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 __FBSDID("$FreeBSD$"); 42 43 #include "opt_bpf.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_pf.h" 47 #include "opt_sctp.h" 48 49 #include <sys/param.h> 50 #include <sys/bus.h> 51 #include <sys/endian.h> 52 #include <sys/gsb_crc32.h> 53 #include <sys/hash.h> 54 #include <sys/interrupt.h> 55 #include <sys/kernel.h> 56 #include <sys/kthread.h> 57 #include <sys/limits.h> 58 #include <sys/mbuf.h> 59 #include <sys/md5.h> 60 #include <sys/random.h> 61 #include <sys/refcount.h> 62 #include <sys/sdt.h> 63 #include <sys/socket.h> 64 #include <sys/sysctl.h> 65 #include <sys/taskqueue.h> 66 #include <sys/ucred.h> 67 68 #include <net/if.h> 69 #include <net/if_var.h> 70 #include <net/if_private.h> 71 #include <net/if_types.h> 72 #include <net/if_vlan_var.h> 73 #include <net/route.h> 74 #include <net/route/nhop.h> 75 #include <net/vnet.h> 76 77 #include <net/pfil.h> 78 #include <net/pfvar.h> 79 #include <net/if_pflog.h> 80 #include <net/if_pfsync.h> 81 82 #include <netinet/in_pcb.h> 83 #include <netinet/in_var.h> 84 #include <netinet/in_fib.h> 85 #include <netinet/ip.h> 86 #include <netinet/ip_fw.h> 87 #include <netinet/ip_icmp.h> 88 #include <netinet/icmp_var.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/tcp.h> 91 #include <netinet/tcp_fsm.h> 92 #include <netinet/tcp_seq.h> 93 #include <netinet/tcp_timer.h> 94 #include <netinet/tcp_var.h> 95 #include <netinet/udp.h> 96 #include <netinet/udp_var.h> 97 98 /* dummynet */ 99 #include <netinet/ip_dummynet.h> 100 #include <netinet/ip_fw.h> 101 #include <netpfil/ipfw/dn_heap.h> 102 #include <netpfil/ipfw/ip_fw_private.h> 103 #include <netpfil/ipfw/ip_dn_private.h> 104 105 #ifdef INET6 106 #include <netinet/ip6.h> 107 #include <netinet/icmp6.h> 108 #include <netinet6/nd6.h> 109 #include <netinet6/ip6_var.h> 110 #include <netinet6/in6_pcb.h> 111 #include <netinet6/in6_fib.h> 112 #include <netinet6/scope6_var.h> 113 #endif /* INET6 */ 114 115 #if defined(SCTP) || defined(SCTP_SUPPORT) 116 #include <netinet/sctp_crc32.h> 117 #endif 118 119 #include <machine/in_cksum.h> 120 #include <security/mac/mac_framework.h> 121 122 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 123 124 SDT_PROVIDER_DEFINE(pf); 125 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 126 "struct pf_kstate *"); 127 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 130 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 131 "struct pf_kstate *"); 132 133 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 134 "struct mbuf *"); 135 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 136 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 137 "int", "struct pf_keth_rule *", "char *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 139 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 140 "int", "struct pf_keth_rule *"); 141 142 /* 143 * Global variables 144 */ 145 146 /* state tables */ 147 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 148 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 149 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 150 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 151 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 152 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 153 VNET_DEFINE(struct pf_kstatus, pf_status); 154 155 VNET_DEFINE(u_int32_t, ticket_altqs_active); 156 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 157 VNET_DEFINE(int, altqs_inactive_open); 158 VNET_DEFINE(u_int32_t, ticket_pabuf); 159 160 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 161 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 162 VNET_DEFINE(u_char, pf_tcp_secret[16]); 163 #define V_pf_tcp_secret VNET(pf_tcp_secret) 164 VNET_DEFINE(int, pf_tcp_secret_init); 165 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 166 VNET_DEFINE(int, pf_tcp_iss_off); 167 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 168 VNET_DECLARE(int, pf_vnet_active); 169 #define V_pf_vnet_active VNET(pf_vnet_active) 170 171 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 172 #define V_pf_purge_idx VNET(pf_purge_idx) 173 174 #ifdef PF_WANT_32_TO_64_COUNTER 175 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 176 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 177 178 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 179 VNET_DEFINE(size_t, pf_allrulecount); 180 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 181 #endif 182 183 /* 184 * Queue for pf_intr() sends. 185 */ 186 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 187 struct pf_send_entry { 188 STAILQ_ENTRY(pf_send_entry) pfse_next; 189 struct mbuf *pfse_m; 190 enum { 191 PFSE_IP, 192 PFSE_IP6, 193 PFSE_ICMP, 194 PFSE_ICMP6, 195 } pfse_type; 196 struct { 197 int type; 198 int code; 199 int mtu; 200 } icmpopts; 201 }; 202 203 STAILQ_HEAD(pf_send_head, pf_send_entry); 204 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 205 #define V_pf_sendqueue VNET(pf_sendqueue) 206 207 static struct mtx_padalign pf_sendqueue_mtx; 208 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 209 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 210 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 211 212 /* 213 * Queue for pf_overload_task() tasks. 214 */ 215 struct pf_overload_entry { 216 SLIST_ENTRY(pf_overload_entry) next; 217 struct pf_addr addr; 218 sa_family_t af; 219 uint8_t dir; 220 struct pf_krule *rule; 221 }; 222 223 SLIST_HEAD(pf_overload_head, pf_overload_entry); 224 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 225 #define V_pf_overloadqueue VNET(pf_overloadqueue) 226 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 227 #define V_pf_overloadtask VNET(pf_overloadtask) 228 229 static struct mtx_padalign pf_overloadqueue_mtx; 230 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 231 "pf overload/flush queue", MTX_DEF); 232 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 233 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 234 235 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 236 struct mtx_padalign pf_unlnkdrules_mtx; 237 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 238 MTX_DEF); 239 240 struct sx pf_config_lock; 241 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 242 243 struct mtx_padalign pf_table_stats_lock; 244 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 245 MTX_DEF); 246 247 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 248 #define V_pf_sources_z VNET(pf_sources_z) 249 uma_zone_t pf_mtag_z; 250 VNET_DEFINE(uma_zone_t, pf_state_z); 251 VNET_DEFINE(uma_zone_t, pf_state_key_z); 252 253 VNET_DEFINE(struct unrhdr64, pf_stateid); 254 255 static void pf_src_tree_remove_state(struct pf_kstate *); 256 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 257 u_int32_t); 258 static void pf_add_threshold(struct pf_threshold *); 259 static int pf_check_threshold(struct pf_threshold *); 260 261 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 262 u_int16_t *, u_int16_t *, struct pf_addr *, 263 u_int16_t, u_int8_t, sa_family_t); 264 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 265 struct tcphdr *, struct pf_state_peer *); 266 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 267 struct pf_addr *, struct pf_addr *, u_int16_t, 268 u_int16_t *, u_int16_t *, u_int16_t *, 269 u_int16_t *, u_int8_t, sa_family_t); 270 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 271 sa_family_t, struct pf_krule *, int); 272 static void pf_detach_state(struct pf_kstate *); 273 static int pf_state_key_attach(struct pf_state_key *, 274 struct pf_state_key *, struct pf_kstate *); 275 static void pf_state_key_detach(struct pf_kstate *, int); 276 static int pf_state_key_ctor(void *, int, void *, int); 277 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 278 void pf_rule_to_actions(struct pf_krule *, 279 struct pf_rule_actions *); 280 static int pf_dummynet(struct pf_pdesc *, int, struct pf_kstate *, 281 struct pf_krule *, struct mbuf **); 282 static int pf_dummynet_route(struct pf_pdesc *, int, 283 struct pf_kstate *, struct pf_krule *, 284 struct ifnet *, struct sockaddr *, struct mbuf **); 285 static int pf_test_eth_rule(int, struct pfi_kkif *, 286 struct mbuf **); 287 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 288 int, struct pfi_kkif *, struct mbuf *, int, 289 struct pf_pdesc *, struct pf_krule **, 290 struct pf_kruleset **, struct inpcb *); 291 static int pf_create_state(struct pf_krule *, struct pf_krule *, 292 struct pf_krule *, struct pf_pdesc *, 293 struct pf_ksrc_node *, struct pf_state_key *, 294 struct pf_state_key *, struct mbuf *, int, 295 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 296 struct pf_kstate **, int, u_int16_t, u_int16_t, 297 int, struct pf_krule_slist *); 298 static int pf_test_fragment(struct pf_krule **, int, 299 struct pfi_kkif *, struct mbuf *, void *, 300 struct pf_pdesc *, struct pf_krule **, 301 struct pf_kruleset **); 302 static int pf_tcp_track_full(struct pf_kstate **, 303 struct pfi_kkif *, struct mbuf *, int, 304 struct pf_pdesc *, u_short *, int *); 305 static int pf_tcp_track_sloppy(struct pf_kstate **, 306 struct pf_pdesc *, u_short *); 307 static int pf_test_state_tcp(struct pf_kstate **, int, 308 struct pfi_kkif *, struct mbuf *, int, 309 void *, struct pf_pdesc *, u_short *); 310 static int pf_test_state_udp(struct pf_kstate **, int, 311 struct pfi_kkif *, struct mbuf *, int, 312 void *, struct pf_pdesc *); 313 