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