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_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 132 "struct pf_krule *", "struct mbuf *", "int"); 133 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 134 "struct pf_sctp_source *"); 135 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 136 "struct pf_kstate *", "struct pf_sctp_source *"); 137 138 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 139 "struct mbuf *"); 140 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 141 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 142 "int", "struct pf_keth_rule *", "char *"); 143 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 144 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 145 "int", "struct pf_keth_rule *"); 146 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 147 148 /* 149 * Global variables 150 */ 151 152 /* state tables */ 153 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 154 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 155 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 156 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 157 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 158 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 159 VNET_DEFINE(struct pf_kstatus, pf_status); 160 161 VNET_DEFINE(u_int32_t, ticket_altqs_active); 162 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 163 VNET_DEFINE(int, altqs_inactive_open); 164 VNET_DEFINE(u_int32_t, ticket_pabuf); 165 166 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 167 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 168 VNET_DEFINE(u_char, pf_tcp_secret[16]); 169 #define V_pf_tcp_secret VNET(pf_tcp_secret) 170 VNET_DEFINE(int, pf_tcp_secret_init); 171 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 172 VNET_DEFINE(int, pf_tcp_iss_off); 173 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 174 VNET_DECLARE(int, pf_vnet_active); 175 #define V_pf_vnet_active VNET(pf_vnet_active) 176 177 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 178 #define V_pf_purge_idx VNET(pf_purge_idx) 179 180 #ifdef PF_WANT_32_TO_64_COUNTER 181 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 182 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 183 184 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 185 VNET_DEFINE(size_t, pf_allrulecount); 186 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 187 #endif 188 189 struct pf_sctp_endpoint; 190 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 191 struct pf_sctp_source { 192 sa_family_t af; 193 struct pf_addr addr; 194 TAILQ_ENTRY(pf_sctp_source) entry; 195 }; 196 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 197 struct pf_sctp_endpoint 198 { 199 uint32_t v_tag; 200 struct pf_sctp_sources sources; 201 RB_ENTRY(pf_sctp_endpoint) entry; 202 }; 203 static int 204 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 205 { 206 return (a->v_tag - b->v_tag); 207 } 208 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 209 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 210 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 211 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 212 static struct mtx_padalign pf_sctp_endpoints_mtx; 213 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 214 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 215 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 216 217 /* 218 * Queue for pf_intr() sends. 219 */ 220 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 221 struct pf_send_entry { 222 STAILQ_ENTRY(pf_send_entry) pfse_next; 223 struct mbuf *pfse_m; 224 enum { 225 PFSE_IP, 226 PFSE_IP6, 227 PFSE_ICMP, 228 PFSE_ICMP6, 229 } pfse_type; 230 struct { 231 int type; 232 int code; 233 int mtu; 234 } icmpopts; 235 }; 236 237 STAILQ_HEAD(pf_send_head, pf_send_entry); 238 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 239 #define V_pf_sendqueue VNET(pf_sendqueue) 240 241 static struct mtx_padalign pf_sendqueue_mtx; 242 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 243 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 244 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 245 246 /* 247 * Queue for pf_overload_task() tasks. 248 */ 249 struct pf_overload_entry { 250 SLIST_ENTRY(pf_overload_entry) next; 251 struct pf_addr addr; 252 sa_family_t af; 253 uint8_t dir; 254 struct pf_krule *rule; 255 }; 256 257 SLIST_HEAD(pf_overload_head, pf_overload_entry); 258 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 259 #define V_pf_overloadqueue VNET(pf_overloadqueue) 260 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 261 #define V_pf_overloadtask VNET(pf_overloadtask) 262 263 static struct mtx_padalign pf_overloadqueue_mtx; 264 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 265 "pf overload/flush queue", MTX_DEF); 266 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 267 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 268 269 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 270 struct mtx_padalign pf_unlnkdrules_mtx; 271 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 272 MTX_DEF); 273 274 struct sx pf_config_lock; 275 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 276 277 struct mtx_padalign pf_table_stats_lock; 278 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 279 MTX_DEF); 280 281 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 282 #define V_pf_sources_z VNET(pf_sources_z) 283 uma_zone_t pf_mtag_z; 284 VNET_DEFINE(uma_zone_t, pf_state_z); 285 VNET_DEFINE(uma_zone_t, pf_state_key_z); 286 VNET_DEFINE(uma_zone_t, pf_udp_mapping_z); 287 288 VNET_DEFINE(struct unrhdr64, pf_stateid); 289 290 static void pf_src_tree_remove_state(struct pf_kstate *); 291 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 292 u_int32_t); 293 static void pf_add_threshold(struct pf_threshold *); 294 static int pf_check_threshold(struct pf_threshold *); 295 296 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 297 u_int16_t *, u_int16_t *, struct pf_addr *, 298 u_int16_t, u_int8_t, sa_family_t); 299 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 300 struct tcphdr *, struct pf_state_peer *); 301 int pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *, 302 int *, u_int16_t *, u_int16_t *); 303 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 304 struct pf_addr *, struct pf_addr *, u_int16_t, 305 u_int16_t *, u_int16_t *, u_int16_t *, 306 u_int16_t *, u_int8_t, sa_family_t); 307 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 308 sa_family_t, struct pf_krule *, int); 309 static void pf_detach_state(struct pf_kstate *); 310 static int pf_state_key_attach(struct pf_state_key *, 311 struct pf_state_key *, struct pf_kstate *); 312 static void pf_state_key_detach(struct pf_kstate *, int); 313 static int pf_state_key_ctor(void *, int, void *, int); 314 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 315 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 316 struct pf_mtag *pf_mtag); 317 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 318 struct pf_krule *, struct mbuf **); 319 static int pf_dummynet_route(struct pf_pdesc *, 320 struct pf_kstate *, struct pf_krule *, 321 struct ifnet *, struct sockaddr *, struct mbuf **); 322 static int pf_test_eth_rule(int, struct pfi_kkif *, 323 struct mbuf **); 324 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 325 struct pfi_kkif *, struct mbuf *, int, 326 struct pf_pdesc *, struct pf_krule **, 327 struct pf_kruleset **, struct inpcb *, int); 328 static int pf_create_state(struct pf_krule *, struct pf_krule *, 329 struct pf_krule *, struct pf_pdesc *, 330 struct pf_ksrc_node *, struct pf_state_key *, 331 struct pf_state_key *, struct mbuf *, int, 332 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 333 struct pf_kstate **, int, u_int16_t, u_int16_t, 334 int, struct pf_krule_slist *, struct pf_udp_mapping *); 335 static int pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *, 336 int, struct pf_state_key_cmp *, int, struct pf_addr *, 337 int, struct pf_addr *, int); 338 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 339 struct mbuf *, void *, struct pf_pdesc *, 340 struct pf_krule **, struct pf_kruleset **); 341 static int pf_tcp_track_full(struct pf_kstate **, 342 struct pfi_kkif *, struct mbuf *, int, 343 struct pf_pdesc *, u_short *, int *); 344 static int pf_tcp_track_sloppy(struct pf_kstate **, 345 struct pf_pdesc *, u_short *); 346 static int pf_test_state_tcp(struct pf_kstate **, 347 struct pfi_kkif *, struct mbuf *, int, 348 void *, struct pf_pdesc *, u_short *); 349 static int pf_test_state_udp(struct pf_kstate **, 350 struct pfi_kkif *, struct mbuf *, int, 351 void *, struct pf_pdesc *); 352 int pf_icmp_state_lookup(struct pf_state_key_cmp *, 353 struct pf_pdesc *, struct pf_kstate **, struct mbuf *, 354 int, int, struct pfi_kkif *, u_int16_t, u_int16_t, 355 int, int *, int, int); 356 static int pf_test_state_icmp(struct pf_kstate **, 357 struct pfi_kkif *, struct mbuf *, int, 358 void *, struct pf_pdesc *, u_short *); 359 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 360 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 361 struct pfi_kkif *, struct pf_kstate *, int); 362 static int pf_test_state_sctp(struct pf_kstate **, 363 struct pfi_kkif *, struct mbuf *, int, 364 void *, struct pf_pdesc *, u_short *); 365 static int pf_test_state_other(struct pf_kstate **, 366 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 367 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 368 int, u_int16_t); 369 static int pf_check_proto_cksum(struct mbuf *, int, int, 370 u_int8_t, sa_family_t); 371 static void pf_print_state_parts(struct pf_kstate *, 372 struct pf_state_key *, struct pf_state_key *); 373 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 374 bool, u_int8_t); 375 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 376 const struct pf_state_key_cmp *, u_int); 377 static int pf_src_connlimit(struct pf_kstate **); 378 static int pf_match_rcvif(struct mbuf *, struct pf_krule *); 379 static void pf_overload_task(void *v, int pending); 380 static u_short pf_insert_src_node(struct pf_ksrc_node **, 381 struct pf_krule *, struct pf_addr *, sa_family_t); 382 static u_int pf_purge_expired_states(u_int, int); 383 static void pf_purge_unlinked_rules(void); 384 static int pf_mtag_uminit(void *, int, int); 385 static void pf_mtag_free(struct m_tag *); 386 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 387 int, struct pf_state_key *); 388 #ifdef INET 389 static void pf_route(struct mbuf **, struct pf_krule *, 390 struct ifnet *, struct pf_kstate *, 391 struct pf_pdesc *, struct inpcb *); 392 #endif /* INET */ 393 #ifdef INET6 394 static void pf_change_a6(struct pf_addr *, u_int16_t *, 395 struct pf_addr *, u_int8_t); 396 static void pf_route6(struct mbuf **, struct pf_krule *, 397 struct ifnet *, struct pf_kstate *, 398 struct pf_pdesc *, struct inpcb *); 399 #endif /* INET6 */ 400 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 401 402 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 403 404 extern int pf_end_threads; 405 extern struct proc *pf_purge_proc; 406 407 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 408 409 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK }; 410 411 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 412 do { \ 413 struct pf_state_key *nk; \ 414 if ((pd->dir) == PF_OUT) \ 415 nk = (_s)->key[PF_SK_STACK]; \ 416 else \ 417 nk = (_s)->key[PF_SK_WIRE]; \ 418 pf_packet_rework_nat(_m, _pd, _off, nk); \ 419 } while (0) 420 421 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 422 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 423 424 #define STATE_LOOKUP(i, k, s, pd) \ 425 do { \ 426 (s) = pf_find_state((i), (k), (pd->dir)); \ 427 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 428 if ((s) == NULL) \ 429 return (PF_DROP); \ 430 if (PACKET_LOOPED(pd)) \ 431 return (PF_PASS); \ 432 } while (0) 433 434 static struct pfi_kkif * 435 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 436 { 437 SDT_PROBE2(pf, ip, , bound_iface, st, k); 438 439 /* Floating unless otherwise specified. */ 440 if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND)) 441 return (V_pfi_all); 442 443 /* 444 * Initially set to all, because we don't know what interface we'll be 445 * sending this out when we create the state. 446 */ 447 if (st->rule.ptr->rt == PF_REPLYTO) 448 return (V_pfi_all); 449 450 /* Don't overrule the interface for states created on incoming packets. */ 451 if (st->direction == PF_IN) 452 return (k); 453 454 /* No route-to, so don't overrule. */ 455 if (st->rt != PF_ROUTETO) 456 return (k); 457 458 /* Bind to the route-to interface. */ 459 return (st->rt_kif); 460 } 461 462 #define STATE_INC_COUNTERS(s) \ 463 do { \ 464 struct pf_krule_item *mrm; \ 465 counter_u64_add(s->rule.ptr->states_cur, 1); \ 466 counter_u64_add(s->rule.ptr->states_tot, 1); \ 467 if (s->anchor.ptr != NULL) { \ 468 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 469 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 470 } \ 471 if (s->nat_rule.ptr != NULL) { \ 472 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 473 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 474 } \ 475 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 476 counter_u64_add(mrm->r->states_cur, 1); \ 477 counter_u64_add(mrm->r->states_tot, 1); \ 478 } \ 479 } while (0) 480 481 #define STATE_DEC_COUNTERS(s) \ 482 do { \ 483 struct pf_krule_item *mrm; \ 484 if (s->nat_rule.ptr != NULL) \ 485 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 486 if (s->anchor.ptr != NULL) \ 487 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 488 counter_u64_add(s->rule.ptr->states_cur, -1); \ 489 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 490 counter_u64_add(mrm->r->states_cur, -1); \ 491 } while (0) 492 493 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 494 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 495 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 496 VNET_DEFINE(struct pf_idhash *, pf_idhash); 497 VNET_DEFINE(struct pf_srchash *, pf_srchash); 498 VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash); 499 VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping); 500 501 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 502 "pf(4)"); 503 504 VNET_DEFINE(u_long, pf_hashmask); 505 VNET_DEFINE(u_long, pf_srchashmask); 506 VNET_DEFINE(u_long, pf_udpendpointhashmask); 507 VNET_DEFINE_STATIC(u_long, pf_hashsize); 508 #define V_pf_hashsize VNET(pf_hashsize) 509 VNET_DEFINE_STATIC(u_long, pf_srchashsize); 510 #define V_pf_srchashsize VNET(pf_srchashsize) 511 VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize); 512 #define V_pf_udpendpointhashsize VNET(pf_udpendpointhashsize) 513 u_long pf_ioctl_maxcount = 65535; 514 515 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 516 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 517 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 518 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 519 SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, 520 &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable"); 521 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 522 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 523 524 VNET_DEFINE(void *, pf_swi_cookie); 525 VNET_DEFINE(struct intr_event *, pf_swi_ie); 526 527 VNET_DEFINE(uint32_t, pf_hashseed); 528 #define V_pf_hashseed VNET(pf_hashseed) 529 530 static void 531 pf_sctp_checksum(struct mbuf *m, int off) 532 { 533 uint32_t sum = 0; 534 535 /* Zero out the checksum, to enable recalculation. */ 536 m_copyback(m, off + offsetof(struct sctphdr, checksum), 537 sizeof(sum), (caddr_t)&sum); 538 539 sum = sctp_calculate_cksum(m, off); 540 541 m_copyback(m, off + offsetof(struct sctphdr, checksum), 542 sizeof(sum), (caddr_t)&sum); 543 } 544 545 int 546 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 547 { 548 549 switch (af) { 550 #ifdef INET 551 case AF_INET: 552 if (a->addr32[0] > b->addr32[0]) 553 return (1); 554 if (a->addr32[0] < b->addr32[0]) 555 return (-1); 556 break; 557 #endif /* INET */ 558 #ifdef INET6 559 case AF_INET6: 560 if (a->addr32[3] > b->addr32[3]) 561 return (1); 562 if (a->addr32[3] < b->addr32[3]) 563 return (-1); 564 if (a->addr32[2] > b->addr32[2]) 565 return (1); 566 if (a->addr32[2] < b->addr32[2]) 567 return (-1); 568 if (a->addr32[1] > b->addr32[1]) 569 return (1); 570 if (a->addr32[1] < b->addr32[1]) 571 return (-1); 572 if (a->addr32[0] > b->addr32[0]) 573 return (1); 574 if (a->addr32[0] < b->addr32[0]) 575 return (-1); 576 break; 577 #endif /* INET6 */ 578 default: 579 panic("%s: unknown address family %u", __func__, af); 580 } 581 return (0); 582 } 583 584 static void 585 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 586 struct pf_state_key *nk) 587 { 588 589 switch (pd->proto) { 590 case IPPROTO_TCP: { 591 struct tcphdr *th = &pd->hdr.tcp; 592 593 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 594 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 595 &th->th_sum, &nk->addr[pd->sidx], 596 nk->port[pd->sidx], 0, pd->af); 597 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 598 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 599 &th->th_sum, &nk->addr[pd->didx], 600 nk->port[pd->didx], 0, pd->af); 601 m_copyback(m, off, sizeof(*th), (caddr_t)th); 602 break; 603 } 604 case IPPROTO_UDP: { 605 struct udphdr *uh = &pd->hdr.udp; 606 607 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 608 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 609 &uh->uh_sum, &nk->addr[pd->sidx], 610 nk->port[pd->sidx], 1, pd->af); 611 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 612 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 613 &uh->uh_sum, &nk->addr[pd->didx], 614 nk->port[pd->didx], 1, pd->af); 615 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 616 break; 617 } 618 case IPPROTO_SCTP: { 619 struct sctphdr *sh = &pd->hdr.sctp; 620 uint16_t checksum = 0; 621 622 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 623 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 624 &checksum, &nk->addr[pd->sidx], 625 nk->port[pd->sidx], 1, pd->af); 626 } 627 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 628 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 629 &checksum, &nk->addr[pd->didx], 630 nk->port[pd->didx], 1, pd->af); 631 } 632 633 break; 634 } 635 case IPPROTO_ICMP: { 636 struct icmp *ih = &pd->hdr.icmp; 637 638 if (nk->port[pd->sidx] != ih->icmp_id) { 639 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 640 ih->icmp_cksum, ih->icmp_id, 641 nk->port[pd->sidx], 0); 642 ih->icmp_id = nk->port[pd->sidx]; 643 pd->sport = &ih->icmp_id; 644 645 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 646 } 647 /* FALLTHROUGH */ 648 } 649 default: 650 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 651 switch (pd->af) { 652 case AF_INET: 653 pf_change_a(&pd->src->v4.s_addr, 654 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 655 0); 656 break; 657 case AF_INET6: 658 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 659 break; 660 } 661 } 662 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 663 switch (pd->af) { 664 case AF_INET: 665 pf_change_a(&pd->dst->v4.s_addr, 666 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 667 0); 668 break; 669 case AF_INET6: 670 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 671 break; 672 } 673 } 674 break; 675 } 676 } 677 678 static __inline uint32_t 679 pf_hashkey(const struct pf_state_key *sk) 680 { 681 uint32_t h; 682 683 h = murmur3_32_hash32((const uint32_t *)sk, 684 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 685 V_pf_hashseed); 686 687 return (h & V_pf_hashmask); 688 } 689 690 static __inline uint32_t 691 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 692 { 693 uint32_t h; 694 695 switch (af) { 696 case AF_INET: 697 h = murmur3_32_hash32((uint32_t *)&addr->v4, 698 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 699 break; 700 case AF_INET6: 701 h = murmur3_32_hash32((uint32_t *)&addr->v6, 702 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 703 break; 704 default: 705 panic("%s: unknown address family %u", __func__, af); 706 } 707 708 return (h & V_pf_srchashmask); 709 } 710 711 static inline uint32_t 712 pf_hashudpendpoint(struct pf_udp_endpoint *endpoint) 713 { 714 uint32_t h; 715 716 h = murmur3_32_hash32((uint32_t *)endpoint, 717 sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t), 718 V_pf_hashseed); 719 return (h & V_pf_udpendpointhashmask); 720 } 721 722 #ifdef ALTQ 723 static int 724 pf_state_hash(struct pf_kstate *s) 725 { 726 u_int32_t hv = (intptr_t)s / sizeof(*s); 727 728 hv ^= crc32(&s->src, sizeof(s->src)); 729 hv ^= crc32(&s->dst, sizeof(s->dst)); 730 if (hv == 0) 731 hv = 1; 732 return (hv); 733 } 734 #endif 735 736 static __inline void 737 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 738 { 739 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 740 s->dst.state = newstate; 741 if (which == PF_PEER_DST) 742 return; 743 if (s->src.state == newstate) 744 return; 745 if (s->creatorid == V_pf_status.hostid && 746 s->key[PF_SK_STACK] != NULL && 747 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 748 !(TCPS_HAVEESTABLISHED(s->src.state) || 749 s->src.state == TCPS_CLOSED) && 750 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 751 atomic_add_32(&V_pf_status.states_halfopen, -1); 752 753 s->src.state = newstate; 754 } 755 756 #ifdef INET6 757 void 758 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 759 { 760 switch (af) { 761 #ifdef INET 762 case AF_INET: 763 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 764 break; 765 #endif /* INET */ 766 case AF_INET6: 767 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 768 break; 769 } 770 } 771 #endif /* INET6 */ 772 773 static void 774 pf_init_threshold(struct pf_threshold *threshold, 775 u_int32_t limit, u_int32_t seconds) 776 { 777 threshold->limit = limit * PF_THRESHOLD_MULT; 778 threshold->seconds = seconds; 779 threshold->count = 0; 780 threshold->last = time_uptime; 781 } 782 783 static void 784 pf_add_threshold(struct pf_threshold *threshold) 785 { 786 u_int32_t t = time_uptime, diff = t - threshold->last; 787 788 if (diff >= threshold->seconds) 789 threshold->count = 0; 790 else 791 threshold->count -= threshold->count * diff / 792 threshold->seconds; 793 threshold->count += PF_THRESHOLD_MULT; 794 threshold->last = t; 795 } 796 797 static int 798 pf_check_threshold(struct pf_threshold *threshold) 799 { 800 return (threshold->count > threshold->limit); 801 } 802 803 static int 804 pf_src_connlimit(struct pf_kstate **state) 805 { 806 struct pf_overload_entry *pfoe; 807 int bad = 0; 808 809 PF_STATE_LOCK_ASSERT(*state); 810 /* 811 * XXXKS: The src node is accessed unlocked! 812 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 813 */ 814 815 (*state)->src_node->conn++; 816 (*state)->src.tcp_est = 1; 817 pf_add_threshold(&(*state)->src_node->conn_rate); 818 819 if ((*state)->rule.ptr->max_src_conn && 820 (*state)->rule.ptr->max_src_conn < 821 (*state)->src_node->conn) { 822 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 823 bad++; 824 } 825 826 if ((*state)->rule.ptr->max_src_conn_rate.limit && 827 pf_check_threshold(&(*state)->src_node->conn_rate)) { 828 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 829 bad++; 830 } 831 832 if (!bad) 833 return (0); 834 835 /* Kill this state. */ 836 (*state)->timeout = PFTM_PURGE; 837 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 838 839 if ((*state)->rule.ptr->overload_tbl == NULL) 840 return (1); 841 842 /* Schedule overloading and flushing task. */ 843 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 844 if (pfoe == NULL) 845 return (1); /* too bad :( */ 846 847 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 848 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 849 pfoe->rule = (*state)->rule.ptr; 850 pfoe->dir = (*state)->direction; 851 PF_OVERLOADQ_LOCK(); 852 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 853 PF_OVERLOADQ_UNLOCK(); 854 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 855 856 return (1); 857 } 858 859 static void 860 pf_overload_task(void *v, int pending) 861 { 862 struct pf_overload_head queue; 863 struct pfr_addr p; 864 struct pf_overload_entry *pfoe, *pfoe1; 865 uint32_t killed = 0; 866 867 CURVNET_SET((struct vnet *)v); 868 869 PF_OVERLOADQ_LOCK(); 870 queue = V_pf_overloadqueue; 871 SLIST_INIT(&V_pf_overloadqueue); 872 PF_OVERLOADQ_UNLOCK(); 873 874 bzero(&p, sizeof(p)); 875 SLIST_FOREACH(pfoe, &queue, next) { 876 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 877 if (V_pf_status.debug >= PF_DEBUG_MISC) { 878 printf("%s: blocking address ", __func__); 879 pf_print_host(&pfoe->addr, 0, pfoe->af); 880 printf("\n"); 881 } 882 883 p.pfra_af = pfoe->af; 884 switch (pfoe->af) { 885 #ifdef INET 886 case AF_INET: 887 p.pfra_net = 32; 888 p.pfra_ip4addr = pfoe->addr.v4; 889 break; 890 #endif 891 #ifdef INET6 892 case AF_INET6: 893 p.pfra_net = 128; 894 p.pfra_ip6addr = pfoe->addr.v6; 895 break; 896 #endif 897 } 898 899 PF_RULES_WLOCK(); 900 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 901 PF_RULES_WUNLOCK(); 902 } 903 904 /* 905 * Remove those entries, that don't need flushing. 906 */ 907 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 908 if (pfoe->rule->flush == 0) { 909 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 910 free(pfoe, M_PFTEMP); 911 } else 912 counter_u64_add( 913 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 914 915 /* If nothing to flush, return. */ 916 if (SLIST_EMPTY(&queue)) { 917 CURVNET_RESTORE(); 918 return; 919 } 920 921 for (int i = 0; i <= V_pf_hashmask; i++) { 922 struct pf_idhash *ih = &V_pf_idhash[i]; 923 struct pf_state_key *sk; 924 struct pf_kstate *s; 925 926 PF_HASHROW_LOCK(ih); 927 LIST_FOREACH(s, &ih->states, entry) { 928 sk = s->key[PF_SK_WIRE]; 929 SLIST_FOREACH(pfoe, &queue, next) 930 if (sk->af == pfoe->af && 931 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 932 pfoe->rule == s->rule.ptr) && 933 ((pfoe->dir == PF_OUT && 934 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 935 (pfoe->dir == PF_IN && 936 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 937 s->timeout = PFTM_PURGE; 938 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 939 killed++; 940 } 941 } 942 PF_HASHROW_UNLOCK(ih); 943 } 944 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 945 free(pfoe, M_PFTEMP); 946 if (V_pf_status.debug >= PF_DEBUG_MISC) 947 printf("%s: %u states killed", __func__, killed); 948 949 CURVNET_RESTORE(); 950 } 951 952 /* 953 * Can return locked on failure, so that we can consistently 954 * allocate and insert a new one. 955 */ 956 struct pf_ksrc_node * 957 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 958 struct pf_srchash **sh, bool returnlocked) 959 { 960 struct pf_ksrc_node *n; 961 962 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 963 964 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 965 PF_HASHROW_LOCK(*sh); 966 LIST_FOREACH(n, &(*sh)->nodes, entry) 967 if (n->rule.ptr == rule && n->af == af && 968 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 969 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 970 break; 971 972 if (n != NULL) { 973 n->states++; 974 PF_HASHROW_UNLOCK(*sh); 975 } else if (returnlocked == false) 976 PF_HASHROW_UNLOCK(*sh); 977 978 return (n); 979 } 980 981 static void 982 pf_free_src_node(struct pf_ksrc_node *sn) 983 { 984 985 for (int i = 0; i < 2; i++) { 986 counter_u64_free(sn->bytes[i]); 987 counter_u64_free(sn->packets[i]); 988 } 989 uma_zfree(V_pf_sources_z, sn); 990 } 991 992 static u_short 993 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 994 struct pf_addr *src, sa_family_t af) 995 { 996 u_short reason = 0; 997 struct pf_srchash *sh = NULL; 998 999 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 1000 rule->rpool.opts & PF_POOL_STICKYADDR), 1001 ("%s for non-tracking rule %p", __func__, rule)); 1002 1003 if (*sn == NULL) 1004 *sn = pf_find_src_node(src, rule, af, &sh, true); 1005 1006 if (*sn == NULL) { 1007 PF_HASHROW_ASSERT(sh); 1008 1009 if (rule->max_src_nodes && 1010 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 1011 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 1012 PF_HASHROW_UNLOCK(sh); 1013 reason = PFRES_SRCLIMIT; 1014 goto done; 1015 } 1016 1017 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 1018 if ((*sn) == NULL) { 1019 PF_HASHROW_UNLOCK(sh); 1020 reason = PFRES_MEMORY; 1021 goto done; 1022 } 1023 1024 for (int i = 0; i < 2; i++) { 1025 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 1026 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 1027 1028 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 1029 pf_free_src_node(*sn); 1030 PF_HASHROW_UNLOCK(sh); 1031 reason = PFRES_MEMORY; 1032 goto done; 1033 } 1034 } 1035 1036 pf_init_threshold(&(*sn)->conn_rate, 1037 rule->max_src_conn_rate.limit, 1038 rule->max_src_conn_rate.seconds); 1039 1040 MPASS((*sn)->lock == NULL); 1041 (*sn)->lock = &sh->lock; 1042 1043 (*sn)->af = af; 1044 (*sn)->rule.ptr = rule; 1045 PF_ACPY(&(*sn)->addr, src, af); 1046 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1047 (*sn)->creation = time_uptime; 1048 (*sn)->ruletype = rule->action; 1049 (*sn)->states = 1; 1050 if ((*sn)->rule.ptr != NULL) 1051 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 1052 PF_HASHROW_UNLOCK(sh); 1053 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1054 } else { 1055 if (rule->max_src_states && 1056 (*sn)->states >= rule->max_src_states) { 1057 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1058 1); 1059 reason = PFRES_SRCLIMIT; 1060 goto done; 1061 } 1062 } 1063 done: 1064 return (reason); 1065 } 1066 1067 void 1068 pf_unlink_src_node(struct pf_ksrc_node *src) 1069 { 1070 PF_SRC_NODE_LOCK_ASSERT(src); 1071 1072 LIST_REMOVE(src, entry); 1073 if (src->rule.ptr) 1074 counter_u64_add(src->rule.ptr->src_nodes, -1); 1075 } 1076 1077 u_int 1078 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1079 { 1080 struct pf_ksrc_node *sn, *tmp; 1081 u_int count = 0; 1082 1083 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1084 pf_free_src_node(sn); 1085 count++; 1086 } 1087 1088 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1089 1090 return (count); 1091 } 1092 1093 void 1094 pf_mtag_initialize(void) 1095 { 1096 1097 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1098 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1099 UMA_ALIGN_PTR, 0); 1100 } 1101 1102 /* Per-vnet data storage structures initialization. */ 1103 void 1104 pf_initialize(void) 1105 { 1106 struct pf_keyhash *kh; 1107 struct pf_idhash *ih; 1108 struct pf_srchash *sh; 1109 struct pf_udpendpointhash *uh; 1110 u_int i; 1111 1112 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 1113 V_pf_hashsize = PF_HASHSIZ; 1114 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 1115 V_pf_srchashsize = PF_SRCHASHSIZ; 1116 if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize)) 1117 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1118 1119 V_pf_hashseed = arc4random(); 1120 1121 /* States and state keys storage. */ 1122 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1123 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1124 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1125 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1126 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1127 1128 V_pf_state_key_z = uma_zcreate("pf state keys", 1129 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1130 UMA_ALIGN_PTR, 0); 1131 1132 V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), 1133 M_PFHASH, M_NOWAIT | M_ZERO); 1134 V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), 1135 M_PFHASH, M_NOWAIT | M_ZERO); 1136 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1137 printf("pf: Unable to allocate memory for " 1138 "state_hashsize %lu.\n", V_pf_hashsize); 1139 1140 free(V_pf_keyhash, M_PFHASH); 1141 free(V_pf_idhash, M_PFHASH); 1142 1143 V_pf_hashsize = PF_HASHSIZ; 1144 V_pf_keyhash = mallocarray(V_pf_hashsize, 1145 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1146 V_pf_idhash = mallocarray(V_pf_hashsize, 1147 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1148 } 1149 1150 V_pf_hashmask = V_pf_hashsize - 1; 1151 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 1152 i++, kh++, ih++) { 1153 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1154 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1155 } 1156 1157 /* Source nodes. */ 1158 V_pf_sources_z = uma_zcreate("pf source nodes", 1159 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1160 0); 1161 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1162 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1163 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1164 1165 V_pf_srchash = mallocarray(V_pf_srchashsize, 1166 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1167 if (V_pf_srchash == NULL) { 1168 printf("pf: Unable to allocate memory for " 1169 "source_hashsize %lu.