/ *" matched, mark frame 4171 * appropriately. 4172 */ 4173 PF_ETH_ANCHOR_SET_MATCH(f); 4174 *match = 0; 4175 } 4176 f->child = RB_NEXT(pf_keth_anchor_node, 4177 &fr->anchor->children, f->child); 4178 if (f->child != NULL) { 4179 *rs = &f->child->ruleset; 4180 *r = TAILQ_FIRST((*rs)->active.rules); 4181 if (*r == NULL) 4182 continue; 4183 else 4184 break; 4185 } 4186 } 4187 (*depth)--; 4188 if (*depth == 0 && a != NULL) 4189 *a = NULL; 4190 *rs = f->rs; 4191 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4192 quick = fr->quick; 4193 *r = TAILQ_NEXT(fr, entries); 4194 } while (*r == NULL); 4195 4196 return (quick); 4197 } 4198 4199 #ifdef INET6 4200 void 4201 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4202 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4203 { 4204 switch (af) { 4205 #ifdef INET 4206 case AF_INET: 4207 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4208 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4209 break; 4210 #endif /* INET */ 4211 case AF_INET6: 4212 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4213 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4214 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4215 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4216 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4217 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4218 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4219 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4220 break; 4221 } 4222 } 4223 4224 void 4225 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4226 { 4227 switch (af) { 4228 #ifdef INET 4229 case AF_INET: 4230 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4231 break; 4232 #endif /* INET */ 4233 case AF_INET6: 4234 if (addr->addr32[3] == 0xffffffff) { 4235 addr->addr32[3] = 0; 4236 if (addr->addr32[2] == 0xffffffff) { 4237 addr->addr32[2] = 0; 4238 if (addr->addr32[1] == 0xffffffff) { 4239 addr->addr32[1] = 0; 4240 addr->addr32[0] = 4241 htonl(ntohl(addr->addr32[0]) + 1); 4242 } else 4243 addr->addr32[1] = 4244 htonl(ntohl(addr->addr32[1]) + 1); 4245 } else 4246 addr->addr32[2] = 4247 htonl(ntohl(addr->addr32[2]) + 1); 4248 } else 4249 addr->addr32[3] = 4250 htonl(ntohl(addr->addr32[3]) + 1); 4251 break; 4252 } 4253 } 4254 #endif /* INET6 */ 4255 4256 void 4257 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4258 { 4259 /* 4260 * Modern rules use the same flags in rules as they do in states. 4261 */ 4262 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4263 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4264 4265 /* 4266 * Old-style scrub rules have different flags which need to be translated. 4267 */ 4268 if (r->rule_flag & PFRULE_RANDOMID) 4269 a->flags |= PFSTATE_RANDOMID; 4270 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4271 a->flags |= PFSTATE_SETTOS; 4272 a->set_tos = r->set_tos; 4273 } 4274 4275 if (r->qid) 4276 a->qid = r->qid; 4277 if (r->pqid) 4278 a->pqid = r->pqid; 4279 if (r->rtableid >= 0) 4280 a->rtableid = r->rtableid; 4281 a->log |= r->log; 4282 if (r->min_ttl) 4283 a->min_ttl = r->min_ttl; 4284 if (r->max_mss) 4285 a->max_mss = r->max_mss; 4286 if (r->dnpipe) 4287 a->dnpipe = r->dnpipe; 4288 if (r->dnrpipe) 4289 a->dnrpipe = r->dnrpipe; 4290 if (r->dnpipe || r->dnrpipe) { 4291 if (r->free_flags & PFRULE_DN_IS_PIPE) 4292 a->flags |= PFSTATE_DN_IS_PIPE; 4293 else 4294 a->flags &= ~PFSTATE_DN_IS_PIPE; 4295 } 4296 if (r->scrub_flags & PFSTATE_SETPRIO) { 4297 a->set_prio[0] = r->set_prio[0]; 4298 a->set_prio[1] = r->set_prio[1]; 4299 } 4300 } 4301 4302 int 4303 pf_socket_lookup(struct pf_pdesc *pd) 4304 { 4305 struct pf_addr *saddr, *daddr; 4306 u_int16_t sport, dport; 4307 struct inpcbinfo *pi; 4308 struct inpcb *inp; 4309 4310 pd->lookup.uid = UID_MAX; 4311 pd->lookup.gid = GID_MAX; 4312 4313 switch (pd->proto) { 4314 case IPPROTO_TCP: 4315 sport = pd->hdr.tcp.th_sport; 4316 dport = pd->hdr.tcp.th_dport; 4317 pi = &V_tcbinfo; 4318 break; 4319 case IPPROTO_UDP: 4320 sport = pd->hdr.udp.uh_sport; 4321 dport = pd->hdr.udp.uh_dport; 4322 pi = &V_udbinfo; 4323 break; 4324 default: 4325 return (-1); 4326 } 4327 if (pd->dir == PF_IN) { 4328 saddr = pd->src; 4329 daddr = pd->dst; 4330 } else { 4331 u_int16_t p; 4332 4333 p = sport; 4334 sport = dport; 4335 dport = p; 4336 saddr = pd->dst; 4337 daddr = pd->src; 4338 } 4339 switch (pd->af) { 4340 #ifdef INET 4341 case AF_INET: 4342 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4343 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4344 if (inp == NULL) { 4345 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4346 daddr->v4, dport, INPLOOKUP_WILDCARD | 4347 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4348 if (inp == NULL) 4349 return (-1); 4350 } 4351 break; 4352 #endif /* INET */ 4353 #ifdef INET6 4354 case AF_INET6: 4355 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4356 dport, INPLOOKUP_RLOCKPCB, NULL, pd->m); 4357 if (inp == NULL) { 4358 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4359 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4360 INPLOOKUP_RLOCKPCB, NULL, pd->m); 4361 if (inp == NULL) 4362 return (-1); 4363 } 4364 break; 4365 #endif /* INET6 */ 4366 } 4367 INP_RLOCK_ASSERT(inp); 4368 pd->lookup.uid = inp->inp_cred->cr_uid; 4369 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4370 INP_RUNLOCK(inp); 4371 4372 return (1); 4373 } 4374 4375 u_int8_t 4376 pf_get_wscale(struct pf_pdesc *pd) 4377 { 4378 struct tcphdr *th = &pd->hdr.tcp; 4379 int hlen; 4380 u_int8_t hdr[60]; 4381 u_int8_t *opt, optlen; 4382 u_int8_t wscale = 0; 4383 4384 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4385 if (hlen <= sizeof(struct tcphdr)) 4386 return (0); 4387 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4388 return (0); 4389 opt = hdr + sizeof(struct tcphdr); 4390 hlen -= sizeof(struct tcphdr); 4391 while (hlen >= 3) { 4392 switch (*opt) { 4393 case TCPOPT_EOL: 4394 case TCPOPT_NOP: 4395 ++opt; 4396 --hlen; 4397 break; 4398 case TCPOPT_WINDOW: 4399 wscale = opt[2]; 4400 if (wscale > TCP_MAX_WINSHIFT) 4401 wscale = TCP_MAX_WINSHIFT; 4402 wscale |= PF_WSCALE_FLAG; 4403 /* FALLTHROUGH */ 4404 default: 4405 optlen = opt[1]; 4406 if (optlen < 2) 4407 optlen = 2; 4408 hlen -= optlen; 4409 opt += optlen; 4410 break; 4411 } 4412 } 4413 return (wscale); 4414 } 4415 4416 u_int16_t 4417 pf_get_mss(struct pf_pdesc *pd) 4418 { 4419 struct tcphdr *th = &pd->hdr.tcp; 4420 int hlen; 4421 u_int8_t hdr[60]; 4422 u_int8_t *opt, optlen; 4423 u_int16_t mss = V_tcp_mssdflt; 4424 4425 hlen = th->th_off << 2; /* hlen <= sizeof(hdr) */ 4426 if (hlen <= sizeof(struct tcphdr)) 4427 return (0); 4428 if (!pf_pull_hdr(pd->m, pd->off, hdr, hlen, NULL, NULL, pd->af)) 4429 return (0); 4430 opt = hdr + sizeof(struct tcphdr); 4431 hlen -= sizeof(struct tcphdr); 4432 while (hlen >= TCPOLEN_MAXSEG) { 4433 switch (*opt) { 4434 case TCPOPT_EOL: 4435 case TCPOPT_NOP: 4436 ++opt; 4437 --hlen; 4438 break; 4439 case TCPOPT_MAXSEG: 4440 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4441 NTOHS(mss); 4442 /* FALLTHROUGH */ 4443 default: 4444 optlen = opt[1]; 4445 if (optlen < 2) 4446 optlen = 2; 4447 hlen -= optlen; 4448 opt += optlen; 4449 break; 4450 } 4451 } 4452 return (mss); 4453 } 4454 4455 static u_int16_t 4456 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4457 { 4458 struct nhop_object *nh; 4459 #ifdef INET6 4460 struct in6_addr dst6; 4461 uint32_t scopeid; 4462 #endif /* INET6 */ 4463 int hlen = 0; 4464 uint16_t mss = 0; 4465 4466 NET_EPOCH_ASSERT(); 4467 4468 switch (af) { 4469 #ifdef INET 4470 case AF_INET: 4471 hlen = sizeof(struct ip); 4472 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4473 if (nh != NULL) 4474 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4475 break; 4476 #endif /* INET */ 4477 #ifdef INET6 4478 case AF_INET6: 4479 hlen = sizeof(struct ip6_hdr); 4480 in6_splitscope(&addr->v6, &dst6, &scopeid); 4481 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4482 if (nh != NULL) 4483 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4484 break; 4485 #endif /* INET6 */ 4486 } 4487 4488 mss = max(V_tcp_mssdflt, mss); 4489 mss = min(mss, offer); 4490 mss = max(mss, 64); /* sanity - at least max opt space */ 4491 return (mss); 4492 } 4493 4494 static u_int32_t 4495 pf_tcp_iss(struct pf_pdesc *pd) 4496 { 4497 MD5_CTX ctx; 4498 u_int32_t digest[4]; 4499 4500 if (V_pf_tcp_secret_init == 0) { 4501 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4502 MD5Init(&V_pf_tcp_secret_ctx); 4503 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4504 sizeof(V_pf_tcp_secret)); 4505 V_pf_tcp_secret_init = 1; 4506 } 4507 4508 ctx = V_pf_tcp_secret_ctx; 4509 4510 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4511 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4512 switch (pd->af) { 4513 case AF_INET6: 4514 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4515 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4516 break; 4517 case AF_INET: 4518 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4519 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4520 break; 4521 } 4522 MD5Final((u_char *)digest, &ctx); 4523 V_pf_tcp_iss_off += 4096; 4524 #define ISN_RANDOM_INCREMENT (4096 - 1) 4525 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4526 V_pf_tcp_iss_off); 4527 #undef ISN_RANDOM_INCREMENT 4528 } 4529 4530 static bool 4531 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4532 { 4533 bool match = true; 4534 4535 /* Always matches if not set */ 4536 if (! r->isset) 4537 return (!r->neg); 4538 4539 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4540 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4541 match = false; 4542 break; 4543 } 4544 } 4545 4546 return (match ^ r->neg); 4547 } 4548 4549 static int 4550 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4551 { 4552 if (*tag == -1) 4553 *tag = mtag; 4554 4555 return ((!r->match_tag_not && r->match_tag == *tag) || 4556 (r->match_tag_not && r->match_tag != *tag)); 4557 } 4558 4559 static void 4560 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4561 { 4562 /* If we don't have the interface drop the packet. */ 4563 if (ifp == NULL) { 4564 m_freem(m); 4565 return; 4566 } 4567 4568 switch (ifp->if_type) { 4569 case IFT_ETHER: 4570 case IFT_XETHER: 4571 case IFT_L2VLAN: 4572 case IFT_BRIDGE: 4573 case IFT_IEEE8023ADLAG: 4574 break; 4575 default: 4576 m_freem(m); 4577 return; 4578 } 4579 4580 ifp->if_transmit(ifp, m); 4581 } 4582 4583 static int 4584 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4585 { 4586 #ifdef INET 4587 struct ip ip; 4588 #endif 4589 #ifdef INET6 4590 struct ip6_hdr ip6; 4591 #endif 4592 struct mbuf *m = *m0; 4593 struct ether_header *e; 4594 struct pf_keth_rule *r, *rm, *a = NULL; 4595 struct pf_keth_ruleset *ruleset = NULL; 4596 struct pf_mtag *mtag; 4597 struct pf_keth_ruleq *rules; 4598 struct pf_addr *src = NULL, *dst = NULL; 4599 struct pfi_kkif *bridge_to; 4600 sa_family_t af = 0; 4601 uint16_t proto; 4602 int asd = 0, match = 0; 4603 int tag = -1; 4604 uint8_t action; 4605 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4606 4607 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4608 NET_EPOCH_ASSERT(); 4609 4610 PF_RULES_RLOCK_TRACKER; 4611 4612 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4613 4614 mtag = pf_find_mtag(m); 4615 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4616 /* Dummynet re-injects packets after they've 4617 * completed their delay. We've already 4618 * processed them, so pass unconditionally. */ 4619 4620 /* But only once. We may see the packet multiple times (e.g. 4621 * PFIL_IN/PFIL_OUT). */ 4622 pf_dummynet_flag_remove(m, mtag); 4623 4624 return (PF_PASS); 4625 } 4626 4627 ruleset = V_pf_keth; 4628 rules = ck_pr_load_ptr(&ruleset->active.rules); 4629 r = TAILQ_FIRST(rules); 4630 rm = NULL; 4631 4632 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4633 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4634 DPFPRINTF(PF_DEBUG_URGENT, 4635 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4636 ", pullup failed\n")); 4637 return (PF_DROP); 4638 } 4639 e = mtod(m, struct ether_header *); 4640 proto = ntohs(e->ether_type); 4641 4642 switch (proto) { 4643 #ifdef INET 4644 case ETHERTYPE_IP: { 4645 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4646 sizeof(ip))) 4647 return (PF_DROP); 4648 4649 af = AF_INET; 4650 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4651 (caddr_t)&ip); 4652 src = (struct pf_addr *)&ip.ip_src; 4653 dst = (struct pf_addr *)&ip.ip_dst; 4654 break; 4655 } 4656 #endif /* INET */ 4657 #ifdef INET6 4658 case ETHERTYPE_IPV6: { 4659 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4660 sizeof(ip6))) 4661 return (PF_DROP); 4662 4663 af = AF_INET6; 4664 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4665 (caddr_t)&ip6); 4666 src = (struct pf_addr *)&ip6.ip6_src; 4667 dst = (struct pf_addr *)&ip6.ip6_dst; 4668 break; 4669 } 4670 #endif /* INET6 */ 4671 } 4672 4673 PF_RULES_RLOCK(); 4674 4675 while (r != NULL) { 4676 counter_u64_add(r->evaluations, 1); 4677 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4678 4679 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4680 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4681 "kif"); 4682 r = r->skip[PFE_SKIP_IFP].ptr; 4683 } 4684 else if (r->direction && r->direction != dir) { 4685 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4686 "dir"); 4687 r = r->skip[PFE_SKIP_DIR].ptr; 4688 } 4689 else if (r->proto && r->proto != proto) { 4690 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4691 "proto"); 4692 r = r->skip[PFE_SKIP_PROTO].ptr; 4693 } 4694 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4695 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4696 "src"); 4697 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4698 } 4699 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4700 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4701 "dst"); 4702 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4703 } 4704 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4705 r->ipsrc.neg, kif, M_GETFIB(m))) { 4706 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4707 "ip_src"); 4708 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4709 } 4710 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4711 r->ipdst.neg, kif, M_GETFIB(m))) { 4712 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4713 "ip_dst"); 4714 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4715 } 4716 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4717 mtag ? mtag->tag : 0)) { 4718 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4719 "match_tag"); 4720 r = TAILQ_NEXT(r, entries); 4721 } 4722 else { 4723 if (r->tag) 4724 tag = r->tag; 4725 if (r->anchor == NULL) { 4726 /* Rule matches */ 4727 rm = r; 4728 4729 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4730 4731 if (r->quick) 4732 break; 4733 4734 r = TAILQ_NEXT(r, entries); 4735 } else { 4736 pf_step_into_keth_anchor(anchor_stack, &asd, 4737 &ruleset, &r, &a, &match); 4738 } 4739 } 4740 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4741 &ruleset, &r, &a, &match)) 4742 break; 4743 } 4744 4745 r = rm; 4746 4747 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4748 4749 /* Default to pass. */ 4750 if (r == NULL) { 4751 PF_RULES_RUNLOCK(); 4752 return (PF_PASS); 4753 } 4754 4755 /* Execute action. */ 4756 counter_u64_add(r->packets[dir == PF_OUT], 1); 4757 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4758 pf_update_timestamp(r); 4759 4760 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4761 if (r->action == PF_DROP) { 4762 PF_RULES_RUNLOCK(); 4763 return (PF_DROP); 4764 } 4765 4766 if (tag > 0) { 4767 if (mtag == NULL) 4768 mtag = pf_get_mtag(m); 4769 if (mtag == NULL) { 4770 PF_RULES_RUNLOCK(); 4771 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4772 return (PF_DROP); 4773 } 4774 mtag->tag = tag; 4775 } 4776 4777 if (r->qid != 0) { 4778 if (mtag == NULL) 4779 mtag = pf_get_mtag(m); 4780 if (mtag == NULL) { 4781 PF_RULES_RUNLOCK(); 4782 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4783 return (PF_DROP); 4784 } 4785 mtag->qid = r->qid; 4786 } 4787 4788 action = r->action; 4789 bridge_to = r->bridge_to; 4790 4791 /* Dummynet */ 4792 if (r->dnpipe) { 4793 struct ip_fw_args dnflow; 4794 4795 /* Drop packet if dummynet is not loaded. */ 4796 if (ip_dn_io_ptr == NULL) { 4797 PF_RULES_RUNLOCK(); 4798 m_freem(m); 4799 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4800 return (PF_DROP); 4801 } 4802 if (mtag == NULL) 4803 mtag = pf_get_mtag(m); 4804 if (mtag == NULL) { 4805 PF_RULES_RUNLOCK(); 4806 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4807 return (PF_DROP); 4808 } 4809 4810 bzero(&dnflow, sizeof(dnflow)); 4811 4812 /* We don't have port numbers here, so we set 0. That means 4813 * that we'll be somewhat limited in distinguishing flows (i.e. 4814 * only based on IP addresses, not based on port numbers), but 4815 * it's better than nothing. */ 4816 dnflow.f_id.dst_port = 0; 4817 dnflow.f_id.src_port = 0; 4818 dnflow.f_id.proto = 0; 4819 4820 dnflow.rule.info = r->dnpipe; 4821 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4822 if (r->dnflags & PFRULE_DN_IS_PIPE) 4823 dnflow.rule.info |= IPFW_IS_PIPE; 4824 4825 dnflow.f_id.extra = dnflow.rule.info; 4826 4827 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4828 dnflow.flags |= IPFW_ARGS_ETHER; 4829 dnflow.ifp = kif->pfik_ifp; 4830 4831 switch (af) { 4832 case AF_INET: 4833 dnflow.f_id.addr_type = 4; 4834 dnflow.f_id.src_ip = src->v4.s_addr; 4835 dnflow.f_id.dst_ip = dst->v4.s_addr; 4836 break; 4837 case AF_INET6: 4838 dnflow.flags |= IPFW_ARGS_IP6; 4839 dnflow.f_id.addr_type = 6; 4840 dnflow.f_id.src_ip6 = src->v6; 4841 dnflow.f_id.dst_ip6 = dst->v6; 4842 break; 4843 } 4844 4845 PF_RULES_RUNLOCK(); 4846 4847 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4848 ip_dn_io_ptr(m0, &dnflow); 4849 if (*m0 != NULL) 4850 pf_dummynet_flag_remove(m, mtag); 4851 } else { 4852 PF_RULES_RUNLOCK(); 4853 } 4854 4855 if (action == PF_PASS && bridge_to) { 4856 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4857 *m0 = NULL; /* We've eaten the packet. */ 4858 } 4859 4860 return (action); 4861 } 4862 4863 #define PF_TEST_ATTRIB(t, a)\ 4864 do { \ 4865 if (t) { \ 4866 r = a; \ 4867 goto nextrule; \ 4868 } \ 4869 } while (0) 4870 4871 static int 4872 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, 4873 struct pf_pdesc *pd, struct pf_krule **am, 4874 struct pf_kruleset **rsm, struct inpcb *inp) 4875 { 4876 struct pf_krule *nr = NULL; 4877 struct pf_addr * const saddr = pd->src; 4878 struct pf_addr * const daddr = pd->dst; 4879 struct pf_krule *r, *a = NULL; 4880 struct pf_kruleset *ruleset = NULL; 4881 struct pf_krule_slist match_rules; 4882 struct pf_krule_item *ri; 4883 struct pf_ksrc_node *nsn = NULL; 4884 struct tcphdr *th = &pd->hdr.tcp; 4885 struct pf_state_key *sk = NULL, *nk = NULL; 4886 u_short reason, transerror; 4887 int rewrite = 0; 4888 int tag = -1; 4889 int asd = 0; 4890 int match = 0; 4891 int state_icmp = 0, icmp_dir, multi; 4892 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4893 u_int16_t bproto_sum = 0, bip_sum = 0; 4894 u_int8_t icmptype = 0, icmpcode = 0; 4895 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4896 struct pf_udp_mapping *udp_mapping = NULL; 4897 4898 PF_RULES_RASSERT(); 