static int pf_test_state_icmp(struct pf_kstate **, int, 314 struct pfi_kkif *, struct mbuf *, int, 315 void *, struct pf_pdesc *, u_short *); 316 static int pf_test_state_other(struct pf_kstate **, int, 317 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 318 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 319 int, u_int16_t); 320 static int pf_check_proto_cksum(struct mbuf *, int, int, 321 u_int8_t, sa_family_t); 322 static void pf_print_state_parts(struct pf_kstate *, 323 struct pf_state_key *, struct pf_state_key *); 324 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 325 struct pf_addr_wrap *); 326 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 327 bool, u_int8_t); 328 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 329 struct pf_state_key_cmp *, u_int); 330 static int pf_src_connlimit(struct pf_kstate **); 331 static void pf_overload_task(void *v, int pending); 332 static int pf_insert_src_node(struct pf_ksrc_node **, 333 struct pf_krule *, struct pf_addr *, sa_family_t); 334 static u_int pf_purge_expired_states(u_int, int); 335 static void pf_purge_unlinked_rules(void); 336 static int pf_mtag_uminit(void *, int, int); 337 static void pf_mtag_free(struct m_tag *); 338 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 339 int, struct pf_state_key *); 340 #ifdef INET 341 static void pf_route(struct mbuf **, struct pf_krule *, int, 342 struct ifnet *, struct pf_kstate *, 343 struct pf_pdesc *, struct inpcb *); 344 #endif /* INET */ 345 #ifdef INET6 346 static void pf_change_a6(struct pf_addr *, u_int16_t *, 347 struct pf_addr *, u_int8_t); 348 static void pf_route6(struct mbuf **, struct pf_krule *, int, 349 struct ifnet *, struct pf_kstate *, 350 struct pf_pdesc *, struct inpcb *); 351 #endif /* INET6 */ 352 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 353 354 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 355 356 extern int pf_end_threads; 357 extern struct proc *pf_purge_proc; 358 359 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 360 361 #define PACKET_UNDO_NAT(_m, _pd, _off, _s, _dir) \ 362 do { \ 363 struct pf_state_key *nk; \ 364 if ((_dir) == PF_OUT) \ 365 nk = (_s)->key[PF_SK_STACK]; \ 366 else \ 367 nk = (_s)->key[PF_SK_WIRE]; \ 368 pf_packet_rework_nat(_m, _pd, _off, nk); \ 369 } while (0) 370 371 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 372 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 373 374 #define STATE_LOOKUP(i, k, d, s, pd) \ 375 do { \ 376 (s) = pf_find_state((i), (k), (d)); \ 377 SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s)); \ 378 if ((s) == NULL) \ 379 return (PF_DROP); \ 380 if (PACKET_LOOPED(pd)) \ 381 return (PF_PASS); \ 382 } while (0) 383 384 #define BOUND_IFACE(r, k) \ 385 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 386 387 #define STATE_INC_COUNTERS(s) \ 388 do { \ 389 struct pf_krule_item *mrm; \ 390 counter_u64_add(s->rule.ptr->states_cur, 1); \ 391 counter_u64_add(s->rule.ptr->states_tot, 1); \ 392 if (s->anchor.ptr != NULL) { \ 393 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 394 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 395 } \ 396 if (s->nat_rule.ptr != NULL) { \ 397 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 398 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 399 } \ 400 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 401 counter_u64_add(mrm->r->states_cur, 1); \ 402 counter_u64_add(mrm->r->states_tot, 1); \ 403 } \ 404 } while (0) 405 406 #define STATE_DEC_COUNTERS(s) \ 407 do { \ 408 struct pf_krule_item *mrm; \ 409 if (s->nat_rule.ptr != NULL) \ 410 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 411 if (s->anchor.ptr != NULL) \ 412 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 413 counter_u64_add(s->rule.ptr->states_cur, -1); \ 414 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 415 counter_u64_add(mrm->r->states_cur, -1); \ 416 } while (0) 417 418 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 419 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 420 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 421 VNET_DEFINE(struct pf_idhash *, pf_idhash); 422 VNET_DEFINE(struct pf_srchash *, pf_srchash); 423 424 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 425 "pf(4)"); 426 427 u_long pf_hashmask; 428 u_long pf_srchashmask; 429 static u_long pf_hashsize; 430 static u_long pf_srchashsize; 431 u_long pf_ioctl_maxcount = 65535; 432 433 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 434 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 435 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 436 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 437 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 438 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 439 440 VNET_DEFINE(void *, pf_swi_cookie); 441 VNET_DEFINE(struct intr_event *, pf_swi_ie); 442 443 VNET_DEFINE(uint32_t, pf_hashseed); 444 #define V_pf_hashseed VNET(pf_hashseed) 445 446 int 447 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 448 { 449 450 switch (af) { 451 #ifdef INET 452 case AF_INET: 453 if (a->addr32[0] > b->addr32[0]) 454 return (1); 455 if (a->addr32[0] < b->addr32[0]) 456 return (-1); 457 break; 458 #endif /* INET */ 459 #ifdef INET6 460 case AF_INET6: 461 if (a->addr32[3] > b->addr32[3]) 462 return (1); 463 if (a->addr32[3] < b->addr32[3]) 464 return (-1); 465 if (a->addr32[2] > b->addr32[2]) 466 return (1); 467 if (a->addr32[2] < b->addr32[2]) 468 return (-1); 469 if (a->addr32[1] > b->addr32[1]) 470 return (1); 471 if (a->addr32[1] < b->addr32[1]) 472 return (-1); 473 if (a->addr32[0] > b->addr32[0]) 474 return (1); 475 if (a->addr32[0] < b->addr32[0]) 476 return (-1); 477 break; 478 #endif /* INET6 */ 479 default: 480 panic("%s: unknown address family %u", __func__, af); 481 } 482 return (0); 483 } 484 485 static void 486 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 487 struct pf_state_key *nk) 488 { 489 490 switch (pd->proto) { 491 case IPPROTO_TCP: { 492 struct tcphdr *th = &pd->hdr.tcp; 493 494 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 495 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 496 &th->th_sum, &nk->addr[pd->sidx], 497 nk->port[pd->sidx], 0, pd->af); 498 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 499 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 500 &th->th_sum, &nk->addr[pd->didx], 501 nk->port[pd->didx], 0, pd->af); 502 m_copyback(m, off, sizeof(*th), (caddr_t)th); 503 break; 504 } 505 case IPPROTO_UDP: { 506 struct udphdr *uh = &pd->hdr.udp; 507 508 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 509 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 510 &uh->uh_sum, &nk->addr[pd->sidx], 511 nk->port[pd->sidx], 1, pd->af); 512 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 513 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 514 &uh->uh_sum, &nk->addr[pd->didx], 515 nk->port[pd->didx], 1, pd->af); 516 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 517 break; 518 } 519 case IPPROTO_ICMP: { 520 struct icmp *ih = &pd->hdr.icmp; 521 522 if (nk->port[pd->sidx] != ih->icmp_id) { 523 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 524 ih->icmp_cksum, ih->icmp_id, 525 nk->port[pd->sidx], 0); 526 ih->icmp_id = nk->port[pd->sidx]; 527 pd->sport = &ih->icmp_id; 528 529 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 530 } 531 /* FALLTHROUGH */ 532 } 533 default: 534 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 535 switch (pd->af) { 536 case AF_INET: 537 pf_change_a(&pd->src->v4.s_addr, 538 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 539 0); 540 break; 541 case AF_INET6: 542 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 543 break; 544 } 545 } 546 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 547 switch (pd->af) { 548 case AF_INET: 549 pf_change_a(&pd->dst->v4.s_addr, 550 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 551 0); 552 break; 553 case AF_INET6: 554 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 555 break; 556 } 557 } 558 break; 559 } 560 } 561 562 static __inline uint32_t 563 pf_hashkey(struct pf_state_key *sk) 564 { 565 uint32_t h; 566 567 h = murmur3_32_hash32((uint32_t *)sk, 568 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 569 V_pf_hashseed); 570 571 return (h & pf_hashmask); 572 } 573 574 static __inline uint32_t 575 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 576 { 577 uint32_t h; 578 579 switch (af) { 580 case AF_INET: 581 h = murmur3_32_hash32((uint32_t *)&addr->v4, 582 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 583 break; 584 case AF_INET6: 585 h = murmur3_32_hash32((uint32_t *)&addr->v6, 586 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 587 break; 588 default: 589 panic("%s: unknown address family %u", __func__, af); 590 } 591 592 return (h & pf_srchashmask); 593 } 594 595 #ifdef ALTQ 596 static int 597 pf_state_hash(struct pf_kstate *s) 598 { 599 u_int32_t hv = (intptr_t)s / sizeof(*s); 600 601 hv ^= crc32(&s->src, sizeof(s->src)); 602 hv ^= crc32(&s->dst, sizeof(s->dst)); 603 if (hv == 0) 604 hv = 1; 605 return (hv); 606 } 607 #endif 608 609 static __inline void 610 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 611 { 612 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 613 s->dst.state = newstate; 614 if (which == PF_PEER_DST) 615 return; 616 if (s->src.state == newstate) 617 return; 618 if (s->creatorid == V_pf_status.hostid && 619 s->key[PF_SK_STACK] != NULL && 620 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 621 !