\n", V_pf_srchashsize); 1170 1171 V_pf_srchashsize = PF_SRCHASHSIZ; 1172 V_pf_srchash = mallocarray(V_pf_srchashsize, 1173 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1174 } 1175 1176 V_pf_srchashmask = V_pf_srchashsize - 1; 1177 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 1178 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1179 1180 1181 /* UDP endpoint mappings. */ 1182 V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings", 1183 sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL, 1184 UMA_ALIGN_PTR, 0); 1185 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1186 sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO); 1187 if (V_pf_udpendpointhash == NULL) { 1188 printf("pf: Unable to allocate memory for " 1189 "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize); 1190 1191 V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ; 1192 V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize, 1193 sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO); 1194 } 1195 1196 V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1; 1197 for (i = 0, uh = V_pf_udpendpointhash; 1198 i <= V_pf_udpendpointhashmask; 1199 i++, uh++) { 1200 mtx_init(&uh->lock, "pf_udpendpointhash", NULL, 1201 MTX_DEF | MTX_DUPOK); 1202 } 1203 1204 /* ALTQ */ 1205 TAILQ_INIT(&V_pf_altqs[0]); 1206 TAILQ_INIT(&V_pf_altqs[1]); 1207 TAILQ_INIT(&V_pf_altqs[2]); 1208 TAILQ_INIT(&V_pf_altqs[3]); 1209 TAILQ_INIT(&V_pf_pabuf); 1210 V_pf_altqs_active = &V_pf_altqs[0]; 1211 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1212 V_pf_altqs_inactive = &V_pf_altqs[2]; 1213 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1214 1215 /* Send & overload+flush queues. */ 1216 STAILQ_INIT(&V_pf_sendqueue); 1217 SLIST_INIT(&V_pf_overloadqueue); 1218 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1219 1220 /* Unlinked, but may be referenced rules. */ 1221 TAILQ_INIT(&V_pf_unlinked_rules); 1222 } 1223 1224 void 1225 pf_mtag_cleanup(void) 1226 { 1227 1228 uma_zdestroy(pf_mtag_z); 1229 } 1230 1231 void 1232 pf_cleanup(void) 1233 { 1234 struct pf_keyhash *kh; 1235 struct pf_idhash *ih; 1236 struct pf_srchash *sh; 1237 struct pf_udpendpointhash *uh; 1238 struct pf_send_entry *pfse, *next; 1239 u_int i; 1240 1241 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; 1242 i <= V_pf_hashmask; 1243 i++, kh++, ih++) { 1244 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1245 __func__)); 1246 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1247 __func__)); 1248 mtx_destroy(&kh->lock); 1249 mtx_destroy(&ih->lock); 1250 } 1251 free(V_pf_keyhash, M_PFHASH); 1252 free(V_pf_idhash, M_PFHASH); 1253 1254 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1255 KASSERT(LIST_EMPTY(&sh->nodes), 1256 ("%s: source node hash not empty", __func__)); 1257 mtx_destroy(&sh->lock); 1258 } 1259 free(V_pf_srchash, M_PFHASH); 1260 1261 for (i = 0, uh = V_pf_udpendpointhash; 1262 i <= V_pf_udpendpointhashmask; 1263 i++, uh++) { 1264 KASSERT(LIST_EMPTY(&uh->endpoints), 1265 ("%s: udp endpoint hash not empty", __func__)); 1266 mtx_destroy(&uh->lock); 1267 } 1268 free(V_pf_udpendpointhash, M_PFHASH); 1269 1270 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1271 m_freem(pfse->pfse_m); 1272 free(pfse, M_PFTEMP); 1273 } 1274 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1275 1276 uma_zdestroy(V_pf_sources_z); 1277 uma_zdestroy(V_pf_state_z); 1278 uma_zdestroy(V_pf_state_key_z); 1279 uma_zdestroy(V_pf_udp_mapping_z); 1280 } 1281 1282 static int 1283 pf_mtag_uminit(void *mem, int size, int how) 1284 { 1285 struct m_tag *t; 1286 1287 t = (struct m_tag *)mem; 1288 t->m_tag_cookie = MTAG_ABI_COMPAT; 1289 t->m_tag_id = PACKET_TAG_PF; 1290 t->m_tag_len = sizeof(struct pf_mtag); 1291 t->m_tag_free = pf_mtag_free; 1292 1293 return (0); 1294 } 1295 1296 static void 1297 pf_mtag_free(struct m_tag *t) 1298 { 1299 1300 uma_zfree(pf_mtag_z, t); 1301 } 1302 1303 struct pf_mtag * 1304 pf_get_mtag(struct mbuf *m) 1305 { 1306 struct m_tag *mtag; 1307 1308 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1309 return ((struct pf_mtag *)(mtag + 1)); 1310 1311 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1312 if (mtag == NULL) 1313 return (NULL); 1314 bzero(mtag + 1, sizeof(struct pf_mtag)); 1315 m_tag_prepend(m, mtag); 1316 1317 return ((struct pf_mtag *)(mtag + 1)); 1318 } 1319 1320 static int 1321 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1322 struct pf_kstate *s) 1323 { 1324 struct pf_keyhash *khs, *khw, *kh; 1325 struct pf_state_key *sk, *cur; 1326 struct pf_kstate *si, *olds = NULL; 1327 int idx; 1328 1329 NET_EPOCH_ASSERT(); 1330 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1331 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1332 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1333 1334 /* 1335 * We need to lock hash slots of both keys. To avoid deadlock 1336 * we always lock the slot with lower address first. Unlock order 1337 * isn't important. 1338 * 1339 * We also need to lock ID hash slot before dropping key 1340 * locks. On success we return with ID hash slot locked. 1341 */ 1342 1343 if (skw == sks) { 1344 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1345 PF_HASHROW_LOCK(khs); 1346 } else { 1347 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1348 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1349 if (khs == khw) { 1350 PF_HASHROW_LOCK(khs); 1351 } else if (khs < khw) { 1352 PF_HASHROW_LOCK(khs); 1353 PF_HASHROW_LOCK(khw); 1354 } else { 1355 PF_HASHROW_LOCK(khw); 1356 PF_HASHROW_LOCK(khs); 1357 } 1358 } 1359 1360 #define KEYS_UNLOCK() do { \ 1361 if (khs != khw) { \ 1362 PF_HASHROW_UNLOCK(khs); \ 1363 PF_HASHROW_UNLOCK(khw); \ 1364 } else \ 1365 PF_HASHROW_UNLOCK(khs); \ 1366 } while (0) 1367 1368 /* 1369 * First run: start with wire key. 1370 */ 1371 sk = skw; 1372 kh = khw; 1373 idx = PF_SK_WIRE; 1374 1375 MPASS(s->lock == NULL); 1376 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1377 1378 keyattach: 1379 LIST_FOREACH(cur, &kh->keys, entry) 1380 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1381 break; 1382 1383 if (cur != NULL) { 1384 /* Key exists. Check for same kif, if none, add to key. */ 1385 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1386 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1387 1388 PF_HASHROW_LOCK(ih); 1389 if (si->kif == s->kif && 1390 si->direction == s->direction) { 1391 if (sk->proto == IPPROTO_TCP && 1392 si->src.state >= TCPS_FIN_WAIT_2 && 1393 si->dst.state >= TCPS_FIN_WAIT_2) { 1394 /* 1395 * New state matches an old >FIN_WAIT_2 1396 * state. We can't drop key hash locks, 1397 * thus we can't unlink it properly. 1398 * 1399 * As a workaround we drop it into 1400 * TCPS_CLOSED state, schedule purge 1401 * ASAP and push it into the very end 1402 * of the slot TAILQ, so that it won't 1403 * conflict with our new state. 1404 */ 1405 pf_set_protostate(si, PF_PEER_BOTH, 1406 TCPS_CLOSED); 1407 si->timeout = PFTM_PURGE; 1408 olds = si; 1409 } else { 1410 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1411 printf("pf: %s key attach " 1412 "failed on %s: ", 1413 (idx == PF_SK_WIRE) ? 1414 "wire" : "stack", 1415 s->kif->pfik_name); 1416 pf_print_state_parts(s, 1417 (idx == PF_SK_WIRE) ? 1418 sk : NULL, 1419 (idx == PF_SK_STACK) ? 1420 sk : NULL); 1421 printf(", existing: "); 1422 pf_print_state_parts(si, 1423 (idx == PF_SK_WIRE) ? 1424 sk : NULL, 1425 (idx == PF_SK_STACK) ? 1426 sk : NULL); 1427 printf("\n"); 1428 } 1429 s->timeout = PFTM_UNLINKED; 1430 PF_HASHROW_UNLOCK(ih); 1431 KEYS_UNLOCK(); 1432 uma_zfree(V_pf_state_key_z, sk); 1433 if (idx == PF_SK_STACK) 1434 pf_detach_state(s); 1435 return (EEXIST); /* collision! */ 1436 } 1437 } 1438 PF_HASHROW_UNLOCK(ih); 1439 } 1440 uma_zfree(V_pf_state_key_z, sk); 1441 s->key[idx] = cur; 1442 } else { 1443 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1444 s->key[idx] = sk; 1445 } 1446 1447 stateattach: 1448 /* List is sorted, if-bound states before floating. */ 1449 if (s->kif == V_pfi_all) 1450 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1451 else 1452 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1453 1454 if (olds) { 1455 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1456 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1457 key_list[idx]); 1458 olds = NULL; 1459 } 1460 1461 /* 1462 * Attach done. See how should we (or should not?) 1463 * attach a second key. 1464 */ 1465 if (sks == skw) { 1466 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1467 idx = PF_SK_STACK; 1468 sks = NULL; 1469 goto stateattach; 1470 } else if (sks != NULL) { 1471 /* 1472 * Continue attaching with stack key. 1473 */ 1474 sk = sks; 1475 kh = khs; 1476 idx = PF_SK_STACK; 1477 sks = NULL; 1478 goto keyattach; 1479 } 1480 1481 PF_STATE_LOCK(s); 1482 KEYS_UNLOCK(); 1483 1484 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1485 ("%s failure", __func__)); 1486 1487 return (0); 1488 #undef KEYS_UNLOCK 1489 } 1490 1491 static void 1492 pf_detach_state(struct pf_kstate *s) 1493 { 1494 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1495 struct pf_keyhash *kh; 1496 1497 NET_EPOCH_ASSERT(); 1498 MPASS(s->timeout >= PFTM_MAX); 1499 1500 pf_sctp_multihome_detach_addr(s); 1501 1502 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1503 V_pflow_export_state_ptr(s); 1504 1505 if (sks != NULL) { 1506 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1507 PF_HASHROW_LOCK(kh); 1508 if (s->key[PF_SK_STACK] != NULL) 1509 pf_state_key_detach(s, PF_SK_STACK); 1510 /* 1511 * If both point to same key, then we are done. 1512 */ 1513 if (sks == s->key[PF_SK_WIRE]) { 1514 pf_state_key_detach(s, PF_SK_WIRE); 1515 PF_HASHROW_UNLOCK(kh); 1516 return; 1517 } 1518 PF_HASHROW_UNLOCK(kh); 1519 } 1520 1521 if (s->key[PF_SK_WIRE] != NULL) { 1522 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1523 PF_HASHROW_LOCK(kh); 1524 if (s->key[PF_SK_WIRE] != NULL) 1525 pf_state_key_detach(s, PF_SK_WIRE); 1526 PF_HASHROW_UNLOCK(kh); 1527 } 1528 } 1529 1530 static void 1531 pf_state_key_detach(struct pf_kstate *s, int idx) 1532 { 1533 struct pf_state_key *sk = s->key[idx]; 1534 #ifdef INVARIANTS 1535 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1536 1537 PF_HASHROW_ASSERT(kh); 1538 #endif 1539 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1540 s->key[idx] = NULL; 1541 1542 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1543 LIST_REMOVE(sk, entry); 1544 uma_zfree(V_pf_state_key_z, sk); 1545 } 1546 } 1547 1548 static int 1549 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1550 { 1551 struct pf_state_key *sk = mem; 1552 1553 bzero(sk, sizeof(struct pf_state_key_cmp)); 1554 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1555 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1556 1557 return (0); 1558 } 1559 1560 static int 1561 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1562 struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr, 1563 int didx, struct pf_addr *daddr, int multi) 1564 { 1565 #ifdef INET6 1566 struct nd_neighbor_solicit nd; 1567 struct pf_addr *target; 1568 u_short action, reason; 1569 1570 if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6) 1571 goto copy; 1572 1573 switch (pd->hdr.icmp6.icmp6_type) { 1574 case ND_NEIGHBOR_SOLICIT: 1575 if (multi) 1576 return (-1); 1577 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1578 return (-1); 1579 target = (struct pf_addr *)&nd.nd_ns_target; 1580 daddr = target; 1581 break; 1582 case ND_NEIGHBOR_ADVERT: 1583 if (multi) 1584 return (-1); 1585 if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af)) 1586 return (-1); 1587 target = (struct pf_addr *)&nd.nd_ns_target; 1588 saddr = target; 1589 if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) { 1590 key->addr[didx].addr32[0] = 0; 1591 key->addr[didx].addr32[1] = 0; 1592 key->addr[didx].addr32[2] = 0; 1593 key->addr[didx].addr32[3] = 0; 1594 daddr = NULL; /* overwritten */ 1595 } 1596 break; 1597 default: 1598 if (multi == PF_ICMP_MULTI_LINK) { 1599 key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL; 1600 key->addr[sidx].addr32[1] = 0; 1601 key->addr[sidx].addr32[2] = 0; 1602 key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE; 1603 saddr = NULL; /* overwritten */ 1604 } 1605 } 1606 copy: 1607 #endif 1608 if (saddr) 1609 PF_ACPY(&key->addr[sidx], saddr, pd->af); 1610 if (daddr) 1611 PF_ACPY(&key->addr[didx], daddr, pd->af); 1612 1613 return (0); 1614 } 1615 1616 struct pf_state_key * 1617 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off, 1618 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, 1619 u_int16_t dport) 1620 { 1621 struct pf_state_key *sk; 1622 1623 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1624 if (sk == NULL) 1625 return (NULL); 1626 1627 if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk, 1628 pd->sidx, pd->src, pd->didx, pd->dst, 0)) { 1629 uma_zfree(V_pf_state_key_z, sk); 1630 return (NULL); 1631 } 1632 1633 sk->port[pd->sidx] = sport; 1634 sk->port[pd->didx] = dport; 1635 sk->proto = pd->proto; 1636 sk->af = pd->af; 1637 1638 return (sk); 1639 } 1640 1641 struct pf_state_key * 1642 pf_state_key_clone(const struct pf_state_key *orig) 1643 { 1644 struct pf_state_key *sk; 1645 1646 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1647 if (sk == NULL) 1648 return (NULL); 1649 1650 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1651 1652 return (sk); 1653 } 1654 1655 int 1656 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1657 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1658 { 1659 struct pf_idhash *ih; 1660 struct pf_kstate *cur; 1661 int error; 1662 1663 NET_EPOCH_ASSERT(); 1664 1665 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1666 ("%s: sks not pristine", __func__)); 1667 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1668 ("%s: skw not pristine", __func__)); 1669 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1670 1671 s->kif = kif; 1672 s->orig_kif = orig_kif; 1673 1674 if (s->id == 0 && s->creatorid == 0) { 1675 s->id = alloc_unr64(&V_pf_stateid); 1676 s->id = htobe64(s->id); 1677 s->creatorid = V_pf_status.hostid; 1678 } 1679 1680 /* Returns with ID locked on success. */ 1681 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1682 return (error); 1683 1684 ih = &V_pf_idhash[PF_IDHASH(s)]; 1685 PF_HASHROW_ASSERT(ih); 1686 LIST_FOREACH(cur, &ih->states, entry) 1687 if (cur->id == s->id && cur->creatorid == s->creatorid) 1688 break; 1689 1690 if (cur != NULL) { 1691 s->timeout = PFTM_UNLINKED; 1692 PF_HASHROW_UNLOCK(ih); 1693 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1694 printf("pf: state ID collision: " 1695 "id: %016llx creatorid: %08x\n", 1696 (unsigned long long)be64toh(s->id), 1697 ntohl(s->creatorid)); 1698 } 1699 pf_detach_state(s); 1700 return (EEXIST); 1701 } 1702 LIST_INSERT_HEAD(&ih->states, s, entry); 1703 /* One for keys, one for ID hash. */ 1704 refcount_init(&s->refs, 2); 1705 1706 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1707 if (V_pfsync_insert_state_ptr != NULL) 1708 V_pfsync_insert_state_ptr(s); 1709 1710 /* Returns locked. */ 1711 return (0); 1712 } 1713 1714 /* 1715 * Find state by ID: returns with locked row on success. 1716 */ 1717 struct pf_kstate * 1718 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1719 { 1720 struct pf_idhash *ih; 1721 struct pf_kstate *s; 1722 1723 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1724 1725 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1726 1727 PF_HASHROW_LOCK(ih); 1728 LIST_FOREACH(s, &ih->states, entry) 1729 if (s->id == id && s->creatorid == creatorid) 1730 break; 1731 1732 if (s == NULL) 1733 PF_HASHROW_UNLOCK(ih); 1734 1735 return (s); 1736 } 1737 1738 /* 1739 * Find state by key. 1740 * Returns with ID hash slot locked on success. 1741 */ 1742 static struct pf_kstate * 1743 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key, 1744 u_int dir) 1745 { 1746 struct pf_keyhash *kh; 1747 struct pf_state_key *sk; 1748 struct pf_kstate *s; 1749 int idx; 1750 1751 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1752 1753 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1754 1755 PF_HASHROW_LOCK(kh); 1756 LIST_FOREACH(sk, &kh->keys, entry) 1757 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1758 break; 1759 if (sk == NULL) { 1760 PF_HASHROW_UNLOCK(kh); 1761 return (NULL); 1762 } 1763 1764 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1765 1766 /* List is sorted, if-bound states before floating ones. */ 1767 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1768 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1769 PF_STATE_LOCK(s); 1770 PF_HASHROW_UNLOCK(kh); 1771 if (__predict_false(s->timeout >= PFTM_MAX)) { 1772 /* 1773 * State is either being processed by 1774 * pf_unlink_state() in an other thread, or 1775 * is scheduled for immediate expiry. 1776 */ 1777 PF_STATE_UNLOCK(s); 1778 return (NULL); 1779 } 1780 return (s); 1781 } 1782 PF_HASHROW_UNLOCK(kh); 1783 1784 return (NULL); 1785 } 1786 1787 /* 1788 * Returns with ID hash slot locked on success. 1789 */ 1790 struct pf_kstate * 1791 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more) 1792 { 1793 struct pf_keyhash *kh; 1794 struct pf_state_key *sk; 1795 struct pf_kstate *s, *ret = NULL; 1796 int idx, inout = 0; 1797 1798 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1799 1800 kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)]; 1801 1802 PF_HASHROW_LOCK(kh); 1803 LIST_FOREACH(sk, &kh->keys, entry) 1804 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1805 break; 1806 if (sk == NULL) { 1807 PF_HASHROW_UNLOCK(kh); 1808 return (NULL); 1809 } 1810 switch (dir) { 1811 case PF_IN: 1812 idx = PF_SK_WIRE; 1813 break; 1814 case PF_OUT: 1815 idx = PF_SK_STACK; 1816 break; 1817 case PF_INOUT: 1818 idx = PF_SK_WIRE; 1819 inout = 1; 1820 break; 1821 default: 1822 panic("%s: dir %u", __func__, dir); 1823 } 1824 second_run: 1825 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1826 if (more == NULL) { 1827 PF_STATE_LOCK(s); 1828 PF_HASHROW_UNLOCK(kh); 1829 return (s); 1830 } 1831 1832 if (ret) 1833 (*more)++; 1834 else { 1835 ret = s; 1836 PF_STATE_LOCK(s); 1837 } 1838 } 1839 if (inout == 1) { 1840 inout = 0; 1841 idx = PF_SK_STACK; 1842 goto second_run; 1843 } 1844 PF_HASHROW_UNLOCK(kh); 1845 1846 return (ret); 1847 } 1848 1849 /* 1850 * FIXME 1851 * This routine is inefficient -- locks the state only to unlock immediately on 1852 * return. 1853 * It is racy -- after the state is unlocked nothing stops other threads from 1854 * removing it. 1855 */ 1856 bool 1857 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir) 1858 { 1859 struct pf_kstate *s; 1860 1861 s = pf_find_state_all(key, dir, NULL); 1862 if (s != NULL) { 1863 PF_STATE_UNLOCK(s); 1864 return (true); 1865 } 1866 return (false); 1867 } 1868 1869 struct pf_udp_mapping * 1870 pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port, 1871 struct pf_addr *nat_addr, uint16_t nat_port) 1872 { 1873 struct pf_udp_mapping *mapping; 1874 1875 mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO); 1876 if (mapping == NULL) 1877 return (NULL); 1878 PF_ACPY(&mapping->endpoints[0].addr, src_addr, af); 1879 mapping->endpoints[0].port = src_port; 1880 mapping->endpoints[0].af = af; 1881 mapping->endpoints[0].mapping = mapping; 1882 PF_ACPY(&mapping->endpoints[1].addr, nat_addr, af); 1883 mapping->endpoints[1].port = nat_port; 1884 mapping->endpoints[1].af = af; 1885 mapping->endpoints[1].mapping = mapping; 1886 refcount_init(&mapping->refs, 1); 1887 return (mapping); 1888 } 1889 1890 int 1891 pf_udp_mapping_insert(struct pf_udp_mapping *mapping) 1892 { 1893 struct pf_udpendpointhash *h0, *h1; 1894 struct pf_udp_endpoint *endpoint; 1895 int ret = EEXIST; 1896 1897 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1898 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1899 if (h0 == h1) { 1900 PF_HASHROW_LOCK(h0); 1901 } else if (h0 < h1) { 1902 PF_HASHROW_LOCK(h0); 1903 PF_HASHROW_LOCK(h1); 1904 } else { 1905 PF_HASHROW_LOCK(h1); 1906 PF_HASHROW_LOCK(h0); 1907 } 1908 1909 LIST_FOREACH(endpoint, &h0->endpoints, entry) { 1910 if (bcmp(endpoint, &mapping->endpoints[0], 1911 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1912 break; 1913 } 1914 if (endpoint != NULL) 1915 goto cleanup; 1916 LIST_FOREACH(endpoint, &h1->endpoints, entry) { 1917 if (bcmp(endpoint, &mapping->endpoints[1], 1918 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1919 break; 1920 } 1921 if (endpoint != NULL) 1922 goto cleanup; 1923 LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry); 1924 LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry); 1925 ret = 0; 1926 1927 cleanup: 1928 if (h0 != h1) { 1929 PF_HASHROW_UNLOCK(h0); 1930 PF_HASHROW_UNLOCK(h1); 1931 } else { 1932 PF_HASHROW_UNLOCK(h0); 1933 } 1934 return (ret); 1935 } 1936 1937 void 1938 pf_udp_mapping_release(struct pf_udp_mapping *mapping) 1939 { 1940 /* refcount is synchronized on the source endpoint's row lock */ 1941 struct pf_udpendpointhash *h0, *h1; 1942 1943 if (mapping == NULL) 1944 return; 1945 1946 h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])]; 1947 PF_HASHROW_LOCK(h0); 1948 if (refcount_release(&mapping->refs)) { 1949 LIST_REMOVE(&mapping->endpoints[0], entry); 1950 PF_HASHROW_UNLOCK(h0); 1951 h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])]; 1952 PF_HASHROW_LOCK(h1); 1953 LIST_REMOVE(&mapping->endpoints[1], entry); 1954 PF_HASHROW_UNLOCK(h1); 1955 1956 uma_zfree(V_pf_udp_mapping_z, mapping); 1957 } else { 1958 PF_HASHROW_UNLOCK(h0); 1959 } 1960 } 1961 1962 1963 struct pf_udp_mapping * 1964 pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key) 1965 { 1966 struct pf_udpendpointhash *uh; 1967 struct pf_udp_endpoint *endpoint; 1968 1969 uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)]; 1970 1971 PF_HASHROW_LOCK(uh); 1972 LIST_FOREACH(endpoint, &uh->endpoints, entry) { 1973 if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 && 1974 bcmp(endpoint, &endpoint->mapping->endpoints[0], 1975 sizeof(struct pf_udp_endpoint_cmp)) == 0) 1976 break; 1977 } 1978 if (endpoint == NULL) { 1979 PF_HASHROW_UNLOCK(uh); 1980 return (NULL); 1981 } 1982 refcount_acquire(&endpoint->mapping->refs); 1983 PF_HASHROW_UNLOCK(uh); 1984 return (endpoint->mapping); 1985 } 1986 /* END state table stuff */ 1987 1988 static void 1989 pf_send(struct pf_send_entry *pfse) 1990 { 1991 1992 PF_SENDQ_LOCK(); 1993 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1994 PF_SENDQ_UNLOCK(); 1995 swi_sched(V_pf_swi_cookie, 0); 1996 } 1997 1998 static bool 1999 pf_isforlocal(struct mbuf *m, int af) 2000 { 2001 switch (af) { 2002 #ifdef INET 2003 case AF_INET: { 2004 struct ip *ip = mtod(m, struct ip *); 2005 2006 return (in_localip(ip->ip_dst)); 2007 } 2008 #endif 2009 #ifdef INET6 2010 case AF_INET6: { 2011 struct ip6_hdr *ip6; 2012 struct in6_ifaddr *ia; 2013 ip6 = mtod(m, struct ip6_hdr *); 2014 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 2015 if (ia == NULL) 2016 return (false); 2017 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 2018 } 2019 #endif 2020 default: 2021 panic("Unsupported af %d", af); 2022 } 2023 2024 return (false); 2025 } 2026 2027 int 2028 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type, 2029 int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type) 2030 { 2031 /* 2032 * ICMP types marked with PF_OUT are typically responses to 2033 * PF_IN, and will match states in the opposite direction. 2034 * PF_IN ICMP types need to match a state with that type. 2035 */ 2036 *icmp_dir = PF_OUT; 2037 *multi = PF_ICMP_MULTI_LINK; 2038 /* Queries (and responses) */ 2039 switch (pd->af) { 2040 #ifdef INET 2041 case AF_INET: 2042 switch (type) { 2043 case ICMP_ECHO: 2044 *icmp_dir = PF_IN; 2045 case ICMP_ECHOREPLY: 2046 *virtual_type = ICMP_ECHO; 2047 *virtual_id = pd->hdr.icmp.icmp_id; 2048 break; 2049 2050 case ICMP_TSTAMP: 2051 *icmp_dir = PF_IN; 2052 case ICMP_TSTAMPREPLY: 2053 *virtual_type = ICMP_TSTAMP; 2054 *virtual_id = pd->hdr.icmp.icmp_id; 2055 break; 2056 2057 case ICMP_IREQ: 2058 *icmp_dir = PF_IN; 2059 case ICMP_IREQREPLY: 2060 *virtual_type = ICMP_IREQ; 2061 *virtual_id = pd->hdr.icmp.icmp_id; 2062 break; 2063 2064 case ICMP_MASKREQ: 2065 *icmp_dir = PF_IN; 2066 case ICMP_MASKREPLY: 2067 *virtual_type = ICMP_MASKREQ; 2068 *virtual_id = pd->hdr.icmp.icmp_id; 2069 break; 2070 2071 case ICMP_IPV6_WHEREAREYOU: 2072 *icmp_dir = PF_IN; 2073 case ICMP_IPV6_IAMHERE: 2074 *virtual_type = ICMP_IPV6_WHEREAREYOU; 2075 *virtual_id = 0; /* Nothing sane to match on! */ 2076 break; 2077 2078 case ICMP_MOBILE_REGREQUEST: 2079 *icmp_dir = PF_IN; 2080 case ICMP_MOBILE_REGREPLY: 2081 *virtual_type = ICMP_MOBILE_REGREQUEST; 2082 *virtual_id = 0; /* Nothing sane to match on! */ 2083 break; 2084 2085 case ICMP_ROUTERSOLICIT: 2086 *icmp_dir = PF_IN; 2087 case ICMP_ROUTERADVERT: 2088 *virtual_type = ICMP_ROUTERSOLICIT; 2089 *virtual_id = 0; /* Nothing sane to match on! */ 2090 break; 2091 2092 /* These ICMP types map to other connections */ 2093 case ICMP_UNREACH: 2094 case ICMP_SOURCEQUENCH: 2095 case ICMP_REDIRECT: 2096 case ICMP_TIMXCEED: 2097 case ICMP_PARAMPROB: 2098 /* These will not be used, but set them anyway */ 2099 *icmp_dir = PF_IN; 2100 *virtual_type = type; 2101 *virtual_id = 0; 2102 HTONS(*virtual_type); 2103 return (1); /* These types match to another state */ 2104 2105 /* 2106 * All remaining ICMP types get their own states, 2107 * and will only match in one direction. 2108 */ 2109 default: 2110 *icmp_dir = PF_IN; 2111 *virtual_type = type; 2112 *virtual_id = 0; 2113 break; 2114 } 2115 break; 2116 #endif /* INET */ 2117 #ifdef INET6 2118 case AF_INET6: 2119 switch (type) { 2120 case ICMP6_ECHO_REQUEST: 2121 *icmp_dir = PF_IN; 2122 case ICMP6_ECHO_REPLY: 2123 *virtual_type = ICMP6_ECHO_REQUEST; 2124 *virtual_id = pd->hdr.icmp6.icmp6_id; 2125 break; 2126 2127 case MLD_LISTENER_QUERY: 2128 case MLD_LISTENER_REPORT: { 2129 /* 2130 * Listener Report can be sent by clients 2131 * without an associated Listener Query. 2132 * In addition to that, when Report is sent as a 2133 * reply to a Query its source and destination 2134 * address are different. 2135 */ 2136 *icmp_dir = PF_IN; 2137 *virtual_type = MLD_LISTENER_QUERY; 2138 *virtual_id = 0; 2139 break; 2140 } 2141 case MLD_MTRACE: 2142 *icmp_dir = PF_IN; 2143 case MLD_MTRACE_RESP: 2144 *virtual_type = MLD_MTRACE; 2145 *virtual_id = 0; /* Nothing sane to match on! */ 2146 break; 2147 2148 case ND_NEIGHBOR_SOLICIT: 2149 *icmp_dir = PF_IN; 2150 case ND_NEIGHBOR_ADVERT: { 2151 *virtual_type = ND_NEIGHBOR_SOLICIT; 2152 *virtual_id = 0; 2153 break; 2154 } 2155 2156 /* 2157 * These ICMP types map to other connections. 2158 * ND_REDIRECT can't be in this list because the triggering 2159 * packet header is optional. 2160 */ 2161 case ICMP6_DST_UNREACH: 2162 case ICMP6_PACKET_TOO_BIG: 2163 case ICMP6_TIME_EXCEEDED: 2164 case ICMP6_PARAM_PROB: 2165 /* These will not be used, but set them anyway */ 2166 *icmp_dir = PF_IN; 2167 *virtual_type = type; 2168 *virtual_id = 0; 2169 HTONS(*virtual_type); 2170 return (1); /* These types match to another state */ 2171 /* 2172 * All remaining ICMP6 types get their own states, 2173 * and will only match in one direction. 2174 */ 2175 default: 2176 *icmp_dir = PF_IN; 2177 *virtual_type = type; 2178 *virtual_id = 0; 2179 break; 2180 } 2181 break; 2182 #endif /* INET6 */ 2183 default: 2184 *icmp_dir = PF_IN; 2185 *virtual_type = type; 2186 *virtual_id = 0; 2187 break; 2188 } 2189 HTONS(*virtual_type); 2190 return (0); /* These types match to their own state */ 2191 } 2192 2193 void 2194 pf_intr(void *v) 2195 { 2196 struct epoch_tracker et; 2197 struct pf_send_head queue; 2198 struct pf_send_entry *pfse, *next; 2199 2200 CURVNET_SET((struct vnet *)v); 2201 2202 PF_SENDQ_LOCK(); 2203 queue = V_pf_sendqueue; 2204 STAILQ_INIT(&V_pf_sendqueue); 2205 PF_SENDQ_UNLOCK(); 2206 2207 NET_EPOCH_ENTER(et); 2208 2209 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 2210 switch (pfse->pfse_type) { 2211 #ifdef INET 2212 case PFSE_IP: { 2213 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 2214 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2215 pfse->pfse_m->m_pkthdr.csum_flags |= 2216 CSUM_IP_VALID | CSUM_IP_CHECKED; 2217 ip_input(pfse->pfse_m); 2218 } else { 2219 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2220 NULL); 2221 } 2222 break; 2223 } 2224 case PFSE_ICMP: 2225 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 2226 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 2227 break; 2228 #endif /* INET */ 2229 #ifdef INET6 2230 case PFSE_IP6: 2231 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 2232 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 2233 ip6_input(pfse->pfse_m); 2234 } else { 2235 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 2236 NULL, NULL); 2237 } 2238 break; 2239 case PFSE_ICMP6: 2240 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 2241 pfse->icmpopts.code, pfse->icmpopts.mtu); 2242 break; 2243 #endif /* INET6 */ 2244 default: 2245 panic("%s: unknown type", __func__); 2246 } 2247 free(pfse, M_PFTEMP); 2248 } 2249 NET_EPOCH_EXIT(et); 2250 CURVNET_RESTORE(); 2251 } 2252 2253 #define pf_purge_thread_period (hz / 10) 2254 2255 #ifdef PF_WANT_32_TO_64_COUNTER 2256 static void 2257 pf_status_counter_u64_periodic(void) 2258 { 2259 2260 PF_RULES_RASSERT(); 2261 2262 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 2263 return; 2264 } 2265 2266 for (int i = 0; i < FCNT_MAX; i++) { 2267 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 2268 } 2269 } 2270 2271 static void 2272 pf_kif_counter_u64_periodic(void) 2273 { 2274 struct pfi_kkif *kif; 2275 size_t r, run; 2276 2277 PF_RULES_RASSERT(); 2278 2279 if (__predict_false(V_pf_allkifcount == 0)) { 2280 return; 2281 } 2282 2283 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2284 return; 2285 } 2286 2287 run = V_pf_allkifcount / 10; 2288 if (run < 5) 2289 run = 5; 2290 2291 for (r = 0; r < run; r++) { 2292 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 2293 if (kif == NULL) { 2294 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2295 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 2296 break; 2297 } 2298 2299 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 2300 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 2301 2302 for (int i = 0; i < 2; i++) { 2303 for (int j = 0; j < 2; j++) { 2304 for (int k = 0; k < 2; k++) { 2305 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 2306 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 2307 } 2308 } 2309 } 2310 } 2311 } 2312 2313 static void 2314 pf_rule_counter_u64_periodic(void) 2315 { 2316 struct pf_krule *rule; 2317 size_t r, run; 2318 2319 PF_RULES_RASSERT(); 2320 2321 if (__predict_false(V_pf_allrulecount == 0)) { 2322 return; 2323 } 2324 2325 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 2326 return; 2327 } 2328 2329 run = V_pf_allrulecount / 10; 2330 if (run < 5) 2331 run = 5; 2332 2333 for (r = 0; r < run; r++) { 2334 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 2335 if (rule == NULL) { 2336 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2337 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 2338 break; 2339 } 2340 2341 LIST_REMOVE(V_pf_rulemarker, allrulelist); 2342 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 2343 2344 pf_counter_u64_periodic(&rule->evaluations); 2345 for (int i = 0; i < 2; i++) { 2346 pf_counter_u64_periodic(&rule->packets[i]); 2347 pf_counter_u64_periodic(&rule->bytes[i]); 2348 } 2349 } 2350 } 2351 2352 static void 2353 pf_counter_u64_periodic_main(void) 2354 { 2355 PF_RULES_RLOCK_TRACKER; 2356 2357 V_pf_counter_periodic_iter++; 2358 2359 PF_RULES_RLOCK(); 2360 pf_counter_u64_critical_enter(); 2361 pf_status_counter_u64_periodic(); 2362 pf_kif_counter_u64_periodic(); 2363 pf_rule_counter_u64_periodic(); 2364 pf_counter_u64_critical_exit(); 2365 PF_RULES_RUNLOCK(); 2366 } 2367 #else 2368 #define pf_counter_u64_periodic_main() do { } while (0) 2369 #endif 2370 2371 void 2372 pf_purge_thread(void *unused __unused) 2373 { 2374 struct epoch_tracker et; 2375 2376 VNET_ITERATOR_DECL(vnet_iter); 2377 2378 sx_xlock(&pf_end_lock); 2379 while (pf_end_threads == 0) { 2380 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 2381 2382 VNET_LIST_RLOCK(); 2383 NET_EPOCH_ENTER(et); 2384 VNET_FOREACH(vnet_iter) { 2385 CURVNET_SET(vnet_iter); 2386 2387 /* Wait until V_pf_default_rule is initialized. */ 2388 if (V_pf_vnet_active == 0) { 2389 CURVNET_RESTORE(); 2390 continue; 2391 } 2392 2393 pf_counter_u64_periodic_main(); 2394 2395 /* 2396 * Process 1/interval fraction of the state 2397 * table every run. 2398 */ 2399 V_pf_purge_idx = 2400 pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / 2401 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 2402 2403 /* 2404 * Purge other expired types every 2405 * PFTM_INTERVAL seconds. 2406 */ 2407 if (V_pf_purge_idx == 0) { 2408 /* 2409 * Order is important: 2410 * - states and src nodes reference rules 2411 * - states and rules reference kifs 2412 */ 2413 pf_purge_expired_fragments(); 2414 pf_purge_expired_src_nodes(); 2415 pf_purge_unlinked_rules(); 2416 pfi_kkif_purge(); 2417 } 2418 CURVNET_RESTORE(); 2419 } 2420 NET_EPOCH_EXIT(et); 2421 VNET_LIST_RUNLOCK(); 2422 } 2423 2424 pf_end_threads++; 2425 sx_xunlock(&pf_end_lock); 2426 kproc_exit(0); 2427 } 2428 2429 void 2430 pf_unload_vnet_purge(void) 2431 { 2432 2433 /* 2434 * To cleanse up all kifs and rules we need 2435 * two runs: first one clears reference flags, 2436 * then pf_purge_expired_states() doesn't 2437 * raise them, and then second run frees. 2438 */ 2439 pf_purge_unlinked_rules(); 2440 pfi_kkif_purge(); 2441 2442 /* 2443 * Now purge everything. 2444 */ 2445 pf_purge_expired_states(0, V_pf_hashmask); 2446 pf_purge_fragments(UINT_MAX); 2447 pf_purge_expired_src_nodes(); 2448 2449 /* 2450 * Now all kifs & rules should be unreferenced, 2451 * thus should be successfully freed. 2452 */ 2453 pf_purge_unlinked_rules(); 2454 pfi_kkif_purge(); 2455 } 2456 2457 u_int32_t 2458 pf_state_expires(const struct pf_kstate *state) 2459 { 2460 u_int32_t timeout; 2461 u_int32_t start; 2462 u_int32_t end; 2463 u_int32_t states; 2464 2465 /* handle all PFTM_* > PFTM_MAX here */ 2466 if (state->timeout == PFTM_PURGE) 2467 return (time_uptime); 2468 KASSERT(state->timeout != PFTM_UNLINKED, 2469 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2470 KASSERT((state->timeout < PFTM_MAX), 2471 ("pf_state_expires: timeout > PFTM_MAX")); 2472 timeout = state->rule.ptr->timeout[state->timeout]; 2473 if (!timeout) 2474 timeout = V_pf_default_rule.timeout[state->timeout]; 2475 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2476 if (start && state->rule.ptr != &V_pf_default_rule) { 2477 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2478 states = counter_u64_fetch(state->rule.ptr->states_cur); 2479 } else { 2480 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2481 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2482 states = V_pf_status.states; 2483 } 2484 if (end && states > start && start < end) { 2485 if (states < end) { 2486 timeout = (u_int64_t)timeout * (end - states) / 2487 (end - start); 2488 return ((state->expire / 1000) + timeout); 2489 } 2490 else 2491 return (time_uptime); 2492 } 2493 return ((state->expire / 1000) + timeout); 2494 } 2495 2496 void 2497 pf_purge_expired_src_nodes(void) 2498 { 2499 struct pf_ksrc_node_list freelist; 2500 struct pf_srchash *sh; 2501 struct pf_ksrc_node *cur, *next; 2502 int i; 2503 2504 LIST_INIT(&freelist); 2505 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 2506 PF_HASHROW_LOCK(sh); 2507 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2508 if (cur->states == 0 && cur->expire <= time_uptime) { 2509 pf_unlink_src_node(cur); 2510 LIST_INSERT_HEAD(&freelist, cur, entry); 2511 } else if (cur->rule.ptr != NULL) 2512 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2513 PF_HASHROW_UNLOCK(sh); 2514 } 2515 2516 pf_free_src_nodes(&freelist); 2517 2518 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2519 } 2520 2521 static void 2522 pf_src_tree_remove_state(struct pf_kstate *s) 2523 { 2524 struct pf_ksrc_node *sn; 2525 uint32_t timeout; 2526 2527 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2528 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2529 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2530 2531 if (s->src_node != NULL) { 2532 sn = s->src_node; 2533 PF_SRC_NODE_LOCK(sn); 2534 if (s->src.tcp_est) 2535 --sn->conn; 2536 if (--sn->states == 0) 2537 sn->expire = time_uptime + timeout; 2538 PF_SRC_NODE_UNLOCK(sn); 2539 } 2540 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2541 sn = s->nat_src_node; 2542 PF_SRC_NODE_LOCK(sn); 2543 if (--sn->states == 0) 2544 sn->expire = time_uptime + timeout; 2545 PF_SRC_NODE_UNLOCK(sn); 2546 } 2547 s->src_node = s->nat_src_node = NULL; 2548 } 2549 2550 /* 2551 * Unlink and potentilly free a state. Function may be 2552 * called with ID hash row locked, but always returns 2553 * unlocked, since it needs to go through key hash locking. 