4899 4900 SLIST_INIT(&match_rules); 4901 4902 if (inp != NULL) { 4903 INP_LOCK_ASSERT(inp); 4904 pd->lookup.uid = inp->inp_cred->cr_uid; 4905 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4906 pd->lookup.done = 1; 4907 } 4908 4909 switch (pd->virtual_proto) { 4910 case IPPROTO_TCP: 4911 sport = th->th_sport; 4912 dport = th->th_dport; 4913 break; 4914 case IPPROTO_UDP: 4915 sport = pd->hdr.udp.uh_sport; 4916 dport = pd->hdr.udp.uh_dport; 4917 break; 4918 case IPPROTO_SCTP: 4919 sport = pd->hdr.sctp.src_port; 4920 dport = pd->hdr.sctp.dest_port; 4921 break; 4922 #ifdef INET 4923 case IPPROTO_ICMP: 4924 MPASS(pd->af == AF_INET); 4925 icmptype = pd->hdr.icmp.icmp_type; 4926 icmpcode = pd->hdr.icmp.icmp_code; 4927 state_icmp = pf_icmp_mapping(pd, icmptype, 4928 &icmp_dir, &multi, &virtual_id, &virtual_type); 4929 if (icmp_dir == PF_IN) { 4930 sport = virtual_id; 4931 dport = virtual_type; 4932 } else { 4933 sport = virtual_type; 4934 dport = virtual_id; 4935 } 4936 break; 4937 #endif /* INET */ 4938 #ifdef INET6 4939 case IPPROTO_ICMPV6: 4940 MPASS(pd->af == AF_INET6); 4941 icmptype = pd->hdr.icmp6.icmp6_type; 4942 icmpcode = pd->hdr.icmp6.icmp6_code; 4943 state_icmp = pf_icmp_mapping(pd, icmptype, 4944 &icmp_dir, &multi, &virtual_id, &virtual_type); 4945 if (icmp_dir == PF_IN) { 4946 sport = virtual_id; 4947 dport = virtual_type; 4948 } else { 4949 sport = virtual_type; 4950 dport = virtual_id; 4951 } 4952 4953 break; 4954 #endif /* INET6 */ 4955 default: 4956 sport = dport = 0; 4957 break; 4958 } 4959 4960 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4961 4962 /* check packet for BINAT/NAT/RDR */ 4963 transerror = pf_get_translation(pd, pd->off, &nsn, &sk, 4964 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4965 switch (transerror) { 4966 default: 4967 /* A translation error occurred. */ 4968 REASON_SET(&reason, transerror); 4969 goto cleanup; 4970 case PFRES_MAX: 4971 /* No match. */ 4972 break; 4973 case PFRES_MATCH: 4974 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4975 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4976 4977 if (nr->log) { 4978 PFLOG_PACKET(PF_PASS, PFRES_MATCH, nr, a, 4979 ruleset, pd, 1); 4980 } 4981 4982 if (pd->ip_sum) 4983 bip_sum = *pd->ip_sum; 4984 4985 switch (pd->proto) { 4986 case IPPROTO_TCP: 4987 bproto_sum = th->th_sum; 4988 pd->proto_sum = &th->th_sum; 4989 4990 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 4991 nk->port[pd->sidx] != sport) { 4992 pf_change_ap(pd->m, saddr, &th->th_sport, pd->ip_sum, 4993 &th->th_sum, &nk->addr[pd->sidx], 4994 nk->port[pd->sidx], 0, pd->af); 4995 pd->sport = &th->th_sport; 4996 sport = th->th_sport; 4997 } 4998 4999 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5000 nk->port[pd->didx] != dport) { 5001 pf_change_ap(pd->m, daddr, &th->th_dport, pd->ip_sum, 5002 &th->th_sum, &nk->addr[pd->didx], 5003 nk->port[pd->didx], 0, pd->af); 5004 dport = th->th_dport; 5005 pd->dport = &th->th_dport; 5006 } 5007 rewrite++; 5008 break; 5009 case IPPROTO_UDP: 5010 bproto_sum = pd->hdr.udp.uh_sum; 5011 pd->proto_sum = &pd->hdr.udp.uh_sum; 5012 5013 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5014 nk->port[pd->sidx] != sport) { 5015 pf_change_ap(pd->m, saddr, &pd->hdr.udp.uh_sport, 5016 pd->ip_sum, &pd->hdr.udp.uh_sum, 5017 &nk->addr[pd->sidx], 5018 nk->port[pd->sidx], 1, pd->af); 5019 sport = pd->hdr.udp.uh_sport; 5020 pd->sport = &pd->hdr.udp.uh_sport; 5021 } 5022 5023 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5024 nk->port[pd->didx] != dport) { 5025 pf_change_ap(pd->m, daddr, &pd->hdr.udp.uh_dport, 5026 pd->ip_sum, &pd->hdr.udp.uh_sum, 5027 &nk->addr[pd->didx], 5028 nk->port[pd->didx], 1, pd->af); 5029 dport = pd->hdr.udp.uh_dport; 5030 pd->dport = &pd->hdr.udp.uh_dport; 5031 } 5032 rewrite++; 5033 break; 5034 case IPPROTO_SCTP: { 5035 uint16_t checksum = 0; 5036 5037 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], pd->af) || 5038 nk->port[pd->sidx] != sport) { 5039 pf_change_ap(pd->m, saddr, &pd->hdr.sctp.src_port, 5040 pd->ip_sum, &checksum, 5041 &nk->addr[pd->sidx], 5042 nk->port[pd->sidx], 1, pd->af); 5043 } 5044 if (PF_ANEQ(daddr, &nk->addr[pd->didx], pd->af) || 5045 nk->port[pd->didx] != dport) { 5046 pf_change_ap(pd->m, daddr, &pd->hdr.sctp.dest_port, 5047 pd->ip_sum, &checksum, 5048 &nk->addr[pd->didx], 5049 nk->port[pd->didx], 1, pd->af); 5050 } 5051 break; 5052 } 5053 #ifdef INET 5054 case IPPROTO_ICMP: 5055 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5056 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5057 nk->addr[pd->sidx].v4.s_addr, 0); 5058 5059 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5060 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5061 nk->addr[pd->didx].v4.s_addr, 0); 5062 5063 if (virtual_type == htons(ICMP_ECHO) && 5064 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5065 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5066 pd->hdr.icmp.icmp_cksum, sport, 5067 nk->port[pd->sidx], 0); 5068 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5069 pd->sport = &pd->hdr.icmp.icmp_id; 5070 } 5071 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5072 break; 5073 #endif /* INET */ 5074 #ifdef INET6 5075 case IPPROTO_ICMPV6: 5076 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5077 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5078 &nk->addr[pd->sidx], 0); 5079 5080 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5081 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5082 &nk->addr[pd->didx], 0); 5083 rewrite++; 5084 break; 5085 #endif /* INET */ 5086 default: 5087 switch (pd->af) { 5088 #ifdef INET 5089 case AF_INET: 5090 if (PF_ANEQ(saddr, 5091 &nk->addr[pd->sidx], AF_INET)) 5092 pf_change_a(&saddr->v4.s_addr, 5093 pd->ip_sum, 5094 nk->addr[pd->sidx].v4.s_addr, 0); 5095 5096 if (PF_ANEQ(daddr, 5097 &nk->addr[pd->didx], AF_INET)) 5098 pf_change_a(&daddr->v4.s_addr, 5099 pd->ip_sum, 5100 nk->addr[pd->didx].v4.s_addr, 0); 5101 break; 5102 #endif /* INET */ 5103 #ifdef INET6 5104 case AF_INET6: 5105 if (PF_ANEQ(saddr, 5106 &nk->addr[pd->sidx], AF_INET6)) 5107 PF_ACPY(saddr, &nk->addr[pd->sidx], pd->af); 5108 5109 if (PF_ANEQ(daddr, 5110 &nk->addr[pd->didx], AF_INET6)) 5111 PF_ACPY(daddr, &nk->addr[pd->didx], pd->af); 5112 break; 5113 #endif /* INET */ 5114 } 5115 break; 5116 } 5117 if (nr->natpass) 5118 r = NULL; 5119 } 5120 5121 while (r != NULL) { 5122 pf_counter_u64_add(&r->evaluations, 1); 5123 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot, 5124 r->skip[PF_SKIP_IFP]); 5125 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5126 r->skip[PF_SKIP_DIR]); 5127 PF_TEST_ATTRIB(r->af && r->af != pd->af, 5128 r->skip[PF_SKIP_AF]); 5129 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5130 r->skip[PF_SKIP_PROTO]); 5131 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, saddr, pd->af, 5132 r->src.neg, pd->kif, M_GETFIB(pd->m)), 5133 r->skip[PF_SKIP_SRC_ADDR]); 5134 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, daddr, pd->af, 5135 r->dst.neg, NULL, M_GETFIB(pd->m)), 5136 r->skip[PF_SKIP_DST_ADDR]); 5137 switch (pd->virtual_proto) { 5138 case PF_VPROTO_FRAGMENT: 5139 /* tcp/udp only. port_op always 0 in other cases */ 5140 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5141 TAILQ_NEXT(r, entries)); 5142 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5143 TAILQ_NEXT(r, entries)); 5144 /* icmp only. type/code always 0 in other cases */ 5145 PF_TEST_ATTRIB((r->type || r->code), 5146 TAILQ_NEXT(r, entries)); 5147 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5148 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5149 TAILQ_NEXT(r, entries)); 5150 break; 5151 5152 case IPPROTO_TCP: 5153 PF_TEST_ATTRIB((r->flagset & th->th_flags) != r->flags, 5154 TAILQ_NEXT(r, entries)); 5155 /* FALLTHROUGH */ 5156 case IPPROTO_SCTP: 5157 case IPPROTO_UDP: 5158 /* tcp/udp only. port_op always 0 in other cases */ 5159 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5160 r->src.port[0], r->src.port[1], sport), 5161 r->skip[PF_SKIP_SRC_PORT]); 5162 /* tcp/udp only. port_op always 0 in other cases */ 5163 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5164 r->dst.port[0], r->dst.port[1], dport), 5165 r->skip[PF_SKIP_DST_PORT]); 5166 /* tcp/udp only. uid.op always 0 in other cases */ 5167 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5168 pf_socket_lookup(pd), 1)) && 5169 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5170 pd->lookup.uid), 5171 TAILQ_NEXT(r, entries)); 5172 /* tcp/udp only. gid.op always 0 in other cases */ 5173 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5174 pf_socket_lookup(pd), 1)) && 5175 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5176 pd->lookup.gid), 5177 TAILQ_NEXT(r, entries)); 5178 break; 5179 5180 case IPPROTO_ICMP: 5181 case IPPROTO_ICMPV6: 5182 /* icmp only. type always 0 in other cases */ 5183 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5184 TAILQ_NEXT(r, entries)); 5185 /* icmp only. type always 0 in other cases */ 5186 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5187 TAILQ_NEXT(r, entries)); 5188 break; 5189 5190 default: 5191 break; 5192 } 5193 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5194 TAILQ_NEXT(r, entries)); 5195 PF_TEST_ATTRIB(r->prio && 5196 !pf_match_ieee8021q_pcp(r->prio, pd->m), 5197 TAILQ_NEXT(r, entries)); 5198 PF_TEST_ATTRIB(r->prob && 5199 r->prob <= arc4random(), 5200 TAILQ_NEXT(r, entries)); 5201 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r, &tag, 5202 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5203 TAILQ_NEXT(r, entries)); 5204 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(pd->m, r), 5205 TAILQ_NEXT(r, entries)); 5206 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5207 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5208 TAILQ_NEXT(r, entries)); 5209 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5210 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5211 pf_osfp_fingerprint(pd, th), 5212 r->os_fingerprint)), 5213 TAILQ_NEXT(r, entries)); 5214 /* FALLTHROUGH */ 5215 if (r->tag) 5216 tag = r->tag; 5217 if (r->anchor == NULL) { 5218 if (r->action == PF_MATCH) { 5219 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5220 if (ri == NULL) { 5221 REASON_SET(&reason, PFRES_MEMORY); 5222 goto cleanup; 5223 } 5224 ri->r = r; 5225 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5226 pf_counter_u64_critical_enter(); 5227 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5228 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5229 pf_counter_u64_critical_exit(); 5230 pf_rule_to_actions(r, &pd->act); 5231 if (r->log || pd->act.log & PF_LOG_MATCHES) 5232 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5233 a, ruleset, pd, 1); 5234 } else { 5235 match = 1; 5236 *rm = r; 5237 *am = a; 5238 *rsm = ruleset; 5239 if (pd->act.log & PF_LOG_MATCHES) 5240 PFLOG_PACKET(r->action, PFRES_MATCH, r, 5241 a, ruleset, pd, 1); 5242 } 5243 if ((*rm)->quick) 5244 break; 5245 r = TAILQ_NEXT(r, entries); 5246 } else 5247 pf_step_into_anchor(anchor_stack, &asd, 5248 &ruleset, PF_RULESET_FILTER, &r, &a, 5249 &match); 5250 nextrule: 5251 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5252 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5253 break; 5254 } 5255 r = *rm; 5256 a = *am; 5257 ruleset = *rsm; 5258 5259 REASON_SET(&reason, PFRES_MATCH); 5260 5261 /* apply actions for last matching pass/block rule */ 5262 pf_rule_to_actions(r, &pd->act); 5263 5264 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5265 if (rewrite) 5266 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5267 PFLOG_PACKET(r->action, reason, r, a, ruleset, pd, 1); 5268 } 5269 5270 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5271 (r->action == PF_DROP) && 5272 ((r->rule_flag & PFRULE_RETURNRST) || 5273 (r->rule_flag & PFRULE_RETURNICMP) || 5274 (r->rule_flag & PFRULE_RETURN))) { 5275 pf_return(r, nr, pd, sk, th, bproto_sum, 5276 bip_sum, &reason, r->rtableid); 5277 } 5278 5279 if (r->action == PF_DROP) 5280 goto cleanup; 5281 5282 if (tag > 0 && pf_tag_packet(pd, tag)) { 5283 REASON_SET(&reason, PFRES_MEMORY); 5284 goto cleanup; 5285 } 5286 if (pd->act.rtableid >= 0) 5287 M_SETFIB(pd->m, pd->act.rtableid); 5288 5289 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5290 (!state_icmp && (r->keep_state || nr != NULL || 5291 (pd->flags & PFDESC_TCP_NORM)))) { 5292 int action; 5293 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, 5294 sport, dport, &rewrite, sm, tag, bproto_sum, bip_sum, 5295 &match_rules, udp_mapping); 5296 if (action != PF_PASS) { 5297 pf_udp_mapping_release(udp_mapping); 5298 if (action == PF_DROP && 5299 (r->rule_flag & PFRULE_RETURN)) 5300 pf_return(r, nr, pd, sk, th, 5301 bproto_sum, bip_sum, &reason, 5302 pd->act.rtableid); 5303 return (action); 5304 } 5305 } else { 5306 while ((ri = SLIST_FIRST(&match_rules))) { 5307 SLIST_REMOVE_HEAD(&match_rules, entry); 5308 free(ri, M_PF_RULE_ITEM); 5309 } 5310 5311 uma_zfree(V_pf_state_key_z, sk); 5312 uma_zfree(V_pf_state_key_z, nk); 5313 pf_udp_mapping_release(udp_mapping); 5314 } 5315 5316 /* copy back packet headers if we performed NAT operations */ 5317 if (rewrite) 5318 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5319 5320 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5321 pd->dir == PF_OUT && 5322 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) 5323 /* 5324 * We want the state created, but we dont 5325 * want to send this in case a partner 5326 * firewall has to know about it to allow 5327 * replies through it. 5328 */ 5329 return (PF_DEFER); 5330 5331 return (PF_PASS); 5332 5333 cleanup: 5334 while ((ri = SLIST_FIRST(&match_rules))) { 5335 SLIST_REMOVE_HEAD(&match_rules, entry); 5336 free(ri, M_PF_RULE_ITEM); 5337 } 5338 5339 uma_zfree(V_pf_state_key_z, sk); 5340 uma_zfree(V_pf_state_key_z, nk); 5341 pf_udp_mapping_release(udp_mapping); 5342 5343 return (PF_DROP); 5344 } 5345 5346 static int 5347 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5348 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5349 struct pf_state_key *sk, u_int16_t sport, 5350 u_int16_t dport, int *rewrite, struct pf_kstate **sm, 5351 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, 5352 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5353 { 5354 struct pf_kstate *s = NULL; 5355 struct pf_ksrc_node *sn = NULL; 5356 struct tcphdr *th = &pd->hdr.tcp; 5357 u_int16_t mss = V_tcp_mssdflt; 5358 u_short reason, sn_reason; 5359 struct pf_krule_item *ri; 5360 5361 /* check maximums */ 5362 if (r->max_states && 5363 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5364 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5365 REASON_SET(&reason, PFRES_MAXSTATES); 5366 goto csfailed; 5367 } 5368 /* src node for filter rule */ 5369 if ((r->rule_flag & PFRULE_SRCTRACK || 5370 r->rpool.opts & PF_POOL_STICKYADDR) && 5371 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5372 REASON_SET(&reason, sn_reason); 5373 goto csfailed; 5374 } 5375 /* src node for translation rule */ 5376 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5377 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5378 pd->af)) != 0 ) { 5379 REASON_SET(&reason, sn_reason); 5380 goto csfailed; 5381 } 5382 s = pf_alloc_state(M_NOWAIT); 5383 if (s == NULL) { 5384 REASON_SET(&reason, PFRES_MEMORY); 5385 goto csfailed; 5386 } 5387 s->rule = r; 5388 s->nat_rule = nr; 5389 s->anchor = a; 5390 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5391 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5392 5393 STATE_INC_COUNTERS(s); 5394 if (r->allow_opts) 5395 s->state_flags |= PFSTATE_ALLOWOPTS; 5396 if (r->rule_flag & PFRULE_STATESLOPPY) 5397 s->state_flags |= PFSTATE_SLOPPY; 5398 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5399 s->state_flags |= PFSTATE_SCRUB_TCP; 5400 if ((r->rule_flag & PFRULE_PFLOW) || 5401 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5402 s->state_flags |= PFSTATE_PFLOW; 5403 5404 s->act.log = pd->act.log & PF_LOG_ALL; 5405 s->sync_state = PFSYNC_S_NONE; 5406 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5407 5408 if (nr != NULL) 5409 s->act.log |= nr->log & PF_LOG_ALL; 5410 switch (pd->proto) { 5411 case IPPROTO_TCP: 5412 s->src.seqlo = ntohl(th->th_seq); 5413 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5414 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5415 r->keep_state == PF_STATE_MODULATE) { 5416 /* Generate sequence number modulator */ 5417 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5418 0) 5419 s->src.seqdiff = 1; 5420 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, 5421 htonl(s->src.seqlo + s->src.seqdiff), 0); 5422 *rewrite = 1; 5423 } else 5424 s->src.seqdiff = 0; 5425 if (th->th_flags & TH_SYN) { 5426 s->src.seqhi++; 5427 s->src.wscale = pf_get_wscale(pd); 5428 } 5429 s->src.max_win = MAX(ntohs(th->th_win), 1); 5430 if (s->src.wscale & PF_WSCALE_MASK) { 5431 /* Remove scale factor from initial window */ 5432 int win = s->src.max_win; 5433 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5434 s->src.max_win = (win - 1) >> 5435 (s->src.wscale & PF_WSCALE_MASK); 5436 } 5437 if (th->th_flags & TH_FIN) 5438 s->src.seqhi++; 5439 s->dst.seqhi = 1; 5440 s->dst.max_win = 1; 5441 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5442 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5443 s->timeout = PFTM_TCP_FIRST_PACKET; 5444 atomic_add_32(&V_pf_status.states_halfopen, 1); 5445 break; 5446 case IPPROTO_UDP: 5447 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5448 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5449 s->timeout = PFTM_UDP_FIRST_PACKET; 5450 break; 5451 case IPPROTO_SCTP: 5452 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5453 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5454 s->timeout = PFTM_SCTP_FIRST_PACKET; 5455 break; 5456 case IPPROTO_ICMP: 5457 #ifdef INET6 5458 case IPPROTO_ICMPV6: 5459 #endif 5460 s->timeout = PFTM_ICMP_FIRST_PACKET; 5461 break; 5462 default: 5463 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5464 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5465 s->timeout = PFTM_OTHER_FIRST_PACKET; 5466 } 5467 5468 if (r->rt) { 5469 /* pf_map_addr increases the reason counters */ 5470 if ((reason = pf_map_addr_sn(pd->af, r, pd->src, &s->rt_addr, 5471 &s->rt_kif, NULL, &sn)) != 0) 5472 goto csfailed; 5473 s->rt = r->rt; 5474 } 5475 5476 s->creation = s->expire = pf_get_uptime(); 5477 5478 if (sn != NULL) 5479 s->src_node = sn; 5480 if (nsn != NULL) { 5481 /* XXX We only modify one side for now. */ 5482 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5483 s->nat_src_node = nsn; 5484 } 5485 if (pd->proto == IPPROTO_TCP) { 5486 if (s->state_flags & PFSTATE_SCRUB_TCP && 5487 pf_normalize_tcp_init(pd, th, &s->src, &s->dst)) { 5488 REASON_SET(&reason, PFRES_MEMORY); 5489 goto csfailed; 5490 } 5491 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5492 pf_normalize_tcp_stateful(pd, &reason, th, s, 5493 &s->src, &s->dst, rewrite)) { 5494 /* This really shouldn't happen!!! */ 5495 DPFPRINTF(PF_DEBUG_URGENT, 5496 ("pf_normalize_tcp_stateful failed on first " 5497 "pkt\n")); 5498 goto csfailed; 5499 } 5500 } else if (pd->proto == IPPROTO_SCTP) { 5501 if (pf_normalize_sctp_init(pd, &s->src, &s->dst)) 5502 goto csfailed; 5503 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5504 goto csfailed; 5505 } 5506 s->direction = pd->dir; 5507 5508 /* 5509 * sk/nk could already been setup by pf_get_translation(). 