(TCPS_HAVEESTABLISHED(s->src.state) || 622 s->src.state == TCPS_CLOSED) && 623 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 624 atomic_add_32(&V_pf_status.states_halfopen, -1); 625 626 s->src.state = newstate; 627 } 628 629 #ifdef INET6 630 void 631 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 632 { 633 switch (af) { 634 #ifdef INET 635 case AF_INET: 636 dst->addr32[0] = src->addr32[0]; 637 break; 638 #endif /* INET */ 639 case AF_INET6: 640 dst->addr32[0] = src->addr32[0]; 641 dst->addr32[1] = src->addr32[1]; 642 dst->addr32[2] = src->addr32[2]; 643 dst->addr32[3] = src->addr32[3]; 644 break; 645 } 646 } 647 #endif /* INET6 */ 648 649 static void 650 pf_init_threshold(struct pf_threshold *threshold, 651 u_int32_t limit, u_int32_t seconds) 652 { 653 threshold->limit = limit * PF_THRESHOLD_MULT; 654 threshold->seconds = seconds; 655 threshold->count = 0; 656 threshold->last = time_uptime; 657 } 658 659 static void 660 pf_add_threshold(struct pf_threshold *threshold) 661 { 662 u_int32_t t = time_uptime, diff = t - threshold->last; 663 664 if (diff >= threshold->seconds) 665 threshold->count = 0; 666 else 667 threshold->count -= threshold->count * diff / 668 threshold->seconds; 669 threshold->count += PF_THRESHOLD_MULT; 670 threshold->last = t; 671 } 672 673 static int 674 pf_check_threshold(struct pf_threshold *threshold) 675 { 676 return (threshold->count > threshold->limit); 677 } 678 679 static int 680 pf_src_connlimit(struct pf_kstate **state) 681 { 682 struct pf_overload_entry *pfoe; 683 int bad = 0; 684 685 PF_STATE_LOCK_ASSERT(*state); 686 /* 687 * XXXKS: The src node is accessed unlocked! 688 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 689 */ 690 691 (*state)->src_node->conn++; 692 (*state)->src.tcp_est = 1; 693 pf_add_threshold(&(*state)->src_node->conn_rate); 694 695 if ((*state)->rule.ptr->max_src_conn && 696 (*state)->rule.ptr->max_src_conn < 697 (*state)->src_node->conn) { 698 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 699 bad++; 700 } 701 702 if ((*state)->rule.ptr->max_src_conn_rate.limit && 703 pf_check_threshold(&(*state)->src_node->conn_rate)) { 704 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 705 bad++; 706 } 707 708 if (!bad) 709 return (0); 710 711 /* Kill this state. */ 712 (*state)->timeout = PFTM_PURGE; 713 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 714 715 if ((*state)->rule.ptr->overload_tbl == NULL) 716 return (1); 717 718 /* Schedule overloading and flushing task. */ 719 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 720 if (pfoe == NULL) 721 return (1); /* too bad :( */ 722 723 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 724 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 725 pfoe->rule = (*state)->rule.ptr; 726 pfoe->dir = (*state)->direction; 727 PF_OVERLOADQ_LOCK(); 728 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 729 PF_OVERLOADQ_UNLOCK(); 730 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 731 732 return (1); 733 } 734 735 static void 736 pf_overload_task(void *v, int pending) 737 { 738 struct pf_overload_head queue; 739 struct pfr_addr p; 740 struct pf_overload_entry *pfoe, *pfoe1; 741 uint32_t killed = 0; 742 743 CURVNET_SET((struct vnet *)v); 744 745 PF_OVERLOADQ_LOCK(); 746 queue = V_pf_overloadqueue; 747 SLIST_INIT(&V_pf_overloadqueue); 748 PF_OVERLOADQ_UNLOCK(); 749 750 bzero(&p, sizeof(p)); 751 SLIST_FOREACH(pfoe, &queue, next) { 752 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 753 if (V_pf_status.debug >= PF_DEBUG_MISC) { 754 printf("%s: blocking address ", __func__); 755 pf_print_host(&pfoe->addr, 0, pfoe->af); 756 printf("\n"); 757 } 758 759 p.pfra_af = pfoe->af; 760 switch (pfoe->af) { 761 #ifdef INET 762 case AF_INET: 763 p.pfra_net = 32; 764 p.pfra_ip4addr = pfoe->addr.v4; 765 break; 766 #endif 767 #ifdef INET6 768 case AF_INET6: 769 p.pfra_net = 128; 770 p.pfra_ip6addr = pfoe->addr.v6; 771 break; 772 #endif 773 } 774 775 PF_RULES_WLOCK(); 776 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 777 PF_RULES_WUNLOCK(); 778 } 779 780 /* 781 * Remove those entries, that don't need flushing. 782 */ 783 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 784 if (pfoe->rule->flush == 0) { 785 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 786 free(pfoe, M_PFTEMP); 787 } else 788 counter_u64_add( 789 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 790 791 /* If nothing to flush, return. */ 792 if (SLIST_EMPTY(&queue)) { 793 CURVNET_RESTORE(); 794 return; 795 } 796 797 for (int i = 0; i <= pf_hashmask; i++) { 798 struct pf_idhash *ih = &V_pf_idhash[i]; 799 struct pf_state_key *sk; 800 struct pf_kstate *s; 801 802 PF_HASHROW_LOCK(ih); 803 LIST_FOREACH(s, &ih->states, entry) { 804 sk = s->key[PF_SK_WIRE]; 805 SLIST_FOREACH(pfoe, &queue, next) 806 if (sk->af == pfoe->af && 807 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 808 pfoe->rule == s->rule.ptr) && 809 ((pfoe->dir == PF_OUT && 810 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 811 (pfoe->dir == PF_IN && 812 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 813 s->timeout = PFTM_PURGE; 814 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 815 killed++; 816 } 817 } 818 PF_HASHROW_UNLOCK(ih); 819 } 820 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 821 free(pfoe, M_PFTEMP); 822 if (V_pf_status.debug >= PF_DEBUG_MISC) 823 printf("%s: %u states killed", __func__, killed); 824 825 CURVNET_RESTORE(); 826 } 827 828 /* 829 * Can return locked on failure, so that we can consistently 830 * allocate and insert a new one. 831 */ 832 struct pf_ksrc_node * 833 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 834 struct pf_srchash **sh, bool returnlocked) 835 { 836 struct pf_ksrc_node *n; 837 838 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 839 840 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 841 PF_HASHROW_LOCK(*sh); 842 LIST_FOREACH(n, &(*sh)->nodes, entry) 843 if (n->rule.ptr == rule && n->af == af && 844 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 845 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 846 break; 847 848 if (n != NULL) { 849 n->states++; 850 PF_HASHROW_UNLOCK(*sh); 851 } else if (returnlocked == false) 852 PF_HASHROW_UNLOCK(*sh); 853 854 return (n); 855 } 856 857 static void 858 pf_free_src_node(struct pf_ksrc_node *sn) 859 { 860 861 for (int i = 0; i < 2; i++) { 862 counter_u64_free(sn->bytes[i]); 863 counter_u64_free(sn->packets[i]); 864 } 865 uma_zfree(V_pf_sources_z, sn); 866 } 867 868 static int 869 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 870 struct pf_addr *src, sa_family_t af) 871 { 872 struct pf_srchash *sh = NULL; 873 874 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 875 rule->rpool.opts & PF_POOL_STICKYADDR), 876 ("%s for non-tracking rule %p", __func__, rule)); 877 878 if (*sn == NULL) 879 *sn = pf_find_src_node(src, rule, af, &sh, true); 880 881 if (*sn == NULL) { 882 PF_HASHROW_ASSERT(sh); 883 884 if (!rule->max_src_nodes || 885 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 886 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 887 else 888 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 889 1); 890 if ((*sn) == NULL) { 891 PF_HASHROW_UNLOCK(sh); 892 return (-1); 893 } 894 895 for (int i = 0; i < 2; i++) { 896 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 897 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 898 899 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 900 pf_free_src_node(*sn); 901 PF_HASHROW_UNLOCK(sh); 902 return (-1); 903 } 904 } 905 906 pf_init_threshold(&(*sn)->conn_rate, 907 rule->max_src_conn_rate.limit, 908 rule->max_src_conn_rate.seconds); 909 910 MPASS((*sn)->lock == NULL); 911 (*sn)->lock = &sh->lock; 912 913 (*sn)->af = af; 914 (*sn)->rule.ptr = rule; 915 PF_ACPY(&(*sn)->addr, src, af); 916 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 917 (*sn)->creation = time_uptime; 918 (*sn)->ruletype = rule->action; 919 (*sn)->states = 1; 920 if ((*sn)->rule.ptr != NULL) 921 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 922 PF_HASHROW_UNLOCK(sh); 923 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 924 } else { 925 if (rule->max_src_states && 926 (*sn)->states >= rule->max_src_states) { 927 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 928 1); 929 return (-1); 930 } 931 } 932 return (0); 933 } 934 935 void 936 pf_unlink_src_node(struct pf_ksrc_node *src) 937 { 938 PF_SRC_NODE_LOCK_ASSERT(src); 939 940 LIST_REMOVE(src, entry); 941 if (src->rule.ptr) 942 counter_u64_add(src->rule.ptr->src_nodes, -1); 943 } 944 945 u_int 946 pf_free_src_nodes(struct pf_ksrc_node_list *head) 947 { 948 struct pf_ksrc_node *sn, *tmp; 949 u_int count = 0; 950 951 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 952 pf_free_src_node(sn); 953 count++; 954 } 955 956 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 957 958 return (count); 959 } 960 961 void 962 pf_mtag_initialize(void) 963 { 964 965 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 966 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 967 UMA_ALIGN_PTR, 0); 968 } 969 970 /* Per-vnet data storage structures initialization. */ 971 void 972 pf_initialize(void) 973 { 974 struct pf_keyhash *kh; 975 struct pf_idhash *ih; 976 struct pf_srchash *sh; 977 u_int i; 978 979 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 980 pf_hashsize = PF_HASHSIZ; 981 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 982 pf_srchashsize = PF_SRCHASHSIZ; 983 984 V_pf_hashseed = arc4random(); 985 986 /* States and state keys storage. */ 987 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 988 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 989 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 990 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 991 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 992 993 V_pf_state_key_z = uma_zcreate("pf state keys", 994 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 995 UMA_ALIGN_PTR, 0); 996 997 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 998 M_PFHASH, M_NOWAIT | M_ZERO); 999 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1000 M_PFHASH, M_NOWAIT | M_ZERO); 1001 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1002 printf("pf: Unable to allocate memory for " 1003 "state_hashsize %lu.