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, sctp, multihome, test, "struct pfi_kkif *", 130 "struct pf_krule *", "struct mbuf *", "int"); 131 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 132 "struct pf_sctp_source *"); 133 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 134 "struct pf_kstate *", "struct pf_sctp_source *"); 135 136 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 137 "struct mbuf *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 139 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 140 "int", "struct pf_keth_rule *", "char *"); 141 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 142 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 143 "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 145 146 /* 147 * Global variables 148 */ 149 150 /* state tables */ 151 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 152 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 153 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 154 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 157 VNET_DEFINE(struct pf_kstatus, pf_status); 158 159 VNET_DEFINE(u_int32_t, ticket_altqs_active); 160 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 161 VNET_DEFINE(int, altqs_inactive_open); 162 VNET_DEFINE(u_int32_t, ticket_pabuf); 163 164 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 165 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 166 VNET_DEFINE(u_char, pf_tcp_secret[16]); 167 #define V_pf_tcp_secret VNET(pf_tcp_secret) 168 VNET_DEFINE(int, pf_tcp_secret_init); 169 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 170 VNET_DEFINE(int, pf_tcp_iss_off); 171 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 172 VNET_DECLARE(int, pf_vnet_active); 173 #define V_pf_vnet_active VNET(pf_vnet_active) 174 175 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 176 #define V_pf_purge_idx VNET(pf_purge_idx) 177 178 #ifdef PF_WANT_32_TO_64_COUNTER 179 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 180 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 181 182 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 183 VNET_DEFINE(size_t, pf_allrulecount); 184 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 185 #endif 186 187 struct pf_sctp_endpoint; 188 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 189 struct pf_sctp_source { 190 sa_family_t af; 191 struct pf_addr addr; 192 TAILQ_ENTRY(pf_sctp_source) entry; 193 }; 194 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 195 struct pf_sctp_endpoint 196 { 197 uint32_t v_tag; 198 struct pf_sctp_sources sources; 199 RB_ENTRY(pf_sctp_endpoint) entry; 200 }; 201 static int 202 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 203 { 204 return (a->v_tag - b->v_tag); 205 } 206 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 207 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 208 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 209 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 210 static struct mtx_padalign pf_sctp_endpoints_mtx; 211 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 212 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 213 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 214 215 /* 216 * Queue for pf_intr() sends. 217 */ 218 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 219 struct pf_send_entry { 220 STAILQ_ENTRY(pf_send_entry) pfse_next; 221 struct mbuf *pfse_m; 222 enum { 223 PFSE_IP, 224 PFSE_IP6, 225 PFSE_ICMP, 226 PFSE_ICMP6, 227 } pfse_type; 228 struct { 229 int type; 230 int code; 231 int mtu; 232 } icmpopts; 233 }; 234 235 STAILQ_HEAD(pf_send_head, pf_send_entry); 236 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 237 #define V_pf_sendqueue VNET(pf_sendqueue) 238 239 static struct mtx_padalign pf_sendqueue_mtx; 240 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 241 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 242 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 243 244 /* 245 * Queue for pf_overload_task() tasks. 246 */ 247 struct pf_overload_entry { 248 SLIST_ENTRY(pf_overload_entry) next; 249 struct pf_addr addr; 250 sa_family_t af; 251 uint8_t dir; 252 struct pf_krule *rule; 253 }; 254 255 SLIST_HEAD(pf_overload_head, pf_overload_entry); 256 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 257 #define V_pf_overloadqueue VNET(pf_overloadqueue) 258 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 259 #define V_pf_overloadtask VNET(pf_overloadtask) 260 261 static struct mtx_padalign pf_overloadqueue_mtx; 262 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 263 "pf overload/flush queue", MTX_DEF); 264 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 265 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 266 267 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 268 struct mtx_padalign pf_unlnkdrules_mtx; 269 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 270 MTX_DEF); 271 272 struct sx pf_config_lock; 273 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 274 275 struct mtx_padalign pf_table_stats_lock; 276 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 277 MTX_DEF); 278 279 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 280 #define V_pf_sources_z VNET(pf_sources_z) 281 uma_zone_t pf_mtag_z; 282 VNET_DEFINE(uma_zone_t, pf_state_z); 283 VNET_DEFINE(uma_zone_t, pf_state_key_z); 284 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 285 286 VNET_DEFINE(struct unrhdr64, pf_stateid); 287 288 static void pf_src_tree_remove_state(struct pf_kstate *); 289 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 290 u_int32_t); 291 static void pf_add_threshold(struct pf_threshold *); 292 static int pf_check_threshold(struct pf_threshold *); 293 294 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 295 u_int16_t *, u_int16_t *, struct pf_addr *, 296 u_int16_t, u_int8_t, sa_family_t); 297 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 298 struct tcphdr *, struct pf_state_peer *); 299 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 300 int *, u_int16_t *, u_int16_t *); 301 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 302 struct pf_addr *, struct pf_addr *, u_int16_t, 303 u_int16_t *, u_int16_t *, u_int16_t *, 304 u_int16_t *, u_int8_t, sa_family_t); 305 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 306 sa_family_t, struct pf_krule *, int); 307 static void pf_detach_state(struct pf_kstate *); 308 static int pf_state_key_attach(struct pf_state_key *, 309 struct pf_state_key *, struct pf_kstate *); 310 static void pf_state_key_detach(struct pf_kstate *, int); 311 static int pf_state_key_ctor(void *, int, void *, int); 312 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 313 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 314 struct pf_mtag *pf_mtag); 315 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 316 struct pf_krule *, struct mbuf **); 317 static int pf_dummynet_route(struct pf_pdesc *, 318 struct pf_kstate *, struct pf_krule *, 319 struct ifnet *, struct sockaddr *, struct mbuf **); 320 static int pf_test_eth_rule(int, struct pfi_kkif *, 321 struct mbuf **); 322 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 323 struct pfi_kkif *, struct mbuf *, int, 324 struct pf_pdesc *, struct pf_krule **, 325 struct pf_kruleset **, struct inpcb *, int); 326 static int pf_create_state(struct pf_krule *, struct pf_krule *, 327 struct pf_krule *, struct pf_pdesc *, 328 struct pf_ksrc_node *, struct pf_state_key *, 329 struct pf_state_key *, struct mbuf *, int, 330 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 331 struct pf_kstate **, int, u_int16_t, u_int16_t, 332 int, struct pf_krule_slist *, struct pf_udp_mapping *); 333 static int pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *, 334 int, struct pf_state_key_cmp *, int, struct pf_addr *, 335 int, struct pf_addr *, int); 336 static int pf_tcp_track_full(struct pf_kstate **, 337 struct pfi_kkif *, struct mbuf *, int, 338 struct pf_pdesc *, u_short *, int *); 339 static int pf_tcp_track_sloppy(struct pf_kstate **, 340 struct pf_pdesc *, u_short *); 341 static int pf_test_state_tcp(struct pf_kstate **, 342 struct pfi_kkif *, struct mbuf *, int, 343 struct pf_pdesc *, u_short *); 344 static int pf_test_state_udp(struct pf_kstate **, 345 struct pfi_kkif *, struct mbuf *, int, 346 struct pf_pdesc *); 347 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 348 struct pf_pdesc *, struct pf_kstate **, struct mbuf *, 349 int, int, struct pfi_kkif *, u_int16_t, u_int16_t, 350 int, int *, int, int); 351 static int pf_test_state_icmp(struct pf_kstate **, 352 struct pfi_kkif *, struct mbuf *, int, 353 struct pf_pdesc *, u_short *); 354 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 355 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 356 struct pfi_kkif *, struct pf_kstate *, int); 357 static int pf_test_state_sctp(struct pf_kstate **, 358 struct pfi_kkif *, struct mbuf *, int, 359 struct pf_pdesc *, u_short *); 360 static int pf_test_state_other(struct pf_kstate **, 361 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 362 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 363 int, u_int16_t); 364 static int pf_check_proto_cksum(struct mbuf *, int, int, 365 u_int8_t, sa_family_t); 366 static void pf_print_state_parts(struct pf_kstate *, 367 struct pf_state_key *, struct pf_state_key *); 368 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 369 bool, u_int8_t); 370 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 371 const struct pf_state_key_cmp *, u_int); 372 static int pf_src_connlimit(struct pf_kstate **); 373 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 374 static void pf_counters_inc(int, 375 struct pf_pdesc *, struct pfi_kkif *, 376 struct pf_kstate *, struct pf_krule *, 377 struct pf_krule *); 378 static void pf_overload_task(void *v, int pending); 379 static u_short pf_insert_src_node(struct pf_ksrc_node **, 380 struct pf_krule *, struct pf_addr *, sa_family_t); 381 static u_int pf_purge_expired_states(u_int, int); 382 static void pf_purge_unlinked_rules(void); 383 static int pf_mtag_uminit(void *, int, int); 384 static void pf_mtag_free(struct m_tag *); 385 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 386 int, struct pf_state_key *); 387 #ifdef INET 388 static void pf_route(struct mbuf **, struct pf_krule *, 389 struct ifnet *, struct pf_kstate *, 390 struct pf_pdesc *, struct inpcb *); 391 #endif /* INET */ 392 #ifdef INET6 393 static void pf_change_a6(struct pf_addr *, u_int16_t *, 394 struct pf_addr *, u_int8_t); 395 static void pf_route6(struct mbuf **, struct pf_krule *, 396 struct ifnet *, struct pf_kstate *, 397 struct pf_pdesc *, struct inpcb *); 398 #endif /* INET6 */ 399 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 400 401 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 402 403 extern int pf_end_threads; 404 extern struct proc *pf_purge_proc; 405 406 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 407 408 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 409 410 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 411 do { \ 412 struct pf_state_key *nk; \ 413 if ((pd->dir) == PF_OUT) \ 414 nk = (_s)->key[PF_SK_STACK]; \ 415 else \ 416 nk = (_s)->key[PF_SK_WIRE]; \ 417 pf_packet_rework_nat(_m, _pd, _off, nk); \ 418 } while (0) 419 420 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 421 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 422 423 #define STATE_LOOKUP(i, k, s, pd) \ 424 do { \ 425 (s) = pf_find_state((i), (k), (pd->dir)); \ 426 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 427 if ((s) == NULL) \ 428 return (PF_DROP); \ 429 if (PACKET_LOOPED(pd)) \ 430 return (PF_PASS); \ 431 } while (0) 432 433 static struct pfi_kkif * 434 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 435 { 436 SDT_PROBE2(pf, ip, , bound_iface, st, k); 437 438 /* Floating unless otherwise specified. */ 439 if (! (st->rule->rule_flag & PFRULE_IFBOUND)) 440 return (V_pfi_all); 441 442 /* 443 * Initially set to all, because we don't know what interface we'll be 444 * sending this out when we create the state. 445 */ 446 if (st->rule->rt == PF_REPLYTO) 447 return (V_pfi_all); 448 449 /* Don't overrule the interface for states created on incoming packets. */ 450 if (st->direction == PF_IN) 451 return (k); 452 453 /* No route-to, so don't overrule. */ 454 if (st->rt != PF_ROUTETO) 455 return (k); 456 457 /* Bind to the route-to interface. */ 458 return (st->rt_kif); 459 } 460 461 #define STATE_INC_COUNTERS(s) \ 462 do { \ 463 struct pf_krule_item *mrm; \ 464 counter_u64_add(s->rule->states_cur, 1); \ 465 counter_u64_add(s->rule->states_tot, 1); \ 466 if (s->anchor != NULL) { \ 467 counter_u64_add(s->anchor->states_cur, 1); \ 468 counter_u64_add(s->anchor->states_tot, 1); \ 469 } \ 470 if (s->nat_rule != NULL) { \ 471 counter_u64_add(s->nat_rule->states_cur, 1);\ 472 counter_u64_add(s->nat_rule->states_tot, 1);\ 473 } \ 474 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 475 counter_u64_add(mrm->r->states_cur, 1); \ 476 counter_u64_add(mrm->r->states_tot, 1); \ 477 } \ 478 } while (0) 479 480 #define STATE_DEC_COUNTERS(s) \ 481 do { \ 482 struct pf_krule_item *mrm; \ 483 if (s->nat_rule != NULL) \ 484 counter_u64_add(s->nat_rule->states_cur, -1);\ 485 if (s->anchor != NULL) \ 486 counter_u64_add(s->anchor->states_cur, -1); \ 487 counter_u64_add(s->rule->states_cur, -1); \ 488 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 489 counter_u64_add(mrm->r->states_cur, -1); \ 490 } while (0) 491 492 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 493 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 494 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 495 VNET_DEFINE(struct pf_idhash *, pf_idhash); 496 VNET_DEFINE(struct pf_srchash *, pf_srchash); 497 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 498 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 499 500 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 501 "pf(4)"); 502 503 VNET_DEFINE(u_long, pf_hashmask); 504 VNET_DEFINE(u_long, pf_srchashmask); 505 VNET_DEFINE(u_long, pf_udpendpointhashmask); 506 VNET_DEFINE_STATIC(u_long, pf_hashsize); 507 #define V_pf_hashsize VNET(pf_hashsize) 508 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 509 #define V_pf_srchashsize VNET(pf_srchashsize) 510 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 511 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 512 u_long pf_ioctl_maxcount = 65535; 513 514 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 515 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 516 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 517 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 518 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 519 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 520 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 521 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 522 523 VNET_DEFINE(void *, pf_swi_cookie); 524 VNET_DEFINE(struct intr_event *, pf_swi_ie); 525 526 VNET_DEFINE(uint32_t, pf_hashseed); 527 #define V_pf_hashseed VNET(pf_hashseed) 528 529 static void 530 pf_sctp_checksum(struct mbuf *m, int off) 531 { 532 uint32_t sum = 0; 533 534 /* Zero out the checksum, to enable recalculation. */ 535 m_copyback(m, off + offsetof(struct sctphdr, checksum), 536 sizeof(sum), (caddr_t)&sum); 537 538 sum = sctp_calculate_cksum(m, off); 539 540 m_copyback(m, off + offsetof(struct sctphdr, checksum), 541 sizeof(sum), (caddr_t)&sum); 542 } 543 544 int 545 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 546 { 547 548 switch (af) { 549 #ifdef INET 550 case AF_INET: 551 if (a->addr32[0] > b->addr32[0]) 552 return (1); 553 if (a->addr32[0] < b->addr32[0]) 554 return (-1); 555 break; 556 #endif /* INET */ 557 #ifdef INET6 558 case AF_INET6: 559 if (a->addr32[3] > b->addr32[3]) 560 return (1); 561 if (a->addr32[3] < b->addr32[3]) 562 return (-1); 563 if (a->addr32[2] > b->addr32[2]) 564 return (1); 565 if (a->addr32[2] < b->addr32[2]) 566 return (-1); 567 if (a->addr32[1] > b->addr32[1]) 568 return (1); 569 if (a->addr32[1] < b->addr32[1]) 570 return (-1); 571 if (a->addr32[0] > b->addr32[0]) 572 return (1); 573 if (a->addr32[0] < b->addr32[0]) 574 return (-1); 575 break; 576 #endif /* INET6 */ 577 default: 578 panic("%s: unknown address family %u", __func__, af); 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 case AF_INET: 588 return IN_LOOPBACK(ntohl(addr->v4.s_addr)); 589 case AF_INET6: 590 return IN6_IS_ADDR_LOOPBACK(&addr->v6); 591 default: 592 panic("Unknown af %d", af); 593 } 594 } 595 596 static void 597 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 598 struct pf_state_key *nk) 599 { 600 601 switch (pd->proto) { 602 case IPPROTO_TCP: { 603 struct tcphdr *th = &pd->hdr.tcp; 604 605 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 606 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 607 &th->th_sum, &nk->addr[pd->sidx], 608 nk->port[pd->sidx], 0, pd->af); 609 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 610 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 611 &th->th_sum, &nk->addr[pd->didx], 612 nk->port[pd->didx], 0, pd->af); 613 m_copyback(m, off, sizeof(*th), (caddr_t)th); 614 break; 615 } 616 case IPPROTO_UDP: { 617 struct udphdr *uh = &pd->hdr.udp; 618 619 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 620 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 621 &uh->uh_sum, &nk->addr[pd->sidx], 622 nk->port[pd->sidx], 1, pd->af); 623 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 624 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 625 &uh->uh_sum, &nk->addr[pd->didx], 626 nk->port[pd->didx], 1, pd->af); 627 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 628 break; 629 } 630 case IPPROTO_SCTP: { 631 struct sctphdr *sh = &pd->hdr.sctp; 632 uint16_t checksum = 0; 633 634 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 635 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 636 &checksum, &nk->addr[pd->sidx], 637 nk->port[pd->sidx], 1, pd->af); 638 } 639 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 640 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 641 &checksum, &nk->addr[pd->didx], 642 nk->port[pd->didx], 1, pd->af); 643 } 644 645 break; 646 } 647 case IPPROTO_ICMP: { 648 struct icmp *ih = &pd->hdr.icmp; 649 650 if (nk->port[pd->sidx] != ih->icmp_id) { 651 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 652 ih->icmp_cksum, ih->icmp_id, 653 nk->port[pd->sidx], 0); 654 ih->icmp_id = nk->port[pd->sidx]; 655 pd->sport = &ih->icmp_id; 656 657 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 658 } 659 /* FALLTHROUGH */ 660 } 661 default: 662 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 663 switch (pd->af) { 664 case AF_INET: 665 pf_change_a(&pd->src->v4.s_addr, 666 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 667 0); 668 break; 669 case AF_INET6: 670 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 671 break; 672 } 673 } 674 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 675 switch (pd->af) { 676 case AF_INET: 677 pf_change_a(&pd->dst->v4.s_addr, 678 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 679 0); 680 break; 681 case AF_INET6: 682 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 683 break; 684 } 685 } 686 break; 687 } 688 } 689 690 static __inline uint32_t 691 pf_hashkey(const struct pf_state_key *sk) 692 { 693 uint32_t h; 694 695 h = murmur3_32_hash32((const uint32_t *)sk, 696 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 697 V_pf_hashseed); 698 699 return (h & V_pf_hashmask); 700 } 701 702 static __inline uint32_t 703 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 704 { 705 uint32_t h; 706 707 switch (af) { 708 case AF_INET: 709 h = murmur3_32_hash32((uint32_t *)&addr->v4, 710 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 711 break; 712 case AF_INET6: 713 h = murmur3_32_hash32((uint32_t *)&addr->v6, 714 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 715 break; 716 default: 717 panic("%s: unknown address family %u", __func__, af); 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, sk); 1445 if (idx == PF_SK_STACK) 1446 pf_detach_state(s); 1447 return (EEXIST); /* collision! */ 1448 } 1449 } 1450 PF_HASHROW_UNLOCK(ih); 1451 } 1452 uma_zfree(V_pf_state_key_z, sk); 1453 s->key[idx] = cur; 1454 } else { 1455 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1456 s->key[idx] = sk; 1457 } 1458 1459 stateattach: 1460 /* List is sorted, if-bound states before floating. */ 1461 if (s->kif == V_pfi_all) 1462 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1463 else 1464 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1465 1466 if (olds) { 1467 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1468 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1469 key_list[idx]); 1470 olds = NULL; 1471 } 1472 1473 /* 1474 * Attach done. See how should we (or should not?) 1475 * attach a second key. 1476 */ 1477 if (sks == skw) { 1478 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1479 idx = PF_SK_STACK; 1480 sks = NULL; 1481 goto stateattach; 1482 } else if (sks != NULL) { 1483 /* 1484 * Continue attaching with stack key. 1485 */ 1486 sk = sks; 1487 kh = khs; 1488 idx = PF_SK_STACK; 1489 sks = NULL; 1490 goto keyattach; 1491 } 1492 1493 PF_STATE_LOCK(s); 1494 KEYS_UNLOCK(); 1495 1496 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1497 ("%s failure", __func__)); 1498 1499 return (0); 1500 #undef KEYS_UNLOCK 1501 } 1502 1503 static void 1504 pf_detach_state(struct pf_kstate *s) 1505 { 1506 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1507 struct pf_keyhash *kh; 1508 1509 NET_EPOCH_ASSERT(); 1510 MPASS(s->timeout >= PFTM_MAX); 1511 1512 pf_sctp_multihome_detach_addr(s); 1513 1514 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1515 V_pflow_export_state_ptr(s); 1516 1517 if (sks != NULL) { 1518 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1519 PF_HASHROW_LOCK(kh); 1520 if (s->key[PF_SK_STACK] != NULL) 1521 pf_state_key_detach(s, PF_SK_STACK); 1522 /* 1523 * If both point to same key, then we are done. 1524 */ 1525 if (sks == s->key[PF_SK_WIRE]) { 1526 pf_state_key_detach(s, PF_SK_WIRE); 1527 PF_HASHROW_UNLOCK(kh); 1528 return; 1529 } 1530 PF_HASHROW_UNLOCK(kh); 1531 } 1532 1533 if (s->key[PF_SK_WIRE] != NULL) { 1534 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1535 PF_HASHROW_LOCK(kh); 1536 if (s->key[PF_SK_WIRE] != NULL) 1537 pf_state_key_detach(s, PF_SK_WIRE); 1538 PF_HASHROW_UNLOCK(kh); 1539 } 1540 } 1541 1542 static void 1543 pf_state_key_detach(struct pf_kstate *s, int idx) 1544 { 1545 struct pf_state_key *sk = s->key[idx]; 1546 #ifdef INVARIANTS 1547 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1548 1549 PF_HASHROW_ASSERT(kh); 1550 #endif 1551 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1552 s->key[idx] = NULL; 1553 1554 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1555 LIST_REMOVE(sk, entry); 1556 uma_zfree(V_pf_state_key_z, sk); 1557 } 1558 } 1559 1560 static int 1561 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1562 { 1563 struct pf_state_key *sk = mem; 1564 1565 bzero(sk, sizeof(struct pf_state_key_cmp)); 1566 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1567 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1568 1569 return (0); 1570 } 1571 1572 static int 1573 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1574 struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr, 1575 int didx, struct pf_addr *daddr, int multi) 1576 { 1577 #ifdef INET6 1578 struct nd_neighbor_solicit nd; 1579 struct pf_addr *target; 1580 u_short action, reason; 1581 1582 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1583 goto copy; 1584 1585 switch (pd->hdr.icmp6.icmp6_type) { 1586 case ND_NEIGHBOR_SOLICIT: 1587 if (multi) 1588 return (-1); 1589 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1590 return (-1); 1591 target = (struct pf_addr *)&nd.nd_ns_target; 1592 daddr = target; 1593 break; 1594 case ND_NEIGHBOR_ADVERT: 1595 if (multi) 1596 return (-1); 1597 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1598 return (-1); 1599 target = (struct pf_addr *)&nd.nd_ns_target; 1600 saddr = target; 1601 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1602 key->addr[didx].addr32[0] = 0; 1603 key->addr[didx].addr32[1] = 0; 1604 key->addr[didx].addr32[2] = 0; 1605 key->addr[didx].addr32[3] = 0; 1606 daddr = NULL; /* overwritten */ 1607 } 1608 break; 1609 default: 1610 if (multi == PF_ICMP_MULTI_LINK) { 1611 key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1612 key->addr[sidx].addr32[1] = 0; 1613 key->addr[sidx].addr32[2] = 0; 1614 key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1615 saddr = NULL; /* overwritten */ 1616 } 1617 } 1618 copy: 1619 #endif 1620 if (saddr) 1621 PF_ACPY(&key->addr[sidx], saddr, pd->af); 1622 if (daddr) 1623 PF_ACPY(&key->addr[didx], daddr, pd->af); 1624 1625 return (0); 1626 } 1627 1628 struct pf_state_key * 1629 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1630 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1631 u_int16_t dport) 1632 { 1633 struct pf_state_key *sk; 1634 1635 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1636 if (sk == NULL) 1637 return (NULL); 1638 1639 if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk, 1640 pd->sidx, pd->src, pd->didx, pd->dst, 0)) { 1641 uma_zfree(V_pf_state_key_z, sk); 1642 return (NULL); 1643 } 1644 1645 sk->port[pd->sidx] = sport; 1646 sk->port[pd->didx] = dport; 1647 sk->proto = pd->proto; 1648 sk->af = pd->af; 1649 1650 return (sk); 1651 } 1652 1653 struct pf_state_key * 1654 pf_state_key_clone(const struct pf_state_key *orig) 1655 { 1656 struct pf_state_key *sk; 1657 1658 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1659 if (sk == NULL) 1660 return (NULL); 1661 1662 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1663 1664 return (sk); 1665 } 1666 1667 int 1668 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1669 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1670 { 1671 struct pf_idhash *ih; 1672 struct pf_kstate *cur; 1673 int error; 1674 1675 NET_EPOCH_ASSERT(); 1676 1677 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1678 ("%s: sks not pristine", __func__)); 1679 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1680 ("%s: skw not pristine", __func__)); 1681 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1682 1683 s->kif = kif; 1684 s->orig_kif = orig_kif; 1685 1686 if (s->id == 0 && s->creatorid == 0) { 1687 s->id = alloc_unr64(&V_pf_stateid); 1688 s->id = htobe64(s->id); 1689 s->creatorid = V_pf_status.hostid; 1690 } 1691 1692 /* Returns with ID locked on success. */ 1693 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1694 return (error); 1695 1696 ih = &V_pf_idhash[PF_IDHASH(s)]; 1697 PF_HASHROW_ASSERT(ih); 1698 LIST_FOREACH(cur, &ih->states, entry) 1699 if (cur->id == s->id && cur->creatorid == s->creatorid) 1700 break; 1701 1702 if (cur != NULL) { 1703 s->timeout = PFTM_UNLINKED; 1704 PF_HASHROW_UNLOCK(ih); 1705 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1706 printf("pf: state ID collision: " 1707 "id: %016llx creatorid: %08x\n", 1708 (unsigned long long)be64toh(s->id), 1709 ntohl(s->creatorid)); 1710 } 1711 pf_detach_state(s); 1712 return (EEXIST); 1713 } 1714 LIST_INSERT_HEAD(&ih->states, s, entry); 1715 /* One for keys, one for ID hash. */ 1716 refcount_init(&s->refs, 2); 1717 1718 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1719 if (V_pfsync_insert_state_ptr != NULL) 1720 V_pfsync_insert_state_ptr(s); 1721 1722 /* Returns locked. */ 1723 return (0); 1724 } 1725 1726 /* 1727 * Find state by ID: returns with locked row on success. 1728 */ 1729 struct pf_kstate * 1730 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1731 { 1732 struct pf_idhash *ih; 1733 struct pf_kstate *s; 1734 1735 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1736 1737 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1738 1739 PF_HASHROW_LOCK(ih); 1740 LIST_FOREACH(s, &ih->states, entry) 1741 if (s->id == id && s->creatorid == creatorid) 1742 break; 1743 1744 if (s == NULL) 1745 PF_HASHROW_UNLOCK(ih); 1746 1747 return (s); 1748 } 1749 1750 /* 1751 * Find state by key. 1752 * Returns with ID hash slot locked on success. 1753 */ 1754 static struct pf_kstate * 1755 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1756 u_int dir) 1757 { 1758 struct pf_keyhash *kh; 1759 struct pf_state_key *sk; 1760 struct pf_kstate *s; 1761 int idx; 1762 1763 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1764 1765 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1766 1767 PF_HASHROW_LOCK(kh); 1768 LIST_FOREACH(sk, &kh->keys, entry) 1769 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1770 break; 1771 if (sk == NULL) { 1772 PF_HASHROW_UNLOCK(kh); 1773 return (NULL); 1774 } 1775 1776 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1777 1778 /* List is sorted, if-bound states before floating ones. */ 1779 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1780 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1781 PF_STATE_LOCK(s); 1782 PF_HASHROW_UNLOCK(kh); 1783 if (__predict_false(s->timeout >= PFTM_MAX)) { 1784 /* 1785 * State is either being processed by 1786 * pf_unlink_state() in an other thread, or 1787 * is scheduled for immediate expiry. 1788 */ 1789 PF_STATE_UNLOCK(s); 1790 return (NULL); 1791 } 1792 return (s); 1793 } 1794 PF_HASHROW_UNLOCK(kh); 1795 1796 return (NULL); 1797 } 1798 1799 /* 1800 * Returns with ID hash slot locked on success. 1801 */ 1802 struct pf_kstate * 1803 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1804 { 1805 struct pf_keyhash *kh; 1806 struct pf_state_key *sk; 1807 struct pf_kstate *s, *ret = NULL; 1808 int idx, inout = 0; 1809 1810 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1811 1812 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1813 1814 PF_HASHROW_LOCK(kh); 1815 LIST_FOREACH(sk, &kh->keys, entry) 1816 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1817 break; 1818 if (sk == NULL) { 1819 PF_HASHROW_UNLOCK(kh); 1820 return (NULL); 1821 } 1822 switch (dir) { 1823 case PF_IN: 1824 idx = PF_SK_WIRE; 1825 break; 1826 case PF_OUT: 1827 idx = PF_SK_STACK; 1828 break; 1829 case PF_INOUT: 1830 idx = PF_SK_WIRE; 1831 inout = 1; 1832 break; 1833 default: 1834 panic("%s: dir %u", __func__, dir); 1835 } 1836 second_run: 1837 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1838 if (more == NULL) { 1839 PF_STATE_LOCK(s); 1840 PF_HASHROW_UNLOCK(kh); 1841 return (s); 1842 } 1843 1844 if (ret) 1845 (*more)++; 1846 else { 1847 ret = s; 1848 PF_STATE_LOCK(s); 1849 } 1850 } 1851 if (inout == 1) { 1852 inout = 0; 1853 idx = PF_SK_STACK; 1854 goto second_run; 1855 } 1856 PF_HASHROW_UNLOCK(kh); 1857 1858 return (ret); 1859 } 1860 1861 /* 1862 * FIXME 1863 * This routine is inefficient -- locks the state only to unlock immediately on 1864 * return. 