5510 */ 5511 if (nr == NULL) { 5512 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5513 __func__, nr, sk, nk)); 5514 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 5515 if (sk == NULL) 5516 goto csfailed; 5517 nk = sk; 5518 } else 5519 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5520 __func__, nr, sk, nk)); 5521 5522 /* Swap sk/nk for PF_OUT. */ 5523 if (pf_state_insert(BOUND_IFACE(s, pd->kif), pd->kif, 5524 (pd->dir == PF_IN) ? sk : nk, 5525 (pd->dir == PF_IN) ? nk : sk, s)) { 5526 REASON_SET(&reason, PFRES_STATEINS); 5527 goto drop; 5528 } else 5529 *sm = s; 5530 5531 if (tag > 0) 5532 s->tag = tag; 5533 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5534 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5535 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5536 /* undo NAT changes, if they have taken place */ 5537 if (nr != NULL) { 5538 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5539 if (pd->dir == PF_OUT) 5540 skt = s->key[PF_SK_STACK]; 5541 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5542 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5543 if (pd->sport) 5544 *pd->sport = skt->port[pd->sidx]; 5545 if (pd->dport) 5546 *pd->dport = skt->port[pd->didx]; 5547 if (pd->proto_sum) 5548 *pd->proto_sum = bproto_sum; 5549 if (pd->ip_sum) 5550 *pd->ip_sum = bip_sum; 5551 m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any); 5552 } 5553 s->src.seqhi = htonl(arc4random()); 5554 /* Find mss option */ 5555 int rtid = M_GETFIB(pd->m); 5556 mss = pf_get_mss(pd); 5557 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5558 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5559 s->src.mss = mss; 5560 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5561 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5562 TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0, 5563 pd->act.rtableid); 5564 REASON_SET(&reason, PFRES_SYNPROXY); 5565 return (PF_SYNPROXY_DROP); 5566 } 5567 5568 s->udp_mapping = udp_mapping; 5569 5570 return (PF_PASS); 5571 5572 csfailed: 5573 while ((ri = SLIST_FIRST(match_rules))) { 5574 SLIST_REMOVE_HEAD(match_rules, entry); 5575 free(ri, M_PF_RULE_ITEM); 5576 } 5577 5578 uma_zfree(V_pf_state_key_z, sk); 5579 uma_zfree(V_pf_state_key_z, nk); 5580 5581 if (sn != NULL) { 5582 PF_SRC_NODE_LOCK(sn); 5583 if (--sn->states == 0 && sn->expire == 0) { 5584 pf_unlink_src_node(sn); 5585 uma_zfree(V_pf_sources_z, sn); 5586 counter_u64_add( 5587 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5588 } 5589 PF_SRC_NODE_UNLOCK(sn); 5590 } 5591 5592 if (nsn != sn && nsn != NULL) { 5593 PF_SRC_NODE_LOCK(nsn); 5594 if (--nsn->states == 0 && nsn->expire == 0) { 5595 pf_unlink_src_node(nsn); 5596 uma_zfree(V_pf_sources_z, nsn); 5597 counter_u64_add( 5598 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5599 } 5600 PF_SRC_NODE_UNLOCK(nsn); 5601 } 5602 5603 drop: 5604 if (s != NULL) { 5605 pf_src_tree_remove_state(s); 5606 s->timeout = PFTM_UNLINKED; 5607 STATE_DEC_COUNTERS(s); 5608 pf_free_state(s); 5609 } 5610 5611 return (PF_DROP); 5612 } 5613 5614 static int 5615 pf_tcp_track_full(struct pf_kstate **state, struct pf_pdesc *pd, 5616 u_short *reason, int *copyback) 5617 { 5618 struct tcphdr *th = &pd->hdr.tcp; 5619 struct pf_state_peer *src, *dst; 5620 u_int16_t win = ntohs(th->th_win); 5621 u_int32_t ack, end, data_end, seq, orig_seq; 5622 u_int8_t sws, dws, psrc, pdst; 5623 int ackskew; 5624 5625 if (pd->dir == (*state)->direction) { 5626 src = &(*state)->src; 5627 dst = &(*state)->dst; 5628 psrc = PF_PEER_SRC; 5629 pdst = PF_PEER_DST; 5630 } else { 5631 src = &(*state)->dst; 5632 dst = &(*state)->src; 5633 psrc = PF_PEER_DST; 5634 pdst = PF_PEER_SRC; 5635 } 5636 5637 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5638 sws = src->wscale & PF_WSCALE_MASK; 5639 dws = dst->wscale & PF_WSCALE_MASK; 5640 } else 5641 sws = dws = 0; 5642 5643 /* 5644 * Sequence tracking algorithm from Guido van Rooij's paper: 5645 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5646 * tcp_filtering.ps 5647 */ 5648 5649 orig_seq = seq = ntohl(th->th_seq); 5650 if (src->seqlo == 0) { 5651 /* First packet from this end. Set its state */ 5652 5653 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5654 src->scrub == NULL) { 5655 if (pf_normalize_tcp_init(pd, th, src, dst)) { 5656 REASON_SET(reason, PFRES_MEMORY); 5657 return (PF_DROP); 5658 } 5659 } 5660 5661 /* Deferred generation of sequence number modulator */ 5662 if (dst->seqdiff && !src->seqdiff) { 5663 /* use random iss for the TCP server */ 5664 while ((src->seqdiff = arc4random() - seq) == 0) 5665 ; 5666 ack = ntohl(th->th_ack) - dst->seqdiff; 5667 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5668 src->seqdiff), 0); 5669 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5670 *copyback = 1; 5671 } else { 5672 ack = ntohl(th->th_ack); 5673 } 5674 5675 end = seq + pd->p_len; 5676 if (th->th_flags & TH_SYN) { 5677 end++; 5678 if (dst->wscale & PF_WSCALE_FLAG) { 5679 src->wscale = pf_get_wscale(pd); 5680 if (src->wscale & PF_WSCALE_FLAG) { 5681 /* Remove scale factor from initial 5682 * window */ 5683 sws = src->wscale & PF_WSCALE_MASK; 5684 win = ((u_int32_t)win + (1 << sws) - 1) 5685 >> sws; 5686 dws = dst->wscale & PF_WSCALE_MASK; 5687 } else { 5688 /* fixup other window */ 5689 dst->max_win = MIN(TCP_MAXWIN, 5690 (u_int32_t)dst->max_win << 5691 (dst->wscale & PF_WSCALE_MASK)); 5692 /* in case of a retrans SYN|ACK */ 5693 dst->wscale = 0; 5694 } 5695 } 5696 } 5697 data_end = end; 5698 if (th->th_flags & TH_FIN) 5699 end++; 5700 5701 src->seqlo = seq; 5702 if (src->state < TCPS_SYN_SENT) 5703 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5704 5705 /* 5706 * May need to slide the window (seqhi may have been set by 5707 * the crappy stack check or if we picked up the connection 5708 * after establishment) 5709 */ 5710 if (src->seqhi == 1 || 5711 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5712 src->seqhi = end + MAX(1, dst->max_win << dws); 5713 if (win > src->max_win) 5714 src->max_win = win; 5715 5716 } else { 5717 ack = ntohl(th->th_ack) - dst->seqdiff; 5718 if (src->seqdiff) { 5719 /* Modulate sequence numbers */ 5720 pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq + 5721 src->seqdiff), 0); 5722 pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5723 *copyback = 1; 5724 } 5725 end = seq + pd->p_len; 5726 if (th->th_flags & TH_SYN) 5727 end++; 5728 data_end = end; 5729 if (th->th_flags & TH_FIN) 5730 end++; 5731 } 5732 5733 if ((th->th_flags & TH_ACK) == 0) { 5734 /* Let it pass through the ack skew check */ 5735 ack = dst->seqlo; 5736 } else if ((ack == 0 && 5737 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5738 /* broken tcp stacks do not set ack */ 5739 (dst->state < TCPS_SYN_SENT)) { 5740 /* 5741 * Many stacks (ours included) will set the ACK number in an 5742 * FIN|ACK if the SYN times out -- no sequence to ACK. 5743 */ 5744 ack = dst->seqlo; 5745 } 5746 5747 if (seq == end) { 5748 /* Ease sequencing restrictions on no data packets */ 5749 seq = src->seqlo; 5750 data_end = end = seq; 5751 } 5752 5753 ackskew = dst->seqlo - ack; 5754 5755 /* 5756 * Need to demodulate the sequence numbers in any TCP SACK options 5757 * (Selective ACK). We could optionally validate the SACK values 5758 * against the current ACK window, either forwards or backwards, but 5759 * I'm not confident that SACK has been implemented properly 5760 * everywhere. It wouldn't surprise me if several stacks accidentally 5761 * SACK too far backwards of previously ACKed data. There really aren't 5762 * any security implications of bad SACKing unless the target stack 5763 * doesn't validate the option length correctly. Someone trying to 5764 * spoof into a TCP connection won't bother blindly sending SACK 5765 * options anyway. 5766 */ 5767 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5768 if (pf_modulate_sack(pd, th, dst)) 5769 *copyback = 1; 5770 } 5771 5772 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5773 if (SEQ_GEQ(src->seqhi, data_end) && 5774 /* Last octet inside other's window space */ 5775 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5776 /* Retrans: not more than one window back */ 5777 (ackskew >= -MAXACKWINDOW) && 5778 /* Acking not more than one reassembled fragment backwards */ 5779 (ackskew <= (MAXACKWINDOW << sws)) && 5780 /* Acking not more than one window forward */ 5781 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5782 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5783 /* Require an exact/+1 sequence match on resets when possible */ 5784 5785 if (dst->scrub || src->scrub) { 5786 if (pf_normalize_tcp_stateful(pd, reason, th, 5787 *state, src, dst, copyback)) 5788 return (PF_DROP); 5789 } 5790 5791 /* update max window */ 5792 if (src->max_win < win) 5793 src->max_win = win; 5794 /* synchronize sequencing */ 5795 if (SEQ_GT(end, src->seqlo)) 5796 src->seqlo = end; 5797 /* slide the window of what the other end can send */ 5798 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5799 dst->seqhi = ack + MAX((win << sws), 1); 5800 5801 /* update states */ 5802 if (th->th_flags & TH_SYN) 5803 if (src->state < TCPS_SYN_SENT) 5804 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5805 if (th->th_flags & TH_FIN) 5806 if (src->state < TCPS_CLOSING) 5807 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5808 if (th->th_flags & TH_ACK) { 5809 if (dst->state == TCPS_SYN_SENT) { 5810 pf_set_protostate(*state, pdst, 5811 TCPS_ESTABLISHED); 5812 if (src->state == TCPS_ESTABLISHED && 5813 (*state)->src_node != NULL && 5814 pf_src_connlimit(state)) { 5815 REASON_SET(reason, PFRES_SRCLIMIT); 5816 return (PF_DROP); 5817 } 5818 } else if (dst->state == TCPS_CLOSING) 5819 pf_set_protostate(*state, pdst, 5820 TCPS_FIN_WAIT_2); 5821 } 5822 if (th->th_flags & TH_RST) 5823 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5824 5825 /* update expire time */ 5826 (*state)->expire = pf_get_uptime(); 5827 if (src->state >= TCPS_FIN_WAIT_2 && 5828 dst->state >= TCPS_FIN_WAIT_2) 5829 (*state)->timeout = PFTM_TCP_CLOSED; 5830 else if (src->state >= TCPS_CLOSING && 5831 dst->state >= TCPS_CLOSING) 5832 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5833 else if (src->state < TCPS_ESTABLISHED || 5834 dst->state < TCPS_ESTABLISHED) 5835 (*state)->timeout = PFTM_TCP_OPENING; 5836 else if (src->state >= TCPS_CLOSING || 5837 dst->state >= TCPS_CLOSING) 5838 (*state)->timeout = PFTM_TCP_CLOSING; 5839 else 5840 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5841 5842 /* Fall through to PASS packet */ 5843 5844 } else if ((dst->state < TCPS_SYN_SENT || 5845 dst->state >= TCPS_FIN_WAIT_2 || 5846 src->state >= TCPS_FIN_WAIT_2) && 5847 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5848 /* Within a window forward of the originating packet */ 5849 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5850 /* Within a window backward of the originating packet */ 5851 5852 /* 5853 * This currently handles three situations: 5854 * 1) Stupid stacks will shotgun SYNs before their peer 5855 * replies. 5856 * 2) When PF catches an already established stream (the 5857 * firewall rebooted, the state table was flushed, routes 5858 * changed...) 5859 * 3) Packets get funky immediately after the connection 5860 * closes (this should catch Solaris spurious ACK|FINs 5861 * that web servers like to spew after a close) 5862 * 5863 * This must be a little more careful than the above code 5864 * since packet floods will also be caught here. We don't 5865 * update the TTL here to mitigate the damage of a packet 5866 * flood and so the same code can handle awkward establishment 5867 * and a loosened connection close. 5868 * In the establishment case, a correct peer response will 5869 * validate the connection, go through the normal state code 5870 * and keep updating the state TTL. 5871 */ 5872 5873 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5874 printf("pf: loose state match: "); 5875 pf_print_state(*state); 5876 pf_print_flags(th->th_flags); 5877 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5878 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5879 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5880 (unsigned long long)(*state)->packets[1], 5881 pd->dir == PF_IN ? "in" : "out", 5882 pd->dir == (*state)->direction ? "fwd" : "rev"); 5883 } 5884 5885 if (dst->scrub || src->scrub) { 5886 if (pf_normalize_tcp_stateful(pd, reason, th, 5887 *state, src, dst, copyback)) 5888 return (PF_DROP); 5889 } 5890 5891 /* update max window */ 5892 if (src->max_win < win) 5893 src->max_win = win; 5894 /* synchronize sequencing */ 5895 if (SEQ_GT(end, src->seqlo)) 5896 src->seqlo = end; 5897 /* slide the window of what the other end can send */ 5898 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5899 dst->seqhi = ack + MAX((win << sws), 1); 5900 5901 /* 5902 * Cannot set dst->seqhi here since this could be a shotgunned 5903 * SYN and not an already established connection. 5904 */ 5905 5906 if (th->th_flags & TH_FIN) 5907 if (src->state < TCPS_CLOSING) 5908 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5909 if (th->th_flags & TH_RST) 5910 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5911 5912 /* Fall through to PASS packet */ 5913 5914 } else { 5915 if ((*state)->dst.state == TCPS_SYN_SENT && 5916 (*state)->src.state == TCPS_SYN_SENT) { 5917 /* Send RST for state mismatches during handshake */ 5918 if (!(th->th_flags & TH_RST)) 5919 pf_send_tcp((*state)->rule, pd->af, 5920 pd->dst, pd->src, th->th_dport, 5921 th->th_sport, ntohl(th->th_ack), 0, 5922 TH_RST, 0, 0, 5923 (*state)->rule->return_ttl, M_SKIP_FIREWALL, 5924 0, 0, (*state)->act.rtableid); 5925 src->seqlo = 0; 5926 src->seqhi = 1; 5927 src->max_win = 1; 5928 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5929 printf("pf: BAD state: "); 5930 pf_print_state(*state); 5931 pf_print_flags(th->th_flags); 5932 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5933 "pkts=%llu:%llu dir=%s,%s\n", 5934 seq, orig_seq, ack, pd->p_len, ackskew, 5935 (unsigned long long)(*state)->packets[0], 5936 (unsigned long long)(*state)->packets[1], 5937 pd->dir == PF_IN ? "in" : "out", 5938 pd->dir == (*state)->direction ? "fwd" : "rev"); 5939 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5940 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5941 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5942 ' ': '2', 5943 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5944 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5945 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5946 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5947 } 5948 REASON_SET(reason, PFRES_BADSTATE); 5949 return (PF_DROP); 5950 } 5951 5952 return (PF_PASS); 5953 } 5954 5955 static int 5956 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5957 { 5958 struct tcphdr *th = &pd->hdr.tcp; 5959 struct pf_state_peer *src, *dst; 5960 u_int8_t psrc, pdst; 5961 5962 if (pd->dir == (*state)->direction) { 5963 src = &(*state)->src; 5964 dst = &(*state)->dst; 5965 psrc = PF_PEER_SRC; 5966 pdst = PF_PEER_DST; 5967 } else { 5968 src = &(*state)->dst; 5969 dst = &(*state)->src; 5970 psrc = PF_PEER_DST; 5971 pdst = PF_PEER_SRC; 5972 } 5973 5974 if (th->th_flags & TH_SYN) 5975 if (src->state < TCPS_SYN_SENT) 5976 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5977 if (th->th_flags & TH_FIN) 5978 if (src->state < TCPS_CLOSING) 5979 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5980 if (th->th_flags & TH_ACK) { 5981 if (dst->state == TCPS_SYN_SENT) { 5982 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5983 if (src->state == TCPS_ESTABLISHED && 5984 (*state)->src_node != NULL && 5985 pf_src_connlimit(state)) { 5986 REASON_SET(reason, PFRES_SRCLIMIT); 5987 return (PF_DROP); 5988 } 5989 } else if (dst->state == TCPS_CLOSING) { 5990 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5991 } else if (src->state == TCPS_SYN_SENT && 5992 dst->state < TCPS_SYN_SENT) { 5993 /* 5994 * Handle a special sloppy case where we only see one 5995 * half of the connection. If there is a ACK after 5996 * the initial SYN without ever seeing a packet from 5997 * the destination, set the connection to established. 5998 */ 5999 pf_set_protostate(*state, PF_PEER_BOTH, 6000 TCPS_ESTABLISHED); 6001 dst->state = src->state = TCPS_ESTABLISHED; 6002 if ((*state)->src_node != NULL && 6003 pf_src_connlimit(state)) { 6004 REASON_SET(reason, PFRES_SRCLIMIT); 6005 return (PF_DROP); 6006 } 6007 } else if (src->state == TCPS_CLOSING && 6008 dst->state == TCPS_ESTABLISHED && 6009 dst->seqlo == 0) { 6010 /* 6011 * Handle the closing of half connections where we 6012 * don't see the full bidirectional FIN/ACK+ACK 6013 * handshake. 