\n", pf_hashsize); 1004 1005 free(V_pf_keyhash, M_PFHASH); 1006 free(V_pf_idhash, M_PFHASH); 1007 1008 pf_hashsize = PF_HASHSIZ; 1009 V_pf_keyhash = mallocarray(pf_hashsize, 1010 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1011 V_pf_idhash = mallocarray(pf_hashsize, 1012 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1013 } 1014 1015 pf_hashmask = pf_hashsize - 1; 1016 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1017 i++, kh++, ih++) { 1018 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1019 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1020 } 1021 1022 /* Source nodes. */ 1023 V_pf_sources_z = uma_zcreate("pf source nodes", 1024 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1025 0); 1026 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1027 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1028 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1029 1030 V_pf_srchash = mallocarray(pf_srchashsize, 1031 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1032 if (V_pf_srchash == NULL) { 1033 printf("pf: Unable to allocate memory for " 1034 "source_hashsize %lu.\n", pf_srchashsize); 1035 1036 pf_srchashsize = PF_SRCHASHSIZ; 1037 V_pf_srchash = mallocarray(pf_srchashsize, 1038 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1039 } 1040 1041 pf_srchashmask = pf_srchashsize - 1; 1042 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1043 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1044 1045 /* ALTQ */ 1046 TAILQ_INIT(&V_pf_altqs[0]); 1047 TAILQ_INIT(&V_pf_altqs[1]); 1048 TAILQ_INIT(&V_pf_altqs[2]); 1049 TAILQ_INIT(&V_pf_altqs[3]); 1050 TAILQ_INIT(&V_pf_pabuf); 1051 V_pf_altqs_active = &V_pf_altqs[0]; 1052 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1053 V_pf_altqs_inactive = &V_pf_altqs[2]; 1054 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1055 1056 /* Send & overload+flush queues. */ 1057 STAILQ_INIT(&V_pf_sendqueue); 1058 SLIST_INIT(&V_pf_overloadqueue); 1059 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1060 1061 /* Unlinked, but may be referenced rules. */ 1062 TAILQ_INIT(&V_pf_unlinked_rules); 1063 } 1064 1065 void 1066 pf_mtag_cleanup(void) 1067 { 1068 1069 uma_zdestroy(pf_mtag_z); 1070 } 1071 1072 void 1073 pf_cleanup(void) 1074 { 1075 struct pf_keyhash *kh; 1076 struct pf_idhash *ih; 1077 struct pf_srchash *sh; 1078 struct pf_send_entry *pfse, *next; 1079 u_int i; 1080 1081 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1082 i++, kh++, ih++) { 1083 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1084 __func__)); 1085 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1086 __func__)); 1087 mtx_destroy(&kh->lock); 1088 mtx_destroy(&ih->lock); 1089 } 1090 free(V_pf_keyhash, M_PFHASH); 1091 free(V_pf_idhash, M_PFHASH); 1092 1093 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1094 KASSERT(LIST_EMPTY(&sh->nodes), 1095 ("%s: source node hash not empty", __func__)); 1096 mtx_destroy(&sh->lock); 1097 } 1098 free(V_pf_srchash, M_PFHASH); 1099 1100 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1101 m_freem(pfse->pfse_m); 1102 free(pfse, M_PFTEMP); 1103 } 1104 1105 uma_zdestroy(V_pf_sources_z); 1106 uma_zdestroy(V_pf_state_z); 1107 uma_zdestroy(V_pf_state_key_z); 1108 } 1109 1110 static int 1111 pf_mtag_uminit(void *mem, int size, int how) 1112 { 1113 struct m_tag *t; 1114 1115 t = (struct m_tag *)mem; 1116 t->m_tag_cookie = MTAG_ABI_COMPAT; 1117 t->m_tag_id = PACKET_TAG_PF; 1118 t->m_tag_len = sizeof(struct pf_mtag); 1119 t->m_tag_free = pf_mtag_free; 1120 1121 return (0); 1122 } 1123 1124 static void 1125 pf_mtag_free(struct m_tag *t) 1126 { 1127 1128 uma_zfree(pf_mtag_z, t); 1129 } 1130 1131 struct pf_mtag * 1132 pf_get_mtag(struct mbuf *m) 1133 { 1134 struct m_tag *mtag; 1135 1136 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1137 return ((struct pf_mtag *)(mtag + 1)); 1138 1139 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1140 if (mtag == NULL) 1141 return (NULL); 1142 bzero(mtag + 1, sizeof(struct pf_mtag)); 1143 m_tag_prepend(m, mtag); 1144 1145 return ((struct pf_mtag *)(mtag + 1)); 1146 } 1147 1148 static int 1149 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1150 struct pf_kstate *s) 1151 { 1152 struct pf_keyhash *khs, *khw, *kh; 1153 struct pf_state_key *sk, *cur; 1154 struct pf_kstate *si, *olds = NULL; 1155 int idx; 1156 1157 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1158 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1159 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1160 1161 /* 1162 * We need to lock hash slots of both keys. To avoid deadlock 1163 * we always lock the slot with lower address first. Unlock order 1164 * isn't important. 1165 * 1166 * We also need to lock ID hash slot before dropping key 1167 * locks. On success we return with ID hash slot locked. 1168 */ 1169 1170 if (skw == sks) { 1171 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1172 PF_HASHROW_LOCK(khs); 1173 } else { 1174 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1175 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1176 if (khs == khw) { 1177 PF_HASHROW_LOCK(khs); 1178 } else if (khs < khw) { 1179 PF_HASHROW_LOCK(khs); 1180 PF_HASHROW_LOCK(khw); 1181 } else { 1182 PF_HASHROW_LOCK(khw); 1183 PF_HASHROW_LOCK(khs); 1184 } 1185 } 1186 1187 #define KEYS_UNLOCK() do { \ 1188 if (khs != khw) { \ 1189 PF_HASHROW_UNLOCK(khs); \ 1190 PF_HASHROW_UNLOCK(khw); \ 1191 } else \ 1192 PF_HASHROW_UNLOCK(khs); \ 1193 } while (0) 1194 1195 /* 1196 * First run: start with wire key. 1197 */ 1198 sk = skw; 1199 kh = khw; 1200 idx = PF_SK_WIRE; 1201 1202 MPASS(s->lock == NULL); 1203 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1204 1205 keyattach: 1206 LIST_FOREACH(cur, &kh->keys, entry) 1207 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1208 break; 1209 1210 if (cur != NULL) { 1211 /* Key exists. Check for same kif, if none, add to key. */ 1212 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1213 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1214 1215 PF_HASHROW_LOCK(ih); 1216 if (si->kif == s->kif && 1217 si->direction == s->direction) { 1218 if (sk->proto == IPPROTO_TCP && 1219 si->src.state >= TCPS_FIN_WAIT_2 && 1220 si->dst.state >= TCPS_FIN_WAIT_2) { 1221 /* 1222 * New state matches an old >FIN_WAIT_2 1223 * state. We can't drop key hash locks, 1224 * thus we can't unlink it properly. 1225 * 1226 * As a workaround we drop it into 1227 * TCPS_CLOSED state, schedule purge 1228 * ASAP and push it into the very end 1229 * of the slot TAILQ, so that it won't 1230 * conflict with our new state. 1231 */ 1232 pf_set_protostate(si, PF_PEER_BOTH, 1233 TCPS_CLOSED); 1234 si->timeout = PFTM_PURGE; 1235 olds = si; 1236 } else { 1237 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1238 printf("pf: %s key attach " 1239 "failed on %s: ", 1240 (idx == PF_SK_WIRE) ? 1241 "wire" : "stack", 1242 s->kif->pfik_name); 1243 pf_print_state_parts(s, 1244 (idx == PF_SK_WIRE) ? 1245 sk : NULL, 1246 (idx == PF_SK_STACK) ? 1247 sk : NULL); 1248 printf(", existing: "); 1249 pf_print_state_parts(si, 1250 (idx == PF_SK_WIRE) ? 1251 sk : NULL, 1252 (idx == PF_SK_STACK) ? 