1865 * It is racy -- after the state is unlocked nothing stops other threads from 1866 * removing it. 1867 */ 1868 bool 1869 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1870 { 1871 struct pf_kstate *s; 1872 1873 s = pf_find_state_all(key, dir, NULL); 1874 if (s != NULL) { 1875 PF_STATE_UNLOCK(s); 1876 return (true); 1877 } 1878 return (false); 1879 } 1880 1881 struct pf_udp_mapping * 1882 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1883 struct pf_addr *nat_addr, uint16_t nat_port) 1884 { 1885 struct pf_udp_mapping *mapping; 1886 1887 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1888 if (mapping == NULL) 1889 return (NULL); 1890 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1891 mapping->endpoints[0].port = src_port; 1892 mapping->endpoints[0].af = af; 1893 mapping->endpoints[0].mapping = mapping; 1894 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1895 mapping->endpoints[1].port = nat_port; 1896 mapping->endpoints[1].af = af; 1897 mapping->endpoints[1].mapping = mapping; 1898 refcount_init(&mapping->refs, 1); 1899 return (mapping); 1900 } 1901 1902 int 1903 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1904 { 1905 struct pf_udpendpointhash *h0, *h1; 1906 struct pf_udp_endpoint *endpoint; 1907 int ret = EEXIST; 1908 1909 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1910 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1911 if (h0 == h1) { 1912 PF_HASHROW_LOCK(h0); 1913 } else if (h0 < h1) { 1914 PF_HASHROW_LOCK(h0); 1915 PF_HASHROW_LOCK(h1); 1916 } else { 1917 PF_HASHROW_LOCK(h1); 1918 PF_HASHROW_LOCK(h0); 1919 } 1920 1921 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1922 if (bcmp(endpoint, &mapping->endpoints[0], 1923 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1924 break; 1925 } 1926 if (endpoint != NULL) 1927 goto cleanup; 1928 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1929 if (bcmp(endpoint, &mapping->endpoints[1], 1930 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1931 break; 1932 } 1933 if (endpoint != NULL) 1934 goto cleanup; 1935 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1936 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1937 ret = 0; 1938 1939 cleanup: 1940 if (h0 != h1) { 1941 PF_HASHROW_UNLOCK(h0); 1942 PF_HASHROW_UNLOCK(h1); 1943 } else { 1944 PF_HASHROW_UNLOCK(h0); 1945 } 1946 return (ret); 1947 } 1948 1949 void 1950 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1951 { 1952 /* refcount is synchronized on the source endpoint's row lock */ 1953 struct pf_udpendpointhash *h0, *h1; 1954 1955 if (mapping == NULL) 1956 return; 1957 1958 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1959 PF_HASHROW_LOCK(h0); 1960 if (refcount_release(&mapping->refs)) { 1961 LIST_REMOVE(&mapping->endpoints[0], entry); 1962 PF_HASHROW_UNLOCK(h0); 1963 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1964 PF_HASHROW_LOCK(h1); 1965 LIST_REMOVE(&mapping->endpoints[1], entry); 1966 PF_HASHROW_UNLOCK(h1); 1967 1968 uma_zfree(V_pf_udp_mapping_z, mapping); 1969 } else { 1970 PF_HASHROW_UNLOCK(h0); 1971 } 1972 } 1973 1974 1975 struct pf_udp_mapping * 1976 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1977 { 1978 struct pf_udpendpointhash *uh; 1979 struct pf_udp_endpoint *endpoint; 1980 1981 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1982 1983 PF_HASHROW_LOCK(uh); 1984 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1985 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1986 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1987 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1988 break; 1989 } 1990 if (endpoint == NULL) { 1991 PF_HASHROW_UNLOCK(uh); 1992 return (NULL); 1993 } 1994 refcount_acquire(&endpoint->mapping->refs); 1995 PF_HASHROW_UNLOCK(uh); 1996 return (endpoint->mapping); 1997 } 1998 /* END state table stuff */ 1999 2000 static void 2001 pf_send(struct pf_send_entry *pfse) 2002 { 2003 2004 PF_SENDQ_LOCK(); 2005 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 2006 PF_SENDQ_UNLOCK(); 2007 swi_sched(V_pf_swi_cookie, 0); 2008 } 2009 2010 static bool 2011 pf_isforlocal(struct mbuf *m, int af) 2012 { 2013 switch (af) { 2014 #ifdef INET 2015 case AF_INET: { 2016 struct ip *ip = mtod(m, struct ip *); 2017 2018 return (in_localip(ip->ip_dst)); 2019 } 2020 #endif 2021 #ifdef INET6 2022 case AF_INET6: { 2023 struct ip6_hdr *ip6; 2024 struct in6_ifaddr *ia; 2025 ip6 = mtod(m, struct ip6_hdr *); 2026 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2027 if (ia == NULL) 2028 return (false); 2029 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2030 } 2031 #endif 2032 default: 2033 panic("Unsupported af %d", af); 2034 } 2035 2036 return (false); 2037 } 2038 2039 int 2040 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2041 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2042 { 2043 /* 2044 * ICMP types marked with PF_OUT are typically responses to 2045 * PF_IN, and will match states in the opposite direction. 2046 * PF_IN ICMP types need to match a state with that type. 2047 */ 2048 *icmp_dir = PF_OUT; 2049 *multi = PF_ICMP_MULTI_LINK; 2050 /* Queries (and responses) */ 2051 switch (pd->af) { 2052 #ifdef INET 2053 case AF_INET: 2054 switch (type) { 2055 case ICMP_ECHO: 2056 *icmp_dir = PF_IN; 2057 case ICMP_ECHOREPLY: 2058 *virtual_type = ICMP_ECHO; 2059 *virtual_id = pd->hdr.icmp.icmp_id; 2060 break; 2061 2062 case ICMP_TSTAMP: 2063 *icmp_dir = PF_IN; 2064 case ICMP_TSTAMPREPLY: 2065 *virtual_type = ICMP_TSTAMP; 2066 *virtual_id = pd->hdr.icmp.icmp_id; 2067 break; 2068 2069 case ICMP_IREQ: 2070 *icmp_dir = PF_IN; 2071 case ICMP_IREQREPLY: 2072 *virtual_type = ICMP_IREQ; 2073 *virtual_id = pd->hdr.icmp.icmp_id; 2074 break; 2075 2076 case ICMP_MASKREQ: 2077 *icmp_dir = PF_IN; 2078 case ICMP_MASKREPLY: 2079 *virtual_type = ICMP_MASKREQ; 2080 *virtual_id = pd->hdr.icmp.icmp_id; 2081 break; 2082 2083 case ICMP_IPV6_WHEREAREYOU: 2084 *icmp_dir = PF_IN; 2085 case ICMP_IPV6_IAMHERE: 2086 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2087 *virtual_id = 0; /* Nothing sane to match on! */ 2088 break; 2089 2090 case ICMP_MOBILE_REGREQUEST: 2091 *icmp_dir = PF_IN; 2092 case ICMP_MOBILE_REGREPLY: 2093 *virtual_type = ICMP_MOBILE_REGREQUEST; 2094 *virtual_id = 0; /* Nothing sane to match on! */ 2095 break; 2096 2097 case ICMP_ROUTERSOLICIT: 2098 *icmp_dir = PF_IN; 2099 case ICMP_ROUTERADVERT: 2100 *virtual_type = ICMP_ROUTERSOLICIT; 2101 *virtual_id = 0; /* Nothing sane to match on! */ 2102 break; 2103 2104 /* These ICMP types map to other connections */ 2105 case ICMP_UNREACH: 2106 case ICMP_SOURCEQUENCH: 2107 case ICMP_REDIRECT: 2108 case ICMP_TIMXCEED: 2109 case ICMP_PARAMPROB: 2110 /* These will not be used, but set them anyway */ 2111 *icmp_dir = PF_IN; 2112 *virtual_type = type; 2113 *virtual_id = 0; 2114 HTONS(*virtual_type); 2115 return (1); /* These types match to another state */ 2116 2117 /* 2118 * All remaining ICMP types get their own states, 2119 * and will only match in one direction. 2120 */ 2121 default: 2122 *icmp_dir = PF_IN; 2123 *virtual_type = type; 2124 *virtual_id = 0; 2125 break; 2126 } 2127 break; 2128 #endif /* INET */ 2129 #ifdef INET6 2130 case AF_INET6: 2131 switch (type) { 2132 case ICMP6_ECHO_REQUEST: 2133 *icmp_dir = PF_IN; 2134 case ICMP6_ECHO_REPLY: 2135 *virtual_type = ICMP6_ECHO_REQUEST; 2136 *virtual_id = pd->hdr.icmp6.icmp6_id; 2137 break; 2138 2139 case MLD_LISTENER_QUERY: 2140 case MLD_LISTENER_REPORT: { 2141 /* 2142 * Listener Report can be sent by clients 2143 * without an associated Listener Query. 2144 * In addition to that, when Report is sent as a 2145 * reply to a Query its source and destination 2146 * address are different. 2147 */ 2148 *icmp_dir = PF_IN; 2149 *virtual_type = MLD_LISTENER_QUERY; 2150 *virtual_id = 0; 2151 break; 2152 } 2153 case MLD_MTRACE: 2154 *icmp_dir = PF_IN; 2155 case MLD_MTRACE_RESP: 2156 *virtual_type = MLD_MTRACE; 2157 *virtual_id = 0; /* Nothing sane to match on! */ 2158 break; 2159 2160 case ND_NEIGHBOR_SOLICIT: 2161 *icmp_dir = PF_IN; 2162 case ND_NEIGHBOR_ADVERT: { 2163 *virtual_type = ND_NEIGHBOR_SOLICIT; 2164 *virtual_id = 0; 2165 break; 2166 } 2167 2168 /* 2169 * These ICMP types map to other connections. 2170 * ND_REDIRECT can't be in this list because the triggering 2171 * packet header is optional. 2172 */ 2173 case ICMP6_DST_UNREACH: 2174 case ICMP6_PACKET_TOO_BIG: 2175 case ICMP6_TIME_EXCEEDED: 2176 case ICMP6_PARAM_PROB: 2177 /* These will not be used, but set them anyway */ 2178 *icmp_dir = PF_IN; 2179 *virtual_type = type; 2180 *virtual_id = 0; 2181 HTONS(*virtual_type); 2182 return (1); /* These types match to another state */ 2183 /* 2184 * All remaining ICMP6 types get their own states, 2185 * and will only match in one direction. 2186 */ 2187 default: 2188 *icmp_dir = PF_IN; 2189 *virtual_type = type; 2190 *virtual_id = 0; 2191 break; 2192 } 2193 break; 2194 #endif /* INET6 */ 2195 default: 2196 *icmp_dir = PF_IN; 2197 *virtual_type = type; 2198 *virtual_id = 0; 2199 break; 2200 } 2201 HTONS(*virtual_type); 2202 return (0); /* These types match to their own state */ 2203 } 2204 2205 void 2206 pf_intr(void *v) 2207 { 2208 struct epoch_tracker et; 2209 struct pf_send_head queue; 2210 struct pf_send_entry *pfse, *next; 2211 2212 CURVNET_SET((struct vnet *)v); 2213 2214 PF_SENDQ_LOCK(); 2215 queue = V_pf_sendqueue; 2216 STAILQ_INIT(&V_pf_sendqueue); 2217 PF_SENDQ_UNLOCK(); 2218 2219 NET_EPOCH_ENTER(et); 2220 2221 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2222 switch (pfse->pfse_type) { 2223 #ifdef INET 2224 case PFSE_IP: { 2225 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2226 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2227 pfse->pfse_m->m_pkthdr.csum_flags |= 2228 CSUM_IP_VALID | CSUM_IP_CHECKED; 2229 ip_input(pfse->pfse_m); 2230 } else { 2231 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2232 NULL); 2233 } 2234 break; 2235 } 2236 case PFSE_ICMP: 2237 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2238 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2239 break; 2240 #endif /* INET */ 2241 #ifdef INET6 2242 case PFSE_IP6: 2243 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2244 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2245 ip6_input(pfse->pfse_m); 2246 } else { 2247 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2248 NULL, NULL); 2249 } 2250 break; 2251 case PFSE_ICMP6: 2252 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2253 pfse->icmpopts.code, pfse->icmpopts.mtu); 2254 break; 2255 #endif /* INET6 */ 2256 default: 2257 panic("%s: unknown type", __func__); 2258 } 2259 free(pfse, M_PFTEMP); 2260 } 2261 NET_EPOCH_EXIT(et); 2262 CURVNET_RESTORE(); 2263 } 2264 2265 #define pf_purge_thread_period (hz / 10) 2266 2267 #ifdef PF_WANT_32_TO_64_COUNTER 2268 static void 2269 pf_status_counter_u64_periodic(void) 2270 { 2271 2272 PF_RULES_RASSERT(); 2273 2274 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2275 return; 2276 } 2277 2278 for (int i = 0; i < FCNT_MAX; i++) { 2279 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2280 } 2281 } 2282 2283 static void 2284 pf_kif_counter_u64_periodic(void) 2285 { 2286 struct pfi_kkif *kif; 2287 size_t r, run; 2288 2289 PF_RULES_RASSERT(); 2290 2291 if (__predict_false(V_pf_allkifcount == 0)) { 2292 return; 2293 } 2294 2295 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2296 return; 2297 } 2298 2299 run = V_pf_allkifcount / 10; 2300 if (run < 5) 2301 run = 5; 2302 2303 for (r = 0; r < run; r++) { 2304 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2305 if (kif == NULL) { 2306 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2307 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2308 break; 2309 } 2310 2311 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2312 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2313 2314 for (int i = 0; i < 2; i++) { 2315 for (int j = 0; j < 2; j++) { 2316 for (int k = 0; k < 2; k++) { 2317 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2318 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2319 } 2320 } 2321 } 2322 } 2323 } 2324 2325 static void 2326 pf_rule_counter_u64_periodic(void) 2327 { 2328 struct pf_krule *rule; 2329 size_t r, run; 2330 2331 PF_RULES_RASSERT(); 2332 2333 if (__predict_false(V_pf_allrulecount == 0)) { 2334 return; 2335 } 2336 2337 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2338 return; 2339 } 2340 2341 run = V_pf_allrulecount / 10; 2342 if (run < 5) 2343 run = 5; 2344 2345 for (r = 0; r < run; r++) { 2346 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2347 if (rule == NULL) { 2348 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2349 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2350 break; 2351 } 2352 2353 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2354 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2355 2356 pf_counter_u64_periodic(&rule->evaluations); 2357 for (int i = 0; i < 2; i++) { 2358 pf_counter_u64_periodic(&rule->packets[i]); 2359 pf_counter_u64_periodic(&rule->bytes[i]); 2360 } 2361 } 2362 } 2363 2364 static void 2365 pf_counter_u64_periodic_main(void) 2366 { 2367 PF_RULES_RLOCK_TRACKER; 2368 2369 V_pf_counter_periodic_iter++; 2370 2371 PF_RULES_RLOCK(); 2372 pf_counter_u64_critical_enter(); 2373 pf_status_counter_u64_periodic(); 2374 pf_kif_counter_u64_periodic(); 2375 pf_rule_counter_u64_periodic(); 2376 pf_counter_u64_critical_exit(); 2377 PF_RULES_RUNLOCK(); 2378 } 2379 #else 2380 #define pf_counter_u64_periodic_main() do { } while (0) 2381 #endif 2382 2383 void 2384 pf_purge_thread(void *unused __unused) 2385 { 2386 struct epoch_tracker et; 2387 2388 VNET_ITERATOR_DECL(vnet_iter); 2389 2390 sx_xlock(&pf_end_lock); 2391 while (pf_end_threads == 0) { 2392 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2393 2394 VNET_LIST_RLOCK(); 2395 NET_EPOCH_ENTER(et); 2396 VNET_FOREACH(vnet_iter) { 2397 CURVNET_SET(vnet_iter); 2398 2399 /* Wait until V_pf_default_rule is initialized. */ 2400 if (V_pf_vnet_active == 0) { 2401 CURVNET_RESTORE(); 2402 continue; 2403 } 2404 2405 pf_counter_u64_periodic_main(); 2406 2407 /* 2408 * Process 1/interval fraction of the state 2409 * table every run. 2410 */ 2411 V_pf_purge_idx = 2412 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2413 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2414 2415 /* 2416 * Purge other expired types every 2417 * PFTM_INTERVAL seconds. 2418 */ 2419 if (V_pf_purge_idx == 0) { 2420 /* 2421 * Order is important: 2422 * - states and src nodes reference rules 2423 * - states and rules reference kifs 2424 */ 2425 pf_purge_expired_fragments(); 2426 pf_purge_expired_src_nodes(); 2427 pf_purge_unlinked_rules(); 2428 pfi_kkif_purge(); 2429 } 2430 CURVNET_RESTORE(); 2431 } 2432 NET_EPOCH_EXIT(et); 2433 VNET_LIST_RUNLOCK(); 2434 } 2435 2436 pf_end_threads++; 2437 sx_xunlock(&pf_end_lock); 2438 kproc_exit(0); 2439 } 2440 2441 void 2442 pf_unload_vnet_purge(void) 2443 { 2444 2445 /* 2446 * To cleanse up all kifs and rules we need 2447 * two runs: first one clears reference flags, 2448 * then pf_purge_expired_states() doesn't 2449 * raise them, and then second run frees. 2450 */ 2451 pf_purge_unlinked_rules(); 2452 pfi_kkif_purge(); 2453 2454 /* 2455 * Now purge everything. 2456 */ 2457 pf_purge_expired_states(0, V_pf_hashmask); 2458 pf_purge_fragments(UINT_MAX); 2459 pf_purge_expired_src_nodes(); 2460 2461 /* 2462 * Now all kifs & rules should be unreferenced, 2463 * thus should be successfully freed. 2464 */ 2465 pf_purge_unlinked_rules(); 2466 pfi_kkif_purge(); 2467 } 2468 2469 u_int32_t 2470 pf_state_expires(const struct pf_kstate *state) 2471 { 2472 u_int32_t timeout; 2473 u_int32_t start; 2474 u_int32_t end; 2475 u_int32_t states; 2476 2477 /* handle all PFTM_* > PFTM_MAX here */ 2478 if (state->timeout == PFTM_PURGE) 2479 return (time_uptime); 2480 KASSERT(state->timeout != PFTM_UNLINKED, 2481 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2482 KASSERT((state->timeout < PFTM_MAX), 2483 ("pf_state_expires: timeout > PFTM_MAX")); 2484 timeout = state->rule->timeout[state->timeout]; 2485 if (!timeout) 2486 timeout = V_pf_default_rule.timeout[state->timeout]; 2487 start = state->rule->timeout[PFTM_ADAPTIVE_START]; 2488 if (start && state->rule != &V_pf_default_rule) { 2489 end = state->rule->timeout[PFTM_ADAPTIVE_END]; 2490 states = counter_u64_fetch(state->rule->states_cur); 2491 } else { 2492 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2493 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2494 states = V_pf_status.states; 2495 } 2496 if (end && states > start && start < end) { 2497 if (states < end) { 2498 timeout = (u_int64_t)timeout * (end - states) / 2499 (end - start); 2500 return ((state->expire / 1000) + timeout); 2501 } 2502 else 2503 return (time_uptime); 2504 } 2505 return ((state->expire / 1000) + timeout); 2506 } 2507 2508 void 2509 pf_purge_expired_src_nodes(void) 2510 { 2511 struct pf_ksrc_node_list freelist; 2512 struct pf_srchash *sh; 2513 struct pf_ksrc_node *cur, *next; 2514 int i; 2515 2516 LIST_INIT(&freelist); 2517 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2518 PF_HASHROW_LOCK(sh); 2519 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2520 if (cur->states == 0 && cur->expire <= time_uptime) { 2521 pf_unlink_src_node(cur); 2522 LIST_INSERT_HEAD(&freelist, cur, entry); 2523 } else if (cur->rule != NULL) 2524 cur->rule->rule_ref |= PFRULE_REFS; 2525 PF_HASHROW_UNLOCK(sh); 2526 } 2527 2528 pf_free_src_nodes(&freelist); 2529 2530 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2531 } 2532 2533 static void 2534 pf_src_tree_remove_state(struct pf_kstate *s) 2535 { 2536 struct pf_ksrc_node *sn; 2537 uint32_t timeout; 2538 2539 timeout = s->rule->timeout[PFTM_SRC_NODE] ? 2540 s->rule->timeout[PFTM_SRC_NODE] : 2541 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2542 2543 if (s->src_node != NULL) { 2544 sn = s->src_node; 2545 PF_SRC_NODE_LOCK(sn); 2546 if (s->src.tcp_est) 2547 --sn->conn; 2548 if (--sn->states == 0) 2549 sn->expire = time_uptime + timeout; 2550 PF_SRC_NODE_UNLOCK(sn); 2551 } 2552 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2553 sn = s->nat_src_node; 2554 PF_SRC_NODE_LOCK(sn); 2555 if (--sn->states == 0) 2556 sn->expire = time_uptime + timeout; 2557 PF_SRC_NODE_UNLOCK(sn); 2558 } 2559 s->src_node = s->nat_src_node = NULL; 2560 } 2561 2562 /* 2563 * Unlink and potentilly free a state. Function may be 2564 * called with ID hash row locked, but always returns 2565 * unlocked, since it needs to go through key hash locking. 2566 */ 2567 int 2568 pf_unlink_state(struct pf_kstate *s) 2569 { 2570 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2571 2572 NET_EPOCH_ASSERT(); 2573 PF_HASHROW_ASSERT(ih); 2574 2575 if (s->timeout == PFTM_UNLINKED) { 2576 /* 2577 * State is being processed 2578 * by pf_unlink_state() in 2579 * an other thread. 2580 */ 2581 PF_HASHROW_UNLOCK(ih); 2582 return (0); /* XXXGL: undefined actually */ 2583 } 2584 2585 if (s->src.state == PF_TCPS_PROXY_DST) { 2586 /* XXX wire key the right one? */ 2587 pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af, 2588 &s->key[PF_SK_WIRE]->addr[1], 2589 &s->key[PF_SK_WIRE]->addr[0], 2590 s->key[PF_SK_WIRE]->port[1], 2591 s->key[PF_SK_WIRE]->port[0], 2592 s->src.seqhi, s->src.seqlo + 1, 2593 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2594 } 2595 2596 LIST_REMOVE(s, entry); 2597 pf_src_tree_remove_state(s); 2598 2599 if (V_pfsync_delete_state_ptr != NULL) 2600 V_pfsync_delete_state_ptr(s); 2601 2602 STATE_DEC_COUNTERS(s); 2603 2604 s->timeout = PFTM_UNLINKED; 2605 2606 /* Ensure we remove it from the list of halfopen states, if needed. */ 2607 if (s->key[PF_SK_STACK] != NULL && 2608 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2609 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2610 2611 PF_HASHROW_UNLOCK(ih); 2612 2613 pf_detach_state(s); 2614 2615 pf_udp_mapping_release(s->udp_mapping); 2616 2617 /* pf_state_insert() initialises refs to 2 */ 2618 return (pf_release_staten(s, 2)); 2619 } 2620 2621 struct pf_kstate * 2622 pf_alloc_state(int flags) 2623 { 2624 2625 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2626 } 2627 2628 void 2629 pf_free_state(struct pf_kstate *cur) 2630 { 2631 struct pf_krule_item *ri; 2632 2633 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2634 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2635 cur->timeout)); 2636 2637 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2638 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2639 free(ri, M_PF_RULE_ITEM); 2640 } 2641 2642 pf_normalize_tcp_cleanup(cur); 2643 uma_zfree(V_pf_state_z, cur); 2644 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2645 } 2646 2647 /* 2648 * Called only from pf_purge_thread(), thus serialized. 2649 */ 2650 static u_int 2651 pf_purge_expired_states(u_int i, int maxcheck) 2652 { 2653 struct pf_idhash *ih; 2654 struct pf_kstate *s; 2655 struct pf_krule_item *mrm; 2656 size_t count __unused; 2657 2658 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2659 2660 /* 2661 * Go through hash and unlink states that expire now. 2662 */ 2663 while (maxcheck > 0) { 2664 count = 0; 2665 ih = &V_pf_idhash[i]; 2666 2667 /* only take the lock if we expect to do work */ 2668 if (!LIST_EMPTY(&ih->states)) { 2669 relock: 2670 PF_HASHROW_LOCK(ih); 2671 LIST_FOREACH(s, &ih->states, entry) { 2672 if (pf_state_expires(s) <= time_uptime) { 2673 V_pf_status.states -= 2674 pf_unlink_state(s); 2675 goto relock; 2676 } 2677 s->rule->rule_ref |= PFRULE_REFS; 2678 if (s->nat_rule != NULL) 2679 s->nat_rule->rule_ref |= PFRULE_REFS; 2680 if (s->anchor != NULL) 2681 s->anchor->rule_ref |= PFRULE_REFS; 2682 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2683 SLIST_FOREACH(mrm, &s->match_rules, entry) 2684 mrm->r->rule_ref |= PFRULE_REFS; 2685 if (s->rt_kif) 2686 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2687 count++; 2688 } 2689 PF_HASHROW_UNLOCK(ih); 2690 } 2691 2692 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2693 2694 /* Return when we hit end of hash. */ 2695 if (++i > V_pf_hashmask) { 2696 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2697 return (0); 2698 } 2699 2700 maxcheck--; 2701 } 2702 2703 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2704 2705 return (i); 2706 } 2707 2708 static void 2709 pf_purge_unlinked_rules(void) 2710 { 2711 struct pf_krulequeue tmpq; 2712 struct pf_krule *r, *r1; 2713 2714 /* 2715 * If we have overloading task pending, then we'd 2716 * better skip purging this time. There is a tiny 2717 * probability that overloading task references 2718 * an already unlinked rule. 2719 */ 2720 PF_OVERLOADQ_LOCK(); 2721 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2722 PF_OVERLOADQ_UNLOCK(); 2723 return; 2724 } 2725 PF_OVERLOADQ_UNLOCK(); 2726 2727 /* 2728 * Do naive mark-and-sweep garbage collecting of old rules. 2729 * Reference flag is raised by pf_purge_expired_states() 2730 * and pf_purge_expired_src_nodes(). 2731 * 2732 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2733 * use a temporary queue. 2734 */ 2735 TAILQ_INIT(&tmpq); 2736 PF_UNLNKDRULES_LOCK(); 2737 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2738 if (!(r->rule_ref & PFRULE_REFS)) { 2739 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2740 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2741 } else 2742 r->rule_ref &= ~PFRULE_REFS; 2743 } 2744 PF_UNLNKDRULES_UNLOCK(); 2745 2746 if (!TAILQ_EMPTY(&tmpq)) { 2747 PF_CONFIG_LOCK(); 2748 PF_RULES_WLOCK(); 2749 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2750 TAILQ_REMOVE(&tmpq, r, entries); 2751 pf_free_rule(r); 2752 } 2753 PF_RULES_WUNLOCK(); 2754 PF_CONFIG_UNLOCK(); 2755 } 2756 } 2757 2758 void 2759 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2760 { 2761 switch (af) { 2762 #ifdef INET 2763 case AF_INET: { 2764 u_int32_t a = ntohl(addr->addr32[0]); 2765 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2766 (a>>8)&255, a&255); 2767 if (p) { 2768 p = ntohs(p); 2769 printf(":%u", p); 2770 } 2771 break; 2772 } 2773 #endif /* INET */ 2774 #ifdef INET6 2775 case AF_INET6: { 2776 u_int16_t b; 2777 u_int8_t i, curstart, curend, maxstart, maxend; 2778 curstart = curend = maxstart = maxend = 255; 2779 for (i = 0; i < 8; i++) { 2780 if (!addr->addr16[i]) { 2781 if (curstart == 255) 2782 curstart = i; 2783 curend = i; 2784 } else { 2785 if ((curend - curstart) > 2786 (maxend - maxstart)) { 2787 maxstart = curstart; 2788 maxend = curend; 2789 } 2790 curstart = curend = 255; 2791 } 2792 } 2793 if ((curend - curstart) > 2794 (maxend - maxstart)) { 2795 maxstart = curstart; 2796 maxend = curend; 2797 } 2798 for (i = 0; i < 8; i++) { 2799 if (i >= maxstart && i <= maxend) { 2800 if (i == 0) 2801 printf(":"); 2802 if (i == maxend) 2803 printf(":"); 2804 } else { 2805 b = ntohs(addr->addr16[i]); 2806 printf("%x", b); 2807 if (i < 7) 2808 printf(":"); 2809 } 2810 } 2811 if (p) { 2812 p = ntohs(p); 2813 printf("[%u]", p); 2814 } 2815 break; 2816 } 2817 #endif /* INET6 */ 2818 } 2819 } 2820 2821 void 2822 pf_print_state(struct pf_kstate *s) 2823 { 2824 pf_print_state_parts(s, NULL, NULL); 2825 } 2826 2827 static void 2828 pf_print_state_parts(struct pf_kstate *s, 2829 struct pf_state_key *skwp, struct pf_state_key *sksp) 2830 { 2831 struct pf_state_key *skw, *sks; 2832 u_int8_t proto, dir; 2833 2834 /* Do our best to fill these, but they're skipped if NULL */ 2835 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2836 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2837 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2838 dir = s ? s->direction : 0; 2839 2840 switch (proto) { 2841 case IPPROTO_IPV4: 2842 printf("IPv4"); 2843 break; 2844 case IPPROTO_IPV6: 2845 printf("IPv6"); 2846 break; 2847 case IPPROTO_TCP: 2848 printf("TCP"); 2849 break; 2850 case IPPROTO_UDP: 2851 printf("UDP"); 2852 break; 2853 case IPPROTO_ICMP: 2854 printf("ICMP"); 2855 break; 2856 case IPPROTO_ICMPV6: 2857 printf("ICMPv6"); 2858 break; 2859 default: 2860 printf("%u", proto); 2861 break; 2862 } 2863 switch (dir) { 2864 case PF_IN: 2865 printf(" in"); 2866 break; 2867 case PF_OUT: 2868 printf(" out"); 2869 break; 2870 } 2871 if (skw) { 2872 printf(" wire: "); 2873 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2874 printf(" "); 2875 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2876 } 2877 if (sks) { 2878 printf(" stack: "); 2879 if (sks != skw) { 2880 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2881 printf(" "); 2882 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2883 } else 2884 printf("-"); 2885 } 2886 if (s) { 2887 if (proto == IPPROTO_TCP) { 2888 printf(" [lo=%u high=%u win=%u modulator=%u", 2889 s->src.seqlo, s->src.seqhi, 2890 s->src.max_win, s->src.seqdiff); 2891 if (s->src.wscale && s->dst.wscale) 2892 printf(" wscale=%u", 2893 s->src.wscale & PF_WSCALE_MASK); 2894 printf("]"); 2895 printf(" [lo=%u high=%u win=%u modulator=%u", 2896 s->dst.seqlo, s->dst.seqhi, 2897 s->dst.max_win, s->dst.seqdiff); 2898 if (s->src.wscale && s->dst.wscale) 2899 printf(" wscale=%u", 2900 s->dst.wscale & PF_WSCALE_MASK); 2901 printf("]"); 2902 } 2903 printf(" %u:%u", s->src.state, s->dst.state); 2904 if (s->rule) 2905 printf(" @%d", s->rule->nr); 2906 } 2907 } 2908 2909 void 2910 pf_print_flags(u_int8_t f) 2911 { 2912 if (f) 2913 printf(" "); 2914 if (f & TH_FIN) 2915 printf("F"); 2916 if (f & TH_SYN) 2917 printf("S"); 2918 if (f & TH_RST) 2919 printf("R"); 2920 if (f & TH_PUSH) 2921 printf("P"); 2922 if (f & TH_ACK) 2923 printf("A"); 2924 if (f & TH_URG) 2925 printf("U"); 2926 if (f & TH_ECE) 2927 printf("E"); 2928 if (f & TH_CWR) 2929 printf("W"); 2930 } 2931 2932 #define PF_SET_SKIP_STEPS(i) \ 2933 do { \ 2934 while (head[i] != cur) { \ 2935 head[i]->skip[i] = cur; \ 2936 head[i] = TAILQ_NEXT(head[i], entries); \ 2937 } \ 2938 } while (0) 2939 2940 void 2941 pf_calc_skip_steps(struct pf_krulequeue *rules) 2942 { 2943 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2944 int i; 2945 2946 cur = TAILQ_FIRST(rules); 2947 prev = cur; 2948 for (i = 0; i < PF_SKIP_COUNT; ++i) 2949 head[i] = cur; 2950 while (cur != NULL) { 2951 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2952 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2953 if (cur->direction != prev->direction) 2954 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2955 if (cur->af != prev->af) 2956 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2957 if (cur->proto != prev->proto) 2958 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2959 if (cur->src.neg != prev->src.neg || 2960 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2961 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2962 if (cur->dst.neg != prev->dst.neg || 2963 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2964 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2965 if (cur->src.port[0] != prev->src.port[0] || 2966 cur->src.port[1] != prev->src.port[1] || 2967 cur->src.port_op != prev->src.port_op) 2968 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2969 if (cur->dst.port[0] != prev->dst.port[0] || 2970 cur->dst.port[1] != prev->dst.port[1] || 2971 cur->dst.port_op != prev->dst.port_op) 2972 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2973 2974 prev = cur; 2975 cur = TAILQ_NEXT(cur, entries); 2976 } 2977 for (i = 0; i < PF_SKIP_COUNT; ++i) 2978 PF_SET_SKIP_STEPS(i); 2979 } 2980 2981 int 2982 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2983 { 2984 if (aw1->type != aw2->type) 2985 return (1); 2986 switch (aw1->type) { 2987 case PF_ADDR_ADDRMASK: 2988 case PF_ADDR_RANGE: 2989 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2990 return (1); 2991 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2992 return (1); 2993 return (0); 2994 case PF_ADDR_DYNIFTL: 2995 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2996 case PF_ADDR_NOROUTE: 2997 case PF_ADDR_URPFFAILED: 2998 return (0); 2999 case PF_ADDR_TABLE: 3000 return (aw1->p.tbl != aw2->p.tbl); 3001 default: 3002 printf("invalid address type: %d\n", aw1->type); 3003 return (1); 3004 } 3005 } 3006 3007 /** 3008 * Checksum updates are a little complicated because the checksum in the TCP/UDP 3009 * header isn't always a full checksum. In some cases (i.e. output) it's a 3010 * pseudo-header checksum, which is a partial checksum over src/dst IP 3011 * addresses, protocol number and length. 