6014 */ 6015 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6016 } 6017 } 6018 if (th->th_flags & TH_RST) 6019 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6020 6021 /* update expire time */ 6022 (*state)->expire = pf_get_uptime(); 6023 if (src->state >= TCPS_FIN_WAIT_2 && 6024 dst->state >= TCPS_FIN_WAIT_2) 6025 (*state)->timeout = PFTM_TCP_CLOSED; 6026 else if (src->state >= TCPS_CLOSING && 6027 dst->state >= TCPS_CLOSING) 6028 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6029 else if (src->state < TCPS_ESTABLISHED || 6030 dst->state < TCPS_ESTABLISHED) 6031 (*state)->timeout = PFTM_TCP_OPENING; 6032 else if (src->state >= TCPS_CLOSING || 6033 dst->state >= TCPS_CLOSING) 6034 (*state)->timeout = PFTM_TCP_CLOSING; 6035 else 6036 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6037 6038 return (PF_PASS); 6039 } 6040 6041 static int 6042 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6043 { 6044 struct pf_state_key *sk = (*state)->key[pd->didx]; 6045 struct tcphdr *th = &pd->hdr.tcp; 6046 6047 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6048 if (pd->dir != (*state)->direction) { 6049 REASON_SET(reason, PFRES_SYNPROXY); 6050 return (PF_SYNPROXY_DROP); 6051 } 6052 if (th->th_flags & TH_SYN) { 6053 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6054 REASON_SET(reason, PFRES_SYNPROXY); 6055 return (PF_DROP); 6056 } 6057 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6058 pd->src, th->th_dport, th->th_sport, 6059 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6060 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 6061 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6062 REASON_SET(reason, PFRES_SYNPROXY); 6063 return (PF_SYNPROXY_DROP); 6064 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6065 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6066 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6067 REASON_SET(reason, PFRES_SYNPROXY); 6068 return (PF_DROP); 6069 } else if ((*state)->src_node != NULL && 6070 pf_src_connlimit(state)) { 6071 REASON_SET(reason, PFRES_SRCLIMIT); 6072 return (PF_DROP); 6073 } else 6074 pf_set_protostate(*state, PF_PEER_SRC, 6075 PF_TCPS_PROXY_DST); 6076 } 6077 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6078 if (pd->dir == (*state)->direction) { 6079 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6080 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6081 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6082 REASON_SET(reason, PFRES_SYNPROXY); 6083 return (PF_DROP); 6084 } 6085 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6086 if ((*state)->dst.seqhi == 1) 6087 (*state)->dst.seqhi = htonl(arc4random()); 6088 pf_send_tcp((*state)->rule, pd->af, 6089 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6090 sk->port[pd->sidx], sk->port[pd->didx], 6091 (*state)->dst.seqhi, 0, TH_SYN, 0, 6092 (*state)->src.mss, 0, 6093 (*state)->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0, 6094 (*state)->tag, 0, (*state)->act.rtableid); 6095 REASON_SET(reason, PFRES_SYNPROXY); 6096 return (PF_SYNPROXY_DROP); 6097 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6098 (TH_SYN|TH_ACK)) || 6099 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6100 REASON_SET(reason, PFRES_SYNPROXY); 6101 return (PF_DROP); 6102 } else { 6103 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6104 (*state)->dst.seqlo = ntohl(th->th_seq); 6105 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6106 pd->src, th->th_dport, th->th_sport, 6107 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6108 TH_ACK, (*state)->src.max_win, 0, 0, 0, 6109 (*state)->tag, 0, (*state)->act.rtableid); 6110 pf_send_tcp((*state)->rule, pd->af, 6111 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6112 sk->port[pd->sidx], sk->port[pd->didx], 6113 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6114 TH_ACK, (*state)->dst.max_win, 0, 0, 6115 M_SKIP_FIREWALL, 0, 0, (*state)->act.rtableid); 6116 (*state)->src.seqdiff = (*state)->dst.seqhi - 6117 (*state)->src.seqlo; 6118 (*state)->dst.seqdiff = (*state)->src.seqhi - 6119 (*state)->dst.seqlo; 6120 (*state)->src.seqhi = (*state)->src.seqlo + 6121 (*state)->dst.max_win; 6122 (*state)->dst.seqhi = (*state)->dst.seqlo + 6123 (*state)->src.max_win; 6124 (*state)->src.wscale = (*state)->dst.wscale = 0; 6125 pf_set_protostate(*state, PF_PEER_BOTH, 6126 TCPS_ESTABLISHED); 6127 REASON_SET(reason, PFRES_SYNPROXY); 6128 return (PF_SYNPROXY_DROP); 6129 } 6130 } 6131 6132 return (PF_PASS); 6133 } 6134 6135 static int 6136 pf_test_state_tcp(struct pf_kstate **state, struct pf_pdesc *pd, 6137 u_short *reason) 6138 { 6139 struct pf_state_key_cmp key; 6140 struct tcphdr *th = &pd->hdr.tcp; 6141 int copyback = 0; 6142 int action; 6143 struct pf_state_peer *src, *dst; 6144 6145 bzero(&key, sizeof(key)); 6146 key.af = pd->af; 6147 key.proto = IPPROTO_TCP; 6148 if (pd->dir == PF_IN) { /* wire side, straight */ 6149 PF_ACPY(&key.addr[0], pd->src, key.af); 6150 PF_ACPY(&key.addr[1], pd->dst, key.af); 6151 key.port[0] = th->th_sport; 6152 key.port[1] = th->th_dport; 6153 } else { /* stack side, reverse */ 6154 PF_ACPY(&key.addr[1], pd->src, key.af); 6155 PF_ACPY(&key.addr[0], pd->dst, key.af); 6156 key.port[1] = th->th_sport; 6157 key.port[0] = th->th_dport; 6158 } 6159 6160 STATE_LOOKUP(&key, *state, pd); 6161 6162 if (pd->dir == (*state)->direction) { 6163 src = &(*state)->src; 6164 dst = &(*state)->dst; 6165 } else { 6166 src = &(*state)->dst; 6167 dst = &(*state)->src; 6168 } 6169 6170 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6171 return (action); 6172 6173 if (dst->state >= TCPS_FIN_WAIT_2 && 6174 src->state >= TCPS_FIN_WAIT_2 && 6175 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6176 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6177 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6178 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6179 printf("pf: state reuse "); 6180 pf_print_state(*state); 6181 pf_print_flags(th->th_flags); 6182 printf("\n"); 6183 } 6184 /* XXX make sure it's the same direction ?? */ 6185 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6186 pf_unlink_state(*state); 6187 *state = NULL; 6188 return (PF_DROP); 6189 } 6190 6191 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6192 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6193 return (PF_DROP); 6194 } else { 6195 if (pf_tcp_track_full(state, pd, reason, 6196 ©back) == PF_DROP) 6197 return (PF_DROP); 6198 } 6199 6200 /* translate source/destination address, if necessary */ 6201 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6202 struct pf_state_key *nk = (*state)->key[pd->didx]; 6203 6204 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6205 nk->port[pd->sidx] != th->th_sport) 6206 pf_change_ap(pd->m, pd->src, &th->th_sport, 6207 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6208 nk->port[pd->sidx], 0, pd->af); 6209 6210 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6211 nk->port[pd->didx] != th->th_dport) 6212 pf_change_ap(pd->m, pd->dst, &th->th_dport, 6213 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6214 nk->port[pd->didx], 0, pd->af); 6215 copyback = 1; 6216 } 6217 6218 /* Copyback sequence modulation or stateful scrub changes if needed */ 6219 if (copyback) 6220 m_copyback(pd->m, pd->off, sizeof(*th), (caddr_t)th); 6221 6222 return (PF_PASS); 6223 } 6224 6225 static int 6226 pf_test_state_udp(struct pf_kstate **state, struct pf_pdesc *pd) 6227 { 6228 struct pf_state_peer *src, *dst; 6229 struct pf_state_key_cmp key; 6230 struct udphdr *uh = &pd->hdr.udp; 6231 uint8_t psrc, pdst; 6232 6233 bzero(&key, sizeof(key)); 6234 key.af = pd->af; 6235 key.proto = IPPROTO_UDP; 6236 if (pd->dir == PF_IN) { /* wire side, straight */ 6237 PF_ACPY(&key.addr[0], pd->src, key.af); 6238 PF_ACPY(&key.addr[1], pd->dst, key.af); 6239 key.port[0] = uh->uh_sport; 6240 key.port[1] = uh->uh_dport; 6241 } else { /* stack side, reverse */ 6242 PF_ACPY(&key.addr[1], pd->src, key.af); 6243 PF_ACPY(&key.addr[0], pd->dst, key.af); 6244 key.port[1] = uh->uh_sport; 6245 key.port[0] = uh->uh_dport; 6246 } 6247 6248 STATE_LOOKUP(&key, *state, pd); 6249 6250 if (pd->dir == (*state)->direction) { 6251 src = &(*state)->src; 6252 dst = &(*state)->dst; 6253 psrc = PF_PEER_SRC; 6254 pdst = PF_PEER_DST; 6255 } else { 6256 src = &(*state)->dst; 6257 dst = &(*state)->src; 6258 psrc = PF_PEER_DST; 6259 pdst = PF_PEER_SRC; 6260 } 6261 6262 /* update states */ 6263 if (src->state < PFUDPS_SINGLE) 6264 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6265 if (dst->state == PFUDPS_SINGLE) 6266 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6267 6268 /* update expire time */ 6269 (*state)->expire = pf_get_uptime(); 6270 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6271 (*state)->timeout = PFTM_UDP_MULTIPLE; 6272 else 6273 (*state)->timeout = PFTM_UDP_SINGLE; 6274 6275 /* translate source/destination address, if necessary */ 6276 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6277 struct pf_state_key *nk = (*state)->key[pd->didx]; 6278 6279 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6280 nk->port[pd->sidx] != uh->uh_sport) 6281 pf_change_ap(pd->m, pd->src, &uh->uh_sport, pd->ip_sum, 6282 &uh->uh_sum, &nk->addr[pd->sidx], 6283 nk->port[pd->sidx], 1, pd->af); 6284 6285 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6286 nk->port[pd->didx] != uh->uh_dport) 6287 pf_change_ap(pd->m, pd->dst, &uh->uh_dport, pd->ip_sum, 6288 &uh->uh_sum, &nk->addr[pd->didx], 6289 nk->port[pd->didx], 1, pd->af); 6290 m_copyback(pd->m, pd->off, sizeof(*uh), (caddr_t)uh); 6291 } 6292 6293 return (PF_PASS); 6294 } 6295 6296 static int 6297 pf_test_state_sctp(struct pf_kstate **state, struct pf_pdesc *pd, 6298 u_short *reason) 6299 { 6300 struct pf_state_key_cmp key; 6301 struct pf_state_peer *src, *dst; 6302 struct sctphdr *sh = &pd->hdr.sctp; 6303 u_int8_t psrc; //, pdst; 6304 6305 bzero(&key, sizeof(key)); 6306 key.af = pd->af; 6307 key.proto = IPPROTO_SCTP; 6308 if (pd->dir == PF_IN) { /* wire side, straight */ 6309 PF_ACPY(&key.addr[0], pd->src, key.af); 6310 PF_ACPY(&key.addr[1], pd->dst, key.af); 6311 key.port[0] = sh->src_port; 6312 key.port[1] = sh->dest_port; 6313 } else { /* stack side, reverse */ 6314 PF_ACPY(&key.addr[1], pd->src, key.af); 6315 PF_ACPY(&key.addr[0], pd->dst, key.af); 6316 key.port[1] = sh->src_port; 6317 key.port[0] = sh->dest_port; 6318 } 6319 6320 STATE_LOOKUP(&key, *state, pd); 6321 6322 if (pd->dir == (*state)->direction) { 6323 src = &(*state)->src; 6324 dst = &(*state)->dst; 6325 psrc = PF_PEER_SRC; 6326 } else { 6327 src = &(*state)->dst; 6328 dst = &(*state)->src; 6329 psrc = PF_PEER_DST; 6330 } 6331 6332 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6333 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6334 pd->sctp_flags & PFDESC_SCTP_INIT) { 6335 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6336 pf_unlink_state(*state); 6337 *state = NULL; 6338 return (PF_DROP); 6339 } 6340 6341 /* Track state. */ 6342 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6343 if (src->state < SCTP_COOKIE_WAIT) { 6344 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6345 (*state)->timeout = PFTM_SCTP_OPENING; 6346 } 6347 } 6348 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6349 MPASS(dst->scrub != NULL); 6350 if (dst->scrub->pfss_v_tag == 0) 6351 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6352 } 6353 6354 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6355 if (src->state < SCTP_ESTABLISHED) { 6356 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6357 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6358 } 6359 } 6360 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6361 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6362 if (src->state < SCTP_SHUTDOWN_PENDING) { 6363 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6364 (*state)->timeout = PFTM_SCTP_CLOSING; 6365 } 6366 } 6367 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6368 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6369 (*state)->timeout = PFTM_SCTP_CLOSED; 6370 } 6371 6372 if (src->scrub != NULL) { 6373 if (src->scrub->pfss_v_tag == 0) { 6374 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6375 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6376 return (PF_DROP); 6377 } 6378 6379 (*state)->expire = pf_get_uptime(); 6380 6381 /* translate source/destination address, if necessary */ 6382 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6383 uint16_t checksum = 0; 6384 struct pf_state_key *nk = (*state)->key[pd->didx]; 6385 6386 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6387 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6388 pf_change_ap(pd->m, pd->src, &pd->hdr.sctp.src_port, 6389 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6390 nk->port[pd->sidx], 1, pd->af); 6391 } 6392 6393 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6394 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6395 pf_change_ap(pd->m, pd->dst, &pd->hdr.sctp.dest_port, 6396 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6397 nk->port[pd->didx], 1, pd->af); 6398 } 6399 } 6400 6401 return (PF_PASS); 6402 } 6403 6404 static void 6405 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6406 { 6407 struct pf_sctp_endpoint key; 6408 struct pf_sctp_endpoint *ep; 6409 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6410 struct pf_sctp_source *i, *tmp; 6411 6412 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6413 return; 6414 6415 PF_SCTP_ENDPOINTS_LOCK(); 6416 6417 key.v_tag = s->dst.scrub->pfss_v_tag; 6418 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6419 if (ep != NULL) { 6420 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6421 if (pf_addr_cmp(&i->addr, 6422 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6423 s->key[PF_SK_WIRE]->af) == 0) { 6424 SDT_PROBE3(pf, sctp, multihome, remove, 6425 key.v_tag, s, i); 6426 TAILQ_REMOVE(&ep->sources, i, entry); 6427 free(i, M_PFTEMP); 6428 break; 6429 } 6430 } 6431 6432 if (TAILQ_EMPTY(&ep->sources)) { 6433 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6434 free(ep, M_PFTEMP); 6435 } 6436 } 6437 6438 /* Other direction. */ 6439 key.v_tag = s->src.scrub->pfss_v_tag; 6440 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6441 if (ep != NULL) { 6442 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6443 if (pf_addr_cmp(&i->addr, 6444 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6445 s->key[PF_SK_WIRE]->af) == 0) { 6446 SDT_PROBE3(pf, sctp, multihome, remove, 6447 key.v_tag, s, i); 6448 TAILQ_REMOVE(&ep->sources, i, entry); 6449 free(i, M_PFTEMP); 6450 break; 6451 } 6452 } 6453 6454 if (TAILQ_EMPTY(&ep->sources)) { 6455 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6456 free(ep, M_PFTEMP); 6457 } 6458 } 6459 6460 PF_SCTP_ENDPOINTS_UNLOCK(); 6461 } 6462 6463 static void 6464 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6465 { 6466 struct pf_sctp_endpoint key = { 6467 .v_tag = v_tag, 6468 }; 6469 struct pf_sctp_source *i; 6470 struct pf_sctp_endpoint *ep; 6471 6472 PF_SCTP_ENDPOINTS_LOCK(); 6473 6474 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6475 if (ep == NULL) { 6476 ep = malloc(sizeof(struct pf_sctp_endpoint), 6477 M_PFTEMP, M_NOWAIT); 6478 if (ep == NULL) { 6479 PF_SCTP_ENDPOINTS_UNLOCK(); 6480 return; 6481 } 6482 6483 ep->v_tag = v_tag; 6484 TAILQ_INIT(&ep->sources); 6485 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6486 } 6487 6488 /* Avoid inserting duplicates. */ 6489 TAILQ_FOREACH(i, &ep->sources, entry) { 6490 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6491 PF_SCTP_ENDPOINTS_UNLOCK(); 6492 return; 6493 } 6494 } 6495 6496 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6497 if (i == NULL) { 6498 PF_SCTP_ENDPOINTS_UNLOCK(); 6499 return; 6500 } 6501 6502 i->af = pd->af; 6503 memcpy(&i->addr, a, sizeof(*a)); 6504 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6505 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6506 6507 PF_SCTP_ENDPOINTS_UNLOCK(); 6508 } 6509 6510 static void 6511 pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif, 6512 struct pf_kstate *s, int action) 6513 { 6514 struct pf_sctp_multihome_job *j, *tmp; 6515 struct pf_sctp_source *i; 6516 int ret __unused; 6517 struct pf_kstate *sm = NULL; 6518 struct pf_krule *ra = NULL; 6519 struct pf_krule *r = &V_pf_default_rule; 6520 struct pf_kruleset *rs = NULL; 6521 bool do_extra = true; 6522 6523 PF_RULES_RLOCK_TRACKER; 6524 6525 again: 6526 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6527 if (s == NULL || action != PF_PASS) 6528 goto free; 6529 6530 /* Confirm we don't recurse here. */ 6531 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6532 6533 switch (j->op) { 6534 case SCTP_ADD_IP_ADDRESS: { 6535 uint32_t v_tag = pd->sctp_initiate_tag; 6536 6537 if (v_tag == 0) { 6538 if (s->direction == pd->dir) 6539 v_tag = s->src.scrub->pfss_v_tag; 6540 else 6541 v_tag = s->dst.scrub->pfss_v_tag; 6542 } 6543 6544 /* 6545 * Avoid duplicating states. We'll already have 6546 * created a state based on the source address of 6547 * the packet, but SCTP endpoints may also list this 6548 * address again in the INIT(_ACK) parameters. 6549 */ 6550 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6551 break; 6552 } 6553 6554 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6555 PF_RULES_RLOCK(); 6556 sm = NULL; 6557 /* 6558 * New connections need to be floating, because 6559 * we cannot know what interfaces it will use. 6560 * That's why we pass V_pfi_all rather than kif. 6561 */ 6562 j->pd.kif = V_pfi_all; 6563 ret = pf_test_rule(&r, &sm, 6564 &j->pd, &ra, &rs, NULL); 6565 PF_RULES_RUNLOCK(); 6566 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret); 6567 if (ret != PF_DROP && sm != NULL) { 6568 /* Inherit v_tag values. */ 6569 if (sm->direction == s->direction) { 6570 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6571 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6572 } else { 6573 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6574 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6575 } 6576 PF_STATE_UNLOCK(sm); 6577 } else { 6578 /* If we try duplicate inserts? */ 6579 break; 6580 } 6581 6582 /* Only add the address if we've actually allowed the state. */ 6583 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6584 6585 if (! do_extra) { 6586 break; 6587 } 6588 /* 6589 * We need to do this for each of our source addresses. 6590 * Find those based on the verification tag. 6591 */ 6592 struct pf_sctp_endpoint key = { 6593 .v_tag = pd->hdr.sctp.v_tag, 6594 }; 6595 struct pf_sctp_endpoint *ep; 6596 6597 PF_SCTP_ENDPOINTS_LOCK(); 6598 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6599 if (ep == NULL) { 6600 PF_SCTP_ENDPOINTS_UNLOCK(); 6601 break; 6602 } 6603 MPASS(ep != NULL); 6604 6605 TAILQ_FOREACH(i, &ep->sources, entry) { 6606 struct pf_sctp_multihome_job *nj; 6607 6608 /* SCTP can intermingle IPv4 and IPv6. */ 6609 if (i->af != pd->af) 6610 continue; 6611 6612 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6613 if (! nj) { 6614 continue; 6615 } 6616 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6617 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6618 nj->pd.src = &nj->src; 6619 // New destination address! 6620 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6621 nj->pd.dst = &nj->dst; 6622 nj->pd.m = j->pd.m; 6623 nj->op = j->op; 6624 6625 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6626 } 6627 PF_SCTP_ENDPOINTS_UNLOCK(); 6628 6629 break; 6630 } 6631 case SCTP_DEL_IP_ADDRESS: { 6632 struct pf_state_key_cmp key; 6633 uint8_t psrc; 6634 6635 bzero(&key, sizeof(key)); 6636 key.af = j->pd.af; 6637 key.proto = IPPROTO_SCTP; 6638 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6639 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6640 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6641 key.port[0] = j->pd.hdr.sctp.src_port; 6642 key.port[1] = j->pd.hdr.sctp.dest_port; 6643 } else { /* stack side, reverse */ 6644 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6645 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6646 key.port[1] = j->pd.hdr.sctp.src_port; 6647 key.port[0] = j->pd.hdr.sctp.dest_port; 6648 } 6649 6650 sm = pf_find_state(kif, &key, j->pd.dir); 6651 if (sm != NULL) { 6652 PF_STATE_LOCK_ASSERT(sm); 6653 if (j->pd.dir == sm->direction) { 6654 psrc = PF_PEER_SRC; 6655 } else { 6656 psrc = PF_PEER_DST; 6657 } 6658 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6659 sm->timeout = PFTM_SCTP_CLOSING; 6660 PF_STATE_UNLOCK(sm); 6661 } 6662 break; 6663 default: 6664 panic("Unknown op %#x", j->op); 6665 } 6666 } 6667 6668 free: 6669 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6670 free(j, M_PFTEMP); 6671 } 6672 6673 /* We may have inserted extra work while processing the list. */ 6674 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6675 do_extra = false; 6676 goto again; 6677 } 6678 } 6679 6680 static int 6681 pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op) 6682 { 6683 int off = 0; 6684 struct pf_sctp_multihome_job *job; 6685 6686 while (off < len) { 6687 struct sctp_paramhdr h; 6688 6689 if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL, 6690 pd->af)) 6691 return (PF_DROP); 6692 6693 /* Parameters are at least 4 bytes. */ 6694 if (ntohs(h.param_length) < 4) 6695 return (PF_DROP); 6696 6697 switch (ntohs(h.param_type)) { 6698 case SCTP_IPV4_ADDRESS: { 6699 struct in_addr t; 6700 6701 if (ntohs(h.param_length) != 6702 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6703 return (PF_DROP); 6704 6705 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6706 NULL, NULL, pd->af)) 6707 return (PF_DROP); 6708 6709 if (in_nullhost(t)) 6710 t.s_addr = pd->src->v4.s_addr; 6711 6712 /* 6713 * We hold the state lock (idhash) here, which means 6714 * that we can't acquire the keyhash, or we'll get a 6715 * LOR (and potentially double-lock things too). We also 6716 * can't release the state lock here, so instead we'll 6717 * enqueue this for async handling. 