2554 */ 2555 int 2556 pf_unlink_state(struct pf_kstate *s) 2557 { 2558 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2559 2560 NET_EPOCH_ASSERT(); 2561 PF_HASHROW_ASSERT(ih); 2562 2563 if (s->timeout == PFTM_UNLINKED) { 2564 /* 2565 * State is being processed 2566 * by pf_unlink_state() in 2567 * an other thread. 2568 */ 2569 PF_HASHROW_UNLOCK(ih); 2570 return (0); /* XXXGL: undefined actually */ 2571 } 2572 2573 if (s->src.state == PF_TCPS_PROXY_DST) { 2574 /* XXX wire key the right one? */ 2575 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2576 &s->key[PF_SK_WIRE]->addr[1], 2577 &s->key[PF_SK_WIRE]->addr[0], 2578 s->key[PF_SK_WIRE]->port[1], 2579 s->key[PF_SK_WIRE]->port[0], 2580 s->src.seqhi, s->src.seqlo + 1, 2581 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2582 } 2583 2584 LIST_REMOVE(s, entry); 2585 pf_src_tree_remove_state(s); 2586 2587 if (V_pfsync_delete_state_ptr != NULL) 2588 V_pfsync_delete_state_ptr(s); 2589 2590 STATE_DEC_COUNTERS(s); 2591 2592 s->timeout = PFTM_UNLINKED; 2593 2594 /* Ensure we remove it from the list of halfopen states, if needed. */ 2595 if (s->key[PF_SK_STACK] != NULL && 2596 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2597 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2598 2599 PF_HASHROW_UNLOCK(ih); 2600 2601 pf_detach_state(s); 2602 2603 pf_udp_mapping_release(s->udp_mapping); 2604 2605 /* pf_state_insert() initialises refs to 2 */ 2606 return (pf_release_staten(s, 2)); 2607 } 2608 2609 struct pf_kstate * 2610 pf_alloc_state(int flags) 2611 { 2612 2613 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2614 } 2615 2616 void 2617 pf_free_state(struct pf_kstate *cur) 2618 { 2619 struct pf_krule_item *ri; 2620 2621 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2622 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2623 cur->timeout)); 2624 2625 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2626 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2627 free(ri, M_PF_RULE_ITEM); 2628 } 2629 2630 pf_normalize_tcp_cleanup(cur); 2631 uma_zfree(V_pf_state_z, cur); 2632 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2633 } 2634 2635 /* 2636 * Called only from pf_purge_thread(), thus serialized. 2637 */ 2638 static u_int 2639 pf_purge_expired_states(u_int i, int maxcheck) 2640 { 2641 struct pf_idhash *ih; 2642 struct pf_kstate *s; 2643 struct pf_krule_item *mrm; 2644 size_t count __unused; 2645 2646 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2647 2648 /* 2649 * Go through hash and unlink states that expire now. 2650 */ 2651 while (maxcheck > 0) { 2652 count = 0; 2653 ih = &V_pf_idhash[i]; 2654 2655 /* only take the lock if we expect to do work */ 2656 if (!LIST_EMPTY(&ih->states)) { 2657 relock: 2658 PF_HASHROW_LOCK(ih); 2659 LIST_FOREACH(s, &ih->states, entry) { 2660 if (pf_state_expires(s) <= time_uptime) { 2661 V_pf_status.states -= 2662 pf_unlink_state(s); 2663 goto relock; 2664 } 2665 s->rule.ptr->rule_ref |= PFRULE_REFS; 2666 if (s->nat_rule.ptr != NULL) 2667 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2668 if (s->anchor.ptr != NULL) 2669 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2670 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2671 SLIST_FOREACH(mrm, &s->match_rules, entry) 2672 mrm->r->rule_ref |= PFRULE_REFS; 2673 if (s->rt_kif) 2674 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2675 count++; 2676 } 2677 PF_HASHROW_UNLOCK(ih); 2678 } 2679 2680 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2681 2682 /* Return when we hit end of hash. */ 2683 if (++i > V_pf_hashmask) { 2684 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2685 return (0); 2686 } 2687 2688 maxcheck--; 2689 } 2690 2691 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2692 2693 return (i); 2694 } 2695 2696 static void 2697 pf_purge_unlinked_rules(void) 2698 { 2699 struct pf_krulequeue tmpq; 2700 struct pf_krule *r, *r1; 2701 2702 /* 2703 * If we have overloading task pending, then we'd 2704 * better skip purging this time. There is a tiny 2705 * probability that overloading task references 2706 * an already unlinked rule. 2707 */ 2708 PF_OVERLOADQ_LOCK(); 2709 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2710 PF_OVERLOADQ_UNLOCK(); 2711 return; 2712 } 2713 PF_OVERLOADQ_UNLOCK(); 2714 2715 /* 2716 * Do naive mark-and-sweep garbage collecting of old rules. 2717 * Reference flag is raised by pf_purge_expired_states() 2718 * and pf_purge_expired_src_nodes(). 2719 * 2720 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2721 * use a temporary queue. 2722 */ 2723 TAILQ_INIT(&tmpq); 2724 PF_UNLNKDRULES_LOCK(); 2725 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2726 if (!(r->rule_ref & PFRULE_REFS)) { 2727 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2728 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2729 } else 2730 r->rule_ref &= ~PFRULE_REFS; 2731 } 2732 PF_UNLNKDRULES_UNLOCK(); 2733 2734 if (!TAILQ_EMPTY(&tmpq)) { 2735 PF_CONFIG_LOCK(); 2736 PF_RULES_WLOCK(); 2737 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2738 TAILQ_REMOVE(&tmpq, r, entries); 2739 pf_free_rule(r); 2740 } 2741 PF_RULES_WUNLOCK(); 2742 PF_CONFIG_UNLOCK(); 2743 } 2744 } 2745 2746 void 2747 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2748 { 2749 switch (af) { 2750 #ifdef INET 2751 case AF_INET: { 2752 u_int32_t a = ntohl(addr->addr32[0]); 2753 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2754 (a>>8)&255, a&255); 2755 if (p) { 2756 p = ntohs(p); 2757 printf(":%u", p); 2758 } 2759 break; 2760 } 2761 #endif /* INET */ 2762 #ifdef INET6 2763 case AF_INET6: { 2764 u_int16_t b; 2765 u_int8_t i, curstart, curend, maxstart, maxend; 2766 curstart = curend = maxstart = maxend = 255; 2767 for (i = 0; i < 8; i++) { 2768 if (!addr->addr16[i]) { 2769 if (curstart == 255) 2770 curstart = i; 2771 curend = i; 2772 } else { 2773 if ((curend - curstart) > 2774 (maxend - maxstart)) { 2775 maxstart = curstart; 2776 maxend = curend; 2777 } 2778 curstart = curend = 255; 2779 } 2780 } 2781 if ((curend - curstart) > 2782 (maxend - maxstart)) { 2783 maxstart = curstart; 2784 maxend = curend; 2785 } 2786 for (i = 0; i < 8; i++) { 2787 if (i >= maxstart && i <= maxend) { 2788 if (i == 0) 2789 printf(":"); 2790 if (i == maxend) 2791 printf(":"); 2792 } else { 2793 b = ntohs(addr->addr16[i]); 2794 printf("%x", b); 2795 if (i < 7) 2796 printf(":"); 2797 } 2798 } 2799 if (p) { 2800 p = ntohs(p); 2801 printf("[%u]", p); 2802 } 2803 break; 2804 } 2805 #endif /* INET6 */ 2806 } 2807 } 2808 2809 void 2810 pf_print_state(struct pf_kstate *s) 2811 { 2812 pf_print_state_parts(s, NULL, NULL); 2813 } 2814 2815 static void 2816 pf_print_state_parts(struct pf_kstate *s, 2817 struct pf_state_key *skwp, struct pf_state_key *sksp) 2818 { 2819 struct pf_state_key *skw, *sks; 2820 u_int8_t proto, dir; 2821 2822 /* Do our best to fill these, but they're skipped if NULL */ 2823 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2824 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2825 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2826 dir = s ? s->direction : 0; 2827 2828 switch (proto) { 2829 case IPPROTO_IPV4: 2830 printf("IPv4"); 2831 break; 2832 case IPPROTO_IPV6: 2833 printf("IPv6"); 2834 break; 2835 case IPPROTO_TCP: 2836 printf("TCP"); 2837 break; 2838 case IPPROTO_UDP: 2839 printf("UDP"); 2840 break; 2841 case IPPROTO_ICMP: 2842 printf("ICMP"); 2843 break; 2844 case IPPROTO_ICMPV6: 2845 printf("ICMPv6"); 2846 break; 2847 default: 2848 printf("%u", proto); 2849 break; 2850 } 2851 switch (dir) { 2852 case PF_IN: 2853 printf(" in"); 2854 break; 2855 case PF_OUT: 2856 printf(" out"); 2857 break; 2858 } 2859 if (skw) { 2860 printf(" wire: "); 2861 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2862 printf(" "); 2863 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2864 } 2865 if (sks) { 2866 printf(" stack: "); 2867 if (sks != skw) { 2868 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2869 printf(" "); 2870 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2871 } else 2872 printf("-"); 2873 } 2874 if (s) { 2875 if (proto == IPPROTO_TCP) { 2876 printf(" [lo=%u high=%u win=%u modulator=%u", 2877 s->src.seqlo, s->src.seqhi, 2878 s->src.max_win, s->src.seqdiff); 2879 if (s->src.wscale && s->dst.wscale) 2880 printf(" wscale=%u", 2881 s->src.wscale & PF_WSCALE_MASK); 2882 printf("]"); 2883 printf(" [lo=%u high=%u win=%u modulator=%u", 2884 s->dst.seqlo, s->dst.seqhi, 2885 s->dst.max_win, s->dst.seqdiff); 2886 if (s->src.wscale && s->dst.wscale) 2887 printf(" wscale=%u", 2888 s->dst.wscale & PF_WSCALE_MASK); 2889 printf("]"); 2890 } 2891 printf(" %u:%u", s->src.state, s->dst.state); 2892 if (s->rule.ptr) 2893 printf(" @%d", s->rule.ptr->nr); 2894 } 2895 } 2896 2897 void 2898 pf_print_flags(u_int8_t f) 2899 { 2900 if (f) 2901 printf(" "); 2902 if (f & TH_FIN) 2903 printf("F"); 2904 if (f & TH_SYN) 2905 printf("S"); 2906 if (f & TH_RST) 2907 printf("R"); 2908 if (f & TH_PUSH) 2909 printf("P"); 2910 if (f & TH_ACK) 2911 printf("A"); 2912 if (f & TH_URG) 2913 printf("U"); 2914 if (f & TH_ECE) 2915 printf("E"); 2916 if (f & TH_CWR) 2917 printf("W"); 2918 } 2919 2920 #define PF_SET_SKIP_STEPS(i) \ 2921 do { \ 2922 while (head[i] != cur) { \ 2923 head[i]->skip[i].ptr = cur; \ 2924 head[i] = TAILQ_NEXT(head[i], entries); \ 2925 } \ 2926 } while (0) 2927 2928 void 2929 pf_calc_skip_steps(struct pf_krulequeue *rules) 2930 { 2931 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2932 int i; 2933 2934 cur = TAILQ_FIRST(rules); 2935 prev = cur; 2936 for (i = 0; i < PF_SKIP_COUNT; ++i) 2937 head[i] = cur; 2938 while (cur != NULL) { 2939 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2940 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2941 if (cur->direction != prev->direction) 2942 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2943 if (cur->af != prev->af) 2944 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2945 if (cur->proto != prev->proto) 2946 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2947 if (cur->src.neg != prev->src.neg || 2948 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2949 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2950 if (cur->src.port[0] != prev->src.port[0] || 2951 cur->src.port[1] != prev->src.port[1] || 2952 cur->src.port_op != prev->src.port_op) 2953 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2954 if (cur->dst.neg != prev->dst.neg || 2955 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2956 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2957 if (cur->dst.port[0] != prev->dst.port[0] || 2958 cur->dst.port[1] != prev->dst.port[1] || 2959 cur->dst.port_op != prev->dst.port_op) 2960 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2961 2962 prev = cur; 2963 cur = TAILQ_NEXT(cur, entries); 2964 } 2965 for (i = 0; i < PF_SKIP_COUNT; ++i) 2966 PF_SET_SKIP_STEPS(i); 2967 } 2968 2969 int 2970 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2971 { 2972 if (aw1->type != aw2->type) 2973 return (1); 2974 switch (aw1->type) { 2975 case PF_ADDR_ADDRMASK: 2976 case PF_ADDR_RANGE: 2977 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2978 return (1); 2979 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2980 return (1); 2981 return (0); 2982 case PF_ADDR_DYNIFTL: 2983 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2984 case PF_ADDR_NOROUTE: 2985 case PF_ADDR_URPFFAILED: 2986 return (0); 2987 case PF_ADDR_TABLE: 2988 return (aw1->p.tbl != aw2->p.tbl); 2989 default: 2990 printf("invalid address type: %d\n", aw1->type); 2991 return (1); 2992 } 2993 } 2994 2995 /** 2996 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2997 * header isn't always a full checksum. In some cases (i.e. output) it's a 2998 * pseudo-header checksum, which is a partial checksum over src/dst IP 2999 * addresses, protocol number and length. 3000 * 3001 * That means we have the following cases: 3002 * * Input or forwarding: we don't have TSO, the checksum fields are full 3003 * checksums, we need to update the checksum whenever we change anything. 3004 * * Output (i.e. the checksum is a pseudo-header checksum): 3005 * x The field being updated is src/dst address or affects the length of 3006 * the packet. We need to update the pseudo-header checksum (note that this 3007 * checksum is not ones' complement). 3008 * x Some other field is being modified (e.g. src/dst port numbers): We 3009 * don't have to update anything. 3010 **/ 3011 u_int16_t 3012 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 3013 { 3014 u_int32_t x; 3015 3016 x = cksum + old - new; 3017 x = (x + (x >> 16)) & 0xffff; 3018 3019 /* optimise: eliminate a branch when not udp */ 3020 if (udp && cksum == 0x0000) 3021 return cksum; 3022 if (udp && x == 0x0000) 3023 x = 0xffff; 3024 3025 return (u_int16_t)(x); 3026 } 3027 3028 static void 3029 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 3030 u_int8_t udp) 3031 { 3032 u_int16_t old = htons(hi ? (*f << 8) : *f); 3033 u_int16_t new = htons(hi ? ( v << 8) : v); 3034 3035 if (*f == v) 3036 return; 3037 3038 *f = v; 3039 3040 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3041 return; 3042 3043 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 3044 } 3045 3046 void 3047 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 3048 bool hi, u_int8_t udp) 3049 { 3050 u_int8_t *fb = (u_int8_t *)f; 3051 u_int8_t *vb = (u_int8_t *)&v; 3052 3053 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3054 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3055 } 3056 3057 void 3058 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 3059 bool hi, u_int8_t udp) 3060 { 3061 u_int8_t *fb = (u_int8_t *)f; 3062 u_int8_t *vb = (u_int8_t *)&v; 3063 3064 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3065 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3066 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 3067 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 3068 } 3069 3070 u_int16_t 3071 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 3072 u_int16_t new, u_int8_t udp) 3073 { 3074 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3075 return (cksum); 3076 3077 return (pf_cksum_fixup(cksum, old, new, udp)); 3078 } 3079 3080 static void 3081 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 3082 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 3083 sa_family_t af) 3084 { 3085 struct pf_addr ao; 3086 u_int16_t po = *p; 3087 3088 PF_ACPY(&ao, a, af); 3089 PF_ACPY(a, an, af); 3090 3091 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 3092 *pc = ~*pc; 3093 3094 *p = pn; 3095 3096 switch (af) { 3097 #ifdef INET 3098 case AF_INET: 3099 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3100 ao.addr16[0], an->addr16[0], 0), 3101 ao.addr16[1], an->addr16[1], 0); 3102 *p = pn; 3103 3104 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 3105 ao.addr16[0], an->addr16[0], u), 3106 ao.addr16[1], an->addr16[1], u); 3107 3108 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3109 break; 3110 #endif /* INET */ 3111 #ifdef INET6 3112 case AF_INET6: 3113 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3114 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3115 pf_cksum_fixup(pf_cksum_fixup(*pc, 3116 ao.addr16[0], an->addr16[0], u), 3117 ao.addr16[1], an->addr16[1], u), 3118 ao.addr16[2], an->addr16[2], u), 3119 ao.addr16[3], an->addr16[3], u), 3120 ao.addr16[4], an->addr16[4], u), 3121 ao.addr16[5], an->addr16[5], u), 3122 ao.addr16[6], an->addr16[6], u), 3123 ao.addr16[7], an->addr16[7], u); 3124 3125 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 3126 break; 3127 #endif /* INET6 */ 3128 } 3129 3130 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 3131 CSUM_DELAY_DATA_IPV6)) { 3132 *pc = ~*pc; 3133 if (! *pc) 3134 *pc = 0xffff; 3135 } 3136 } 3137 3138 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 3139 void 3140 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 3141 { 3142 u_int32_t ao; 3143 3144 memcpy(&ao, a, sizeof(ao)); 3145 memcpy(a, &an, sizeof(u_int32_t)); 3146 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 3147 ao % 65536, an % 65536, u); 3148 } 3149 3150 void 3151 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 3152 { 3153 u_int32_t ao; 3154 3155 memcpy(&ao, a, sizeof(ao)); 3156 memcpy(a, &an, sizeof(u_int32_t)); 3157 3158 *c = pf_proto_cksum_fixup(m, 3159 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 3160 ao % 65536, an % 65536, udp); 3161 } 3162 3163 #ifdef INET6 3164 static void 3165 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 3166 { 3167 struct pf_addr ao; 3168 3169 PF_ACPY(&ao, a, AF_INET6); 3170 PF_ACPY(a, an, AF_INET6); 3171 3172 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3173 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3174 pf_cksum_fixup(pf_cksum_fixup(*c, 3175 ao.addr16[0], an->addr16[0], u), 3176 ao.addr16[1], an->addr16[1], u), 3177 ao.addr16[2], an->addr16[2], u), 3178 ao.addr16[3], an->addr16[3], u), 3179 ao.addr16[4], an->addr16[4], u), 3180 ao.addr16[5], an->addr16[5], u), 3181 ao.addr16[6], an->addr16[6], u), 3182 ao.addr16[7], an->addr16[7], u); 3183 } 3184 #endif /* INET6 */ 3185 3186 static void 3187 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 3188 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 3189 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 3190 { 3191 struct pf_addr oia, ooa; 3192 3193 PF_ACPY(&oia, ia, af); 3194 if (oa) 3195 PF_ACPY(&ooa, oa, af); 3196 3197 /* Change inner protocol port, fix inner protocol checksum. */ 3198 if (ip != NULL) { 3199 u_int16_t oip = *ip; 3200 u_int32_t opc; 3201 3202 if (pc != NULL) 3203 opc = *pc; 3204 *ip = np; 3205 if (pc != NULL) 3206 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 3207 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 3208 if (pc != NULL) 3209 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 3210 } 3211 /* Change inner ip address, fix inner ip and icmp checksums. */ 3212 PF_ACPY(ia, na, af); 3213 switch (af) { 3214 #ifdef INET 3215 case AF_INET: { 3216 u_int32_t oh2c = *h2c; 3217 3218 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 3219 oia.addr16[0], ia->addr16[0], 0), 3220 oia.addr16[1], ia->addr16[1], 0); 3221 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 3222 oia.addr16[0], ia->addr16[0], 0), 3223 oia.addr16[1], ia->addr16[1], 0); 3224 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 3225 break; 3226 } 3227 #endif /* INET */ 3228 #ifdef INET6 3229 case AF_INET6: 3230 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3231 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3232 pf_cksum_fixup(pf_cksum_fixup(*ic, 3233 oia.addr16[0], ia->addr16[0], u), 3234 oia.addr16[1], ia->addr16[1], u), 3235 oia.addr16[2], ia->addr16[2], u), 3236 oia.addr16[3], ia->addr16[3], u), 3237 oia.addr16[4], ia->addr16[4], u), 3238 oia.addr16[5], ia->addr16[5], u), 3239 oia.addr16[6], ia->addr16[6], u), 3240 oia.addr16[7], ia->addr16[7], u); 3241 break; 3242 #endif /* INET6 */ 3243 } 3244 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 3245 if (oa) { 3246 PF_ACPY(oa, na, af); 3247 switch (af) { 3248 #ifdef INET 3249 case AF_INET: 3250 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 3251 ooa.addr16[0], oa->addr16[0], 0), 3252 ooa.addr16[1], oa->addr16[1], 0); 3253 break; 3254 #endif /* INET */ 3255 #ifdef INET6 3256 case AF_INET6: 3257 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3258 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 3259 pf_cksum_fixup(pf_cksum_fixup(*ic, 3260 ooa.addr16[0], oa->addr16[0], u), 3261 ooa.addr16[1], oa->addr16[1], u), 3262 ooa.addr16[2], oa->addr16[2], u), 3263 ooa.addr16[3], oa->addr16[3], u), 3264 ooa.addr16[4], oa->addr16[4], u), 3265 ooa.addr16[5], oa->addr16[5], u), 3266 ooa.addr16[6], oa->addr16[6], u), 3267 ooa.addr16[7], oa->addr16[7], u); 3268 break; 3269 #endif /* INET6 */ 3270 } 3271 } 3272 } 3273 3274 /* 3275 * Need to modulate the sequence numbers in the TCP SACK option 3276 * (credits to Krzysztof Pfaff for report and patch) 3277 */ 3278 static int 3279 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 3280 struct tcphdr *th, struct pf_state_peer *dst) 3281 { 3282 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 3283 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 3284 int copyback = 0, i, olen; 3285 struct sackblk sack; 3286 3287 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 3288 if (hlen < TCPOLEN_SACKLEN || 3289 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 3290 return 0; 3291 3292 while (hlen >= TCPOLEN_SACKLEN) { 3293 size_t startoff = opt - opts; 3294 olen = opt[1]; 3295 switch (*opt) { 3296 case TCPOPT_EOL: /* FALLTHROUGH */ 3297 case TCPOPT_NOP: 3298 opt++; 3299 hlen--; 3300 break; 3301 case TCPOPT_SACK: 3302 if (olen > hlen) 3303 olen = hlen; 3304 if (olen >= TCPOLEN_SACKLEN) { 3305 for (i = 2; i + TCPOLEN_SACK <= olen; 3306 i += TCPOLEN_SACK) { 3307 memcpy(&sack, &opt[i], sizeof(sack)); 3308 pf_patch_32_unaligned(m, 3309 &th->th_sum, &sack.start, 3310 htonl(ntohl(sack.start) - dst->seqdiff), 3311 PF_ALGNMNT(startoff), 3312 0); 3313 pf_patch_32_unaligned(m, &th->th_sum, 3314 &sack.end, 3315 htonl(ntohl(sack.end) - dst->seqdiff), 3316 PF_ALGNMNT(startoff), 3317 0); 3318 memcpy(&opt[i], &sack, sizeof(sack)); 3319 } 3320 copyback = 1; 3321 } 3322 /* FALLTHROUGH */ 3323 default: 3324 if (olen < 2) 3325 olen = 2; 3326 hlen -= olen; 3327 opt += olen; 3328 } 3329 } 3330 3331 if (copyback) 3332 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 3333 return (copyback); 3334 } 3335 3336 struct mbuf * 3337 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 3338 const struct pf_addr *saddr, const struct pf_addr *daddr, 3339 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3340 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3341 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3342 { 3343 struct mbuf *m; 3344 int len, tlen; 3345 #ifdef INET 3346 struct ip *h = NULL; 3347 #endif /* INET */ 3348 #ifdef INET6 3349 struct ip6_hdr *h6 = NULL; 3350 #endif /* INET6 */ 3351 struct tcphdr *th; 3352 char *opt; 3353 struct pf_mtag *pf_mtag; 3354 3355 len = 0; 3356 th = NULL; 3357 3358 /* maximum segment size tcp option */ 3359 tlen = sizeof(struct tcphdr); 3360 if (mss) 3361 tlen += 4; 3362 3363 switch (af) { 3364 #ifdef INET 3365 case AF_INET: 3366 len = sizeof(struct ip) + tlen; 3367 break; 3368 #endif /* INET */ 3369 #ifdef INET6 3370 case AF_INET6: 3371 len = sizeof(struct ip6_hdr) + tlen; 3372 break; 3373 #endif /* INET6 */ 3374 default: 3375 panic("%s: unsupported af %d", __func__, af); 3376 } 3377 3378 m = m_gethdr(M_NOWAIT, MT_DATA); 3379 if (m == NULL) 3380 return (NULL); 3381 3382 #ifdef MAC 3383 mac_netinet_firewall_send(m); 3384 #endif 3385 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 3386 m_freem(m); 3387 return (NULL); 3388 } 3389 if (skip_firewall) 3390 m->m_flags |= M_SKIP_FIREWALL; 3391 pf_mtag->tag = mtag_tag; 3392 pf_mtag->flags = mtag_flags; 3393 3394 if (rtableid >= 0) 3395 M_SETFIB(m, rtableid); 3396 3397 #ifdef ALTQ 3398 if (r != NULL && r->qid) { 3399 pf_mtag->qid = r->qid; 3400 3401 /* add hints for ecn */ 3402 pf_mtag->hdr = mtod(m, struct ip *); 3403 } 3404 #endif /* ALTQ */ 3405 m->m_data += max_linkhdr; 3406 m->m_pkthdr.len = m->m_len = len; 3407 /* The rest of the stack assumes a rcvif, so provide one. 3408 * This is a locally generated packet, so .. close enough. */ 3409 m->m_pkthdr.rcvif = V_loif; 3410 bzero(m->m_data, len); 3411 switch (af) { 3412 #ifdef INET 3413 case AF_INET: 3414 h = mtod(m, struct ip *); 3415 3416 /* IP header fields included in the TCP checksum */ 3417 h->ip_p = IPPROTO_TCP; 3418 h->ip_len = htons(tlen); 3419 h->ip_src.s_addr = saddr->v4.s_addr; 3420 h->ip_dst.s_addr = daddr->v4.s_addr; 3421 3422 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3423 break; 3424 #endif /* INET */ 3425 #ifdef INET6 3426 case AF_INET6: 3427 h6 = mtod(m, struct ip6_hdr *); 3428 3429 /* IP header fields included in the TCP checksum */ 3430 h6->ip6_nxt = IPPROTO_TCP; 3431 h6->ip6_plen = htons(tlen); 3432 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3433 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3434 3435 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3436 break; 3437 #endif /* INET6 */ 3438 } 3439 3440 /* TCP header */ 3441 th->th_sport = sport; 3442 th->th_dport = dport; 3443 th->th_seq = htonl(seq); 3444 th->th_ack = htonl(ack); 3445 th->th_off = tlen >> 2; 3446 th->th_flags = tcp_flags; 3447 th->th_win = htons(win); 3448 3449 if (mss) { 3450 opt = (char *)(th + 1); 3451 opt[0] = TCPOPT_MAXSEG; 3452 opt[1] = 4; 3453 HTONS(mss); 3454 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3455 } 3456 3457 switch (af) { 3458 #ifdef INET 3459 case AF_INET: 3460 /* TCP checksum */ 3461 th->th_sum = in_cksum(m, len); 3462 3463 /* Finish the IP header */ 3464 h->ip_v = 4; 3465 h->ip_hl = sizeof(*h) >> 2; 3466 h->ip_tos = IPTOS_LOWDELAY; 3467 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3468 h->ip_len = htons(len); 3469 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3470 h->ip_sum = 0; 3471 break; 3472 #endif /* INET */ 3473 #ifdef INET6 3474 case AF_INET6: 3475 /* TCP checksum */ 3476 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3477 sizeof(struct ip6_hdr), tlen); 3478 3479 h6->ip6_vfc |= IPV6_VERSION; 3480 h6->ip6_hlim = IPV6_DEFHLIM; 3481 break; 3482 #endif /* INET6 */ 3483 } 3484 3485 return (m); 3486 } 3487 3488 static void 3489 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3490 uint8_t ttl, int rtableid) 3491 { 3492 struct mbuf *m; 3493 #ifdef INET 3494 struct ip *h = NULL; 3495 #endif /* INET */ 3496 #ifdef INET6 3497 struct ip6_hdr *h6 = NULL; 3498 #endif /* INET6 */ 3499 struct sctphdr *hdr; 3500 struct sctp_chunkhdr *chunk; 3501 struct pf_send_entry *pfse; 3502 int off = 0; 3503 3504 MPASS(af == pd->af); 3505 3506 m = m_gethdr(M_NOWAIT, MT_DATA); 3507 if (m == NULL) 3508 return; 3509 3510 m->m_data += max_linkhdr; 3511 m->m_flags |= M_SKIP_FIREWALL; 3512 /* The rest of the stack assumes a rcvif, so provide one. 3513 * This is a locally generated packet, so .. close enough. */ 3514 m->m_pkthdr.rcvif = V_loif; 3515 3516 /* IPv4|6 header */ 3517 switch (af) { 3518 #ifdef INET 3519 case AF_INET: 3520 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3521 3522 h = mtod(m, struct ip *); 3523 3524 /* IP header fields included in the TCP checksum */ 3525 3526 h->ip_p = IPPROTO_SCTP; 3527 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3528 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3529 h->ip_src = pd->dst->v4; 3530 h->ip_dst = pd->src->v4; 3531 3532 off += sizeof(struct ip); 3533 break; 3534 #endif /* INET */ 3535 #ifdef INET6 3536 case AF_INET6: 3537 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3538 3539 h6 = mtod(m, struct ip6_hdr *); 3540 3541 /* IP header fields included in the TCP checksum */ 3542 h6->ip6_vfc |= IPV6_VERSION; 3543 h6->ip6_nxt = IPPROTO_SCTP; 3544 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3545 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3546 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3547 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3548 3549 off += sizeof(struct ip6_hdr); 3550 break; 3551 #endif /* INET6 */ 3552 } 3553 3554 /* SCTP header */ 3555 hdr = mtodo(m, off); 3556 3557 hdr->src_port = pd->hdr.sctp.dest_port; 3558 hdr->dest_port = pd->hdr.sctp.src_port; 3559 hdr->v_tag = pd->sctp_initiate_tag; 3560 hdr->checksum = 0; 3561 3562 /* Abort chunk. */ 3563 off += sizeof(struct sctphdr); 3564 chunk = mtodo(m, off); 3565 3566 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3567 chunk->chunk_length = htons(sizeof(*chunk)); 3568 3569 /* SCTP checksum */ 3570 off += sizeof(*chunk); 3571 m->m_pkthdr.len = m->m_len = off; 3572 3573 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3574 3575 if (rtableid >= 0) 3576 M_SETFIB(m, rtableid); 3577 3578 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3579 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3580 if (pfse == NULL) { 3581 m_freem(m); 3582 return; 3583 } 3584 3585 switch (af) { 3586 #ifdef INET 3587 case AF_INET: 3588 pfse->pfse_type = PFSE_IP; 3589 break; 3590 #endif /* INET */ 3591 #ifdef INET6 3592 case AF_INET6: 3593 pfse->pfse_type = PFSE_IP6; 3594 break; 3595 #endif /* INET6 */ 3596 } 3597 3598 pfse->pfse_m = m; 3599 pf_send(pfse); 3600 } 3601 3602 void 3603 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3604 const struct pf_addr *saddr, const struct pf_addr *daddr, 3605 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3606 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3607 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3608 { 3609 struct pf_send_entry *pfse; 3610 struct mbuf *m; 3611 3612 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3613 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3614 if (m == NULL) 3615 return; 3616 3617 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3618 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3619 if (pfse == NULL) { 3620 m_freem(m); 3621 return; 3622 } 3623 3624 switch (af) { 3625 #ifdef INET 3626 case AF_INET: 3627 pfse->pfse_type = PFSE_IP; 3628 break; 3629 #endif /* INET */ 3630 #ifdef INET6 3631 case AF_INET6: 3632 pfse->pfse_type = PFSE_IP6; 3633 break; 3634 #endif /* INET6 */ 3635 } 3636 3637 pfse->pfse_m = m; 3638 pf_send(pfse); 3639 } 3640 3641 static void 3642 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3643 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3644 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3645 u_short *reason, int rtableid) 3646 { 3647 struct pf_addr * const saddr = pd->src; 3648 struct pf_addr * const daddr = pd->dst; 3649 sa_family_t af = pd->af; 3650 3651 /* undo NAT changes, if they have taken place */ 3652 if (nr != NULL) { 3653 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3654 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3655 if (pd->sport) 3656 *pd->sport = sk->port[pd->sidx]; 3657 if (pd->dport) 3658 *pd->dport = sk->port[pd->didx]; 3659 if (pd->proto_sum) 3660 *pd->proto_sum = bproto_sum; 3661 if (pd->ip_sum) 3662 *pd->ip_sum = bip_sum; 3663 m_copyback(m, off, hdrlen, pd->hdr.any); 3664 } 3665 if (pd->proto == IPPROTO_TCP && 3666 ((r->rule_flag & PFRULE_RETURNRST) || 3667 (r->rule_flag & PFRULE_RETURN)) && 3668 !(th->th_flags & TH_RST)) { 3669 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3670 int len = 0; 3671 #ifdef INET 3672 struct ip *h4; 3673 #endif 3674 #ifdef INET6 3675 struct ip6_hdr *h6; 3676 #endif 3677 3678 switch (af) { 3679 #ifdef INET 3680 case AF_INET: 3681 h4 = mtod(m, struct ip *); 3682 len = ntohs(h4->ip_len) - off; 3683 break; 3684 #endif 3685 #ifdef INET6 3686 case AF_INET6: 3687 h6 = mtod(m, struct ip6_hdr *); 3688 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3689 break; 3690 #endif 3691 } 3692 3693 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3694 REASON_SET(reason, PFRES_PROTCKSUM); 3695 else { 3696 if (th->th_flags & TH_SYN) 3697 ack++; 3698 if (th->th_flags & TH_FIN) 3699 ack++; 3700 pf_send_tcp(r, af, pd->dst, 3701 pd->src, th->th_dport, th->th_sport, 3702 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3703 r->return_ttl, true, 0, 0, rtableid); 3704 } 3705 } else if (pd->proto == IPPROTO_SCTP && 3706 (r->rule_flag & PFRULE_RETURN)) { 3707 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3708 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3709 r->return_icmp) 3710 pf_send_icmp(m, r->return_icmp >> 8, 3711 r->return_icmp & 255, af, r, rtableid); 3712 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3713 r->return_icmp6) 3714 pf_send_icmp(m, r->return_icmp6 >> 8, 3715 r->return_icmp6 & 255, af, r, rtableid); 3716 } 3717 3718 static int 3719 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3720 { 3721 struct m_tag *mtag; 3722 u_int8_t mpcp; 3723 3724 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3725 if (mtag == NULL) 3726 return (0); 3727 3728 if (prio == PF_PRIO_ZERO) 3729 prio = 0; 3730 3731 mpcp = *(uint8_t *)(mtag + 1); 3732 3733 return (mpcp == prio); 3734 } 3735 3736 static int 3737 pf_icmp_to_bandlim(uint8_t type) 3738 { 3739 switch (type) { 3740 case ICMP_ECHO: 3741 case ICMP_ECHOREPLY: 3742 return (BANDLIM_ICMP_ECHO); 3743 case ICMP_TSTAMP: 3744 case ICMP_TSTAMPREPLY: 3745 return (BANDLIM_ICMP_TSTAMP); 3746 case ICMP_UNREACH: 3747 default: 3748 return (BANDLIM_ICMP_UNREACH); 3749 } 3750 } 3751 3752 static void 3753 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3754 struct pf_krule *r, int rtableid) 3755 { 3756 struct pf_send_entry *pfse; 3757 struct mbuf *m0; 3758 struct pf_mtag *pf_mtag; 3759 3760 /* ICMP packet rate limitation. */ 3761 #ifdef INET6 3762 if (af == AF_INET6) { 3763 if (icmp6_ratelimit(NULL, type, code)) 3764 return; 3765 } 3766 #endif 3767 #ifdef INET 3768 if (af == AF_INET) { 3769 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3770 return; 3771 } 3772 #endif 3773 3774 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3775 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3776 if (pfse == NULL) 3777 return; 3778 3779 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3780 free(pfse, M_PFTEMP); 3781 return; 3782 } 3783 3784 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3785 free(pfse, M_PFTEMP); 3786 return; 3787 } 3788 /* XXX: revisit */ 3789 m0->m_flags |= M_SKIP_FIREWALL; 3790 3791 if (rtableid >= 0) 3792 M_SETFIB(m0, rtableid); 3793 3794 #ifdef ALTQ 3795 if (r->qid) { 3796 pf_mtag->qid = r->qid; 3797 /* add hints for ecn */ 3798 pf_mtag->hdr = mtod(m0, struct ip *); 3799 } 3800 #endif /* ALTQ */ 3801 3802 switch (af) { 3803 #ifdef INET 3804 case AF_INET: 3805 pfse->pfse_type = PFSE_ICMP; 3806 break; 3807 #endif /* INET */ 3808 #ifdef INET6 3809 case AF_INET6: 3810 pfse->pfse_type = PFSE_ICMP6; 3811 break; 3812 #endif /* INET6 */ 3813 } 3814 pfse->pfse_m = m0; 3815 pfse->icmpopts.type = type; 3816 pfse->icmpopts.code = code; 3817 pf_send(pfse); 3818 } 3819 3820 /* 3821 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3822 * If n is 0, they match if they are equal. If n is != 0, they match if they 3823 * are different. 