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