6718 * There's a relatively small race here, in that a 6719 * packet using the new addresses could arrive already, 6720 * but that's just though luck for it. 6721 */ 6722 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6723 if (! job) 6724 return (PF_DROP); 6725 6726 memcpy(&job->pd, pd, sizeof(*pd)); 6727 6728 // New source address! 6729 memcpy(&job->src, &t, sizeof(t)); 6730 job->pd.src = &job->src; 6731 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6732 job->pd.dst = &job->dst; 6733 job->pd.m = pd->m; 6734 job->op = op; 6735 6736 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6737 break; 6738 } 6739 #ifdef INET6 6740 case SCTP_IPV6_ADDRESS: { 6741 struct in6_addr t; 6742 6743 if (ntohs(h.param_length) != 6744 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6745 return (PF_DROP); 6746 6747 if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t), 6748 NULL, NULL, pd->af)) 6749 return (PF_DROP); 6750 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6751 break; 6752 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6753 memcpy(&t, &pd->src->v6, sizeof(t)); 6754 6755 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6756 if (! job) 6757 return (PF_DROP); 6758 6759 memcpy(&job->pd, pd, sizeof(*pd)); 6760 memcpy(&job->src, &t, sizeof(t)); 6761 job->pd.src = &job->src; 6762 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6763 job->pd.dst = &job->dst; 6764 job->pd.m = pd->m; 6765 job->op = op; 6766 6767 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6768 break; 6769 } 6770 #endif 6771 case SCTP_ADD_IP_ADDRESS: { 6772 int ret; 6773 struct sctp_asconf_paramhdr ah; 6774 6775 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6776 NULL, NULL, pd->af)) 6777 return (PF_DROP); 6778 6779 ret = pf_multihome_scan(start + off + sizeof(ah), 6780 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6781 SCTP_ADD_IP_ADDRESS); 6782 if (ret != PF_PASS) 6783 return (ret); 6784 break; 6785 } 6786 case SCTP_DEL_IP_ADDRESS: { 6787 int ret; 6788 struct sctp_asconf_paramhdr ah; 6789 6790 if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah), 6791 NULL, NULL, pd->af)) 6792 return (PF_DROP); 6793 ret = pf_multihome_scan(start + off + sizeof(ah), 6794 ntohs(ah.ph.param_length) - sizeof(ah), pd, 6795 SCTP_DEL_IP_ADDRESS); 6796 if (ret != PF_PASS) 6797 return (ret); 6798 break; 6799 } 6800 default: 6801 break; 6802 } 6803 6804 off += roundup(ntohs(h.param_length), 4); 6805 } 6806 6807 return (PF_PASS); 6808 } 6809 int 6810 pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd) 6811 { 6812 start += sizeof(struct sctp_init_chunk); 6813 len -= sizeof(struct sctp_init_chunk); 6814 6815 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6816 } 6817 6818 int 6819 pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd) 6820 { 6821 start += sizeof(struct sctp_asconf_chunk); 6822 len -= sizeof(struct sctp_asconf_chunk); 6823 6824 return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS)); 6825 } 6826 6827 int 6828 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6829 struct pf_kstate **state, int direction, 6830 u_int16_t icmpid, u_int16_t type, int icmp_dir, 6831 int *iidx, int multi, int inner) 6832 { 6833 key->af = pd->af; 6834 key->proto = pd->proto; 6835 if (icmp_dir == PF_IN) { 6836 *iidx = pd->sidx; 6837 key->port[pd->sidx] = icmpid; 6838 key->port[pd->didx] = type; 6839 } else { 6840 *iidx = pd->didx; 6841 key->port[pd->sidx] = type; 6842 key->port[pd->didx] = icmpid; 6843 } 6844 if (pf_state_key_addr_setup(pd, key, multi)) 6845 return (PF_DROP); 6846 6847 STATE_LOOKUP(key, *state, pd); 6848 6849 if ((*state)->state_flags & PFSTATE_SLOPPY) 6850 return (-1); 6851 6852 /* Is this ICMP message flowing in right direction? */ 6853 if ((*state)->rule->type && 6854 (((!inner && (*state)->direction == direction) || 6855 (inner && (*state)->direction != direction)) ? 6856 PF_IN : PF_OUT) != icmp_dir) { 6857 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6858 printf("pf: icmp type %d in wrong direction (%d): ", 6859 ntohs(type), icmp_dir); 6860 pf_print_state(*state); 6861 printf("\n"); 6862 } 6863 PF_STATE_UNLOCK(*state); 6864 *state = NULL; 6865 return (PF_DROP); 6866 } 6867 return (-1); 6868 } 6869 6870 static int 6871 pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd, 6872 u_short *reason) 6873 { 6874 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6875 u_int16_t *icmpsum, virtual_id, virtual_type; 6876 u_int8_t icmptype, icmpcode; 6877 int icmp_dir, iidx, ret, multi; 6878 struct pf_state_key_cmp key; 6879 #ifdef INET 6880 u_int16_t icmpid; 6881 #endif 6882 6883 MPASS(*state == NULL); 6884 6885 bzero(&key, sizeof(key)); 6886 switch (pd->proto) { 6887 #ifdef INET 6888 case IPPROTO_ICMP: 6889 icmptype = pd->hdr.icmp.icmp_type; 6890 icmpcode = pd->hdr.icmp.icmp_code; 6891 icmpid = pd->hdr.icmp.icmp_id; 6892 icmpsum = &pd->hdr.icmp.icmp_cksum; 6893 break; 6894 #endif /* INET */ 6895 #ifdef INET6 6896 case IPPROTO_ICMPV6: 6897 icmptype = pd->hdr.icmp6.icmp6_type; 6898 icmpcode = pd->hdr.icmp6.icmp6_code; 6899 #ifdef INET 6900 icmpid = pd->hdr.icmp6.icmp6_id; 6901 #endif 6902 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6903 break; 6904 #endif /* INET6 */ 6905 } 6906 6907 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6908 &virtual_id, &virtual_type) == 0) { 6909 /* 6910 * ICMP query/reply message not related to a TCP/UDP packet. 6911 * Search for an ICMP state. 6912 */ 6913 ret = pf_icmp_state_lookup(&key, pd, state, pd->dir, 6914 virtual_id, virtual_type, icmp_dir, &iidx, 6915 PF_ICMP_MULTI_NONE, 0); 6916 if (ret >= 0) { 6917 MPASS(*state == NULL); 6918 if (ret == PF_DROP && pd->af == AF_INET6 && 6919 icmp_dir == PF_OUT) { 6920 ret = pf_icmp_state_lookup(&key, pd, state, 6921 pd->dir, virtual_id, virtual_type, 6922 icmp_dir, &iidx, multi, 0); 6923 if (ret >= 0) { 6924 MPASS(*state == NULL); 6925 return (ret); 6926 } 6927 } else 6928 return (ret); 6929 } 6930 6931 (*state)->expire = pf_get_uptime(); 6932 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6933 6934 /* translate source/destination address, if necessary */ 6935 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6936 struct pf_state_key *nk = (*state)->key[pd->didx]; 6937 6938 switch (pd->af) { 6939 #ifdef INET 6940 case AF_INET: 6941 if (PF_ANEQ(pd->src, 6942 &nk->addr[pd->sidx], AF_INET)) 6943 pf_change_a(&saddr->v4.s_addr, 6944 pd->ip_sum, 6945 nk->addr[pd->sidx].v4.s_addr, 0); 6946 6947 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6948 AF_INET)) 6949 pf_change_a(&daddr->v4.s_addr, 6950 pd->ip_sum, 6951 nk->addr[pd->didx].v4.s_addr, 0); 6952 6953 if (nk->port[iidx] != 6954 pd->hdr.icmp.icmp_id) { 6955 pd->hdr.icmp.icmp_cksum = 6956 pf_cksum_fixup( 6957 pd->hdr.icmp.icmp_cksum, icmpid, 6958 nk->port[iidx], 0); 6959 pd->hdr.icmp.icmp_id = 6960 nk->port[iidx]; 6961 } 6962 6963 m_copyback(pd->m, pd->off, ICMP_MINLEN, 6964 (caddr_t )&pd->hdr.icmp); 6965 break; 6966 #endif /* INET */ 6967 #ifdef INET6 6968 case AF_INET6: 6969 if (PF_ANEQ(pd->src, 6970 &nk->addr[pd->sidx], AF_INET6)) 6971 pf_change_a6(saddr, 6972 &pd->hdr.icmp6.icmp6_cksum, 6973 &nk->addr[pd->sidx], 0); 6974 6975 if (PF_ANEQ(pd->dst, 6976 &nk->addr[pd->didx], AF_INET6)) 6977 pf_change_a6(daddr, 6978 &pd->hdr.icmp6.icmp6_cksum, 6979 &nk->addr[pd->didx], 0); 6980 6981 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 6982 (caddr_t )&pd->hdr.icmp6); 6983 break; 6984 #endif /* INET6 */ 6985 } 6986 } 6987 return (PF_PASS); 6988 6989 } else { 6990 /* 6991 * ICMP error message in response to a TCP/UDP packet. 6992 * Extract the inner TCP/UDP header and search for that state. 6993 */ 6994 6995 struct pf_pdesc pd2; 6996 bzero(&pd2, sizeof pd2); 6997 #ifdef INET 6998 struct ip h2; 6999 #endif /* INET */ 7000 #ifdef INET6 7001 struct ip6_hdr h2_6; 7002 int fragoff2, extoff2; 7003 u_int32_t jumbolen; 7004 #endif /* INET6 */ 7005 int ipoff2 = 0; 7006 7007 pd2.af = pd->af; 7008 pd2.dir = pd->dir; 7009 /* Payload packet is from the opposite direction. */ 7010 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7011 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7012 pd2.m = pd->m; 7013 switch (pd->af) { 7014 #ifdef INET 7015 case AF_INET: 7016 /* offset of h2 in mbuf chain */ 7017 ipoff2 = pd->off + ICMP_MINLEN; 7018 7019 if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2), 7020 NULL, reason, pd2.af)) { 7021 DPFPRINTF(PF_DEBUG_MISC, 7022 ("pf: ICMP error message too short " 7023 "(ip)\n")); 7024 return (PF_DROP); 7025 } 7026 /* 7027 * ICMP error messages don't refer to non-first 7028 * fragments 7029 */ 7030 if (h2.ip_off & htons(IP_OFFMASK)) { 7031 REASON_SET(reason, PFRES_FRAG); 7032 return (PF_DROP); 7033 } 7034 7035 /* offset of protocol header that follows h2 */ 7036 pd2.off = ipoff2 + (h2.ip_hl << 2); 7037 7038 pd2.proto = h2.ip_p; 7039 pd2.src = (struct pf_addr *)&h2.ip_src; 7040 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7041 pd2.ip_sum = &h2.ip_sum; 7042 break; 7043 #endif /* INET */ 7044 #ifdef INET6 7045 case AF_INET6: 7046 ipoff2 = pd->off + sizeof(struct icmp6_hdr); 7047 7048 if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6), 7049 NULL, reason, pd2.af)) { 7050 DPFPRINTF(PF_DEBUG_MISC, 7051 ("pf: ICMP error message too short " 7052 "(ip6)\n")); 7053 return (PF_DROP); 7054 } 7055 pd2.off = ipoff2; 7056 if (pf_walk_header6(pd->m, &h2_6, &pd2.off, &extoff2, 7057 &fragoff2, &pd2.proto, &jumbolen, 7058 reason) != PF_PASS) 7059 return (PF_DROP); 7060 7061 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7062 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7063 pd2.ip_sum = NULL; 7064 break; 7065 #endif /* INET6 */ 7066 } 7067 7068 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7069 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7070 printf("pf: BAD ICMP %d:%d outer dst: ", 7071 icmptype, icmpcode); 7072 pf_print_host(pd->src, 0, pd->af); 7073 printf(" -> "); 7074 pf_print_host(pd->dst, 0, pd->af); 7075 printf(" inner src: "); 7076 pf_print_host(pd2.src, 0, pd2.af); 7077 printf(" -> "); 7078 pf_print_host(pd2.dst, 0, pd2.af); 7079 printf("\n"); 7080 } 7081 REASON_SET(reason, PFRES_BADSTATE); 7082 return (PF_DROP); 7083 } 7084 7085 switch (pd2.proto) { 7086 case IPPROTO_TCP: { 7087 struct tcphdr th; 7088 u_int32_t seq; 7089 struct pf_state_peer *src, *dst; 7090 u_int8_t dws; 7091 int copyback = 0; 7092 7093 /* 7094 * Only the first 8 bytes of the TCP header can be 7095 * expected. Don't access any TCP header fields after 7096 * th_seq, an ackskew test is not possible. 7097 */ 7098 if (!pf_pull_hdr(pd->m, pd2.off, &th, 8, NULL, reason, 7099 pd2.af)) { 7100 DPFPRINTF(PF_DEBUG_MISC, 7101 ("pf: ICMP error message too short " 7102 "(tcp)\n")); 7103 return (PF_DROP); 7104 } 7105 7106 key.af = pd2.af; 7107 key.proto = IPPROTO_TCP; 7108 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7109 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7110 key.port[pd2.sidx] = th.th_sport; 7111 key.port[pd2.didx] = th.th_dport; 7112 7113 STATE_LOOKUP(&key, *state, pd); 7114 7115 if (pd->dir == (*state)->direction) { 7116 src = &(*state)->dst; 7117 dst = &(*state)->src; 7118 } else { 7119 src = &(*state)->src; 7120 dst = &(*state)->dst; 7121 } 7122 7123 if (src->wscale && dst->wscale) 7124 dws = dst->wscale & PF_WSCALE_MASK; 7125 else 7126 dws = 0; 7127 7128 /* Demodulate sequence number */ 7129 seq = ntohl(th.th_seq) - src->seqdiff; 7130 if (src->seqdiff) { 7131 pf_change_a(&th.th_seq, icmpsum, 7132 htonl(seq), 0); 7133 copyback = 1; 7134 } 7135 7136 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7137 (!SEQ_GEQ(src->seqhi, seq) || 7138 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7139 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7140 printf("pf: BAD ICMP %d:%d ", 7141 icmptype, icmpcode); 7142 pf_print_host(pd->src, 0, pd->af); 7143 printf(" -> "); 7144 pf_print_host(pd->dst, 0, pd->af); 7145 printf(" state: "); 7146 pf_print_state(*state); 7147 printf(" seq=%u\n", seq); 7148 } 7149 REASON_SET(reason, PFRES_BADSTATE); 7150 return (PF_DROP); 7151 } else { 7152 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7153 printf("pf: OK ICMP %d:%d ", 7154 icmptype, icmpcode); 7155 pf_print_host(pd->src, 0, pd->af); 7156 printf(" -> "); 7157 pf_print_host(pd->dst, 0, pd->af); 7158 printf(" state: "); 7159 pf_print_state(*state); 7160 printf(" seq=%u\n", seq); 7161 } 7162 } 7163 7164 /* translate source/destination address, if necessary */ 7165 if ((*state)->key[PF_SK_WIRE] != 7166 (*state)->key[PF_SK_STACK]) { 7167 struct pf_state_key *nk = 7168 (*state)->key[pd->didx]; 7169 7170 if (PF_ANEQ(pd2.src, 7171 &nk->addr[pd2.sidx], pd2.af) || 7172 nk->port[pd2.sidx] != th.th_sport) 7173 pf_change_icmp(pd2.src, &th.th_sport, 7174 daddr, &nk->addr[pd2.sidx], 7175 nk->port[pd2.sidx], NULL, 7176 pd2.ip_sum, icmpsum, 7177 pd->ip_sum, 0, pd2.af); 7178 7179 if (PF_ANEQ(pd2.dst, 7180 &nk->addr[pd2.didx], pd2.af) || 7181 nk->port[pd2.didx] != th.th_dport) 7182 pf_change_icmp(pd2.dst, &th.th_dport, 7183 saddr, &nk->addr[pd2.didx], 7184 nk->port[pd2.didx], NULL, 7185 pd2.ip_sum, icmpsum, 7186 pd->ip_sum, 0, pd2.af); 7187 copyback = 1; 7188 } 7189 7190 if (copyback) { 7191 switch (pd2.af) { 7192 #ifdef INET 7193 case AF_INET: 7194 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7195 (caddr_t )&pd->hdr.icmp); 7196 m_copyback(pd->m, ipoff2, sizeof(h2), 7197 (caddr_t )&h2); 7198 break; 7199 #endif /* INET */ 7200 #ifdef INET6 7201 case AF_INET6: 7202 m_copyback(pd->m, pd->off, 7203 sizeof(struct icmp6_hdr), 7204 (caddr_t )&pd->hdr.icmp6); 7205 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7206 (caddr_t )&h2_6); 7207 break; 7208 #endif /* INET6 */ 7209 } 7210 m_copyback(pd->m, pd2.off, 8, (caddr_t)&th); 7211 } 7212 7213 return (PF_PASS); 7214 break; 7215 } 7216 case IPPROTO_UDP: { 7217 struct udphdr uh; 7218 7219 if (!pf_pull_hdr(pd->m, pd2.off, &uh, sizeof(uh), 7220 NULL, reason, pd2.af)) { 7221 DPFPRINTF(PF_DEBUG_MISC, 7222 ("pf: ICMP error message too short " 7223 "(udp)\n")); 7224 return (PF_DROP); 7225 } 7226 7227 key.af = pd2.af; 7228 key.proto = IPPROTO_UDP; 7229 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7230 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7231 key.port[pd2.sidx] = uh.uh_sport; 7232 key.port[pd2.didx] = uh.uh_dport; 7233 7234 STATE_LOOKUP(&key, *state, pd); 7235 7236 /* translate source/destination address, if necessary */ 7237 if ((*state)->key[PF_SK_WIRE] != 7238 (*state)->key[PF_SK_STACK]) { 7239 struct pf_state_key *nk = 7240 (*state)->key[pd->didx]; 7241 7242 if (PF_ANEQ(pd2.src, 7243 &nk->addr[pd2.sidx], pd2.af) || 7244 nk->port[pd2.sidx] != uh.uh_sport) 7245 pf_change_icmp(pd2.src, &uh.uh_sport, 7246 daddr, &nk->addr[pd2.sidx], 7247 nk->port[pd2.sidx], &uh.uh_sum, 7248 pd2.ip_sum, icmpsum, 7249 pd->ip_sum, 1, pd2.af); 7250 7251 if (PF_ANEQ(pd2.dst, 7252 &nk->addr[pd2.didx], pd2.af) || 7253 nk->port[pd2.didx] != uh.uh_dport) 7254 pf_change_icmp(pd2.dst, &uh.uh_dport, 7255 saddr, &nk->addr[pd2.didx], 7256 nk->port[pd2.didx], &uh.uh_sum, 7257 pd2.ip_sum, icmpsum, 7258 pd->ip_sum, 1, pd2.af); 7259 7260 switch (pd2.af) { 7261 #ifdef INET 7262 case AF_INET: 7263 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7264 (caddr_t )&pd->hdr.icmp); 7265 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7266 break; 7267 #endif /* INET */ 7268 #ifdef INET6 7269 case AF_INET6: 7270 m_copyback(pd->m, pd->off, 7271 sizeof(struct icmp6_hdr), 7272 (caddr_t )&pd->hdr.icmp6); 7273 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7274 (caddr_t )&h2_6); 7275 break; 7276 #endif /* INET6 */ 7277 } 7278 m_copyback(pd->m, pd2.off, sizeof(uh), (caddr_t)&uh); 7279 } 7280 return (PF_PASS); 7281 break; 7282 } 7283 #ifdef INET 7284 case IPPROTO_ICMP: { 7285 struct icmp *iih = &pd2.hdr.icmp; 7286 7287 if (pd2.af != AF_INET) { 7288 REASON_SET(reason, PFRES_NORM); 7289 return (PF_DROP); 7290 } 7291 7292 if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN, 7293 NULL, reason, pd2.af)) { 7294 DPFPRINTF(PF_DEBUG_MISC, 7295 ("pf: ICMP error message too short i" 7296 "(icmp)\n")); 7297 return (PF_DROP); 7298 } 7299 7300 icmpid = iih->icmp_id; 7301 pf_icmp_mapping(&pd2, iih->icmp_type, 7302 &icmp_dir, &multi, &virtual_id, &virtual_type); 7303 7304 ret = pf_icmp_state_lookup(&key, &pd2, state, 7305 pd2.dir, virtual_id, virtual_type, 7306 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7307 if (ret >= 0) { 7308 MPASS(*state == NULL); 7309 return (ret); 7310 } 7311 7312 /* translate source/destination address, if necessary */ 7313 if ((*state)->key[PF_SK_WIRE] != 7314 (*state)->key[PF_SK_STACK]) { 7315 struct pf_state_key *nk = 7316 (*state)->key[pd->didx]; 7317 7318 if (PF_ANEQ(pd2.src, 7319 &nk->addr[pd2.sidx], pd2.af) || 7320 (virtual_type == htons(ICMP_ECHO) && 7321 nk->port[iidx] != iih->icmp_id)) 7322 pf_change_icmp(pd2.src, 7323 (virtual_type == htons(ICMP_ECHO)) ? 7324 &iih->icmp_id : NULL, 7325 daddr, &nk->addr[pd2.sidx], 7326 (virtual_type == htons(ICMP_ECHO)) ? 