3824 */ 3825 int 3826 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3827 struct pf_addr *b, sa_family_t af) 3828 { 3829 int match = 0; 3830 3831 switch (af) { 3832 #ifdef INET 3833 case AF_INET: 3834 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3835 match++; 3836 break; 3837 #endif /* INET */ 3838 #ifdef INET6 3839 case AF_INET6: 3840 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3841 match++; 3842 break; 3843 #endif /* INET6 */ 3844 } 3845 if (match) { 3846 if (n) 3847 return (0); 3848 else 3849 return (1); 3850 } else { 3851 if (n) 3852 return (1); 3853 else 3854 return (0); 3855 } 3856 } 3857 3858 /* 3859 * Return 1 if b <= a <= e, otherwise return 0. 3860 */ 3861 int 3862 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3863 struct pf_addr *a, sa_family_t af) 3864 { 3865 switch (af) { 3866 #ifdef INET 3867 case AF_INET: 3868 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3869 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3870 return (0); 3871 break; 3872 #endif /* INET */ 3873 #ifdef INET6 3874 case AF_INET6: { 3875 int i; 3876 3877 /* check a >= b */ 3878 for (i = 0; i < 4; ++i) 3879 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3880 break; 3881 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3882 return (0); 3883 /* check a <= e */ 3884 for (i = 0; i < 4; ++i) 3885 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3886 break; 3887 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3888 return (0); 3889 break; 3890 } 3891 #endif /* INET6 */ 3892 } 3893 return (1); 3894 } 3895 3896 static int 3897 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3898 { 3899 switch (op) { 3900 case PF_OP_IRG: 3901 return ((p > a1) && (p < a2)); 3902 case PF_OP_XRG: 3903 return ((p < a1) || (p > a2)); 3904 case PF_OP_RRG: 3905 return ((p >= a1) && (p <= a2)); 3906 case PF_OP_EQ: 3907 return (p == a1); 3908 case PF_OP_NE: 3909 return (p != a1); 3910 case PF_OP_LT: 3911 return (p < a1); 3912 case PF_OP_LE: 3913 return (p <= a1); 3914 case PF_OP_GT: 3915 return (p > a1); 3916 case PF_OP_GE: 3917 return (p >= a1); 3918 } 3919 return (0); /* never reached */ 3920 } 3921 3922 int 3923 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3924 { 3925 NTOHS(a1); 3926 NTOHS(a2); 3927 NTOHS(p); 3928 return (pf_match(op, a1, a2, p)); 3929 } 3930 3931 static int 3932 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3933 { 3934 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3935 return (0); 3936 return (pf_match(op, a1, a2, u)); 3937 } 3938 3939 static int 3940 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3941 { 3942 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3943 return (0); 3944 return (pf_match(op, a1, a2, g)); 3945 } 3946 3947 int 3948 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3949 { 3950 if (*tag == -1) 3951 *tag = mtag; 3952 3953 return ((!r->match_tag_not && r->match_tag == *tag) || 3954 (r->match_tag_not && r->match_tag != *tag)); 3955 } 3956 3957 static int 3958 pf_match_rcvif(struct mbuf *m, struct pf_krule *r) 3959 { 3960 struct ifnet *ifp = m->m_pkthdr.rcvif; 3961 struct pfi_kkif *kif; 3962 3963 if (ifp == NULL) 3964 return (0); 3965 3966 kif = (struct pfi_kkif *)ifp->if_pf_kif; 3967 3968 if (kif == NULL) { 3969 DPFPRINTF(PF_DEBUG_URGENT, 3970 ("pf_test_via: kif == NULL, @%d via %s\n", r->nr, 3971 r->rcv_ifname)); 3972 return (0); 3973 } 3974 3975 return (pfi_kkif_match(r->rcv_kif, kif)); 3976 } 3977 3978 int 3979 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3980 { 3981 3982 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3983 3984 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3985 return (ENOMEM); 3986 3987 pd->pf_mtag->tag = tag; 3988 3989 return (0); 3990 } 3991 3992 #define PF_ANCHOR_STACKSIZE 32 3993 struct pf_kanchor_stackframe { 3994 struct pf_kruleset *rs; 3995 struct pf_krule *r; /* XXX: + match bit */ 3996 struct pf_kanchor *child; 3997 }; 3998 3999 /* 4000 * XXX: We rely on malloc(9) returning pointer aligned addresses. 4001 */ 4002 #define PF_ANCHORSTACK_MATCH 0x00000001 4003 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 4004 4005 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4006 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 4007 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4008 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4009 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4010 } while (0) 4011 4012 void 4013 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4014 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4015 int *match) 4016 { 4017 struct pf_kanchor_stackframe *f; 4018 4019 PF_RULES_RASSERT(); 4020 4021 if (match) 4022 *match = 0; 4023 if (*depth >= PF_ANCHOR_STACKSIZE) { 4024 printf("%s: anchor stack overflow on %s\n", 4025 __func__, (*r)->anchor->name); 4026 *r = TAILQ_NEXT(*r, entries); 4027 return; 4028 } else if (*depth == 0 && a != NULL) 4029 *a = *r; 4030 f = stack + (*depth)++; 4031 f->rs = *rs; 4032 f->r = *r; 4033 if ((*r)->anchor_wildcard) { 4034 struct pf_kanchor_node *parent = &(*r)->anchor->children; 4035 4036 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 4037 *r = NULL; 4038 return; 4039 } 4040 *rs = &f->child->ruleset; 4041 } else { 4042 f->child = NULL; 4043 *rs = &(*r)->anchor->ruleset; 4044 } 4045 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4046 } 4047 4048 int 4049 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 4050 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 4051 int *match) 4052 { 4053 struct pf_kanchor_stackframe *f; 4054 struct pf_krule *fr; 4055 int quick = 0; 4056 4057 PF_RULES_RASSERT(); 4058 4059 do { 4060 if (*depth <= 0) 4061 break; 4062 f = stack + *depth - 1; 4063 fr = PF_ANCHOR_RULE(f); 4064 if (f->child != NULL) { 4065 /* 4066 * This block traverses through 4067 * a wildcard anchor. 4068 */ 4069 if (match != NULL && *match) { 4070 /* 4071 * If any of "*" matched, then 4072 * "foo/ *" matched, mark frame 4073 * appropriately. 4074 */ 4075 PF_ANCHOR_SET_MATCH(f); 4076 *match = 0; 4077 } 4078 f->child = RB_NEXT(pf_kanchor_node, 4079 &fr->anchor->children, f->child); 4080 if (f->child != NULL) { 4081 *rs = &f->child->ruleset; 4082 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 4083 if (*r == NULL) 4084 continue; 4085 else 4086 break; 4087 } 4088 } 4089 (*depth)--; 4090 if (*depth == 0 && a != NULL) 4091 *a = NULL; 4092 *rs = f->rs; 4093 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 4094 quick = fr->quick; 4095 *r = TAILQ_NEXT(fr, entries); 4096 } while (*r == NULL); 4097 4098 return (quick); 4099 } 4100 4101 struct pf_keth_anchor_stackframe { 4102 struct pf_keth_ruleset *rs; 4103 struct pf_keth_rule *r; /* XXX: + match bit */ 4104 struct pf_keth_anchor *child; 4105 }; 4106 4107 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 4108 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 4109 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 4110 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 4111 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 4112 } while (0) 4113 4114 void 4115 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4116 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4117 struct pf_keth_rule **a, int *match) 4118 { 4119 struct pf_keth_anchor_stackframe *f; 4120 4121 NET_EPOCH_ASSERT(); 4122 4123 if (match) 4124 *match = 0; 4125 if (*depth >= PF_ANCHOR_STACKSIZE) { 4126 printf("%s: anchor stack overflow on %s\n", 4127 __func__, (*r)->anchor->name); 4128 *r = TAILQ_NEXT(*r, entries); 4129 return; 4130 } else if (*depth == 0 && a != NULL) 4131 *a = *r; 4132 f = stack + (*depth)++; 4133 f->rs = *rs; 4134 f->r = *r; 4135 if ((*r)->anchor_wildcard) { 4136 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 4137 4138 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 4139 *r = NULL; 4140 return; 4141 } 4142 *rs = &f->child->ruleset; 4143 } else { 4144 f->child = NULL; 4145 *rs = &(*r)->anchor->ruleset; 4146 } 4147 *r = TAILQ_FIRST((*rs)->active.rules); 4148 } 4149 4150 int 4151 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 4152 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 4153 struct pf_keth_rule **a, int *match) 4154 { 4155 struct pf_keth_anchor_stackframe *f; 4156 struct pf_keth_rule *fr; 4157 int quick = 0; 4158 4159 NET_EPOCH_ASSERT(); 4160 4161 do { 4162 if (*depth <= 0) 4163 break; 4164 f = stack + *depth - 1; 4165 fr = PF_ETH_ANCHOR_RULE(f); 4166 if (f->child != NULL) { 4167 /* 4168 * This block traverses through 4169 * a wildcard anchor. 4170 */ 4171 if (match != NULL && *match) { 4172 /* 4173 * If any of "*" matched, then 4174 * "foo/ *" matched, mark frame 4175 * appropriately. 4176 */ 4177 PF_ETH_ANCHOR_SET_MATCH(f); 4178 *match = 0; 4179 } 4180 f->child = RB_NEXT(pf_keth_anchor_node, 4181 &fr->anchor->children, f->child); 4182 if (f->child != NULL) { 4183 *rs = &f->child->ruleset; 4184 *r = TAILQ_FIRST((*rs)->active.rules); 4185 if (*r == NULL) 4186 continue; 4187 else 4188 break; 4189 } 4190 } 4191 (*depth)--; 4192 if (*depth == 0 && a != NULL) 4193 *a = NULL; 4194 *rs = f->rs; 4195 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 4196 quick = fr->quick; 4197 *r = TAILQ_NEXT(fr, entries); 4198 } while (*r == NULL); 4199 4200 return (quick); 4201 } 4202 4203 #ifdef INET6 4204 void 4205 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 4206 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 4207 { 4208 switch (af) { 4209 #ifdef INET 4210 case AF_INET: 4211 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4212 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4213 break; 4214 #endif /* INET */ 4215 case AF_INET6: 4216 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 4217 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 4218 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 4219 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 4220 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 4221 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 4222 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 4223 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 4224 break; 4225 } 4226 } 4227 4228 void 4229 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 4230 { 4231 switch (af) { 4232 #ifdef INET 4233 case AF_INET: 4234 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 4235 break; 4236 #endif /* INET */ 4237 case AF_INET6: 4238 if (addr->addr32[3] == 0xffffffff) { 4239 addr->addr32[3] = 0; 4240 if (addr->addr32[2] == 0xffffffff) { 4241 addr->addr32[2] = 0; 4242 if (addr->addr32[1] == 0xffffffff) { 4243 addr->addr32[1] = 0; 4244 addr->addr32[0] = 4245 htonl(ntohl(addr->addr32[0]) + 1); 4246 } else 4247 addr->addr32[1] = 4248 htonl(ntohl(addr->addr32[1]) + 1); 4249 } else 4250 addr->addr32[2] = 4251 htonl(ntohl(addr->addr32[2]) + 1); 4252 } else 4253 addr->addr32[3] = 4254 htonl(ntohl(addr->addr32[3]) + 1); 4255 break; 4256 } 4257 } 4258 #endif /* INET6 */ 4259 4260 void 4261 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 4262 { 4263 /* 4264 * Modern rules use the same flags in rules as they do in states. 4265 */ 4266 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 4267 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 4268 4269 /* 4270 * Old-style scrub rules have different flags which need to be translated. 4271 */ 4272 if (r->rule_flag & PFRULE_RANDOMID) 4273 a->flags |= PFSTATE_RANDOMID; 4274 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 4275 a->flags |= PFSTATE_SETTOS; 4276 a->set_tos = r->set_tos; 4277 } 4278 4279 if (r->qid) 4280 a->qid = r->qid; 4281 if (r->pqid) 4282 a->pqid = r->pqid; 4283 if (r->rtableid >= 0) 4284 a->rtableid = r->rtableid; 4285 a->log |= r->log; 4286 if (r->min_ttl) 4287 a->min_ttl = r->min_ttl; 4288 if (r->max_mss) 4289 a->max_mss = r->max_mss; 4290 if (r->dnpipe) 4291 a->dnpipe = r->dnpipe; 4292 if (r->dnrpipe) 4293 a->dnrpipe = r->dnrpipe; 4294 if (r->dnpipe || r->dnrpipe) { 4295 if (r->free_flags & PFRULE_DN_IS_PIPE) 4296 a->flags |= PFSTATE_DN_IS_PIPE; 4297 else 4298 a->flags &= ~PFSTATE_DN_IS_PIPE; 4299 } 4300 if (r->scrub_flags & PFSTATE_SETPRIO) { 4301 a->set_prio[0] = r->set_prio[0]; 4302 a->set_prio[1] = r->set_prio[1]; 4303 } 4304 } 4305 4306 int 4307 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 4308 { 4309 struct pf_addr *saddr, *daddr; 4310 u_int16_t sport, dport; 4311 struct inpcbinfo *pi; 4312 struct inpcb *inp; 4313 4314 pd->lookup.uid = UID_MAX; 4315 pd->lookup.gid = GID_MAX; 4316 4317 switch (pd->proto) { 4318 case IPPROTO_TCP: 4319 sport = pd->hdr.tcp.th_sport; 4320 dport = pd->hdr.tcp.th_dport; 4321 pi = &V_tcbinfo; 4322 break; 4323 case IPPROTO_UDP: 4324 sport = pd->hdr.udp.uh_sport; 4325 dport = pd->hdr.udp.uh_dport; 4326 pi = &V_udbinfo; 4327 break; 4328 default: 4329 return (-1); 4330 } 4331 if (pd->dir == PF_IN) { 4332 saddr = pd->src; 4333 daddr = pd->dst; 4334 } else { 4335 u_int16_t p; 4336 4337 p = sport; 4338 sport = dport; 4339 dport = p; 4340 saddr = pd->dst; 4341 daddr = pd->src; 4342 } 4343 switch (pd->af) { 4344 #ifdef INET 4345 case AF_INET: 4346 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 4347 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4348 if (inp == NULL) { 4349 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 4350 daddr->v4, dport, INPLOOKUP_WILDCARD | 4351 INPLOOKUP_RLOCKPCB, NULL, m); 4352 if (inp == NULL) 4353 return (-1); 4354 } 4355 break; 4356 #endif /* INET */ 4357 #ifdef INET6 4358 case AF_INET6: 4359 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 4360 dport, INPLOOKUP_RLOCKPCB, NULL, m); 4361 if (inp == NULL) { 4362 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 4363 &daddr->v6, dport, INPLOOKUP_WILDCARD | 4364 INPLOOKUP_RLOCKPCB, NULL, m); 4365 if (inp == NULL) 4366 return (-1); 4367 } 4368 break; 4369 #endif /* INET6 */ 4370 4371 default: 4372 return (-1); 4373 } 4374 INP_RLOCK_ASSERT(inp); 4375 pd->lookup.uid = inp->inp_cred->cr_uid; 4376 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4377 INP_RUNLOCK(inp); 4378 4379 return (1); 4380 } 4381 4382 u_int8_t 4383 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4384 { 4385 int hlen; 4386 u_int8_t hdr[60]; 4387 u_int8_t *opt, optlen; 4388 u_int8_t wscale = 0; 4389 4390 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4391 if (hlen <= sizeof(struct tcphdr)) 4392 return (0); 4393 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4394 return (0); 4395 opt = hdr + sizeof(struct tcphdr); 4396 hlen -= sizeof(struct tcphdr); 4397 while (hlen >= 3) { 4398 switch (*opt) { 4399 case TCPOPT_EOL: 4400 case TCPOPT_NOP: 4401 ++opt; 4402 --hlen; 4403 break; 4404 case TCPOPT_WINDOW: 4405 wscale = opt[2]; 4406 if (wscale > TCP_MAX_WINSHIFT) 4407 wscale = TCP_MAX_WINSHIFT; 4408 wscale |= PF_WSCALE_FLAG; 4409 /* FALLTHROUGH */ 4410 default: 4411 optlen = opt[1]; 4412 if (optlen < 2) 4413 optlen = 2; 4414 hlen -= optlen; 4415 opt += optlen; 4416 break; 4417 } 4418 } 4419 return (wscale); 4420 } 4421 4422 u_int16_t 4423 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 4424 { 4425 int hlen; 4426 u_int8_t hdr[60]; 4427 u_int8_t *opt, optlen; 4428 u_int16_t mss = V_tcp_mssdflt; 4429 4430 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 4431 if (hlen <= sizeof(struct tcphdr)) 4432 return (0); 4433 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 4434 return (0); 4435 opt = hdr + sizeof(struct tcphdr); 4436 hlen -= sizeof(struct tcphdr); 4437 while (hlen >= TCPOLEN_MAXSEG) { 4438 switch (*opt) { 4439 case TCPOPT_EOL: 4440 case TCPOPT_NOP: 4441 ++opt; 4442 --hlen; 4443 break; 4444 case TCPOPT_MAXSEG: 4445 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4446 NTOHS(mss); 4447 /* FALLTHROUGH */ 4448 default: 4449 optlen = opt[1]; 4450 if (optlen < 2) 4451 optlen = 2; 4452 hlen -= optlen; 4453 opt += optlen; 4454 break; 4455 } 4456 } 4457 return (mss); 4458 } 4459 4460 static u_int16_t 4461 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4462 { 4463 struct nhop_object *nh; 4464 #ifdef INET6 4465 struct in6_addr dst6; 4466 uint32_t scopeid; 4467 #endif /* INET6 */ 4468 int hlen = 0; 4469 uint16_t mss = 0; 4470 4471 NET_EPOCH_ASSERT(); 4472 4473 switch (af) { 4474 #ifdef INET 4475 case AF_INET: 4476 hlen = sizeof(struct ip); 4477 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4478 if (nh != NULL) 4479 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4480 break; 4481 #endif /* INET */ 4482 #ifdef INET6 4483 case AF_INET6: 4484 hlen = sizeof(struct ip6_hdr); 4485 in6_splitscope(&addr->v6, &dst6, &scopeid); 4486 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4487 if (nh != NULL) 4488 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4489 break; 4490 #endif /* INET6 */ 4491 } 4492 4493 mss = max(V_tcp_mssdflt, mss); 4494 mss = min(mss, offer); 4495 mss = max(mss, 64); /* sanity - at least max opt space */ 4496 return (mss); 4497 } 4498 4499 static u_int32_t 4500 pf_tcp_iss(struct pf_pdesc *pd) 4501 { 4502 MD5_CTX ctx; 4503 u_int32_t digest[4]; 4504 4505 if (V_pf_tcp_secret_init == 0) { 4506 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4507 MD5Init(&V_pf_tcp_secret_ctx); 4508 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4509 sizeof(V_pf_tcp_secret)); 4510 V_pf_tcp_secret_init = 1; 4511 } 4512 4513 ctx = V_pf_tcp_secret_ctx; 4514 4515 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4516 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4517 if (pd->af == AF_INET6) { 4518 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4519 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4520 } else { 4521 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4522 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4523 } 4524 MD5Final((u_char *)digest, &ctx); 4525 V_pf_tcp_iss_off += 4096; 4526 #define ISN_RANDOM_INCREMENT (4096 - 1) 4527 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4528 V_pf_tcp_iss_off); 4529 #undef ISN_RANDOM_INCREMENT 4530 } 4531 4532 static bool 4533 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4534 { 4535 bool match = true; 4536 4537 /* Always matches if not set */ 4538 if (! r->isset) 4539 return (!r->neg); 4540 4541 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4542 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4543 match = false; 4544 break; 4545 } 4546 } 4547 4548 return (match ^ r->neg); 4549 } 4550 4551 static int 4552 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4553 { 4554 if (*tag == -1) 4555 *tag = mtag; 4556 4557 return ((!r->match_tag_not && r->match_tag == *tag) || 4558 (r->match_tag_not && r->match_tag != *tag)); 4559 } 4560 4561 static void 4562 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4563 { 4564 /* If we don't have the interface drop the packet. */ 4565 if (ifp == NULL) { 4566 m_freem(m); 4567 return; 4568 } 4569 4570 switch (ifp->if_type) { 4571 case IFT_ETHER: 4572 case IFT_XETHER: 4573 case IFT_L2VLAN: 4574 case IFT_BRIDGE: 4575 case IFT_IEEE8023ADLAG: 4576 break; 4577 default: 4578 m_freem(m); 4579 return; 4580 } 4581 4582 ifp->if_transmit(ifp, m); 4583 } 4584 4585 static int 4586 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4587 { 4588 #ifdef INET 4589 struct ip ip; 4590 #endif 4591 #ifdef INET6 4592 struct ip6_hdr ip6; 4593 #endif 4594 struct mbuf *m = *m0; 4595 struct ether_header *e; 4596 struct pf_keth_rule *r, *rm, *a = NULL; 4597 struct pf_keth_ruleset *ruleset = NULL; 4598 struct pf_mtag *mtag; 4599 struct pf_keth_ruleq *rules; 4600 struct pf_addr *src = NULL, *dst = NULL; 4601 struct pfi_kkif *bridge_to; 4602 sa_family_t af = 0; 4603 uint16_t proto; 4604 int asd = 0, match = 0; 4605 int tag = -1; 4606 uint8_t action; 4607 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4608 4609 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4610 NET_EPOCH_ASSERT(); 4611 4612 PF_RULES_RLOCK_TRACKER; 4613 4614 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4615 4616 mtag = pf_find_mtag(m); 4617 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4618 /* Dummynet re-injects packets after they've 4619 * completed their delay. We've already 4620 * processed them, so pass unconditionally. */ 4621 4622 /* But only once. We may see the packet multiple times (e.g. 4623 * PFIL_IN/PFIL_OUT). */ 4624 pf_dummynet_flag_remove(m, mtag); 4625 4626 return (PF_PASS); 4627 } 4628 4629 ruleset = V_pf_keth; 4630 rules = ck_pr_load_ptr(&ruleset->active.rules); 4631 r = TAILQ_FIRST(rules); 4632 rm = NULL; 4633 4634 if (__predict_false(m->m_len < sizeof(struct ether_header)) && 4635 (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) { 4636 DPFPRINTF(PF_DEBUG_URGENT, 4637 ("pf_test_eth_rule: m_len < sizeof(struct ether_header)" 4638 ", pullup failed\n")); 4639 return (PF_DROP); 4640 } 4641 e = mtod(m, struct ether_header *); 4642 proto = ntohs(e->ether_type); 4643 4644 switch (proto) { 4645 #ifdef INET 4646 case ETHERTYPE_IP: { 4647 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4648 sizeof(ip))) 4649 return (PF_DROP); 4650 4651 af = AF_INET; 4652 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4653 (caddr_t)&ip); 4654 src = (struct pf_addr *)&ip.ip_src; 4655 dst = (struct pf_addr *)&ip.ip_dst; 4656 break; 4657 } 4658 #endif /* INET */ 4659 #ifdef INET6 4660 case ETHERTYPE_IPV6: { 4661 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4662 sizeof(ip6))) 4663 return (PF_DROP); 4664 4665 af = AF_INET6; 4666 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4667 (caddr_t)&ip6); 4668 src = (struct pf_addr *)&ip6.ip6_src; 4669 dst = (struct pf_addr *)&ip6.ip6_dst; 4670 break; 4671 } 4672 #endif /* INET6 */ 4673 } 4674 4675 PF_RULES_RLOCK(); 4676 4677 while (r != NULL) { 4678 counter_u64_add(r->evaluations, 1); 4679 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4680 4681 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4682 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4683 "kif"); 4684 r = r->skip[PFE_SKIP_IFP].ptr; 4685 } 4686 else if (r->direction && r->direction != dir) { 4687 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4688 "dir"); 4689 r = r->skip[PFE_SKIP_DIR].ptr; 4690 } 4691 else if (r->proto && r->proto != proto) { 4692 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4693 "proto"); 4694 r = r->skip[PFE_SKIP_PROTO].ptr; 4695 } 4696 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4697 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4698 "src"); 4699 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4700 } 4701 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4702 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4703 "dst"); 4704 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4705 } 4706 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4707 r->ipsrc.neg, kif, M_GETFIB(m))) { 4708 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4709 "ip_src"); 4710 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4711 } 4712 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4713 r->ipdst.neg, kif, M_GETFIB(m))) { 4714 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4715 "ip_dst"); 4716 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4717 } 4718 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4719 mtag ? mtag->tag : 0)) { 4720 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4721 "match_tag"); 4722 r = TAILQ_NEXT(r, entries); 4723 } 4724 else { 4725 if (r->tag) 4726 tag = r->tag; 4727 if (r->anchor == NULL) { 4728 /* Rule matches */ 4729 rm = r; 4730 4731 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4732 4733 if (r->quick) 4734 break; 4735 4736 r = TAILQ_NEXT(r, entries); 4737 } else { 4738 pf_step_into_keth_anchor(anchor_stack, &asd, 4739 &ruleset, &r, &a, &match); 4740 } 4741 } 4742 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4743 &ruleset, &r, &a, &match)) 4744 break; 4745 } 4746 4747 r = rm; 4748 4749 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4750 4751 /* Default to pass. */ 4752 if (r == NULL) { 4753 PF_RULES_RUNLOCK(); 4754 return (PF_PASS); 4755 } 4756 4757 /* Execute action. */ 4758 counter_u64_add(r->packets[dir == PF_OUT], 1); 4759 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4760 pf_update_timestamp(r); 4761 4762 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4763 if (r->action == PF_DROP) { 4764 PF_RULES_RUNLOCK(); 4765 return (PF_DROP); 4766 } 4767 4768 if (tag > 0) { 4769 if (mtag == NULL) 4770 mtag = pf_get_mtag(m); 4771 if (mtag == NULL) { 4772 PF_RULES_RUNLOCK(); 4773 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4774 return (PF_DROP); 4775 } 4776 mtag->tag = tag; 4777 } 4778 4779 if (r->qid != 0) { 4780 if (mtag == NULL) 4781 mtag = pf_get_mtag(m); 4782 if (mtag == NULL) { 4783 PF_RULES_RUNLOCK(); 4784 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4785 return (PF_DROP); 4786 } 4787 mtag->qid = r->qid; 4788 } 4789 4790 action = r->action; 4791 bridge_to = r->bridge_to; 4792 4793 /* Dummynet */ 4794 if (r->dnpipe) { 4795 struct ip_fw_args dnflow; 4796 4797 /* Drop packet if dummynet is not loaded. */ 4798 if (ip_dn_io_ptr == NULL) { 4799 PF_RULES_RUNLOCK(); 4800 m_freem(m); 4801 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4802 return (PF_DROP); 4803 } 4804 if (mtag == NULL) 4805 mtag = pf_get_mtag(m); 4806 if (mtag == NULL) { 4807 PF_RULES_RUNLOCK(); 4808 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4809 return (PF_DROP); 4810 } 4811 4812 bzero(&dnflow, sizeof(dnflow)); 4813 4814 /* We don't have port numbers here, so we set 0. That means 4815 * that we'll be somewhat limited in distinguishing flows (i.e. 4816 * only based on IP addresses, not based on port numbers), but 4817 * it's better than nothing. */ 4818 dnflow.f_id.dst_port = 0; 4819 dnflow.f_id.src_port = 0; 4820 dnflow.f_id.proto = 0; 4821 4822 dnflow.rule.info = r->dnpipe; 4823 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4824 if (r->dnflags & PFRULE_DN_IS_PIPE) 4825 dnflow.rule.info |= IPFW_IS_PIPE; 4826 4827 dnflow.f_id.extra = dnflow.rule.info; 4828 4829 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4830 dnflow.flags |= IPFW_ARGS_ETHER; 4831 dnflow.ifp = kif->pfik_ifp; 4832 4833 switch (af) { 4834 case AF_INET: 4835 dnflow.f_id.addr_type = 4; 4836 dnflow.f_id.src_ip = src->v4.s_addr; 4837 dnflow.f_id.dst_ip = dst->v4.s_addr; 4838 break; 4839 case AF_INET6: 4840 dnflow.flags |= IPFW_ARGS_IP6; 4841 dnflow.f_id.addr_type = 6; 4842 dnflow.f_id.src_ip6 = src->v6; 4843 dnflow.f_id.dst_ip6 = dst->v6; 4844 break; 4845 } 4846 4847 PF_RULES_RUNLOCK(); 4848 4849 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4850 ip_dn_io_ptr(m0, &dnflow); 4851 if (*m0 != NULL) 4852 pf_dummynet_flag_remove(m, mtag); 4853 } else { 4854 PF_RULES_RUNLOCK(); 4855 } 4856 4857 if (action == PF_PASS && bridge_to) { 4858 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4859 *m0 = NULL; /* We've eaten the packet. */ 4860 } 4861 4862 return (action); 4863 } 4864 4865 static int 4866 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4867 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4868 struct pf_kruleset **rsm, struct inpcb *inp, int hdrlen) 4869 { 4870 struct pf_krule *nr = NULL; 4871 struct pf_addr * const saddr = pd->src; 4872 struct pf_addr * const daddr = pd->dst; 4873 sa_family_t af = pd->af; 4874 struct pf_krule *r, *a = NULL; 4875 struct pf_kruleset *ruleset = NULL; 4876 struct pf_krule_slist match_rules; 4877 struct pf_krule_item *ri; 4878 struct pf_ksrc_node *nsn = NULL; 4879 struct tcphdr *th = &pd->hdr.tcp; 4880 struct pf_state_key *sk = NULL, *nk = NULL; 4881 u_short reason, transerror; 4882 int rewrite = 0; 4883 int tag = -1; 4884 int asd = 0; 4885 int match = 0; 4886 int state_icmp = 0, icmp_dir, multi; 4887 u_int16_t sport = 0, dport = 0, virtual_type, virtual_id; 4888 u_int16_t bproto_sum = 0, bip_sum = 0; 4889 u_int8_t icmptype = 0, icmpcode = 0; 4890 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4891 struct pf_udp_mapping *udp_mapping = NULL; 4892 4893 PF_RULES_RASSERT(); 4894 4895 SLIST_INIT(&match_rules); 4896 4897 if (inp != NULL) { 4898 INP_LOCK_ASSERT(inp); 4899 pd->lookup.uid = inp->inp_cred->cr_uid; 4900 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4901 pd->lookup.done = 1; 4902 } 4903 4904 switch (pd->proto) { 4905 case IPPROTO_TCP: 4906 sport = th->th_sport; 4907 dport = th->th_dport; 4908 break; 4909 case IPPROTO_UDP: 4910 sport = pd->hdr.udp.uh_sport; 4911 dport = pd->hdr.udp.uh_dport; 4912 break; 4913 case IPPROTO_SCTP: 4914 sport = pd->hdr.sctp.src_port; 4915 dport = pd->hdr.sctp.dest_port; 4916 break; 4917 #ifdef INET 4918 case IPPROTO_ICMP: 4919 if (pd->af != AF_INET) 4920 break; 4921 icmptype = pd->hdr.icmp.icmp_type; 4922 icmpcode = pd->hdr.icmp.icmp_code; 4923 state_icmp = pf_icmp_mapping(pd, icmptype, 4924 &icmp_dir, &multi, &virtual_id, &virtual_type); 4925 if (icmp_dir == PF_IN) { 4926 sport = virtual_id; 4927 dport = virtual_type; 4928 } else { 4929 sport = virtual_type; 4930 dport = virtual_id; 4931 } 4932 break; 4933 #endif /* INET */ 4934 #ifdef INET6 4935 case IPPROTO_ICMPV6: 4936 if (af != AF_INET6) 4937 break; 4938 icmptype = pd->hdr.icmp6.icmp6_type; 4939 icmpcode = pd->hdr.icmp6.icmp6_code; 4940 state_icmp = pf_icmp_mapping(pd, icmptype, 4941 &icmp_dir, &multi, &virtual_id, &virtual_type); 4942 if (icmp_dir == PF_IN) { 4943 sport = virtual_id; 4944 dport = virtual_type; 4945 } else { 4946 sport = virtual_type; 4947 dport = virtual_id; 4948 } 4949 4950 break; 4951 #endif /* INET6 */ 4952 default: 4953 sport = dport = 0; 4954 break; 4955 } 4956 4957 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4958 4959 /* check packet for BINAT/NAT/RDR */ 4960 transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4961 &nk, saddr, daddr, sport, dport, anchor_stack, &nr, &udp_mapping); 4962 switch (transerror) { 4963 default: 4964 /* A translation error occurred. */ 4965 REASON_SET(&reason, transerror); 4966 goto cleanup; 4967 case PFRES_MAX: 4968 /* No match. */ 4969 break; 4970 case PFRES_MATCH: 4971 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4972 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4973 4974 if (nr->log) { 4975 PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a, 4976 ruleset, pd, 1); 4977 } 4978 4979 if (pd->ip_sum) 4980 bip_sum = *pd->ip_sum; 4981 4982 switch (pd->proto) { 4983 case IPPROTO_TCP: 4984 bproto_sum = th->th_sum; 4985 pd->proto_sum = &th->th_sum; 4986 4987 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4988 nk->port[pd->sidx] != sport) { 4989 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4990 &th->th_sum, &nk->addr[pd->sidx], 4991 nk->port[pd->sidx], 0, af); 4992 pd->sport = &th->th_sport; 4993 sport = th->th_sport; 4994 } 4995 4996 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4997 nk->port[pd->didx] != dport) { 4998 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4999 &th->th_sum, &nk->addr[pd->didx], 5000 nk->port[pd->didx], 0, af); 5001 dport = th->th_dport; 5002 pd->dport = &th->th_dport; 5003 } 5004 rewrite++; 5005 break; 5006 case IPPROTO_UDP: 5007 bproto_sum = pd->hdr.udp.uh_sum; 5008 pd->proto_sum = &pd->hdr.udp.uh_sum; 5009 5010 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5011 nk->port[pd->sidx] != sport) { 5012 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 5013 pd->ip_sum, &pd->hdr.udp.uh_sum, 5014 &nk->addr[pd->sidx], 5015 nk->port[pd->sidx], 1, af); 5016 sport = pd->hdr.udp.uh_sport; 5017 pd->sport = &pd->hdr.udp.uh_sport; 5018 } 5019 5020 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5021 nk->port[pd->didx] != dport) { 5022 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 5023 pd->ip_sum, &pd->hdr.udp.uh_sum, 5024 &nk->addr[pd->didx], 5025 nk->port[pd->didx], 1, af); 5026 dport = pd->hdr.udp.uh_dport; 5027 pd->dport = &pd->hdr.udp.uh_dport; 5028 } 5029 rewrite++; 5030 break; 5031 case IPPROTO_SCTP: { 5032 uint16_t checksum = 0; 5033 5034 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 5035 nk->port[pd->sidx] != sport) { 5036 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 5037 pd->ip_sum, &checksum, 5038 &nk->addr[pd->sidx], 5039 nk->port[pd->sidx], 1, af); 5040 } 5041 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 5042 nk->port[pd->didx] != dport) { 5043 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 5044 pd->ip_sum, &checksum, 5045 &nk->addr[pd->didx], 5046 nk->port[pd->didx], 1, af); 5047 } 5048 break; 5049 } 5050 #ifdef INET 5051 case IPPROTO_ICMP: 5052 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 5053 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 5054 nk->addr[pd->sidx].v4.s_addr, 0); 5055 5056 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 5057 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 5058 nk->addr[pd->didx].v4.s_addr, 0); 5059 5060 if (virtual_type == htons(ICMP_ECHO) && 5061 nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) { 5062 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 5063 pd->hdr.icmp.icmp_cksum, sport, 5064 nk->port[pd->sidx], 0); 5065 pd->hdr.icmp.icmp_id = nk->port[pd->sidx]; 5066 pd->sport = &pd->hdr.icmp.icmp_id; 5067 } 5068 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5069 break; 5070 #endif /* INET */ 5071 #ifdef INET6 5072 case IPPROTO_ICMPV6: 5073 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 5074 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 5075 &nk->addr[pd->sidx], 0); 5076 5077 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 5078 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 5079 &nk->addr[pd->didx], 0); 5080 rewrite++; 5081 break; 5082 #endif /* INET */ 5083 default: 5084 switch (af) { 5085 #ifdef INET 5086 case AF_INET: 5087 if (PF_ANEQ(saddr, 5088 &nk->addr[pd->sidx], AF_INET)) 5089 pf_change_a(&saddr->v4.s_addr, 5090 pd->ip_sum, 5091 nk->addr[pd->sidx].v4.s_addr, 0); 5092 5093 if (PF_ANEQ(daddr, 5094 &nk->addr[pd->didx], AF_INET)) 5095 pf_change_a(&daddr->v4.s_addr, 5096 pd->ip_sum, 5097 nk->addr[pd->didx].v4.s_addr, 0); 5098 break; 5099 #endif /* INET */ 5100 #ifdef INET6 5101 case AF_INET6: 5102 if (PF_ANEQ(saddr, 5103 &nk->addr[pd->sidx], AF_INET6)) 5104 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 5105 5106 if (PF_ANEQ(daddr, 5107 &nk->addr[pd->didx], AF_INET6)) 5108 PF_ACPY(daddr, &nk->addr[pd->didx], af); 5109 break; 5110 #endif /* INET */ 5111 } 5112 break; 5113 } 5114 if (nr->natpass) 5115 r = NULL; 5116 pd->nat_rule = nr; 5117 } 5118 5119 while (r != NULL) { 5120 pf_counter_u64_add(&r->evaluations, 1); 5121 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5122 r = r->skip[PF_SKIP_IFP].ptr; 5123 else if (r->direction && r->direction != pd->dir) 5124 r = r->skip[PF_SKIP_DIR].ptr; 5125 else if (r->af && r->af != af) 5126 r = r->skip[PF_SKIP_AF].ptr; 5127 else if (r->proto && r->proto != pd->proto) 5128 r = r->skip[PF_SKIP_PROTO].ptr; 5129 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 5130 r->src.neg, kif, M_GETFIB(m))) 5131 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5132 /* tcp/udp only. port_op always 0 in other cases */ 5133 else if (r->src.port_op && !pf_match_port(r->src.port_op, 5134 r->src.port[0], r->src.port[1], sport)) 5135 r = r->skip[PF_SKIP_SRC_PORT].ptr; 5136 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 5137 r->dst.neg, NULL, M_GETFIB(m))) 5138 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5139 /* tcp/udp only. port_op always 0 in other cases */ 5140 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 5141 r->dst.port[0], r->dst.port[1], dport)) 5142 r = r->skip[PF_SKIP_DST_PORT].ptr; 5143 /* icmp only. type always 0 in other cases */ 5144 else if (r->type && r->type != icmptype + 1) 5145 r = TAILQ_NEXT(r, entries); 5146 /* icmp only. type always 0 in other cases */ 5147 else if (r->code && r->code != icmpcode + 1) 5148 r = TAILQ_NEXT(r, entries); 5149 else if (r->tos && !