1253 sk : NULL); 1254 printf("\n"); 1255 } 1256 PF_HASHROW_UNLOCK(ih); 1257 KEYS_UNLOCK(); 1258 uma_zfree(V_pf_state_key_z, sk); 1259 if (idx == PF_SK_STACK) 1260 pf_detach_state(s); 1261 return (EEXIST); /* collision! */ 1262 } 1263 } 1264 PF_HASHROW_UNLOCK(ih); 1265 } 1266 uma_zfree(V_pf_state_key_z, sk); 1267 s->key[idx] = cur; 1268 } else { 1269 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1270 s->key[idx] = sk; 1271 } 1272 1273 stateattach: 1274 /* List is sorted, if-bound states before floating. */ 1275 if (s->kif == V_pfi_all) 1276 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1277 else 1278 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1279 1280 if (olds) { 1281 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1282 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1283 key_list[idx]); 1284 olds = NULL; 1285 } 1286 1287 /* 1288 * Attach done. See how should we (or should not?) 1289 * attach a second key. 1290 */ 1291 if (sks == skw) { 1292 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1293 idx = PF_SK_STACK; 1294 sks = NULL; 1295 goto stateattach; 1296 } else if (sks != NULL) { 1297 /* 1298 * Continue attaching with stack key. 1299 */ 1300 sk = sks; 1301 kh = khs; 1302 idx = PF_SK_STACK; 1303 sks = NULL; 1304 goto keyattach; 1305 } 1306 1307 PF_STATE_LOCK(s); 1308 KEYS_UNLOCK(); 1309 1310 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1311 ("%s failure", __func__)); 1312 1313 return (0); 1314 #undef KEYS_UNLOCK 1315 } 1316 1317 static void 1318 pf_detach_state(struct pf_kstate *s) 1319 { 1320 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1321 struct pf_keyhash *kh; 1322 1323 if (sks != NULL) { 1324 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1325 PF_HASHROW_LOCK(kh); 1326 if (s->key[PF_SK_STACK] != NULL) 1327 pf_state_key_detach(s, PF_SK_STACK); 1328 /* 1329 * If both point to same key, then we are done. 1330 */ 1331 if (sks == s->key[PF_SK_WIRE]) { 1332 pf_state_key_detach(s, PF_SK_WIRE); 1333 PF_HASHROW_UNLOCK(kh); 1334 return; 1335 } 1336 PF_HASHROW_UNLOCK(kh); 1337 } 1338 1339 if (s->key[PF_SK_WIRE] != NULL) { 1340 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1341 PF_HASHROW_LOCK(kh); 1342 if (s->key[PF_SK_WIRE] != NULL) 1343 pf_state_key_detach(s, PF_SK_WIRE); 1344 PF_HASHROW_UNLOCK(kh); 1345 } 1346 } 1347 1348 static void 1349 pf_state_key_detach(struct pf_kstate *s, int idx) 1350 { 1351 struct pf_state_key *sk = s->key[idx]; 1352 #ifdef INVARIANTS 1353 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1354 1355 PF_HASHROW_ASSERT(kh); 1356 #endif 1357 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1358 s->key[idx] = NULL; 1359 1360 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1361 LIST_REMOVE(sk, entry); 1362 uma_zfree(V_pf_state_key_z, sk); 1363 } 1364 } 1365 1366 static int 1367 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1368 { 1369 struct pf_state_key *sk = mem; 1370 1371 bzero(sk, sizeof(struct pf_state_key_cmp)); 1372 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1373 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1374 1375 return (0); 1376 } 1377 1378 struct pf_state_key * 1379 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1380 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1381 { 1382 struct pf_state_key *sk; 1383 1384 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1385 if (sk == NULL) 1386 return (NULL); 1387 1388 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1389 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1390 sk->port[pd->sidx] = sport; 1391 sk->port[pd->didx] = dport; 1392 sk->proto = pd->proto; 1393 sk->af = pd->af; 1394 1395 return (sk); 1396 } 1397 1398 struct pf_state_key * 1399 pf_state_key_clone(struct pf_state_key *orig) 1400 { 1401 struct pf_state_key *sk; 1402 1403 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1404 if (sk == NULL) 1405 return (NULL); 1406 1407 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1408 1409 return (sk); 1410 } 1411 1412 int 1413 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1414 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1415 { 1416 struct pf_idhash *ih; 1417 struct pf_kstate *cur; 1418 int error; 1419 1420 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1421 ("%s: sks not pristine", __func__)); 1422 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1423 ("%s: skw not pristine", __func__)); 1424 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1425 1426 s->kif = kif; 1427 s->orig_kif = orig_kif; 1428 1429 if (s->id == 0 && s->creatorid == 0) { 1430 s->id = alloc_unr64(&V_pf_stateid); 1431 s->id = htobe64(s->id); 1432 s->creatorid = V_pf_status.hostid; 1433 } 1434 1435 /* Returns with ID locked on success. */ 1436 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1437 return (error); 1438 1439 ih = &V_pf_idhash[PF_IDHASH(s)]; 1440 PF_HASHROW_ASSERT(ih); 1441 LIST_FOREACH(cur, &ih->states, entry) 1442 if (cur->id == s->id && cur->creatorid == s->creatorid) 1443 break; 1444 1445 if (cur != NULL) { 1446 PF_HASHROW_UNLOCK(ih); 1447 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1448 printf("pf: state ID collision: " 1449 "id: %016llx creatorid: %08x\n", 1450 (unsigned long long)be64toh(s->id), 1451 ntohl(s->creatorid)); 1452 } 1453 pf_detach_state(s); 1454 return (EEXIST); 1455 } 1456 LIST_INSERT_HEAD(&ih->states, s, entry); 1457 /* One for keys, one for ID hash. */ 1458 refcount_init(&s->refs, 2); 1459 1460 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1461 if (V_pfsync_insert_state_ptr != NULL) 1462 V_pfsync_insert_state_ptr(s); 1463 1464 /* Returns locked. */ 1465 return (0); 1466 } 1467 1468 /* 1469 * Find state by ID: returns with locked row on success. 1470 */ 1471 struct pf_kstate * 1472 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1473 { 1474 struct pf_idhash *ih; 1475 struct pf_kstate *s; 1476 1477 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1478 1479 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1480 1481 PF_HASHROW_LOCK(ih); 1482 LIST_FOREACH(s, &ih->states, entry) 1483 if (s->id == id && s->creatorid == creatorid) 1484 break; 1485 1486 if (s == NULL) 1487 PF_HASHROW_UNLOCK(ih); 1488 1489 return (s); 1490 } 1491 1492 /* 1493 * Find state by key. 1494 * Returns with ID hash slot locked on success. 1495 */ 1496 static struct pf_kstate * 1497 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1498 { 1499 struct pf_keyhash *kh; 1500 struct pf_state_key *sk; 1501 struct pf_kstate *s; 1502 int idx; 1503 1504 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1505 1506 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1507 1508 PF_HASHROW_LOCK(kh); 1509 LIST_FOREACH(sk, &kh->keys, entry) 1510 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1511 break; 1512 if (sk == NULL) { 1513 PF_HASHROW_UNLOCK(kh); 1514 return (NULL); 1515 } 1516 1517 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1518 1519 /* List is sorted, if-bound states before floating ones. */ 1520 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1521 if (s->kif == V_pfi_all || s->kif == kif) { 1522 PF_STATE_LOCK(s); 1523 PF_HASHROW_UNLOCK(kh); 1524 if (__predict_false(s->timeout >= PFTM_MAX)) { 1525 /* 1526 * State is either being processed by 1527 * pf_unlink_state() in an other thread, or 1528 * is scheduled for immediate expiry. 1529 */ 1530 PF_STATE_UNLOCK(s); 1531 return (NULL); 1532 } 1533 return (s); 1534 } 1535 PF_HASHROW_UNLOCK(kh); 1536 1537 return (NULL); 1538 } 1539 1540 /* 1541 * Returns with ID hash slot locked on success. 1542 */ 1543 struct pf_kstate * 1544 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1545 { 1546 struct pf_keyhash *kh; 1547 struct pf_state_key *sk; 1548 struct pf_kstate *s, *ret = NULL; 1549 int idx, inout = 0; 1550 1551 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1552 1553 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1554 1555 PF_HASHROW_LOCK(kh); 1556 LIST_FOREACH(sk, &kh->keys, entry) 1557 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1558 break; 1559 if (sk == NULL) { 1560 PF_HASHROW_UNLOCK(kh); 1561 return (NULL); 1562 } 1563 switch (dir) { 1564 case PF_IN: 1565 idx = PF_SK_WIRE; 1566 break; 1567 case PF_OUT: 1568 idx = PF_SK_STACK; 1569 break; 1570 case PF_INOUT: 1571 idx = PF_SK_WIRE; 1572 inout = 1; 1573 break; 1574 default: 1575 panic("%s: dir %u", __func__, dir); 1576 } 1577 second_run: 1578 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1579 if (more == NULL) { 1580 PF_STATE_LOCK(s); 1581 PF_HASHROW_UNLOCK(kh); 1582 return (s); 1583 } 1584 1585 if (ret) 1586 (*more)++; 1587 else { 1588 ret = s; 1589 PF_STATE_LOCK(s); 1590 } 1591 } 1592 if (inout == 1) { 1593 inout = 0; 1594 idx = PF_SK_STACK; 1595 goto second_run; 1596 } 1597 PF_HASHROW_UNLOCK(kh); 1598 1599 return (ret); 1600 } 1601 1602 /* 1603 * FIXME 1604 * This routine is inefficient -- locks the state only to unlock immediately on 1605 * return. 1606 * It is racy -- after the state is unlocked nothing stops other threads from 1607 * removing it. 1608 */ 1609 bool 1610 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1611 { 1612 struct pf_kstate *s; 1613 1614 s = pf_find_state_all(key, dir, NULL); 1615 if (s != NULL) { 1616 PF_STATE_UNLOCK(s); 1617 return (true); 1618 } 1619 return (false); 1620 } 1621 1622 /* END state table stuff */ 1623 1624 static void 1625 pf_send(struct pf_send_entry *pfse) 1626 { 1627 1628 PF_SENDQ_LOCK(); 1629 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1630 PF_SENDQ_UNLOCK(); 1631 swi_sched(V_pf_swi_cookie, 0); 1632 } 1633 1634 static bool 1635 pf_isforlocal(struct mbuf *m, int af) 1636 { 1637 switch (af) { 1638 #ifdef INET 1639 case AF_INET: { 1640 struct ip *ip = mtod(m, struct ip *); 1641 1642 return (in_localip(ip->ip_dst)); 1643 } 1644 #endif 1645 #ifdef INET6 1646 case AF_INET6: { 1647 struct ip6_hdr *ip6; 1648 struct in6_ifaddr *ia; 1649 ip6 = mtod(m, struct ip6_hdr *); 1650 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1651 if (ia == NULL) 1652 return (false); 1653 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1654 } 1655 #endif 1656 default: 1657 panic("Unsupported af %d", af); 1658 } 1659 1660 return (false); 1661 } 1662 1663 void 1664 pf_intr(void *v) 1665 { 1666 struct epoch_tracker et; 1667 struct pf_send_head queue; 1668 struct pf_send_entry *pfse, *next; 1669 1670 CURVNET_SET((struct vnet *)v); 1671 1672 PF_SENDQ_LOCK(); 1673 queue = V_pf_sendqueue; 1674 STAILQ_INIT(&V_pf_sendqueue); 1675 PF_SENDQ_UNLOCK(); 1676 1677 NET_EPOCH_ENTER(et); 1678 1679 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1680 switch (pfse->pfse_type) { 1681 #ifdef INET 1682 case PFSE_IP: { 1683 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1684 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1685 pfse->pfse_m->m_pkthdr.csum_flags |= 1686 CSUM_IP_VALID | CSUM_IP_CHECKED; 1687 ip_input(pfse->pfse_m); 1688 } else { 1689 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1690 NULL); 1691 } 1692 break; 1693 } 1694 case PFSE_ICMP: 1695 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1696 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1697 break; 1698 #endif /* INET */ 1699 #ifdef INET6 1700 case PFSE_IP6: 1701 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1702 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1703 ip6_input(pfse->pfse_m); 1704 } else { 1705 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1706 NULL, NULL); 1707 } 1708 break; 1709 case PFSE_ICMP6: 1710 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1711 pfse->icmpopts.code, pfse->icmpopts.mtu); 1712 break; 1713 #endif /* INET6 */ 1714 default: 1715 panic("%s: unknown type", __func__); 1716 } 1717 free(pfse, M_PFTEMP); 1718 } 1719 NET_EPOCH_EXIT(et); 1720 CURVNET_RESTORE(); 1721 } 1722 1723 #define pf_purge_thread_period (hz / 10) 1724 1725 #ifdef PF_WANT_32_TO_64_COUNTER 1726 static void 1727 pf_status_counter_u64_periodic(void) 1728 { 1729 1730 PF_RULES_RASSERT(); 1731 1732 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1733 return; 1734 } 1735 1736 for (int i = 0; i < FCNT_MAX; i++) { 1737 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1738 } 1739 } 1740 1741 static void 1742 pf_kif_counter_u64_periodic(void) 1743 { 1744 struct pfi_kkif *kif; 1745 size_t r, run; 1746 1747 PF_RULES_RASSERT(); 1748 1749 if (__predict_false(V_pf_allkifcount == 0)) { 1750 return; 1751 } 1752 1753 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1754 return; 1755 } 1756 1757 run = V_pf_allkifcount / 10; 1758 if (run < 5) 1759 run = 5; 1760 1761 for (r = 0; r < run; r++) { 1762 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1763 if (kif == NULL) { 1764 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1765 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1766 break; 1767 } 1768 1769 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1770 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1771 1772 for (int i = 0; i < 2; i++) { 1773 for (int j = 0; j < 2; j++) { 1774 for (int k = 0; k < 2; k++) { 1775 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1776 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1777 } 1778 } 1779 } 1780 } 1781 } 1782 1783 static void 1784 pf_rule_counter_u64_periodic(void) 1785 { 1786 struct pf_krule *rule; 1787 size_t r, run; 1788 1789 PF_RULES_RASSERT(); 1790 1791 if (__predict_false(V_pf_allrulecount == 0)) { 1792 return; 1793 } 1794 1795 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1796 return; 1797 } 1798 1799 run = V_pf_allrulecount / 10; 1800 if (run < 5) 1801 run = 5; 1802 1803 for (r = 0; r < run; r++) { 1804 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1805 if (rule == NULL) { 1806 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1807 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1808 break; 1809 } 1810 1811 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1812 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1813 1814 pf_counter_u64_periodic(&rule->evaluations); 1815 for (int i = 0; i < 2; i++) { 1816 pf_counter_u64_periodic(&rule->packets[i]); 1817 pf_counter_u64_periodic(&rule->bytes[i]); 1818 } 1819 } 1820 } 1821 1822 static void 1823 pf_counter_u64_periodic_main(void) 1824 { 1825 PF_RULES_RLOCK_TRACKER; 1826 1827 V_pf_counter_periodic_iter++; 1828 1829 PF_RULES_RLOCK(); 1830 pf_counter_u64_critical_enter(); 1831 pf_status_counter_u64_periodic(); 1832 pf_kif_counter_u64_periodic(); 1833 pf_rule_counter_u64_periodic(); 1834 pf_counter_u64_critical_exit(); 1835 PF_RULES_RUNLOCK(); 1836 } 1837 #else 1838 #define pf_counter_u64_periodic_main() do { } while (0) 1839 #endif 1840 1841 void 1842 pf_purge_thread(void *unused __unused) 1843 { 1844 VNET_ITERATOR_DECL(vnet_iter); 1845 1846 sx_xlock(&pf_end_lock); 1847 while (pf_end_threads == 0) { 1848 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1849 1850 VNET_LIST_RLOCK(); 1851 VNET_FOREACH(vnet_iter) { 1852 CURVNET_SET(vnet_iter); 1853 1854 /* Wait until V_pf_default_rule is initialized. */ 1855 if (V_pf_vnet_active == 0) { 1856 CURVNET_RESTORE(); 1857 continue; 1858 } 1859 1860 pf_counter_u64_periodic_main(); 1861 1862 /* 1863 * Process 1/interval fraction of the state 1864 * table every run. 1865 */ 1866 V_pf_purge_idx = 1867 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1868 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1869 1870 /* 1871 * Purge other expired types every 1872 * PFTM_INTERVAL seconds. 1873 */ 1874 if (V_pf_purge_idx == 0) { 1875 /* 1876 * Order is important: 1877 * - states and src nodes reference rules 1878 * - states and rules reference kifs 1879 */ 1880 pf_purge_expired_fragments(); 1881 pf_purge_expired_src_nodes(); 1882 pf_purge_unlinked_rules(); 1883 pfi_kkif_purge(); 1884 } 1885 CURVNET_RESTORE(); 1886 } 1887 VNET_LIST_RUNLOCK(); 1888 } 1889 1890 pf_end_threads++; 1891 sx_xunlock(&pf_end_lock); 1892 kproc_exit(0); 1893 } 1894 1895 void 1896 pf_unload_vnet_purge(void) 1897 { 1898 1899 /* 1900 * To cleanse up all kifs and rules we need 1901 * two runs: first one clears reference flags, 1902 * then pf_purge_expired_states() doesn't 1903 * raise them, and then second run frees. 1904 */ 1905 pf_purge_unlinked_rules(); 1906 pfi_kkif_purge(); 1907 1908 /* 1909 * Now purge everything. 1910 */ 1911 pf_purge_expired_states(0, pf_hashmask); 1912 pf_purge_fragments(UINT_MAX); 1913 pf_purge_expired_src_nodes(); 1914 1915 /* 1916 * Now all kifs & rules should be unreferenced, 1917 * thus should be successfully freed. 1918 */ 1919 pf_purge_unlinked_rules(); 1920 pfi_kkif_purge(); 1921 } 1922 1923 u_int32_t 1924 pf_state_expires(const struct pf_kstate *state) 1925 { 1926 u_int32_t timeout; 1927 u_int32_t start; 1928 u_int32_t end; 1929 u_int32_t states; 1930 1931 /* handle all PFTM_* > PFTM_MAX here */ 1932 if (state->timeout == PFTM_PURGE) 1933 return (time_uptime); 1934 KASSERT(state->timeout != PFTM_UNLINKED, 1935 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1936 KASSERT((state->timeout < PFTM_MAX), 1937 ("pf_state_expires: timeout > PFTM_MAX")); 1938 timeout = state->rule.ptr->timeout[state->timeout]; 1939 if (!timeout) 1940 timeout = V_pf_default_rule.timeout[state->timeout]; 1941 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1942 if (start && state->rule.ptr != &V_pf_default_rule) { 1943 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1944 states = counter_u64_fetch(state->rule.ptr->states_cur); 1945 } else { 1946 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1947 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1948 states = V_pf_status.states; 1949 } 1950 if (end && states > start && start < end) { 1951 if (states < end) { 1952 timeout = (u_int64_t)timeout * (end - states) / 1953 (end - start); 1954 return (state->expire + timeout); 1955 } 1956 else 1957 return (time_uptime); 1958 } 1959 return (state->expire + timeout); 1960 } 1961 1962 void 1963 pf_purge_expired_src_nodes(void) 1964 { 1965 struct pf_ksrc_node_list freelist; 1966 struct pf_srchash *sh; 1967 struct pf_ksrc_node *cur, *next; 1968 int i; 1969 1970 LIST_INIT(&freelist); 1971 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1972 PF_HASHROW_LOCK(sh); 1973 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1974 if (cur->states == 0 && cur->expire <= time_uptime) { 1975 pf_unlink_src_node(cur); 1976 LIST_INSERT_HEAD(&freelist, cur, entry); 1977 } else if (cur->rule.ptr != NULL) 1978 cur->rule.ptr->rule_ref |= PFRULE_REFS; 1979 PF_HASHROW_UNLOCK(sh); 1980 } 1981 1982 pf_free_src_nodes(&freelist); 1983 1984 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 1985 } 1986 1987 static void 1988 pf_src_tree_remove_state(struct pf_kstate *s) 1989 { 1990 struct pf_ksrc_node *sn; 1991 uint32_t timeout; 1992 1993 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 1994 s->rule.ptr->timeout[PFTM_SRC_NODE] : 1995 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1996 1997 if (s->src_node != NULL) { 1998 sn = s->src_node; 1999 PF_SRC_NODE_LOCK(sn); 2000 if (s->src.tcp_est) 2001 --sn->conn; 2002 if (--sn->states == 0) 2003 sn->expire = time_uptime + timeout; 2004 PF_SRC_NODE_UNLOCK(sn); 2005 } 2006 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2007 sn = s->nat_src_node; 2008 PF_SRC_NODE_LOCK(sn); 2009 if (--sn->states == 0) 2010 sn->expire = time_uptime + timeout; 2011 PF_SRC_NODE_UNLOCK(sn); 2012 } 2013 s->src_node = s->nat_src_node = NULL; 2014 } 2015 2016 /* 2017 * Unlink and potentilly free a state. Function may be 2018 * called with ID hash row locked, but always returns 2019 * unlocked, since it needs to go through key hash locking. 2020 */ 2021 int 2022 pf_unlink_state(struct pf_kstate *s) 2023 { 2024 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2025 2026 PF_HASHROW_ASSERT(ih); 2027 2028 if (s->timeout == PFTM_UNLINKED) { 2029 /* 2030 * State is being processed 2031 * by pf_unlink_state() in 2032 * an other thread. 