7327 nk->port[iidx] : 0, NULL, 7328 pd2.ip_sum, icmpsum, 7329 pd->ip_sum, 0, AF_INET); 7330 7331 if (PF_ANEQ(pd2.dst, 7332 &nk->addr[pd2.didx], pd2.af)) 7333 pf_change_icmp(pd2.dst, NULL, NULL, 7334 &nk->addr[pd2.didx], 0, NULL, 7335 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7336 AF_INET); 7337 7338 m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7339 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7340 m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih); 7341 } 7342 return (PF_PASS); 7343 break; 7344 } 7345 #endif /* INET */ 7346 #ifdef INET6 7347 case IPPROTO_ICMPV6: { 7348 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7349 7350 if (pd2.af != AF_INET6) { 7351 REASON_SET(reason, PFRES_NORM); 7352 return (PF_DROP); 7353 } 7354 7355 if (!pf_pull_hdr(pd->m, pd2.off, iih, 7356 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7357 DPFPRINTF(PF_DEBUG_MISC, 7358 ("pf: ICMP error message too short " 7359 "(icmp6)\n")); 7360 return (PF_DROP); 7361 } 7362 7363 pf_icmp_mapping(&pd2, iih->icmp6_type, 7364 &icmp_dir, &multi, &virtual_id, &virtual_type); 7365 7366 ret = pf_icmp_state_lookup(&key, &pd2, state, 7367 pd->dir, virtual_id, virtual_type, 7368 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7369 if (ret >= 0) { 7370 MPASS(*state == NULL); 7371 if (ret == PF_DROP && pd2.af == AF_INET6 && 7372 icmp_dir == PF_OUT) { 7373 ret = pf_icmp_state_lookup(&key, &pd2, 7374 state, pd->dir, 7375 virtual_id, virtual_type, 7376 icmp_dir, &iidx, multi, 1); 7377 if (ret >= 0) { 7378 MPASS(*state == NULL); 7379 return (ret); 7380 } 7381 } else 7382 return (ret); 7383 } 7384 7385 /* translate source/destination address, if necessary */ 7386 if ((*state)->key[PF_SK_WIRE] != 7387 (*state)->key[PF_SK_STACK]) { 7388 struct pf_state_key *nk = 7389 (*state)->key[pd->didx]; 7390 7391 if (PF_ANEQ(pd2.src, 7392 &nk->addr[pd2.sidx], pd2.af) || 7393 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7394 nk->port[pd2.sidx] != iih->icmp6_id)) 7395 pf_change_icmp(pd2.src, 7396 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7397 ? &iih->icmp6_id : NULL, 7398 daddr, &nk->addr[pd2.sidx], 7399 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7400 ? nk->port[iidx] : 0, NULL, 7401 pd2.ip_sum, icmpsum, 7402 pd->ip_sum, 0, AF_INET6); 7403 7404 if (PF_ANEQ(pd2.dst, 7405 &nk->addr[pd2.didx], pd2.af)) 7406 pf_change_icmp(pd2.dst, NULL, NULL, 7407 &nk->addr[pd2.didx], 0, NULL, 7408 pd2.ip_sum, icmpsum, 7409 pd->ip_sum, 0, AF_INET6); 7410 7411 m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr), 7412 (caddr_t)&pd->hdr.icmp6); 7413 m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7414 m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr), 7415 (caddr_t)iih); 7416 } 7417 return (PF_PASS); 7418 break; 7419 } 7420 #endif /* INET6 */ 7421 default: { 7422 key.af = pd2.af; 7423 key.proto = pd2.proto; 7424 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7425 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7426 key.port[0] = key.port[1] = 0; 7427 7428 STATE_LOOKUP(&key, *state, pd); 7429 7430 /* translate source/destination address, if necessary */ 7431 if ((*state)->key[PF_SK_WIRE] != 7432 (*state)->key[PF_SK_STACK]) { 7433 struct pf_state_key *nk = 7434 (*state)->key[pd->didx]; 7435 7436 if (PF_ANEQ(pd2.src, 7437 &nk->addr[pd2.sidx], pd2.af)) 7438 pf_change_icmp(pd2.src, NULL, daddr, 7439 &nk->addr[pd2.sidx], 0, NULL, 7440 pd2.ip_sum, icmpsum, 7441 pd->ip_sum, 0, pd2.af); 7442 7443 if (PF_ANEQ(pd2.dst, 7444 &nk->addr[pd2.didx], pd2.af)) 7445 pf_change_icmp(pd2.dst, NULL, saddr, 7446 &nk->addr[pd2.didx], 0, NULL, 7447 pd2.ip_sum, icmpsum, 7448 pd->ip_sum, 0, pd2.af); 7449 7450 switch (pd2.af) { 7451 #ifdef INET 7452 case AF_INET: 7453 m_copyback(pd->m, pd->off, ICMP_MINLEN, 7454 (caddr_t)&pd->hdr.icmp); 7455 m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2); 7456 break; 7457 #endif /* INET */ 7458 #ifdef INET6 7459 case AF_INET6: 7460 m_copyback(pd->m, pd->off, 7461 sizeof(struct icmp6_hdr), 7462 (caddr_t )&pd->hdr.icmp6); 7463 m_copyback(pd->m, ipoff2, sizeof(h2_6), 7464 (caddr_t )&h2_6); 7465 break; 7466 #endif /* INET6 */ 7467 } 7468 } 7469 return (PF_PASS); 7470 break; 7471 } 7472 } 7473 } 7474 } 7475 7476 static int 7477 pf_test_state_other(struct pf_kstate **state, struct pf_pdesc *pd) 7478 { 7479 struct pf_state_peer *src, *dst; 7480 struct pf_state_key_cmp key; 7481 uint8_t psrc, pdst; 7482 7483 bzero(&key, sizeof(key)); 7484 key.af = pd->af; 7485 key.proto = pd->proto; 7486 if (pd->dir == PF_IN) { 7487 PF_ACPY(&key.addr[0], pd->src, key.af); 7488 PF_ACPY(&key.addr[1], pd->dst, key.af); 7489 key.port[0] = key.port[1] = 0; 7490 } else { 7491 PF_ACPY(&key.addr[1], pd->src, key.af); 7492 PF_ACPY(&key.addr[0], pd->dst, key.af); 7493 key.port[1] = key.port[0] = 0; 7494 } 7495 7496 STATE_LOOKUP(&key, *state, pd); 7497 7498 if (pd->dir == (*state)->direction) { 7499 src = &(*state)->src; 7500 dst = &(*state)->dst; 7501 psrc = PF_PEER_SRC; 7502 pdst = PF_PEER_DST; 7503 } else { 7504 src = &(*state)->dst; 7505 dst = &(*state)->src; 7506 psrc = PF_PEER_DST; 7507 pdst = PF_PEER_SRC; 7508 } 7509 7510 /* update states */ 7511 if (src->state < PFOTHERS_SINGLE) 7512 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7513 if (dst->state == PFOTHERS_SINGLE) 7514 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7515 7516 /* update expire time */ 7517 (*state)->expire = pf_get_uptime(); 7518 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7519 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7520 else 7521 (*state)->timeout = PFTM_OTHER_SINGLE; 7522 7523 /* translate source/destination address, if necessary */ 7524 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7525 struct pf_state_key *nk = (*state)->key[pd->didx]; 7526 7527 KASSERT(nk, ("%s: nk is null", __func__)); 7528 KASSERT(pd, ("%s: pd is null", __func__)); 7529 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7530 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7531 switch (pd->af) { 7532 #ifdef INET 7533 case AF_INET: 7534 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7535 pf_change_a(&pd->src->v4.s_addr, 7536 pd->ip_sum, 7537 nk->addr[pd->sidx].v4.s_addr, 7538 0); 7539 7540 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7541 pf_change_a(&pd->dst->v4.s_addr, 7542 pd->ip_sum, 7543 nk->addr[pd->didx].v4.s_addr, 7544 0); 7545 7546 break; 7547 #endif /* INET */ 7548 #ifdef INET6 7549 case AF_INET6: 7550 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7551 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7552 7553 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7554 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7555 #endif /* INET6 */ 7556 } 7557 } 7558 return (PF_PASS); 7559 } 7560 7561 /* 7562 * ipoff and off are measured from the start of the mbuf chain. 7563 * h must be at "ipoff" on the mbuf chain. 7564 */ 7565 void * 7566 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7567 u_short *actionp, u_short *reasonp, sa_family_t af) 7568 { 7569 switch (af) { 7570 #ifdef INET 7571 case AF_INET: { 7572 const struct ip *h = mtod(m, struct ip *); 7573 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7574 7575 if (fragoff) { 7576 if (fragoff >= len) 7577 ACTION_SET(actionp, PF_PASS); 7578 else { 7579 ACTION_SET(actionp, PF_DROP); 7580 REASON_SET(reasonp, PFRES_FRAG); 7581 } 7582 return (NULL); 7583 } 7584 if (m->m_pkthdr.len < off + len || 7585 ntohs(h->ip_len) < off + len) { 7586 ACTION_SET(actionp, PF_DROP); 7587 REASON_SET(reasonp, PFRES_SHORT); 7588 return (NULL); 7589 } 7590 break; 7591 } 7592 #endif /* INET */ 7593 #ifdef INET6 7594 case AF_INET6: { 7595 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7596 7597 if (m->m_pkthdr.len < off + len || 7598 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7599 (unsigned)(off + len)) { 7600 ACTION_SET(actionp, PF_DROP); 7601 REASON_SET(reasonp, PFRES_SHORT); 7602 return (NULL); 7603 } 7604 break; 7605 } 7606 #endif /* INET6 */ 7607 } 7608 m_copydata(m, off, len, p); 7609 return (p); 7610 } 7611 7612 int 7613 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7614 int rtableid) 7615 { 7616 struct ifnet *ifp; 7617 7618 /* 7619 * Skip check for addresses with embedded interface scope, 7620 * as they would always match anyway. 7621 */ 7622 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7623 return (1); 7624 7625 if (af != AF_INET && af != AF_INET6) 7626 return (0); 7627 7628 if (kif == V_pfi_all) 7629 return (1); 7630 7631 /* Skip checks for ipsec interfaces */ 7632 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7633 return (1); 7634 7635 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7636 7637 switch (af) { 7638 #ifdef INET6 7639 case AF_INET6: 7640 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7641 ifp)); 7642 #endif 7643 #ifdef INET 7644 case AF_INET: 7645 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7646 ifp)); 7647 #endif 7648 } 7649 7650 return (0); 7651 } 7652 7653 #ifdef INET 7654 static void 7655 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7656 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7657 { 7658 struct mbuf *m0, *m1, *md; 7659 struct sockaddr_in dst; 7660 struct ip *ip; 7661 struct pfi_kkif *nkif = NULL; 7662 struct ifnet *ifp = NULL; 7663 struct pf_addr naddr; 7664 int error = 0; 7665 uint16_t ip_len, ip_off; 7666 uint16_t tmp; 7667 int r_rt, r_dir; 7668 7669 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7670 7671 SDT_PROBE4(pf, ip, route_to, entry, *m, pd, s, oifp); 7672 7673 if (s) { 7674 r_rt = s->rt; 7675 r_dir = s->direction; 7676 } else { 7677 r_rt = r->rt; 7678 r_dir = r->direction; 7679 } 7680 7681 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7682 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7683 __func__)); 7684 7685 if ((pd->pf_mtag == NULL && 7686 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7687 pd->pf_mtag->routed++ > 3) { 7688 m0 = *m; 7689 *m = NULL; 7690 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7691 goto bad_locked; 7692 } 7693 7694 if (r_rt == PF_DUPTO) { 7695 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7696 if (s == NULL) { 7697 ifp = r->rpool.cur->kif ? 7698 r->rpool.cur->kif->pfik_ifp : NULL; 7699 } else { 7700 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7701 /* If pfsync'd */ 7702 if (ifp == NULL && r->rpool.cur != NULL) 7703 ifp = r->rpool.cur->kif ? 7704 r->rpool.cur->kif->pfik_ifp : NULL; 7705 PF_STATE_UNLOCK(s); 7706 } 7707 if (ifp == oifp) { 7708 /* When the 2nd interface is not skipped */ 7709 return; 7710 } else { 7711 m0 = *m; 7712 *m = NULL; 7713 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7714 goto bad; 7715 } 7716 } else { 7717 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7718 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7719 if (s) 7720 PF_STATE_UNLOCK(s); 7721 return; 7722 } 7723 } 7724 } else { 7725 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7726 pf_dummynet(pd, s, r, m); 7727 if (s) 7728 PF_STATE_UNLOCK(s); 7729 return; 7730 } 7731 m0 = *m; 7732 } 7733 7734 ip = mtod(m0, struct ip *); 7735 7736 bzero(&dst, sizeof(dst)); 7737 dst.sin_family = AF_INET; 7738 dst.sin_len = sizeof(dst); 7739 dst.sin_addr = ip->ip_dst; 7740 7741 bzero(&naddr, sizeof(naddr)); 7742 7743 if (s == NULL) { 7744 if (TAILQ_EMPTY(&r->rpool.list)) { 7745 DPFPRINTF(PF_DEBUG_URGENT, 7746 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7747 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7748 goto bad_locked; 7749 } 7750 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7751 &naddr, &nkif, NULL); 7752 if (!PF_AZERO(&naddr, AF_INET)) 7753 dst.sin_addr.s_addr = naddr.v4.s_addr; 7754 ifp = nkif ? nkif->pfik_ifp : NULL; 7755 } else { 7756 struct pfi_kkif *kif; 7757 7758 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7759 dst.sin_addr.s_addr = 7760 s->rt_addr.v4.s_addr; 7761 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7762 kif = s->rt_kif; 7763 /* If pfsync'd */ 7764 if (ifp == NULL && r->rpool.cur != NULL) { 7765 ifp = r->rpool.cur->kif ? 7766 r->rpool.cur->kif->pfik_ifp : NULL; 7767 kif = r->rpool.cur->kif; 7768 } 7769 if (ifp != NULL && kif != NULL && 7770 r->rule_flag & PFRULE_IFBOUND && 7771 r->rt == PF_REPLYTO && 7772 s->kif == V_pfi_all) { 7773 s->kif = kif; 7774 s->orig_kif = oifp->if_pf_kif; 7775 } 7776 7777 PF_STATE_UNLOCK(s); 7778 } 7779 7780 if (ifp == NULL) { 7781 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7782 goto bad; 7783 } 7784 7785 if (pd->dir == PF_IN) { 7786 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 7787 &pd->act) != PF_PASS) { 7788 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7789 goto bad; 7790 } else if (m0 == NULL) { 7791 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7792 goto done; 7793 } 7794 if (m0->m_len < sizeof(struct ip)) { 7795 DPFPRINTF(PF_DEBUG_URGENT, 7796 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7797 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7798 goto bad; 7799 } 7800 ip = mtod(m0, struct ip *); 7801 } 7802 7803 if (ifp->if_flags & IFF_LOOPBACK) 7804 m0->m_flags |= M_SKIP_FIREWALL; 7805 7806 ip_len = ntohs(ip->ip_len); 7807 ip_off = ntohs(ip->ip_off); 7808 7809 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7810 m0->m_pkthdr.csum_flags |= CSUM_IP; 7811 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7812 in_delayed_cksum(m0); 7813 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7814 } 7815 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7816 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7817 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7818 } 7819 7820 if (pd->dir == PF_IN) { 7821 /* 7822 * Make sure dummynet gets the correct direction, in case it needs to 7823 * re-inject later. 7824 */ 7825 pd->dir = PF_OUT; 7826 7827 /* 7828 * The following processing is actually the rest of the inbound processing, even 7829 * though we've marked it as outbound (so we don't look through dummynet) and it 7830 * happens after the outbound processing (pf_test(PF_OUT) above). 7831 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7832 * conclusion about what direction it's processing, and we can't fix it or it 7833 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7834 * decision will pick the right pipe, and everything will mostly work as expected. 7835 */ 7836 tmp = pd->act.dnrpipe; 7837 pd->act.dnrpipe = pd->act.dnpipe; 7838 pd->act.dnpipe = tmp; 7839 } 7840 7841 /* 7842 * If small enough for interface, or the interface will take 7843 * care of the fragmentation for us, we can just send directly. 7844 */ 7845 if (ip_len <= ifp->if_mtu || 7846 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7847 ip->ip_sum = 0; 7848 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7849 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7850 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7851 } 7852 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7853 7854 md = m0; 7855 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7856 if (md != NULL) { 7857 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7858 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 7859 } 7860 goto done; 7861 } 7862 7863 /* Balk when DF bit is set or the interface didn't support TSO. */ 7864 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7865 error = EMSGSIZE; 7866 KMOD_IPSTAT_INC(ips_cantfrag); 7867 if (r_rt != PF_DUPTO) { 7868 if (s && s->nat_rule != NULL) 7869 PACKET_UNDO_NAT(m0, pd, 7870 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7871 s); 7872 7873 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7874 ifp->if_mtu); 7875 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7876 goto done; 7877 } else { 7878 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7879 goto bad; 7880 } 7881 } 7882 7883 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7884 if (error) { 7885 SDT_PROBE1(pf, ip, route_to, drop, __LINE__); 7886 goto bad; 7887 } 7888 7889 for (; m0; m0 = m1) { 7890 m1 = m0->m_nextpkt; 7891 m0->m_nextpkt = NULL; 7892 if (error == 0) { 7893 m_clrprotoflags(m0); 7894 md = m0; 7895 pd->pf_mtag = pf_find_mtag(md); 7896 error = pf_dummynet_route(pd, s, r, ifp, 7897 sintosa(&dst), &md); 7898 if (md != NULL) { 7899 error = (*ifp->if_output)(ifp, md, 7900 sintosa(&dst), NULL); 7901 SDT_PROBE2(pf, ip, route_to, output, ifp, error); 7902 } 7903 } else 7904 m_freem(m0); 7905 } 7906 7907 if (error == 0) 7908 KMOD_IPSTAT_INC(ips_fragmented); 7909 7910 done: 7911 if (r_rt != PF_DUPTO) 7912 *m = NULL; 7913 return; 7914 7915 bad_locked: 7916 if (s) 7917 PF_STATE_UNLOCK(s); 7918 bad: 7919 m_freem(m0); 7920 goto done; 7921 } 7922 #endif /* INET */ 7923 7924 #ifdef INET6 7925 static void 7926 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7927 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7928 { 7929 struct mbuf *m0, *md; 7930 struct sockaddr_in6 dst; 7931 struct ip6_hdr *ip6; 7932 struct pfi_kkif *nkif = NULL; 7933 struct ifnet *ifp = NULL; 7934 struct pf_addr naddr; 7935 int r_rt, r_dir; 7936 7937 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7938 7939 SDT_PROBE4(pf, ip6, route_to, entry, *m, pd, s, oifp); 7940 7941 if (s) { 7942 r_rt = s->rt; 7943 r_dir = s->direction; 7944 } else { 7945 r_rt = r->rt; 7946 r_dir = r->direction; 7947 } 7948 7949 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7950 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7951 __func__)); 7952 7953 if ((pd->pf_mtag == NULL && 7954 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7955 pd->pf_mtag->routed++ > 3) { 7956 m0 = *m; 7957 *m = NULL; 7958 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 7959 goto bad_locked; 7960 } 7961 7962 if (r_rt == PF_DUPTO) { 7963 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7964 if (s == NULL) { 7965 ifp = r->rpool.cur->kif ? 7966 r->rpool.cur->kif->pfik_ifp : NULL; 7967 } else { 7968 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7969 /* If pfsync'd */ 7970 if (ifp == NULL && r->rpool.cur != NULL) 7971 ifp = r->rpool.cur->kif ? 7972 r->rpool.cur->kif->pfik_ifp : NULL; 7973 PF_STATE_UNLOCK(s); 7974 } 7975 if (ifp == oifp) { 7976 /* When the 2nd interface is not skipped */ 7977 return; 7978 } else { 7979 m0 = *m; 7980 *m = NULL; 7981 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 7982 goto bad; 7983 } 7984 } else { 7985 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7986 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7987 if (s) 7988 PF_STATE_UNLOCK(s); 7989 return; 7990 } 7991 } 7992 } else { 7993 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7994 pf_dummynet(pd, s, r, m); 7995 if (s) 7996 PF_STATE_UNLOCK(s); 7997 return; 7998 } 7999 m0 = *m; 8000 } 8001 8002 ip6 = mtod(m0, struct ip6_hdr *); 8003 8004 bzero(&dst, sizeof(dst)); 8005 dst.sin6_family = AF_INET6; 8006 dst.sin6_len = sizeof(dst); 8007 dst.sin6_addr = ip6->ip6_dst; 8008 8009 bzero(&naddr, sizeof(naddr)); 8010 8011 if (s == NULL) { 8012 if (TAILQ_EMPTY(&r->rpool.list)) { 8013 DPFPRINTF(PF_DEBUG_URGENT, 8014 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 8015 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8016 goto bad_locked; 8017 } 8018 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 8019 &naddr, &nkif, NULL); 8020 if (!PF_AZERO(&naddr, AF_INET6)) 8021 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8022 &naddr, AF_INET6); 8023 ifp = nkif ? nkif->pfik_ifp : NULL; 8024 } else { 8025 struct pfi_kkif *kif; 8026 8027 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 8028 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8029 &s->rt_addr, AF_INET6); 8030 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8031 kif = s->rt_kif; 8032 /* If pfsync'd */ 8033 if (ifp == NULL && r->rpool.cur != NULL) { 8034 ifp = r->rpool.cur->kif ? 8035 r->rpool.cur->kif->pfik_ifp : NULL; 8036 kif = r->rpool.cur->kif; 8037 } 8038 if (ifp != NULL && kif != NULL && 8039 r->rule_flag & PFRULE_IFBOUND && 8040 r->rt == PF_REPLYTO && 8041 s->kif == V_pfi_all) { 8042 s->kif = kif; 8043 s->orig_kif = oifp->if_pf_kif; 8044 } 8045 } 8046 8047 if (s) 8048 PF_STATE_UNLOCK(s); 8049 8050 if (ifp == NULL) { 8051 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8052 goto bad; 8053 } 8054 8055 if (pd->dir == PF_IN) { 8056 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8057 &pd->act) != PF_PASS) { 8058 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8059 goto bad; 8060 } else if (m0 == NULL) { 8061 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8062 goto done; 8063 } 8064 if (m0->m_len < sizeof(struct ip6_hdr)) { 8065 DPFPRINTF(PF_DEBUG_URGENT, 8066 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8067 __func__)); 8068 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8069 goto bad; 8070 } 8071 ip6 = mtod(m0, struct ip6_hdr *); 8072 } 8073 8074 if (ifp->if_flags & IFF_LOOPBACK) 8075 m0->m_flags |= M_SKIP_FIREWALL; 8076 8077 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8078 ~ifp->if_hwassist) { 8079 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8080 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8081 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8082 } 8083 8084 /* 8085 * If the packet is too large for the outgoing interface, 8086 * send back an icmp6 error. 