(r->tos == pd->tos)) 5150 r = TAILQ_NEXT(r, entries); 5151 else if (r->rule_flag & PFRULE_FRAGMENT) 5152 r = TAILQ_NEXT(r, entries); 5153 else if (pd->proto == IPPROTO_TCP && 5154 (r->flagset & th->th_flags) != r->flags) 5155 r = TAILQ_NEXT(r, entries); 5156 /* tcp/udp only. uid.op always 0 in other cases */ 5157 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 5158 pf_socket_lookup(pd, m), 1)) && 5159 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 5160 pd->lookup.uid)) 5161 r = TAILQ_NEXT(r, entries); 5162 /* tcp/udp only. gid.op always 0 in other cases */ 5163 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 5164 pf_socket_lookup(pd, m), 1)) && 5165 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 5166 pd->lookup.gid)) 5167 r = TAILQ_NEXT(r, entries); 5168 else if (r->prio && 5169 !pf_match_ieee8021q_pcp(r->prio, m)) 5170 r = TAILQ_NEXT(r, entries); 5171 else if (r->prob && 5172 r->prob <= arc4random()) 5173 r = TAILQ_NEXT(r, entries); 5174 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5175 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5176 r = TAILQ_NEXT(r, entries); 5177 else if (r->rcv_kif && !pf_match_rcvif(m, r)) 5178 r = TAILQ_NEXT(r, entries); 5179 else if (r->os_fingerprint != PF_OSFP_ANY && 5180 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 5181 pf_osfp_fingerprint(pd, m, off, th), 5182 r->os_fingerprint))) 5183 r = TAILQ_NEXT(r, entries); 5184 else { 5185 if (r->tag) 5186 tag = r->tag; 5187 if (r->anchor == NULL) { 5188 if (r->action == PF_MATCH) { 5189 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5190 if (ri == NULL) { 5191 REASON_SET(&reason, PFRES_MEMORY); 5192 goto cleanup; 5193 } 5194 ri->r = r; 5195 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5196 pf_counter_u64_critical_enter(); 5197 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5198 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5199 pf_counter_u64_critical_exit(); 5200 pf_rule_to_actions(r, &pd->act); 5201 if (r->log || pd->act.log & PF_LOG_MATCHES) 5202 PFLOG_PACKET(kif, m, af, 5203 r->action, PFRES_MATCH, r, 5204 a, ruleset, pd, 1); 5205 } else { 5206 match = 1; 5207 *rm = r; 5208 *am = a; 5209 *rsm = ruleset; 5210 if (pd->act.log & PF_LOG_MATCHES) 5211 PFLOG_PACKET(kif, m, af, 5212 r->action, PFRES_MATCH, r, 5213 a, ruleset, pd, 1); 5214 } 5215 if ((*rm)->quick) 5216 break; 5217 r = TAILQ_NEXT(r, entries); 5218 } else 5219 pf_step_into_anchor(anchor_stack, &asd, 5220 &ruleset, PF_RULESET_FILTER, &r, &a, 5221 &match); 5222 } 5223 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5224 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5225 break; 5226 } 5227 r = *rm; 5228 a = *am; 5229 ruleset = *rsm; 5230 5231 REASON_SET(&reason, PFRES_MATCH); 5232 5233 /* apply actions for last matching pass/block rule */ 5234 pf_rule_to_actions(r, &pd->act); 5235 5236 if (r->log || pd->act.log & PF_LOG_MATCHES) { 5237 if (rewrite) 5238 m_copyback(m, off, hdrlen, pd->hdr.any); 5239 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 5240 } 5241 5242 if ((r->action == PF_DROP) && 5243 ((r->rule_flag & PFRULE_RETURNRST) || 5244 (r->rule_flag & PFRULE_RETURNICMP) || 5245 (r->rule_flag & PFRULE_RETURN))) { 5246 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 5247 bip_sum, hdrlen, &reason, r->rtableid); 5248 } 5249 5250 if (r->action == PF_DROP) 5251 goto cleanup; 5252 5253 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5254 REASON_SET(&reason, PFRES_MEMORY); 5255 goto cleanup; 5256 } 5257 if (pd->act.rtableid >= 0) 5258 M_SETFIB(m, pd->act.rtableid); 5259 5260 if (!state_icmp && (r->keep_state || nr != NULL || 5261 (pd->flags & PFDESC_TCP_NORM))) { 5262 int action; 5263 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 5264 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 5265 hdrlen, &match_rules, udp_mapping); 5266 if (action != PF_PASS) { 5267 pf_udp_mapping_release(udp_mapping); 5268 if (action == PF_DROP && 5269 (r->rule_flag & PFRULE_RETURN)) 5270 pf_return(r, nr, pd, sk, off, m, th, kif, 5271 bproto_sum, bip_sum, hdrlen, &reason, 5272 pd->act.rtableid); 5273 return (action); 5274 } 5275 } else { 5276 while ((ri = SLIST_FIRST(&match_rules))) { 5277 SLIST_REMOVE_HEAD(&match_rules, entry); 5278 free(ri, M_PF_RULE_ITEM); 5279 } 5280 5281 uma_zfree(V_pf_state_key_z, sk); 5282 uma_zfree(V_pf_state_key_z, nk); 5283 pf_udp_mapping_release(udp_mapping); 5284 } 5285 5286 /* copy back packet headers if we performed NAT operations */ 5287 if (rewrite) 5288 m_copyback(m, off, hdrlen, pd->hdr.any); 5289 5290 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 5291 pd->dir == PF_OUT && 5292 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 5293 /* 5294 * We want the state created, but we dont 5295 * want to send this in case a partner 5296 * firewall has to know about it to allow 5297 * replies through it. 5298 */ 5299 return (PF_DEFER); 5300 5301 return (PF_PASS); 5302 5303 cleanup: 5304 while ((ri = SLIST_FIRST(&match_rules))) { 5305 SLIST_REMOVE_HEAD(&match_rules, entry); 5306 free(ri, M_PF_RULE_ITEM); 5307 } 5308 5309 uma_zfree(V_pf_state_key_z, sk); 5310 uma_zfree(V_pf_state_key_z, nk); 5311 pf_udp_mapping_release(udp_mapping); 5312 5313 return (PF_DROP); 5314 } 5315 5316 static int 5317 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 5318 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 5319 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 5320 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 5321 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 5322 struct pf_krule_slist *match_rules, struct pf_udp_mapping *udp_mapping) 5323 { 5324 struct pf_kstate *s = NULL; 5325 struct pf_ksrc_node *sn = NULL; 5326 struct tcphdr *th = &pd->hdr.tcp; 5327 u_int16_t mss = V_tcp_mssdflt; 5328 u_short reason, sn_reason; 5329 struct pf_krule_item *ri; 5330 5331 /* check maximums */ 5332 if (r->max_states && 5333 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 5334 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 5335 REASON_SET(&reason, PFRES_MAXSTATES); 5336 goto csfailed; 5337 } 5338 /* src node for filter rule */ 5339 if ((r->rule_flag & PFRULE_SRCTRACK || 5340 r->rpool.opts & PF_POOL_STICKYADDR) && 5341 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 5342 REASON_SET(&reason, sn_reason); 5343 goto csfailed; 5344 } 5345 /* src node for translation rule */ 5346 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 5347 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 5348 pd->af)) != 0 ) { 5349 REASON_SET(&reason, sn_reason); 5350 goto csfailed; 5351 } 5352 s = pf_alloc_state(M_NOWAIT); 5353 if (s == NULL) { 5354 REASON_SET(&reason, PFRES_MEMORY); 5355 goto csfailed; 5356 } 5357 s->rule.ptr = r; 5358 s->nat_rule.ptr = nr; 5359 s->anchor.ptr = a; 5360 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 5361 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 5362 5363 STATE_INC_COUNTERS(s); 5364 if (r->allow_opts) 5365 s->state_flags |= PFSTATE_ALLOWOPTS; 5366 if (r->rule_flag & PFRULE_STATESLOPPY) 5367 s->state_flags |= PFSTATE_SLOPPY; 5368 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 5369 s->state_flags |= PFSTATE_SCRUB_TCP; 5370 if ((r->rule_flag & PFRULE_PFLOW) || 5371 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 5372 s->state_flags |= PFSTATE_PFLOW; 5373 5374 s->act.log = pd->act.log & PF_LOG_ALL; 5375 s->sync_state = PFSYNC_S_NONE; 5376 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 5377 5378 if (nr != NULL) 5379 s->act.log |= nr->log & PF_LOG_ALL; 5380 switch (pd->proto) { 5381 case IPPROTO_TCP: 5382 s->src.seqlo = ntohl(th->th_seq); 5383 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 5384 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 5385 r->keep_state == PF_STATE_MODULATE) { 5386 /* Generate sequence number modulator */ 5387 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 5388 0) 5389 s->src.seqdiff = 1; 5390 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 5391 htonl(s->src.seqlo + s->src.seqdiff), 0); 5392 *rewrite = 1; 5393 } else 5394 s->src.seqdiff = 0; 5395 if (th->th_flags & TH_SYN) { 5396 s->src.seqhi++; 5397 s->src.wscale = pf_get_wscale(m, off, 5398 th->th_off, pd->af); 5399 } 5400 s->src.max_win = MAX(ntohs(th->th_win), 1); 5401 if (s->src.wscale & PF_WSCALE_MASK) { 5402 /* Remove scale factor from initial window */ 5403 int win = s->src.max_win; 5404 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 5405 s->src.max_win = (win - 1) >> 5406 (s->src.wscale & PF_WSCALE_MASK); 5407 } 5408 if (th->th_flags & TH_FIN) 5409 s->src.seqhi++; 5410 s->dst.seqhi = 1; 5411 s->dst.max_win = 1; 5412 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 5413 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 5414 s->timeout = PFTM_TCP_FIRST_PACKET; 5415 atomic_add_32(&V_pf_status.states_halfopen, 1); 5416 break; 5417 case IPPROTO_UDP: 5418 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 5419 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 5420 s->timeout = PFTM_UDP_FIRST_PACKET; 5421 break; 5422 case IPPROTO_SCTP: 5423 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 5424 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 5425 s->timeout = PFTM_SCTP_FIRST_PACKET; 5426 break; 5427 case IPPROTO_ICMP: 5428 #ifdef INET6 5429 case IPPROTO_ICMPV6: 5430 #endif 5431 s->timeout = PFTM_ICMP_FIRST_PACKET; 5432 break; 5433 default: 5434 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 5435 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 5436 s->timeout = PFTM_OTHER_FIRST_PACKET; 5437 } 5438 5439 if (r->rt) { 5440 /* pf_map_addr increases the reason counters */ 5441 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 5442 &s->rt_kif, NULL, &sn)) != 0) 5443 goto csfailed; 5444 s->rt = r->rt; 5445 } 5446 5447 s->creation = s->expire = pf_get_uptime(); 5448 5449 if (sn != NULL) 5450 s->src_node = sn; 5451 if (nsn != NULL) { 5452 /* XXX We only modify one side for now. */ 5453 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 5454 s->nat_src_node = nsn; 5455 } 5456 if (pd->proto == IPPROTO_TCP) { 5457 if (s->state_flags & PFSTATE_SCRUB_TCP && 5458 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 5459 REASON_SET(&reason, PFRES_MEMORY); 5460 goto csfailed; 5461 } 5462 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 5463 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 5464 &s->src, &s->dst, rewrite)) { 5465 /* This really shouldn't happen!!! */ 5466 DPFPRINTF(PF_DEBUG_URGENT, 5467 ("pf_normalize_tcp_stateful failed on first " 5468 "pkt\n")); 5469 goto csfailed; 5470 } 5471 } else if (pd->proto == IPPROTO_SCTP) { 5472 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5473 goto csfailed; 5474 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5475 goto csfailed; 5476 } 5477 s->direction = pd->dir; 5478 5479 /* 5480 * sk/nk could already been setup by pf_get_translation(). 5481 */ 5482 if (nr == NULL) { 5483 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5484 __func__, nr, sk, nk)); 5485 sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport); 5486 if (sk == NULL) 5487 goto csfailed; 5488 nk = sk; 5489 } else 5490 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5491 __func__, nr, sk, nk)); 5492 5493 /* Swap sk/nk for PF_OUT. */ 5494 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5495 (pd->dir == PF_IN) ? sk : nk, 5496 (pd->dir == PF_IN) ? nk : sk, s)) { 5497 REASON_SET(&reason, PFRES_STATEINS); 5498 goto drop; 5499 } else 5500 *sm = s; 5501 5502 if (tag > 0) 5503 s->tag = tag; 5504 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5505 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5506 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5507 /* undo NAT changes, if they have taken place */ 5508 if (nr != NULL) { 5509 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5510 if (pd->dir == PF_OUT) 5511 skt = s->key[PF_SK_STACK]; 5512 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5513 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5514 if (pd->sport) 5515 *pd->sport = skt->port[pd->sidx]; 5516 if (pd->dport) 5517 *pd->dport = skt->port[pd->didx]; 5518 if (pd->proto_sum) 5519 *pd->proto_sum = bproto_sum; 5520 if (pd->ip_sum) 5521 *pd->ip_sum = bip_sum; 5522 m_copyback(m, off, hdrlen, pd->hdr.any); 5523 } 5524 s->src.seqhi = htonl(arc4random()); 5525 /* Find mss option */ 5526 int rtid = M_GETFIB(m); 5527 mss = pf_get_mss(m, off, th->th_off, pd->af); 5528 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5529 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5530 s->src.mss = mss; 5531 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5532 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5533 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5534 pd->act.rtableid); 5535 REASON_SET(&reason, PFRES_SYNPROXY); 5536 return (PF_SYNPROXY_DROP); 5537 } 5538 5539 s->udp_mapping = udp_mapping; 5540 5541 return (PF_PASS); 5542 5543 csfailed: 5544 while ((ri = SLIST_FIRST(match_rules))) { 5545 SLIST_REMOVE_HEAD(match_rules, entry); 5546 free(ri, M_PF_RULE_ITEM); 5547 } 5548 5549 uma_zfree(V_pf_state_key_z, sk); 5550 uma_zfree(V_pf_state_key_z, nk); 5551 5552 if (sn != NULL) { 5553 PF_SRC_NODE_LOCK(sn); 5554 if (--sn->states == 0 && sn->expire == 0) { 5555 pf_unlink_src_node(sn); 5556 uma_zfree(V_pf_sources_z, sn); 5557 counter_u64_add( 5558 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5559 } 5560 PF_SRC_NODE_UNLOCK(sn); 5561 } 5562 5563 if (nsn != sn && nsn != NULL) { 5564 PF_SRC_NODE_LOCK(nsn); 5565 if (--nsn->states == 0 && nsn->expire == 0) { 5566 pf_unlink_src_node(nsn); 5567 uma_zfree(V_pf_sources_z, nsn); 5568 counter_u64_add( 5569 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5570 } 5571 PF_SRC_NODE_UNLOCK(nsn); 5572 } 5573 5574 drop: 5575 if (s != NULL) { 5576 pf_src_tree_remove_state(s); 5577 s->timeout = PFTM_UNLINKED; 5578 STATE_DEC_COUNTERS(s); 5579 pf_free_state(s); 5580 } 5581 5582 return (PF_DROP); 5583 } 5584 5585 static int 5586 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5587 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5588 struct pf_kruleset **rsm) 5589 { 5590 struct pf_krule *r, *a = NULL; 5591 struct pf_kruleset *ruleset = NULL; 5592 struct pf_krule_slist match_rules; 5593 struct pf_krule_item *ri; 5594 sa_family_t af = pd->af; 5595 u_short reason; 5596 int tag = -1; 5597 int asd = 0; 5598 int match = 0; 5599 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5600 5601 PF_RULES_RASSERT(); 5602 5603 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5604 SLIST_INIT(&match_rules); 5605 while (r != NULL) { 5606 pf_counter_u64_add(&r->evaluations, 1); 5607 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5608 r = r->skip[PF_SKIP_IFP].ptr; 5609 else if (r->direction && r->direction != pd->dir) 5610 r = r->skip[PF_SKIP_DIR].ptr; 5611 else if (r->af && r->af != af) 5612 r = r->skip[PF_SKIP_AF].ptr; 5613 else if (r->proto && r->proto != pd->proto) 5614 r = r->skip[PF_SKIP_PROTO].ptr; 5615 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5616 r->src.neg, kif, M_GETFIB(m))) 5617 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5618 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5619 r->dst.neg, NULL, M_GETFIB(m))) 5620 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5621 else if (r->tos && !(r->tos == pd->tos)) 5622 r = TAILQ_NEXT(r, entries); 5623 else if (r->os_fingerprint != PF_OSFP_ANY) 5624 r = TAILQ_NEXT(r, entries); 5625 else if (pd->proto == IPPROTO_UDP && 5626 (r->src.port_op || r->dst.port_op)) 5627 r = TAILQ_NEXT(r, entries); 5628 else if (pd->proto == IPPROTO_TCP && 5629 (r->src.port_op || r->dst.port_op || r->flagset)) 5630 r = TAILQ_NEXT(r, entries); 5631 else if ((pd->proto == IPPROTO_ICMP || 5632 pd->proto == IPPROTO_ICMPV6) && 5633 (r->type || r->code)) 5634 r = TAILQ_NEXT(r, entries); 5635 else if (r->prio && 5636 !pf_match_ieee8021q_pcp(r->prio, m)) 5637 r = TAILQ_NEXT(r, entries); 5638 else if (r->prob && r->prob <= 5639 (arc4random() % (UINT_MAX - 1) + 1)) 5640 r = TAILQ_NEXT(r, entries); 5641 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5642 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5643 r = TAILQ_NEXT(r, entries); 5644 else { 5645 if (r->anchor == NULL) { 5646 if (r->action == PF_MATCH) { 5647 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5648 if (ri == NULL) { 5649 REASON_SET(&reason, PFRES_MEMORY); 5650 goto cleanup; 5651 } 5652 ri->r = r; 5653 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5654 pf_counter_u64_critical_enter(); 5655 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5656 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5657 pf_counter_u64_critical_exit(); 5658 pf_rule_to_actions(r, &pd->act); 5659 if (r->log) 5660 PFLOG_PACKET(kif, m, af, 5661 r->action, PFRES_MATCH, r, 5662 a, ruleset, pd, 1); 5663 } else { 5664 match = 1; 5665 *rm = r; 5666 *am = a; 5667 *rsm = ruleset; 5668 } 5669 if ((*rm)->quick) 5670 break; 5671 r = TAILQ_NEXT(r, entries); 5672 } else 5673 pf_step_into_anchor(anchor_stack, &asd, 5674 &ruleset, PF_RULESET_FILTER, &r, &a, 5675 &match); 5676 } 5677 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5678 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5679 break; 5680 } 5681 r = *rm; 5682 a = *am; 5683 ruleset = *rsm; 5684 5685 REASON_SET(&reason, PFRES_MATCH); 5686 5687 /* apply actions for last matching pass/block rule */ 5688 pf_rule_to_actions(r, &pd->act); 5689 5690 if (r->log) 5691 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 5692 5693 if (r->action != PF_PASS) 5694 return (PF_DROP); 5695 5696 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5697 REASON_SET(&reason, PFRES_MEMORY); 5698 goto cleanup; 5699 } 5700 5701 return (PF_PASS); 5702 5703 cleanup: 5704 while ((ri = SLIST_FIRST(&match_rules))) { 5705 SLIST_REMOVE_HEAD(&match_rules, entry); 5706 free(ri, M_PF_RULE_ITEM); 5707 } 5708 5709 return (PF_DROP); 5710 } 5711 5712 static int 5713 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5714 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5715 int *copyback) 5716 { 5717 struct tcphdr *th = &pd->hdr.tcp; 5718 struct pf_state_peer *src, *dst; 5719 u_int16_t win = ntohs(th->th_win); 5720 u_int32_t ack, end, data_end, seq, orig_seq; 5721 u_int8_t sws, dws, psrc, pdst; 5722 int ackskew; 5723 5724 if (pd->dir == (*state)->direction) { 5725 src = &(*state)->src; 5726 dst = &(*state)->dst; 5727 psrc = PF_PEER_SRC; 5728 pdst = PF_PEER_DST; 5729 } else { 5730 src = &(*state)->dst; 5731 dst = &(*state)->src; 5732 psrc = PF_PEER_DST; 5733 pdst = PF_PEER_SRC; 5734 } 5735 5736 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5737 sws = src->wscale & PF_WSCALE_MASK; 5738 dws = dst->wscale & PF_WSCALE_MASK; 5739 } else 5740 sws = dws = 0; 5741 5742 /* 5743 * Sequence tracking algorithm from Guido van Rooij's paper: 5744 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5745 * tcp_filtering.ps 5746 */ 5747 5748 orig_seq = seq = ntohl(th->th_seq); 5749 if (src->seqlo == 0) { 5750 /* First packet from this end. Set its state */ 5751 5752 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5753 src->scrub == NULL) { 5754 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5755 REASON_SET(reason, PFRES_MEMORY); 5756 return (PF_DROP); 5757 } 5758 } 5759 5760 /* Deferred generation of sequence number modulator */ 5761 if (dst->seqdiff && !src->seqdiff) { 5762 /* use random iss for the TCP server */ 5763 while ((src->seqdiff = arc4random() - seq) == 0) 5764 ; 5765 ack = ntohl(th->th_ack) - dst->seqdiff; 5766 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5767 src->seqdiff), 0); 5768 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5769 *copyback = 1; 5770 } else { 5771 ack = ntohl(th->th_ack); 5772 } 5773 5774 end = seq + pd->p_len; 5775 if (th->th_flags & TH_SYN) { 5776 end++; 5777 if (dst->wscale & PF_WSCALE_FLAG) { 5778 src->wscale = pf_get_wscale(m, off, th->th_off, 5779 pd->af); 5780 if (src->wscale & PF_WSCALE_FLAG) { 5781 /* Remove scale factor from initial 5782 * window */ 5783 sws = src->wscale & PF_WSCALE_MASK; 5784 win = ((u_int32_t)win + (1 << sws) - 1) 5785 >> sws; 5786 dws = dst->wscale & PF_WSCALE_MASK; 5787 } else { 5788 /* fixup other window */ 5789 dst->max_win = MIN(TCP_MAXWIN, 5790 (u_int32_t)dst->max_win << 5791 (dst->wscale & PF_WSCALE_MASK)); 5792 /* in case of a retrans SYN|ACK */ 5793 dst->wscale = 0; 5794 } 5795 } 5796 } 5797 data_end = end; 5798 if (th->th_flags & TH_FIN) 5799 end++; 5800 5801 src->seqlo = seq; 5802 if (src->state < TCPS_SYN_SENT) 5803 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5804 5805 /* 5806 * May need to slide the window (seqhi may have been set by 5807 * the crappy stack check or if we picked up the connection 5808 * after establishment) 5809 */ 5810 if (src->seqhi == 1 || 5811 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5812 src->seqhi = end + MAX(1, dst->max_win << dws); 5813 if (win > src->max_win) 5814 src->max_win = win; 5815 5816 } else { 5817 ack = ntohl(th->th_ack) - dst->seqdiff; 5818 if (src->seqdiff) { 5819 /* Modulate sequence numbers */ 5820 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5821 src->seqdiff), 0); 5822 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5823 *copyback = 1; 5824 } 5825 end = seq + pd->p_len; 5826 if (th->th_flags & TH_SYN) 5827 end++; 5828 data_end = end; 5829 if (th->th_flags & TH_FIN) 5830 end++; 5831 } 5832 5833 if ((th->th_flags & TH_ACK) == 0) { 5834 /* Let it pass through the ack skew check */ 5835 ack = dst->seqlo; 5836 } else if ((ack == 0 && 5837 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5838 /* broken tcp stacks do not set ack */ 5839 (dst->state < TCPS_SYN_SENT)) { 5840 /* 5841 * Many stacks (ours included) will set the ACK number in an 5842 * FIN|ACK if the SYN times out -- no sequence to ACK. 5843 */ 5844 ack = dst->seqlo; 5845 } 5846 5847 if (seq == end) { 5848 /* Ease sequencing restrictions on no data packets */ 5849 seq = src->seqlo; 5850 data_end = end = seq; 5851 } 5852 5853 ackskew = dst->seqlo - ack; 5854 5855 /* 5856 * Need to demodulate the sequence numbers in any TCP SACK options 5857 * (Selective ACK). We could optionally validate the SACK values 5858 * against the current ACK window, either forwards or backwards, but 5859 * I'm not confident that SACK has been implemented properly 5860 * everywhere. It wouldn't surprise me if several stacks accidentally 5861 * SACK too far backwards of previously ACKed data. There really aren't 5862 * any security implications of bad SACKing unless the target stack 5863 * doesn't validate the option length correctly. Someone trying to 5864 * spoof into a TCP connection won't bother blindly sending SACK 5865 * options anyway. 5866 */ 5867 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5868 if (pf_modulate_sack(m, off, pd, th, dst)) 5869 *copyback = 1; 5870 } 5871 5872 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5873 if (SEQ_GEQ(src->seqhi, data_end) && 5874 /* Last octet inside other's window space */ 5875 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5876 /* Retrans: not more than one window back */ 5877 (ackskew >= -MAXACKWINDOW) && 5878 /* Acking not more than one reassembled fragment backwards */ 5879 (ackskew <= (MAXACKWINDOW << sws)) && 5880 /* Acking not more than one window forward */ 5881 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5882 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5883 /* Require an exact/+1 sequence match on resets when possible */ 5884 5885 if (dst->scrub || src->scrub) { 5886 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5887 *state, src, dst, copyback)) 5888 return (PF_DROP); 5889 } 5890 5891 /* update max window */ 5892 if (src->max_win < win) 5893 src->max_win = win; 5894 /* synchronize sequencing */ 5895 if (SEQ_GT(end, src->seqlo)) 5896 src->seqlo = end; 5897 /* slide the window of what the other end can send */ 5898 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5899 dst->seqhi = ack + MAX((win << sws), 1); 5900 5901 /* update states */ 5902 if (th->th_flags & TH_SYN) 5903 if (src->state < TCPS_SYN_SENT) 5904 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5905 if (th->th_flags & TH_FIN) 5906 if (src->state < TCPS_CLOSING) 5907 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5908 if (th->th_flags & TH_ACK) { 5909 if (dst->state == TCPS_SYN_SENT) { 5910 pf_set_protostate(*state, pdst, 5911 TCPS_ESTABLISHED); 5912 if (src->state == TCPS_ESTABLISHED && 5913 (*state)->src_node != NULL && 5914 pf_src_connlimit(state)) { 5915 REASON_SET(reason, PFRES_SRCLIMIT); 5916 return (PF_DROP); 5917 } 5918 } else if (dst->state == TCPS_CLOSING) 5919 pf_set_protostate(*state, pdst, 5920 TCPS_FIN_WAIT_2); 5921 } 5922 if (th->th_flags & TH_RST) 5923 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5924 5925 /* update expire time */ 5926 (*state)->expire = pf_get_uptime(); 5927 if (src->state >= TCPS_FIN_WAIT_2 && 5928 dst->state >= TCPS_FIN_WAIT_2) 5929 (*state)->timeout = PFTM_TCP_CLOSED; 5930 else if (src->state >= TCPS_CLOSING && 5931 dst->state >= TCPS_CLOSING) 5932 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5933 else if (src->state < TCPS_ESTABLISHED || 5934 dst->state < TCPS_ESTABLISHED) 5935 (*state)->timeout = PFTM_TCP_OPENING; 5936 else if (src->state >= TCPS_CLOSING || 5937 dst->state >= TCPS_CLOSING) 5938 (*state)->timeout = PFTM_TCP_CLOSING; 5939 else 5940 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5941 5942 /* Fall through to PASS packet */ 5943 5944 } else if ((dst->state < TCPS_SYN_SENT || 5945 dst->state >= TCPS_FIN_WAIT_2 || 5946 src->state >= TCPS_FIN_WAIT_2) && 5947 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) && 5948 /* Within a window forward of the originating packet */ 5949 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5950 /* Within a window backward of the originating packet */ 5951 5952 /* 5953 * This currently handles three situations: 5954 * 1) Stupid stacks will shotgun SYNs before their peer 5955 * replies. 5956 * 2) When PF catches an already established stream (the 5957 * firewall rebooted, the state table was flushed, routes 5958 * changed...) 5959 * 3) Packets get funky immediately after the connection 5960 * closes (this should catch Solaris spurious ACK|FINs 5961 * that web servers like to spew after a close) 5962 * 5963 * This must be a little more careful than the above code 5964 * since packet floods will also be caught here. We don't 5965 * update the TTL here to mitigate the damage of a packet 5966 * flood and so the same code can handle awkward establishment 5967 * and a loosened connection close. 5968 * In the establishment case, a correct peer response will 5969 * validate the connection, go through the normal state code 5970 * and keep updating the state TTL. 5971 */ 5972 5973 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5974 printf("pf: loose state match: "); 5975 pf_print_state(*state); 5976 pf_print_flags(th->th_flags); 5977 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5978 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5979 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5980 (unsigned long long)(*state)->packets[1], 5981 pd->dir == PF_IN ? "in" : "out", 5982 pd->dir == (*state)->direction ? "fwd" : "rev"); 5983 } 5984 5985 if (dst->scrub || src->scrub) { 5986 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5987 *state, src, dst, copyback)) 5988 return (PF_DROP); 5989 } 5990 5991 /* update max window */ 5992 if (src->max_win < win) 5993 src->max_win = win; 5994 /* synchronize sequencing */ 5995 if (SEQ_GT(end, src->seqlo)) 5996 src->seqlo = end; 5997 /* slide the window of what the other end can send */ 5998 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5999 dst->seqhi = ack + MAX((win << sws), 1); 6000 6001 /* 6002 * Cannot set dst->seqhi here since this could be a shotgunned 6003 * SYN and not an already established connection. 6004 */ 6005 6006 if (th->th_flags & TH_FIN) 6007 if (src->state < TCPS_CLOSING) 6008 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6009 if (th->th_flags & TH_RST) 6010 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6011 6012 /* Fall through to PASS packet */ 6013 6014 } else { 6015 if ((*state)->dst.state == TCPS_SYN_SENT && 6016 (*state)->src.state == TCPS_SYN_SENT) { 6017 /* Send RST for state mismatches during handshake */ 6018 if (!(th->th_flags & TH_RST)) 6019 pf_send_tcp((*state)->rule.ptr, pd->af, 6020 pd->dst, pd->src, th->th_dport, 6021 th->th_sport, ntohl(th->th_ack), 0, 6022 TH_RST, 0, 0, 6023 (*state)->rule.ptr->return_ttl, true, 0, 0, 6024 (*state)->act.rtableid); 6025 src->seqlo = 0; 6026 src->seqhi = 1; 6027 src->max_win = 1; 6028 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 6029 printf("pf: BAD state: "); 6030 pf_print_state(*state); 6031 pf_print_flags(th->th_flags); 6032 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 6033 "pkts=%llu:%llu dir=%s,%s\n", 6034 seq, orig_seq, ack, pd->p_len, ackskew, 6035 (unsigned long long)(*state)->packets[0], 6036 (unsigned long long)(*state)->packets[1], 6037 pd->dir == PF_IN ? "in" : "out", 6038 pd->dir == (*state)->direction ? "fwd" : "rev"); 6039 printf("pf: State failure on: %c %c %c %c | %c %c\n", 6040 SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1', 6041 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 6042 ' ': '2', 6043 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 6044 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 6045 SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5', 6046 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 6047 } 6048 REASON_SET(reason, PFRES_BADSTATE); 6049 return (PF_DROP); 6050 } 6051 6052 return (PF_PASS); 6053 } 6054 6055 static int 6056 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 6057 { 6058 struct tcphdr *th = &pd->hdr.tcp; 6059 struct pf_state_peer *src, *dst; 6060 u_int8_t psrc, pdst; 6061 6062 if (pd->dir == (*state)->direction) { 6063 src = &(*state)->src; 6064 dst = &(*state)->dst; 6065 psrc = PF_PEER_SRC; 6066 pdst = PF_PEER_DST; 6067 } else { 6068 src = &(*state)->dst; 6069 dst = &(*state)->src; 6070 psrc = PF_PEER_DST; 6071 pdst = PF_PEER_SRC; 6072 } 6073 6074 if (th->th_flags & TH_SYN) 6075 if (src->state < TCPS_SYN_SENT) 6076 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 6077 if (th->th_flags & TH_FIN) 6078 if (src->state < TCPS_CLOSING) 6079 pf_set_protostate(*state, psrc, TCPS_CLOSING); 6080 if (th->th_flags & TH_ACK) { 6081 if (dst->state == TCPS_SYN_SENT) { 6082 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 6083 if (src->state == TCPS_ESTABLISHED && 6084 (*state)->src_node != NULL && 6085 pf_src_connlimit(state)) { 6086 REASON_SET(reason, PFRES_SRCLIMIT); 6087 return (PF_DROP); 6088 } 6089 } else if (dst->state == TCPS_CLOSING) { 6090 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 6091 } else if (src->state == TCPS_SYN_SENT && 6092 dst->state < TCPS_SYN_SENT) { 6093 /* 6094 * Handle a special sloppy case where we only see one 6095 * half of the connection. If there is a ACK after 6096 * the initial SYN without ever seeing a packet from 6097 * the destination, set the connection to established. 6098 */ 6099 pf_set_protostate(*state, PF_PEER_BOTH, 6100 TCPS_ESTABLISHED); 6101 dst->state = src->state = TCPS_ESTABLISHED; 6102 if ((*state)->src_node != NULL && 6103 pf_src_connlimit(state)) { 6104 REASON_SET(reason, PFRES_SRCLIMIT); 6105 return (PF_DROP); 6106 } 6107 } else if (src->state == TCPS_CLOSING && 6108 dst->state == TCPS_ESTABLISHED && 6109 dst->seqlo == 0) { 6110 /* 6111 * Handle the closing of half connections where we 6112 * don't see the full bidirectional FIN/ACK+ACK 6113 * handshake. 