3012 * 3013 * That means we have the following cases: 3014 * * Input or forwarding: we don't have TSO, the checksum fields are full 3015 * checksums, we need to update the checksum whenever we change anything. 3016 * * Output (i.e. the checksum is a pseudo-header checksum): 3017 * x The field being updated is src/dst address or affects the length of 3018 * the packet. We need to update the pseudo-header checksum (note that this 3019 * checksum is not ones' complement). 3020 * x Some other field is being modified (e.g. src/dst port numbers): We 3021 * don't have to update anything. 3022 **/ 3023 u_int16_t 3024 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3025 { 3026 u_int32_t x; 3027 3028 x = cksum + old - new; 3029 x = (x + (x >> 16)) & 0xffff; 3030 3031 /* optimise: eliminate a branch when not udp */ 3032 if (udp && cksum == 0x0000) 3033 return cksum; 3034 if (udp && x == 0x0000) 3035 x = 0xffff; 3036 3037 return (u_int16_t)(x); 3038 } 3039 3040 static void 3041 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3042 u_int8_t udp) 3043 { 3044 u_int16_t old = htons(hi ? (*f << 8) : *f); 3045 u_int16_t new = htons(hi ? ( v << 8) : v); 3046 3047 if (*f == v) 3048 return; 3049 3050 *f = v; 3051 3052 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3053 return; 3054 3055 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3056 } 3057 3058 void 3059 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3060 bool hi, u_int8_t udp) 3061 { 3062 u_int8_t *fb = (u_int8_t *)f; 3063 u_int8_t *vb = (u_int8_t *)&v; 3064 3065 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3066 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3067 } 3068 3069 void 3070 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3071 bool hi, u_int8_t udp) 3072 { 3073 u_int8_t *fb = (u_int8_t *)f; 3074 u_int8_t *vb = (u_int8_t *)&v; 3075 3076 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3077 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3078 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3079 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3080 } 3081 3082 u_int16_t 3083 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3084 u_int16_t new, u_int8_t udp) 3085 { 3086 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3087 return (cksum); 3088 3089 return (pf_cksum_fixup(cksum, old, new, udp)); 3090 } 3091 3092 static void 3093 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3094 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3095 sa_family_t af) 3096 { 3097 struct pf_addr ao; 3098 u_int16_t po = *p; 3099 3100 PF_ACPY(&ao, a, af); 3101 PF_ACPY(a, an, af); 3102 3103 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3104 *pc = ~*pc; 3105 3106 *p = pn; 3107 3108 switch (af) { 3109 #ifdef INET 3110 case AF_INET: 3111 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3112 ao.addr16[0], an->addr16[0], 0), 3113 ao.addr16[1], an->addr16[1], 0); 3114 *p = pn; 3115 3116 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3117 ao.addr16[0], an->addr16[0], u), 3118 ao.addr16[1], an->addr16[1], u); 3119 3120 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3121 break; 3122 #endif /* INET */ 3123 #ifdef INET6 3124 case AF_INET6: 3125 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3126 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3127 pf_cksum_fixup(pf_cksum_fixup(*pc, 3128 ao.addr16[0], an->addr16[0], u), 3129 ao.addr16[1], an->addr16[1], u), 3130 ao.addr16[2], an->addr16[2], u), 3131 ao.addr16[3], an->addr16[3], u), 3132 ao.addr16[4], an->addr16[4], u), 3133 ao.addr16[5], an->addr16[5], u), 3134 ao.addr16[6], an->addr16[6], u), 3135 ao.addr16[7], an->addr16[7], u); 3136 3137 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3138 break; 3139 #endif /* INET6 */ 3140 } 3141 3142 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3143 CSUM_DELAY_DATA_IPV6)) { 3144 *pc = ~*pc; 3145 if (! *pc) 3146 *pc = 0xffff; 3147 } 3148 } 3149 3150 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3151 void 3152 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3153 { 3154 u_int32_t ao; 3155 3156 memcpy(&ao, a, sizeof(ao)); 3157 memcpy(a, &an, sizeof(u_int32_t)); 3158 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3159 ao % 65536, an % 65536, u); 3160 } 3161 3162 void 3163 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3164 { 3165 u_int32_t ao; 3166 3167 memcpy(&ao, a, sizeof(ao)); 3168 memcpy(a, &an, sizeof(u_int32_t)); 3169 3170 *c = pf_proto_cksum_fixup(m, 3171 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3172 ao % 65536, an % 65536, udp); 3173 } 3174 3175 #ifdef INET6 3176 static void 3177 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3178 { 3179 struct pf_addr ao; 3180 3181 PF_ACPY(&ao, a, AF_INET6); 3182 PF_ACPY(a, an, AF_INET6); 3183 3184 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3185 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3186 pf_cksum_fixup(pf_cksum_fixup(*c, 3187 ao.addr16[0], an->addr16[0], u), 3188 ao.addr16[1], an->addr16[1], u), 3189 ao.addr16[2], an->addr16[2], u), 3190 ao.addr16[3], an->addr16[3], u), 3191 ao.addr16[4], an->addr16[4], u), 3192 ao.addr16[5], an->addr16[5], u), 3193 ao.addr16[6], an->addr16[6], u), 3194 ao.addr16[7], an->addr16[7], u); 3195 } 3196 #endif /* INET6 */ 3197 3198 static void 3199 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3200 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3201 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3202 { 3203 struct pf_addr oia, ooa; 3204 3205 PF_ACPY(&oia, ia, af); 3206 if (oa) 3207 PF_ACPY(&ooa, oa, af); 3208 3209 /* Change inner protocol port, fix inner protocol checksum. */ 3210 if (ip != NULL) { 3211 u_int16_t oip = *ip; 3212 u_int32_t opc; 3213 3214 if (pc != NULL) 3215 opc = *pc; 3216 *ip = np; 3217 if (pc != NULL) 3218 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3219 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3220 if (pc != NULL) 3221 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3222 } 3223 /* Change inner ip address, fix inner ip and icmp checksums. */ 3224 PF_ACPY(ia, na, af); 3225 switch (af) { 3226 #ifdef INET 3227 case AF_INET: { 3228 u_int32_t oh2c = *h2c; 3229 3230 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3231 oia.addr16[0], ia->addr16[0], 0), 3232 oia.addr16[1], ia->addr16[1], 0); 3233 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3234 oia.addr16[0], ia->addr16[0], 0), 3235 oia.addr16[1], ia->addr16[1], 0); 3236 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3237 break; 3238 } 3239 #endif /* INET */ 3240 #ifdef INET6 3241 case AF_INET6: 3242 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3243 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3244 pf_cksum_fixup(pf_cksum_fixup(*ic, 3245 oia.addr16[0], ia->addr16[0], u), 3246 oia.addr16[1], ia->addr16[1], u), 3247 oia.addr16[2], ia->addr16[2], u), 3248 oia.addr16[3], ia->addr16[3], u), 3249 oia.addr16[4], ia->addr16[4], u), 3250 oia.addr16[5], ia->addr16[5], u), 3251 oia.addr16[6], ia->addr16[6], u), 3252 oia.addr16[7], ia->addr16[7], u); 3253 break; 3254 #endif /* INET6 */ 3255 } 3256 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3257 if (oa) { 3258 PF_ACPY(oa, na, af); 3259 switch (af) { 3260 #ifdef INET 3261 case AF_INET: 3262 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3263 ooa.addr16[0], oa->addr16[0], 0), 3264 ooa.addr16[1], oa->addr16[1], 0); 3265 break; 3266 #endif /* INET */ 3267 #ifdef INET6 3268 case AF_INET6: 3269 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3270 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3271 pf_cksum_fixup(pf_cksum_fixup(*ic, 3272 ooa.addr16[0], oa->addr16[0], u), 3273 ooa.addr16[1], oa->addr16[1], u), 3274 ooa.addr16[2], oa->addr16[2], u), 3275 ooa.addr16[3], oa->addr16[3], u), 3276 ooa.addr16[4], oa->addr16[4], u), 3277 ooa.addr16[5], oa->addr16[5], u), 3278 ooa.addr16[6], oa->addr16[6], u), 3279 ooa.addr16[7], oa->addr16[7], u); 3280 break; 3281 #endif /* INET6 */ 3282 } 3283 } 3284 } 3285 3286 /* 3287 * Need to modulate the sequence numbers in the TCP SACK option 3288 * (credits to Krzysztof Pfaff for report and patch) 3289 */ 3290 static int 3291 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 3292 struct tcphdr *th, struct pf_state_peer *dst) 3293 { 3294 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3295 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3296 int copyback = 0, i, olen; 3297 struct sackblk sack; 3298 3299 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3300 if (hlen < TCPOLEN_SACKLEN || 3301 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3302 return 0; 3303 3304 while (hlen >= TCPOLEN_SACKLEN) { 3305 size_t startoff = opt - opts; 3306 olen = opt[1]; 3307 switch (*opt) { 3308 case TCPOPT_EOL: /* FALLTHROUGH */ 3309 case TCPOPT_NOP: 3310 opt++; 3311 hlen--; 3312 break; 3313 case TCPOPT_SACK: 3314 if (olen > hlen) 3315 olen = hlen; 3316 if (olen >= TCPOLEN_SACKLEN) { 3317 for (i = 2; i + TCPOLEN_SACK <= olen; 3318 i += TCPOLEN_SACK) { 3319 memcpy(&sack, &opt[i], sizeof(sack)); 3320 pf_patch_32_unaligned(m, 3321 &th->th_sum, &sack.start, 3322 htonl(ntohl(sack.start) - dst->seqdiff), 3323 PF_ALGNMNT(startoff), 3324 0); 3325 pf_patch_32_unaligned(m, &th->th_sum, 3326 &sack.end, 3327 htonl(ntohl(sack.end) - dst->seqdiff), 3328 PF_ALGNMNT(startoff), 3329 0); 3330 memcpy(&opt[i], &sack, sizeof(sack)); 3331 } 3332 copyback = 1; 3333 } 3334 /* FALLTHROUGH */ 3335 default: 3336 if (olen < 2) 3337 olen = 2; 3338 hlen -= olen; 3339 opt += olen; 3340 } 3341 } 3342 3343 if (copyback) 3344 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 3345 return (copyback); 3346 } 3347 3348 struct mbuf * 3349 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3350 const struct pf_addr *saddr, const struct pf_addr *daddr, 3351 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3352 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3353 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3354 { 3355 struct mbuf *m; 3356 int len, tlen; 3357 #ifdef INET 3358 struct ip *h = NULL; 3359 #endif /* INET */ 3360 #ifdef INET6 3361 struct ip6_hdr *h6 = NULL; 3362 #endif /* INET6 */ 3363 struct tcphdr *th; 3364 char *opt; 3365 struct pf_mtag *pf_mtag; 3366 3367 len = 0; 3368 th = NULL; 3369 3370 /* maximum segment size tcp option */ 3371 tlen = sizeof(struct tcphdr); 3372 if (mss) 3373 tlen += 4; 3374 3375 switch (af) { 3376 #ifdef INET 3377 case AF_INET: 3378 len = sizeof(struct ip) + tlen; 3379 break; 3380 #endif /* INET */ 3381 #ifdef INET6 3382 case AF_INET6: 3383 len = sizeof(struct ip6_hdr) + tlen; 3384 break; 3385 #endif /* INET6 */ 3386 default: 3387 panic("%s: unsupported af %d", __func__, af); 3388 } 3389 3390 m = m_gethdr(M_NOWAIT, MT_DATA); 3391 if (m == NULL) 3392 return (NULL); 3393 3394 #ifdef MAC 3395 mac_netinet_firewall_send(m); 3396 #endif 3397 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3398 m_freem(m); 3399 return (NULL); 3400 } 3401 if (skip_firewall) 3402 m->m_flags |= M_SKIP_FIREWALL; 3403 pf_mtag->tag = mtag_tag; 3404 pf_mtag->flags = mtag_flags; 3405 3406 if (rtableid >= 0) 3407 M_SETFIB(m, rtableid); 3408 3409 #ifdef ALTQ 3410 if (r != NULL && r->qid) { 3411 pf_mtag->qid = r->qid; 3412 3413 /* add hints for ecn */ 3414 pf_mtag->hdr = mtod(m, struct ip *); 3415 } 3416 #endif /* ALTQ */ 3417 m->m_data += max_linkhdr; 3418 m->m_pkthdr.len = m->m_len = len; 3419 /* The rest of the stack assumes a rcvif, so provide one. 3420 * This is a locally generated packet, so .. close enough. */ 3421 m->m_pkthdr.rcvif = V_loif; 3422 bzero(m->m_data, len); 3423 switch (af) { 3424 #ifdef INET 3425 case AF_INET: 3426 h = mtod(m, struct ip *); 3427 3428 /* IP header fields included in the TCP checksum */ 3429 h->ip_p = IPPROTO_TCP; 3430 h->ip_len = htons(tlen); 3431 h->ip_src.s_addr = saddr->v4.s_addr; 3432 h->ip_dst.s_addr = daddr->v4.s_addr; 3433 3434 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3435 break; 3436 #endif /* INET */ 3437 #ifdef INET6 3438 case AF_INET6: 3439 h6 = mtod(m, struct ip6_hdr *); 3440 3441 /* IP header fields included in the TCP checksum */ 3442 h6->ip6_nxt = IPPROTO_TCP; 3443 h6->ip6_plen = htons(tlen); 3444 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3445 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3446 3447 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3448 break; 3449 #endif /* INET6 */ 3450 } 3451 3452 /* TCP header */ 3453 th->th_sport = sport; 3454 th->th_dport = dport; 3455 th->th_seq = htonl(seq); 3456 th->th_ack = htonl(ack); 3457 th->th_off = tlen >> 2; 3458 th->th_flags = tcp_flags; 3459 th->th_win = htons(win); 3460 3461 if (mss) { 3462 opt = (char *)(th + 1); 3463 opt[0] = TCPOPT_MAXSEG; 3464 opt[1] = 4; 3465 HTONS(mss); 3466 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3467 } 3468 3469 switch (af) { 3470 #ifdef INET 3471 case AF_INET: 3472 /* TCP checksum */ 3473 th->th_sum = in_cksum(m, len); 3474 3475 /* Finish the IP header */ 3476 h->ip_v = 4; 3477 h->ip_hl = sizeof(*h) >> 2; 3478 h->ip_tos = IPTOS_LOWDELAY; 3479 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3480 h->ip_len = htons(len); 3481 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3482 h->ip_sum = 0; 3483 break; 3484 #endif /* INET */ 3485 #ifdef INET6 3486 case AF_INET6: 3487 /* TCP checksum */ 3488 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3489 sizeof(struct ip6_hdr), tlen); 3490 3491 h6->ip6_vfc |= IPV6_VERSION; 3492 h6->ip6_hlim = IPV6_DEFHLIM; 3493 break; 3494 #endif /* INET6 */ 3495 } 3496 3497 return (m); 3498 } 3499 3500 static void 3501 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3502 uint8_t ttl, int rtableid) 3503 { 3504 struct mbuf *m; 3505 #ifdef INET 3506 struct ip *h = NULL; 3507 #endif /* INET */ 3508 #ifdef INET6 3509 struct ip6_hdr *h6 = NULL; 3510 #endif /* INET6 */ 3511 struct sctphdr *hdr; 3512 struct sctp_chunkhdr *chunk; 3513 struct pf_send_entry *pfse; 3514 int off = 0; 3515 3516 MPASS(af == pd->af); 3517 3518 m = m_gethdr(M_NOWAIT, MT_DATA); 3519 if (m == NULL) 3520 return; 3521 3522 m->m_data += max_linkhdr; 3523 m->m_flags |= M_SKIP_FIREWALL; 3524 /* The rest of the stack assumes a rcvif, so provide one. 3525 * This is a locally generated packet, so .. close enough. */ 3526 m->m_pkthdr.rcvif = V_loif; 3527 3528 /* IPv4|6 header */ 3529 switch (af) { 3530 #ifdef INET 3531 case AF_INET: 3532 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3533 3534 h = mtod(m, struct ip *); 3535 3536 /* IP header fields included in the TCP checksum */ 3537 3538 h->ip_p = IPPROTO_SCTP; 3539 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3540 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3541 h->ip_src = pd->dst->v4; 3542 h->ip_dst = pd->src->v4; 3543 3544 off += sizeof(struct ip); 3545 break; 3546 #endif /* INET */ 3547 #ifdef INET6 3548 case AF_INET6: 3549 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3550 3551 h6 = mtod(m, struct ip6_hdr *); 3552 3553 /* IP header fields included in the TCP checksum */ 3554 h6->ip6_vfc |= IPV6_VERSION; 3555 h6->ip6_nxt = IPPROTO_SCTP; 3556 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3557 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3558 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3559 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3560 3561 off += sizeof(struct ip6_hdr); 3562 break; 3563 #endif /* INET6 */ 3564 } 3565 3566 /* SCTP header */ 3567 hdr = mtodo(m, off); 3568 3569 hdr->src_port = pd->hdr.sctp.dest_port; 3570 hdr->dest_port = pd->hdr.sctp.src_port; 3571 hdr->v_tag = pd->sctp_initiate_tag; 3572 hdr->checksum = 0; 3573 3574 /* Abort chunk. */ 3575 off += sizeof(struct sctphdr); 3576 chunk = mtodo(m, off); 3577 3578 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3579 chunk->chunk_length = htons(sizeof(*chunk)); 3580 3581 /* SCTP checksum */ 3582 off += sizeof(*chunk); 3583 m->m_pkthdr.len = m->m_len = off; 3584 3585 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3586 3587 if (rtableid >= 0) 3588 M_SETFIB(m, rtableid); 3589 3590 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3591 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3592 if (pfse == NULL) { 3593 m_freem(m); 3594 return; 3595 } 3596 3597 switch (af) { 3598 #ifdef INET 3599 case AF_INET: 3600 pfse->pfse_type = PFSE_IP; 3601 break; 3602 #endif /* INET */ 3603 #ifdef INET6 3604 case AF_INET6: 3605 pfse->pfse_type = PFSE_IP6; 3606 break; 3607 #endif /* INET6 */ 3608 } 3609 3610 pfse->pfse_m = m; 3611 pf_send(pfse); 3612 } 3613 3614 void 3615 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3616 const struct pf_addr *saddr, const struct pf_addr *daddr, 3617 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3618 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3619 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3620 { 3621 struct pf_send_entry *pfse; 3622 struct mbuf *m; 3623 3624 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3625 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3626 if (m == NULL) 3627 return; 3628 3629 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3630 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3631 if (pfse == NULL) { 3632 m_freem(m); 3633 return; 3634 } 3635 3636 switch (af) { 3637 #ifdef INET 3638 case AF_INET: 3639 pfse->pfse_type = PFSE_IP; 3640 break; 3641 #endif /* INET */ 3642 #ifdef INET6 3643 case AF_INET6: 3644 pfse->pfse_type = PFSE_IP6; 3645 break; 3646 #endif /* INET6 */ 3647 } 3648 3649 pfse->pfse_m = m; 3650 pf_send(pfse); 3651 } 3652 3653 static void 3654 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3655 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3656 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3657 u_short *reason, int rtableid) 3658 { 3659 struct pf_addr * const saddr = pd->src; 3660 struct pf_addr * const daddr = pd->dst; 3661 sa_family_t af = pd->af; 3662 3663 /* undo NAT changes, if they have taken place */ 3664 if (nr != NULL) { 3665 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3666 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3667 if (pd->sport) 3668 *pd->sport = sk->port[pd->sidx]; 3669 if (pd->dport) 3670 *pd->dport = sk->port[pd->didx]; 3671 if (pd->proto_sum) 3672 *pd->proto_sum = bproto_sum; 3673 if (pd->ip_sum) 3674 *pd->ip_sum = bip_sum; 3675 m_copyback(m, off, hdrlen, pd->hdr.any); 3676 } 3677 if (pd->proto == IPPROTO_TCP && 3678 ((r->rule_flag & PFRULE_RETURNRST) || 3679 (r->rule_flag & PFRULE_RETURN)) && 3680 !(th->th_flags & TH_RST)) { 3681 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3682 int len = 0; 3683 #ifdef INET 3684 struct ip *h4; 3685 #endif 3686 #ifdef INET6 3687 struct ip6_hdr *h6; 3688 #endif 3689 3690 switch (af) { 3691 #ifdef INET 3692 case AF_INET: 3693 h4 = mtod(m, struct ip *); 3694 len = ntohs(h4->ip_len) - off; 3695 break; 3696 #endif 3697 #ifdef INET6 3698 case AF_INET6: 3699 h6 = mtod(m, struct ip6_hdr *); 3700 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3701 break; 3702 #endif 3703 } 3704 3705 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3706 REASON_SET(reason, PFRES_PROTCKSUM); 3707 else { 3708 if (th->th_flags & TH_SYN) 3709 ack++; 3710 if (th->th_flags & TH_FIN) 3711 ack++; 3712 pf_send_tcp(r, af, pd->dst, 3713 pd->src, th->th_dport, th->th_sport, 3714 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3715 r->return_ttl, true, 0, 0, rtableid); 3716 } 3717 } else if (pd->proto == IPPROTO_SCTP && 3718 (r->rule_flag & PFRULE_RETURN)) { 3719 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3720 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3721 r->return_icmp) 3722 pf_send_icmp(m, r->return_icmp >> 8, 3723 r->return_icmp & 255, af, r, rtableid); 3724 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3725 r->return_icmp6) 3726 pf_send_icmp(m, r->return_icmp6 >> 8, 3727 r->return_icmp6 & 255, af, r, rtableid); 3728 } 3729 3730 static int 3731 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3732 { 3733 struct m_tag *mtag; 3734 u_int8_t mpcp; 3735 3736 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3737 if (mtag == NULL) 3738 return (0); 3739 3740 if (prio == PF_PRIO_ZERO) 3741 prio = 0; 3742 3743 mpcp = *(uint8_t *)(mtag + 1); 3744 3745 return (mpcp == prio); 3746 } 3747 3748 static int 3749 pf_icmp_to_bandlim(uint8_t type) 3750 { 3751 switch (type) { 3752 case ICMP_ECHO: 3753 case ICMP_ECHOREPLY: 3754 return (BANDLIM_ICMP_ECHO); 3755 case ICMP_TSTAMP: 3756 case ICMP_TSTAMPREPLY: 3757 return (BANDLIM_ICMP_TSTAMP); 3758 case ICMP_UNREACH: 3759 default: 3760 return (BANDLIM_ICMP_UNREACH); 3761 } 3762 } 3763 3764 static void 3765 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3766 struct pf_krule *r, int rtableid) 3767 { 3768 struct pf_send_entry *pfse; 3769 struct mbuf *m0; 3770 struct pf_mtag *pf_mtag; 3771 3772 /* ICMP packet rate limitation. */ 3773 #ifdef INET6 3774 if (af == AF_INET6) { 3775 if (icmp6_ratelimit(NULL, type, code)) 3776 return; 3777 } 3778 #endif 3779 #ifdef INET 3780 if (af == AF_INET) { 3781 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3782 return; 3783 } 3784 #endif 3785 3786 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3787 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3788 if (pfse == NULL) 3789 return; 3790 3791 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3792 free(pfse, M_PFTEMP); 3793 return; 3794 } 3795 3796 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3797 free(pfse, M_PFTEMP); 3798 return; 3799 } 3800 /* XXX: revisit */ 3801 m0->m_flags |= M_SKIP_FIREWALL; 3802 3803 if (rtableid >= 0) 3804 M_SETFIB(m0, rtableid); 3805 3806 #ifdef ALTQ 3807 if (r->qid) { 3808 pf_mtag->qid = r->qid; 3809 /* add hints for ecn */ 3810 pf_mtag->hdr = mtod(m0, struct ip *); 3811 } 3812 #endif /* ALTQ */ 3813 3814 switch (af) { 3815 #ifdef INET 3816 case AF_INET: 3817 pfse->pfse_type = PFSE_ICMP; 3818 break; 3819 #endif /* INET */ 3820 #ifdef INET6 3821 case AF_INET6: 3822 pfse->pfse_type = PFSE_ICMP6; 3823 break; 3824 #endif /* INET6 */ 3825 } 3826 pfse->pfse_m = m0; 3827 pfse->icmpopts.type = type; 3828 pfse->icmpopts.code = code; 3829 pf_send(pfse); 3830 } 3831 3832 /* 3833 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3834 * If n is 0, they match if they are equal. If n is != 0, they match if they 3835 * are different. 3836 */ 3837 int 3838 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3839 struct pf_addr *b, sa_family_t af) 3840 { 3841 int match = 0; 3842 3843 switch (af) { 3844 #ifdef INET 3845 case AF_INET: 3846 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3847 match++; 3848 break; 3849 #endif /* INET */ 3850 #ifdef INET6 3851 case AF_INET6: 3852 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3853 match++; 3854 break; 3855 #endif /* INET6 */ 3856 } 3857 if (match) { 3858 if (n) 3859 return (0); 3860 else 3861 return (1); 3862 } else { 3863 if (n) 3864 return (1); 3865 else 3866 return (0); 3867 } 3868 } 3869 3870 /* 3871 * Return 1 if b <= a <= e, otherwise return 0. 3872 */ 3873 int 3874 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3875 struct pf_addr *a, sa_family_t af) 3876 { 3877 switch (af) { 3878 #ifdef INET 3879 case AF_INET: 3880 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3881 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3882 return (0); 3883 break; 3884 #endif /* INET */ 3885 #ifdef INET6 3886 case AF_INET6: { 3887 int i; 3888 3889 /* check a >= b */ 3890 for (i = 0; i < 4; ++i) 3891 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3892 break; 3893 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3894 return (0); 3895 /* check a <= e */ 3896 for (i = 0; i < 4; ++i) 3897 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3898 break; 3899 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3900 return (0); 3901 break; 3902 } 3903 #endif /* INET6 */ 3904 } 3905 return (1); 3906 } 3907 3908 static int 3909 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3910 { 3911 switch (op) { 3912 case PF_OP_IRG: 3913 return ((p > a1) && (p < a2)); 3914 case PF_OP_XRG: 3915 return ((p < a1) || (p > a2)); 3916 case PF_OP_RRG: 3917 return ((p >= a1) && (p <= a2)); 3918 case PF_OP_EQ: 3919 return (p == a1); 3920 case PF_OP_NE: 3921 return (p != a1); 3922 case PF_OP_LT: 3923 return (p < a1); 3924 case PF_OP_LE: 3925 return (p <= a1); 3926 case PF_OP_GT: 3927 return (p > a1); 3928 case PF_OP_GE: 3929 return (p >= a1); 3930 } 3931 return (0); /* never reached */ 3932 } 3933 3934 int 3935 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3936 { 3937 NTOHS(a1); 3938 NTOHS(a2); 3939 NTOHS(p); 3940 return (pf_match(op, a1, a2, p)); 3941 } 3942 3943 static int 3944 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3945 { 3946 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3947 return (0); 3948 return (pf_match(op, a1, a2, u)); 3949 } 3950 3951 static int 3952 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3953 { 3954 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3955 return (0); 3956 return (pf_match(op, a1, a2, g)); 3957 } 3958 3959 int 3960 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3961 { 3962 if (*tag == -1) 3963 *tag = mtag; 3964 3965 return ((!r->match_tag_not && r->match_tag == *tag) || 3966 (r->match_tag_not && r->match_tag != *tag)); 3967 } 3968 3969 static int 3970 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3971 { 3972 struct ifnet *ifp = m->m_pkthdr.rcvif; 3973 struct pfi_kkif *kif; 3974 3975 if (ifp == NULL) 3976 return (0); 3977 3978 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3979 3980 if (kif == NULL) { 3981 DPFPRINTF(PF_DEBUG_URGENT, 3982 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3983 r->rcv_ifname)); 3984 return (0); 3985 } 3986 3987 return (pfi_kkif_match(r->rcv_kif, kif)); 3988 } 3989 3990 int 3991 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3992 { 3993 3994 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3995 3996 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3997 return (ENOMEM); 3998 3999 pd->pf_mtag->tag = tag; 4000 4001 return (0); 4002 } 4003 4004 #define PF_ANCHOR_STACKSIZE 32 4005 struct pf_kanchor_stackframe { 4006 struct pf_kruleset *rs; 4007 struct pf_krule *r; /* XXX: + match bit */ 4008 struct pf_kanchor *child; 4009 }; 4010 4011 /* 4012 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4013 */ 4014 #define PF_ANCHORSTACK_MATCH 0x00000001 4015 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4016 4017 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4018 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4019 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4020 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4021 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4022 } while (0) 4023 4024 void 4025 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4026 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4027 int *match) 4028 { 4029 struct pf_kanchor_stackframe *f; 4030 4031 PF_RULES_RASSERT(); 4032 4033 if (match) 4034 *match = 0; 4035 if (*depth >= PF_ANCHOR_STACKSIZE) { 4036 printf("%s: anchor stack overflow on %s\n", 4037 __func__, (*r)->anchor->name); 4038 *r = TAILQ_NEXT(*r, entries); 4039 return; 4040 } else if (*depth == 0 && a != NULL) 4041 *a = *r; 4042 f = stack + (*depth)++; 4043 f->rs = *rs; 4044 f->r = *r; 4045 if ((*r)->anchor_wildcard) { 4046 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4047 4048 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4049 *r = NULL; 4050 return; 4051 } 4052 *rs = &f->child->ruleset; 4053 } else { 4054 f->child = NULL; 4055 *rs = &(*r)->anchor->ruleset; 4056 } 4057 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4058 } 4059 4060 int 4061 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4062 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4063 int *match) 4064 { 4065 struct pf_kanchor_stackframe *f; 4066 struct pf_krule *fr; 4067 int quick = 0; 4068 4069 PF_RULES_RASSERT(); 4070 4071 do { 4072 if (*depth <= 0) 4073 break; 4074 f = stack + *depth - 1; 4075 fr = PF_ANCHOR_RULE(f); 4076 if (f->child != NULL) { 4077 /* 4078 * This block traverses through 4079 * a wildcard anchor. 4080 */ 4081 if (match != NULL && *match) { 4082 /* 4083 * If any of "*" matched, then 4084 * "foo/ *" matched, mark frame 4085 * appropriately. 4086 */ 4087 PF_ANCHOR_SET_MATCH(f); 4088 *match = 0; 4089 } 4090 f->child = RB_NEXT(pf_kanchor_node, 4091 &fr->anchor->children, f->child); 4092 if (f->child != NULL) { 4093 *rs = &f->child->ruleset; 4094 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4095 if (*r == NULL) 4096 continue; 4097 else 4098 break; 4099 } 4100 } 4101 (*depth)--; 4102 if (*depth == 0 && a != NULL) 4103 *a = NULL; 4104 *rs = f->rs; 4105 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4106 quick = fr->quick; 4107 *r = TAILQ_NEXT(fr, entries); 4108 } while (*r == NULL); 4109 4110 return (quick); 4111 } 4112 4113 struct pf_keth_anchor_stackframe { 4114 struct pf_keth_ruleset *rs; 4115 struct pf_keth_rule *r; /* XXX: + match bit */ 4116 struct pf_keth_anchor *child; 4117 }; 4118 4119 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4120 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4121 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4122 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4123 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4124 } while (0) 4125 4126 void 4127 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4128 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4129 struct pf_keth_rule **a, int *match) 4130 { 4131 struct pf_keth_anchor_stackframe *f; 4132 4133 NET_EPOCH_ASSERT(); 4134 4135 if (match) 4136 *match = 0; 4137 if (*depth >= PF_ANCHOR_STACKSIZE) { 4138 printf("%s: anchor stack overflow on %s\n", 4139 __func__, (*r)->anchor->name); 4140 *r = TAILQ_NEXT(*r, entries); 4141 return; 4142 } else if (*depth == 0 && a != NULL) 4143 *a = *r; 4144 f = stack + (*depth)++; 4145 f->rs = *rs; 4146 f->r = *r; 4147 if ((*r)->anchor_wildcard) { 4148 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4149 4150 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4151 *r = NULL; 4152 return; 4153 } 4154 *rs = &f->child->ruleset; 4155 } else { 4156 f->child = NULL; 4157 *rs = &(*r)->anchor->ruleset; 4158 } 4159 *r = TAILQ_FIRST((*rs)->active.rules); 4160 } 4161 4162 int 4163 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4164 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4165 struct pf_keth_rule **a, int *match) 4166 { 4167 struct pf_keth_anchor_stackframe *f; 4168 struct pf_keth_rule *fr; 4169 int quick = 0; 4170 4171 NET_EPOCH_ASSERT(); 4172 4173 do { 4174 if (*depth <= 0) 4175 break; 4176 f = stack + *depth - 1; 4177 fr = PF_ETH_ANCHOR_RULE(f); 4178 if (f->child != NULL) { 4179 /* 4180 * This block traverses through 4181 * a wildcard anchor. 