8087 */ 8088 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8089 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8090 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8091 md = m0; 8092 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8093 if (md != NULL) { 8094 int ret __sdt_used; 8095 ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8096 SDT_PROBE2(pf, ip6, route_to, output, ifp, ret); 8097 } 8098 } 8099 else { 8100 in6_ifstat_inc(ifp, ifs6_in_toobig); 8101 if (r_rt != PF_DUPTO) { 8102 if (s && s->nat_rule != NULL) 8103 PACKET_UNDO_NAT(m0, pd, 8104 ((caddr_t)ip6 - m0->m_data) + 8105 sizeof(struct ip6_hdr), s); 8106 8107 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8108 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8109 } else { 8110 SDT_PROBE1(pf, ip6, route_to, drop, __LINE__); 8111 goto bad; 8112 } 8113 } 8114 8115 done: 8116 if (r_rt != PF_DUPTO) 8117 *m = NULL; 8118 return; 8119 8120 bad_locked: 8121 if (s) 8122 PF_STATE_UNLOCK(s); 8123 bad: 8124 m_freem(m0); 8125 goto done; 8126 } 8127 #endif /* INET6 */ 8128 8129 /* 8130 * FreeBSD supports cksum offloads for the following drivers. 8131 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8132 * 8133 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8134 * network driver performed cksum including pseudo header, need to verify 8135 * csum_data 8136 * CSUM_DATA_VALID : 8137 * network driver performed cksum, needs to additional pseudo header 8138 * cksum computation with partial csum_data(i.e. lack of H/W support for 8139 * pseudo header, for instance sk(4) and possibly gem(4)) 8140 * 8141 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8142 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8143 * TCP/UDP layer. 8144 * Also, set csum_data to 0xffff to force cksum validation. 8145 */ 8146 static int 8147 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8148 { 8149 u_int16_t sum = 0; 8150 int hw_assist = 0; 8151 struct ip *ip; 8152 8153 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8154 return (1); 8155 if (m->m_pkthdr.len < off + len) 8156 return (1); 8157 8158 switch (p) { 8159 case IPPROTO_TCP: 8160 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8161 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8162 sum = m->m_pkthdr.csum_data; 8163 } else { 8164 ip = mtod(m, struct ip *); 8165 sum = in_pseudo(ip->ip_src.s_addr, 8166 ip->ip_dst.s_addr, htonl((u_short)len + 8167 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8168 } 8169 sum ^= 0xffff; 8170 ++hw_assist; 8171 } 8172 break; 8173 case IPPROTO_UDP: 8174 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8175 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8176 sum = m->m_pkthdr.csum_data; 8177 } else { 8178 ip = mtod(m, struct ip *); 8179 sum = in_pseudo(ip->ip_src.s_addr, 8180 ip->ip_dst.s_addr, htonl((u_short)len + 8181 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8182 } 8183 sum ^= 0xffff; 8184 ++hw_assist; 8185 } 8186 break; 8187 case IPPROTO_ICMP: 8188 #ifdef INET6 8189 case IPPROTO_ICMPV6: 8190 #endif /* INET6 */ 8191 break; 8192 default: 8193 return (1); 8194 } 8195 8196 if (!hw_assist) { 8197 switch (af) { 8198 case AF_INET: 8199 if (p == IPPROTO_ICMP) { 8200 if (m->m_len < off) 8201 return (1); 8202 m->m_data += off; 8203 m->m_len -= off; 8204 sum = in_cksum(m, len); 8205 m->m_data -= off; 8206 m->m_len += off; 8207 } else { 8208 if (m->m_len < sizeof(struct ip)) 8209 return (1); 8210 sum = in4_cksum(m, p, off, len); 8211 } 8212 break; 8213 #ifdef INET6 8214 case AF_INET6: 8215 if (m->m_len < sizeof(struct ip6_hdr)) 8216 return (1); 8217 sum = in6_cksum(m, p, off, len); 8218 break; 8219 #endif /* INET6 */ 8220 } 8221 } 8222 if (sum) { 8223 switch (p) { 8224 case IPPROTO_TCP: 8225 { 8226 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8227 break; 8228 } 8229 case IPPROTO_UDP: 8230 { 8231 KMOD_UDPSTAT_INC(udps_badsum); 8232 break; 8233 } 8234 #ifdef INET 8235 case IPPROTO_ICMP: 8236 { 8237 KMOD_ICMPSTAT_INC(icps_checksum); 8238 break; 8239 } 8240 #endif 8241 #ifdef INET6 8242 case IPPROTO_ICMPV6: 8243 { 8244 KMOD_ICMP6STAT_INC(icp6s_checksum); 8245 break; 8246 } 8247 #endif /* INET6 */ 8248 } 8249 return (1); 8250 } else { 8251 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8252 m->m_pkthdr.csum_flags |= 8253 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8254 m->m_pkthdr.csum_data = 0xffff; 8255 } 8256 } 8257 return (0); 8258 } 8259 8260 static bool 8261 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8262 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8263 { 8264 int dndir = r->direction; 8265 8266 if (s && dndir == PF_INOUT) { 8267 dndir = s->direction; 8268 } else if (dndir == PF_INOUT) { 8269 /* Assume primary direction. Happens when we've set dnpipe in 8270 * the ethernet level code. */ 8271 dndir = pd->dir; 8272 } 8273 8274 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8275 return (false); 8276 8277 memset(dnflow, 0, sizeof(*dnflow)); 8278 8279 if (pd->dport != NULL) 8280 dnflow->f_id.dst_port = ntohs(*pd->dport); 8281 if (pd->sport != NULL) 8282 dnflow->f_id.src_port = ntohs(*pd->sport); 8283 8284 if (pd->dir == PF_IN) 8285 dnflow->flags |= IPFW_ARGS_IN; 8286 else 8287 dnflow->flags |= IPFW_ARGS_OUT; 8288 8289 if (pd->dir != dndir && pd->act.dnrpipe) { 8290 dnflow->rule.info = pd->act.dnrpipe; 8291 } 8292 else if (pd->dir == dndir && pd->act.dnpipe) { 8293 dnflow->rule.info = pd->act.dnpipe; 8294 } 8295 else { 8296 return (false); 8297 } 8298 8299 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8300 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8301 dnflow->rule.info |= IPFW_IS_PIPE; 8302 8303 dnflow->f_id.proto = pd->proto; 8304 dnflow->f_id.extra = dnflow->rule.info; 8305 switch (pd->af) { 8306 case AF_INET: 8307 dnflow->f_id.addr_type = 4; 8308 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8309 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8310 break; 8311 case AF_INET6: 8312 dnflow->flags |= IPFW_ARGS_IP6; 8313 dnflow->f_id.addr_type = 6; 8314 dnflow->f_id.src_ip6 = pd->src->v6; 8315 dnflow->f_id.dst_ip6 = pd->dst->v6; 8316 break; 8317 } 8318 8319 return (true); 8320 } 8321 8322 int 8323 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8324 struct inpcb *inp) 8325 { 8326 struct pfi_kkif *kif; 8327 struct mbuf *m = *m0; 8328 8329 M_ASSERTPKTHDR(m); 8330 MPASS(ifp->if_vnet == curvnet); 8331 NET_EPOCH_ASSERT(); 8332 8333 if (!V_pf_status.running) 8334 return (PF_PASS); 8335 8336 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8337 8338 if (kif == NULL) { 8339 DPFPRINTF(PF_DEBUG_URGENT, 8340 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8341 return (PF_DROP); 8342 } 8343 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8344 return (PF_PASS); 8345 8346 if (m->m_flags & M_SKIP_FIREWALL) 8347 return (PF_PASS); 8348 8349 if (__predict_false(! M_WRITABLE(*m0))) { 8350 m = *m0 = m_unshare(*m0, M_NOWAIT); 8351 if (*m0 == NULL) 8352 return (PF_DROP); 8353 } 8354 8355 /* Stateless! */ 8356 return (pf_test_eth_rule(dir, kif, m0)); 8357 } 8358 8359 static __inline void 8360 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8361 { 8362 struct m_tag *mtag; 8363 8364 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8365 8366 /* dummynet adds this tag, but pf does not need it, 8367 * and keeping it creates unexpected behavior, 8368 * e.g. in case of divert(4) usage right after dummynet. */ 8369 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8370 if (mtag != NULL) 8371 m_tag_delete(m, mtag); 8372 } 8373 8374 static int 8375 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8376 struct pf_krule *r, struct mbuf **m0) 8377 { 8378 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8379 } 8380 8381 static int 8382 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8383 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8384 struct mbuf **m0) 8385 { 8386 NET_EPOCH_ASSERT(); 8387 8388 if (pd->act.dnpipe || pd->act.dnrpipe) { 8389 struct ip_fw_args dnflow; 8390 if (ip_dn_io_ptr == NULL) { 8391 m_freem(*m0); 8392 *m0 = NULL; 8393 return (ENOMEM); 8394 } 8395 8396 if (pd->pf_mtag == NULL && 8397 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8398 m_freem(*m0); 8399 *m0 = NULL; 8400 return (ENOMEM); 8401 } 8402 8403 if (ifp != NULL) { 8404 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8405 8406 pd->pf_mtag->if_index = ifp->if_index; 8407 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8408 8409 MPASS(sa != NULL); 8410 8411 switch (pd->af) { 8412 case AF_INET: 8413 memcpy(&pd->pf_mtag->dst, sa, 8414 sizeof(struct sockaddr_in)); 8415 break; 8416 case AF_INET6: 8417 memcpy(&pd->pf_mtag->dst, sa, 8418 sizeof(struct sockaddr_in6)); 8419 break; 8420 } 8421 } 8422 8423 if (s != NULL && s->nat_rule != NULL && 8424 s->nat_rule->action == PF_RDR && 8425 ( 8426 #ifdef INET 8427 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8428 #endif 8429 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8430 /* 8431 * If we're redirecting to loopback mark this packet 8432 * as being local. Otherwise it might get dropped 8433 * if dummynet re-injects. 8434 */ 8435 (*m0)->m_pkthdr.rcvif = V_loif; 8436 } 8437 8438 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8439 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8440 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8441 ip_dn_io_ptr(m0, &dnflow); 8442 if (*m0 != NULL) { 8443 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8444 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8445 } 8446 } 8447 } 8448 8449 return (0); 8450 } 8451 8452 #ifdef INET6 8453 static int 8454 pf_walk_option6(struct mbuf *m, int off, int end, uint32_t *jumbolen, 8455 u_short *reason) 8456 { 8457 struct ip6_opt opt; 8458 struct ip6_opt_jumbo jumbo; 8459 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 8460 8461 while (off < end) { 8462 if (!pf_pull_hdr(m, off, &opt.ip6o_type, sizeof(opt.ip6o_type), 8463 NULL, reason, AF_INET6)) { 8464 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt type")); 8465 return (PF_DROP); 8466 } 8467 if (opt.ip6o_type == IP6OPT_PAD1) { 8468 off++; 8469 continue; 8470 } 8471 if (!pf_pull_hdr(m, off, &opt, sizeof(opt), NULL, reason, 8472 AF_INET6)) { 8473 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short opt")); 8474 return (PF_DROP); 8475 } 8476 if (off + sizeof(opt) + opt.ip6o_len > end) { 8477 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 long opt")); 8478 REASON_SET(reason, PFRES_IPOPTIONS); 8479 return (PF_DROP); 8480 } 8481 switch (opt.ip6o_type) { 8482 case IP6OPT_JUMBO: 8483 if (*jumbolen != 0) { 8484 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple jumbo")); 8485 REASON_SET(reason, PFRES_IPOPTIONS); 8486 return (PF_DROP); 8487 } 8488 if (ntohs(h->ip6_plen) != 0) { 8489 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 bad jumbo plen")); 8490 REASON_SET(reason, PFRES_IPOPTIONS); 8491 return (PF_DROP); 8492 } 8493 if (!pf_pull_hdr(m, off, &jumbo, sizeof(jumbo), NULL, 8494 reason, AF_INET6)) { 8495 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbo")); 8496 return (PF_DROP); 8497 } 8498 memcpy(jumbolen, jumbo.ip6oj_jumbo_len, 8499 sizeof(*jumbolen)); 8500 *jumbolen = ntohl(*jumbolen); 8501 if (*jumbolen < IPV6_MAXPACKET) { 8502 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short jumbolen")); 8503 REASON_SET(reason, PFRES_IPOPTIONS); 8504 return (PF_DROP); 8505 } 8506 break; 8507 default: 8508 break; 8509 } 8510 off += sizeof(opt) + opt.ip6o_len; 8511 } 8512 8513 return (PF_PASS); 8514 } 8515 8516 int 8517 pf_walk_header6(struct mbuf *m, struct ip6_hdr *h, int *off, int *extoff, 8518 int *fragoff, uint8_t *nxt, uint32_t *jumbolen, u_short *reason) 8519 { 8520 struct ip6_frag frag; 8521 struct ip6_ext ext; 8522 struct ip6_rthdr rthdr; 8523 int rthdr_cnt = 0; 8524 8525 *off += sizeof(struct ip6_hdr); 8526 *extoff = *fragoff = 0; 8527 *nxt = h->ip6_nxt; 8528 *jumbolen = 0; 8529 for (;;) { 8530 switch (*nxt) { 8531 case IPPROTO_FRAGMENT: 8532 if (*fragoff != 0) { 8533 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple fragment")); 8534 REASON_SET(reason, PFRES_FRAG); 8535 return (PF_DROP); 8536 } 8537 /* jumbo payload packets cannot be fragmented */ 8538 if (*jumbolen != 0) { 8539 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 fragmented jumbo")); 8540 REASON_SET(reason, PFRES_FRAG); 8541 return (PF_DROP); 8542 } 8543 if (!pf_pull_hdr(m, *off, &frag, sizeof(frag), NULL, 8544 reason, AF_INET6)) { 8545 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short fragment")); 8546 return (PF_DROP); 8547 } 8548 *fragoff = *off; 8549 /* stop walking over non initial fragments */ 8550 if ((frag.ip6f_offlg & IP6F_OFF_MASK) != 0) 8551 return (PF_PASS); 8552 *off += sizeof(frag); 8553 *nxt = frag.ip6f_nxt; 8554 break; 8555 case IPPROTO_ROUTING: 8556 if (rthdr_cnt++) { 8557 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 multiple rthdr")); 8558 REASON_SET(reason, PFRES_IPOPTIONS); 8559 return (PF_DROP); 8560 } 8561 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), NULL, 8562 reason, AF_INET6)) { 8563 /* fragments may be short */ 8564 if (*fragoff != 0) { 8565 *off = *fragoff; 8566 *nxt = IPPROTO_FRAGMENT; 8567 return (PF_PASS); 8568 } 8569 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short rthdr")); 8570 return (PF_DROP); 8571 } 8572 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8573 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 rthdr0")); 8574 REASON_SET(reason, PFRES_IPOPTIONS); 8575 return (PF_DROP); 8576 } 8577 /* FALLTHROUGH */ 8578 case IPPROTO_AH: 8579 case IPPROTO_HOPOPTS: 8580 case IPPROTO_DSTOPTS: 8581 if (!pf_pull_hdr(m, *off, &ext, sizeof(ext), NULL, 8582 reason, AF_INET6)) { 8583 /* fragments may be short */ 8584 if (*fragoff != 0) { 8585 *off = *fragoff; 8586 *nxt = IPPROTO_FRAGMENT; 8587 return (PF_PASS); 8588 } 8589 DPFPRINTF(PF_DEBUG_MISC, ("IPv6 short exthdr")); 8590 return (PF_DROP); 8591 } 8592 /* reassembly needs the ext header before the frag */ 8593 if (*fragoff == 0) 8594 *extoff = *off; 8595 if (*nxt == IPPROTO_HOPOPTS && *fragoff == 0) { 8596 if (pf_walk_option6(m, *off + sizeof(ext), 8597 *off + (ext.ip6e_len + 1) * 8, jumbolen, 8598 reason) != PF_PASS) 8599 return (PF_DROP); 8600 if (ntohs(h->ip6_plen) == 0 && *jumbolen != 0) { 8601 DPFPRINTF(PF_DEBUG_MISC, 8602 ("IPv6 missing jumbo")); 8603 REASON_SET(reason, PFRES_IPOPTIONS); 8604 return (PF_DROP); 8605 } 8606 } 8607 if (*nxt == IPPROTO_AH) 8608 *off += (ext.ip6e_len + 2) * 4; 8609 else 8610 *off += (ext.ip6e_len + 1) * 8; 8611 *nxt = ext.ip6e_nxt; 8612 break; 8613 case IPPROTO_TCP: 8614 case IPPROTO_UDP: 8615 case IPPROTO_SCTP: 8616 case IPPROTO_ICMPV6: 8617 /* fragments may be short, ignore inner header then */ 8618 if (*fragoff != 0 && ntohs(h->ip6_plen) < *off + 8619 (*nxt == IPPROTO_TCP ? sizeof(struct tcphdr) : 8620 *nxt == IPPROTO_UDP ? sizeof(struct udphdr) : 8621 *nxt == IPPROTO_SCTP ? sizeof(struct sctphdr) : 8622 sizeof(struct icmp6_hdr))) { 8623 *off = *fragoff; 8624 *nxt = IPPROTO_FRAGMENT; 8625 } 8626 /* FALLTHROUGH */ 8627 default: 8628 return (PF_PASS); 8629 } 8630 } 8631 } 8632 #endif 8633 8634 static void 8635 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 8636 { 8637 memset(pd, 0, sizeof(*pd)); 8638 pd->pf_mtag = pf_find_mtag(m); 8639 pd->m = m; 8640 } 8641 8642 static int 8643 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 8644 u_short *action, u_short *reason, struct pfi_kkif *kif, 8645 struct pf_rule_actions *default_actions) 8646 { 8647 pd->af = af; 8648 pd->dir = dir; 8649 pd->kif = kif; 8650 pd->m = *m0; 8651 pd->sidx = (dir == PF_IN) ? 0 : 1; 8652 pd->didx = (dir == PF_IN) ? 1 : 0; 8653 8654 TAILQ_INIT(&pd->sctp_multihome_jobs); 8655 if (default_actions != NULL) 8656 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8657 8658 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8659 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8660 pd->act.flags = pd->pf_mtag->dnflags; 8661 } 8662 8663 switch (af) { 8664 #ifdef INET 8665 case AF_INET: { 8666 struct ip *h; 8667 8668 if (__predict_false((*m0)->m_len < sizeof(struct ip)) && 8669 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8670 DPFPRINTF(PF_DEBUG_URGENT, 8671 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8672 *action = PF_DROP; 8673 REASON_SET(reason, PFRES_SHORT); 8674 return (-1); 8675 } 8676 8677 if (pf_normalize_ip(m0, reason, pd) != PF_PASS) { 8678 /* We do IP header normalization and packet reassembly here */ 8679 *action = PF_DROP; 8680 return (-1); 8681 } 8682 pd->m = *m0; 8683 8684 h = mtod(pd->m, struct ip *); 8685 pd->off = h->ip_hl << 2; 8686 if (pd->off < (int)sizeof(*h)) { 8687 *action = PF_DROP; 8688 REASON_SET(reason, PFRES_SHORT); 8689 return (-1); 8690 } 8691 pd->src = (struct pf_addr *)&h->ip_src; 8692 pd->dst = (struct pf_addr *)&h->ip_dst; 8693 pd->ip_sum = &h->ip_sum; 8694 pd->proto_sum = NULL; 8695 pd->virtual_proto = pd->proto = h->ip_p; 8696 pd->tos = h->ip_tos; 8697 pd->ttl = h->ip_ttl; 8698 pd->tot_len = ntohs(h->ip_len); 8699 pd->act.rtableid = -1; 8700 8701 if (h->ip_hl > 5) /* has options */ 8702 pd->badopts++; 8703 8704 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 8705 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8706 8707 break; 8708 } 8709 #endif 8710 #ifdef INET6 8711 case AF_INET6: { 8712 struct ip6_hdr *h; 8713 int fragoff; 8714 uint32_t jumbolen; 8715 uint8_t nxt; 8716 8717 if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) && 8718 (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8719 DPFPRINTF(PF_DEBUG_URGENT, 8720 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8721 ", pullup failed\n")); 8722 *action = PF_DROP; 8723 REASON_SET(reason, PFRES_SHORT); 8724 return (-1); 8725 } 8726 8727 h = mtod(pd->m, struct ip6_hdr *); 8728 pd->off = 0; 8729 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8730 &jumbolen, reason) != PF_PASS) { 8731 *action = PF_DROP; 8732 return (-1); 8733 } 8734 8735 h = mtod(pd->m, struct ip6_hdr *); 8736 pd->src = (struct pf_addr *)&h->ip6_src; 8737 pd->dst = (struct pf_addr *)&h->ip6_dst; 8738 pd->ip_sum = NULL; 8739 pd->proto_sum = NULL; 8740 pd->tos = IPV6_DSCP(h); 8741 pd->ttl = h->ip6_hlim; 8742 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8743 pd->virtual_proto = pd->proto = h->ip6_nxt; 8744 pd->act.rtableid = -1; 8745 8746 if (fragoff != 0) 8747 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8748 8749 /* 8750 * we do not support jumbogram. if we keep going, zero ip6_plen 8751 * will do something bad, so drop the packet for now. 8752 */ 8753 if (htons(h->ip6_plen) == 0) { 8754 *action = PF_DROP; 8755 return (-1); 8756 } 8757 8758 /* We do IP header normalization and packet reassembly here */ 8759 if (pf_normalize_ip6(m0, fragoff, reason, pd) != 8760 PF_PASS) { 8761 *action = PF_DROP; 8762 return (-1); 8763 } 8764 pd->m = *m0; 8765 if (pd->m == NULL) { 8766 /* packet sits in reassembly queue, no error */ 8767 *action = PF_PASS; 8768 return (-1); 8769 } 8770 8771 /* Update pointers into the packet. */ 8772 h = mtod(pd->m, struct ip6_hdr *); 8773 pd->src = (struct pf_addr *)&h->ip6_src; 8774 pd->dst = (struct pf_addr *)&h->ip6_dst; 8775 8776 /* 8777 * Reassembly may have changed the next protocol from fragment 8778 * to something else, so update. 