6114 */ 6115 pf_set_protostate(*state, pdst, TCPS_CLOSING); 6116 } 6117 } 6118 if (th->th_flags & TH_RST) 6119 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 6120 6121 /* update expire time */ 6122 (*state)->expire = pf_get_uptime(); 6123 if (src->state >= TCPS_FIN_WAIT_2 && 6124 dst->state >= TCPS_FIN_WAIT_2) 6125 (*state)->timeout = PFTM_TCP_CLOSED; 6126 else if (src->state >= TCPS_CLOSING && 6127 dst->state >= TCPS_CLOSING) 6128 (*state)->timeout = PFTM_TCP_FIN_WAIT; 6129 else if (src->state < TCPS_ESTABLISHED || 6130 dst->state < TCPS_ESTABLISHED) 6131 (*state)->timeout = PFTM_TCP_OPENING; 6132 else if (src->state >= TCPS_CLOSING || 6133 dst->state >= TCPS_CLOSING) 6134 (*state)->timeout = PFTM_TCP_CLOSING; 6135 else 6136 (*state)->timeout = PFTM_TCP_ESTABLISHED; 6137 6138 return (PF_PASS); 6139 } 6140 6141 static int 6142 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 6143 { 6144 struct pf_state_key *sk = (*state)->key[pd->didx]; 6145 struct tcphdr *th = &pd->hdr.tcp; 6146 6147 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 6148 if (pd->dir != (*state)->direction) { 6149 REASON_SET(reason, PFRES_SYNPROXY); 6150 return (PF_SYNPROXY_DROP); 6151 } 6152 if (th->th_flags & TH_SYN) { 6153 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 6154 REASON_SET(reason, PFRES_SYNPROXY); 6155 return (PF_DROP); 6156 } 6157 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 6158 pd->src, th->th_dport, th->th_sport, 6159 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 6160 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 6161 (*state)->act.rtableid); 6162 REASON_SET(reason, PFRES_SYNPROXY); 6163 return (PF_SYNPROXY_DROP); 6164 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 6165 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6166 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6167 REASON_SET(reason, PFRES_SYNPROXY); 6168 return (PF_DROP); 6169 } else if ((*state)->src_node != NULL && 6170 pf_src_connlimit(state)) { 6171 REASON_SET(reason, PFRES_SRCLIMIT); 6172 return (PF_DROP); 6173 } else 6174 pf_set_protostate(*state, PF_PEER_SRC, 6175 PF_TCPS_PROXY_DST); 6176 } 6177 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 6178 if (pd->dir == (*state)->direction) { 6179 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 6180 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 6181 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 6182 REASON_SET(reason, PFRES_SYNPROXY); 6183 return (PF_DROP); 6184 } 6185 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 6186 if ((*state)->dst.seqhi == 1) 6187 (*state)->dst.seqhi = htonl(arc4random()); 6188 pf_send_tcp((*state)->rule.ptr, pd->af, 6189 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6190 sk->port[pd->sidx], sk->port[pd->didx], 6191 (*state)->dst.seqhi, 0, TH_SYN, 0, 6192 (*state)->src.mss, 0, false, (*state)->tag, 0, 6193 (*state)->act.rtableid); 6194 REASON_SET(reason, PFRES_SYNPROXY); 6195 return (PF_SYNPROXY_DROP); 6196 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 6197 (TH_SYN|TH_ACK)) || 6198 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 6199 REASON_SET(reason, PFRES_SYNPROXY); 6200 return (PF_DROP); 6201 } else { 6202 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 6203 (*state)->dst.seqlo = ntohl(th->th_seq); 6204 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 6205 pd->src, th->th_dport, th->th_sport, 6206 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 6207 TH_ACK, (*state)->src.max_win, 0, 0, false, 6208 (*state)->tag, 0, (*state)->act.rtableid); 6209 pf_send_tcp((*state)->rule.ptr, pd->af, 6210 &sk->addr[pd->sidx], &sk->addr[pd->didx], 6211 sk->port[pd->sidx], sk->port[pd->didx], 6212 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 6213 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 6214 (*state)->act.rtableid); 6215 (*state)->src.seqdiff = (*state)->dst.seqhi - 6216 (*state)->src.seqlo; 6217 (*state)->dst.seqdiff = (*state)->src.seqhi - 6218 (*state)->dst.seqlo; 6219 (*state)->src.seqhi = (*state)->src.seqlo + 6220 (*state)->dst.max_win; 6221 (*state)->dst.seqhi = (*state)->dst.seqlo + 6222 (*state)->src.max_win; 6223 (*state)->src.wscale = (*state)->dst.wscale = 0; 6224 pf_set_protostate(*state, PF_PEER_BOTH, 6225 TCPS_ESTABLISHED); 6226 REASON_SET(reason, PFRES_SYNPROXY); 6227 return (PF_SYNPROXY_DROP); 6228 } 6229 } 6230 6231 return (PF_PASS); 6232 } 6233 6234 static int 6235 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 6236 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 6237 u_short *reason) 6238 { 6239 struct pf_state_key_cmp key; 6240 struct tcphdr *th = &pd->hdr.tcp; 6241 int copyback = 0; 6242 int action; 6243 struct pf_state_peer *src, *dst; 6244 6245 bzero(&key, sizeof(key)); 6246 key.af = pd->af; 6247 key.proto = IPPROTO_TCP; 6248 if (pd->dir == PF_IN) { /* wire side, straight */ 6249 PF_ACPY(&key.addr[0], pd->src, key.af); 6250 PF_ACPY(&key.addr[1], pd->dst, key.af); 6251 key.port[0] = th->th_sport; 6252 key.port[1] = th->th_dport; 6253 } else { /* stack side, reverse */ 6254 PF_ACPY(&key.addr[1], pd->src, key.af); 6255 PF_ACPY(&key.addr[0], pd->dst, key.af); 6256 key.port[1] = th->th_sport; 6257 key.port[0] = th->th_dport; 6258 } 6259 6260 STATE_LOOKUP(kif, &key, *state, pd); 6261 6262 if (pd->dir == (*state)->direction) { 6263 src = &(*state)->src; 6264 dst = &(*state)->dst; 6265 } else { 6266 src = &(*state)->dst; 6267 dst = &(*state)->src; 6268 } 6269 6270 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 6271 return (action); 6272 6273 if (dst->state >= TCPS_FIN_WAIT_2 && 6274 src->state >= TCPS_FIN_WAIT_2 && 6275 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 6276 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 6277 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 6278 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6279 printf("pf: state reuse "); 6280 pf_print_state(*state); 6281 pf_print_flags(th->th_flags); 6282 printf("\n"); 6283 } 6284 /* XXX make sure it's the same direction ?? */ 6285 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 6286 pf_unlink_state(*state); 6287 *state = NULL; 6288 return (PF_DROP); 6289 } 6290 6291 if ((*state)->state_flags & PFSTATE_SLOPPY) { 6292 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 6293 return (PF_DROP); 6294 } else { 6295 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 6296 ©back) == PF_DROP) 6297 return (PF_DROP); 6298 } 6299 6300 /* translate source/destination address, if necessary */ 6301 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6302 struct pf_state_key *nk = (*state)->key[pd->didx]; 6303 6304 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6305 nk->port[pd->sidx] != th->th_sport) 6306 pf_change_ap(m, pd->src, &th->th_sport, 6307 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 6308 nk->port[pd->sidx], 0, pd->af); 6309 6310 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6311 nk->port[pd->didx] != th->th_dport) 6312 pf_change_ap(m, pd->dst, &th->th_dport, 6313 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 6314 nk->port[pd->didx], 0, pd->af); 6315 copyback = 1; 6316 } 6317 6318 /* Copyback sequence modulation or stateful scrub changes if needed */ 6319 if (copyback) 6320 m_copyback(m, off, sizeof(*th), (caddr_t)th); 6321 6322 return (PF_PASS); 6323 } 6324 6325 static int 6326 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 6327 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 6328 { 6329 struct pf_state_peer *src, *dst; 6330 struct pf_state_key_cmp key; 6331 struct udphdr *uh = &pd->hdr.udp; 6332 uint8_t psrc, pdst; 6333 6334 bzero(&key, sizeof(key)); 6335 key.af = pd->af; 6336 key.proto = IPPROTO_UDP; 6337 if (pd->dir == PF_IN) { /* wire side, straight */ 6338 PF_ACPY(&key.addr[0], pd->src, key.af); 6339 PF_ACPY(&key.addr[1], pd->dst, key.af); 6340 key.port[0] = uh->uh_sport; 6341 key.port[1] = uh->uh_dport; 6342 } else { /* stack side, reverse */ 6343 PF_ACPY(&key.addr[1], pd->src, key.af); 6344 PF_ACPY(&key.addr[0], pd->dst, key.af); 6345 key.port[1] = uh->uh_sport; 6346 key.port[0] = uh->uh_dport; 6347 } 6348 6349 STATE_LOOKUP(kif, &key, *state, pd); 6350 6351 if (pd->dir == (*state)->direction) { 6352 src = &(*state)->src; 6353 dst = &(*state)->dst; 6354 psrc = PF_PEER_SRC; 6355 pdst = PF_PEER_DST; 6356 } else { 6357 src = &(*state)->dst; 6358 dst = &(*state)->src; 6359 psrc = PF_PEER_DST; 6360 pdst = PF_PEER_SRC; 6361 } 6362 6363 /* update states */ 6364 if (src->state < PFUDPS_SINGLE) 6365 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 6366 if (dst->state == PFUDPS_SINGLE) 6367 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 6368 6369 /* update expire time */ 6370 (*state)->expire = pf_get_uptime(); 6371 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 6372 (*state)->timeout = PFTM_UDP_MULTIPLE; 6373 else 6374 (*state)->timeout = PFTM_UDP_SINGLE; 6375 6376 /* translate source/destination address, if necessary */ 6377 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6378 struct pf_state_key *nk = (*state)->key[pd->didx]; 6379 6380 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6381 nk->port[pd->sidx] != uh->uh_sport) 6382 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 6383 &uh->uh_sum, &nk->addr[pd->sidx], 6384 nk->port[pd->sidx], 1, pd->af); 6385 6386 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6387 nk->port[pd->didx] != uh->uh_dport) 6388 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 6389 &uh->uh_sum, &nk->addr[pd->didx], 6390 nk->port[pd->didx], 1, pd->af); 6391 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 6392 } 6393 6394 return (PF_PASS); 6395 } 6396 6397 static int 6398 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 6399 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6400 { 6401 struct pf_state_key_cmp key; 6402 struct pf_state_peer *src, *dst; 6403 struct sctphdr *sh = &pd->hdr.sctp; 6404 u_int8_t psrc; //, pdst; 6405 6406 bzero(&key, sizeof(key)); 6407 key.af = pd->af; 6408 key.proto = IPPROTO_SCTP; 6409 if (pd->dir == PF_IN) { /* wire side, straight */ 6410 PF_ACPY(&key.addr[0], pd->src, key.af); 6411 PF_ACPY(&key.addr[1], pd->dst, key.af); 6412 key.port[0] = sh->src_port; 6413 key.port[1] = sh->dest_port; 6414 } else { /* stack side, reverse */ 6415 PF_ACPY(&key.addr[1], pd->src, key.af); 6416 PF_ACPY(&key.addr[0], pd->dst, key.af); 6417 key.port[1] = sh->src_port; 6418 key.port[0] = sh->dest_port; 6419 } 6420 6421 STATE_LOOKUP(kif, &key, *state, pd); 6422 6423 if (pd->dir == (*state)->direction) { 6424 src = &(*state)->src; 6425 dst = &(*state)->dst; 6426 psrc = PF_PEER_SRC; 6427 } else { 6428 src = &(*state)->dst; 6429 dst = &(*state)->src; 6430 psrc = PF_PEER_DST; 6431 } 6432 6433 if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) && 6434 (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) && 6435 pd->sctp_flags & PFDESC_SCTP_INIT) { 6436 pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED); 6437 pf_unlink_state(*state); 6438 *state = NULL; 6439 return (PF_DROP); 6440 } 6441 6442 /* Track state. */ 6443 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 6444 if (src->state < SCTP_COOKIE_WAIT) { 6445 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 6446 (*state)->timeout = PFTM_SCTP_OPENING; 6447 } 6448 } 6449 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 6450 MPASS(dst->scrub != NULL); 6451 if (dst->scrub->pfss_v_tag == 0) 6452 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 6453 } 6454 6455 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 6456 if (src->state < SCTP_ESTABLISHED) { 6457 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 6458 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 6459 } 6460 } 6461 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 6462 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6463 if (src->state < SCTP_SHUTDOWN_PENDING) { 6464 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 6465 (*state)->timeout = PFTM_SCTP_CLOSING; 6466 } 6467 } 6468 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 6469 pf_set_protostate(*state, psrc, SCTP_CLOSED); 6470 (*state)->timeout = PFTM_SCTP_CLOSED; 6471 } 6472 6473 if (src->scrub != NULL) { 6474 if (src->scrub->pfss_v_tag == 0) { 6475 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 6476 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 6477 return (PF_DROP); 6478 } 6479 6480 (*state)->expire = pf_get_uptime(); 6481 6482 /* translate source/destination address, if necessary */ 6483 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6484 uint16_t checksum = 0; 6485 struct pf_state_key *nk = (*state)->key[pd->didx]; 6486 6487 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6488 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6489 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6490 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6491 nk->port[pd->sidx], 1, pd->af); 6492 } 6493 6494 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6495 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6496 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6497 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6498 nk->port[pd->didx], 1, pd->af); 6499 } 6500 } 6501 6502 return (PF_PASS); 6503 } 6504 6505 static void 6506 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6507 { 6508 struct pf_sctp_endpoint key; 6509 struct pf_sctp_endpoint *ep; 6510 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6511 struct pf_sctp_source *i, *tmp; 6512 6513 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6514 return; 6515 6516 PF_SCTP_ENDPOINTS_LOCK(); 6517 6518 key.v_tag = s->dst.scrub->pfss_v_tag; 6519 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6520 if (ep != NULL) { 6521 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6522 if (pf_addr_cmp(&i->addr, 6523 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6524 s->key[PF_SK_WIRE]->af) == 0) { 6525 SDT_PROBE3(pf, sctp, multihome, remove, 6526 key.v_tag, s, i); 6527 TAILQ_REMOVE(&ep->sources, i, entry); 6528 free(i, M_PFTEMP); 6529 break; 6530 } 6531 } 6532 6533 if (TAILQ_EMPTY(&ep->sources)) { 6534 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6535 free(ep, M_PFTEMP); 6536 } 6537 } 6538 6539 /* Other direction. */ 6540 key.v_tag = s->src.scrub->pfss_v_tag; 6541 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6542 if (ep != NULL) { 6543 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6544 if (pf_addr_cmp(&i->addr, 6545 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6546 s->key[PF_SK_WIRE]->af) == 0) { 6547 SDT_PROBE3(pf, sctp, multihome, remove, 6548 key.v_tag, s, i); 6549 TAILQ_REMOVE(&ep->sources, i, entry); 6550 free(i, M_PFTEMP); 6551 break; 6552 } 6553 } 6554 6555 if (TAILQ_EMPTY(&ep->sources)) { 6556 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6557 free(ep, M_PFTEMP); 6558 } 6559 } 6560 6561 PF_SCTP_ENDPOINTS_UNLOCK(); 6562 } 6563 6564 static void 6565 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6566 { 6567 struct pf_sctp_endpoint key = { 6568 .v_tag = v_tag, 6569 }; 6570 struct pf_sctp_source *i; 6571 struct pf_sctp_endpoint *ep; 6572 6573 PF_SCTP_ENDPOINTS_LOCK(); 6574 6575 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6576 if (ep == NULL) { 6577 ep = malloc(sizeof(struct pf_sctp_endpoint), 6578 M_PFTEMP, M_NOWAIT); 6579 if (ep == NULL) { 6580 PF_SCTP_ENDPOINTS_UNLOCK(); 6581 return; 6582 } 6583 6584 ep->v_tag = v_tag; 6585 TAILQ_INIT(&ep->sources); 6586 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6587 } 6588 6589 /* Avoid inserting duplicates. */ 6590 TAILQ_FOREACH(i, &ep->sources, entry) { 6591 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6592 PF_SCTP_ENDPOINTS_UNLOCK(); 6593 return; 6594 } 6595 } 6596 6597 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6598 if (i == NULL) { 6599 PF_SCTP_ENDPOINTS_UNLOCK(); 6600 return; 6601 } 6602 6603 i->af = pd->af; 6604 memcpy(&i->addr, a, sizeof(*a)); 6605 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6606 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6607 6608 PF_SCTP_ENDPOINTS_UNLOCK(); 6609 } 6610 6611 static void 6612 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6613 struct pf_kstate *s, int action) 6614 { 6615 struct pf_sctp_multihome_job *j, *tmp; 6616 struct pf_sctp_source *i; 6617 int ret __unused; 6618 struct pf_kstate *sm = NULL; 6619 struct pf_krule *ra = NULL; 6620 struct pf_krule *r = &V_pf_default_rule; 6621 struct pf_kruleset *rs = NULL; 6622 bool do_extra = true; 6623 6624 PF_RULES_RLOCK_TRACKER; 6625 6626 again: 6627 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6628 if (s == NULL || action != PF_PASS) 6629 goto free; 6630 6631 /* Confirm we don't recurse here. */ 6632 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6633 6634 switch (j->op) { 6635 case SCTP_ADD_IP_ADDRESS: { 6636 uint32_t v_tag = pd->sctp_initiate_tag; 6637 6638 if (v_tag == 0) { 6639 if (s->direction == pd->dir) 6640 v_tag = s->src.scrub->pfss_v_tag; 6641 else 6642 v_tag = s->dst.scrub->pfss_v_tag; 6643 } 6644 6645 /* 6646 * Avoid duplicating states. We'll already have 6647 * created a state based on the source address of 6648 * the packet, but SCTP endpoints may also list this 6649 * address again in the INIT(_ACK) parameters. 6650 */ 6651 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6652 break; 6653 } 6654 6655 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6656 PF_RULES_RLOCK(); 6657 sm = NULL; 6658 /* 6659 * New connections need to be floating, because 6660 * we cannot know what interfaces it will use. 6661 * That's why we pass V_pfi_all rather than kif. 6662 */ 6663 ret = pf_test_rule(&r, &sm, V_pfi_all, 6664 j->m, off, &j->pd, &ra, &rs, NULL, 6665 sizeof(j->pd.hdr.sctp)); 6666 PF_RULES_RUNLOCK(); 6667 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6668 if (ret != PF_DROP && sm != NULL) { 6669 /* Inherit v_tag values. */ 6670 if (sm->direction == s->direction) { 6671 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6672 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6673 } else { 6674 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6675 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6676 } 6677 PF_STATE_UNLOCK(sm); 6678 } else { 6679 /* If we try duplicate inserts? */ 6680 break; 6681 } 6682 6683 /* Only add the address if we've actually allowed the state. */ 6684 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6685 6686 if (! do_extra) { 6687 break; 6688 } 6689 /* 6690 * We need to do this for each of our source addresses. 6691 * Find those based on the verification tag. 6692 */ 6693 struct pf_sctp_endpoint key = { 6694 .v_tag = pd->hdr.sctp.v_tag, 6695 }; 6696 struct pf_sctp_endpoint *ep; 6697 6698 PF_SCTP_ENDPOINTS_LOCK(); 6699 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6700 if (ep == NULL) { 6701 PF_SCTP_ENDPOINTS_UNLOCK(); 6702 break; 6703 } 6704 MPASS(ep != NULL); 6705 6706 TAILQ_FOREACH(i, &ep->sources, entry) { 6707 struct pf_sctp_multihome_job *nj; 6708 6709 /* SCTP can intermingle IPv4 and IPv6. */ 6710 if (i->af != pd->af) 6711 continue; 6712 6713 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6714 if (! nj) { 6715 continue; 6716 } 6717 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6718 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6719 nj->pd.src = &nj->src; 6720 // New destination address! 6721 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6722 nj->pd.dst = &nj->dst; 6723 nj->m = j->m; 6724 nj->op = j->op; 6725 6726 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6727 } 6728 PF_SCTP_ENDPOINTS_UNLOCK(); 6729 6730 break; 6731 } 6732 case SCTP_DEL_IP_ADDRESS: { 6733 struct pf_state_key_cmp key; 6734 uint8_t psrc; 6735 6736 bzero(&key, sizeof(key)); 6737 key.af = j->pd.af; 6738 key.proto = IPPROTO_SCTP; 6739 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6740 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6741 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6742 key.port[0] = j->pd.hdr.sctp.src_port; 6743 key.port[1] = j->pd.hdr.sctp.dest_port; 6744 } else { /* stack side, reverse */ 6745 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6746 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6747 key.port[1] = j->pd.hdr.sctp.src_port; 6748 key.port[0] = j->pd.hdr.sctp.dest_port; 6749 } 6750 6751 sm = pf_find_state(kif, &key, j->pd.dir); 6752 if (sm != NULL) { 6753 PF_STATE_LOCK_ASSERT(sm); 6754 if (j->pd.dir == sm->direction) { 6755 psrc = PF_PEER_SRC; 6756 } else { 6757 psrc = PF_PEER_DST; 6758 } 6759 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6760 sm->timeout = PFTM_SCTP_CLOSING; 6761 PF_STATE_UNLOCK(sm); 6762 } 6763 break; 6764 default: 6765 panic("Unknown op %#x", j->op); 6766 } 6767 } 6768 6769 free: 6770 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6771 free(j, M_PFTEMP); 6772 } 6773 6774 /* We may have inserted extra work while processing the list. */ 6775 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6776 do_extra = false; 6777 goto again; 6778 } 6779 } 6780 6781 static int 6782 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6783 struct pfi_kkif *kif, int op) 6784 { 6785 int off = 0; 6786 struct pf_sctp_multihome_job *job; 6787 6788 while (off < len) { 6789 struct sctp_paramhdr h; 6790 6791 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6792 pd->af)) 6793 return (PF_DROP); 6794 6795 /* Parameters are at least 4 bytes. */ 6796 if (ntohs(h.param_length) < 4) 6797 return (PF_DROP); 6798 6799 switch (ntohs(h.param_type)) { 6800 case SCTP_IPV4_ADDRESS: { 6801 struct in_addr t; 6802 6803 if (ntohs(h.param_length) != 6804 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6805 return (PF_DROP); 6806 6807 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6808 NULL, NULL, pd->af)) 6809 return (PF_DROP); 6810 6811 if (in_nullhost(t)) 6812 t.s_addr = pd->src->v4.s_addr; 6813 6814 /* 6815 * We hold the state lock (idhash) here, which means 6816 * that we can't acquire the keyhash, or we'll get a 6817 * LOR (and potentially double-lock things too). We also 6818 * can't release the state lock here, so instead we'll 6819 * enqueue this for async handling. 6820 * There's a relatively small race here, in that a 6821 * packet using the new addresses could arrive already, 6822 * but that's just though luck for it. 6823 */ 6824 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6825 if (! job) 6826 return (PF_DROP); 6827 6828 memcpy(&job->pd, pd, sizeof(*pd)); 6829 6830 // New source address! 6831 memcpy(&job->src, &t, sizeof(t)); 6832 job->pd.src = &job->src; 6833 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6834 job->pd.dst = &job->dst; 6835 job->m = m; 6836 job->op = op; 6837 6838 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6839 break; 6840 } 6841 #ifdef INET6 6842 case SCTP_IPV6_ADDRESS: { 6843 struct in6_addr t; 6844 6845 if (ntohs(h.param_length) != 6846 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6847 return (PF_DROP); 6848 6849 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6850 NULL, NULL, pd->af)) 6851 return (PF_DROP); 6852 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6853 break; 6854 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6855 memcpy(&t, &pd->src->v6, sizeof(t)); 6856 6857 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6858 if (! job) 6859 return (PF_DROP); 6860 6861 memcpy(&job->pd, pd, sizeof(*pd)); 6862 memcpy(&job->src, &t, sizeof(t)); 6863 job->pd.src = &job->src; 6864 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6865 job->pd.dst = &job->dst; 6866 job->m = m; 6867 job->op = op; 6868 6869 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6870 break; 6871 } 6872 #endif 6873 case SCTP_ADD_IP_ADDRESS: { 6874 int ret; 6875 struct sctp_asconf_paramhdr ah; 6876 6877 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6878 NULL, NULL, pd->af)) 6879 return (PF_DROP); 6880 6881 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6882 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6883 SCTP_ADD_IP_ADDRESS); 6884 if (ret != PF_PASS) 6885 return (ret); 6886 break; 6887 } 6888 case SCTP_DEL_IP_ADDRESS: { 6889 int ret; 6890 struct sctp_asconf_paramhdr ah; 6891 6892 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6893 NULL, NULL, pd->af)) 6894 return (PF_DROP); 6895 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6896 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6897 SCTP_DEL_IP_ADDRESS); 6898 if (ret != PF_PASS) 6899 return (ret); 6900 break; 6901 } 6902 default: 6903 break; 6904 } 6905 6906 off += roundup(ntohs(h.param_length), 4); 6907 } 6908 6909 return (PF_PASS); 6910 } 6911 int 6912 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6913 struct pfi_kkif *kif) 6914 { 6915 start += sizeof(struct sctp_init_chunk); 6916 len -= sizeof(struct sctp_init_chunk); 6917 6918 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6919 } 6920 6921 int 6922 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6923 struct pf_pdesc *pd, struct pfi_kkif *kif) 6924 { 6925 start += sizeof(struct sctp_asconf_chunk); 6926 len -= sizeof(struct sctp_asconf_chunk); 6927 6928 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6929 } 6930 6931 int 6932 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd, 6933 struct pf_kstate **state, struct mbuf *m, int off, int direction, 6934 struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir, 6935 int *iidx, int multi, int inner) 6936 { 6937 key->af = pd->af; 6938 key->proto = pd->proto; 6939 if (icmp_dir == PF_IN) { 6940 *iidx = pd->sidx; 6941 key->port[pd->sidx] = icmpid; 6942 key->port[pd->didx] = type; 6943 } else { 6944 *iidx = pd->didx; 6945 key->port[pd->sidx] = type; 6946 key->port[pd->didx] = icmpid; 6947 } 6948 if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src, 6949 pd->didx, pd->dst, multi)) 6950 return (PF_DROP); 6951 6952 STATE_LOOKUP(kif, key, *state, pd); 6953 6954 if ((*state)->state_flags & PFSTATE_SLOPPY) 6955 return (-1); 6956 6957 /* Is this ICMP message flowing in right direction? */ 6958 if ((*state)->rule.ptr->type && 6959 (((!inner && (*state)->direction == direction) || 6960 (inner && (*state)->direction != direction)) ? 6961 PF_IN : PF_OUT) != icmp_dir) { 6962 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6963 printf("pf: icmp type %d in wrong direction (%d): ", 6964 ntohs(type), icmp_dir); 6965 pf_print_state(*state); 6966 printf("\n"); 6967 } 6968 PF_STATE_UNLOCK(*state); 6969 *state = NULL; 6970 return (PF_DROP); 6971 } 6972 return (-1); 6973 } 6974 6975 static int 6976 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6977 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6978 { 6979 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6980 u_int16_t *icmpsum, virtual_id, virtual_type; 6981 u_int8_t icmptype, icmpcode; 6982 int icmp_dir, iidx, ret, multi; 6983 struct pf_state_key_cmp key; 6984 #ifdef INET 6985 u_int16_t icmpid; 6986 #endif 6987 6988 MPASS(*state == NULL); 6989 6990 bzero(&key, sizeof(key)); 6991 switch (pd->proto) { 6992 #ifdef INET 6993 case IPPROTO_ICMP: 6994 icmptype = pd->hdr.icmp.icmp_type; 6995 icmpcode = pd->hdr.icmp.icmp_code; 6996 icmpid = pd->hdr.icmp.icmp_id; 6997 icmpsum = &pd->hdr.icmp.icmp_cksum; 6998 break; 6999 #endif /* INET */ 7000 #ifdef INET6 7001 case IPPROTO_ICMPV6: 7002 icmptype = pd->hdr.icmp6.icmp6_type; 7003 icmpcode = pd->hdr.icmp6.icmp6_code; 7004 #ifdef INET 7005 icmpid = pd->hdr.icmp6.icmp6_id; 7006 #endif 7007 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 7008 break; 7009 #endif /* INET6 */ 7010 } 7011 7012 if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi, 7013 &virtual_id, &virtual_type) == 0) { 7014 /* 7015 * ICMP query/reply message not related to a TCP/UDP packet. 7016 * Search for an ICMP state. 7017 */ 7018 ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir, 7019 kif, virtual_id, virtual_type, icmp_dir, &iidx, 7020 PF_ICMP_MULTI_NONE, 0); 7021 if (ret >= 0) { 7022 MPASS(*state == NULL); 7023 if (ret == PF_DROP && pd->af == AF_INET6 && 7024 icmp_dir == PF_OUT) { 7025 ret = pf_icmp_state_lookup(&key, pd, state, m, off, 7026 pd->dir, kif, virtual_id, virtual_type, 7027 icmp_dir, &iidx, multi, 0); 7028 if (ret >= 0) { 7029 MPASS(*state == NULL); 7030 return (ret); 7031 } 7032 } else 7033 return (ret); 7034 } 7035 7036 (*state)->expire = pf_get_uptime(); 7037 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 7038 7039 /* translate source/destination address, if necessary */ 7040 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7041 struct pf_state_key *nk = (*state)->key[pd->didx]; 7042 7043 switch (pd->af) { 7044 #ifdef INET 7045 case AF_INET: 7046 if (PF_ANEQ(pd->src, 7047 &nk->addr[pd->sidx], AF_INET)) 7048 pf_change_a(&saddr->v4.s_addr, 7049 pd->ip_sum, 7050 nk->addr[pd->sidx].v4.s_addr, 0); 7051 7052 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 7053 AF_INET)) 7054 pf_change_a(&daddr->v4.s_addr, 7055 pd->ip_sum, 7056 nk->addr[pd->didx].v4.s_addr, 0); 7057 7058 if (nk->port[iidx] != 7059 pd->hdr.icmp.icmp_id) { 7060 pd->hdr.icmp.icmp_cksum = 7061 pf_cksum_fixup( 7062 pd->hdr.icmp.icmp_cksum, icmpid, 7063 nk->port[iidx], 0); 7064 pd->hdr.icmp.icmp_id = 7065 nk->port[iidx]; 7066 } 7067 7068 m_copyback(m, off, ICMP_MINLEN, 7069 (caddr_t )&pd->hdr.icmp); 7070 break; 7071 #endif /* INET */ 7072 #ifdef INET6 7073 case AF_INET6: 7074 if (PF_ANEQ(pd->src, 7075 &nk->addr[pd->sidx], AF_INET6)) 7076 pf_change_a6(saddr, 7077 &pd->hdr.icmp6.icmp6_cksum, 7078 &nk->addr[pd->sidx], 0); 7079 7080 if (PF_ANEQ(pd->dst, 7081 &nk->addr[pd->didx], AF_INET6)) 7082 pf_change_a6(daddr, 7083 &pd->hdr.icmp6.icmp6_cksum, 7084 &nk->addr[pd->didx], 0); 7085 7086 m_copyback(m, off, sizeof(struct icmp6_hdr), 7087 (caddr_t )&pd->hdr.icmp6); 7088 break; 7089 #endif /* INET6 */ 7090 } 7091 } 7092 return (PF_PASS); 7093 7094 } else { 7095 /* 7096 * ICMP error message in response to a TCP/UDP packet. 7097 * Extract the inner TCP/UDP header and search for that state. 7098 */ 7099 7100 struct pf_pdesc pd2; 7101 bzero(&pd2, sizeof pd2); 7102 #ifdef INET 7103 struct ip h2; 7104 #endif /* INET */ 7105 #ifdef INET6 7106 struct ip6_hdr h2_6; 7107 int terminal = 0; 7108 #endif /* INET6 */ 7109 int ipoff2 = 0; 7110 int off2 = 0; 7111 7112 pd2.af = pd->af; 7113 pd2.dir = pd->dir; 7114 /* Payload packet is from the opposite direction. */ 7115 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 7116 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 7117 switch (pd->af) { 7118 #ifdef INET 7119 case AF_INET: 7120 /* offset of h2 in mbuf chain */ 7121 ipoff2 = off + ICMP_MINLEN; 7122 7123 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 7124 NULL, reason, pd2.af)) { 7125 DPFPRINTF(PF_DEBUG_MISC, 7126 ("pf: ICMP error message too short " 7127 "(ip)\n")); 7128 return (PF_DROP); 7129 } 7130 /* 7131 * ICMP error messages don't refer to non-first 7132 * fragments 7133 */ 7134 if (h2.ip_off & htons(IP_OFFMASK)) { 7135 REASON_SET(reason, PFRES_FRAG); 7136 return (PF_DROP); 7137 } 7138 7139 /* offset of protocol header that follows h2 */ 7140 off2 = ipoff2 + (h2.ip_hl << 2); 7141 7142 pd2.proto = h2.ip_p; 7143 pd2.src = (struct pf_addr *)&h2.ip_src; 7144 pd2.dst = (struct pf_addr *)&h2.ip_dst; 7145 pd2.ip_sum = &h2.ip_sum; 7146 break; 7147 #endif /* INET */ 7148 #ifdef INET6 7149 case AF_INET6: 7150 ipoff2 = off + sizeof(struct icmp6_hdr); 7151 7152 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 7153 NULL, reason, pd2.af)) { 7154 DPFPRINTF(PF_DEBUG_MISC, 7155 ("pf: ICMP error message too short " 7156 "(ip6)\n")); 7157 return (PF_DROP); 7158 } 7159 pd2.proto = h2_6.ip6_nxt; 7160 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 7161 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 7162 pd2.ip_sum = NULL; 7163 off2 = ipoff2 + sizeof(h2_6); 7164 do { 7165 switch (pd2.proto) { 7166 case IPPROTO_FRAGMENT: 7167 /* 7168 * ICMPv6 error messages for 7169 * non-first fragments 7170 */ 7171 REASON_SET(reason, PFRES_FRAG); 7172 return (PF_DROP); 7173 case IPPROTO_AH: 7174 case IPPROTO_HOPOPTS: 7175 case IPPROTO_ROUTING: 7176 case IPPROTO_DSTOPTS: { 7177 /* get next header and header length */ 7178 struct ip6_ext opt6; 7179 7180 if (!pf_pull_hdr(m, off2, &opt6, 7181 sizeof(opt6), NULL, reason, 7182 pd2.af)) { 7183 DPFPRINTF(PF_DEBUG_MISC, 7184 ("pf: ICMPv6 short opt\n")); 7185 return (PF_DROP); 7186 } 7187 if (pd2.proto == IPPROTO_AH) 7188 off2 += (opt6.ip6e_len + 2) * 4; 7189 else 7190 off2 += (opt6.ip6e_len + 1) * 8; 7191 pd2.proto = opt6.ip6e_nxt; 7192 /* goto the next header */ 7193 break; 7194 } 7195 default: 7196 terminal++; 7197 break; 7198 } 7199 } while (!terminal); 7200 break; 7201 #endif /* INET6 */ 7202 } 7203 7204 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 7205 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7206 printf("pf: BAD ICMP %d:%d outer dst: ", 7207 icmptype, icmpcode); 7208 pf_print_host(pd->src, 0, pd->af); 7209 printf(" -> "); 7210 pf_print_host(pd->dst, 0, pd->af); 7211 printf(" inner src: "); 7212 pf_print_host(pd2.src, 0, pd2.af); 7213 printf(" -> "); 7214 pf_print_host(pd2.dst, 0, pd2.af); 7215 printf("\n"); 7216 } 7217 REASON_SET(reason, PFRES_BADSTATE); 7218 return (PF_DROP); 7219 } 7220 7221 switch (pd2.proto) { 7222 case IPPROTO_TCP: { 7223 struct tcphdr th; 7224 u_int32_t seq; 7225 struct pf_state_peer *src, *dst; 7226 u_int8_t dws; 7227 int copyback = 0; 7228 7229 /* 7230 * Only the first 8 bytes of the TCP header can be 7231 * expected. Don't access any TCP header fields after 7232 * th_seq, an ackskew test is not possible. 