4182 */ 4183 if (match != NULL && *match) { 4184 /* 4185 * If any of "*" matched, then 4186 * "foo/ *" matched, mark frame 4187 * appropriately. 4188 */ 4189 PF_ETH_ANCHOR_SET_MATCH(f); 4190 *match = 0; 4191 } 4192 f->child = RB_NEXT(pf_keth_anchor_node, 4193 &fr->anchor->children, f->child); 4194 if (f->child != NULL) { 4195 *rs = &f->child->ruleset; 4196 *r = TAILQ_FIRST((*rs)->active.rules); 4197 if (*r == NULL) 4198 continue; 4199 else 4200 break; 4201 } 4202 } 4203 (*depth)--; 4204 if (*depth == 0 && a != NULL) 4205 *a = NULL; 4206 *rs = f->rs; 4207 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4208 quick = fr->quick; 4209 *r = TAILQ_NEXT(fr, entries); 4210 } while (*r == NULL); 4211 4212 return (quick); 4213 } 4214 4215 #ifdef INET6 4216 void 4217 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4218 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4219 { 4220 switch (af) { 4221 #ifdef INET 4222 case AF_INET: 4223 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4224 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4225 break; 4226 #endif /* INET */ 4227 case AF_INET6: 4228 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4229 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4230 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4231 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4232 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4233 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4234 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4235 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4236 break; 4237 } 4238 } 4239 4240 void 4241 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4242 { 4243 switch (af) { 4244 #ifdef INET 4245 case AF_INET: 4246 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4247 break; 4248 #endif /* INET */ 4249 case AF_INET6: 4250 if (addr->addr32[3] == 0xffffffff) { 4251 addr->addr32[3] = 0; 4252 if (addr->addr32[2] == 0xffffffff) { 4253 addr->addr32[2] = 0; 4254 if (addr->addr32[1] == 0xffffffff) { 4255 addr->addr32[1] = 0; 4256 addr->addr32[0] = 4257 htonl(ntohl(addr->addr32[0]) + 1); 4258 } else 4259 addr->addr32[1] = 4260 htonl(ntohl(addr->addr32[1]) + 1); 4261 } else 4262 addr->addr32[2] = 4263 htonl(ntohl(addr->addr32[2]) + 1); 4264 } else 4265 addr->addr32[3] = 4266 htonl(ntohl(addr->addr32[3]) + 1); 4267 break; 4268 } 4269 } 4270 #endif /* INET6 */ 4271 4272 void 4273 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4274 { 4275 /* 4276 * Modern rules use the same flags in rules as they do in states. 4277 */ 4278 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4279 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4280 4281 /* 4282 * Old-style scrub rules have different flags which need to be translated. 4283 */ 4284 if (r->rule_flag & PFRULE_RANDOMID) 4285 a->flags |= PFSTATE_RANDOMID; 4286 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4287 a->flags |= PFSTATE_SETTOS; 4288 a->set_tos = r->set_tos; 4289 } 4290 4291 if (r->qid) 4292 a->qid = r->qid; 4293 if (r->pqid) 4294 a->pqid = r->pqid; 4295 if (r->rtableid >= 0) 4296 a->rtableid = r->rtableid; 4297 a->log |= r->log; 4298 if (r->min_ttl) 4299 a->min_ttl = r->min_ttl; 4300 if (r->max_mss) 4301 a->max_mss = r->max_mss; 4302 if (r->dnpipe) 4303 a->dnpipe = r->dnpipe; 4304 if (r->dnrpipe) 4305 a->dnrpipe = r->dnrpipe; 4306 if (r->dnpipe || r->dnrpipe) { 4307 if (r->free_flags & PFRULE_DN_IS_PIPE) 4308 a->flags |= PFSTATE_DN_IS_PIPE; 4309 else 4310 a->flags &= ~PFSTATE_DN_IS_PIPE; 4311 } 4312 if (r->scrub_flags & PFSTATE_SETPRIO) { 4313 a->set_prio[0] = r->set_prio[0]; 4314 a->set_prio[1] = r->set_prio[1]; 4315 } 4316 } 4317 4318 int 4319 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 4320 { 4321 struct pf_addr *saddr, *daddr; 4322 u_int16_t sport, dport; 4323 struct inpcbinfo *pi; 4324 struct inpcb *inp; 4325 4326 pd->lookup.uid = UID_MAX; 4327 pd->lookup.gid = GID_MAX; 4328 4329 switch (pd->proto) { 4330 case IPPROTO_TCP: 4331 sport = pd->hdr.tcp.th_sport; 4332 dport = pd->hdr.tcp.th_dport; 4333 pi = &V_tcbinfo; 4334 break; 4335 case IPPROTO_UDP: 4336 sport = pd->hdr.udp.uh_sport; 4337 dport = pd->hdr.udp.uh_dport; 4338 pi = &V_udbinfo; 4339 break; 4340 default: 4341 return (-1); 4342 } 4343 if (pd->dir == PF_IN) { 4344 saddr = pd->src; 4345 daddr = pd->dst; 4346 } else { 4347 u_int16_t p; 4348 4349 p = sport; 4350 sport = dport; 4351 dport = p; 4352 saddr = pd->dst; 4353 daddr = pd->src; 4354 } 4355 switch (pd->af) { 4356 #ifdef INET 4357 case AF_INET: 4358 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4359 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4360 if (inp == NULL) { 4361 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4362 daddr->v4, dport, INPLOOKUP_WILDCARD | 4363 INPLOOKUP_RLOCKPCB, NULL, m); 4364 if (inp == NULL) 4365 return (-1); 4366 } 4367 break; 4368 #endif /* INET */ 4369 #ifdef INET6 4370 case AF_INET6: 4371 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4372 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4373 if (inp == NULL) { 4374 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4375 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4376 INPLOOKUP_RLOCKPCB, NULL, m); 4377 if (inp == NULL) 4378 return (-1); 4379 } 4380 break; 4381 #endif /* INET6 */ 4382 4383 default: 4384 return (-1); 4385 } 4386 INP_RLOCK_ASSERT(inp); 4387 pd->lookup.uid = inp->inp_cred->cr_uid; 4388 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4389 INP_RUNLOCK(inp); 4390 4391 return (1); 4392 } 4393 4394 u_int8_t 4395 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4396 { 4397 int hlen; 4398 u_int8_t hdr[60]; 4399 u_int8_t *opt, optlen; 4400 u_int8_t wscale = 0; 4401 4402 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4403 if (hlen <= sizeof(struct tcphdr)) 4404 return (0); 4405 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4406 return (0); 4407 opt = hdr + sizeof(struct tcphdr); 4408 hlen -= sizeof(struct tcphdr); 4409 while (hlen >= 3) { 4410 switch (*opt) { 4411 case TCPOPT_EOL: 4412 case TCPOPT_NOP: 4413 ++opt; 4414 --hlen; 4415 break; 4416 case TCPOPT_WINDOW: 4417 wscale = opt[2]; 4418 if (wscale > TCP_MAX_WINSHIFT) 4419 wscale = TCP_MAX_WINSHIFT; 4420 wscale |= PF_WSCALE_FLAG; 4421 /* FALLTHROUGH */ 4422 default: 4423 optlen = opt[1]; 4424 if (optlen < 2) 4425 optlen = 2; 4426 hlen -= optlen; 4427 opt += optlen; 4428 break; 4429 } 4430 } 4431 return (wscale); 4432 } 4433 4434 u_int16_t 4435 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4436 { 4437 int hlen; 4438 u_int8_t hdr[60]; 4439 u_int8_t *opt, optlen; 4440 u_int16_t mss = V_tcp_mssdflt; 4441 4442 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4443 if (hlen <= sizeof(struct tcphdr)) 4444 return (0); 4445 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4446 return (0); 4447 opt = hdr + sizeof(struct tcphdr); 4448 hlen -= sizeof(struct tcphdr); 4449 while (hlen >= TCPOLEN_MAXSEG) { 4450 switch (*opt) { 4451 case TCPOPT_EOL: 4452 case TCPOPT_NOP: 4453 ++opt; 4454 --hlen; 4455 break; 4456 case TCPOPT_MAXSEG: 4457 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4458 NTOHS(mss); 4459 /* FALLTHROUGH */ 4460 default: 4461 optlen = opt[1]; 4462 if (optlen < 2) 4463 optlen = 2; 4464 hlen -= optlen; 4465 opt += optlen; 4466 break; 4467 } 4468 } 4469 return (mss); 4470 } 4471 4472 static u_int16_t 4473 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4474 { 4475 struct nhop_object *nh; 4476 #ifdef INET6 4477 struct in6_addr dst6; 4478 uint32_t scopeid; 4479 #endif /* INET6 */ 4480 int hlen = 0; 4481 uint16_t mss = 0; 4482 4483 NET_EPOCH_ASSERT(); 4484 4485 switch (af) { 4486 #ifdef INET 4487 case AF_INET: 4488 hlen = sizeof(struct ip); 4489 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4490 if (nh != NULL) 4491 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4492 break; 4493 #endif /* INET */ 4494 #ifdef INET6 4495 case AF_INET6: 4496 hlen = sizeof(struct ip6_hdr); 4497 in6_splitscope(&addr->v6, &dst6, &scopeid); 4498 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4499 if (nh != NULL) 4500 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4501 break; 4502 #endif /* INET6 */ 4503 } 4504 4505 mss = max(V_tcp_mssdflt, mss); 4506 mss = min(mss, offer); 4507 mss = max(mss, 64); /* sanity - at least max opt space */ 4508 return (mss); 4509 } 4510 4511 static u_int32_t 4512 pf_tcp_iss(struct pf_pdesc *pd) 4513 { 4514 MD5_CTX ctx; 4515 u_int32_t digest[4]; 4516 4517 if (V_pf_tcp_secret_init == 0) { 4518 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4519 MD5Init(&V_pf_tcp_secret_ctx); 4520 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4521 sizeof(V_pf_tcp_secret)); 4522 V_pf_tcp_secret_init = 1; 4523 } 4524 4525 ctx = V_pf_tcp_secret_ctx; 4526 4527 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4528 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4529 if (pd->af == AF_INET6) { 4530 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4531 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4532 } else { 4533 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4534 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4535 } 4536 MD5Final((u_char *)digest, &ctx); 4537 V_pf_tcp_iss_off += 4096; 4538 #define ISN_RANDOM_INCREMENT (4096 - 1) 4539 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4540 V_pf_tcp_iss_off); 4541 #undef ISN_RANDOM_INCREMENT 4542 } 4543 4544 static bool 4545 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4546 { 4547 bool match = true; 4548 4549 /* Always matches if not set */ 4550 if (! r->isset) 4551 return (!r->neg); 4552 4553 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4554 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4555 match = false; 4556 break; 4557 } 4558 } 4559 4560 return (match ^ r->neg); 4561 } 4562 4563 static int 4564 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4565 { 4566 if (*tag == -1) 4567 *tag = mtag; 4568 4569 return ((!r->match_tag_not && r->match_tag == *tag) || 4570 (r->match_tag_not && r->match_tag != *tag)); 4571 } 4572 4573 static void 4574 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4575 { 4576 /* If we don't have the interface drop the packet. */ 4577 if (ifp == NULL) { 4578 m_freem(m); 4579 return; 4580 } 4581 4582 switch (ifp->if_type) { 4583 case IFT_ETHER: 4584 case IFT_XETHER: 4585 case IFT_L2VLAN: 4586 case IFT_BRIDGE: 4587 case IFT_IEEE8023ADLAG: 4588 break; 4589 default: 4590 m_freem(m); 4591 return; 4592 } 4593 4594 ifp->if_transmit(ifp, m); 4595 } 4596 4597 static int 4598 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4599 { 4600 #ifdef INET 4601 struct ip ip; 4602 #endif 4603 #ifdef INET6 4604 struct ip6_hdr ip6; 4605 #endif 4606 struct mbuf *m = *m0; 4607 struct ether_header *e; 4608 struct pf_keth_rule *r, *rm, *a = NULL; 4609 struct pf_keth_ruleset *ruleset = NULL; 4610 struct pf_mtag *mtag; 4611 struct pf_keth_ruleq *rules; 4612 struct pf_addr *src = NULL, *dst = NULL; 4613 struct pfi_kkif *bridge_to; 4614 sa_family_t af = 0; 4615 uint16_t proto; 4616 int asd = 0, match = 0; 4617 int tag = -1; 4618 uint8_t action; 4619 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4620 4621 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4622 NET_EPOCH_ASSERT(); 4623 4624 PF_RULES_RLOCK_TRACKER; 4625 4626 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4627 4628 mtag = pf_find_mtag(m); 4629 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4630 /* Dummynet re-injects packets after they've 4631 * completed their delay. We've already 4632 * processed them, so pass unconditionally. */ 4633 4634 /* But only once. We may see the packet multiple times (e.g. 4635 * PFIL_IN/PFIL_OUT). */ 4636 pf_dummynet_flag_remove(m, mtag); 4637 4638 return (PF_PASS); 4639 } 4640 4641 ruleset = V_pf_keth; 4642 rules = ck_pr_load_ptr(&ruleset->active.rules); 4643 r = TAILQ_FIRST(rules); 4644 rm = NULL; 4645 4646 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4647 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4648 DPFPRINTF(PF_DEBUG_URGENT, 4649 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4650 ", pullup failed\n")); 4651 return (PF_DROP); 4652 } 4653 e = mtod(m, struct ether_header *); 4654 proto = ntohs(e->ether_type); 4655 4656 switch (proto) { 4657 #ifdef INET 4658 case ETHERTYPE_IP: { 4659 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4660 sizeof(ip))) 4661 return (PF_DROP); 4662 4663 af = AF_INET; 4664 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4665 (caddr_t)&ip); 4666 src = (struct pf_addr *)&ip.ip_src; 4667 dst = (struct pf_addr *)&ip.ip_dst; 4668 break; 4669 } 4670 #endif /* INET */ 4671 #ifdef INET6 4672 case ETHERTYPE_IPV6: { 4673 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4674 sizeof(ip6))) 4675 return (PF_DROP); 4676 4677 af = AF_INET6; 4678 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4679 (caddr_t)&ip6); 4680 src = (struct pf_addr *)&ip6.ip6_src; 4681 dst = (struct pf_addr *)&ip6.ip6_dst; 4682 break; 4683 } 4684 #endif /* INET6 */ 4685 } 4686 4687 PF_RULES_RLOCK(); 4688 4689 while (r != NULL) { 4690 counter_u64_add(r->evaluations, 1); 4691 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4692 4693 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4694 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4695 "kif"); 4696 r = r->skip[PFE_SKIP_IFP].ptr; 4697 } 4698 else if (r->direction && r->direction != dir) { 4699 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4700 "dir"); 4701 r = r->skip[PFE_SKIP_DIR].ptr; 4702 } 4703 else if (r->proto && r->proto != proto) { 4704 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4705 "proto"); 4706 r = r->skip[PFE_SKIP_PROTO].ptr; 4707 } 4708 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4709 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4710 "src"); 4711 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4712 } 4713 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4714 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4715 "dst"); 4716 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4717 } 4718 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4719 r->ipsrc.neg, kif, M_GETFIB(m))) { 4720 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4721 "ip_src"); 4722 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4723 } 4724 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4725 r->ipdst.neg, kif, M_GETFIB(m))) { 4726 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4727 "ip_dst"); 4728 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4729 } 4730 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4731 mtag ? mtag->tag : 0)) { 4732 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4733 "match_tag"); 4734 r = TAILQ_NEXT(r, entries); 4735 } 4736 else { 4737 if (r->tag) 4738 tag = r->tag; 4739 if (r->anchor == NULL) { 4740 /* Rule matches */ 4741 rm = r; 4742 4743 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4744 4745 if (r->quick) 4746 break; 4747 4748 r = TAILQ_NEXT(r, entries); 4749 } else { 4750 pf_step_into_keth_anchor(anchor_stack, &asd, 4751 &ruleset, &r, &a, &match); 4752 } 4753 } 4754 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4755 &ruleset, &r, &a, &match)) 4756 break; 4757 } 4758 4759 r = rm; 4760 4761 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4762 4763 /* Default to pass. */ 4764 if (r == NULL) { 4765 PF_RULES_RUNLOCK(); 4766 return (PF_PASS); 4767 } 4768 4769 /* Execute action. */ 4770 counter_u64_add(r->packets[dir == PF_OUT], 1); 4771 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4772 pf_update_timestamp(r); 4773 4774 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4775 if (r->action == PF_DROP) { 4776 PF_RULES_RUNLOCK(); 4777 return (PF_DROP); 4778 } 4779 4780 if (tag > 0) { 4781 if (mtag == NULL) 4782 mtag = pf_get_mtag(m); 4783 if (mtag == NULL) { 4784 PF_RULES_RUNLOCK(); 4785 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4786 return (PF_DROP); 4787 } 4788 mtag->tag = tag; 4789 } 4790 4791 if (r->qid != 0) { 4792 if (mtag == NULL) 4793 mtag = pf_get_mtag(m); 4794 if (mtag == NULL) { 4795 PF_RULES_RUNLOCK(); 4796 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4797 return (PF_DROP); 4798 } 4799 mtag->qid = r->qid; 4800 } 4801 4802 action = r->action; 4803 bridge_to = r->bridge_to; 4804 4805 /* Dummynet */ 4806 if (r->dnpipe) { 4807 struct ip_fw_args dnflow; 4808 4809 /* Drop packet if dummynet is not loaded. */ 4810 if (ip_dn_io_ptr == NULL) { 4811 PF_RULES_RUNLOCK(); 4812 m_freem(m); 4813 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4814 return (PF_DROP); 4815 } 4816 if (mtag == NULL) 4817 mtag = pf_get_mtag(m); 4818 if (mtag == NULL) { 4819 PF_RULES_RUNLOCK(); 4820 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4821 return (PF_DROP); 4822 } 4823 4824 bzero(&dnflow, sizeof(dnflow)); 4825 4826 /* We don't have port numbers here, so we set 0. That means 4827 * that we'll be somewhat limited in distinguishing flows (i.e. 4828 * only based on IP addresses, not based on port numbers), but 4829 * it's better than nothing. */ 4830 dnflow.f_id.dst_port = 0; 4831 dnflow.f_id.src_port = 0; 4832 dnflow.f_id.proto = 0; 4833 4834 dnflow.rule.info = r->dnpipe; 4835 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4836 if (r->dnflags & PFRULE_DN_IS_PIPE) 4837 dnflow.rule.info |= IPFW_IS_PIPE; 4838 4839 dnflow.f_id.extra = dnflow.rule.info; 4840 4841 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4842 dnflow.flags |= IPFW_ARGS_ETHER; 4843 dnflow.ifp = kif->pfik_ifp; 4844 4845 switch (af) { 4846 case AF_INET: 4847 dnflow.f_id.addr_type = 4; 4848 dnflow.f_id.src_ip = src->v4.s_addr; 4849 dnflow.f_id.dst_ip = dst->v4.s_addr; 4850 break; 4851 case AF_INET6: 4852 dnflow.flags |= IPFW_ARGS_IP6; 4853 dnflow.f_id.addr_type = 6; 4854 dnflow.f_id.src_ip6 = src->v6; 4855 dnflow.f_id.dst_ip6 = dst->v6; 4856 break; 4857 } 4858 4859 PF_RULES_RUNLOCK(); 4860 4861 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4862 ip_dn_io_ptr(m0, &dnflow); 4863 if (*m0 != NULL) 4864 pf_dummynet_flag_remove(m, mtag); 4865 } else { 4866 PF_RULES_RUNLOCK(); 4867 } 4868 4869 if (action == PF_PASS && bridge_to) { 4870 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4871 *m0 = NULL; /* We've eaten the packet. */ 4872 } 4873 4874 return (action); 4875 } 4876 4877 #define PF_TEST_ATTRIB(t, a)\ 4878 do { \ 4879 if (t) { \ 4880 r = a; \ 4881 goto nextrule; \ 4882 } \ 4883 } while (0) 4884 4885 static int 4886 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4887 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4888 struct pf_kruleset **rsm, struct inpcb *inp, int hdrlen) 4889 { 4890 struct pf_krule *nr = NULL; 4891 struct pf_addr * const saddr = pd->src; 4892 struct pf_addr * const daddr = pd->dst; 4893 sa_family_t af = pd->af; 4894 struct pf_krule *r, *a = NULL; 4895 struct pf_kruleset *ruleset = NULL; 4896 struct pf_krule_slist match_rules; 4897 struct pf_krule_item *ri; 4898 struct pf_ksrc_node *nsn = NULL; 4899 struct tcphdr *th = &pd->hdr.tcp; 4900 struct pf_state_key *sk = NULL, *nk = NULL; 4901 u_short reason, transerror; 4902 int rewrite = 0; 4903 int tag = -1; 4904 int asd = 0; 4905 int match = 0; 4906 int state_icmp = 0, icmp_dir, multi; 4907 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4908 u_int16_t bproto_sum = 0, bip_sum = 0; 4909 u_int8_t icmptype = 0, icmpcode = 0; 4910 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4911 struct pf_udp_mapping *udp_mapping = NULL; 4912 4913 PF_RULES_RASSERT(); 4914 4915 SLIST_INIT(&match_rules); 4916 4917 if (inp != NULL) { 4918 INP_LOCK_ASSERT(inp); 4919 pd->lookup.uid = inp->inp_cred->cr_uid; 4920 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4921 pd->lookup.done = 1; 4922 } 4923 4924 switch (pd->virtual_proto) { 4925 case IPPROTO_TCP: 4926 sport = th->th_sport; 4927 dport = th->th_dport; 4928 break; 4929 case IPPROTO_UDP: 4930 sport = pd->hdr.udp.uh_sport; 4931 dport = pd->hdr.udp.uh_dport; 4932 break; 4933 case IPPROTO_SCTP: 4934 sport = pd->hdr.sctp.src_port; 4935 dport = pd->hdr.sctp.dest_port; 4936 break; 4937 #ifdef INET 4938 case IPPROTO_ICMP: 4939 MPASS(af == AF_INET); 4940 icmptype = pd->hdr.icmp.icmp_type; 4941 icmpcode = pd->hdr.icmp.icmp_code; 4942 state_icmp = pf_icmp_mapping(pd, icmptype, 4943 &icmp_dir, &multi, &virtual_id, &virtual_type); 4944 if (icmp_dir == PF_IN) { 4945 sport = virtual_id; 4946 dport = virtual_type; 4947 } else { 4948 sport = virtual_type; 4949 dport = virtual_id; 4950 } 4951 break; 4952 #endif /* INET */ 4953 #ifdef INET6 4954 case IPPROTO_ICMPV6: 4955 MPASS(af == AF_INET6); 4956 icmptype = pd->hdr.icmp6.icmp6_type; 4957 icmpcode = pd->hdr.icmp6.icmp6_code; 4958 state_icmp = pf_icmp_mapping(pd, icmptype, 4959 &icmp_dir, &multi, &virtual_id, &virtual_type); 4960 if (icmp_dir == PF_IN) { 4961 sport = virtual_id; 4962 dport = virtual_type; 4963 } else { 4964 sport = virtual_type; 4965 dport = virtual_id; 4966 } 4967 4968 break; 4969 #endif /* INET6 */ 4970 default: 4971 sport = dport = 0; 4972 break; 4973 } 4974 4975 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4976 4977 /* check packet for BINAT/NAT/RDR */ 4978 transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4979 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4980 switch (transerror) { 4981 default: 4982 /* A translation error occurred. */ 4983 REASON_SET(&reason, transerror); 4984 goto cleanup; 4985 case PFRES_MAX: 4986 /* No match. */ 4987 break; 4988 case PFRES_MATCH: 4989 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4990 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4991 4992 if (nr->log) { 4993 PFLOG_PACKET(kif, m, PF_PASS, PFRES_MATCH, nr, a, 4994 ruleset, pd, 1); 4995 } 4996 4997 if (pd->ip_sum) 4998 bip_sum = *pd->ip_sum; 4999 5000 switch (pd->proto) { 5001 case IPPROTO_TCP: 5002 bproto_sum = th->th_sum; 5003 pd->proto_sum = &th->th_sum; 5004 5005 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5006 nk->port[pd->sidx] != sport) { 5007 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 5008 &th->th_sum, &nk->addr[pd->sidx], 5009 nk->port[pd->sidx], 0, af); 5010 pd->sport = &th->th_sport; 5011 sport = th->th_sport; 5012 } 5013 5014 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5015 nk->port[pd->didx] != dport) { 5016 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 5017 &th->th_sum, &nk->addr[pd->didx], 5018 nk->port[pd->didx], 0, af); 5019 dport = th->th_dport; 5020 pd->dport = &th->th_dport; 5021 } 5022 rewrite++; 5023 break; 5024 case IPPROTO_UDP: 5025 bproto_sum = pd->hdr.udp.uh_sum; 5026 pd->proto_sum = &pd->hdr.udp.uh_sum; 5027 5028 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5029 nk->port[pd->sidx] != sport) { 5030 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 5031 pd->ip_sum, &pd->hdr.udp.uh_sum, 5032 &nk->addr[pd->sidx], 5033 nk->port[pd->sidx], 1, af); 5034 sport = pd->hdr.udp.uh_sport; 5035 pd->sport = &pd->hdr.udp.uh_sport; 5036 } 5037 5038 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5039 nk->port[pd->didx] != dport) { 5040 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 5041 pd->ip_sum, &pd->hdr.udp.uh_sum, 5042 &nk->addr[pd->didx], 5043 nk->port[pd->didx], 1, af); 5044 dport = pd->hdr.udp.uh_dport; 5045 pd->dport = &pd->hdr.udp.uh_dport; 5046 } 5047 rewrite++; 5048 break; 5049 case IPPROTO_SCTP: { 5050 uint16_t checksum = 0; 5051 5052 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5053 nk->port[pd->sidx] != sport) { 5054 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 5055 pd->ip_sum, &checksum, 5056 &nk->addr[pd->sidx], 5057 nk->port[pd->sidx], 1, af); 5058 } 5059 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5060 nk->port[pd->didx] != dport) { 5061 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 5062 pd->ip_sum, &checksum, 5063 &nk->addr[pd->didx], 5064 nk->port[pd->didx], 1, af); 5065 } 5066 break; 5067 } 5068 #ifdef INET 5069 case IPPROTO_ICMP: 5070 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5071 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5072 nk->addr[pd->sidx].v4.s_addr, 0); 5073 5074 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5075 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5076 nk->addr[pd->didx].v4.s_addr, 0); 5077 5078 if (virtual_type == htons(ICMP_ECHO) && 5079 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5080 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5081 pd->hdr.icmp.icmp_cksum, sport, 5082 nk->port[pd->sidx], 0); 5083 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5084 pd->sport = &pd->hdr.icmp.icmp_id; 5085 } 5086 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5087 break; 5088 #endif /* INET */ 5089 #ifdef INET6 5090 case IPPROTO_ICMPV6: 5091 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5092 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5093 &nk->addr[pd->sidx], 0); 5094 5095 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5096 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5097 &nk->addr[pd->didx], 0); 5098 rewrite++; 5099 break; 5100 #endif /* INET */ 5101 default: 5102 switch (af) { 5103 #ifdef INET 5104 case AF_INET: 5105 if (PF_ANEQ(saddr, 5106 &nk->addr[pd->sidx], AF_INET)) 5107 pf_change_a(&saddr->v4.s_addr, 5108 pd->ip_sum, 5109 nk->addr[pd->sidx].v4.s_addr, 0); 5110 5111 if (PF_ANEQ(daddr, 5112 &nk->addr[pd->didx], AF_INET)) 5113 pf_change_a(&daddr->v4.s_addr, 5114 pd->ip_sum, 5115 nk->addr[pd->didx].v4.s_addr, 0); 5116 break; 5117 #endif /* INET */ 5118 #ifdef INET6 5119 case AF_INET6: 5120 if (PF_ANEQ(saddr, 5121 &nk->addr[pd->sidx], AF_INET6)) 5122 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 5123 5124 if (PF_ANEQ(daddr, 5125 &nk->addr[pd->didx], AF_INET6)) 5126 PF_ACPY(daddr, &nk->addr[pd->didx], af); 5127 break; 5128 #endif /* INET */ 5129 } 5130 break; 5131 } 5132 if (nr->natpass) 5133 r = NULL; 5134 } 5135 5136 while (r != NULL) { 5137 pf_counter_u64_add(&r->evaluations, 1); 5138 PF_TEST_ATTRIB(pfi_kkif_match(r->kif, kif) == r->ifnot, 5139 r->skip[PF_SKIP_IFP]); 5140 PF_TEST_ATTRIB(r->direction && r->direction != pd->dir, 5141 r->skip[PF_SKIP_DIR]); 5142 PF_TEST_ATTRIB(r->af && r->af != af, 5143 r->skip[PF_SKIP_AF]); 5144 PF_TEST_ATTRIB(r->proto && r->proto != pd->proto, 5145 r->skip[PF_SKIP_PROTO]); 5146 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, saddr, af, 5147 r->src.neg, kif, M_GETFIB(m)), 5148 r->skip[PF_SKIP_SRC_ADDR]); 5149 PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, daddr, af, 5150 r->dst.neg, NULL, M_GETFIB(m)), 5151 r->skip[PF_SKIP_DST_ADDR]); 5152 switch (pd->virtual_proto) { 5153 case PF_VPROTO_FRAGMENT: 5154 /* tcp/udp only. port_op always 0 in other cases */ 5155 PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op), 5156 TAILQ_NEXT(r, entries)); 5157 PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset), 5158 TAILQ_NEXT(r, entries)); 5159 /* icmp only. type/code always 0 in other cases */ 5160 PF_TEST_ATTRIB((r->type || r->code), 5161 TAILQ_NEXT(r, entries)); 5162 /* tcp/udp only. {uid|gid}.op always 0 in other cases */ 5163 PF_TEST_ATTRIB((r->gid.op || r->uid.op), 5164 TAILQ_NEXT(r, entries)); 5165 break; 5166 5167 case IPPROTO_TCP: 5168 PF_TEST_ATTRIB((r->flagset & th->th_flags) != r->flags, 5169 TAILQ_NEXT(r, entries)); 5170 /* FALLTHROUGH */ 5171 case IPPROTO_SCTP: 5172 case IPPROTO_UDP: 5173 /* tcp/udp only. port_op always 0 in other cases */ 5174 PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op, 5175 r->src.port[0], r->src.port[1], sport), 5176 r->skip[PF_SKIP_SRC_PORT]); 5177 /* tcp/udp only. port_op always 0 in other cases */ 5178 PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op, 5179 r->dst.port[0], r->dst.port[1], dport), 5180 r->skip[PF_SKIP_DST_PORT]); 5181 /* tcp/udp only. uid.op always 0 in other cases */ 5182 PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done = 5183 pf_socket_lookup(pd, m), 1)) && 5184 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5185 pd->lookup.uid), 5186 TAILQ_NEXT(r, entries)); 5187 /* tcp/udp only. gid.op always 0 in other cases */ 5188 PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done = 5189 pf_socket_lookup(pd, m), 1)) && 5190 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5191 pd->lookup.gid), 5192 TAILQ_NEXT(r, entries)); 5193 break; 5194 5195 case IPPROTO_ICMP: 5196 case IPPROTO_ICMPV6: 5197 /* icmp only. type always 0 in other cases */ 5198 PF_TEST_ATTRIB(r->type && r->type != icmptype + 1, 5199 TAILQ_NEXT(r, entries)); 5200 /* icmp only. type always 0 in other cases */ 5201 PF_TEST_ATTRIB(r->code && r->code != icmpcode + 1, 5202 TAILQ_NEXT(r, entries)); 5203 break; 5204 5205 default: 5206 break; 5207 } 5208 PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos), 5209 TAILQ_NEXT(r, entries)); 5210 PF_TEST_ATTRIB(r->prio && 5211 !pf_match_ieee8021q_pcp(r->prio, m), 5212 TAILQ_NEXT(r, entries)); 5213 PF_TEST_ATTRIB(r->prob && 5214 r->prob <= arc4random(), 5215 TAILQ_NEXT(r, entries)); 5216 PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(m, r, &tag, 5217 pd->pf_mtag ? pd->pf_mtag->tag : 0), 5218 TAILQ_NEXT(r, entries)); 5219 PF_TEST_ATTRIB(r->rcv_kif && !pf_match_rcvif(m, r), 5220 TAILQ_NEXT(r, entries)); 5221 PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT && 5222 pd->virtual_proto != PF_VPROTO_FRAGMENT), 5223 TAILQ_NEXT(r, entries)); 5224 PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY && 5225 (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match( 5226 pf_osfp_fingerprint(pd, m, off, th), 5227 r->os_fingerprint)), 5228 TAILQ_NEXT(r, entries)); 5229 /* FALLTHROUGH */ 5230 if (r->tag) 5231 tag = r->tag; 5232 if (r->anchor == NULL) { 5233 if (r->action == PF_MATCH) { 5234 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5235 if (ri == NULL) { 5236 REASON_SET(&reason, PFRES_MEMORY); 5237 goto cleanup; 5238 } 5239 ri->r = r; 5240 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5241 pf_counter_u64_critical_enter(); 5242 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5243 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5244 pf_counter_u64_critical_exit(); 5245 pf_rule_to_actions(r, &pd->act); 5246 if (r->log || pd->act.log & PF_LOG_MATCHES) 5247 PFLOG_PACKET(kif, m, 5248 r->action, PFRES_MATCH, r, 5249 a, ruleset, pd, 1); 5250 } else { 5251 match = 1; 5252 *rm = r; 5253 *am = a; 5254 *rsm = ruleset; 5255 if (pd->act.log & PF_LOG_MATCHES) 5256 PFLOG_PACKET(kif, m, 5257 r->action, PFRES_MATCH, r, 5258 a, ruleset, pd, 1); 5259 } 5260 if ((*rm)->quick) 5261 break; 5262 r = TAILQ_NEXT(r, entries); 5263 } else 5264 pf_step_into_anchor(anchor_stack, &asd, 5265 &ruleset, PF_RULESET_FILTER, &r, &a, 5266 &match); 5267 nextrule: 5268 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5269 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5270 break; 5271 } 5272 r = *rm; 5273 a = *am; 5274 ruleset = *rsm; 5275 5276 REASON_SET(&reason, PFRES_MATCH); 5277 5278 /* apply actions for last matching pass/block rule */ 5279 pf_rule_to_actions(r, &pd->act); 5280 5281 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5282 if (rewrite) 5283 m_copyback(m, off, hdrlen, pd->hdr.any); 5284 PFLOG_PACKET(kif, m, r->action, reason, r, a, ruleset, pd, 1); 5285 } 5286 5287 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5288 (r->action == PF_DROP) && 5289 ((r->rule_flag & PFRULE_RETURNRST) || 5290 (r->rule_flag & PFRULE_RETURNICMP) || 5291 (r->rule_flag & PFRULE_RETURN))) { 5292 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 5293 bip_sum, hdrlen, &reason, r->rtableid); 5294 } 5295 5296 if (r->action == PF_DROP) 5297 goto cleanup; 5298 5299 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5300 REASON_SET(&reason, PFRES_MEMORY); 5301 goto cleanup; 5302 } 5303 if (pd->act.rtableid >= 0) 5304 M_SETFIB(m, pd->act.rtableid); 5305 5306 if (pd->virtual_proto != PF_VPROTO_FRAGMENT && 5307 (!state_icmp && (r->keep_state || nr != NULL || 5308 (pd->flags & PFDESC_TCP_NORM)))) { 5309 int action; 5310 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 5311 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 5312 hdrlen, &match_rules, udp_mapping); 5313 if (action != PF_PASS) { 5314 pf_udp_mapping_release(udp_mapping); 5315 if (action == PF_DROP && 5316 (r->rule_flag & PFRULE_RETURN)) 5317 pf_return(r, nr, pd, sk, off, m, th, kif, 5318 bproto_sum, bip_sum, hdrlen, &reason, 5319 pd->act.rtableid); 5320 return (action); 5321 } 5322 } else { 5323 while ((ri = SLIST_FIRST(&match_rules))) { 5324 SLIST_REMOVE_HEAD(&match_rules, entry); 5325 free(ri, M_PF_RULE_ITEM); 5326 } 5327 5328 uma_zfree(V_pf_state_key_z, sk); 5329 uma_zfree(V_pf_state_key_z, nk); 5330 pf_udp_mapping_release(udp_mapping); 5331 } 5332 5333 /* copy back packet headers if we performed NAT operations */ 5334 if (rewrite) 5335 m_copyback(m, off, hdrlen, pd->hdr.any); 5336 5337 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5338 pd->dir == PF_OUT && 5339 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 5340 /* 5341 * We want the state created, but we dont 5342 * want to send this in case a partner 5343 * firewall has to know about it to allow 5344 * replies through it. 5345 */ 5346 return (PF_DEFER); 5347 5348 return (PF_PASS); 5349 5350 cleanup: 5351 while ((ri = SLIST_FIRST(&match_rules))) { 5352 SLIST_REMOVE_HEAD(&match_rules, entry); 5353 free(ri, M_PF_RULE_ITEM); 5354 } 5355 5356 uma_zfree(V_pf_state_key_z, sk); 5357 uma_zfree(V_pf_state_key_z, nk); 5358 pf_udp_mapping_release(udp_mapping); 5359 5360 return (PF_DROP); 5361 } 5362 5363 static int 5364 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5365 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5366 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 5367 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 5368 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 5369 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5370 { 5371 struct pf_kstate *s = NULL; 5372 struct pf_ksrc_node *sn = NULL; 5373 struct tcphdr *th = &pd->hdr.tcp; 5374 u_int16_t mss = V_tcp_mssdflt; 5375 u_short reason, sn_reason; 5376 struct pf_krule_item *ri; 5377 5378 /* check maximums */ 5379 if (r->max_states && 5380 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5381 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5382 REASON_SET(&reason, PFRES_MAXSTATES); 5383 goto csfailed; 5384 } 5385 /* src node for filter rule */ 5386 if ((r->rule_flag & PFRULE_SRCTRACK || 5387 r->rpool.opts & PF_POOL_STICKYADDR) && 5388 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5389 REASON_SET(&reason, sn_reason); 5390 goto csfailed; 5391 } 5392 /* src node for translation rule */ 5393 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5394 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5395 pd->af)) != 0 ) { 5396 REASON_SET(&reason, sn_reason); 5397 goto csfailed; 5398 } 5399 s = pf_alloc_state(M_NOWAIT); 5400 if (s == NULL) { 5401 REASON_SET(&reason, PFRES_MEMORY); 5402 goto csfailed; 5403 } 5404 s->rule = r; 5405 s->nat_rule = nr; 5406 s->anchor = a; 5407 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5408 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5409 5410 STATE_INC_COUNTERS(s); 5411 if (r->allow_opts) 5412 s->state_flags |= PFSTATE_ALLOWOPTS; 5413 if (r->rule_flag & PFRULE_STATESLOPPY) 5414 s->state_flags |= PFSTATE_SLOPPY; 5415 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5416 s->state_flags |= PFSTATE_SCRUB_TCP; 5417 if ((r->rule_flag & PFRULE_PFLOW) || 5418 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5419 s->state_flags |= PFSTATE_PFLOW; 5420 5421 s->act.log = pd->act.log & PF_LOG_ALL; 5422 s->sync_state = PFSYNC_S_NONE; 5423 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5424 5425 if (nr != NULL) 5426 s->act.log |= nr->log & PF_LOG_ALL; 5427 switch (pd->proto) { 5428 case IPPROTO_TCP: 5429 s->src.seqlo = ntohl(th->th_seq); 5430 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5431 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5432 r->keep_state == PF_STATE_MODULATE) { 5433 /* Generate sequence number modulator */ 5434 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5435 0) 5436 s->src.seqdiff = 1; 5437 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 5438 htonl(s->src.seqlo + s->src.seqdiff), 0); 5439 *rewrite = 1; 5440 } else 5441 s->src.seqdiff = 0; 5442 if (th->th_flags & TH_SYN) { 5443 s->src.seqhi++; 5444 s->src.wscale = pf_get_wscale(m, off, 5445 th->th_off, pd->af); 5446 } 5447 s->src.max_win = MAX(ntohs(th->th_win), 1); 5448 if (s->src.wscale & PF_WSCALE_MASK) { 5449 /* Remove scale factor from initial window */ 5450 int win = s->src.max_win; 5451 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5452 s->src.max_win = (win - 1) >> 5453 (s->src.wscale & PF_WSCALE_MASK); 5454 } 5455 if (th->th_flags & TH_FIN) 5456 s->src.seqhi++; 5457 s->dst.seqhi = 1; 5458 s->dst.max_win = 1; 5459 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5460 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5461 s->timeout = PFTM_TCP_FIRST_PACKET; 5462 atomic_add_32(&V_pf_status.states_halfopen, 1); 5463 break; 5464 case IPPROTO_UDP: 5465 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5466 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5467 s->timeout = PFTM_UDP_FIRST_PACKET; 5468 break; 5469 case IPPROTO_SCTP: 5470 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5471 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5472 s->timeout = PFTM_SCTP_FIRST_PACKET; 5473 break; 5474 case IPPROTO_ICMP: 5475 #ifdef INET6 5476 case IPPROTO_ICMPV6: 5477 #endif 5478 s->timeout = PFTM_ICMP_FIRST_PACKET; 5479 break; 5480 default: 5481 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5482 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5483 s->timeout = PFTM_OTHER_FIRST_PACKET; 5484 } 5485 5486 if (r->rt) { 5487 /* pf_map_addr increases the reason counters */ 5488 if ((reason = pf_map_addr_sn(pd->af, r, pd->src, &s->rt_addr, 5489 &s->rt_kif, NULL, &sn)) != 0) 5490 goto csfailed; 5491 s->rt = r->rt; 5492 } 5493 5494 s->creation = s->expire = pf_get_uptime(); 5495 5496 if (sn != NULL) 5497 s->src_node = sn; 5498 if (nsn != NULL) { 5499 /* XXX We only modify one side for now. */ 5500 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5501 s->nat_src_node = nsn; 5502 } 5503 if (pd->proto == IPPROTO_TCP) { 5504 if (s->state_flags & PFSTATE_SCRUB_TCP && 5505 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 5506 REASON_SET(&reason, PFRES_MEMORY); 5507 goto csfailed; 5508 } 5509 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5510 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 5511 &s->src, &s->dst, rewrite)) { 5512 /* This really shouldn't happen!!! */ 5513 DPFPRINTF(PF_DEBUG_URGENT, 5514 ("pf_normalize_tcp_stateful failed on first " 5515 "pkt\n")); 5516 goto csfailed; 5517 } 5518 } else if (pd->proto == IPPROTO_SCTP) { 5519 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5520 goto csfailed; 5521 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5522 goto csfailed; 5523 } 5524 s->direction = pd->dir; 5525 5526 /* 5527 * sk/nk could already been setup by pf_get_translation(). 5528 */ 5529 if (nr == NULL) { 5530 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5531 __func__, nr, sk, nk)); 5532 sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport); 5533 if (sk == NULL) 5534 goto csfailed; 5535 nk = sk; 5536 } else 5537 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5538 __func__, nr, sk, nk)); 5539 5540 /* Swap sk/nk for PF_OUT. */ 5541 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5542 (pd->dir == PF_IN) ? sk : nk, 5543 (pd->dir == PF_IN) ? nk : sk, s)) { 5544 REASON_SET(&reason, PFRES_STATEINS); 5545 goto drop; 5546 } else 5547 *sm = s; 5548 5549 if (tag > 0) 5550 s->tag = tag; 5551 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5552 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5553 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5554 /* undo NAT changes, if they have taken place */ 5555 if (nr != NULL) { 5556 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5557 if (pd->dir == PF_OUT) 5558 skt = s->key[PF_SK_STACK]; 5559 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5560 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5561 if (pd->sport) 5562 *pd->sport = skt->port[pd->sidx]; 5563 if (pd->dport) 5564 *pd->dport = skt->port[pd->didx]; 5565 if (pd->proto_sum) 5566 *pd->proto_sum = bproto_sum; 5567 if (pd->ip_sum) 5568 *pd->ip_sum = bip_sum; 5569 m_copyback(m, off, hdrlen, pd->hdr.any); 5570 } 5571 s->src.seqhi = htonl(arc4random()); 5572 /* Find mss option */ 5573 int rtid = M_GETFIB(m); 5574 mss = pf_get_mss(m, off, th->th_off, pd->af); 5575 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5576 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5577 s->src.mss = mss; 5578 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5579 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5580 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5581 pd->act.rtableid); 5582 REASON_SET(&reason, PFRES_SYNPROXY); 5583 return (PF_SYNPROXY_DROP); 5584 } 5585 5586 s->udp_mapping = udp_mapping; 5587 5588 return (PF_PASS); 5589 5590 csfailed: 5591 while ((ri = SLIST_FIRST(match_rules))) { 5592 SLIST_REMOVE_HEAD(match_rules, entry); 5593 free(ri, M_PF_RULE_ITEM); 5594 } 5595 5596 uma_zfree(V_pf_state_key_z, sk); 5597 uma_zfree(V_pf_state_key_z, nk); 5598 5599 if (sn != NULL) { 5600 PF_SRC_NODE_LOCK(sn); 5601 if (--sn->states == 0 && sn->expire == 0) { 5602 pf_unlink_src_node(sn); 5603 uma_zfree(V_pf_sources_z, sn); 5604 counter_u64_add( 5605 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5606 } 5607 PF_SRC_NODE_UNLOCK(sn); 5608 } 5609 5610 if (nsn != sn && nsn != NULL) { 5611 PF_SRC_NODE_LOCK(nsn); 5612 if (--nsn->states == 0 && nsn->expire == 0) { 5613 pf_unlink_src_node(nsn); 5614 uma_zfree(V_pf_sources_z, nsn); 5615 counter_u64_add( 5616 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5617 } 5618 PF_SRC_NODE_UNLOCK(nsn); 5619 } 5620 5621 drop: 5622 if (s != NULL) { 5623 pf_src_tree_remove_state(s); 5624 s->timeout = PFTM_UNLINKED; 5625 STATE_DEC_COUNTERS(s); 5626 pf_free_state(s); 5627 } 5628 5629 return (PF_DROP); 5630 } 5631 5632 static int 5633 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5634 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5635 int *copyback) 5636 { 5637 struct tcphdr *th = &pd->hdr.tcp; 5638 struct pf_state_peer *src, *dst; 5639 u_int16_t win = ntohs(th->th_win); 5640 u_int32_t ack, end, data_end, seq, orig_seq; 5641 u_int8_t sws, dws, psrc, pdst; 5642 int ackskew; 5643 5644 if (pd->dir == (*state)->direction) { 5645 src = &(*state)->src; 5646 dst = &(*state)->dst; 5647 psrc = PF_PEER_SRC; 5648 pdst = PF_PEER_DST; 5649 } else { 5650 src = &(*state)->dst; 5651 dst = &(*state)->src; 5652 psrc = PF_PEER_DST; 5653 pdst = PF_PEER_SRC; 5654 } 5655 5656 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5657 sws = src->wscale & PF_WSCALE_MASK; 5658 dws = dst->wscale & PF_WSCALE_MASK; 5659 } else 5660 sws = dws = 0; 5661 5662 /* 5663 * Sequence tracking algorithm from Guido van Rooij's paper: 5664 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5665 * tcp_filtering.ps 5666 */ 5667 5668 orig_seq = seq = ntohl(th->th_seq); 5669 if (src->seqlo == 0) { 5670 /* First packet from this end. Set its state */ 5671 5672 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5673 src->scrub == NULL) { 5674 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5675 REASON_SET(reason, PFRES_MEMORY); 5676 return (PF_DROP); 5677 } 5678 } 5679 5680 /* Deferred generation of sequence number modulator */ 5681 if (dst->seqdiff && !src->seqdiff) { 5682 /* use random iss for the TCP server */ 5683 while ((src->seqdiff = arc4random() - seq) == 0) 5684 ; 5685 ack = ntohl(th->th_ack) - dst->seqdiff; 5686 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5687 src->seqdiff), 0); 5688 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5689 *copyback = 1; 5690 } else { 5691 ack = ntohl(th->th_ack); 5692 } 5693 5694 end = seq + pd->p_len; 5695 if (th->th_flags & TH_SYN) { 5696 end++; 5697 if (dst->wscale & PF_WSCALE_FLAG) { 5698 src->wscale = pf_get_wscale(m, off, th->th_off, 5699 pd->af); 5700 if (src->wscale & PF_WSCALE_FLAG) { 5701 /* Remove scale factor from initial 5702 * window */ 5703 sws = src->wscale & PF_WSCALE_MASK; 5704 win = ((u_int32_t)win + (1 << sws) - 1) 5705 >> sws; 5706 dws = dst->wscale & PF_WSCALE_MASK; 5707 } else { 5708 /* fixup other window */ 5709 dst->max_win = MIN(TCP_MAXWIN, 5710 (u_int32_t)dst->max_win << 5711 (dst->wscale & PF_WSCALE_MASK)); 5712 /* in case of a retrans SYN|ACK */ 5713 dst->wscale = 0; 5714 } 5715 } 5716 } 5717 data_end = end; 5718 if (th->th_flags & TH_FIN) 5719 end++; 5720 5721 src->seqlo = seq; 5722 if (src->state < TCPS_SYN_SENT) 5723 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5724 5725 /* 5726 * May need to slide the window (seqhi may have been set by 5727 * the crappy stack check or if we picked up the connection 5728 * after establishment) 5729 */ 5730 if (src->seqhi == 1 || 5731 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5732 src->seqhi = end + MAX(1, dst->max_win << dws); 5733 if (win > src->max_win) 5734 src->max_win = win; 5735 5736 } else { 5737 ack = ntohl(th->th_ack) - dst->seqdiff; 5738 if (src->seqdiff) { 5739 /* Modulate sequence numbers */ 5740 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5741 src->seqdiff), 0); 5742 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5743 *copyback = 1; 5744 } 5745 end = seq + pd->p_len; 5746 if (th->th_flags & TH_SYN) 5747 end++; 5748 data_end = end; 5749 if (th->th_flags & TH_FIN) 5750 end++; 5751 } 5752 5753 if ((th->th_flags & TH_ACK) == 0) { 5754 /* Let it pass through the ack skew check */ 5755 ack = dst->seqlo; 5756 } else if ((ack == 0 && 5757 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5758 /* broken tcp stacks do not set ack */ 5759 (dst->state < TCPS_SYN_SENT)) { 5760 /* 5761 * Many stacks (ours included) will set the ACK number in an 5762 * FIN|ACK if the SYN times out -- no sequence to ACK. 5763 */ 5764 ack = dst->seqlo; 5765 } 5766 5767 if (seq == end) { 5768 /* Ease sequencing restrictions on no data packets */ 5769 seq = src->seqlo; 5770 data_end = end = seq; 5771 } 5772 5773 ackskew = dst->seqlo - ack; 5774 5775 /* 5776 * Need to demodulate the sequence numbers in any TCP SACK options 5777 * (Selective ACK). We could optionally validate the SACK values 5778 * against the current ACK window, either forwards or backwards, but 5779 * I'm not confident that SACK has been implemented properly 5780 * everywhere. It wouldn't surprise me if several stacks accidentally 5781 * SACK too far backwards of previously ACKed data. There really aren't 5782 * any security implications of bad SACKing unless the target stack 5783 * doesn't validate the option length correctly. Someone trying to 5784 * spoof into a TCP connection won't bother blindly sending SACK 5785 * options anyway. 5786 */ 5787 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5788 if (pf_modulate_sack(m, off, pd, th, dst)) 5789 *copyback = 1; 5790 } 5791 5792 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5793 if (SEQ_GEQ(src->seqhi, data_end) && 5794 /* Last octet inside other's window space */ 5795 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5796 /* Retrans: not more than one window back */ 5797 (ackskew >= -MAXACKWINDOW) && 5798 /* Acking not more than one reassembled fragment backwards */ 5799 (ackskew <= (MAXACKWINDOW << sws)) && 5800 /* Acking not more than one window forward */ 5801 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5802 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5803 /* Require an exact/+1 sequence match on resets when possible */ 5804 5805 if (dst->scrub || src->scrub) { 5806 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5807 *state, src, dst, copyback)) 5808 return (PF_DROP); 5809 } 5810 5811 /* update max window */ 5812 if (src->max_win < win) 5813 src->max_win = win; 5814 /* synchronize sequencing */ 5815 if (SEQ_GT(end, src->seqlo)) 5816 src->seqlo = end; 5817 /* slide the window of what the other end can send */ 5818 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5819 dst->seqhi = ack + MAX((win << sws), 1); 5820 5821 /* update states */ 5822 if (th->th_flags & TH_SYN) 5823 if (src->state < TCPS_SYN_SENT) 5824 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5825 if (th->th_flags & TH_FIN) 5826 if (src->state < TCPS_CLOSING) 5827 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5828 if (th->th_flags & TH_ACK) { 5829 if (dst->state == TCPS_SYN_SENT) { 5830 pf_set_protostate(*state, pdst, 5831 TCPS_ESTABLISHED); 5832 if (src->state == TCPS_ESTABLISHED && 5833 (*state)->src_node != NULL && 5834 pf_src_connlimit(state)) { 5835 REASON_SET(reason, PFRES_SRCLIMIT); 5836 return (PF_DROP); 5837 } 5838 } else if (dst->state == TCPS_CLOSING) 5839 pf_set_protostate(*state, pdst, 5840 TCPS_FIN_WAIT_2); 5841 } 5842 if (th->th_flags & TH_RST) 5843 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5844 5845 /* update expire time */ 5846 (*state)->expire = pf_get_uptime(); 5847 if (src->state >= TCPS_FIN_WAIT_2 && 5848 dst->state >= TCPS_FIN_WAIT_2) 5849 (*state)->timeout = PFTM_TCP_CLOSED; 5850 else if (src->state >= TCPS_CLOSING && 5851 dst->state >= TCPS_CLOSING) 5852 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5853 else if (src->state < TCPS_ESTABLISHED || 5854 dst->state < TCPS_ESTABLISHED) 5855 (*state)->timeout = PFTM_TCP_OPENING; 5856 else if (src->state >= TCPS_CLOSING || 5857 dst->state >= TCPS_CLOSING) 5858 (*state)->timeout = PFTM_TCP_CLOSING; 5859 else 5860 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5861 5862 /* Fall through to PASS packet */ 5863 5864 } else if ((dst->state < TCPS_SYN_SENT || 5865 dst->state >= TCPS_FIN_WAIT_2 || 5866 src->state >= TCPS_FIN_WAIT_2) && 5867 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5868 /* Within a window forward of the originating packet */ 5869 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5870 /* Within a window backward of the originating packet */ 5871 5872 /* 5873 * This currently handles three situations: 5874 * 1) Stupid stacks will shotgun SYNs before their peer 5875 * replies. 5876 * 2) When PF catches an already established stream (the 5877 * firewall rebooted, the state table was flushed, routes 5878 * changed...) 5879 * 3) Packets get funky immediately after the connection 5880 * closes (this should catch Solaris spurious ACK|FINs 5881 * that web servers like to spew after a close) 5882 * 5883 * This must be a little more careful than the above code 5884 * since packet floods will also be caught here. We don't 5885 * update the TTL here to mitigate the damage of a packet 5886 * flood and so the same code can handle awkward establishment 5887 * and a loosened connection close. 5888 * In the establishment case, a correct peer response will 5889 * validate the connection, go through the normal state code 5890 * and keep updating the state TTL. 5891 */ 5892 5893 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5894 printf("pf: loose state match: "); 5895 pf_print_state(*state); 5896 pf_print_flags(th->th_flags); 5897 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5898 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5899 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5900 (unsigned long long)(*state)->packets[1], 5901 pd->dir == PF_IN ? "in" : "out", 5902 pd->dir == (*state)->direction ? "fwd" : "rev"); 5903 } 5904 5905 if (dst->scrub || src->scrub) { 5906 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5907 *state, src, dst, copyback)) 5908 return (PF_DROP); 5909 } 5910 5911 /* update max window */ 5912 if (src->max_win < win) 5913 src->max_win = win; 5914 /* synchronize sequencing */ 5915 if (SEQ_GT(end, src->seqlo)) 5916 src->seqlo = end; 5917 /* slide the window of what the other end can send */ 5918 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5919 dst->seqhi = ack + MAX((win << sws), 1); 5920 5921 /* 5922 * Cannot set dst->seqhi here since this could be a shotgunned 5923 * SYN and not an already established connection. 5924 */ 5925 5926 if (th->th_flags & TH_FIN) 5927 if (src->state < TCPS_CLOSING) 5928 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5929 if (th->th_flags & TH_RST) 5930 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5931 5932 /* Fall through to PASS packet */ 5933 5934 } else { 5935 if ((*state)->dst.state == TCPS_SYN_SENT && 5936 (*state)->src.state == TCPS_SYN_SENT) { 5937 /* Send RST for state mismatches during handshake */ 5938 if (!(th->th_flags & TH_RST)) 5939 pf_send_tcp((*state)->rule, pd->af, 5940 pd->dst, pd->src, th->th_dport, 5941 th->th_sport, ntohl(th->th_ack), 0, 5942 TH_RST, 0, 0, 5943 (*state)->rule->return_ttl, true, 0, 0, 5944 (*state)->act.rtableid); 5945 src->seqlo = 0; 5946 src->seqhi = 1; 5947 src->max_win = 1; 5948 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5949 printf("pf: BAD state: "); 5950 pf_print_state(*state); 5951 pf_print_flags(th->th_flags); 5952 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5953 "pkts=%llu:%llu dir=%s,%s\n", 5954 seq, orig_seq, ack, pd->p_len, ackskew, 5955 (unsigned long long)(*state)->packets[0], 5956 (unsigned long long)(*state)->packets[1], 5957 pd->dir == PF_IN ? "in" : "out", 5958 pd->dir == (*state)->direction ? "fwd" : "rev"); 5959 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5960 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 5961 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5962 ' ': '2', 5963 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5964 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5965 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 5966 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5967 } 5968 REASON_SET(reason, PFRES_BADSTATE); 5969 return (PF_DROP); 5970 } 5971 5972 return (PF_PASS); 5973 } 5974 5975 static int 5976 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5977 { 5978 struct tcphdr *th = &pd->hdr.tcp; 5979 struct pf_state_peer *src, *dst; 5980 u_int8_t psrc, pdst; 5981 5982 if (pd->dir == (*state)->direction) { 5983 src = &(*state)->src; 5984 dst = &(*state)->dst; 5985 psrc = PF_PEER_SRC; 5986 pdst = PF_PEER_DST; 5987 } else { 5988 src = &(*state)->dst; 5989 dst = &(*state)->src; 5990 psrc = PF_PEER_DST; 5991 pdst = PF_PEER_SRC; 5992 } 5993 5994 if (th->th_flags & TH_SYN) 5995 if (src->state < TCPS_SYN_SENT) 5996 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5997 if (th->th_flags & TH_FIN) 5998 if (src->state < TCPS_CLOSING) 5999 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6000 if (th->th_flags & TH_ACK) { 6001 if (dst->state == TCPS_SYN_SENT) { 6002 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6003 if (src->state == TCPS_ESTABLISHED && 6004 (*state)->src_node != NULL && 6005 pf_src_connlimit(state)) { 6006 REASON_SET(reason, PFRES_SRCLIMIT); 6007 return (PF_DROP); 6008 } 6009 } else if (dst->state == TCPS_CLOSING) { 6010 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6011 } else if (src->state == TCPS_SYN_SENT && 6012 dst->state < TCPS_SYN_SENT) { 6013 /* 6014 * Handle a special sloppy case where we only see one 6015 * half of the connection. If there is a ACK after 6016 * the initial SYN without ever seeing a packet from 6017 * the destination, set the connection to established. 6018 */ 6019 pf_set_protostate(*state, PF_PEER_BOTH, 6020 TCPS_ESTABLISHED); 6021 dst->state = src->state = TCPS_ESTABLISHED; 6022 if ((*state)->src_node != NULL && 6023 pf_src_connlimit(state)) { 6024 REASON_SET(reason, PFRES_SRCLIMIT); 6025 return (PF_DROP); 6026 } 6027 } else if (src->state == TCPS_CLOSING && 6028 dst->state == TCPS_ESTABLISHED && 6029 dst->seqlo == 0) { 6030 /* 6031 * Handle the closing of half connections where we 6032 * don't see the full bidirectional FIN/ACK+ACK 6033 * handshake. 