8779 */ 8780 pd->virtual_proto = pd->proto = h->ip6_nxt; 8781 pd->off = 0; 8782 8783 if (pf_walk_header6(pd->m, h, &pd->off, &pd->extoff, &fragoff, &nxt, 8784 &jumbolen, reason) != PF_PASS) { 8785 *action = PF_DROP; 8786 return (-1); 8787 } 8788 8789 if (fragoff != 0) 8790 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8791 8792 break; 8793 } 8794 #endif 8795 default: 8796 panic("pf_setup_pdesc called with illegal af %u", af); 8797 } 8798 8799 switch (pd->virtual_proto) { 8800 case IPPROTO_TCP: { 8801 struct tcphdr *th = &pd->hdr.tcp; 8802 8803 if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action, 8804 reason, af)) { 8805 *action = PF_DROP; 8806 REASON_SET(reason, PFRES_SHORT); 8807 return (-1); 8808 } 8809 pd->hdrlen = sizeof(*th); 8810 pd->p_len = pd->tot_len - pd->off - (th->th_off << 2); 8811 pd->sport = &th->th_sport; 8812 pd->dport = &th->th_dport; 8813 break; 8814 } 8815 case IPPROTO_UDP: { 8816 struct udphdr *uh = &pd->hdr.udp; 8817 8818 if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action, 8819 reason, af)) { 8820 *action = PF_DROP; 8821 REASON_SET(reason, PFRES_SHORT); 8822 return (-1); 8823 } 8824 pd->hdrlen = sizeof(*uh); 8825 if (uh->uh_dport == 0 || 8826 ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off || 8827 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8828 *action = PF_DROP; 8829 REASON_SET(reason, PFRES_SHORT); 8830 return (-1); 8831 } 8832 pd->sport = &uh->uh_sport; 8833 pd->dport = &uh->uh_dport; 8834 break; 8835 } 8836 case IPPROTO_SCTP: { 8837 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8838 action, reason, af)) { 8839 *action = PF_DROP; 8840 REASON_SET(reason, PFRES_SHORT); 8841 return (-1); 8842 } 8843 pd->hdrlen = sizeof(pd->hdr.sctp); 8844 pd->p_len = pd->tot_len - pd->off; 8845 8846 pd->sport = &pd->hdr.sctp.src_port; 8847 pd->dport = &pd->hdr.sctp.dest_port; 8848 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8849 *action = PF_DROP; 8850 REASON_SET(reason, PFRES_SHORT); 8851 return (-1); 8852 } 8853 if (pf_scan_sctp(pd) != PF_PASS) { 8854 *action = PF_DROP; 8855 REASON_SET(reason, PFRES_SHORT); 8856 return (-1); 8857 } 8858 break; 8859 } 8860 case IPPROTO_ICMP: { 8861 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN, 8862 action, reason, af)) { 8863 *action = PF_DROP; 8864 REASON_SET(reason, PFRES_SHORT); 8865 return (-1); 8866 } 8867 pd->hdrlen = ICMP_MINLEN; 8868 break; 8869 } 8870 #ifdef INET6 8871 case IPPROTO_ICMPV6: { 8872 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8873 8874 if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8875 action, reason, af)) { 8876 *action = PF_DROP; 8877 REASON_SET(reason, PFRES_SHORT); 8878 return (-1); 8879 } 8880 /* ICMP headers we look further into to match state */ 8881 switch (pd->hdr.icmp6.icmp6_type) { 8882 case MLD_LISTENER_QUERY: 8883 case MLD_LISTENER_REPORT: 8884 icmp_hlen = sizeof(struct mld_hdr); 8885 break; 8886 case ND_NEIGHBOR_SOLICIT: 8887 case ND_NEIGHBOR_ADVERT: 8888 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8889 break; 8890 } 8891 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8892 !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen, 8893 action, reason, af)) { 8894 *action = PF_DROP; 8895 REASON_SET(reason, PFRES_SHORT); 8896 return (-1); 8897 } 8898 pd->hdrlen = icmp_hlen; 8899 break; 8900 } 8901 #endif 8902 } 8903 return (0); 8904 } 8905 8906 static void 8907 pf_counters_inc(int action, struct pf_pdesc *pd, 8908 struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a) 8909 { 8910 struct pf_krule *tr; 8911 int dir = pd->dir; 8912 int dirndx; 8913 8914 pf_counter_u64_critical_enter(); 8915 pf_counter_u64_add_protected( 8916 &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8917 pd->tot_len); 8918 pf_counter_u64_add_protected( 8919 &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8920 1); 8921 8922 if (action == PF_PASS || r->action == PF_DROP) { 8923 dirndx = (dir == PF_OUT); 8924 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8925 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8926 pf_update_timestamp(r); 8927 8928 if (a != NULL) { 8929 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8930 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8931 } 8932 if (s != NULL) { 8933 struct pf_krule_item *ri; 8934 8935 if (s->nat_rule != NULL) { 8936 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 8937 1); 8938 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 8939 pd->tot_len); 8940 } 8941 if (s->src_node != NULL) { 8942 counter_u64_add(s->src_node->packets[dirndx], 8943 1); 8944 counter_u64_add(s->src_node->bytes[dirndx], 8945 pd->tot_len); 8946 } 8947 if (s->nat_src_node != NULL) { 8948 counter_u64_add(s->nat_src_node->packets[dirndx], 8949 1); 8950 counter_u64_add(s->nat_src_node->bytes[dirndx], 8951 pd->tot_len); 8952 } 8953 dirndx = (dir == s->direction) ? 0 : 1; 8954 s->packets[dirndx]++; 8955 s->bytes[dirndx] += pd->tot_len; 8956 8957 SLIST_FOREACH(ri, &s->match_rules, entry) { 8958 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8959 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8960 } 8961 } 8962 8963 tr = r; 8964 if (s != NULL && s->nat_rule != NULL && 8965 r == &V_pf_default_rule) 8966 tr = s->nat_rule; 8967 8968 if (tr->src.addr.type == PF_ADDR_TABLE) 8969 pfr_update_stats(tr->src.addr.p.tbl, 8970 (s == NULL) ? pd->src : 8971 &s->key[(s->direction == PF_IN)]-> 8972 addr[(s->direction == PF_OUT)], 8973 pd->af, pd->tot_len, dir == PF_OUT, 8974 r->action == PF_PASS, tr->src.neg); 8975 if (tr->dst.addr.type == PF_ADDR_TABLE) 8976 pfr_update_stats(tr->dst.addr.p.tbl, 8977 (s == NULL) ? pd->dst : 8978 &s->key[(s->direction == PF_IN)]-> 8979 addr[(s->direction == PF_IN)], 8980 pd->af, pd->tot_len, dir == PF_OUT, 8981 r->action == PF_PASS, tr->dst.neg); 8982 } 8983 pf_counter_u64_critical_exit(); 8984 } 8985 8986 #if defined(INET) || defined(INET6) 8987 int 8988 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8989 struct inpcb *inp, struct pf_rule_actions *default_actions) 8990 { 8991 struct pfi_kkif *kif; 8992 u_short action, reason = 0; 8993 struct m_tag *mtag; 8994 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8995 struct pf_kstate *s = NULL; 8996 struct pf_kruleset *ruleset = NULL; 8997 struct pf_pdesc pd; 8998 int use_2nd_queue = 0; 8999 uint16_t tag; 9000 uint8_t rt; 9001 9002 PF_RULES_RLOCK_TRACKER; 9003 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 9004 M_ASSERTPKTHDR(*m0); 9005 9006 if (!V_pf_status.running) 9007 return (PF_PASS); 9008 9009 PF_RULES_RLOCK(); 9010 9011 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9012 9013 if (__predict_false(kif == NULL)) { 9014 DPFPRINTF(PF_DEBUG_URGENT, 9015 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 9016 PF_RULES_RUNLOCK(); 9017 return (PF_DROP); 9018 } 9019 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 9020 PF_RULES_RUNLOCK(); 9021 return (PF_PASS); 9022 } 9023 9024 if ((*m0)->m_flags & M_SKIP_FIREWALL) { 9025 PF_RULES_RUNLOCK(); 9026 return (PF_PASS); 9027 } 9028 9029 #ifdef INET6 9030 /* 9031 * If we end up changing IP addresses (e.g. binat) the stack may get 9032 * confused and fail to send the icmp6 packet too big error. Just send 9033 * it here, before we do any NAT. 9034 */ 9035 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 9036 IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) { 9037 PF_RULES_RUNLOCK(); 9038 icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 9039 *m0 = NULL; 9040 return (PF_DROP); 9041 } 9042 #endif 9043 9044 if (__predict_false(! M_WRITABLE(*m0))) { 9045 *m0 = m_unshare(*m0, M_NOWAIT); 9046 if (*m0 == NULL) { 9047 PF_RULES_RUNLOCK(); 9048 return (PF_DROP); 9049 } 9050 } 9051 9052 pf_init_pdesc(&pd, *m0); 9053 9054 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9055 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9056 9057 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9058 pd.pf_mtag->if_idxgen); 9059 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9060 PF_RULES_RUNLOCK(); 9061 m_freem(*m0); 9062 *m0 = NULL; 9063 return (PF_PASS); 9064 } 9065 PF_RULES_RUNLOCK(); 9066 (ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL); 9067 *m0 = NULL; 9068 return (PF_PASS); 9069 } 9070 9071 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9072 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9073 /* Dummynet re-injects packets after they've 9074 * completed their delay. We've already 9075 * processed them, so pass unconditionally. */ 9076 9077 /* But only once. We may see the packet multiple times (e.g. 9078 * PFIL_IN/PFIL_OUT). */ 9079 pf_dummynet_flag_remove(pd.m, pd.pf_mtag); 9080 PF_RULES_RUNLOCK(); 9081 9082 return (PF_PASS); 9083 } 9084 9085 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, 9086 kif, default_actions) == -1) { 9087 if (action != PF_PASS) 9088 pd.act.log |= PF_LOG_FORCE; 9089 goto done; 9090 } 9091 9092 if (__predict_false(ip_divert_ptr != NULL) && 9093 ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 9094 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 9095 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 9096 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 9097 if (pd.pf_mtag == NULL && 9098 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9099 action = PF_DROP; 9100 goto done; 9101 } 9102 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 9103 } 9104 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 9105 pd.m->m_flags |= M_FASTFWD_OURS; 9106 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9107 } 9108 m_tag_delete(pd.m, mtag); 9109 9110 mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL); 9111 if (mtag != NULL) 9112 m_tag_delete(pd.m, mtag); 9113 } 9114 9115 switch (pd.virtual_proto) { 9116 case PF_VPROTO_FRAGMENT: 9117 /* 9118 * handle fragments that aren't reassembled by 9119 * normalization 9120 */ 9121 if (kif == NULL || r == NULL) /* pflog */ 9122 action = PF_DROP; 9123 else 9124 action = pf_test_rule(&r, &s, &pd, &a, 9125 &ruleset, inp); 9126 if (action != PF_PASS) 9127 REASON_SET(&reason, PFRES_FRAG); 9128 break; 9129 9130 case IPPROTO_TCP: { 9131 /* Respond to SYN with a syncookie. */ 9132 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9133 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9134 pf_syncookie_send(&pd); 9135 action = PF_DROP; 9136 break; 9137 } 9138 9139 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9140 use_2nd_queue = 1; 9141 action = pf_normalize_tcp(&pd); 9142 if (action == PF_DROP) 9143 goto done; 9144 action = pf_test_state_tcp(&s, &pd, &reason); 9145 if (action == PF_PASS) { 9146 if (V_pfsync_update_state_ptr != NULL) 9147 V_pfsync_update_state_ptr(s); 9148 r = s->rule; 9149 a = s->anchor; 9150 } else if (s == NULL) { 9151 /* Validate remote SYN|ACK, re-create original SYN if 9152 * valid. */ 9153 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9154 TH_ACK && pf_syncookie_validate(&pd) && 9155 pd.dir == PF_IN) { 9156 struct mbuf *msyn; 9157 9158 msyn = pf_syncookie_recreate_syn(&pd); 9159 if (msyn == NULL) { 9160 action = PF_DROP; 9161 break; 9162 } 9163 9164 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 9165 &pd.act); 9166 m_freem(msyn); 9167 if (action != PF_PASS) 9168 break; 9169 9170 action = pf_test_state_tcp(&s, &pd, &reason); 9171 if (action != PF_PASS || s == NULL) { 9172 action = PF_DROP; 9173 break; 9174 } 9175 9176 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9177 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9178 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9179 action = pf_synproxy(&pd, &s, &reason); 9180 break; 9181 } else { 9182 action = pf_test_rule(&r, &s, &pd, 9183 &a, &ruleset, inp); 9184 } 9185 } 9186 break; 9187 } 9188 9189 case IPPROTO_UDP: { 9190 action = pf_test_state_udp(&s, &pd); 9191 if (action == PF_PASS) { 9192 if (V_pfsync_update_state_ptr != NULL) 9193 V_pfsync_update_state_ptr(s); 9194 r = s->rule; 9195 a = s->anchor; 9196 } else if (s == NULL) 9197 action = pf_test_rule(&r, &s, &pd, 9198 &a, &ruleset, inp); 9199 break; 9200 } 9201 9202 case IPPROTO_SCTP: { 9203 action = pf_normalize_sctp(&pd); 9204 if (action == PF_DROP) 9205 goto done; 9206 action = pf_test_state_sctp(&s, &pd, &reason); 9207 if (action == PF_PASS) { 9208 if (V_pfsync_update_state_ptr != NULL) 9209 V_pfsync_update_state_ptr(s); 9210 r = s->rule; 9211 a = s->anchor; 9212 } else if (s == NULL) { 9213 action = pf_test_rule(&r, &s, 9214 &pd, &a, &ruleset, inp); 9215 } 9216 break; 9217 } 9218 9219 case IPPROTO_ICMP: { 9220 if (af != AF_INET) { 9221 action = PF_DROP; 9222 REASON_SET(&reason, PFRES_NORM); 9223 DPFPRINTF(PF_DEBUG_MISC, 9224 ("dropping IPv6 packet with ICMPv4 payload")); 9225 goto done; 9226 } 9227 action = pf_test_state_icmp(&s, &pd, &reason); 9228 if (action == PF_PASS) { 9229 if (V_pfsync_update_state_ptr != NULL) 9230 V_pfsync_update_state_ptr(s); 9231 r = s->rule; 9232 a = s->anchor; 9233 } else if (s == NULL) 9234 action = pf_test_rule(&r, &s, &pd, 9235 &a, &ruleset, inp); 9236 break; 9237 } 9238 9239 case IPPROTO_ICMPV6: { 9240 if (af != AF_INET6) { 9241 action = PF_DROP; 9242 REASON_SET(&reason, PFRES_NORM); 9243 DPFPRINTF(PF_DEBUG_MISC, 9244 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9245 goto done; 9246 } 9247 action = pf_test_state_icmp(&s, &pd, &reason); 9248 if (action == PF_PASS) { 9249 if (V_pfsync_update_state_ptr != NULL) 9250 V_pfsync_update_state_ptr(s); 9251 r = s->rule; 9252 a = s->anchor; 9253 } else if (s == NULL) 9254 action = pf_test_rule(&r, &s, &pd, 9255 &a, &ruleset, inp); 9256 break; 9257 } 9258 9259 default: 9260 action = pf_test_state_other(&s, &pd); 9261 if (action == PF_PASS) { 9262 if (V_pfsync_update_state_ptr != NULL) 9263 V_pfsync_update_state_ptr(s); 9264 r = s->rule; 9265 a = s->anchor; 9266 } else if (s == NULL) 9267 action = pf_test_rule(&r, &s, &pd, 9268 &a, &ruleset, inp); 9269 break; 9270 } 9271 9272 done: 9273 PF_RULES_RUNLOCK(); 9274 9275 if (pd.m == NULL) 9276 goto eat_pkt; 9277 9278 if (action == PF_PASS && pd.badopts && 9279 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9280 action = PF_DROP; 9281 REASON_SET(&reason, PFRES_IPOPTIONS); 9282 pd.act.log = PF_LOG_FORCE; 9283 DPFPRINTF(PF_DEBUG_MISC, 9284 ("pf: dropping packet with dangerous headers\n")); 9285 } 9286 9287 if (s) { 9288 uint8_t log = pd.act.log; 9289 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9290 pd.act.log |= log; 9291 tag = s->tag; 9292 rt = s->rt; 9293 } else { 9294 tag = r->tag; 9295 rt = r->rt; 9296 } 9297 9298 if (tag > 0 && pf_tag_packet(&pd, tag)) { 9299 action = PF_DROP; 9300 REASON_SET(&reason, PFRES_MEMORY); 9301 } 9302 9303 pf_scrub(&pd); 9304 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9305 pf_normalize_mss(&pd); 9306 9307 if (pd.act.rtableid >= 0) 9308 M_SETFIB(pd.m, pd.act.rtableid); 9309 9310 if (pd.act.flags & PFSTATE_SETPRIO) { 9311 if (pd.tos & IPTOS_LOWDELAY) 9312 use_2nd_queue = 1; 9313 if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) { 9314 action = PF_DROP; 9315 REASON_SET(&reason, PFRES_MEMORY); 9316 pd.act.log = PF_LOG_FORCE; 9317 DPFPRINTF(PF_DEBUG_MISC, 9318 ("pf: failed to allocate 802.1q mtag\n")); 9319 } 9320 } 9321 9322 #ifdef ALTQ 9323 if (action == PF_PASS && pd.act.qid) { 9324 if (pd.pf_mtag == NULL && 9325 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9326 action = PF_DROP; 9327 REASON_SET(&reason, PFRES_MEMORY); 9328 } else { 9329 if (s != NULL) 9330 pd.pf_mtag->qid_hash = pf_state_hash(s); 9331 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9332 pd.pf_mtag->qid = pd.act.pqid; 9333 else 9334 pd.pf_mtag->qid = pd.act.qid; 9335 /* Add hints for ecn. */ 9336 pd.pf_mtag->hdr = mtod(pd.m, void *); 9337 } 9338 } 9339 #endif /* ALTQ */ 9340 9341 /* 9342 * connections redirected to loopback should not match sockets 9343 * bound specifically to loopback due to security implications, 9344 * see tcp_input() and in_pcblookup_listen(). 9345 */ 9346 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9347 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 9348 (s->nat_rule->action == PF_RDR || 9349 s->nat_rule->action == PF_BINAT) && 9350 pf_is_loopback(af, pd.dst)) 9351 pd.m->m_flags |= M_SKIP_FIREWALL; 9352 9353 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 9354 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 9355 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9356 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9357 if (mtag != NULL) { 9358 ((struct pf_divert_mtag *)(mtag+1))->port = 9359 ntohs(r->divert.port); 9360 ((struct pf_divert_mtag *)(mtag+1))->idir = 9361 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9362 PF_DIVERT_MTAG_DIR_OUT; 9363 9364 if (s) 9365 PF_STATE_UNLOCK(s); 9366 9367 m_tag_prepend(pd.m, mtag); 9368 if (pd.m->m_flags & M_FASTFWD_OURS) { 9369 if (pd.pf_mtag == NULL && 9370 ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) { 9371 action = PF_DROP; 9372 REASON_SET(&reason, PFRES_MEMORY); 9373 pd.act.log = PF_LOG_FORCE; 9374 DPFPRINTF(PF_DEBUG_MISC, 9375 ("pf: failed to allocate tag\n")); 9376 } else { 9377 pd.pf_mtag->flags |= 9378 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9379 pd.m->m_flags &= ~M_FASTFWD_OURS; 9380 } 9381 } 9382 ip_divert_ptr(*m0, dir == PF_IN); 9383 *m0 = NULL; 9384 9385 return (action); 9386 } else { 9387 /* XXX: ipfw has the same behaviour! */ 9388 action = PF_DROP; 9389 REASON_SET(&reason, PFRES_MEMORY); 9390 pd.act.log = PF_LOG_FORCE; 9391 DPFPRINTF(PF_DEBUG_MISC, 9392 ("pf: failed to allocate divert tag\n")); 9393 } 9394 } 9395 /* XXX: Anybody working on it?! */ 9396 if (af == AF_INET6 && r->divert.port) 9397 printf("pf: divert(9) is not supported for IPv6\n"); 9398 9399 /* this flag will need revising if the pkt is forwarded */ 9400 if (pd.pf_mtag) 9401 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9402 9403 if (pd.act.log) { 9404 struct pf_krule *lr; 9405 struct pf_krule_item *ri; 9406 9407 if (s != NULL && s->nat_rule != NULL && 9408 s->nat_rule->log & PF_LOG_ALL) 9409 lr = s->nat_rule; 9410 else 9411 lr = r; 9412 9413 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9414 PFLOG_PACKET(action, reason, lr, a, 9415 ruleset, &pd, (s == NULL)); 9416 if (s) { 9417 SLIST_FOREACH(ri, &s->match_rules, entry) 9418 if (ri->r->log & PF_LOG_ALL) 9419 PFLOG_PACKET(action, 9420 reason, ri->r, a, ruleset, &pd, 0); 9421 } 9422 } 9423 9424 pf_counters_inc(action, &pd, s, r, a); 9425 9426 switch (action) { 9427 case PF_SYNPROXY_DROP: 9428 m_freem(*m0); 9429 case PF_DEFER: 9430 *m0 = NULL; 9431 action = PF_PASS; 9432 break; 9433 case PF_DROP: 9434 m_freem(*m0); 9435 *m0 = NULL; 9436 break; 9437 default: 9438 if (rt) { 9439 switch (af) { 9440 #ifdef INET 9441 case AF_INET: 9442 /* pf_route() returns unlocked. */ 9443 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9444 break; 9445 #endif 9446 #ifdef INET6 9447 case AF_INET6: 9448 /* pf_route6() returns unlocked. */ 9449 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9450 break; 9451 #endif 9452 } 9453 goto out; 9454 } 9455 if (pf_dummynet(&pd, s, r, m0) != 0) { 9456 action = PF_DROP; 9457 REASON_SET(&reason, PFRES_MEMORY); 9458 } 9459 break; 9460 } 9461 9462 eat_pkt: 9463 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9464 9465 if (s && action != PF_DROP) { 9466 if (!s->if_index_in && dir == PF_IN) 9467 s->if_index_in = ifp->if_index; 9468 else if (!s->if_index_out && dir == PF_OUT) 9469 s->if_index_out = ifp->if_index; 9470 } 9471 9472 if (s) 9473 PF_STATE_UNLOCK(s); 9474 9475 #ifdef INET6 9476 /* If reassembled packet passed, create new fragments. */ 9477 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 9478 (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9479 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9480 #endif 9481 9482 out: 9483 pf_sctp_multihome_delayed(&pd, kif, s, action); 9484 9485 return (action); 9486 } 9487 #endif /* INET || INET6 */ 9488