7233 */ 7234 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 7235 pd2.af)) { 7236 DPFPRINTF(PF_DEBUG_MISC, 7237 ("pf: ICMP error message too short " 7238 "(tcp)\n")); 7239 return (PF_DROP); 7240 } 7241 7242 key.af = pd2.af; 7243 key.proto = IPPROTO_TCP; 7244 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7245 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7246 key.port[pd2.sidx] = th.th_sport; 7247 key.port[pd2.didx] = th.th_dport; 7248 7249 STATE_LOOKUP(kif, &key, *state, pd); 7250 7251 if (pd->dir == (*state)->direction) { 7252 src = &(*state)->dst; 7253 dst = &(*state)->src; 7254 } else { 7255 src = &(*state)->src; 7256 dst = &(*state)->dst; 7257 } 7258 7259 if (src->wscale && dst->wscale) 7260 dws = dst->wscale & PF_WSCALE_MASK; 7261 else 7262 dws = 0; 7263 7264 /* Demodulate sequence number */ 7265 seq = ntohl(th.th_seq) - src->seqdiff; 7266 if (src->seqdiff) { 7267 pf_change_a(&th.th_seq, icmpsum, 7268 htonl(seq), 0); 7269 copyback = 1; 7270 } 7271 7272 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 7273 (!SEQ_GEQ(src->seqhi, seq) || 7274 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 7275 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7276 printf("pf: BAD ICMP %d:%d ", 7277 icmptype, icmpcode); 7278 pf_print_host(pd->src, 0, pd->af); 7279 printf(" -> "); 7280 pf_print_host(pd->dst, 0, pd->af); 7281 printf(" state: "); 7282 pf_print_state(*state); 7283 printf(" seq=%u\n", seq); 7284 } 7285 REASON_SET(reason, PFRES_BADSTATE); 7286 return (PF_DROP); 7287 } else { 7288 if (V_pf_status.debug >= PF_DEBUG_MISC) { 7289 printf("pf: OK ICMP %d:%d ", 7290 icmptype, icmpcode); 7291 pf_print_host(pd->src, 0, pd->af); 7292 printf(" -> "); 7293 pf_print_host(pd->dst, 0, pd->af); 7294 printf(" state: "); 7295 pf_print_state(*state); 7296 printf(" seq=%u\n", seq); 7297 } 7298 } 7299 7300 /* translate source/destination address, if necessary */ 7301 if ((*state)->key[PF_SK_WIRE] != 7302 (*state)->key[PF_SK_STACK]) { 7303 struct pf_state_key *nk = 7304 (*state)->key[pd->didx]; 7305 7306 if (PF_ANEQ(pd2.src, 7307 &nk->addr[pd2.sidx], pd2.af) || 7308 nk->port[pd2.sidx] != th.th_sport) 7309 pf_change_icmp(pd2.src, &th.th_sport, 7310 daddr, &nk->addr[pd2.sidx], 7311 nk->port[pd2.sidx], NULL, 7312 pd2.ip_sum, icmpsum, 7313 pd->ip_sum, 0, pd2.af); 7314 7315 if (PF_ANEQ(pd2.dst, 7316 &nk->addr[pd2.didx], pd2.af) || 7317 nk->port[pd2.didx] != th.th_dport) 7318 pf_change_icmp(pd2.dst, &th.th_dport, 7319 saddr, &nk->addr[pd2.didx], 7320 nk->port[pd2.didx], NULL, 7321 pd2.ip_sum, icmpsum, 7322 pd->ip_sum, 0, pd2.af); 7323 copyback = 1; 7324 } 7325 7326 if (copyback) { 7327 switch (pd2.af) { 7328 #ifdef INET 7329 case AF_INET: 7330 m_copyback(m, off, ICMP_MINLEN, 7331 (caddr_t )&pd->hdr.icmp); 7332 m_copyback(m, ipoff2, sizeof(h2), 7333 (caddr_t )&h2); 7334 break; 7335 #endif /* INET */ 7336 #ifdef INET6 7337 case AF_INET6: 7338 m_copyback(m, off, 7339 sizeof(struct icmp6_hdr), 7340 (caddr_t )&pd->hdr.icmp6); 7341 m_copyback(m, ipoff2, sizeof(h2_6), 7342 (caddr_t )&h2_6); 7343 break; 7344 #endif /* INET6 */ 7345 } 7346 m_copyback(m, off2, 8, (caddr_t)&th); 7347 } 7348 7349 return (PF_PASS); 7350 break; 7351 } 7352 case IPPROTO_UDP: { 7353 struct udphdr uh; 7354 7355 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 7356 NULL, reason, pd2.af)) { 7357 DPFPRINTF(PF_DEBUG_MISC, 7358 ("pf: ICMP error message too short " 7359 "(udp)\n")); 7360 return (PF_DROP); 7361 } 7362 7363 key.af = pd2.af; 7364 key.proto = IPPROTO_UDP; 7365 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7366 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7367 key.port[pd2.sidx] = uh.uh_sport; 7368 key.port[pd2.didx] = uh.uh_dport; 7369 7370 STATE_LOOKUP(kif, &key, *state, pd); 7371 7372 /* translate source/destination address, if necessary */ 7373 if ((*state)->key[PF_SK_WIRE] != 7374 (*state)->key[PF_SK_STACK]) { 7375 struct pf_state_key *nk = 7376 (*state)->key[pd->didx]; 7377 7378 if (PF_ANEQ(pd2.src, 7379 &nk->addr[pd2.sidx], pd2.af) || 7380 nk->port[pd2.sidx] != uh.uh_sport) 7381 pf_change_icmp(pd2.src, &uh.uh_sport, 7382 daddr, &nk->addr[pd2.sidx], 7383 nk->port[pd2.sidx], &uh.uh_sum, 7384 pd2.ip_sum, icmpsum, 7385 pd->ip_sum, 1, pd2.af); 7386 7387 if (PF_ANEQ(pd2.dst, 7388 &nk->addr[pd2.didx], pd2.af) || 7389 nk->port[pd2.didx] != uh.uh_dport) 7390 pf_change_icmp(pd2.dst, &uh.uh_dport, 7391 saddr, &nk->addr[pd2.didx], 7392 nk->port[pd2.didx], &uh.uh_sum, 7393 pd2.ip_sum, icmpsum, 7394 pd->ip_sum, 1, pd2.af); 7395 7396 switch (pd2.af) { 7397 #ifdef INET 7398 case AF_INET: 7399 m_copyback(m, off, ICMP_MINLEN, 7400 (caddr_t )&pd->hdr.icmp); 7401 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7402 break; 7403 #endif /* INET */ 7404 #ifdef INET6 7405 case AF_INET6: 7406 m_copyback(m, off, 7407 sizeof(struct icmp6_hdr), 7408 (caddr_t )&pd->hdr.icmp6); 7409 m_copyback(m, ipoff2, sizeof(h2_6), 7410 (caddr_t )&h2_6); 7411 break; 7412 #endif /* INET6 */ 7413 } 7414 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 7415 } 7416 return (PF_PASS); 7417 break; 7418 } 7419 #ifdef INET 7420 case IPPROTO_ICMP: { 7421 struct icmp *iih = &pd2.hdr.icmp; 7422 7423 if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN, 7424 NULL, reason, pd2.af)) { 7425 DPFPRINTF(PF_DEBUG_MISC, 7426 ("pf: ICMP error message too short i" 7427 "(icmp)\n")); 7428 return (PF_DROP); 7429 } 7430 7431 icmpid = iih->icmp_id; 7432 pf_icmp_mapping(&pd2, iih->icmp_type, 7433 &icmp_dir, &multi, &virtual_id, &virtual_type); 7434 7435 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7436 pd2.dir, kif, virtual_id, virtual_type, 7437 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7438 if (ret >= 0) { 7439 MPASS(*state == NULL); 7440 return (ret); 7441 } 7442 7443 /* translate source/destination address, if necessary */ 7444 if ((*state)->key[PF_SK_WIRE] != 7445 (*state)->key[PF_SK_STACK]) { 7446 struct pf_state_key *nk = 7447 (*state)->key[pd->didx]; 7448 7449 if (PF_ANEQ(pd2.src, 7450 &nk->addr[pd2.sidx], pd2.af) || 7451 (virtual_type == htons(ICMP_ECHO) && 7452 nk->port[iidx] != iih->icmp_id)) 7453 pf_change_icmp(pd2.src, 7454 (virtual_type == htons(ICMP_ECHO)) ? 7455 &iih->icmp_id : NULL, 7456 daddr, &nk->addr[pd2.sidx], 7457 (virtual_type == htons(ICMP_ECHO)) ? 7458 nk->port[iidx] : 0, NULL, 7459 pd2.ip_sum, icmpsum, 7460 pd->ip_sum, 0, AF_INET); 7461 7462 if (PF_ANEQ(pd2.dst, 7463 &nk->addr[pd2.didx], pd2.af)) 7464 pf_change_icmp(pd2.dst, NULL, NULL, 7465 &nk->addr[pd2.didx], 0, NULL, 7466 pd2.ip_sum, icmpsum, pd->ip_sum, 0, 7467 AF_INET); 7468 7469 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 7470 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7471 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih); 7472 } 7473 return (PF_PASS); 7474 break; 7475 } 7476 #endif /* INET */ 7477 #ifdef INET6 7478 case IPPROTO_ICMPV6: { 7479 struct icmp6_hdr *iih = &pd2.hdr.icmp6; 7480 7481 if (!pf_pull_hdr(m, off2, iih, 7482 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 7483 DPFPRINTF(PF_DEBUG_MISC, 7484 ("pf: ICMP error message too short " 7485 "(icmp6)\n")); 7486 return (PF_DROP); 7487 } 7488 7489 pf_icmp_mapping(&pd2, iih->icmp6_type, 7490 &icmp_dir, &multi, &virtual_id, &virtual_type); 7491 7492 ret = pf_icmp_state_lookup(&key, &pd2, state, m, off, 7493 pd->dir, kif, virtual_id, virtual_type, 7494 icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1); 7495 if (ret >= 0) { 7496 MPASS(*state == NULL); 7497 if (ret == PF_DROP && pd2.af == AF_INET6 && 7498 icmp_dir == PF_OUT) { 7499 ret = pf_icmp_state_lookup(&key, &pd2, 7500 state, m, off, pd->dir, kif, 7501 virtual_id, virtual_type, 7502 icmp_dir, &iidx, multi, 1); 7503 if (ret >= 0) { 7504 MPASS(*state == NULL); 7505 return (ret); 7506 } 7507 } else 7508 return (ret); 7509 } 7510 7511 /* translate source/destination address, if necessary */ 7512 if ((*state)->key[PF_SK_WIRE] != 7513 (*state)->key[PF_SK_STACK]) { 7514 struct pf_state_key *nk = 7515 (*state)->key[pd->didx]; 7516 7517 if (PF_ANEQ(pd2.src, 7518 &nk->addr[pd2.sidx], pd2.af) || 7519 ((virtual_type == htons(ICMP6_ECHO_REQUEST)) && 7520 nk->port[pd2.sidx] != iih->icmp6_id)) 7521 pf_change_icmp(pd2.src, 7522 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7523 ? &iih->icmp6_id : NULL, 7524 daddr, &nk->addr[pd2.sidx], 7525 (virtual_type == htons(ICMP6_ECHO_REQUEST)) 7526 ? nk->port[iidx] : 0, NULL, 7527 pd2.ip_sum, icmpsum, 7528 pd->ip_sum, 0, AF_INET6); 7529 7530 if (PF_ANEQ(pd2.dst, 7531 &nk->addr[pd2.didx], pd2.af)) 7532 pf_change_icmp(pd2.dst, NULL, NULL, 7533 &nk->addr[pd2.didx], 0, NULL, 7534 pd2.ip_sum, icmpsum, 7535 pd->ip_sum, 0, AF_INET6); 7536 7537 m_copyback(m, off, sizeof(struct icmp6_hdr), 7538 (caddr_t)&pd->hdr.icmp6); 7539 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 7540 m_copyback(m, off2, sizeof(struct icmp6_hdr), 7541 (caddr_t)iih); 7542 } 7543 return (PF_PASS); 7544 break; 7545 } 7546 #endif /* INET6 */ 7547 default: { 7548 key.af = pd2.af; 7549 key.proto = pd2.proto; 7550 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7551 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7552 key.port[0] = key.port[1] = 0; 7553 7554 STATE_LOOKUP(kif, &key, *state, pd); 7555 7556 /* translate source/destination address, if necessary */ 7557 if ((*state)->key[PF_SK_WIRE] != 7558 (*state)->key[PF_SK_STACK]) { 7559 struct pf_state_key *nk = 7560 (*state)->key[pd->didx]; 7561 7562 if (PF_ANEQ(pd2.src, 7563 &nk->addr[pd2.sidx], pd2.af)) 7564 pf_change_icmp(pd2.src, NULL, daddr, 7565 &nk->addr[pd2.sidx], 0, NULL, 7566 pd2.ip_sum, icmpsum, 7567 pd->ip_sum, 0, pd2.af); 7568 7569 if (PF_ANEQ(pd2.dst, 7570 &nk->addr[pd2.didx], pd2.af)) 7571 pf_change_icmp(pd2.dst, NULL, saddr, 7572 &nk->addr[pd2.didx], 0, NULL, 7573 pd2.ip_sum, icmpsum, 7574 pd->ip_sum, 0, pd2.af); 7575 7576 switch (pd2.af) { 7577 #ifdef INET 7578 case AF_INET: 7579 m_copyback(m, off, ICMP_MINLEN, 7580 (caddr_t)&pd->hdr.icmp); 7581 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7582 break; 7583 #endif /* INET */ 7584 #ifdef INET6 7585 case AF_INET6: 7586 m_copyback(m, off, 7587 sizeof(struct icmp6_hdr), 7588 (caddr_t )&pd->hdr.icmp6); 7589 m_copyback(m, ipoff2, sizeof(h2_6), 7590 (caddr_t )&h2_6); 7591 break; 7592 #endif /* INET6 */ 7593 } 7594 } 7595 return (PF_PASS); 7596 break; 7597 } 7598 } 7599 } 7600 } 7601 7602 static int 7603 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7604 struct mbuf *m, struct pf_pdesc *pd) 7605 { 7606 struct pf_state_peer *src, *dst; 7607 struct pf_state_key_cmp key; 7608 uint8_t psrc, pdst; 7609 7610 bzero(&key, sizeof(key)); 7611 key.af = pd->af; 7612 key.proto = pd->proto; 7613 if (pd->dir == PF_IN) { 7614 PF_ACPY(&key.addr[0], pd->src, key.af); 7615 PF_ACPY(&key.addr[1], pd->dst, key.af); 7616 key.port[0] = key.port[1] = 0; 7617 } else { 7618 PF_ACPY(&key.addr[1], pd->src, key.af); 7619 PF_ACPY(&key.addr[0], pd->dst, key.af); 7620 key.port[1] = key.port[0] = 0; 7621 } 7622 7623 STATE_LOOKUP(kif, &key, *state, pd); 7624 7625 if (pd->dir == (*state)->direction) { 7626 src = &(*state)->src; 7627 dst = &(*state)->dst; 7628 psrc = PF_PEER_SRC; 7629 pdst = PF_PEER_DST; 7630 } else { 7631 src = &(*state)->dst; 7632 dst = &(*state)->src; 7633 psrc = PF_PEER_DST; 7634 pdst = PF_PEER_SRC; 7635 } 7636 7637 /* update states */ 7638 if (src->state < PFOTHERS_SINGLE) 7639 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7640 if (dst->state == PFOTHERS_SINGLE) 7641 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7642 7643 /* update expire time */ 7644 (*state)->expire = pf_get_uptime(); 7645 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7646 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7647 else 7648 (*state)->timeout = PFTM_OTHER_SINGLE; 7649 7650 /* translate source/destination address, if necessary */ 7651 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7652 struct pf_state_key *nk = (*state)->key[pd->didx]; 7653 7654 KASSERT(nk, ("%s: nk is null", __func__)); 7655 KASSERT(pd, ("%s: pd is null", __func__)); 7656 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7657 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7658 switch (pd->af) { 7659 #ifdef INET 7660 case AF_INET: 7661 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7662 pf_change_a(&pd->src->v4.s_addr, 7663 pd->ip_sum, 7664 nk->addr[pd->sidx].v4.s_addr, 7665 0); 7666 7667 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7668 pf_change_a(&pd->dst->v4.s_addr, 7669 pd->ip_sum, 7670 nk->addr[pd->didx].v4.s_addr, 7671 0); 7672 7673 break; 7674 #endif /* INET */ 7675 #ifdef INET6 7676 case AF_INET6: 7677 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) 7678 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7679 7680 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) 7681 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7682 #endif /* INET6 */ 7683 } 7684 } 7685 return (PF_PASS); 7686 } 7687 7688 /* 7689 * ipoff and off are measured from the start of the mbuf chain. 7690 * h must be at "ipoff" on the mbuf chain. 7691 */ 7692 void * 7693 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len, 7694 u_short *actionp, u_short *reasonp, sa_family_t af) 7695 { 7696 switch (af) { 7697 #ifdef INET 7698 case AF_INET: { 7699 const struct ip *h = mtod(m, struct ip *); 7700 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7701 7702 if (fragoff) { 7703 if (fragoff >= len) 7704 ACTION_SET(actionp, PF_PASS); 7705 else { 7706 ACTION_SET(actionp, PF_DROP); 7707 REASON_SET(reasonp, PFRES_FRAG); 7708 } 7709 return (NULL); 7710 } 7711 if (m->m_pkthdr.len < off + len || 7712 ntohs(h->ip_len) < off + len) { 7713 ACTION_SET(actionp, PF_DROP); 7714 REASON_SET(reasonp, PFRES_SHORT); 7715 return (NULL); 7716 } 7717 break; 7718 } 7719 #endif /* INET */ 7720 #ifdef INET6 7721 case AF_INET6: { 7722 const struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7723 7724 if (m->m_pkthdr.len < off + len || 7725 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7726 (unsigned)(off + len)) { 7727 ACTION_SET(actionp, PF_DROP); 7728 REASON_SET(reasonp, PFRES_SHORT); 7729 return (NULL); 7730 } 7731 break; 7732 } 7733 #endif /* INET6 */ 7734 } 7735 m_copydata(m, off, len, p); 7736 return (p); 7737 } 7738 7739 int 7740 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7741 int rtableid) 7742 { 7743 struct ifnet *ifp; 7744 7745 /* 7746 * Skip check for addresses with embedded interface scope, 7747 * as they would always match anyway. 7748 */ 7749 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7750 return (1); 7751 7752 if (af != AF_INET && af != AF_INET6) 7753 return (0); 7754 7755 if (kif == V_pfi_all) 7756 return (1); 7757 7758 /* Skip checks for ipsec interfaces */ 7759 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7760 return (1); 7761 7762 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7763 7764 switch (af) { 7765 #ifdef INET6 7766 case AF_INET6: 7767 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7768 ifp)); 7769 #endif 7770 #ifdef INET 7771 case AF_INET: 7772 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7773 ifp)); 7774 #endif 7775 } 7776 7777 return (0); 7778 } 7779 7780 #ifdef INET 7781 static void 7782 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7783 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7784 { 7785 struct mbuf *m0, *m1, *md; 7786 struct sockaddr_in dst; 7787 struct ip *ip; 7788 struct pfi_kkif *nkif = NULL; 7789 struct ifnet *ifp = NULL; 7790 struct pf_addr naddr; 7791 struct pf_ksrc_node *sn = NULL; 7792 int error = 0; 7793 uint16_t ip_len, ip_off; 7794 uint16_t tmp; 7795 int r_rt, r_dir; 7796 7797 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7798 7799 if (s) { 7800 r_rt = s->rt; 7801 r_dir = s->direction; 7802 } else { 7803 r_rt = r->rt; 7804 r_dir = r->direction; 7805 } 7806 7807 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7808 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7809 __func__)); 7810 7811 if ((pd->pf_mtag == NULL && 7812 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7813 pd->pf_mtag->routed++ > 3) { 7814 m0 = *m; 7815 *m = NULL; 7816 goto bad_locked; 7817 } 7818 7819 if (r_rt == PF_DUPTO) { 7820 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7821 if (s == NULL) { 7822 ifp = r->rpool.cur->kif ? 7823 r->rpool.cur->kif->pfik_ifp : NULL; 7824 } else { 7825 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7826 /* If pfsync'd */ 7827 if (ifp == NULL && r->rpool.cur != NULL) 7828 ifp = r->rpool.cur->kif ? 7829 r->rpool.cur->kif->pfik_ifp : NULL; 7830 PF_STATE_UNLOCK(s); 7831 } 7832 if (ifp == oifp) { 7833 /* When the 2nd interface is not skipped */ 7834 return; 7835 } else { 7836 m0 = *m; 7837 *m = NULL; 7838 goto bad; 7839 } 7840 } else { 7841 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7842 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7843 if (s) 7844 PF_STATE_UNLOCK(s); 7845 return; 7846 } 7847 } 7848 } else { 7849 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7850 pf_dummynet(pd, s, r, m); 7851 if (s) 7852 PF_STATE_UNLOCK(s); 7853 return; 7854 } 7855 m0 = *m; 7856 } 7857 7858 ip = mtod(m0, struct ip *); 7859 7860 bzero(&dst, sizeof(dst)); 7861 dst.sin_family = AF_INET; 7862 dst.sin_len = sizeof(dst); 7863 dst.sin_addr = ip->ip_dst; 7864 7865 bzero(&naddr, sizeof(naddr)); 7866 7867 if (s == NULL) { 7868 if (TAILQ_EMPTY(&r->rpool.list)) { 7869 DPFPRINTF(PF_DEBUG_URGENT, 7870 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7871 goto bad_locked; 7872 } 7873 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7874 &naddr, &nkif, NULL, &sn); 7875 if (!PF_AZERO(&naddr, AF_INET)) 7876 dst.sin_addr.s_addr = naddr.v4.s_addr; 7877 ifp = nkif ? nkif->pfik_ifp : NULL; 7878 } else { 7879 struct pfi_kkif *kif; 7880 7881 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7882 dst.sin_addr.s_addr = 7883 s->rt_addr.v4.s_addr; 7884 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7885 kif = s->rt_kif; 7886 /* If pfsync'd */ 7887 if (ifp == NULL && r->rpool.cur != NULL) { 7888 ifp = r->rpool.cur->kif ? 7889 r->rpool.cur->kif->pfik_ifp : NULL; 7890 kif = r->rpool.cur->kif; 7891 } 7892 if (ifp != NULL && kif != NULL && 7893 r->rule_flag & PFRULE_IFBOUND && 7894 r->rt == PF_REPLYTO && 7895 s->kif == V_pfi_all) { 7896 s->kif = kif; 7897 s->orig_kif = oifp->if_pf_kif; 7898 } 7899 7900 PF_STATE_UNLOCK(s); 7901 } 7902 7903 if (ifp == NULL) 7904 goto bad; 7905 7906 if (pd->dir == PF_IN) { 7907 if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7908 goto bad; 7909 else if (m0 == NULL) 7910 goto done; 7911 if (m0->m_len < sizeof(struct ip)) { 7912 DPFPRINTF(PF_DEBUG_URGENT, 7913 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7914 goto bad; 7915 } 7916 ip = mtod(m0, struct ip *); 7917 } 7918 7919 if (ifp->if_flags & IFF_LOOPBACK) 7920 m0->m_flags |= M_SKIP_FIREWALL; 7921 7922 ip_len = ntohs(ip->ip_len); 7923 ip_off = ntohs(ip->ip_off); 7924 7925 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7926 m0->m_pkthdr.csum_flags |= CSUM_IP; 7927 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7928 in_delayed_cksum(m0); 7929 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7930 } 7931 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7932 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7933 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7934 } 7935 7936 if (pd->dir == PF_IN) { 7937 /* 7938 * Make sure dummynet gets the correct direction, in case it needs to 7939 * re-inject later. 7940 */ 7941 pd->dir = PF_OUT; 7942 7943 /* 7944 * The following processing is actually the rest of the inbound processing, even 7945 * though we've marked it as outbound (so we don't look through dummynet) and it 7946 * happens after the outbound processing (pf_test(PF_OUT) above). 7947 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7948 * conclusion about what direction it's processing, and we can't fix it or it 7949 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7950 * decision will pick the right pipe, and everything will mostly work as expected. 7951 */ 7952 tmp = pd->act.dnrpipe; 7953 pd->act.dnrpipe = pd->act.dnpipe; 7954 pd->act.dnpipe = tmp; 7955 } 7956 7957 /* 7958 * If small enough for interface, or the interface will take 7959 * care of the fragmentation for us, we can just send directly. 7960 */ 7961 if (ip_len <= ifp->if_mtu || 7962 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7963 ip->ip_sum = 0; 7964 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7965 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7966 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7967 } 7968 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7969 7970 md = m0; 7971 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7972 if (md != NULL) 7973 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7974 goto done; 7975 } 7976 7977 /* Balk when DF bit is set or the interface didn't support TSO. */ 7978 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7979 error = EMSGSIZE; 7980 KMOD_IPSTAT_INC(ips_cantfrag); 7981 if (r_rt != PF_DUPTO) { 7982 if (s && pd->nat_rule != NULL) 7983 PACKET_UNDO_NAT(m0, pd, 7984 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7985 s); 7986 7987 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7988 ifp->if_mtu); 7989 goto done; 7990 } else 7991 goto bad; 7992 } 7993 7994 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7995 if (error) 7996 goto bad; 7997 7998 for (; m0; m0 = m1) { 7999 m1 = m0->m_nextpkt; 8000 m0->m_nextpkt = NULL; 8001 if (error == 0) { 8002 m_clrprotoflags(m0); 8003 md = m0; 8004 pd->pf_mtag = pf_find_mtag(md); 8005 error = pf_dummynet_route(pd, s, r, ifp, 8006 sintosa(&dst), &md); 8007 if (md != NULL) 8008 error = (*ifp->if_output)(ifp, md, 8009 sintosa(&dst), NULL); 8010 } else 8011 m_freem(m0); 8012 } 8013 8014 if (error == 0) 8015 KMOD_IPSTAT_INC(ips_fragmented); 8016 8017 done: 8018 if (r_rt != PF_DUPTO) 8019 *m = NULL; 8020 return; 8021 8022 bad_locked: 8023 if (s) 8024 PF_STATE_UNLOCK(s); 8025 bad: 8026 m_freem(m0); 8027 goto done; 8028 } 8029 #endif /* INET */ 8030 8031 #ifdef INET6 8032 static void 8033 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 8034 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 8035 { 8036 struct mbuf *m0, *md; 8037 struct sockaddr_in6 dst; 8038 struct ip6_hdr *ip6; 8039 struct pfi_kkif *nkif = NULL; 8040 struct ifnet *ifp = NULL; 8041 struct pf_addr naddr; 8042 struct pf_ksrc_node *sn = NULL; 8043 int r_rt, r_dir; 8044 8045 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 8046 8047 if (s) { 8048 r_rt = s->rt; 8049 r_dir = s->direction; 8050 } else { 8051 r_rt = r->rt; 8052 r_dir = r->direction; 8053 } 8054 8055 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 8056 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 8057 __func__)); 8058 8059 if ((pd->pf_mtag == NULL && 8060 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 8061 pd->pf_mtag->routed++ > 3) { 8062 m0 = *m; 8063 *m = NULL; 8064 goto bad_locked; 8065 } 8066 8067 if (r_rt == PF_DUPTO) { 8068 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 8069 if (s == NULL) { 8070 ifp = r->rpool.cur->kif ? 8071 r->rpool.cur->kif->pfik_ifp : NULL; 8072 } else { 8073 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8074 /* If pfsync'd */ 8075 if (ifp == NULL && r->rpool.cur != NULL) 8076 ifp = r->rpool.cur->kif ? 8077 r->rpool.cur->kif->pfik_ifp : NULL; 8078 PF_STATE_UNLOCK(s); 8079 } 8080 if (ifp == oifp) { 8081 /* When the 2nd interface is not skipped */ 8082 return; 8083 } else { 8084 m0 = *m; 8085 *m = NULL; 8086 goto bad; 8087 } 8088 } else { 8089 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 8090 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 8091 if (s) 8092 PF_STATE_UNLOCK(s); 8093 return; 8094 } 8095 } 8096 } else { 8097 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 8098 pf_dummynet(pd, s, r, m); 8099 if (s) 8100 PF_STATE_UNLOCK(s); 8101 return; 8102 } 8103 m0 = *m; 8104 } 8105 8106 ip6 = mtod(m0, struct ip6_hdr *); 8107 8108 bzero(&dst, sizeof(dst)); 8109 dst.sin6_family = AF_INET6; 8110 dst.sin6_len = sizeof(dst); 8111 dst.sin6_addr = ip6->ip6_dst; 8112 8113 bzero(&naddr, sizeof(naddr)); 8114 8115 if (s == NULL) { 8116 if (TAILQ_EMPTY(&r->rpool.list)) { 8117 DPFPRINTF(PF_DEBUG_URGENT, 8118 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 8119 goto bad_locked; 8120 } 8121 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 8122 &naddr, &nkif, NULL, &sn); 8123 if (!PF_AZERO(&naddr, AF_INET6)) 8124 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8125 &naddr, AF_INET6); 8126 ifp = nkif ? nkif->pfik_ifp : NULL; 8127 } else { 8128 struct pfi_kkif *kif; 8129 8130 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 8131 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 8132 &s->rt_addr, AF_INET6); 8133 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 8134 kif = s->rt_kif; 8135 /* If pfsync'd */ 8136 if (ifp == NULL && r->rpool.cur != NULL) { 8137 ifp = r->rpool.cur->kif ? 8138 r->rpool.cur->kif->pfik_ifp : NULL; 8139 kif = r->rpool.cur->kif; 8140 } 8141 if (ifp != NULL && kif != NULL && 8142 r->rule_flag & PFRULE_IFBOUND && 8143 r->rt == PF_REPLYTO && 8144 s->kif == V_pfi_all) { 8145 s->kif = kif; 8146 s->orig_kif = oifp->if_pf_kif; 8147 } 8148 } 8149 8150 if (s) 8151 PF_STATE_UNLOCK(s); 8152 8153 if (ifp == NULL) 8154 goto bad; 8155 8156 if (pd->dir == PF_IN) { 8157 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 8158 goto bad; 8159 else if (m0 == NULL) 8160 goto done; 8161 if (m0->m_len < sizeof(struct ip6_hdr)) { 8162 DPFPRINTF(PF_DEBUG_URGENT, 8163 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 8164 __func__)); 8165 goto bad; 8166 } 8167 ip6 = mtod(m0, struct ip6_hdr *); 8168 } 8169 8170 if (ifp->if_flags & IFF_LOOPBACK) 8171 m0->m_flags |= M_SKIP_FIREWALL; 8172 8173 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 8174 ~ifp->if_hwassist) { 8175 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 8176 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 8177 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 8178 } 8179 8180 /* 8181 * If the packet is too large for the outgoing interface, 8182 * send back an icmp6 error. 8183 */ 8184 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 8185 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 8186 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 8187 md = m0; 8188 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 8189 if (md != NULL) 8190 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 8191 } 8192 else { 8193 in6_ifstat_inc(ifp, ifs6_in_toobig); 8194 if (r_rt != PF_DUPTO) { 8195 if (s && pd->nat_rule != NULL) 8196 PACKET_UNDO_NAT(m0, pd, 8197 ((caddr_t)ip6 - m0->m_data) + 8198 sizeof(struct ip6_hdr), s); 8199 8200 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 8201 } else 8202 goto bad; 8203 } 8204 8205 done: 8206 if (r_rt != PF_DUPTO) 8207 *m = NULL; 8208 return; 8209 8210 bad_locked: 8211 if (s) 8212 PF_STATE_UNLOCK(s); 8213 bad: 8214 m_freem(m0); 8215 goto done; 8216 } 8217 #endif /* INET6 */ 8218 8219 /* 8220 * FreeBSD supports cksum offloads for the following drivers. 8221 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 8222 * 8223 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 8224 * network driver performed cksum including pseudo header, need to verify 8225 * csum_data 8226 * CSUM_DATA_VALID : 8227 * network driver performed cksum, needs to additional pseudo header 8228 * cksum computation with partial csum_data(i.e. lack of H/W support for 8229 * pseudo header, for instance sk(4) and possibly gem(4)) 8230 * 8231 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 8232 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 8233 * TCP/UDP layer. 8234 * Also, set csum_data to 0xffff to force cksum validation. 8235 */ 8236 static int 8237 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 8238 { 8239 u_int16_t sum = 0; 8240 int hw_assist = 0; 8241 struct ip *ip; 8242 8243 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 8244 return (1); 8245 if (m->m_pkthdr.len < off + len) 8246 return (1); 8247 8248 switch (p) { 8249 case IPPROTO_TCP: 8250 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8251 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8252 sum = m->m_pkthdr.csum_data; 8253 } else { 8254 ip = mtod(m, struct ip *); 8255 sum = in_pseudo(ip->ip_src.s_addr, 8256 ip->ip_dst.s_addr, htonl((u_short)len + 8257 m->m_pkthdr.csum_data + IPPROTO_TCP)); 8258 } 8259 sum ^= 0xffff; 8260 ++hw_assist; 8261 } 8262 break; 8263 case IPPROTO_UDP: 8264 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 8265 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 8266 sum = m->m_pkthdr.csum_data; 8267 } else { 8268 ip = mtod(m, struct ip *); 8269 sum = in_pseudo(ip->ip_src.s_addr, 8270 ip->ip_dst.s_addr, htonl((u_short)len + 8271 m->m_pkthdr.csum_data + IPPROTO_UDP)); 8272 } 8273 sum ^= 0xffff; 8274 ++hw_assist; 8275 } 8276 break; 8277 case IPPROTO_ICMP: 8278 #ifdef INET6 8279 case IPPROTO_ICMPV6: 8280 #endif /* INET6 */ 8281 break; 8282 default: 8283 return (1); 8284 } 8285 8286 if (!hw_assist) { 8287 switch (af) { 8288 case AF_INET: 8289 if (p == IPPROTO_ICMP) { 8290 if (m->m_len < off) 8291 return (1); 8292 m->m_data += off; 8293 m->m_len -= off; 8294 sum = in_cksum(m, len); 8295 m->m_data -= off; 8296 m->m_len += off; 8297 } else { 8298 if (m->m_len < sizeof(struct ip)) 8299 return (1); 8300 sum = in4_cksum(m, p, off, len); 8301 } 8302 break; 8303 #ifdef INET6 8304 case AF_INET6: 8305 if (m->m_len < sizeof(struct ip6_hdr)) 8306 return (1); 8307 sum = in6_cksum(m, p, off, len); 8308 break; 8309 #endif /* INET6 */ 8310 default: 8311 return (1); 8312 } 8313 } 8314 if (sum) { 8315 switch (p) { 8316 case IPPROTO_TCP: 8317 { 8318 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 8319 break; 8320 } 8321 case IPPROTO_UDP: 8322 { 8323 KMOD_UDPSTAT_INC(udps_badsum); 8324 break; 8325 } 8326 #ifdef INET 8327 case IPPROTO_ICMP: 8328 { 8329 KMOD_ICMPSTAT_INC(icps_checksum); 8330 break; 8331 } 8332 #endif 8333 #ifdef INET6 8334 case IPPROTO_ICMPV6: 8335 { 8336 KMOD_ICMP6STAT_INC(icp6s_checksum); 8337 break; 8338 } 8339 #endif /* INET6 */ 8340 } 8341 return (1); 8342 } else { 8343 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 8344 m->m_pkthdr.csum_flags |= 8345 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 8346 m->m_pkthdr.csum_data = 0xffff; 8347 } 8348 } 8349 return (0); 8350 } 8351 8352 static bool 8353 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 8354 const struct pf_kstate *s, struct ip_fw_args *dnflow) 8355 { 8356 int dndir = r->direction; 8357 8358 if (s && dndir == PF_INOUT) { 8359 dndir = s->direction; 8360 } else if (dndir == PF_INOUT) { 8361 /* Assume primary direction. Happens when we've set dnpipe in 8362 * the ethernet level code. */ 8363 dndir = pd->dir; 8364 } 8365 8366 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 8367 return (false); 8368 8369 memset(dnflow, 0, sizeof(*dnflow)); 8370 8371 if (pd->dport != NULL) 8372 dnflow->f_id.dst_port = ntohs(*pd->dport); 8373 if (pd->sport != NULL) 8374 dnflow->f_id.src_port = ntohs(*pd->sport); 8375 8376 if (pd->dir == PF_IN) 8377 dnflow->flags |= IPFW_ARGS_IN; 8378 else 8379 dnflow->flags |= IPFW_ARGS_OUT; 8380 8381 if (pd->dir != dndir && pd->act.dnrpipe) { 8382 dnflow->rule.info = pd->act.dnrpipe; 8383 } 8384 else if (pd->dir == dndir && pd->act.dnpipe) { 8385 dnflow->rule.info = pd->act.dnpipe; 8386 } 8387 else { 8388 return (false); 8389 } 8390 8391 dnflow->rule.info |= IPFW_IS_DUMMYNET; 8392 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 8393 dnflow->rule.info |= IPFW_IS_PIPE; 8394 8395 dnflow->f_id.proto = pd->proto; 8396 dnflow->f_id.extra = dnflow->rule.info; 8397 switch (pd->af) { 8398 case AF_INET: 8399 dnflow->f_id.addr_type = 4; 8400 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 8401 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 8402 break; 8403 case AF_INET6: 8404 dnflow->flags |= IPFW_ARGS_IP6; 8405 dnflow->f_id.addr_type = 6; 8406 dnflow->f_id.src_ip6 = pd->src->v6; 8407 dnflow->f_id.dst_ip6 = pd->dst->v6; 8408 break; 8409 default: 8410 panic("Invalid AF"); 8411 break; 8412 } 8413 8414 return (true); 8415 } 8416 8417 int 8418 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8419 struct inpcb *inp) 8420 { 8421 struct pfi_kkif *kif; 8422 struct mbuf *m = *m0; 8423 8424 M_ASSERTPKTHDR(m); 8425 MPASS(ifp->if_vnet == curvnet); 8426 NET_EPOCH_ASSERT(); 8427 8428 if (!V_pf_status.running) 8429 return (PF_PASS); 8430 8431 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8432 8433 if (kif == NULL) { 8434 DPFPRINTF(PF_DEBUG_URGENT, 8435 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 8436 return (PF_DROP); 8437 } 8438 if (kif->pfik_flags & PFI_IFLAG_SKIP) 8439 return (PF_PASS); 8440 8441 if (m->m_flags & M_SKIP_FIREWALL) 8442 return (PF_PASS); 8443 8444 if (__predict_false(! M_WRITABLE(*m0))) { 8445 m = *m0 = m_unshare(*m0, M_NOWAIT); 8446 if (*m0 == NULL) 8447 return (PF_DROP); 8448 } 8449 8450 /* Stateless! */ 8451 return (pf_test_eth_rule(dir, kif, m0)); 8452 } 8453 8454 static __inline void 8455 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 8456 { 8457 struct m_tag *mtag; 8458 8459 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8460 8461 /* dummynet adds this tag, but pf does not need it, 8462 * and keeping it creates unexpected behavior, 8463 * e.g. in case of divert(4) usage right after dummynet. */ 8464 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8465 if (mtag != NULL) 8466 m_tag_delete(m, mtag); 8467 } 8468 8469 static int 8470 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 8471 struct pf_krule *r, struct mbuf **m0) 8472 { 8473 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 8474 } 8475 8476 static int 8477 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 8478 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 8479 struct mbuf **m0) 8480 { 8481 NET_EPOCH_ASSERT(); 8482 8483 if (pd->act.dnpipe || pd->act.dnrpipe) { 8484 struct ip_fw_args dnflow; 8485 if (ip_dn_io_ptr == NULL) { 8486 m_freem(*m0); 8487 *m0 = NULL; 8488 return (ENOMEM); 8489 } 8490 8491 if (pd->pf_mtag == NULL && 8492 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 8493 m_freem(*m0); 8494 *m0 = NULL; 8495 return (ENOMEM); 8496 } 8497 8498 if (ifp != NULL) { 8499 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 8500 8501 pd->pf_mtag->if_index = ifp->if_index; 8502 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 8503 8504 MPASS(sa != NULL); 8505 8506 if (pd->af == AF_INET) 8507 memcpy(&pd->pf_mtag->dst, sa, 8508 sizeof(struct sockaddr_in)); 8509 else 8510 memcpy(&pd->pf_mtag->dst, sa, 8511 sizeof(struct sockaddr_in6)); 8512 } 8513 8514 if (s != NULL && s->nat_rule.ptr != NULL && 8515 s->nat_rule.ptr->action == PF_RDR && 8516 ( 8517 #ifdef INET 8518 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 8519 #endif 8520 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 8521 /* 8522 * If we're redirecting to loopback mark this packet 8523 * as being local. Otherwise it might get dropped 8524 * if dummynet re-injects. 8525 */ 8526 (*m0)->m_pkthdr.rcvif = V_loif; 8527 } 8528 8529 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 8530 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 8531 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 8532 ip_dn_io_ptr(m0, &dnflow); 8533 if (*m0 != NULL) { 8534 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8535 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 8536 } 8537 } 8538 } 8539 8540 return (0); 8541 } 8542 8543 int 8544 pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf *m, 8545 u_short *action, u_short *reason, struct pfi_kkif *kif, struct pf_krule **a, 8546 struct pf_krule **r, struct pf_kruleset **ruleset, int *off, int *hdrlen, 8547 struct pf_rule_actions *default_actions) 8548 { 8549 8550 TAILQ_INIT(&pd->sctp_multihome_jobs); 8551 if (default_actions != NULL) 8552 memcpy(&pd->act, default_actions, sizeof(pd->act)); 8553 pd->pf_mtag = pf_find_mtag(m); 8554 8555 if (pd->pf_mtag && pd->pf_mtag->dnpipe) { 8556 pd->act.dnpipe = pd->pf_mtag->dnpipe; 8557 pd->act.flags = pd->pf_mtag->dnflags; 8558 } 8559 8560 pd->af = af; 8561 8562 switch (af) { 8563 #ifdef INET 8564 case AF_INET: { 8565 struct ip *h; 8566 8567 h = mtod(m, struct ip *); 8568 *off = h->ip_hl << 2; 8569 if (*off < (int)sizeof(*h)) { 8570 *action = PF_DROP; 8571 REASON_SET(reason, PFRES_SHORT); 8572 return (-1); 8573 } 8574 pd->src = (struct pf_addr *)&h->ip_src; 8575 pd->dst = (struct pf_addr *)&h->ip_dst; 8576 pd->sport = pd->dport = NULL; 8577 pd->ip_sum = &h->ip_sum; 8578 pd->proto_sum = NULL; 8579 pd->proto = h->ip_p; 8580 pd->dir = dir; 8581 pd->sidx = (dir == PF_IN) ? 0 : 1; 8582 pd->didx = (dir == PF_IN) ? 