6034 */ 6035 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6036 } 6037 } 6038 if (th->th_flags & TH_RST) 6039 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6040 6041 /* update expire time */ 6042 (*state)->expire = pf_get_uptime(); 6043 if (src->state >= TCPS_FIN_WAIT_2 && 6044 dst->state >= TCPS_FIN_WAIT_2) 6045 (*state)->timeout = PFTM_TCP_CLOSED; 6046 else if (src->state >= TCPS_CLOSING && 6047 dst->state >= TCPS_CLOSING) 6048 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6049 else if (src->state < TCPS_ESTABLISHED || 6050 dst->state < TCPS_ESTABLISHED) 6051 (*state)->timeout = PFTM_TCP_OPENING; 6052 else if (src->state >= TCPS_CLOSING || 6053 dst->state >= TCPS_CLOSING) 6054 (*state)->timeout = PFTM_TCP_CLOSING; 6055 else 6056 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6057 6058 return (PF_PASS); 6059 } 6060 6061 static int 6062 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6063 { 6064 struct pf_state_key *sk = (*state)->key[pd->didx]; 6065 struct tcphdr *th = &pd->hdr.tcp; 6066 6067 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6068 if (pd->dir != (*state)->direction) { 6069 REASON_SET(reason, PFRES_SYNPROXY); 6070 return (PF_SYNPROXY_DROP); 6071 } 6072 if (th->th_flags & TH_SYN) { 6073 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6074 REASON_SET(reason, PFRES_SYNPROXY); 6075 return (PF_DROP); 6076 } 6077 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6078 pd->src, th->th_dport, th->th_sport, 6079 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6080 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 6081 (*state)->act.rtableid); 6082 REASON_SET(reason, PFRES_SYNPROXY); 6083 return (PF_SYNPROXY_DROP); 6084 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6085 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6086 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6087 REASON_SET(reason, PFRES_SYNPROXY); 6088 return (PF_DROP); 6089 } else if ((*state)->src_node != NULL && 6090 pf_src_connlimit(state)) { 6091 REASON_SET(reason, PFRES_SRCLIMIT); 6092 return (PF_DROP); 6093 } else 6094 pf_set_protostate(*state, PF_PEER_SRC, 6095 PF_TCPS_PROXY_DST); 6096 } 6097 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6098 if (pd->dir == (*state)->direction) { 6099 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6100 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6101 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6102 REASON_SET(reason, PFRES_SYNPROXY); 6103 return (PF_DROP); 6104 } 6105 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6106 if ((*state)->dst.seqhi == 1) 6107 (*state)->dst.seqhi = htonl(arc4random()); 6108 pf_send_tcp((*state)->rule, pd->af, 6109 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6110 sk->port[pd->sidx], sk->port[pd->didx], 6111 (*state)->dst.seqhi, 0, TH_SYN, 0, 6112 (*state)->src.mss, 0, false, (*state)->tag, 0, 6113 (*state)->act.rtableid); 6114 REASON_SET(reason, PFRES_SYNPROXY); 6115 return (PF_SYNPROXY_DROP); 6116 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6117 (TH_SYN|TH_ACK)) || 6118 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6119 REASON_SET(reason, PFRES_SYNPROXY); 6120 return (PF_DROP); 6121 } else { 6122 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6123 (*state)->dst.seqlo = ntohl(th->th_seq); 6124 pf_send_tcp((*state)->rule, pd->af, pd->dst, 6125 pd->src, th->th_dport, th->th_sport, 6126 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6127 TH_ACK, (*state)->src.max_win, 0, 0, false, 6128 (*state)->tag, 0, (*state)->act.rtableid); 6129 pf_send_tcp((*state)->rule, pd->af, 6130 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6131 sk->port[pd->sidx], sk->port[pd->didx], 6132 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6133 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 6134 (*state)->act.rtableid); 6135 (*state)->src.seqdiff = (*state)->dst.seqhi - 6136 (*state)->src.seqlo; 6137 (*state)->dst.seqdiff = (*state)->src.seqhi - 6138 (*state)->dst.seqlo; 6139 (*state)->src.seqhi = (*state)->src.seqlo + 6140 (*state)->dst.max_win; 6141 (*state)->dst.seqhi = (*state)->dst.seqlo + 6142 (*state)->src.max_win; 6143 (*state)->src.wscale = (*state)->dst.wscale = 0; 6144 pf_set_protostate(*state, PF_PEER_BOTH, 6145 TCPS_ESTABLISHED); 6146 REASON_SET(reason, PFRES_SYNPROXY); 6147 return (PF_SYNPROXY_DROP); 6148 } 6149 } 6150 6151 return (PF_PASS); 6152 } 6153 6154 static int 6155 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 6156 struct mbuf *m, int off, struct pf_pdesc *pd, 6157 u_short *reason) 6158 { 6159 struct pf_state_key_cmp key; 6160 struct tcphdr *th = &pd->hdr.tcp; 6161 int copyback = 0; 6162 int action; 6163 struct pf_state_peer *src, *dst; 6164 6165 bzero(&key, sizeof(key)); 6166 key.af = pd->af; 6167 key.proto = IPPROTO_TCP; 6168 if (pd->dir == PF_IN) { /* wire side, straight */ 6169 PF_ACPY(&key.addr[0], pd->src, key.af); 6170 PF_ACPY(&key.addr[1], pd->dst, key.af); 6171 key.port[0] = th->th_sport; 6172 key.port[1] = th->th_dport; 6173 } else { /* stack side, reverse */ 6174 PF_ACPY(&key.addr[1], pd->src, key.af); 6175 PF_ACPY(&key.addr[0], pd->dst, key.af); 6176 key.port[1] = th->th_sport; 6177 key.port[0] = th->th_dport; 6178 } 6179 6180 STATE_LOOKUP(kif, &key, *state, pd); 6181 6182 if (pd->dir == (*state)->direction) { 6183 src = &(*state)->src; 6184 dst = &(*state)->dst; 6185 } else { 6186 src = &(*state)->dst; 6187 dst = &(*state)->src; 6188 } 6189 6190 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6191 return (action); 6192 6193 if (dst->state >= TCPS_FIN_WAIT_2 && 6194 src->state >= TCPS_FIN_WAIT_2 && 6195 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6196 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6197 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6198 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6199 printf("pf: state reuse "); 6200 pf_print_state(*state); 6201 pf_print_flags(th->th_flags); 6202 printf("\n"); 6203 } 6204 /* XXX make sure it's the same direction ?? */ 6205 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6206 pf_unlink_state(*state); 6207 *state = NULL; 6208 return (PF_DROP); 6209 } 6210 6211 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6212 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6213 return (PF_DROP); 6214 } else { 6215 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 6216 ©back) == PF_DROP) 6217 return (PF_DROP); 6218 } 6219 6220 /* translate source/destination address, if necessary */ 6221 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6222 struct pf_state_key *nk = (*state)->key[pd->didx]; 6223 6224 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6225 nk->port[pd->sidx] != th->th_sport) 6226 pf_change_ap(m, pd->src, &th->th_sport, 6227 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6228 nk->port[pd->sidx], 0, pd->af); 6229 6230 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6231 nk->port[pd->didx] != th->th_dport) 6232 pf_change_ap(m, pd->dst, &th->th_dport, 6233 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6234 nk->port[pd->didx], 0, pd->af); 6235 copyback = 1; 6236 } 6237 6238 /* Copyback sequence modulation or stateful scrub changes if needed */ 6239 if (copyback) 6240 m_copyback(m, off, sizeof(*th), (caddr_t)th); 6241 6242 return (PF_PASS); 6243 } 6244 6245 static int 6246 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 6247 struct mbuf *m, int off, struct pf_pdesc *pd) 6248 { 6249 struct pf_state_peer *src, *dst; 6250 struct pf_state_key_cmp key; 6251 struct udphdr *uh = &pd->hdr.udp; 6252 uint8_t psrc, pdst; 6253 6254 bzero(&key, sizeof(key)); 6255 key.af = pd->af; 6256 key.proto = IPPROTO_UDP; 6257 if (pd->dir == PF_IN) { /* wire side, straight */ 6258 PF_ACPY(&key.addr[0], pd->src, key.af); 6259 PF_ACPY(&key.addr[1], pd->dst, key.af); 6260 key.port[0] = uh->uh_sport; 6261 key.port[1] = uh->uh_dport; 6262 } else { /* stack side, reverse */ 6263 PF_ACPY(&key.addr[1], pd->src, key.af); 6264 PF_ACPY(&key.addr[0], pd->dst, key.af); 6265 key.port[1] = uh->uh_sport; 6266 key.port[0] = uh->uh_dport; 6267 } 6268 6269 STATE_LOOKUP(kif, &key, *state, pd); 6270 6271 if (pd->dir == (*state)->direction) { 6272 src = &(*state)->src; 6273 dst = &(*state)->dst; 6274 psrc = PF_PEER_SRC; 6275 pdst = PF_PEER_DST; 6276 } else { 6277 src = &(*state)->dst; 6278 dst = &(*state)->src; 6279 psrc = PF_PEER_DST; 6280 pdst = PF_PEER_SRC; 6281 } 6282 6283 /* update states */ 6284 if (src->state < PFUDPS_SINGLE) 6285 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6286 if (dst->state == PFUDPS_SINGLE) 6287 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6288 6289 /* update expire time */ 6290 (*state)->expire = pf_get_uptime(); 6291 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6292 (*state)->timeout = PFTM_UDP_MULTIPLE; 6293 else 6294 (*state)->timeout = PFTM_UDP_SINGLE; 6295 6296 /* translate source/destination address, if necessary */ 6297 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6298 struct pf_state_key *nk = (*state)->key[pd->didx]; 6299 6300 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6301 nk->port[pd->sidx] != uh->uh_sport) 6302 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 6303 &uh->uh_sum, &nk->addr[pd->sidx], 6304 nk->port[pd->sidx], 1, pd->af); 6305 6306 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6307 nk->port[pd->didx] != uh->uh_dport) 6308 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 6309 &uh->uh_sum, &nk->addr[pd->didx], 6310 nk->port[pd->didx], 1, pd->af); 6311 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 6312 } 6313 6314 return (PF_PASS); 6315 } 6316 6317 static int 6318 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 6319 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason) 6320 { 6321 struct pf_state_key_cmp key; 6322 struct pf_state_peer *src, *dst; 6323 struct sctphdr *sh = &pd->hdr.sctp; 6324 u_int8_t psrc; //, pdst; 6325 6326 bzero(&key, sizeof(key)); 6327 key.af = pd->af; 6328 key.proto = IPPROTO_SCTP; 6329 if (pd->dir == PF_IN) { /* wire side, straight */ 6330 PF_ACPY(&key.addr[0], pd->src, key.af); 6331 PF_ACPY(&key.addr[1], pd->dst, key.af); 6332 key.port[0] = sh->src_port; 6333 key.port[1] = sh->dest_port; 6334 } else { /* stack side, reverse */ 6335 PF_ACPY(&key.addr[1], pd->src, key.af); 6336 PF_ACPY(&key.addr[0], pd->dst, key.af); 6337 key.port[1] = sh->src_port; 6338 key.port[0] = sh->dest_port; 6339 } 6340 6341 STATE_LOOKUP(kif, &key, *state, pd); 6342 6343 if (pd->dir == (*state)->direction) { 6344 src = &(*state)->src; 6345 dst = &(*state)->dst; 6346 psrc = PF_PEER_SRC; 6347 } else { 6348 src = &(*state)->dst; 6349 dst = &(*state)->src; 6350 psrc = PF_PEER_DST; 6351 } 6352 6353 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6354 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6355 pd->sctp_flags & PFDESC_SCTP_INIT) { 6356 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6357 pf_unlink_state(*state); 6358 *state = NULL; 6359 return (PF_DROP); 6360 } 6361 6362 /* Track state. */ 6363 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6364 if (src->state < SCTP_COOKIE_WAIT) { 6365 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6366 (*state)->timeout = PFTM_SCTP_OPENING; 6367 } 6368 } 6369 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6370 MPASS(dst->scrub != NULL); 6371 if (dst->scrub->pfss_v_tag == 0) 6372 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6373 } 6374 6375 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6376 if (src->state < SCTP_ESTABLISHED) { 6377 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6378 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6379 } 6380 } 6381 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6382 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6383 if (src->state < SCTP_SHUTDOWN_PENDING) { 6384 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6385 (*state)->timeout = PFTM_SCTP_CLOSING; 6386 } 6387 } 6388 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6389 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6390 (*state)->timeout = PFTM_SCTP_CLOSED; 6391 } 6392 6393 if (src->scrub != NULL) { 6394 if (src->scrub->pfss_v_tag == 0) { 6395 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6396 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6397 return (PF_DROP); 6398 } 6399 6400 (*state)->expire = pf_get_uptime(); 6401 6402 /* translate source/destination address, if necessary */ 6403 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6404 uint16_t checksum = 0; 6405 struct pf_state_key *nk = (*state)->key[pd->didx]; 6406 6407 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6408 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6409 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6410 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6411 nk->port[pd->sidx], 1, pd->af); 6412 } 6413 6414 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6415 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6416 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6417 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6418 nk->port[pd->didx], 1, pd->af); 6419 } 6420 } 6421 6422 return (PF_PASS); 6423 } 6424 6425 static void 6426 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6427 { 6428 struct pf_sctp_endpoint key; 6429 struct pf_sctp_endpoint *ep; 6430 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6431 struct pf_sctp_source *i, *tmp; 6432 6433 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6434 return; 6435 6436 PF_SCTP_ENDPOINTS_LOCK(); 6437 6438 key.v_tag = s->dst.scrub->pfss_v_tag; 6439 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6440 if (ep != NULL) { 6441 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6442 if (pf_addr_cmp(&i->addr, 6443 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6444 s->key[PF_SK_WIRE]->af) == 0) { 6445 SDT_PROBE3(pf, sctp, multihome, remove, 6446 key.v_tag, s, i); 6447 TAILQ_REMOVE(&ep->sources, i, entry); 6448 free(i, M_PFTEMP); 6449 break; 6450 } 6451 } 6452 6453 if (TAILQ_EMPTY(&ep->sources)) { 6454 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6455 free(ep, M_PFTEMP); 6456 } 6457 } 6458 6459 /* Other direction. */ 6460 key.v_tag = s->src.scrub->pfss_v_tag; 6461 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6462 if (ep != NULL) { 6463 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6464 if (pf_addr_cmp(&i->addr, 6465 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6466 s->key[PF_SK_WIRE]->af) == 0) { 6467 SDT_PROBE3(pf, sctp, multihome, remove, 6468 key.v_tag, s, i); 6469 TAILQ_REMOVE(&ep->sources, i, entry); 6470 free(i, M_PFTEMP); 6471 break; 6472 } 6473 } 6474 6475 if (TAILQ_EMPTY(&ep->sources)) { 6476 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6477 free(ep, M_PFTEMP); 6478 } 6479 } 6480 6481 PF_SCTP_ENDPOINTS_UNLOCK(); 6482 } 6483 6484 static void 6485 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6486 { 6487 struct pf_sctp_endpoint key = { 6488 .v_tag = v_tag, 6489 }; 6490 struct pf_sctp_source *i; 6491 struct pf_sctp_endpoint *ep; 6492 6493 PF_SCTP_ENDPOINTS_LOCK(); 6494 6495 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6496 if (ep == NULL) { 6497 ep = malloc(sizeof(struct pf_sctp_endpoint), 6498 M_PFTEMP, M_NOWAIT); 6499 if (ep == NULL) { 6500 PF_SCTP_ENDPOINTS_UNLOCK(); 6501 return; 6502 } 6503 6504 ep->v_tag = v_tag; 6505 TAILQ_INIT(&ep->sources); 6506 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6507 } 6508 6509 /* Avoid inserting duplicates. */ 6510 TAILQ_FOREACH(i, &ep->sources, entry) { 6511 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6512 PF_SCTP_ENDPOINTS_UNLOCK(); 6513 return; 6514 } 6515 } 6516 6517 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6518 if (i == NULL) { 6519 PF_SCTP_ENDPOINTS_UNLOCK(); 6520 return; 6521 } 6522 6523 i->af = pd->af; 6524 memcpy(&i->addr, a, sizeof(*a)); 6525 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6526 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6527 6528 PF_SCTP_ENDPOINTS_UNLOCK(); 6529 } 6530 6531 static void 6532 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6533 struct pf_kstate *s, int action) 6534 { 6535 struct pf_sctp_multihome_job *j, *tmp; 6536 struct pf_sctp_source *i; 6537 int ret __unused; 6538 struct pf_kstate *sm = NULL; 6539 struct pf_krule *ra = NULL; 6540 struct pf_krule *r = &V_pf_default_rule; 6541 struct pf_kruleset *rs = NULL; 6542 bool do_extra = true; 6543 6544 PF_RULES_RLOCK_TRACKER; 6545 6546 again: 6547 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6548 if (s == NULL || action != PF_PASS) 6549 goto free; 6550 6551 /* Confirm we don't recurse here. */ 6552 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6553 6554 switch (j->op) { 6555 case SCTP_ADD_IP_ADDRESS: { 6556 uint32_t v_tag = pd->sctp_initiate_tag; 6557 6558 if (v_tag == 0) { 6559 if (s->direction == pd->dir) 6560 v_tag = s->src.scrub->pfss_v_tag; 6561 else 6562 v_tag = s->dst.scrub->pfss_v_tag; 6563 } 6564 6565 /* 6566 * Avoid duplicating states. We'll already have 6567 * created a state based on the source address of 6568 * the packet, but SCTP endpoints may also list this 6569 * address again in the INIT(_ACK) parameters. 6570 */ 6571 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6572 break; 6573 } 6574 6575 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6576 PF_RULES_RLOCK(); 6577 sm = NULL; 6578 /* 6579 * New connections need to be floating, because 6580 * we cannot know what interfaces it will use. 6581 * That's why we pass V_pfi_all rather than kif. 6582 */ 6583 ret = pf_test_rule(&r, &sm, V_pfi_all, 6584 j->m, off, &j->pd, &ra, &rs, NULL, 6585 sizeof(j->pd.hdr.sctp)); 6586 PF_RULES_RUNLOCK(); 6587 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6588 if (ret != PF_DROP && sm != NULL) { 6589 /* Inherit v_tag values. */ 6590 if (sm->direction == s->direction) { 6591 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6592 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6593 } else { 6594 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6595 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6596 } 6597 PF_STATE_UNLOCK(sm); 6598 } else { 6599 /* If we try duplicate inserts? */ 6600 break; 6601 } 6602 6603 /* Only add the address if we've actually allowed the state. */ 6604 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6605 6606 if (! do_extra) { 6607 break; 6608 } 6609 /* 6610 * We need to do this for each of our source addresses. 6611 * Find those based on the verification tag. 6612 */ 6613 struct pf_sctp_endpoint key = { 6614 .v_tag = pd->hdr.sctp.v_tag, 6615 }; 6616 struct pf_sctp_endpoint *ep; 6617 6618 PF_SCTP_ENDPOINTS_LOCK(); 6619 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6620 if (ep == NULL) { 6621 PF_SCTP_ENDPOINTS_UNLOCK(); 6622 break; 6623 } 6624 MPASS(ep != NULL); 6625 6626 TAILQ_FOREACH(i, &ep->sources, entry) { 6627 struct pf_sctp_multihome_job *nj; 6628 6629 /* SCTP can intermingle IPv4 and IPv6. */ 6630 if (i->af != pd->af) 6631 continue; 6632 6633 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6634 if (! nj) { 6635 continue; 6636 } 6637 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6638 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6639 nj->pd.src = &nj->src; 6640 // New destination address! 6641 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6642 nj->pd.dst = &nj->dst; 6643 nj->m = j->m; 6644 nj->op = j->op; 6645 6646 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6647 } 6648 PF_SCTP_ENDPOINTS_UNLOCK(); 6649 6650 break; 6651 } 6652 case SCTP_DEL_IP_ADDRESS: { 6653 struct pf_state_key_cmp key; 6654 uint8_t psrc; 6655 6656 bzero(&key, sizeof(key)); 6657 key.af = j->pd.af; 6658 key.proto = IPPROTO_SCTP; 6659 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6660 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6661 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6662 key.port[0] = j->pd.hdr.sctp.src_port; 6663 key.port[1] = j->pd.hdr.sctp.dest_port; 6664 } else { /* stack side, reverse */ 6665 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6666 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6667 key.port[1] = j->pd.hdr.sctp.src_port; 6668 key.port[0] = j->pd.hdr.sctp.dest_port; 6669 } 6670 6671 sm = pf_find_state(kif, &key, j->pd.dir); 6672 if (sm != NULL) { 6673 PF_STATE_LOCK_ASSERT(sm); 6674 if (j->pd.dir == sm->direction) { 6675 psrc = PF_PEER_SRC; 6676 } else { 6677 psrc = PF_PEER_DST; 6678 } 6679 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6680 sm->timeout = PFTM_SCTP_CLOSING; 6681 PF_STATE_UNLOCK(sm); 6682 } 6683 break; 6684 default: 6685 panic("Unknown op %#x", j->op); 6686 } 6687 } 6688 6689 free: 6690 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6691 free(j, M_PFTEMP); 6692 } 6693 6694 /* We may have inserted extra work while processing the list. */ 6695 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6696 do_extra = false; 6697 goto again; 6698 } 6699 } 6700 6701 static int 6702 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6703 struct pfi_kkif *kif, int op) 6704 { 6705 int off = 0; 6706 struct pf_sctp_multihome_job *job; 6707 6708 while (off < len) { 6709 struct sctp_paramhdr h; 6710 6711 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6712 pd->af)) 6713 return (PF_DROP); 6714 6715 /* Parameters are at least 4 bytes. */ 6716 if (ntohs(h.param_length) < 4) 6717 return (PF_DROP); 6718 6719 switch (ntohs(h.param_type)) { 6720 case SCTP_IPV4_ADDRESS: { 6721 struct in_addr t; 6722 6723 if (ntohs(h.param_length) != 6724 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6725 return (PF_DROP); 6726 6727 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6728 NULL, NULL, pd->af)) 6729 return (PF_DROP); 6730 6731 if (in_nullhost(t)) 6732 t.s_addr = pd->src->v4.s_addr; 6733 6734 /* 6735 * We hold the state lock (idhash) here, which means 6736 * that we can't acquire the keyhash, or we'll get a 6737 * LOR (and potentially double-lock things too). We also 6738 * can't release the state lock here, so instead we'll 6739 * enqueue this for async handling. 6740 * There's a relatively small race here, in that a 6741 * packet using the new addresses could arrive already, 6742 * but that's just though luck for it. 6743 */ 6744 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6745 if (! job) 6746 return (PF_DROP); 6747 6748 memcpy(&job->pd, pd, sizeof(*pd)); 6749 6750 // New source address! 6751 memcpy(&job->src, &t, sizeof(t)); 6752 job->pd.src = &job->src; 6753 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6754 job->pd.dst = &job->dst; 6755 job->m = m; 6756 job->op = op; 6757 6758 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6759 break; 6760 } 6761 #ifdef INET6 6762 case SCTP_IPV6_ADDRESS: { 6763 struct in6_addr t; 6764 6765 if (ntohs(h.param_length) != 6766 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6767 return (PF_DROP); 6768 6769 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6770 NULL, NULL, pd->af)) 6771 return (PF_DROP); 6772 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6773 break; 6774 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6775 memcpy(&t, &pd->src->v6, sizeof(t)); 6776 6777 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6778 if (! job) 6779 return (PF_DROP); 6780 6781 memcpy(&job->pd, pd, sizeof(*pd)); 6782 memcpy(&job->src, &t, sizeof(t)); 6783 job->pd.src = &job->src; 6784 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6785 job->pd.dst = &job->dst; 6786 job->m = m; 6787 job->op = op; 6788 6789 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6790 break; 6791 } 6792 #endif 6793 case SCTP_ADD_IP_ADDRESS: { 6794 int ret; 6795 struct sctp_asconf_paramhdr ah; 6796 6797 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6798 NULL, NULL, pd->af)) 6799 return (PF_DROP); 6800 6801 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6802 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6803 SCTP_ADD_IP_ADDRESS); 6804 if (ret != PF_PASS) 6805 return (ret); 6806 break; 6807 } 6808 case SCTP_DEL_IP_ADDRESS: { 6809 int ret; 6810 struct sctp_asconf_paramhdr ah; 6811 6812 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6813 NULL, NULL, pd->af)) 6814 return (PF_DROP); 6815 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6816 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6817 SCTP_DEL_IP_ADDRESS); 6818 if (ret != PF_PASS) 6819 return (ret); 6820 break; 6821 } 6822 default: 6823 break; 6824 } 6825 6826 off += roundup(ntohs(h.param_length), 4); 6827 } 6828 6829 return (PF_PASS); 6830 } 6831 int 6832 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6833 struct pfi_kkif *kif) 6834 { 6835 start += sizeof(struct sctp_init_chunk); 6836 len -= sizeof(struct sctp_init_chunk); 6837 6838 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6839 } 6840 6841 int 6842 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6843 struct pf_pdesc *pd, struct pfi_kkif *kif) 6844 { 6845 start += sizeof(struct sctp_asconf_chunk); 6846 len -= sizeof(struct sctp_asconf_chunk); 6847 6848 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6849 } 6850 6851 int 6852 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6853 struct pf_kstate **state, struct mbuf *m, int off, int direction, 6854 struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir, 6855 int *iidx, int multi, int inner) 6856 { 6857 key->af = pd->af; 6858 key->proto = pd->proto; 6859 if (icmp_dir == PF_IN) { 6860 *iidx = pd->sidx; 6861 key->port[pd->sidx] = icmpid; 6862 key->port[pd->didx] = type; 6863 } else { 6864 *iidx = pd->didx; 6865 key->port[pd->sidx] = type; 6866 key->port[pd->didx] = icmpid; 6867 } 6868 if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src, 6869 pd->didx, pd->dst, multi)) 6870 return (PF_DROP); 6871 6872 STATE_LOOKUP(kif, key, *state, pd); 6873 6874 if ((*state)->state_flags & PFSTATE_SLOPPY) 6875 return (-1); 6876 6877 /* Is this ICMP message flowing in right direction? */ 6878 if ((*state)->rule->type && 6879 (((!inner && (*state)->direction == direction) || 6880 (inner && (*state)->direction != direction)) ? 6881 PF_IN : PF_OUT) != icmp_dir) { 6882 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6883 printf("pf: icmp type %d in wrong direction (%d): ", 6884 ntohs(type), icmp_dir); 6885 pf_print_state(*state); 6886 printf("\n"); 6887 } 6888 PF_STATE_UNLOCK(*state); 6889 *state = NULL; 6890 return (PF_DROP); 6891 } 6892 return (-1); 6893 } 6894 6895 static int 6896 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6897 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason) 6898 { 6899 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6900 u_int16_t *icmpsum, virtual_id, virtual_type; 6901 u_int8_t icmptype, icmpcode; 6902 int icmp_dir, iidx, ret, multi; 6903 struct pf_state_key_cmp key; 6904 #ifdef INET 6905 u_int16_t icmpid; 6906 #endif 6907 6908 MPASS(*state == NULL); 6909 6910 bzero(&key, sizeof(key)); 6911 switch (pd->proto) { 6912 #ifdef INET 6913 case IPPROTO_ICMP: 6914 icmptype = pd->hdr.icmp.icmp_type; 6915 icmpcode = pd->hdr.icmp.icmp_code; 6916 icmpid = pd->hdr.icmp.icmp_id; 6917 icmpsum = &pd->hdr.icmp.icmp_cksum; 6918 break; 6919 #endif /* INET */ 6920 #ifdef INET6 6921 case IPPROTO_ICMPV6: 6922 icmptype = pd->hdr.icmp6.icmp6_type; 6923 icmpcode = pd->hdr.icmp6.icmp6_code; 6924 #ifdef INET 6925 icmpid = pd->hdr.icmp6.icmp6_id; 6926 #endif 6927 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6928 break; 6929 #endif /* INET6 */ 6930 } 6931 6932 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 6933 &virtual_id, &virtual_type) == 0) { 6934 /* 6935 * ICMP query/reply message not related to a TCP/UDP packet. 6936 * Search for an ICMP state. 6937 */ 6938 ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir, 6939 kif, virtual_id, virtual_type, icmp_dir, &iidx, 6940 PF_ICMP_MULTI_NONE, 0); 6941 if (ret >= 0) { 6942 MPASS(*state == NULL); 6943 if (ret == PF_DROP && pd->af == AF_INET6 && 6944 icmp_dir == PF_OUT) { 6945 ret = pf_icmp_state_lookup(&key, pd, state, m, off, 6946 pd->dir, kif, virtual_id, virtual_type, 6947 icmp_dir, &iidx, multi, 0); 6948 if (ret >= 0) { 6949 MPASS(*state == NULL); 6950 return (ret); 6951 } 6952 } else 6953 return (ret); 6954 } 6955 6956 (*state)->expire = pf_get_uptime(); 6957 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6958 6959 /* translate source/destination address, if necessary */ 6960 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6961 struct pf_state_key *nk = (*state)->key[pd->didx]; 6962 6963 switch (pd->af) { 6964 #ifdef INET 6965 case AF_INET: 6966 if (PF_ANEQ(pd->src, 6967 &nk->addr[pd->sidx], AF_INET)) 6968 pf_change_a(&saddr->v4.s_addr, 6969 pd->ip_sum, 6970 nk->addr[pd->sidx].v4.s_addr, 0); 6971 6972 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6973 AF_INET)) 6974 pf_change_a(&daddr->v4.s_addr, 6975 pd->ip_sum, 6976 nk->addr[pd->didx].v4.s_addr, 0); 6977 6978 if (nk->port[iidx] != 6979 pd->hdr.icmp.icmp_id) { 6980 pd->hdr.icmp.icmp_cksum = 6981 pf_cksum_fixup( 6982 pd->hdr.icmp.icmp_cksum, icmpid, 6983 nk->port[iidx], 0); 6984 pd->hdr.icmp.icmp_id = 6985 nk->port[iidx]; 6986 } 6987 6988 m_copyback(m, off, ICMP_MINLEN, 6989 (caddr_t )&pd->hdr.icmp); 6990 break; 6991 #endif /* INET */ 6992 #ifdef INET6 6993 case AF_INET6: 6994 if (PF_ANEQ(pd->src, 6995 &nk->addr[pd->sidx], AF_INET6)) 6996 pf_change_a6(saddr, 6997 &pd->hdr.icmp6.icmp6_cksum, 6998 &nk->addr[pd->sidx], 0); 6999 7000 if (PF_ANEQ(pd->dst, 7001 &nk->addr[pd->didx], AF_INET6)) 7002 pf_change_a6(daddr, 7003 &pd->hdr.icmp6.icmp6_cksum, 7004 &nk->addr[pd->didx], 0); 7005 7006 m_copyback(m, off, sizeof(struct icmp6_hdr), 7007 (caddr_t )&pd->hdr.icmp6); 7008 break; 7009 #endif /* INET6 */ 7010 } 7011 } 7012 return (PF_PASS); 7013 7014 } else { 7015 /* 7016 * ICMP error message in response to a TCP/UDP packet. 7017 * Extract the inner TCP/UDP header and search for that state. 7018 */ 7019 7020 struct pf_pdesc pd2; 7021 bzero(&pd2, sizeof pd2); 7022 #ifdef INET 7023 struct ip h2; 7024 #endif /* INET */ 7025 #ifdef INET6 7026 struct ip6_hdr h2_6; 7027 int terminal = 0; 7028 #endif /* INET6 */ 7029 int ipoff2 = 0; 7030 int off2 = 0; 7031 7032 pd2.af = pd->af; 7033 pd2.dir = pd->dir; 7034 /* Payload packet is from the opposite direction. */ 7035 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7036 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7037 switch (pd->af) { 7038 #ifdef INET 7039 case AF_INET: 7040 /* offset of h2 in mbuf chain */ 7041 ipoff2 = off + ICMP_MINLEN; 7042 7043 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 7044 NULL, reason, pd2.af)) { 7045 DPFPRINTF(PF_DEBUG_MISC, 7046 ("pf: ICMP error message too short " 7047 "(ip)\n")); 7048 return (PF_DROP); 7049 } 7050 /* 7051 * ICMP error messages don't refer to non-first 7052 * fragments 7053 */ 7054 if (h2.ip_off & htons(IP_OFFMASK)) { 7055 REASON_SET(reason, PFRES_FRAG); 7056 return (PF_DROP); 7057 } 7058 7059 /* offset of protocol header that follows h2 */ 7060 off2 = ipoff2 + (h2.ip_hl << 2); 7061 7062 pd2.proto = h2.ip_p; 7063 pd2.src = (struct pf_addr *)&h2.ip_src; 7064 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7065 pd2.ip_sum = &h2.ip_sum; 7066 break; 7067 #endif /* INET */ 7068 #ifdef INET6 7069 case AF_INET6: 7070 ipoff2 = off + sizeof(struct icmp6_hdr); 7071 7072 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 7073 NULL, reason, pd2.af)) { 7074 DPFPRINTF(PF_DEBUG_MISC, 7075 ("pf: ICMP error message too short " 7076 "(ip6)\n")); 7077 return (PF_DROP); 7078 } 7079 pd2.proto = h2_6.ip6_nxt; 7080 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7081 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7082 pd2.ip_sum = NULL; 7083 off2 = ipoff2 + sizeof(h2_6); 7084 do { 7085 switch (pd2.proto) { 7086 case IPPROTO_FRAGMENT: 7087 /* 7088 * ICMPv6 error messages for 7089 * non-first fragments 7090 */ 7091 REASON_SET(reason, PFRES_FRAG); 7092 return (PF_DROP); 7093 case IPPROTO_AH: 7094 case IPPROTO_HOPOPTS: 7095 case IPPROTO_ROUTING: 7096 case IPPROTO_DSTOPTS: { 7097 /* get next header and header length */ 7098 struct ip6_ext opt6; 7099 7100 if (!pf_pull_hdr(m, off2, &opt6, 7101 sizeof(opt6), NULL, reason, 7102 pd2.af)) { 7103 DPFPRINTF(PF_DEBUG_MISC, 7104 ("pf: ICMPv6 short opt\n")); 7105 return (PF_DROP); 7106 } 7107 if (pd2.proto == IPPROTO_AH) 7108 off2 += (opt6.ip6e_len + 2) * 4; 7109 else 7110 off2 += (opt6.ip6e_len + 1) * 8; 7111 pd2.proto = opt6.ip6e_nxt; 7112 /* goto the next header */ 7113 break; 7114 } 7115 default: 7116 terminal++; 7117 break; 7118 } 7119 } while (!terminal); 7120 break; 7121 #endif /* INET6 */ 7122 } 7123 7124 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7125 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7126 printf("pf: BAD ICMP %d:%d outer dst: ", 7127 icmptype, icmpcode); 7128 pf_print_host(pd->src, 0, pd->af); 7129 printf(" -> "); 7130 pf_print_host(pd->dst, 0, pd->af); 7131 printf(" inner src: "); 7132 pf_print_host(pd2.src, 0, pd2.af); 7133 printf(" -> "); 7134 pf_print_host(pd2.dst, 0, pd2.af); 7135 printf("\n"); 7136 } 7137 REASON_SET(reason, PFRES_BADSTATE); 7138 return (PF_DROP); 7139 } 7140 7141 switch (pd2.proto) { 7142 case IPPROTO_TCP: { 7143 struct tcphdr th; 7144 u_int32_t seq; 7145 struct pf_state_peer *src, *dst; 7146 u_int8_t dws; 7147 int copyback = 0; 7148 7149 /* 7150 * Only the first 8 bytes of the TCP header can be 7151 * expected. Don't access any TCP header fields after 7152 * th_seq, an ackskew test is not possible. 