2033 */ 2034 PF_HASHROW_UNLOCK(ih); 2035 return (0); /* XXXGL: undefined actually */ 2036 } 2037 2038 if (s->src.state == PF_TCPS_PROXY_DST) { 2039 /* XXX wire key the right one? */ 2040 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2041 &s->key[PF_SK_WIRE]->addr[1], 2042 &s->key[PF_SK_WIRE]->addr[0], 2043 s->key[PF_SK_WIRE]->port[1], 2044 s->key[PF_SK_WIRE]->port[0], 2045 s->src.seqhi, s->src.seqlo + 1, 2046 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, s->rtableid); 2047 } 2048 2049 LIST_REMOVE(s, entry); 2050 pf_src_tree_remove_state(s); 2051 2052 if (V_pfsync_delete_state_ptr != NULL) 2053 V_pfsync_delete_state_ptr(s); 2054 2055 STATE_DEC_COUNTERS(s); 2056 2057 s->timeout = PFTM_UNLINKED; 2058 2059 /* Ensure we remove it from the list of halfopen states, if needed. */ 2060 if (s->key[PF_SK_STACK] != NULL && 2061 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2062 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2063 2064 PF_HASHROW_UNLOCK(ih); 2065 2066 pf_detach_state(s); 2067 /* pf_state_insert() initialises refs to 2 */ 2068 return (pf_release_staten(s, 2)); 2069 } 2070 2071 struct pf_kstate * 2072 pf_alloc_state(int flags) 2073 { 2074 2075 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2076 } 2077 2078 void 2079 pf_free_state(struct pf_kstate *cur) 2080 { 2081 struct pf_krule_item *ri; 2082 2083 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2084 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2085 cur->timeout)); 2086 2087 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2088 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2089 free(ri, M_PF_RULE_ITEM); 2090 } 2091 2092 pf_normalize_tcp_cleanup(cur); 2093 uma_zfree(V_pf_state_z, cur); 2094 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2095 } 2096 2097 /* 2098 * Called only from pf_purge_thread(), thus serialized. 2099 */ 2100 static u_int 2101 pf_purge_expired_states(u_int i, int maxcheck) 2102 { 2103 struct pf_idhash *ih; 2104 struct pf_kstate *s; 2105 struct pf_krule_item *mrm; 2106 2107 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2108 2109 /* 2110 * Go through hash and unlink states that expire now. 2111 */ 2112 while (maxcheck > 0) { 2113 ih = &V_pf_idhash[i]; 2114 2115 /* only take the lock if we expect to do work */ 2116 if (!LIST_EMPTY(&ih->states)) { 2117 relock: 2118 PF_HASHROW_LOCK(ih); 2119 LIST_FOREACH(s, &ih->states, entry) { 2120 if (pf_state_expires(s) <= time_uptime) { 2121 V_pf_status.states -= 2122 pf_unlink_state(s); 2123 goto relock; 2124 } 2125 s->rule.ptr->rule_ref |= PFRULE_REFS; 2126 if (s->nat_rule.ptr != NULL) 2127 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2128 if (s->anchor.ptr != NULL) 2129 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2130 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2131 SLIST_FOREACH(mrm, &s->match_rules, entry) 2132 mrm->r->rule_ref |= PFRULE_REFS; 2133 if (s->rt_kif) 2134 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2135 } 2136 PF_HASHROW_UNLOCK(ih); 2137 } 2138 2139 /* Return when we hit end of hash. */ 2140 if (++i > pf_hashmask) { 2141 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2142 return (0); 2143 } 2144 2145 maxcheck--; 2146 } 2147 2148 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2149 2150 return (i); 2151 } 2152 2153 static void 2154 pf_purge_unlinked_rules(void) 2155 { 2156 struct pf_krulequeue tmpq; 2157 struct pf_krule *r, *r1; 2158 2159 /* 2160 * If we have overloading task pending, then we'd 2161 * better skip purging this time. There is a tiny 2162 * probability that overloading task references 2163 * an already unlinked rule. 2164 */ 2165 PF_OVERLOADQ_LOCK(); 2166 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2167 PF_OVERLOADQ_UNLOCK(); 2168 return; 2169 } 2170 PF_OVERLOADQ_UNLOCK(); 2171 2172 /* 2173 * Do naive mark-and-sweep garbage collecting of old rules. 2174 * Reference flag is raised by pf_purge_expired_states() 2175 * and pf_purge_expired_src_nodes(). 2176 * 2177 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2178 * use a temporary queue. 2179 */ 2180 TAILQ_INIT(&tmpq); 2181 PF_UNLNKDRULES_LOCK(); 2182 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2183 if (!(r->rule_ref & PFRULE_REFS)) { 2184 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2185 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2186 } else 2187 r->rule_ref &= ~PFRULE_REFS; 2188 } 2189 PF_UNLNKDRULES_UNLOCK(); 2190 2191 if (!TAILQ_EMPTY(&tmpq)) { 2192 PF_CONFIG_LOCK(); 2193 PF_RULES_WLOCK(); 2194 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2195 TAILQ_REMOVE(&tmpq, r, entries); 2196 pf_free_rule(r); 2197 } 2198 PF_RULES_WUNLOCK(); 2199 PF_CONFIG_UNLOCK(); 2200 } 2201 } 2202 2203 void 2204 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2205 { 2206 switch (af) { 2207 #ifdef INET 2208 case AF_INET: { 2209 u_int32_t a = ntohl(addr->addr32[0]); 2210 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2211 (a>>8)&255, a&255); 2212 if (p) { 2213 p = ntohs(p); 2214 printf(":%u", p); 2215 } 2216 break; 2217 } 2218 #endif /* INET */ 2219 #ifdef INET6 2220 case AF_INET6: { 2221 u_int16_t b; 2222 u_int8_t i, curstart, curend, maxstart, maxend; 2223 curstart = curend = maxstart = maxend = 255; 2224 for (i = 0; i < 8; i++) { 2225 if (!addr->addr16[i]) { 2226 if (curstart == 255) 2227 curstart = i; 2228 curend = i; 2229 } else { 2230 if ((curend - curstart) > 2231 (maxend - maxstart)) { 2232 maxstart = curstart; 2233 maxend = curend; 2234 } 2235 curstart = curend = 255; 2236 } 2237 } 2238 if ((curend - curstart) > 2239 (maxend - maxstart)) { 2240 maxstart = curstart; 2241 maxend = curend; 2242 } 2243 for (i = 0; i < 8; i++) { 2244 if (i >= maxstart && i <= maxend) { 2245 if (i == 0) 2246 printf(":"); 2247 if (i == maxend) 2248 printf(":"); 2249 } else { 2250 b = ntohs(addr->addr16[i]); 2251 printf("%x", b); 2252 if (i < 7) 2253 printf(":"); 2254 } 2255 } 2256 if (p) { 2257 p = ntohs(p); 2258 printf("[%u]", p); 2259 } 2260 break; 2261 } 2262 #endif /* INET6 */ 2263 } 2264 } 2265 2266 void 2267 pf_print_state(struct pf_kstate *s) 2268 { 2269 pf_print_state_parts(s, NULL, NULL); 2270 } 2271 2272 static void 2273 pf_print_state_parts(struct pf_kstate *s, 2274 struct pf_state_key *skwp, struct pf_state_key *sksp) 2275 { 2276 struct pf_state_key *skw, *sks; 2277 u_int8_t proto, dir; 2278 2279 /* Do our best to fill these, but they're skipped if NULL */ 2280 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2281 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2282 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2283 dir = s ? s->direction : 0; 2284 2285 switch (proto) { 2286 case IPPROTO_IPV4: 2287 printf("IPv4"); 2288 break; 2289 case IPPROTO_IPV6: 2290 printf("IPv6"); 2291 break; 2292 case IPPROTO_TCP: 2293 printf("TCP"); 2294 break; 2295 case IPPROTO_UDP: 2296 printf("UDP"); 2297 break; 2298 case IPPROTO_ICMP: 2299 printf("ICMP"); 2300 break; 2301 case IPPROTO_ICMPV6: 2302 printf("ICMPv6"); 2303 break; 2304 default: 2305 printf("%u", proto); 2306 break; 2307 } 2308 switch (dir) { 2309 case PF_IN: 2310 printf(" in"); 2311 break; 2312 case PF_OUT: 2313 printf(" out"); 2314 break; 2315 } 2316 if (skw) { 2317 printf(" wire: "); 2318 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2319 printf(" "); 2320 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2321 } 2322 if (sks) { 2323 printf(" stack: "); 2324 if (sks != skw) { 2325 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2326 printf(" "); 2327 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2328 } else 2329 printf("-"); 2330 } 2331 if (s) { 2332 if (proto == IPPROTO_TCP) { 2333 printf(" [lo=%u high=%u win=%u modulator=%u", 2334 s->src.seqlo, s->src.seqhi, 2335 s->src.max_win, s->src.seqdiff); 2336 if (s->src.wscale && s->dst.wscale) 2337 printf(" wscale=%u", 2338 s->src.wscale & PF_WSCALE_MASK); 2339 printf("]"); 2340 printf(" [lo=%u high=%u win=%u modulator=%u", 2341 s->dst.seqlo, s->dst.seqhi, 2342 s->dst.max_win, s->dst.seqdiff); 2343 if (s->src.wscale && s->dst.wscale) 2344 printf(" wscale=%u", 2345 s->dst.wscale & PF_WSCALE_MASK); 2346 printf("]"); 2347 } 2348 printf(" %u:%u", s->src.state, s->dst.state); 2349 } 2350 } 2351 2352 void 2353 pf_print_flags(u_int8_t f) 2354 { 2355 if (f) 2356 printf(" "); 2357 if (f & TH_FIN) 2358 printf("F"); 2359 if (f & TH_SYN) 2360 printf("S"); 2361 if (f & TH_RST) 2362 printf("R"); 2363 if (f & TH_PUSH) 2364 printf("P"); 2365 if (f & TH_ACK) 2366 printf("A"); 2367 if (f & TH_URG) 2368 printf("U"); 2369 if (f & TH_ECE) 2370 printf("E"); 2371 if (f & TH_CWR) 2372 printf("W"); 2373 } 2374 2375 #define PF_SET_SKIP_STEPS(i) \ 2376 do { \ 2377 while (head[i] != cur) { \ 2378 head[i]->skip[i].ptr = cur; \ 2379 head[i] = TAILQ_NEXT(head[i], entries); \ 2380 } \ 2381 } while (0) 2382 2383 void 2384 pf_calc_skip_steps(struct pf_krulequeue *rules) 2385 { 2386 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2387 int i; 2388 2389 cur = TAILQ_FIRST(rules); 2390 prev = cur; 2391 for (i = 0; i < PF_SKIP_COUNT; ++i) 2392 head[i] = cur; 2393 while (cur != NULL) { 2394 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2395 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2396 if (cur->direction != prev->direction) 2397 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2398 if (cur->af != prev->af) 2399 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2400 if (cur->proto != prev->proto) 2401 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2402 if (cur->src.neg != prev->src.neg || 2403 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2404 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2405 if (cur->src.port[0] != prev->src.port[0] || 2406 cur->src.port[1] != prev->src.port[1] || 2407 cur->src.port_op != prev->src.port_op) 2408 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2409 