1 : 0; 8583 pd->tos = h->ip_tos; 8584 pd->tot_len = ntohs(h->ip_len); 8585 pd->act.rtableid = -1; 8586 8587 /* fragments not reassembled handled later */ 8588 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) 8589 return (0); 8590 8591 switch (h->ip_p) { 8592 case IPPROTO_TCP: { 8593 struct tcphdr *th = &pd->hdr.tcp; 8594 8595 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8596 reason, AF_INET)) { 8597 *action = PF_DROP; 8598 REASON_SET(reason, PFRES_SHORT); 8599 return (-1); 8600 } 8601 *hdrlen = sizeof(*th); 8602 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8603 pd->sport = &th->th_sport; 8604 pd->dport = &th->th_dport; 8605 break; 8606 } 8607 case IPPROTO_UDP: { 8608 struct udphdr *uh = &pd->hdr.udp; 8609 8610 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8611 reason, AF_INET)) { 8612 *action = PF_DROP; 8613 REASON_SET(reason, PFRES_SHORT); 8614 return (-1); 8615 } 8616 *hdrlen = sizeof(*uh); 8617 if (uh->uh_dport == 0 || 8618 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8619 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8620 *action = PF_DROP; 8621 REASON_SET(reason, PFRES_SHORT); 8622 return (-1); 8623 } 8624 pd->sport = &uh->uh_sport; 8625 pd->dport = &uh->uh_dport; 8626 break; 8627 } 8628 case IPPROTO_SCTP: { 8629 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8630 action, reason, AF_INET)) { 8631 *action = PF_DROP; 8632 REASON_SET(reason, PFRES_SHORT); 8633 return (-1); 8634 } 8635 *hdrlen = sizeof(pd->hdr.sctp); 8636 pd->p_len = pd->tot_len - *off; 8637 8638 pd->sport = &pd->hdr.sctp.src_port; 8639 pd->dport = &pd->hdr.sctp.dest_port; 8640 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8641 *action = PF_DROP; 8642 REASON_SET(reason, PFRES_SHORT); 8643 return (-1); 8644 } 8645 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8646 *action = PF_DROP; 8647 REASON_SET(reason, PFRES_SHORT); 8648 return (-1); 8649 } 8650 break; 8651 } 8652 case IPPROTO_ICMP: { 8653 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp, ICMP_MINLEN, 8654 action, reason, AF_INET)) { 8655 *action = PF_DROP; 8656 REASON_SET(reason, PFRES_SHORT); 8657 return (-1); 8658 } 8659 *hdrlen = ICMP_MINLEN; 8660 break; 8661 } 8662 } 8663 break; 8664 } 8665 #endif 8666 #ifdef INET6 8667 case AF_INET6: { 8668 struct ip6_hdr *h; 8669 int terminal = 0; 8670 8671 h = mtod(m, struct ip6_hdr *); 8672 pd->src = (struct pf_addr *)&h->ip6_src; 8673 pd->dst = (struct pf_addr *)&h->ip6_dst; 8674 pd->sport = pd->dport = NULL; 8675 pd->ip_sum = NULL; 8676 pd->proto_sum = NULL; 8677 pd->dir = dir; 8678 pd->sidx = (dir == PF_IN) ? 0 : 1; 8679 pd->didx = (dir == PF_IN) ? 1 : 0; 8680 pd->tos = IPV6_DSCP(h); 8681 pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8682 *off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8683 pd->proto = h->ip6_nxt; 8684 pd->act.rtableid = -1; 8685 8686 do { 8687 switch (pd->proto) { 8688 case IPPROTO_FRAGMENT: 8689 if (kif == NULL || r == NULL) /* pflog */ 8690 *action = PF_DROP; 8691 else 8692 *action = pf_test_fragment(r, kif, 8693 m, h, pd, a, ruleset); 8694 if (*action == PF_DROP) 8695 REASON_SET(reason, PFRES_FRAG); 8696 return (-1); 8697 case IPPROTO_ROUTING: { 8698 struct ip6_rthdr rthdr; 8699 8700 if (pd->rh_cnt++) { 8701 DPFPRINTF(PF_DEBUG_MISC, 8702 ("pf: IPv6 more than one rthdr")); 8703 *action = PF_DROP; 8704 REASON_SET(reason, PFRES_IPOPTIONS); 8705 return (-1); 8706 } 8707 if (!pf_pull_hdr(m, *off, &rthdr, sizeof(rthdr), 8708 NULL, reason, pd->af)) { 8709 DPFPRINTF(PF_DEBUG_MISC, 8710 ("pf: IPv6 short rthdr")); 8711 *action = PF_DROP; 8712 REASON_SET(reason, PFRES_SHORT); 8713 return (-1); 8714 } 8715 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8716 DPFPRINTF(PF_DEBUG_MISC, 8717 ("pf: IPv6 rthdr0")); 8718 *action = PF_DROP; 8719 REASON_SET(reason, PFRES_IPOPTIONS); 8720 return (-1); 8721 } 8722 /* FALLTHROUGH */ 8723 } 8724 case IPPROTO_AH: 8725 case IPPROTO_HOPOPTS: 8726 case IPPROTO_DSTOPTS: { 8727 /* get next header and header length */ 8728 struct ip6_ext opt6; 8729 8730 if (!pf_pull_hdr(m, *off, &opt6, sizeof(opt6), 8731 NULL, reason, pd->af)) { 8732 DPFPRINTF(PF_DEBUG_MISC, 8733 ("pf: IPv6 short opt")); 8734 *action = PF_DROP; 8735 return (-1); 8736 } 8737 if (pd->proto == IPPROTO_AH) 8738 *off += (opt6.ip6e_len + 2) * 4; 8739 else 8740 *off += (opt6.ip6e_len + 1) * 8; 8741 pd->proto = opt6.ip6e_nxt; 8742 /* goto the next header */ 8743 break; 8744 } 8745 default: 8746 terminal++; 8747 break; 8748 } 8749 } while (!terminal); 8750 8751 switch (pd->proto) { 8752 case IPPROTO_TCP: { 8753 struct tcphdr *th = &pd->hdr.tcp; 8754 8755 if (!pf_pull_hdr(m, *off, th, sizeof(*th), action, 8756 reason, AF_INET6)) { 8757 *action = PF_DROP; 8758 REASON_SET(reason, PFRES_SHORT); 8759 return (-1); 8760 } 8761 *hdrlen = sizeof(*th); 8762 pd->p_len = pd->tot_len - *off - (th->th_off << 2); 8763 pd->sport = &th->th_sport; 8764 pd->dport = &th->th_dport; 8765 break; 8766 } 8767 case IPPROTO_UDP: { 8768 struct udphdr *uh = &pd->hdr.udp; 8769 8770 if (!pf_pull_hdr(m, *off, uh, sizeof(*uh), action, 8771 reason, AF_INET6)) { 8772 *action = PF_DROP; 8773 REASON_SET(reason, PFRES_SHORT); 8774 return (-1); 8775 } 8776 *hdrlen = sizeof(*uh); 8777 if (uh->uh_dport == 0 || 8778 ntohs(uh->uh_ulen) > m->m_pkthdr.len - *off || 8779 ntohs(uh->uh_ulen) < sizeof(struct udphdr)) { 8780 *action = PF_DROP; 8781 REASON_SET(reason, PFRES_SHORT); 8782 return (-1); 8783 } 8784 pd->sport = &uh->uh_sport; 8785 pd->dport = &uh->uh_dport; 8786 break; 8787 } 8788 case IPPROTO_SCTP: { 8789 if (!pf_pull_hdr(m, *off, &pd->hdr.sctp, sizeof(pd->hdr.sctp), 8790 action, reason, AF_INET6)) { 8791 *action = PF_DROP; 8792 REASON_SET(reason, PFRES_SHORT); 8793 return (-1); 8794 } 8795 *hdrlen = sizeof(pd->hdr.sctp); 8796 pd->p_len = pd->tot_len - *off; 8797 8798 pd->sport = &pd->hdr.sctp.src_port; 8799 pd->dport = &pd->hdr.sctp.dest_port; 8800 if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) { 8801 *action = PF_DROP; 8802 REASON_SET(reason, PFRES_SHORT); 8803 return (-1); 8804 } 8805 if (pf_scan_sctp(m, *off, pd, kif) != PF_PASS) { 8806 *action = PF_DROP; 8807 REASON_SET(reason, PFRES_SHORT); 8808 return (-1); 8809 } 8810 break; 8811 } 8812 case IPPROTO_ICMPV6: { 8813 size_t icmp_hlen = sizeof(struct icmp6_hdr); 8814 8815 if (!pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8816 action, reason, AF_INET6)) { 8817 *action = PF_DROP; 8818 REASON_SET(reason, PFRES_SHORT); 8819 return (-1); 8820 } 8821 /* ICMP headers we look further into to match state */ 8822 switch (pd->hdr.icmp6.icmp6_type) { 8823 case MLD_LISTENER_QUERY: 8824 case MLD_LISTENER_REPORT: 8825 icmp_hlen = sizeof(struct mld_hdr); 8826 break; 8827 case ND_NEIGHBOR_SOLICIT: 8828 case ND_NEIGHBOR_ADVERT: 8829 icmp_hlen = sizeof(struct nd_neighbor_solicit); 8830 break; 8831 } 8832 if (icmp_hlen > sizeof(struct icmp6_hdr) && 8833 !pf_pull_hdr(m, *off, &pd->hdr.icmp6, icmp_hlen, 8834 action, reason, AF_INET6)) { 8835 *action = PF_DROP; 8836 REASON_SET(reason, PFRES_SHORT); 8837 return (-1); 8838 } 8839 *hdrlen = icmp_hlen; 8840 break; 8841 } 8842 } 8843 break; 8844 } 8845 #endif 8846 default: 8847 panic("pf_setup_pdesc called with illegal af %u", af); 8848 } 8849 return (0); 8850 } 8851 8852 #ifdef INET 8853 int 8854 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 8855 struct inpcb *inp, struct pf_rule_actions *default_actions) 8856 { 8857 struct pfi_kkif *kif; 8858 u_short action, reason = 0; 8859 struct mbuf *m = *m0; 8860 struct ip *h = NULL; 8861 struct m_tag *mtag; 8862 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8863 struct pf_kstate *s = NULL; 8864 struct pf_kruleset *ruleset = NULL; 8865 struct pf_pdesc pd; 8866 int off, hdrlen, dirndx, use_2nd_queue = 0; 8867 uint16_t tag; 8868 uint8_t rt; 8869 8870 PF_RULES_RLOCK_TRACKER; 8871 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8872 M_ASSERTPKTHDR(m); 8873 8874 if (!V_pf_status.running) 8875 return (PF_PASS); 8876 8877 PF_RULES_RLOCK(); 8878 8879 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8880 8881 if (__predict_false(kif == NULL)) { 8882 DPFPRINTF(PF_DEBUG_URGENT, 8883 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8884 PF_RULES_RUNLOCK(); 8885 return (PF_DROP); 8886 } 8887 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8888 PF_RULES_RUNLOCK(); 8889 return (PF_PASS); 8890 } 8891 8892 if (m->m_flags & M_SKIP_FIREWALL) { 8893 PF_RULES_RUNLOCK(); 8894 return (PF_PASS); 8895 } 8896 8897 if (__predict_false(! M_WRITABLE(*m0))) { 8898 m = *m0 = m_unshare(*m0, M_NOWAIT); 8899 if (*m0 == NULL) 8900 return (PF_DROP); 8901 } 8902 8903 if (__predict_false(m->m_len < sizeof(struct ip)) && 8904 (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) { 8905 DPFPRINTF(PF_DEBUG_URGENT, 8906 ("pf_test: m_len < sizeof(struct ip), pullup failed\n")); 8907 PF_RULES_RUNLOCK(); 8908 return (PF_DROP); 8909 } 8910 8911 memset(&pd, 0, sizeof(pd)); 8912 pd.dir = dir; 8913 8914 if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 8915 /* We do IP header normalization and packet reassembly here */ 8916 action = PF_DROP; 8917 goto done; 8918 } 8919 m = *m0; /* pf_normalize messes with m0 */ 8920 h = mtod(m, struct ip *); 8921 8922 if (pf_setup_pdesc(AF_INET, dir, &pd, m, &action, &reason, kif, &a, &r, 8923 &ruleset, &off, &hdrlen, default_actions) == -1) { 8924 if (action != PF_PASS) 8925 pd.act.log |= PF_LOG_FORCE; 8926 goto done; 8927 } 8928 8929 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8930 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8931 8932 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8933 pd.pf_mtag->if_idxgen); 8934 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8935 PF_RULES_RUNLOCK(); 8936 m_freem(*m0); 8937 *m0 = NULL; 8938 return (PF_PASS); 8939 } 8940 PF_RULES_RUNLOCK(); 8941 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 8942 *m0 = NULL; 8943 return (PF_PASS); 8944 } 8945 8946 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8947 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8948 /* Dummynet re-injects packets after they've 8949 * completed their delay. We've already 8950 * processed them, so pass unconditionally. */ 8951 8952 /* But only once. We may see the packet multiple times (e.g. 8953 * PFIL_IN/PFIL_OUT). */ 8954 pf_dummynet_flag_remove(m, pd.pf_mtag); 8955 PF_RULES_RUNLOCK(); 8956 8957 return (PF_PASS); 8958 } 8959 8960 if (__predict_false(ip_divert_ptr != NULL) && 8961 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 8962 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 8963 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 8964 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 8965 if (pd.pf_mtag == NULL && 8966 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8967 action = PF_DROP; 8968 goto done; 8969 } 8970 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 8971 } 8972 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 8973 m->m_flags |= M_FASTFWD_OURS; 8974 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8975 } 8976 m_tag_delete(m, mtag); 8977 8978 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8979 if (mtag != NULL) 8980 m_tag_delete(m, mtag); 8981 } 8982 8983 /* handle fragments that didn't get reassembled by normalization */ 8984 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8985 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 8986 goto done; 8987 } 8988 8989 switch (h->ip_p) { 8990 case IPPROTO_TCP: { 8991 8992 /* Respond to SYN with a syncookie. */ 8993 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8994 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8995 pf_syncookie_send(m, off, &pd); 8996 action = PF_DROP; 8997 break; 8998 } 8999 9000 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 9001 use_2nd_queue = 1; 9002 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 9003 if (action == PF_DROP) 9004 goto done; 9005 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 9006 if (action == PF_PASS) { 9007 if (V_pfsync_update_state_ptr != NULL) 9008 V_pfsync_update_state_ptr(s); 9009 r = s->rule.ptr; 9010 a = s->anchor.ptr; 9011 } else if (s == NULL) { 9012 /* Validate remote SYN|ACK, re-create original SYN if 9013 * valid. */ 9014 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9015 TH_ACK && pf_syncookie_validate(&pd) && 9016 pd.dir == PF_IN) { 9017 struct mbuf *msyn; 9018 9019 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 9020 &pd); 9021 if (msyn == NULL) { 9022 action = PF_DROP; 9023 break; 9024 } 9025 9026 action = pf_test(dir, pflags, ifp, &msyn, inp, 9027 &pd.act); 9028 m_freem(msyn); 9029 if (action != PF_PASS) 9030 break; 9031 9032 action = pf_test_state_tcp(&s, kif, m, off, h, 9033 &pd, &reason); 9034 if (action != PF_PASS || s == NULL) { 9035 action = PF_DROP; 9036 break; 9037 } 9038 9039 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9040 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9041 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9042 action = pf_synproxy(&pd, &s, &reason); 9043 break; 9044 } else { 9045 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9046 &a, &ruleset, inp, hdrlen); 9047 } 9048 } 9049 break; 9050 } 9051 9052 case IPPROTO_UDP: { 9053 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 9054 if (action == PF_PASS) { 9055 if (V_pfsync_update_state_ptr != NULL) 9056 V_pfsync_update_state_ptr(s); 9057 r = s->rule.ptr; 9058 a = s->anchor.ptr; 9059 } else if (s == NULL) 9060 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9061 &a, &ruleset, inp, hdrlen); 9062 break; 9063 } 9064 9065 case IPPROTO_SCTP: { 9066 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 9067 if (action == PF_DROP) 9068 goto done; 9069 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 9070 &reason); 9071 if (action == PF_PASS) { 9072 if (V_pfsync_update_state_ptr != NULL) 9073 V_pfsync_update_state_ptr(s); 9074 r = s->rule.ptr; 9075 a = s->anchor.ptr; 9076 } else if (s == NULL) { 9077 action = pf_test_rule(&r, &s, kif, m, off, 9078 &pd, &a, &ruleset, inp, hdrlen); 9079 } 9080 break; 9081 } 9082 9083 case IPPROTO_ICMP: { 9084 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 9085 if (action == PF_PASS) { 9086 if (V_pfsync_update_state_ptr != NULL) 9087 V_pfsync_update_state_ptr(s); 9088 r = s->rule.ptr; 9089 a = s->anchor.ptr; 9090 } else if (s == NULL) 9091 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9092 &a, &ruleset, inp, hdrlen); 9093 break; 9094 } 9095 9096 case IPPROTO_ICMPV6: { 9097 action = PF_DROP; 9098 DPFPRINTF(PF_DEBUG_MISC, 9099 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 9100 goto done; 9101 } 9102 9103 default: 9104 action = pf_test_state_other(&s, kif, m, &pd); 9105 if (action == PF_PASS) { 9106 if (V_pfsync_update_state_ptr != NULL) 9107 V_pfsync_update_state_ptr(s); 9108 r = s->rule.ptr; 9109 a = s->anchor.ptr; 9110 } else if (s == NULL) 9111 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9112 &a, &ruleset, inp, hdrlen); 9113 break; 9114 } 9115 9116 done: 9117 PF_RULES_RUNLOCK(); 9118 if (action == PF_PASS && h->ip_hl > 5 && 9119 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9120 action = PF_DROP; 9121 REASON_SET(&reason, PFRES_IPOPTIONS); 9122 pd.act.log = PF_LOG_FORCE; 9123 DPFPRINTF(PF_DEBUG_MISC, 9124 ("pf: dropping packet with ip options\n")); 9125 } 9126 9127 if (s) { 9128 uint8_t log = pd.act.log; 9129 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9130 pd.act.log |= log; 9131 tag = s->tag; 9132 rt = s->rt; 9133 } else { 9134 tag = r->tag; 9135 rt = r->rt; 9136 } 9137 9138 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9139 action = PF_DROP; 9140 REASON_SET(&reason, PFRES_MEMORY); 9141 } 9142 9143 pf_scrub_ip(&m, &pd); 9144 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9145 pf_normalize_mss(m, off, &pd); 9146 9147 if (pd.act.rtableid >= 0) 9148 M_SETFIB(m, pd.act.rtableid); 9149 9150 if (pd.act.flags & PFSTATE_SETPRIO) { 9151 if (pd.tos & IPTOS_LOWDELAY) 9152 use_2nd_queue = 1; 9153 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9154 action = PF_DROP; 9155 REASON_SET(&reason, PFRES_MEMORY); 9156 pd.act.log = PF_LOG_FORCE; 9157 DPFPRINTF(PF_DEBUG_MISC, 9158 ("pf: failed to allocate 802.1q mtag\n")); 9159 } 9160 } 9161 9162 #ifdef ALTQ 9163 if (action == PF_PASS && pd.act.qid) { 9164 if (pd.pf_mtag == NULL && 9165 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9166 action = PF_DROP; 9167 REASON_SET(&reason, PFRES_MEMORY); 9168 } else { 9169 if (s != NULL) 9170 pd.pf_mtag->qid_hash = pf_state_hash(s); 9171 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 9172 pd.pf_mtag->qid = pd.act.pqid; 9173 else 9174 pd.pf_mtag->qid = pd.act.qid; 9175 /* Add hints for ecn. */ 9176 pd.pf_mtag->hdr = h; 9177 } 9178 } 9179 #endif /* ALTQ */ 9180 9181 /* 9182 * connections redirected to loopback should not match sockets 9183 * bound specifically to loopback due to security implications, 9184 * see tcp_input() and in_pcblookup_listen(). 9185 */ 9186 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9187 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9188 (s->nat_rule.ptr->action == PF_RDR || 9189 s->nat_rule.ptr->action == PF_BINAT) && 9190 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 9191 m->m_flags |= M_SKIP_FIREWALL; 9192 9193 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 9194 r->divert.port && !PACKET_LOOPED(&pd)) { 9195 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 9196 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 9197 if (mtag != NULL) { 9198 ((struct pf_divert_mtag *)(mtag+1))->port = 9199 ntohs(r->divert.port); 9200 ((struct pf_divert_mtag *)(mtag+1))->idir = 9201 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 9202 PF_DIVERT_MTAG_DIR_OUT; 9203 9204 if (s) 9205 PF_STATE_UNLOCK(s); 9206 9207 m_tag_prepend(m, mtag); 9208 if (m->m_flags & M_FASTFWD_OURS) { 9209 if (pd.pf_mtag == NULL && 9210 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9211 action = PF_DROP; 9212 REASON_SET(&reason, PFRES_MEMORY); 9213 pd.act.log = PF_LOG_FORCE; 9214 DPFPRINTF(PF_DEBUG_MISC, 9215 ("pf: failed to allocate tag\n")); 9216 } else { 9217 pd.pf_mtag->flags |= 9218 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 9219 m->m_flags &= ~M_FASTFWD_OURS; 9220 } 9221 } 9222 ip_divert_ptr(*m0, dir == PF_IN); 9223 *m0 = NULL; 9224 9225 return (action); 9226 } else { 9227 /* XXX: ipfw has the same behaviour! */ 9228 action = PF_DROP; 9229 REASON_SET(&reason, PFRES_MEMORY); 9230 pd.act.log = PF_LOG_FORCE; 9231 DPFPRINTF(PF_DEBUG_MISC, 9232 ("pf: failed to allocate divert tag\n")); 9233 } 9234 } 9235 /* this flag will need revising if the pkt is forwarded */ 9236 if (pd.pf_mtag) 9237 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 9238 9239 if (pd.act.log) { 9240 struct pf_krule *lr; 9241 struct pf_krule_item *ri; 9242 9243 if (s != NULL && s->nat_rule.ptr != NULL && 9244 s->nat_rule.ptr->log & PF_LOG_ALL) 9245 lr = s->nat_rule.ptr; 9246 else 9247 lr = r; 9248 9249 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9250 PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a, 9251 ruleset, &pd, (s == NULL)); 9252 if (s) { 9253 SLIST_FOREACH(ri, &s->match_rules, entry) 9254 if (ri->r->log & PF_LOG_ALL) 9255 PFLOG_PACKET(kif, m, AF_INET, action, 9256 reason, ri->r, a, ruleset, &pd, 0); 9257 } 9258 } 9259 9260 pf_counter_u64_critical_enter(); 9261 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 9262 pd.tot_len); 9263 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 9264 1); 9265 9266 if (action == PF_PASS || r->action == PF_DROP) { 9267 dirndx = (dir == PF_OUT); 9268 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 9269 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 9270 pf_update_timestamp(r); 9271 9272 if (a != NULL) { 9273 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 9274 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 9275 } 9276 if (s != NULL) { 9277 struct pf_krule_item *ri; 9278 9279 if (s->nat_rule.ptr != NULL) { 9280 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 9281 1); 9282 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 9283 pd.tot_len); 9284 } 9285 if (s->src_node != NULL) { 9286 counter_u64_add(s->src_node->packets[dirndx], 9287 1); 9288 counter_u64_add(s->src_node->bytes[dirndx], 9289 pd.tot_len); 9290 } 9291 if (s->nat_src_node != NULL) { 9292 counter_u64_add(s->nat_src_node->packets[dirndx], 9293 1); 9294 counter_u64_add(s->nat_src_node->bytes[dirndx], 9295 pd.tot_len); 9296 } 9297 dirndx = (dir == s->direction) ? 0 : 1; 9298 s->packets[dirndx]++; 9299 s->bytes[dirndx] += pd.tot_len; 9300 SLIST_FOREACH(ri, &s->match_rules, entry) { 9301 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 9302 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 9303 } 9304 } 9305 tr = r; 9306 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 9307 if (nr != NULL && r == &V_pf_default_rule) 9308 tr = nr; 9309 if (tr->src.addr.type == PF_ADDR_TABLE) 9310 pfr_update_stats(tr->src.addr.p.tbl, 9311 (s == NULL) ? pd.src : 9312 &s->key[(s->direction == PF_IN)]-> 9313 addr[(s->direction == PF_OUT)], 9314 pd.af, pd.tot_len, dir == PF_OUT, 9315 r->action == PF_PASS, tr->src.neg); 9316 if (tr->dst.addr.type == PF_ADDR_TABLE) 9317 pfr_update_stats(tr->dst.addr.p.tbl, 9318 (s == NULL) ? pd.dst : 9319 &s->key[(s->direction == PF_IN)]-> 9320 addr[(s->direction == PF_IN)], 9321 pd.af, pd.tot_len, dir == PF_OUT, 9322 r->action == PF_PASS, tr->dst.neg); 9323 } 9324 pf_counter_u64_critical_exit(); 9325 9326 switch (action) { 9327 case PF_SYNPROXY_DROP: 9328 m_freem(*m0); 9329 case PF_DEFER: 9330 *m0 = NULL; 9331 action = PF_PASS; 9332 break; 9333 case PF_DROP: 9334 m_freem(*m0); 9335 *m0 = NULL; 9336 break; 9337 default: 9338 /* pf_route() returns unlocked. */ 9339 if (rt) { 9340 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 9341 goto out; 9342 } 9343 if (pf_dummynet(&pd, s, r, m0) != 0) { 9344 action = PF_DROP; 9345 REASON_SET(&reason, PFRES_MEMORY); 9346 } 9347 break; 9348 } 9349 9350 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 9351 9352 if (s && action != PF_DROP) { 9353 if (!s->if_index_in && dir == PF_IN) 9354 s->if_index_in = ifp->if_index; 9355 else if (!s->if_index_out && dir == PF_OUT) 9356 s->if_index_out = ifp->if_index; 9357 } 9358 9359 if (s) 9360 PF_STATE_UNLOCK(s); 9361 9362 out: 9363 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9364 9365 return (action); 9366 } 9367 #endif /* INET */ 9368 9369 #ifdef INET6 9370 int 9371 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 9372 struct pf_rule_actions *default_actions) 9373 { 9374 struct pfi_kkif *kif; 9375 u_short action, reason = 0; 9376 struct mbuf *m = *m0, *n = NULL; 9377 struct m_tag *mtag; 9378 struct ip6_hdr *h = NULL; 9379 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 9380 struct pf_kstate *s = NULL; 9381 struct pf_kruleset *ruleset = NULL; 9382 struct pf_pdesc pd; 9383 int off, hdrlen, dirndx, use_2nd_queue = 0; 9384 uint16_t tag; 9385 uint8_t rt; 9386 9387 PF_RULES_RLOCK_TRACKER; 9388 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 9389 M_ASSERTPKTHDR(m); 9390 9391 if (!V_pf_status.running) 9392 return (PF_PASS); 9393 9394 PF_RULES_RLOCK(); 9395 9396 kif = (struct pfi_kkif *)ifp->if_pf_kif; 9397 if (__predict_false(kif == NULL)) { 9398 DPFPRINTF(PF_DEBUG_URGENT, 9399 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 9400 PF_RULES_RUNLOCK(); 9401 return (PF_DROP); 9402 } 9403 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 9404 PF_RULES_RUNLOCK(); 9405 return (PF_PASS); 9406 } 9407 9408 if (m->m_flags & M_SKIP_FIREWALL) { 9409 PF_RULES_RUNLOCK(); 9410 return (PF_PASS); 9411 } 9412 9413 /* 9414 * If we end up changing IP addresses (e.g. binat) the stack may get 9415 * confused and fail to send the icmp6 packet too big error. Just send 9416 * it here, before we do any NAT. 9417 */ 9418 if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 9419 PF_RULES_RUNLOCK(); 9420 *m0 = NULL; 9421 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 9422 return (PF_DROP); 9423 } 9424 9425 if (__predict_false(! M_WRITABLE(*m0))) { 9426 m = *m0 = m_unshare(*m0, M_NOWAIT); 9427 if (*m0 == NULL) 9428 return (PF_DROP); 9429 } 9430 9431 if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) && 9432 (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) { 9433 DPFPRINTF(PF_DEBUG_URGENT, 9434 ("pf_test6: m_len < sizeof(struct ip6_hdr)" 9435 ", pullup failed\n")); 9436 PF_RULES_RUNLOCK(); 9437 return (PF_DROP); 9438 } 9439 9440 memset(&pd, 0, sizeof(pd)); 9441 pd.dir = dir; 9442 9443 /* We do IP header normalization and packet reassembly here */ 9444 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 9445 action = PF_DROP; 9446 goto done; 9447 } 9448 m = *m0; /* pf_normalize messes with m0 */ 9449 h = mtod(m, struct ip6_hdr *); 9450 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 9451 9452 if (pf_setup_pdesc(AF_INET6, dir, &pd, m, &action, &reason, kif, &a, &r, 9453 &ruleset, &off, &hdrlen, default_actions) == -1) { 9454 if (action != PF_PASS) 9455 pd.act.log |= PF_LOG_FORCE; 9456 goto done; 9457 } 9458 9459 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 9460 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 9461 9462 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 9463 pd.pf_mtag->if_idxgen); 9464 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 9465 PF_RULES_RUNLOCK(); 9466 m_freem(*m0); 9467 *m0 = NULL; 9468 return (PF_PASS); 9469 } 9470 PF_RULES_RUNLOCK(); 9471 nd6_output_ifp(ifp, ifp, m, 9472 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 9473 *m0 = NULL; 9474 return (PF_PASS); 9475 } 9476 9477 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 9478 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 9479 pf_dummynet_flag_remove(m, pd.pf_mtag); 9480 /* Dummynet re-injects packets after they've 9481 * completed their delay. We've already 9482 * processed them, so pass unconditionally. */ 9483 PF_RULES_RUNLOCK(); 9484 return (PF_PASS); 9485 } 9486 9487 /* 9488 * we do not support jumbogram. if we keep going, zero ip6_plen 9489 * will do something bad, so drop the packet for now. 9490 */ 9491 if (htons(h->ip6_plen) == 0) { 9492 action = PF_DROP; 9493 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 9494 goto done; 9495 } 9496 9497 /* if there's no routing header, use unmodified mbuf for checksumming */ 9498 if (!n) 9499 n = m; 9500 9501 switch (pd.proto) { 9502 case IPPROTO_TCP: { 9503 /* Respond to SYN with a syncookie. */ 9504 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 9505 pd.dir == PF_IN && pf_synflood_check(&pd)) { 9506 pf_syncookie_send(m, off, &pd); 9507 action = PF_DROP; 9508 break; 9509 } 9510 9511 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 9512 if (action == PF_DROP) 9513 goto done; 9514 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 9515 if (action == PF_PASS) { 9516 if (V_pfsync_update_state_ptr != NULL) 9517 V_pfsync_update_state_ptr(s); 9518 r = s->rule.ptr; 9519 a = s->anchor.ptr; 9520 } else if (s == NULL) { 9521 /* Validate remote SYN|ACK, re-create original SYN if 9522 * valid. */ 9523 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 9524 TH_ACK && pf_syncookie_validate(&pd) && 9525 pd.dir == PF_IN) { 9526 struct mbuf *msyn; 9527 9528 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 9529 off, &pd); 9530 if (msyn == NULL) { 9531 action = PF_DROP; 9532 break; 9533 } 9534 9535 action = pf_test6(dir, pflags, ifp, &msyn, inp, 9536 &pd.act); 9537 m_freem(msyn); 9538 if (action != PF_PASS) 9539 break; 9540 9541 action = pf_test_state_tcp(&s, kif, m, off, h, 9542 &pd, &reason); 9543 if (action != PF_PASS || s == NULL) { 9544 action = PF_DROP; 9545 break; 9546 } 9547 9548 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 9549 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 9550 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 9551 9552 action = pf_synproxy(&pd, &s, &reason); 9553 break; 9554 } else { 9555 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9556 &a, &ruleset, inp, hdrlen); 9557 } 9558 } 9559 break; 9560 } 9561 9562 case IPPROTO_UDP: { 9563 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 9564 if (action == PF_PASS) { 9565 if (V_pfsync_update_state_ptr != NULL) 9566 V_pfsync_update_state_ptr(s); 9567 r = s->rule.ptr; 9568 a = s->anchor.ptr; 9569 } else if (s == NULL) 9570 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9571 &a, &ruleset, inp, hdrlen); 9572 break; 9573 } 9574 9575 case IPPROTO_SCTP: { 9576 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 9577 if (action == PF_DROP) 9578 goto done; 9579 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 9580 &reason); 9581 if (action == PF_PASS) { 9582 if (V_pfsync_update_state_ptr != NULL) 9583 V_pfsync_update_state_ptr(s); 9584 r = s->rule.ptr; 9585 a = s->anchor.ptr; 9586 } else if (s == NULL) { 9587 action = pf_test_rule(&r, &s, kif, m, off, 9588 &pd, &a, &ruleset, inp, hdrlen); 9589 } 9590 break; 9591 } 9592 9593 case IPPROTO_ICMP: { 9594 action = PF_DROP; 9595 DPFPRINTF(PF_DEBUG_MISC, 9596 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 9597 goto done; 9598 } 9599 9600 case IPPROTO_ICMPV6: { 9601 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 9602 if (action == PF_PASS) { 9603 if (V_pfsync_update_state_ptr != NULL) 9604 V_pfsync_update_state_ptr(s); 9605 r = s->rule.ptr; 9606 a = s->anchor.ptr; 9607 } else if (s == NULL) 9608 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9609 &a, &ruleset, inp, hdrlen); 9610 break; 9611 } 9612 9613 default: 9614 action = pf_test_state_other(&s, kif, m, &pd); 9615 if (action == PF_PASS) { 9616 if (V_pfsync_update_state_ptr != NULL) 9617 V_pfsync_update_state_ptr(s); 9618 r = s->rule.ptr; 9619 a = s->anchor.ptr; 9620 } else if (s == NULL) 9621 action = pf_test_rule(&r, &s, kif, m, off, &pd, 9622 &a, &ruleset, inp, hdrlen); 9623 break; 9624 } 9625 9626 done: 9627 PF_RULES_RUNLOCK(); 9628 if (n != m) { 9629 m_freem(n); 9630 n = NULL; 9631 } 9632 9633 /* handle dangerous IPv6 extension headers. */ 9634 if (action == PF_PASS && pd.rh_cnt && 9635 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 9636 action = PF_DROP; 9637 REASON_SET(&reason, PFRES_IPOPTIONS); 9638 pd.act.log = r->log; 9639 DPFPRINTF(PF_DEBUG_MISC, 9640 ("pf: dropping packet with dangerous v6 headers\n")); 9641 } 9642 9643 if (s) { 9644 uint8_t log = pd.act.log; 9645 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 9646 pd.act.log |= log; 9647 tag = s->tag; 9648 rt = s->rt; 9649 } else { 9650 tag = r->tag; 9651 rt = r->rt; 9652 } 9653 9654 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 9655 action = PF_DROP; 9656 REASON_SET(&reason, PFRES_MEMORY); 9657 } 9658 9659 pf_scrub_ip6(&m, &pd); 9660 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 9661 pf_normalize_mss(m, off, &pd); 9662 9663 if (pd.act.rtableid >= 0) 9664 M_SETFIB(m, pd.act.rtableid); 9665 9666 if (pd.act.flags & PFSTATE_SETPRIO) { 9667 if (pd.tos & IPTOS_LOWDELAY) 9668 use_2nd_queue = 1; 9669 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 9670 action = PF_DROP; 9671 REASON_SET(&reason, PFRES_MEMORY); 9672 pd.act.log = PF_LOG_FORCE; 9673 DPFPRINTF(PF_DEBUG_MISC, 9674 ("pf: failed to allocate 802.1q mtag\n")); 9675 } 9676 } 9677 9678 #ifdef ALTQ 9679 if (action == PF_PASS && pd.act.qid) { 9680 if (pd.pf_mtag == NULL && 9681 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 9682 action = PF_DROP; 9683 REASON_SET(&reason, PFRES_MEMORY); 9684 } else { 9685 if (s != NULL) 9686 pd.pf_mtag->qid_hash = pf_state_hash(s); 9687 if (pd.tos & IPTOS_LOWDELAY) 9688 pd.pf_mtag->qid = pd.act.pqid; 9689 else 9690 pd.pf_mtag->qid = pd.act.qid; 9691 /* Add hints for ecn. */ 9692 pd.pf_mtag->hdr = h; 9693 } 9694 } 9695 #endif /* ALTQ */ 9696 9697 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9698 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9699 (s->nat_rule.ptr->action == PF_RDR || 9700 s->nat_rule.ptr->action == PF_BINAT) && 9701 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 9702 m->m_flags |= M_SKIP_FIREWALL; 9703 9704 /* XXX: Anybody working on it?! */ 9705 if (r->divert.port) 9706 printf("pf: divert(9) is not supported for IPv6\n"); 9707 9708 if (pd.act.log) { 9709 struct pf_krule *lr; 9710 struct pf_krule_item *ri; 9711 9712 if (s != NULL && s->nat_rule.ptr != NULL && 9713 s->nat_rule.ptr->log & PF_LOG_ALL) 9714 lr = s->nat_rule.ptr; 9715 else 9716 lr = r; 9717 9718 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9719 PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset, 9720 &pd, (s == NULL)); 9721 if (s) { 9722 SLIST_FOREACH(ri, &s->match_rules, entry) 9723 if (ri->r->log & PF_LOG_ALL) 9724 PFLOG_PACKET(kif, m, AF_INET6, action, reason, 9725 ri->r, a, ruleset, &pd, 0); 9726 } 9727 } 9728 9729 pf_counter_u64_critical_enter(); 9730 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 9731 pd.tot_len); 9732 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 9733 1); 9734 9735 if (action == PF_PASS || r->action == PF_DROP) { 9736 dirndx = (dir == PF_OUT); 9737 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 9738 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 9739 if (a != NULL) { 9740 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 9741 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 9742 } 9743 if (s != NULL) { 9744 if (s->nat_rule.ptr != NULL) { 9745 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 9746 1); 9747 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 9748 pd.tot_len); 9749 } 9750 if (s->src_node != NULL) { 9751 counter_u64_add(s->src_node->packets[dirndx], 9752 1); 9753 counter_u64_add(s->src_node->bytes[dirndx], 9754 pd.tot_len); 9755 } 9756 if (s->nat_src_node != NULL) { 9757 counter_u64_add(s->nat_src_node->packets[dirndx], 9758 1); 9759 counter_u64_add(s->nat_src_node->bytes[dirndx], 9760 pd.tot_len); 9761 } 9762 dirndx = (dir == s->direction) ? 0 : 1; 9763 s->packets[dirndx]++; 9764 s->bytes[dirndx] += pd.tot_len; 9765 } 9766 tr = r; 9767 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 9768 if (nr != NULL && r == &V_pf_default_rule) 9769 tr = nr; 9770 if (tr->src.addr.type == PF_ADDR_TABLE) 9771 pfr_update_stats(tr->src.addr.p.tbl, 9772 (s == NULL) ? pd.src : 9773 &s->key[(s->direction == PF_IN)]->addr[0], 9774 pd.af, pd.tot_len, dir == PF_OUT, 9775 r->action == PF_PASS, tr->src.neg); 9776 if (tr->dst.addr.type == PF_ADDR_TABLE) 9777 pfr_update_stats(tr->dst.addr.p.tbl, 9778 (s == NULL) ? pd.dst : 9779 &s->key[(s->direction == PF_IN)]->addr[1], 9780 pd.af, pd.tot_len, dir == PF_OUT, 9781 r->action == PF_PASS, tr->dst.neg); 9782 } 9783 pf_counter_u64_critical_exit(); 9784 9785 switch (action) { 9786 case PF_SYNPROXY_DROP: 9787 m_freem(*m0); 9788 case PF_DEFER: 9789 *m0 = NULL; 9790 action = PF_PASS; 9791 break; 9792 case PF_DROP: 9793 m_freem(*m0); 9794 *m0 = NULL; 9795 break; 9796 default: 9797 /* pf_route6() returns unlocked. */ 9798 if (rt) { 9799 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9800 goto out; 9801 } 9802 if (pf_dummynet(&pd, s, r, m0) != 0) { 9803 action = PF_DROP; 9804 REASON_SET(&reason, PFRES_MEMORY); 9805 } 9806 break; 9807 } 9808 9809 if (s && action != PF_DROP) { 9810 if (!s->if_index_in && dir == PF_IN) 9811 s->if_index_in = ifp->if_index; 9812 else if (!s->if_index_out && dir == PF_OUT) 9813 s->if_index_out = ifp->if_index; 9814 } 9815 9816 if (s) 9817 PF_STATE_UNLOCK(s); 9818 9819 /* If reassembled packet passed, create new fragments. */ 9820 if (action == PF_PASS && *m0 && dir == PF_OUT && 9821 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9822 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9823 9824 out: 9825 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 9826 9827 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9828 9829 return (action); 9830 } 9831 #endif /* INET6 */ 9832