7153 */ 7154 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 7155 pd2.af)) { 7156 DPFPRINTF(PF_DEBUG_MISC, 7157 ("pf: ICMP error message too short " 7158 "(tcp)\n")); 7159 return (PF_DROP); 7160 } 7161 7162 key.af = pd2.af; 7163 key.proto = IPPROTO_TCP; 7164 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7165 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7166 key.port[pd2.sidx] = th.th_sport; 7167 key.port[pd2.didx] = th.th_dport; 7168 7169 STATE_LOOKUP(kif, &key, *state, pd); 7170 7171 if (pd->dir == (*state)->direction) { 7172 src = &(*state)->dst; 7173 dst = &(*state)->src; 7174 } else { 7175 src = &(*state)->src; 7176 dst = &(*state)->dst; 7177 } 7178 7179 if (src->wscale && dst->wscale) 7180 dws = dst->wscale & PF_WSCALE_MASK; 7181 else 7182 dws = 0; 7183 7184 /* Demodulate sequence number */ 7185 seq = ntohl(th.th_seq) - src->seqdiff; 7186 if (src->seqdiff) { 7187 pf_change_a(&th.th_seq, icmpsum, 7188 htonl(seq), 0); 7189 copyback = 1; 7190 } 7191 7192 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7193 (!SEQ_GEQ(src->seqhi, seq) || 7194 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7195 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7196 printf("pf: BAD ICMP %d:%d ", 7197 icmptype, icmpcode); 7198 pf_print_host(pd->src, 0, pd->af); 7199 printf(" -> "); 7200 pf_print_host(pd->dst, 0, pd->af); 7201 printf(" state: "); 7202 pf_print_state(*state); 7203 printf(" seq=%u\n", seq); 7204 } 7205 REASON_SET(reason, PFRES_BADSTATE); 7206 return (PF_DROP); 7207 } else { 7208 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7209 printf("pf: OK ICMP %d:%d ", 7210 icmptype, icmpcode); 7211 pf_print_host(pd->src, 0, pd->af); 7212 printf(" -> "); 7213 pf_print_host(pd->dst, 0, pd->af); 7214 printf(" state: "); 7215 pf_print_state(*state); 7216 printf(" seq=%u\n", seq); 7217 } 7218 } 7219 7220 /* translate source/destination address, if necessary */ 7221 if ((*state)->key[PF_SK_WIRE] != 7222 (*state)->key[PF_SK_STACK]) { 7223 struct pf_state_key *nk = 7224 (*state)->key[pd->didx]; 7225 7226 if (PF_ANEQ(pd2.src, 7227 &nk->addr[pd2.sidx], pd2.af) || 7228 nk->port[pd2.sidx] != th.th_sport) 7229 pf_change_icmp(pd2.src, &th.th_sport, 7230 daddr, &nk->addr[pd2.sidx], 7231 nk->port[pd2.sidx], NULL, 7232 pd2.ip_sum, icmpsum, 7233 pd->ip_sum, 0, pd2.af); 7234 7235 if (PF_ANEQ(pd2.dst, 7236 &nk->addr[pd2.didx], pd2.af) || 7237 nk->port[pd2.didx] != th.th_dport) 7238 pf_change_icmp(pd2.dst, &th.th_dport, 7239 saddr, &nk->addr[pd2.didx], 7240 nk->port[pd2.didx], NULL, 7241 pd2.ip_sum, icmpsum, 7242 pd->ip_sum, 0, pd2.af); 7243 copyback = 1; 7244 } 7245 7246 if (copyback) { 7247 switch (pd2.af) { 7248 #ifdef INET 7249 case AF_INET: 7250 m_copyback(m, off, ICMP_MINLEN, 7251 (caddr_t )&pd->hdr.icmp); 7252 m_copyback(m, ipoff2, sizeof(h2), 7253 (caddr_t )&h2); 7254 break; 7255 #endif /* INET */ 7256 #ifdef INET6 7257 case AF_INET6: 7258 m_copyback(m, off, 7259 sizeof(struct icmp6_hdr), 7260 (caddr_t )&pd->hdr.icmp6); 7261 m_copyback(m, ipoff2, sizeof(h2_6), 7262 (caddr_t )&h2_6); 7263 break; 7264 #endif /* INET6 */ 7265 } 7266 m_copyback(m, off2, 8, (caddr_t)&th); 7267 } 7268 7269 return (PF_PASS); 7270 break; 7271 } 7272 case IPPROTO_UDP: { 7273 struct udphdr uh; 7274 7275 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 7276 NULL, reason, pd2.af)) { 7277 DPFPRINTF(PF_DEBUG_MISC, 7278 ("pf: ICMP error message too short " 7279 "(udp)\n")); 7280 return (PF_DROP); 7281 } 7282 7283 key.af = pd2.af; 7284 key.proto = IPPROTO_UDP; 7285 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7286 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7287 key.port[pd2.sidx] = uh.uh_sport; 7288 key.port[pd2.didx] = uh.uh_dport; 7289 7290 STATE_LOOKUP(kif, &key, *state, pd); 7291 7292 /* translate source/destination address, if necessary */ 7293 if ((*state)->key[PF_SK_WIRE] != 7294 (*state)->key[PF_SK_STACK]) { 7295 struct pf_state_key *nk = 7296 (*state)->key[pd->didx]; 7297 7298 if (PF_ANEQ(pd2.src, 7299 &nk->addr[pd2.sidx], pd2.af) || 7300 nk->port[pd2.sidx] != uh.uh_sport) 7301 pf_change_icmp(pd2.src, &uh.uh_sport, 7302 daddr, &nk->addr[pd2.sidx], 7303 nk->port[pd2.sidx], &uh.uh_sum, 7304 pd2.ip_sum, icmpsum, 7305 pd->ip_sum, 1, pd2.af); 7306 7307 if (PF_ANEQ(pd2.dst, 7308 &nk->addr[pd2.didx], pd2.af) || 7309 nk->port[pd2.didx] != uh.uh_dport) 7310 pf_change_icmp(pd2.dst, &uh.uh_dport, 7311 saddr, &nk->addr[pd2.didx], 7312 nk->port[pd2.didx], &uh.uh_sum, 7313 pd2.ip_sum, icmpsum, 7314 pd->ip_sum, 1, pd2.af); 7315 7316 switch (pd2.af) { 7317 #ifdef INET 7318 case AF_INET: 7319 m_copyback(m, off, ICMP_MINLEN, 7320 (caddr_t )&pd->hdr.icmp); 7321 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7322 break; 7323 #endif /* INET */ 7324 #ifdef INET6 7325 case AF_INET6: 7326 m_copyback(m, off, 7327 sizeof(struct icmp6_hdr), 7328 (caddr_t )&pd->hdr.icmp6); 7329 m_copyback(m, ipoff2, sizeof(h2_6), 7330 (caddr_t )&h2_6); 7331 break; 7332 #endif /* INET6 */ 7333 } 7334 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 7335 } 7336 return (PF_PASS); 7337 break; 7338 } 7339 #ifdef INET 7340 case IPPROTO_ICMP: { 7341 struct icmp *iih = &pd2.hdr.icmp; 7342 7343 if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN, 7344 NULL, reason, pd2.af)) { 7345 DPFPRINTF(PF_DEBUG_MISC, 7346 ("pf: ICMP error message too short i" 7347 "(icmp)\n")); 7348 return (PF_DROP); 7349 } 7350 7351 icmpid = iih->icmp_id; 7352 pf_icmp_mapping(&pd2, iih->icmp_type, 7353 &icmp_dir, &multi, &virtual_id, &virtual_type); 7354 7355 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7356 pd2.dir, kif, virtual_id, virtual_type, 7357 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7358 if (ret >= 0) { 7359 MPASS(*state == NULL); 7360 return (ret); 7361 } 7362 7363 /* translate source/destination address, if necessary */ 7364 if ((*state)->key[PF_SK_WIRE] != 7365 (*state)->key[PF_SK_STACK]) { 7366 struct pf_state_key *nk = 7367 (*state)->key[pd->didx]; 7368 7369 if (PF_ANEQ(pd2.src, 7370 &nk->addr[pd2.sidx], pd2.af) || 7371 (virtual_type == htons(ICMP_ECHO) && 7372 nk->port[iidx] != iih->icmp_id)) 7373 pf_change_icmp(pd2.src, 7374 (virtual_type == htons(ICMP_ECHO)) ? 7375 &iih->icmp_id : NULL, 7376 daddr, &nk->addr[pd2.sidx], 7377 (virtual_type == htons(ICMP_ECHO)) ? 7378 nk->port[iidx] : 0, NULL, 7379 pd2.ip_sum, icmpsum, 7380 pd->ip_sum, 0, AF_INET); 7381 7382 if (PF_ANEQ(pd2.dst, 7383 &nk->addr[pd2.didx], pd2.af)) 7384 pf_change_icmp(pd2.dst, NULL, NULL, 7385 &nk->addr[pd2.didx], 0, NULL, 7386 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7387 AF_INET); 7388 7389 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7390 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7391 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih); 7392 } 7393 return (PF_PASS); 7394 break; 7395 } 7396 #endif /* INET */ 7397 #ifdef INET6 7398 case IPPROTO_ICMPV6: { 7399 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7400 7401 if (!pf_pull_hdr(m, off2, iih, 7402 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7403 DPFPRINTF(PF_DEBUG_MISC, 7404 ("pf: ICMP error message too short " 7405 "(icmp6)\n")); 7406 return (PF_DROP); 7407 } 7408 7409 pf_icmp_mapping(&pd2, iih->icmp6_type, 7410 &icmp_dir, &multi, &virtual_id, &virtual_type); 7411 7412 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7413 pd->dir, kif, virtual_id, virtual_type, 7414 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7415 if (ret >= 0) { 7416 MPASS(*state == NULL); 7417 if (ret == PF_DROP && pd2.af == AF_INET6 && 7418 icmp_dir == PF_OUT) { 7419 ret = pf_icmp_state_lookup(&key, &pd2, 7420 state, m, off, pd->dir, kif, 7421 virtual_id, virtual_type, 7422 icmp_dir, &iidx, multi, 1); 7423 if (ret >= 0) { 7424 MPASS(*state == NULL); 7425 return (ret); 7426 } 7427 } else 7428 return (ret); 7429 } 7430 7431 /* translate source/destination address, if necessary */ 7432 if ((*state)->key[PF_SK_WIRE] != 7433 (*state)->key[PF_SK_STACK]) { 7434 struct pf_state_key *nk = 7435 (*state)->key[pd->didx]; 7436 7437 if (PF_ANEQ(pd2.src, 7438 &nk->addr[pd2.sidx], pd2.af) || 7439 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7440 nk->port[pd2.sidx] != iih->icmp6_id)) 7441 pf_change_icmp(pd2.src, 7442 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7443 ? &iih->icmp6_id : NULL, 7444 daddr, &nk->addr[pd2.sidx], 7445 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7446 ? nk->port[iidx] : 0, NULL, 7447 pd2.ip_sum, icmpsum, 7448 pd->ip_sum, 0, AF_INET6); 7449 7450 if (PF_ANEQ(pd2.dst, 7451 &nk->addr[pd2.didx], pd2.af)) 7452 pf_change_icmp(pd2.dst, NULL, NULL, 7453 &nk->addr[pd2.didx], 0, NULL, 7454 pd2.ip_sum, icmpsum, 7455 pd->ip_sum, 0, AF_INET6); 7456 7457 m_copyback(m, off, sizeof(struct icmp6_hdr), 7458 (caddr_t)&pd->hdr.icmp6); 7459 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7460 m_copyback(m, off2, sizeof(struct icmp6_hdr), 7461 (caddr_t)iih); 7462 } 7463 return (PF_PASS); 7464 break; 7465 } 7466 #endif /* INET6 */ 7467 default: { 7468 key.af = pd2.af; 7469 key.proto = pd2.proto; 7470 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7471 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7472 key.port[0] = key.port[1] = 0; 7473 7474 STATE_LOOKUP(kif, &key, *state, pd); 7475 7476 /* translate source/destination address, if necessary */ 7477 if ((*state)->key[PF_SK_WIRE] != 7478 (*state)->key[PF_SK_STACK]) { 7479 struct pf_state_key *nk = 7480 (*state)->key[pd->didx]; 7481 7482 if (PF_ANEQ(pd2.src, 7483 &nk->addr[pd2.sidx], pd2.af)) 7484 pf_change_icmp(pd2.src, NULL, daddr, 7485 &nk->addr[pd2.sidx], 0, NULL, 7486 pd2.ip_sum, icmpsum, 7487 pd->ip_sum, 0, pd2.af); 7488 7489 if (PF_ANEQ(pd2.dst, 7490 &nk->addr[pd2.didx], pd2.af)) 7491 pf_change_icmp(pd2.dst, NULL, saddr, 7492 &nk->addr[pd2.didx], 0, NULL, 7493 pd2.ip_sum, icmpsum, 7494 pd->ip_sum, 0, pd2.af); 7495 7496 switch (pd2.af) { 7497 #ifdef INET 7498 case AF_INET: 7499 m_copyback(m, off, ICMP_MINLEN, 7500 (caddr_t)&pd->hdr.icmp); 7501 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7502 break; 7503 #endif /* INET */ 7504 #ifdef INET6 7505 case AF_INET6: 7506 m_copyback(m, off, 7507 sizeof(struct icmp6_hdr), 7508 (caddr_t )&pd->hdr.icmp6); 7509 m_copyback(m, ipoff2, sizeof(h2_6), 7510 (caddr_t )&h2_6); 7511 break; 7512 #endif /* INET6 */ 7513 } 7514 } 7515 return (PF_PASS); 7516 break; 7517 } 7518 } 7519 } 7520 } 7521 7522 static int 7523 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7524 struct mbuf *m, struct pf_pdesc *pd) 7525 { 7526 struct pf_state_peer *src, *dst; 7527 struct pf_state_key_cmp key; 7528 uint8_t psrc, pdst; 7529 7530 bzero(&key, sizeof(key)); 7531 key.af = pd->af; 7532 key.proto = pd->proto; 7533 if (pd->dir == PF_IN) { 7534 PF_ACPY(&key.addr[0], pd->src, key.af); 7535 PF_ACPY(&key.addr[1], pd->dst, key.af); 7536 key.port[0] = key.port[1] = 0; 7537 } else { 7538 PF_ACPY(&key.addr[1], pd->src, key.af); 7539 PF_ACPY(&key.addr[0], pd->dst, key.af); 7540 key.port[1] = key.port[0] = 0; 7541 } 7542 7543 STATE_LOOKUP(kif, &key, *state, pd); 7544 7545 if (pd->dir == (*state)->direction) { 7546 src = &(*state)->src; 7547 dst = &(*state)->dst; 7548 psrc = PF_PEER_SRC; 7549 pdst = PF_PEER_DST; 7550 } else { 7551 src = &(*state)->dst; 7552 dst = &(*state)->src; 7553 psrc = PF_PEER_DST; 7554 pdst = PF_PEER_SRC; 7555 } 7556 7557 /* update states */ 7558 if (src->state < PFOTHERS_SINGLE) 7559 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7560 if (dst->state == PFOTHERS_SINGLE) 7561 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7562 7563 /* update expire time */ 7564 (*state)->expire = pf_get_uptime(); 7565 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7566 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7567 else 7568 (*state)->timeout = PFTM_OTHER_SINGLE; 7569 7570 /* translate source/destination address, if necessary */ 7571 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7572 struct pf_state_key *nk = (*state)->key[pd->didx]; 7573 7574 KASSERT(nk, ("%s: nk is null", __func__)); 7575 KASSERT(pd, ("%s: pd is null", __func__)); 7576 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7577 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7578 switch (pd->af) { 7579 #ifdef INET 7580 case AF_INET: 7581 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7582 pf_change_a(&pd->src->v4.s_addr, 7583 pd->ip_sum, 7584 nk->addr[pd->sidx].v4.s_addr, 7585 0); 7586 7587 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7588 pf_change_a(&pd->dst->v4.s_addr, 7589 pd->ip_sum, 7590 nk->addr[pd->didx].v4.s_addr, 7591 0); 7592 7593 break; 7594 #endif /* INET */ 7595 #ifdef INET6 7596 case AF_INET6: 7597 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7598 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7599 7600 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7601 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7602 #endif /* INET6 */ 7603 } 7604 } 7605 return (PF_PASS); 7606 } 7607 7608 /* 7609 * ipoff and off are measured from the start of the mbuf chain. 7610 * h must be at "ipoff" on the mbuf chain. 7611 */ 7612 void * 7613 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7614 u_short *actionp, u_short *reasonp, sa_family_t af) 7615 { 7616 switch (af) { 7617 #ifdef INET 7618 case AF_INET: { 7619 const struct ip *h = mtod(m, struct ip *); 7620 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7621 7622 if (fragoff) { 7623 if (fragoff >= len) 7624 ACTION_SET(actionp, PF_PASS); 7625 else { 7626 ACTION_SET(actionp, PF_DROP); 7627 REASON_SET(reasonp, PFRES_FRAG); 7628 } 7629 return (NULL); 7630 } 7631 if (m->m_pkthdr.len < off + len || 7632 ntohs(h->ip_len) < off + len) { 7633 ACTION_SET(actionp, PF_DROP); 7634 REASON_SET(reasonp, PFRES_SHORT); 7635 return (NULL); 7636 } 7637 break; 7638 } 7639 #endif /* INET */ 7640 #ifdef INET6 7641 case AF_INET6: { 7642 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7643 7644 if (m->m_pkthdr.len < off + len || 7645 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7646 (unsigned)(off + len)) { 7647 ACTION_SET(actionp, PF_DROP); 7648 REASON_SET(reasonp, PFRES_SHORT); 7649 return (NULL); 7650 } 7651 break; 7652 } 7653 #endif /* INET6 */ 7654 } 7655 m_copydata(m, off, len, p); 7656 return (p); 7657 } 7658 7659 int 7660 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7661 int rtableid) 7662 { 7663 struct ifnet *ifp; 7664 7665 /* 7666 * Skip check for addresses with embedded interface scope, 7667 * as they would always match anyway. 7668 */ 7669 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7670 return (1); 7671 7672 if (af != AF_INET && af != AF_INET6) 7673 return (0); 7674 7675 if (kif == V_pfi_all) 7676 return (1); 7677 7678 /* Skip checks for ipsec interfaces */ 7679 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7680 return (1); 7681 7682 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7683 7684 switch (af) { 7685 #ifdef INET6 7686 case AF_INET6: 7687 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7688 ifp)); 7689 #endif 7690 #ifdef INET 7691 case AF_INET: 7692 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7693 ifp)); 7694 #endif 7695 } 7696 7697 return (0); 7698 } 7699 7700 #ifdef INET 7701 static void 7702 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7703 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7704 { 7705 struct mbuf *m0, *m1, *md; 7706 struct sockaddr_in dst; 7707 struct ip *ip; 7708 struct pfi_kkif *nkif = NULL; 7709 struct ifnet *ifp = NULL; 7710 struct pf_addr naddr; 7711 int error = 0; 7712 uint16_t ip_len, ip_off; 7713 uint16_t tmp; 7714 int r_rt, r_dir; 7715 7716 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7717 7718 if (s) { 7719 r_rt = s->rt; 7720 r_dir = s->direction; 7721 } else { 7722 r_rt = r->rt; 7723 r_dir = r->direction; 7724 } 7725 7726 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7727 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7728 __func__)); 7729 7730 if ((pd->pf_mtag == NULL && 7731 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7732 pd->pf_mtag->routed++ > 3) { 7733 m0 = *m; 7734 *m = NULL; 7735 goto bad_locked; 7736 } 7737 7738 if (r_rt == PF_DUPTO) { 7739 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7740 if (s == NULL) { 7741 ifp = r->rpool.cur->kif ? 7742 r->rpool.cur->kif->pfik_ifp : NULL; 7743 } else { 7744 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7745 /* If pfsync'd */ 7746 if (ifp == NULL && r->rpool.cur != NULL) 7747 ifp = r->rpool.cur->kif ? 7748 r->rpool.cur->kif->pfik_ifp : NULL; 7749 PF_STATE_UNLOCK(s); 7750 } 7751 if (ifp == oifp) { 7752 /* When the 2nd interface is not skipped */ 7753 return; 7754 } else { 7755 m0 = *m; 7756 *m = NULL; 7757 goto bad; 7758 } 7759 } else { 7760 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7761 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7762 if (s) 7763 PF_STATE_UNLOCK(s); 7764 return; 7765 } 7766 } 7767 } else { 7768 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7769 pf_dummynet(pd, s, r, m); 7770 if (s) 7771 PF_STATE_UNLOCK(s); 7772 return; 7773 } 7774 m0 = *m; 7775 } 7776 7777 ip = mtod(m0, struct ip *); 7778 7779 bzero(&dst, sizeof(dst)); 7780 dst.sin_family = AF_INET; 7781 dst.sin_len = sizeof(dst); 7782 dst.sin_addr = ip->ip_dst; 7783 7784 bzero(&naddr, sizeof(naddr)); 7785 7786 if (s == NULL) { 7787 if (TAILQ_EMPTY(&r->rpool.list)) { 7788 DPFPRINTF(PF_DEBUG_URGENT, 7789 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7790 goto bad_locked; 7791 } 7792 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7793 &naddr, &nkif, NULL); 7794 if (!PF_AZERO(&naddr, AF_INET)) 7795 dst.sin_addr.s_addr = naddr.v4.s_addr; 7796 ifp = nkif ? nkif->pfik_ifp : NULL; 7797 } else { 7798 struct pfi_kkif *kif; 7799 7800 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7801 dst.sin_addr.s_addr = 7802 s->rt_addr.v4.s_addr; 7803 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7804 kif = s->rt_kif; 7805 /* If pfsync'd */ 7806 if (ifp == NULL && r->rpool.cur != NULL) { 7807 ifp = r->rpool.cur->kif ? 7808 r->rpool.cur->kif->pfik_ifp : NULL; 7809 kif = r->rpool.cur->kif; 7810 } 7811 if (ifp != NULL && kif != NULL && 7812 r->rule_flag & PFRULE_IFBOUND && 7813 r->rt == PF_REPLYTO && 7814 s->kif == V_pfi_all) { 7815 s->kif = kif; 7816 s->orig_kif = oifp->if_pf_kif; 7817 } 7818 7819 PF_STATE_UNLOCK(s); 7820 } 7821 7822 if (ifp == NULL) 7823 goto bad; 7824 7825 if (pd->dir == PF_IN) { 7826 if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, 7827 &pd->act) != PF_PASS) 7828 goto bad; 7829 else if (m0 == NULL) 7830 goto done; 7831 if (m0->m_len < sizeof(struct ip)) { 7832 DPFPRINTF(PF_DEBUG_URGENT, 7833 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7834 goto bad; 7835 } 7836 ip = mtod(m0, struct ip *); 7837 } 7838 7839 if (ifp->if_flags & IFF_LOOPBACK) 7840 m0->m_flags |= M_SKIP_FIREWALL; 7841 7842 ip_len = ntohs(ip->ip_len); 7843 ip_off = ntohs(ip->ip_off); 7844 7845 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7846 m0->m_pkthdr.csum_flags |= CSUM_IP; 7847 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7848 in_delayed_cksum(m0); 7849 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7850 } 7851 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7852 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7853 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7854 } 7855 7856 if (pd->dir == PF_IN) { 7857 /* 7858 * Make sure dummynet gets the correct direction, in case it needs to 7859 * re-inject later. 7860 */ 7861 pd->dir = PF_OUT; 7862 7863 /* 7864 * The following processing is actually the rest of the inbound processing, even 7865 * though we've marked it as outbound (so we don't look through dummynet) and it 7866 * happens after the outbound processing (pf_test(PF_OUT) above). 7867 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7868 * conclusion about what direction it's processing, and we can't fix it or it 7869 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7870 * decision will pick the right pipe, and everything will mostly work as expected. 7871 */ 7872 tmp = pd->act.dnrpipe; 7873 pd->act.dnrpipe = pd->act.dnpipe; 7874 pd->act.dnpipe = tmp; 7875 } 7876 7877 /* 7878 * If small enough for interface, or the interface will take 7879 * care of the fragmentation for us, we can just send directly. 7880 */ 7881 if (ip_len <= ifp->if_mtu || 7882 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7883 ip->ip_sum = 0; 7884 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7885 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7886 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7887 } 7888 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7889 7890 md = m0; 7891 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7892 if (md != NULL) 7893 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7894 goto done; 7895 } 7896 7897 /* Balk when DF bit is set or the interface didn't support TSO. */ 7898 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7899 error = EMSGSIZE; 7900 KMOD_IPSTAT_INC(ips_cantfrag); 7901 if (r_rt != PF_DUPTO) { 7902 if (s && s->nat_rule != NULL) 7903 PACKET_UNDO_NAT(m0, pd, 7904 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7905 s); 7906 7907 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7908 ifp->if_mtu); 7909 goto done; 7910 } else 7911 goto bad; 7912 } 7913 7914 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7915 if (error) 7916 goto bad; 7917 7918 for (; m0; m0 = m1) { 7919 m1 = m0->m_nextpkt; 7920 m0->m_nextpkt = NULL; 7921 if (error == 0) { 7922 m_clrprotoflags(m0); 7923 md = m0; 7924 pd->pf_mtag = pf_find_mtag(md); 7925 error = pf_dummynet_route(pd, s, r, ifp, 7926 sintosa(&dst), &md); 7927 if (md != NULL) 7928 error = (*ifp->if_output)(ifp, md, 7929 sintosa(&dst), NULL); 7930 } else 7931 m_freem(m0); 7932 } 7933 7934 if (error == 0) 7935 KMOD_IPSTAT_INC(ips_fragmented); 7936 7937 done: 7938 if (r_rt != PF_DUPTO) 7939 *m = NULL; 7940 return; 7941 7942 bad_locked: 7943 if (s) 7944 PF_STATE_UNLOCK(s); 7945 bad: 7946 m_freem(m0); 7947 goto done; 7948 } 7949 #endif /* INET */ 7950 7951 #ifdef INET6 7952 static void 7953 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7954 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7955 { 7956 struct mbuf *m0, *md; 7957 struct sockaddr_in6 dst; 7958 struct ip6_hdr *ip6; 7959 struct pfi_kkif *nkif = NULL; 7960 struct ifnet *ifp = NULL; 7961 struct pf_addr naddr; 7962 int r_rt, r_dir; 7963 7964 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7965 7966 if (s) { 7967 r_rt = s->rt; 7968 r_dir = s->direction; 7969 } else { 7970 r_rt = r->rt; 7971 r_dir = r->direction; 7972 } 7973 7974 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7975 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7976 __func__)); 7977 7978 if ((pd->pf_mtag == NULL && 7979 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7980 pd->pf_mtag->routed++ > 3) { 7981 m0 = *m; 7982 *m = NULL; 7983 goto bad_locked; 7984 } 7985 7986 if (r_rt == PF_DUPTO) { 7987 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7988 if (s == NULL) { 7989 ifp = r->rpool.cur->kif ? 7990 r->rpool.cur->kif->pfik_ifp : NULL; 7991 } else { 7992 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7993 /* If pfsync'd */ 7994 if (ifp == NULL && r->rpool.cur != NULL) 7995 ifp = r->rpool.cur->kif ? 7996 r->rpool.cur->kif->pfik_ifp : NULL; 7997 PF_STATE_UNLOCK(s); 7998 } 7999 if (ifp == oifp) { 8000 /* When the 2nd interface is not skipped */ 8001 return; 8002 } else { 8003 m0 = *m; 8004 *m = NULL; 8005 goto bad; 8006 } 8007 } else { 8008 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8009 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8010 if (s) 8011 PF_STATE_UNLOCK(s); 8012 return; 8013 } 8014 } 8015 } else { 8016 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8017 pf_dummynet(pd, s, r, m); 8018 if (s) 8019 PF_STATE_UNLOCK(s); 8020 return; 8021 } 8022 m0 = *m; 8023 } 8024 8025 ip6 = mtod(m0, struct ip6_hdr *); 8026 8027 bzero(&dst, sizeof(dst)); 8028 dst.sin6_family = AF_INET6; 8029 dst.sin6_len = sizeof(dst); 8030 dst.sin6_addr = ip6->ip6_dst; 8031 8032 bzero(&naddr, sizeof(naddr)); 8033 8034 if (s == NULL) { 8035 if (TAILQ_EMPTY(&r->rpool.list)) { 8036 DPFPRINTF(PF_DEBUG_URGENT, 8037 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 8038 goto bad_locked; 8039 } 8040 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 8041 &naddr, &nkif, NULL); 8042 if (!PF_AZERO(&naddr, AF_INET6)) 8043 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8044 &naddr, AF_INET6); 8045 ifp = nkif ? nkif->pfik_ifp : NULL; 8046 } else { 8047 struct pfi_kkif *kif; 8048 8049 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 8050 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8051 &s->rt_addr, AF_INET6); 8052 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8053 kif = s->rt_kif; 8054 /* If pfsync'd */ 8055 if (ifp == NULL && r->rpool.cur != NULL) { 8056 ifp = r->rpool.cur->kif ? 8057 r->rpool.cur->kif->pfik_ifp : NULL; 8058 kif = r->rpool.cur->kif; 8059 } 8060 if (ifp != NULL && kif != NULL && 8061 r->rule_flag & PFRULE_IFBOUND && 8062 r->rt == PF_REPLYTO && 8063 s->kif == V_pfi_all) { 8064 s->kif = kif; 8065 s->orig_kif = oifp->if_pf_kif; 8066 } 8067 } 8068 8069 if (s) 8070 PF_STATE_UNLOCK(s); 8071 8072 if (ifp == NULL) 8073 goto bad; 8074 8075 if (pd->dir == PF_IN) { 8076 if (pf_test(AF_INET6, PF_OUT, PFIL_FWD, ifp, &m0, inp, 8077 &pd->act) != PF_PASS) 8078 goto bad; 8079 else if (m0 == NULL) 8080 goto done; 8081 if (m0->m_len < sizeof(struct ip6_hdr)) { 8082 DPFPRINTF(PF_DEBUG_URGENT, 8083 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8084 __func__)); 8085 goto bad; 8086 } 8087 ip6 = mtod(m0, struct ip6_hdr *); 8088 } 8089 8090 if (ifp->if_flags & IFF_LOOPBACK) 8091 m0->m_flags |= M_SKIP_FIREWALL; 8092 8093 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8094 ~ifp->if_hwassist) { 8095 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8096 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8097 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8098 } 8099 8100 /* 8101 * If the packet is too large for the outgoing interface, 8102 * send back an icmp6 error. 8103 */ 8104 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8105 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8106 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8107 md = m0; 8108 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8109 if (md != NULL) 8110 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8111 } 8112 else { 8113 in6_ifstat_inc(ifp, ifs6_in_toobig); 8114 if (r_rt != PF_DUPTO) { 8115 if (s && s->nat_rule != NULL) 8116 PACKET_UNDO_NAT(m0, pd, 8117 ((caddr_t)ip6 - m0->m_data) + 8118 sizeof(struct ip6_hdr), s); 8119 8120 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8121 } else 8122 goto bad; 8123 } 8124 8125 done: 8126 if (r_rt != PF_DUPTO) 8127 *m = NULL; 8128 return; 8129 8130 bad_locked: 8131 if (s) 8132 PF_STATE_UNLOCK(s); 8133 bad: 8134 m_freem(m0); 8135 goto done; 8136 } 8137 #endif /* INET6 */ 8138 8139 /* 8140 * FreeBSD supports cksum offloads for the following drivers. 8141 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8142 * 8143 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8144 * network driver performed cksum including pseudo header, need to verify 8145 * csum_data 8146 * CSUM_DATA_VALID : 8147 * network driver performed cksum, needs to additional pseudo header 8148 * cksum computation with partial csum_data(i.e. lack of H/W support for 8149 * pseudo header, for instance sk(4) and possibly gem(4)) 8150 * 8151 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8152 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8153 * TCP/UDP layer. 8154 * Also, set csum_data to 0xffff to force cksum validation. 8155 */ 8156 static int 8157 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8158 { 8159 u_int16_t sum = 0; 8160 int hw_assist = 0; 8161 struct ip *ip; 8162 8163 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8164 return (1); 8165 if (m->m_pkthdr.len < off + len) 8166 return (1); 8167 8168 switch (p) { 8169 case IPPROTO_TCP: 8170 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8171 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8172 sum = m->m_pkthdr.csum_data; 8173 } else { 8174 ip = mtod(m, struct ip *); 8175 sum = in_pseudo(ip->ip_src.s_addr, 8176 ip->ip_dst.s_addr, htonl((u_short)len + 8177 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8178 } 8179 sum ^= 0xffff; 8180 ++hw_assist; 8181 } 8182 break; 8183 case IPPROTO_UDP: 8184 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8185 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8186 sum = m->m_pkthdr.csum_data; 8187 } else { 8188 ip = mtod(m, struct ip *); 8189 sum = in_pseudo(ip->ip_src.s_addr, 8190 ip->ip_dst.s_addr, htonl((u_short)len + 8191 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8192 } 8193 sum ^= 0xffff; 8194 ++hw_assist; 8195 } 8196 break; 8197 case IPPROTO_ICMP: 8198 #ifdef INET6 8199 case IPPROTO_ICMPV6: 8200 #endif /* INET6 */ 8201 break; 8202 default: 8203 return (1); 8204 } 8205 8206 if (!hw_assist) { 8207 switch (af) { 8208 case AF_INET: 8209 if (p == IPPROTO_ICMP) { 8210 if (m->m_len < off) 8211 return (1); 8212 m->m_data += off; 8213 m->m_len -= off; 8214 sum = in_cksum(m, len); 8215 m->m_data -= off; 8216 m->m_len += off; 8217 } else { 8218 if (m->m_len < sizeof(struct ip)) 8219 return (1); 8220 sum = in4_cksum(m, p, off, len); 8221 } 8222 break; 8223 #ifdef INET6 8224 case AF_INET6: 8225 if (m->m_len < sizeof(struct ip6_hdr)) 8226 return (1); 8227 sum = in6_cksum(m, p, off, len); 8228 break; 8229 #endif /* INET6 */ 8230 default: 8231 return (1); 8232 } 8233 } 8234 if (sum) { 8235 switch (p) { 8236 case IPPROTO_TCP: 8237 { 8238 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8239 break; 8240 } 8241 case IPPROTO_UDP: 8242 { 8243 KMOD_UDPSTAT_INC(udps_badsum); 8244 break; 8245 } 8246 #ifdef INET 8247 case IPPROTO_ICMP: 8248 { 8249 KMOD_ICMPSTAT_INC(icps_checksum); 8250 break; 8251 } 8252 #endif 8253 #ifdef INET6 8254 case IPPROTO_ICMPV6: 8255 { 8256 KMOD_ICMP6STAT_INC(icp6s_checksum); 8257 break; 8258 } 8259 #endif /* INET6 */ 8260 } 8261 return (1); 8262 } else { 8263 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8264 m->m_pkthdr.csum_flags |= 8265 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8266 m->m_pkthdr.csum_data = 0xffff; 8267 } 8268 } 8269 return (0); 8270 } 8271 8272 static bool 8273 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8274 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8275 { 8276 int dndir = r->direction; 8277 8278 if (s && dndir == PF_INOUT) { 8279 dndir = s->direction; 8280 } else if (dndir == PF_INOUT) { 8281 /* Assume primary direction. Happens when we've set dnpipe in 8282 * the ethernet level code. */ 8283 dndir = pd->dir; 8284 } 8285 8286 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8287 return (false); 8288 8289 memset(dnflow, 0, sizeof(*dnflow)); 8290 8291 if (pd->dport != NULL) 8292 dnflow->f_id.dst_port = ntohs(*pd->dport); 8293 if (pd->sport != NULL) 8294 dnflow->f_id.src_port = ntohs(*pd->sport); 8295 8296 if (pd->dir == PF_IN) 8297 dnflow->flags |= IPFW_ARGS_IN; 8298 else 8299 dnflow->flags |= IPFW_ARGS_OUT; 8300 8301 if (pd->dir != dndir && pd->act.dnrpipe) { 8302 dnflow->rule.info = pd->act.dnrpipe; 8303 } 8304 else if (pd->dir == dndir && pd->act.dnpipe) { 8305 dnflow->rule.info = pd->act.dnpipe; 8306 } 8307 else { 8308 return (false); 8309 } 8310 8311 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8312 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8313 dnflow->rule.info |= IPFW_IS_PIPE; 8314 8315 dnflow->f_id.proto = pd->proto; 8316 dnflow->f_id.extra = dnflow->rule.info; 8317 switch (pd->af) { 8318 case AF_INET: 8319 dnflow->f_id.addr_type = 4; 8320 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8321 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8322 break; 8323 case AF_INET6: 8324 dnflow->flags |= IPFW_ARGS_IP6; 8325 dnflow->f_id.addr_type = 6; 8326 dnflow->f_id.src_ip6 = pd->src->v6; 8327 dnflow->f_id.dst_ip6 = pd->dst->v6; 8328 break; 8329 default: 8330 panic("Invalid AF"); 8331 break; 8332 } 8333 8334 return (true); 8335 } 8336 8337 int 8338 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8339 struct inpcb *inp) 8340 { 8341 struct pfi_kkif *kif; 8342 struct mbuf *m = *m0; 8343 8344 M_ASSERTPKTHDR(m); 8345 MPASS(ifp->if_vnet == curvnet); 8346 NET_EPOCH_ASSERT(); 8347 8348 if (!V_pf_status.running) 8349 return (PF_PASS); 8350 8351 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8352 8353 if (kif == NULL) { 8354 DPFPRINTF(PF_DEBUG_URGENT, 8355 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8356 return (PF_DROP); 8357 } 8358 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8359 return (PF_PASS); 8360 8361 if (m->m_flags & M_SKIP_FIREWALL) 8362 return (PF_PASS); 8363 8364 if (__predict_false(! M_WRITABLE(*m0))) { 8365 m = *m0 = m_unshare(*m0, M_NOWAIT); 8366 if (*m0 == NULL) 8367 return (PF_DROP); 8368 } 8369 8370 /* Stateless! */ 8371 return (pf_test_eth_rule(dir, kif, m0)); 8372 } 8373 8374 static __inline void 8375 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8376 { 8377 struct m_tag *mtag; 8378 8379 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8380 8381 /* dummynet adds this tag, but pf does not need it, 8382 * and keeping it creates unexpected behavior, 8383 * e.g. in case of divert(4) usage right after dummynet. */ 8384 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8385 if (mtag != NULL) 8386 m_tag_delete(m, mtag); 8387 } 8388 8389 static int 8390 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8391 struct pf_krule *r, struct mbuf **m0) 8392 { 8393 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8394 } 8395 8396 static int 8397 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8398 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8399 struct mbuf **m0) 8400 { 8401 NET_EPOCH_ASSERT(); 8402 8403 if (pd->act.dnpipe || pd->act.dnrpipe) { 8404 struct ip_fw_args dnflow; 8405 if (ip_dn_io_ptr == NULL) { 8406 m_freem(*m0); 8407 *m0 = NULL; 8408 return (ENOMEM); 8409 } 8410 8411 if (pd->pf_mtag == NULL && 8412 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8413 m_freem(*m0); 8414 *m0 = NULL; 8415 return (ENOMEM); 8416 } 8417 8418 if (ifp != NULL) { 8419 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8420 8421 pd->pf_mtag->if_index = ifp->if_index; 8422 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8423 8424 MPASS(sa != NULL); 8425 8426 if (pd->af == AF_INET) 8427 memcpy(&pd->pf_mtag->dst, sa, 8428 sizeof(struct sockaddr_in)); 8429 else 8430 memcpy(&pd->pf_mtag->dst, sa, 8431 sizeof(struct sockaddr_in6)); 8432 } 8433 8434 if (s != NULL && s->nat_rule != NULL && 8435 s->nat_rule->action == PF_RDR && 8436 ( 8437 #ifdef INET 8438 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8439 #endif 8440 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8441 /* 8442 * If we're redirecting to loopback mark this packet 8443 * as being local. Otherwise it might get dropped 8444 * if dummynet re-injects. 8445 */ 8446 (*m0)->m_pkthdr.rcvif = V_loif; 8447 } 8448 8449 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8450 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8451 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8452 ip_dn_io_ptr(m0, &dnflow); 8453 if (*m0 != NULL) { 8454 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8455 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8456 } 8457 } 8458 } 8459 8460 return (0); 8461 } 8462 8463 static void 8464 pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m) 8465 { 8466 memset(pd, 0, sizeof(*pd)); 8467 pd->pf_mtag = pf_find_mtag(m); 8468 } 8469 8470 static int 8471 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0, 8472 u_short *action, u_short *reason, struct pfi_kkif *kif, struct pf_krule **a, 8473 struct pf_krule **r, struct pf_kstate **s, struct pf_kruleset **ruleset, 8474 int *off, int *hdrlen, struct inpcb *inp, 8475 struct pf_rule_actions *default_actions) 8476 { 8477 struct mbuf *m = *m0; 8478 8479 pd->af = af; 8480 pd->dir = dir; 8481 8482 TAILQ_INIT(&pd->sctp_multihome_jobs); 8483 if (default_actions != NULL) 8484 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8485 8486 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8487 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8488 pd->act.flags = pd->pf_mtag->dnflags; 8489 } 8490 8491 switch (af) { 8492 #ifdef INET 8493 case AF_INET: { 8494 struct ip *h; 8495 8496 if (__predict_false(m->m_len < sizeof(struct ip)) && 8497 (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8498 DPFPRINTF(PF_DEBUG_URGENT, 8499 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8500 *action = PF_DROP; 8501 REASON_SET(reason, PFRES_SHORT); 8502 return (-1); 8503 } 8504 8505 if (pf_normalize_ip(m0, kif, reason, pd) != PF_PASS) { 8506 /* We do IP header normalization and packet reassembly here */ 8507 *action = PF_DROP; 8508 return (-1); 8509 } 8510 m = *m0; 8511 8512 h = mtod(m, struct ip *); 8513 *off = h->ip_hl << 2; 8514 if (*off < (int)sizeof(*h)) { 8515 *action = PF_DROP; 8516 REASON_SET(reason, PFRES_SHORT); 8517 return (-1); 8518 } 8519 pd->src = (struct pf_addr *)&h->ip_src; 8520 pd->dst = (struct pf_addr *)&h->ip_dst; 8521 pd->sport = pd->dport = NULL; 8522 pd->ip_sum = &h->ip_sum; 8523 pd->proto_sum = NULL; 8524 pd->virtual_proto = pd->proto = h->ip_p; 8525 pd->dir = dir; 8526 pd->sidx = (dir == PF_IN) ? 0 : 1; 8527 pd->didx = (dir == PF_IN) ? 1 : 0; 8528 pd->tos = h->ip_tos; 8529 pd->tot_len = ntohs(h->ip_len); 8530 pd->act.rtableid = -1; 8531 8532 if (h->ip_hl > 5) /* has options */ 8533 pd->badopts++; 8534 8535 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8536 /* 8537 * handle fragments that aren't reassembled by 8538 * normalization 8539 */ 8540 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8541 if (kif == NULL || r == NULL) /* pflog */ 8542 *action = PF_DROP; 8543 else 8544 *action = pf_test_rule(r, s, kif, m, *off, 8545 pd, a, ruleset, inp, *hdrlen); 8546 if (*action != PF_PASS) 8547 REASON_SET(reason, PFRES_FRAG); 8548 return (-1); 8549 } 8550 8551 break; 8552 } 8553 #endif 8554 #ifdef INET6 8555 case AF_INET6: { 8556 struct ip6_hdr *h; 8557 int terminal = 0; 8558 8559 if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) && 8560 (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 8561 DPFPRINTF(PF_DEBUG_URGENT, 8562 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 8563 ", pullup failed\n")); 8564 *action = PF_DROP; 8565 REASON_SET(reason, PFRES_SHORT); 8566 return (-1); 8567 } 8568 8569 /* We do IP header normalization and packet reassembly here */ 8570 if (pf_normalize_ip6(m0, kif, reason, pd) != PF_PASS) { 8571 *action = PF_DROP; 8572 return (-1); 8573 } 8574 m = *m0; 8575 8576 h = mtod(m, struct ip6_hdr *); 8577 pd->src = (struct pf_addr *)&h->ip6_src; 8578 pd->dst = (struct pf_addr *)&h->ip6_dst; 8579 pd->sport = pd->dport = NULL; 8580 pd->ip_sum = NULL; 8581 pd->proto_sum = NULL; 8582 pd->dir = dir; 8583 pd->sidx = (dir == PF_IN) ? 0 : 1; 8584 pd->didx = (dir == PF_IN) ? 1 : 0; 8585 pd->tos = IPV6_DSCP(h); 8586 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8587 *off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8588 pd->virtual_proto = pd->proto = h->ip6_nxt; 8589 pd->act.rtableid = -1; 8590 8591 do { 8592 switch (pd->proto) { 8593 case IPPROTO_FRAGMENT: 8594 pd->virtual_proto = PF_VPROTO_FRAGMENT; 8595 if (kif == NULL || r == NULL) /* pflog */ 8596 *action = PF_DROP; 8597 else 8598 *action = pf_test_rule(r, s, kif, m, *off, 8599 pd, a, ruleset, inp, 8600 *hdrlen); 8601 if (*action == PF_DROP) 8602 REASON_SET(reason, PFRES_FRAG); 8603 return (-1); 8604 case IPPROTO_ROUTING: { 8605 struct ip6_rthdr rthdr; 8606 8607 if (pd->badopts++) { 8608 DPFPRINTF(PF_DEBUG_MISC, 8609 ("pf: IPv6 more than one rthdr")); 8610 *action = PF_DROP; 8611 REASON_SET(reason, PFRES_IPOPTIONS); 8612 return (-1); 8613 } 8614 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), 8615 NULL, reason, pd->af)) { 8616 DPFPRINTF(PF_DEBUG_MISC, 8617 ("pf: IPv6 short rthdr")); 8618 *action = PF_DROP; 8619 REASON_SET(reason, PFRES_SHORT); 8620 return (-1); 8621 } 8622 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8623 DPFPRINTF(PF_DEBUG_MISC, 8624 ("pf: IPv6 rthdr0")); 8625 *action = PF_DROP; 8626 REASON_SET(reason, PFRES_IPOPTIONS); 8627 return (-1); 8628 } 8629 /* FALLTHROUGH */ 8630 } 8631 case IPPROTO_AH: 8632 case IPPROTO_HOPOPTS: 8633 case IPPROTO_DSTOPTS: { 8634 /* get next header and header length */ 8635 struct ip6_ext opt6; 8636 8637 if (!pf_pull_hdr(m, *off, &opt6, sizeof(opt6), 8638 NULL, reason, pd->af)) { 8639 DPFPRINTF(PF_DEBUG_MISC, 8640 ("pf: IPv6 short opt")); 8641 *action = PF_DROP; 8642 return (-1); 8643 } 8644 if (pd->proto == IPPROTO_AH) 8645 *off += (opt6.ip6e_len + 2) * 4; 8646 else 8647 *off += (opt6.ip6e_len + 1) * 8; 8648 pd->virtual_proto = pd->proto = opt6.ip6e_nxt; 8649 /* goto the next header */ 8650 break; 8651 } 8652 default: 8653 terminal++; 8654 break; 8655 } 8656 } while (!terminal); 8657 8658 break; 8659 } 8660 #endif 8661 default: 8662 panic("pf_setup_pdesc called with illegal af %u", af); 8663 } 8664 8665 switch (pd->proto) { 8666 case IPPROTO_TCP: { 8667 struct tcphdr *th = &pd->hdr.tcp; 8668 8669 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8670 reason, af)) { 8671 *action = PF_DROP; 8672 REASON_SET(reason, PFRES_SHORT); 8673 return (-1); 8674 } 8675 *hdrlen = sizeof(*th); 8676 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8677 pd->sport = &th->th_sport; 8678 pd->dport = &th->th_dport; 8679 break; 8680 } 8681 case IPPROTO_UDP: { 8682 struct udphdr *uh = &pd->hdr.udp; 8683 8684 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8685 reason, af)) { 8686 *action = PF_DROP; 8687 REASON_SET(reason, PFRES_SHORT); 8688 return (-1); 8689 } 8690 *hdrlen = sizeof(*uh); 8691 if (uh->uh_dport == 0 || 8692 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8693 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8694 *action = PF_DROP; 8695 REASON_SET(reason, PFRES_SHORT); 8696 return (-1); 8697 } 8698 pd->sport = &uh->uh_sport; 8699 pd->dport = &uh->uh_dport; 8700 break; 8701 } 8702 case IPPROTO_SCTP: { 8703 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8704 action, reason, af)) { 8705 *action = PF_DROP; 8706 REASON_SET(reason, PFRES_SHORT); 8707 return (-1); 8708 } 8709 *hdrlen = sizeof(pd->hdr.sctp); 8710 pd->p_len = pd->tot_len - *off; 8711 8712 pd->sport = &pd->hdr.sctp.src_port; 8713 pd->dport = &pd->hdr.sctp.dest_port; 8714 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8715 *action = PF_DROP; 8716 REASON_SET(reason, PFRES_SHORT); 8717 return (-1); 8718 } 8719 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8720 *action = PF_DROP; 8721 REASON_SET(reason, PFRES_SHORT); 8722 return (-1); 8723 } 8724 break; 8725 } 8726 case IPPROTO_ICMP: { 8727 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp, ICMP_MINLEN, 8728 action, reason, af)) { 8729 *action = PF_DROP; 8730 REASON_SET(reason, PFRES_SHORT); 8731 return (-1); 8732 } 8733 *hdrlen = ICMP_MINLEN; 8734 break; 8735 } 8736 #ifdef INET6 8737 case IPPROTO_ICMPV6: { 8738 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8739 8740 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8741 action, reason, af)) { 8742 *action = PF_DROP; 8743 REASON_SET(reason, PFRES_SHORT); 8744 return (-1); 8745 } 8746 /* ICMP headers we look further into to match state */ 8747 switch (pd->hdr.icmp6.icmp6_type) { 8748 case MLD_LISTENER_QUERY: 8749 case MLD_LISTENER_REPORT: 8750 icmp_hlen = sizeof(struct mld_hdr); 8751 break; 8752 case ND_NEIGHBOR_SOLICIT: 8753 case ND_NEIGHBOR_ADVERT: 8754 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8755 break; 8756 } 8757 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8758 !pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8759 action, reason, af)) { 8760 *action = PF_DROP; 8761 REASON_SET(reason, PFRES_SHORT); 8762 return (-1); 8763 } 8764 *hdrlen = icmp_hlen; 8765 break; 8766 } 8767 #endif 8768 } 8769 return (0); 8770 } 8771 8772 static void 8773 pf_counters_inc(int action, struct pf_pdesc *pd, 8774 struct pfi_kkif *kif, struct pf_kstate *s, 8775 struct pf_krule *r, struct pf_krule *a) 8776 { 8777 struct pf_krule *tr; 8778 int dir = pd->dir; 8779 int dirndx; 8780 8781 pf_counter_u64_critical_enter(); 8782 pf_counter_u64_add_protected( 8783 &kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8784 pd->tot_len); 8785 pf_counter_u64_add_protected( 8786 &kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS], 8787 1); 8788 8789 if (action == PF_PASS || r->action == PF_DROP) { 8790 dirndx = (dir == PF_OUT); 8791 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8792 pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len); 8793 pf_update_timestamp(r); 8794 8795 if (a != NULL) { 8796 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8797 pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len); 8798 } 8799 if (s != NULL) { 8800 struct pf_krule_item *ri; 8801 8802 if (s->nat_rule != NULL) { 8803 pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx], 8804 1); 8805 pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx], 8806 pd->tot_len); 8807 } 8808 if (s->src_node != NULL) { 8809 counter_u64_add(s->src_node->packets[dirndx], 8810 1); 8811 counter_u64_add(s->src_node->bytes[dirndx], 8812 pd->tot_len); 8813 } 8814 if (s->nat_src_node != NULL) { 8815 counter_u64_add(s->nat_src_node->packets[dirndx], 8816 1); 8817 counter_u64_add(s->nat_src_node->bytes[dirndx], 8818 pd->tot_len); 8819 } 8820 dirndx = (dir == s->direction) ? 0 : 1; 8821 s->packets[dirndx]++; 8822 s->bytes[dirndx] += pd->tot_len; 8823 8824 SLIST_FOREACH(ri, &s->match_rules, entry) { 8825 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8826 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len); 8827 } 8828 } 8829 8830 tr = r; 8831 if (s != NULL && s->nat_rule != NULL && 8832 r == &V_pf_default_rule) 8833 tr = s->nat_rule; 8834 8835 if (tr->src.addr.type == PF_ADDR_TABLE) 8836 pfr_update_stats(tr->src.addr.p.tbl, 8837 (s == NULL) ? pd->src : 8838 &s->key[(s->direction == PF_IN)]-> 8839 addr[(s->direction == PF_OUT)], 8840 pd->af, pd->tot_len, dir == PF_OUT, 8841 r->action == PF_PASS, tr->src.neg); 8842 if (tr->dst.addr.type == PF_ADDR_TABLE) 8843 pfr_update_stats(tr->dst.addr.p.tbl, 8844 (s == NULL) ? pd->dst : 8845 &s->key[(s->direction == PF_IN)]-> 8846 addr[(s->direction == PF_IN)], 8847 pd->af, pd->tot_len, dir == PF_OUT, 8848 r->action == PF_PASS, tr->dst.neg); 8849 } 8850 pf_counter_u64_critical_exit(); 8851 } 8852 8853 #if defined(INET) || defined(INET6) 8854 int 8855 pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8856 struct inpcb *inp, struct pf_rule_actions *default_actions) 8857 { 8858 struct pfi_kkif *kif; 8859 u_short action, reason = 0; 8860 struct mbuf *m = *m0; 8861 #ifdef INET 8862 struct ip *h = NULL; 8863 #endif 8864 #ifdef INET6 8865 struct ip6_hdr *h6 = NULL; 8866 #endif 8867 struct m_tag *mtag; 8868 struct pf_krule *a = NULL, *r = &V_pf_default_rule; 8869 struct pf_kstate *s = NULL; 8870 struct pf_kruleset *ruleset = NULL; 8871 struct pf_pdesc pd; 8872 int off, hdrlen, use_2nd_queue = 0; 8873 uint16_t tag; 8874 uint8_t rt; 8875 uint8_t ttl; 8876 8877 PF_RULES_RLOCK_TRACKER; 8878 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8879 M_ASSERTPKTHDR(m); 8880 8881 if (!V_pf_status.running) 8882 return (PF_PASS); 8883 8884 PF_RULES_RLOCK(); 8885 8886 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8887 8888 if (__predict_false(kif == NULL)) { 8889 DPFPRINTF(PF_DEBUG_URGENT, 8890 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8891 PF_RULES_RUNLOCK(); 8892 return (PF_DROP); 8893 } 8894 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8895 PF_RULES_RUNLOCK(); 8896 return (PF_PASS); 8897 } 8898 8899 if (m->m_flags & M_SKIP_FIREWALL) { 8900 PF_RULES_RUNLOCK(); 8901 return (PF_PASS); 8902 } 8903 8904 #ifdef INET6 8905 /* 8906 * If we end up changing IP addresses (e.g. binat) the stack may get 8907 * confused and fail to send the icmp6 packet too big error. Just send 8908 * it here, before we do any NAT. 8909 */ 8910 if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD && 8911 IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 8912 PF_RULES_RUNLOCK(); 8913 *m0 = NULL; 8914 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 8915 return (PF_DROP); 8916 } 8917 #endif 8918 8919 if (__predict_false(! M_WRITABLE(*m0))) { 8920 m = *m0 = m_unshare(*m0, M_NOWAIT); 8921 if (*m0 == NULL) 8922 return (PF_DROP); 8923 } 8924 8925 pf_init_pdesc(&pd, m); 8926 8927 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8928 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8929 8930 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8931 pd.pf_mtag->if_idxgen); 8932 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8933 PF_RULES_RUNLOCK(); 8934 m_freem(*m0); 8935 *m0 = NULL; 8936 return (PF_PASS); 8937 } 8938 PF_RULES_RUNLOCK(); 8939 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 8940 *m0 = NULL; 8941 return (PF_PASS); 8942 } 8943 8944 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8945 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8946 /* Dummynet re-injects packets after they've 8947 * completed their delay. We've already 8948 * processed them, so pass unconditionally. */ 8949 8950 /* But only once. We may see the packet multiple times (e.g. 8951 * PFIL_IN/PFIL_OUT). */ 8952 pf_dummynet_flag_remove(m, pd.pf_mtag); 8953 PF_RULES_RUNLOCK(); 8954 8955 return (PF_PASS); 8956 } 8957 8958 if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason, kif, &a, &r, 8959 &s, &ruleset, &off, &hdrlen, inp, default_actions) == -1) { 8960 if (action != PF_PASS) 8961 pd.act.log |= PF_LOG_FORCE; 8962 goto done; 8963 } 8964 m = *m0; 8965 8966 switch (af) { 8967 #ifdef INET 8968 case AF_INET: 8969 h = mtod(m, struct ip *); 8970 ttl = h->ip_ttl; 8971 break; 8972 #endif 8973 #ifdef INET6 8974 case AF_INET6: 8975 h6 = mtod(m, struct ip6_hdr *); 8976 ttl = h6->ip6_hlim; 8977 break; 8978 #endif 8979 default: 8980 panic("Unknown af %d", af); 8981 } 8982 8983 if (__predict_false(ip_divert_ptr != NULL) && 8984 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 8985 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 8986 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 8987 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 8988 if (pd.pf_mtag == NULL && 8989 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8990 action = PF_DROP; 8991 goto done; 8992 } 8993 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 8994 } 8995 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 8996 m->m_flags |= M_FASTFWD_OURS; 8997 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8998 } 8999 m_tag_delete(m, mtag); 9000 9001 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 9002 if (mtag != NULL) 9003 m_tag_delete(m, mtag); 9004 } 9005 9006 #ifdef INET6 9007 /* 9008 * we do not support jumbogram. if we keep going, zero ip6_plen 9009 * will do something bad, so drop the packet for now. 9010 */ 9011 if (af == AF_INET6 && htons(h6->ip6_plen) == 0) { 9012 action = PF_DROP; 9013 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 9014 goto done; 9015 } 9016 #endif 9017 9018 switch (pd.proto) { 9019 case IPPROTO_TCP: { 9020 /* Respond to SYN with a syncookie. */ 9021 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9022 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9023 pf_syncookie_send(m, off, &pd); 9024 action = PF_DROP; 9025 break; 9026 } 9027 9028 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9029 use_2nd_queue = 1; 9030 action = pf_normalize_tcp(kif, m, 0, off, &pd); 9031 if (action == PF_DROP) 9032 goto done; 9033 action = pf_test_state_tcp(&s, kif, m, off, &pd, &reason); 9034 if (action == PF_PASS) { 9035 if (V_pfsync_update_state_ptr != NULL) 9036 V_pfsync_update_state_ptr(s); 9037 r = s->rule; 9038 a = s->anchor; 9039 } else if (s == NULL) { 9040 /* Validate remote SYN|ACK, re-create original SYN if 9041 * valid. */ 9042 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9043 TH_ACK && pf_syncookie_validate(&pd) && 9044 pd.dir == PF_IN) { 9045 struct mbuf *msyn; 9046 9047 msyn = pf_syncookie_recreate_syn(ttl, off, 9048 &pd); 9049 if (msyn == NULL) { 9050 action = PF_DROP; 9051 break; 9052 } 9053 9054 action = pf_test(af, dir, pflags, ifp, &msyn, inp, 9055 &pd.act); 9056 m_freem(msyn); 9057 if (action != PF_PASS) 9058 break; 9059 9060 action = pf_test_state_tcp(&s, kif, m, off, 9061 &pd, &reason); 9062 if (action != PF_PASS || s == NULL) { 9063 action = PF_DROP; 9064 break; 9065 } 9066 9067 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9068 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9069 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9070 action = pf_synproxy(&pd, &s, &reason); 9071 break; 9072 } else { 9073 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9074 &a, &ruleset, inp, hdrlen); 9075 } 9076 } 9077 break; 9078 } 9079 9080 case IPPROTO_UDP: { 9081 action = pf_test_state_udp(&s, kif, m, off, &pd); 9082 if (action == PF_PASS) { 9083 if (V_pfsync_update_state_ptr != NULL) 9084 V_pfsync_update_state_ptr(s); 9085 r = s->rule; 9086 a = s->anchor; 9087 } else if (s == NULL) 9088 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9089 &a, &ruleset, inp, hdrlen); 9090 break; 9091 } 9092 9093 case IPPROTO_SCTP: { 9094 action = pf_normalize_sctp(dir, kif, m, 0, off, &pd); 9095 if (action == PF_DROP) 9096 goto done; 9097 action = pf_test_state_sctp(&s, kif, m, off, &pd, 9098 &reason); 9099 if (action == PF_PASS) { 9100 if (V_pfsync_update_state_ptr != NULL) 9101 V_pfsync_update_state_ptr(s); 9102 r = s->rule; 9103 a = s->anchor; 9104 } else if (s == NULL) { 9105 action = pf_test_rule(&r, &s, kif, m, off, 9106 &pd, &a, &ruleset, inp, hdrlen); 9107 } 9108 break; 9109 } 9110 9111 case IPPROTO_ICMP: { 9112 if (af != AF_INET) { 9113 action = PF_DROP; 9114 DPFPRINTF(PF_DEBUG_MISC, 9115 ("dropping IPv6 packet with ICMPv4 payload")); 9116 goto done; 9117 } 9118 action = pf_test_state_icmp(&s, kif, m, off, &pd, &reason); 9119 if (action == PF_PASS) { 9120 if (V_pfsync_update_state_ptr != NULL) 9121 V_pfsync_update_state_ptr(s); 9122 r = s->rule; 9123 a = s->anchor; 9124 } else if (s == NULL) 9125 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9126 &a, &ruleset, inp, hdrlen); 9127 break; 9128 } 9129 9130 case IPPROTO_ICMPV6: { 9131 if (af != AF_INET6) { 9132 action = PF_DROP; 9133 DPFPRINTF(PF_DEBUG_MISC, 9134 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9135 goto done; 9136 } 9137 action = pf_test_state_icmp(&s, kif, m, off, &pd, &reason); 9138 if (action == PF_PASS) { 9139 if (V_pfsync_update_state_ptr != NULL) 9140 V_pfsync_update_state_ptr(s); 9141 r = s->rule; 9142 a = s->anchor; 9143 } else if (s == NULL) 9144 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9145 &a, &ruleset, inp, hdrlen); 9146 break; 9147 } 9148 9149 default: 9150 action = pf_test_state_other(&s, kif, m, &pd); 9151 if (action == PF_PASS) { 9152 if (V_pfsync_update_state_ptr != NULL) 9153 V_pfsync_update_state_ptr(s); 9154 r = s->rule; 9155 a = s->anchor; 9156 } else if (s == NULL) 9157 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9158 &a, &ruleset, inp, hdrlen); 9159 break; 9160 } 9161 9162 done: 9163 PF_RULES_RUNLOCK(); 9164 9165 if (action == PF_PASS && pd.badopts && 9166 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9167 action = PF_DROP; 9168 REASON_SET(&reason, PFRES_IPOPTIONS); 9169 pd.act.log = PF_LOG_FORCE; 9170 DPFPRINTF(PF_DEBUG_MISC, 9171 ("pf: dropping packet with dangerous headers\n")); 9172 } 9173 9174 if (s) { 9175 uint8_t log = pd.act.log; 9176 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9177 pd.act.log |= log; 9178 tag = s->tag; 9179 rt = s->rt; 9180 } else { 9181 tag = r->tag; 9182 rt = r->rt; 9183 } 9184 9185 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9186 action = PF_DROP; 9187 REASON_SET(&reason, PFRES_MEMORY); 9188 } 9189 9190 pf_scrub(m, &pd); 9191 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9192 pf_normalize_mss(m, off, &pd); 9193 9194 if (pd.act.rtableid >= 0) 9195 M_SETFIB(m, pd.act.rtableid); 9196 9197 if (pd.act.flags & PFSTATE_SETPRIO) { 9198 if (pd.tos & IPTOS_LOWDELAY) 9199 use_2nd_queue = 1; 9200 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9201 action = PF_DROP; 9202 REASON_SET(&reason, PFRES_MEMORY); 9203 pd.act.log = PF_LOG_FORCE; 9204 DPFPRINTF(PF_DEBUG_MISC, 9205 ("pf: failed to allocate 802.1q mtag\n")); 9206 } 9207 } 9208 9209 #ifdef ALTQ 9210 if (action == PF_PASS && pd.act.qid) { 9211 if (pd.pf_mtag == NULL && 9212 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9213 action = PF_DROP; 9214 REASON_SET(&reason, PFRES_MEMORY); 9215 } else { 9216 if (s != NULL) 9217 pd.pf_mtag->qid_hash = pf_state_hash(s); 9218 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9219 pd.pf_mtag->qid = pd.act.pqid; 9220 else 9221 pd.pf_mtag->qid = pd.act.qid; 9222 /* Add hints for ecn. */ 9223 #ifdef INET 9224 if (af == AF_INET) 9225 pd.pf_mtag->hdr = h; 9226 #endif 9227 #ifdef INET6 9228 if (af == AF_INET6) 9229 pd.pf_mtag->hdr = h6; 9230 #endif 9231 } 9232 } 9233 #endif /* ALTQ */ 9234 9235 /* 9236 * connections redirected to loopback should not match sockets 9237 * bound specifically to loopback due to security implications, 9238 * see tcp_input() and in_pcblookup_listen(). 9239 */ 9240 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9241 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL && 9242 (s->nat_rule->action == PF_RDR || 9243 s->nat_rule->action == PF_BINAT) && 9244 pf_is_loopback(af, pd.dst)) 9245 m->m_flags |= M_SKIP_FIREWALL; 9246 9247 if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) && 9248 action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) { 9249 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9250 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9251 if (mtag != NULL) { 9252 ((struct pf_divert_mtag *)(mtag+1))->port = 9253 ntohs(r->divert.port); 9254 ((struct pf_divert_mtag *)(mtag+1))->idir = 9255 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9256 PF_DIVERT_MTAG_DIR_OUT; 9257 9258 if (s) 9259 PF_STATE_UNLOCK(s); 9260 9261 m_tag_prepend(m, mtag); 9262 if (m->m_flags & M_FASTFWD_OURS) { 9263 if (pd.pf_mtag == NULL && 9264 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9265 action = PF_DROP; 9266 REASON_SET(&reason, PFRES_MEMORY); 9267 pd.act.log = PF_LOG_FORCE; 9268 DPFPRINTF(PF_DEBUG_MISC, 9269 ("pf: failed to allocate tag\n")); 9270 } else { 9271 pd.pf_mtag->flags |= 9272 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9273 m->m_flags &= ~M_FASTFWD_OURS; 9274 } 9275 } 9276 ip_divert_ptr(*m0, dir == PF_IN); 9277 *m0 = NULL; 9278 9279 return (action); 9280 } else { 9281 /* XXX: ipfw has the same behaviour! */ 9282 action = PF_DROP; 9283 REASON_SET(&reason, PFRES_MEMORY); 9284 pd.act.log = PF_LOG_FORCE; 9285 DPFPRINTF(PF_DEBUG_MISC, 9286 ("pf: failed to allocate divert tag\n")); 9287 } 9288 } 9289 /* XXX: Anybody working on it?! */ 9290 if (af == AF_INET6 && r->divert.port) 9291 printf("pf: divert(9) is not supported for IPv6\n"); 9292 9293 /* this flag will need revising if the pkt is forwarded */ 9294 if (pd.pf_mtag) 9295 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9296 9297 if (pd.act.log) { 9298 struct pf_krule *lr; 9299 struct pf_krule_item *ri; 9300 9301 if (s != NULL && s->nat_rule != NULL && 9302 s->nat_rule->log & PF_LOG_ALL) 9303 lr = s->nat_rule; 9304 else 9305 lr = r; 9306 9307 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9308 PFLOG_PACKET(kif, m, action, reason, lr, a, 9309 ruleset, &pd, (s == NULL)); 9310 if (s) { 9311 SLIST_FOREACH(ri, &s->match_rules, entry) 9312 if (ri->r->log & PF_LOG_ALL) 9313 PFLOG_PACKET(kif, m, action, 9314 reason, ri->r, a, ruleset, &pd, 0); 9315 } 9316 } 9317 9318 pf_counters_inc(action, &pd, kif, s, r, a); 9319 9320 switch (action) { 9321 case PF_SYNPROXY_DROP: 9322 m_freem(*m0); 9323 case PF_DEFER: 9324 *m0 = NULL; 9325 action = PF_PASS; 9326 break; 9327 case PF_DROP: 9328 m_freem(*m0); 9329 *m0 = NULL; 9330 break; 9331 default: 9332 if (rt) { 9333 switch (af) { 9334 #ifdef INET 9335 case AF_INET: 9336 /* pf_route() returns unlocked. */ 9337 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9338 break; 9339 #endif 9340 #ifdef INET6 9341 case AF_INET6: 9342 /* pf_route6() returns unlocked. */ 9343 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9344 break; 9345 #endif 9346 default: 9347 panic("Unknown af %d", af); 9348 } 9349 goto out; 9350 } 9351 if (pf_dummynet(&pd, s, r, m0) != 0) { 9352 action = PF_DROP; 9353 REASON_SET(&reason, PFRES_MEMORY); 9354 } 9355 break; 9356 } 9357 9358 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9359 9360 if (s && action != PF_DROP) { 9361 if (!s->if_index_in && dir == PF_IN) 9362 s->if_index_in = ifp->if_index; 9363 else if (!s->if_index_out && dir == PF_OUT) 9364 s->if_index_out = ifp->if_index; 9365 } 9366 9367 if (s) 9368 PF_STATE_UNLOCK(s); 9369 9370 #ifdef INET6 9371 /* If reassembled packet passed, create new fragments. */ 9372 if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT && 9373 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9374 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9375 #endif 9376 9377 out: 9378 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9379 9380 return (action); 9381 } 9382 #endif /* INET || INET6 */ 9383