if (cur->dst.neg != prev->dst.neg || 2410 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2411 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2412 if (cur->dst.port[0] != prev->dst.port[0] || 2413 cur->dst.port[1] != prev->dst.port[1] || 2414 cur->dst.port_op != prev->dst.port_op) 2415 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2416 2417 prev = cur; 2418 cur = TAILQ_NEXT(cur, entries); 2419 } 2420 for (i = 0; i < PF_SKIP_COUNT; ++i) 2421 PF_SET_SKIP_STEPS(i); 2422 } 2423 2424 static int 2425 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2426 { 2427 if (aw1->type != aw2->type) 2428 return (1); 2429 switch (aw1->type) { 2430 case PF_ADDR_ADDRMASK: 2431 case PF_ADDR_RANGE: 2432 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2433 return (1); 2434 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2435 return (1); 2436 return (0); 2437 case PF_ADDR_DYNIFTL: 2438 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2439 case PF_ADDR_NOROUTE: 2440 case PF_ADDR_URPFFAILED: 2441 return (0); 2442 case PF_ADDR_TABLE: 2443 return (aw1->p.tbl != aw2->p.tbl); 2444 default: 2445 printf("invalid address type: %d\n", aw1->type); 2446 return (1); 2447 } 2448 } 2449 2450 /** 2451 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2452 * header isn't always a full checksum. In some cases (i.e. output) it's a 2453 * pseudo-header checksum, which is a partial checksum over src/dst IP 2454 * addresses, protocol number and length. 2455 * 2456 * That means we have the following cases: 2457 * * Input or forwarding: we don't have TSO, the checksum fields are full 2458 * checksums, we need to update the checksum whenever we change anything. 2459 * * Output (i.e. the checksum is a pseudo-header checksum): 2460 * x The field being updated is src/dst address or affects the length of 2461 * the packet. We need to update the pseudo-header checksum (note that this 2462 * checksum is not ones' complement). 2463 * x Some other field is being modified (e.g. src/dst port numbers): We 2464 * don't have to update anything. 2465 **/ 2466 u_int16_t 2467 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2468 { 2469 u_int32_t x; 2470 2471 x = cksum + old - new; 2472 x = (x + (x >> 16)) & 0xffff; 2473 2474 /* optimise: eliminate a branch when not udp */ 2475 if (udp && cksum == 0x0000) 2476 return cksum; 2477 if (udp && x == 0x0000) 2478 x = 0xffff; 2479 2480 return (u_int16_t)(x); 2481 } 2482 2483 static void 2484 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2485 u_int8_t udp) 2486 { 2487 u_int16_t old = htons(hi ? (*f << 8) : *f); 2488 u_int16_t new = htons(hi ? ( v << 8) : v); 2489 2490 if (*f == v) 2491 return; 2492 2493 *f = v; 2494 2495 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2496 return; 2497 2498 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2499 } 2500 2501 void 2502 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2503 bool hi, u_int8_t udp) 2504 { 2505 u_int8_t *fb = (u_int8_t *)f; 2506 u_int8_t *vb = (u_int8_t *)&v; 2507 2508 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2509 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2510 } 2511 2512 void 2513 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2514 bool hi, u_int8_t udp) 2515 { 2516 u_int8_t *fb = (u_int8_t *)f; 2517 u_int8_t *vb = (u_int8_t *)&v; 2518 2519 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2520 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2521 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2522 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2523 } 2524 2525 u_int16_t 2526 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2527 u_int16_t new, u_int8_t udp) 2528 { 2529 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2530 return (cksum); 2531 2532 return (pf_cksum_fixup(cksum, old, new, udp)); 2533 } 2534 2535 static void 2536 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2537 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2538 sa_family_t af) 2539 { 2540 struct pf_addr ao; 2541 u_int16_t po = *p; 2542 2543 PF_ACPY(&ao, a, af); 2544 PF_ACPY(a, an, af); 2545 2546 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2547 *pc = ~*pc; 2548 2549 *p = pn; 2550 2551 switch (af) { 2552 #ifdef INET 2553 case AF_INET: 2554 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2555 ao.addr16[0], an->addr16[0], 0), 2556 ao.addr16[1], an->addr16[1], 0); 2557 *p = pn; 2558 2559 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2560 ao.addr16[0], an->addr16[0], u), 2561 ao.addr16[1], an->addr16[1], u); 2562 2563 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2564 break; 2565 #endif /* INET */ 2566 #ifdef INET6 2567 case AF_INET6: 2568 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2569 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2570 pf_cksum_fixup(pf_cksum_fixup(*pc, 2571 ao.addr16[0], an->addr16[0], u), 2572 ao.addr16[1], an->addr16[1], u), 2573 ao.addr16[2], an->addr16[2], u), 2574 ao.addr16[3], an->addr16[3], u), 2575 ao.addr16[4], an->addr16[4], u), 2576 ao.addr16[5], an->addr16[5], u), 2577 ao.addr16[6], an->addr16[6], u), 2578 ao.addr16[7], an->addr16[7], u); 2579 2580 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2581 break; 2582 #endif /* INET6 */ 2583 } 2584 2585 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2586 CSUM_DELAY_DATA_IPV6)) { 2587 *pc = ~*pc; 2588 if (! *pc) 2589 *pc = 0xffff; 2590 } 2591 } 2592 2593 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2594 void 2595 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2596 { 2597 u_int32_t ao; 2598 2599 memcpy(&ao, a, sizeof(ao)); 2600 memcpy(a, &an, sizeof(u_int32_t)); 2601 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2602 ao % 65536, an % 65536, u); 2603 } 2604 2605 void 2606 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2607 { 2608 u_int32_t ao; 2609 2610 memcpy(&ao, a, sizeof(ao)); 2611 memcpy(a, &an, sizeof(u_int32_t)); 2612 2613 *c = pf_proto_cksum_fixup(m, 2614 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2615 ao % 65536, an % 65536, udp); 2616 } 2617 2618 #ifdef INET6 2619 static void 2620 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2621 { 2622 struct pf_addr ao; 2623 2624 PF_ACPY(&ao, a, AF_INET6); 2625 PF_ACPY(a, an, AF_INET6); 2626 2627 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2628 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2629 pf_cksum_fixup(pf_cksum_fixup(*c, 2630 ao.addr16[0], an->addr16[0], u), 2631 ao.addr16[1], an->addr16[1], u), 2632 ao.addr16[2], an->addr16[2], u), 2633 ao.addr16[3], an->addr16[3], u), 2634 ao.addr16[4], an->addr16[4], u), 2635 ao.addr16[5], an->addr16[5], u), 2636 ao.addr16[6], an->addr16[6], u), 2637 ao.addr16[7], an->addr16[7], u); 2638 } 2639 #endif /* INET6 */ 2640 2641 static void 2642 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2643 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2644 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2645 { 2646 struct pf_addr oia, ooa; 2647 2648 PF_ACPY(&oia, ia, af); 2649 if (oa) 2650 PF_ACPY(&ooa, oa, af); 2651 2652 /* Change inner protocol port, fix inner protocol checksum. */ 2653 if (ip != NULL) { 2654 u_int16_t oip = *ip; 2655 u_int32_t opc; 2656 2657 if (pc != NULL) 2658 opc = *pc; 2659 *ip = np; 2660 if (pc != NULL) 2661 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2662 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2663 if (pc != NULL) 2664 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2665 } 2666 /* Change inner ip address, fix inner ip and icmp checksums. */ 2667 PF_ACPY(ia, na, af); 2668 switch (af) { 2669 #ifdef INET 2670 case AF_INET: { 2671 u_int32_t oh2c = *h2c; 2672 2673 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2674 oia.addr16[0], ia->addr16[0], 0), 2675 oia.addr16[1], ia->addr16[1], 0); 2676 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2677 oia.addr16[0], ia->addr16[0], 0), 2678 oia.addr16[1], ia->addr16[1], 0); 2679 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2680 break; 2681 } 2682 #endif /* INET */ 2683 #ifdef INET6 2684 case AF_INET6: 2685 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2686 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2687 pf_cksum_fixup(pf_cksum_fixup(*ic, 2688 oia.addr16[0], ia->addr16[0], u), 2689 oia.addr16[1], ia->addr16[1], u), 2690 oia.addr16[2], ia->addr16[2], u), 2691 oia.addr16[3], ia->addr16[3], u), 2692 oia.addr16[4], ia->addr16[4], u), 2693 oia.addr16[5], ia->addr16[5], u), 2694 oia.addr16[6], ia->addr16[6], u), 2695 oia.addr16[7], ia->addr16[7], u); 2696 break; 2697 #endif /* INET6 */ 2698 } 2699 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2700 if (oa) { 2701 PF_ACPY(oa, na, af); 2702 switch (af) { 2703 #ifdef INET 2704 case AF_INET: 2705 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2706 ooa.addr16[0], oa->addr16[0], 0), 2707 ooa.addr16[1], oa->addr16[1], 0); 2708 break; 2709 #endif /* INET */ 2710 #ifdef INET6 2711 case AF_INET6: 2712 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2713 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2714 pf_cksum_fixup(pf_cksum_fixup(*ic, 2715 ooa.addr16[0], oa->addr16[0], u), 2716 ooa.addr16[1], oa->addr16[1], u), 2717 ooa.addr16[2], oa->addr16[2], u), 2718 ooa.addr16[3], oa->addr16[3], u), 2719 ooa.addr16[4], oa->addr16[4], u), 2720 ooa.addr16[5], oa->addr16[5], u), 2721 ooa.addr16[6], oa->addr16[6], u), 2722 ooa.addr16[7], oa->addr16[7], u); 2723 break; 2724 #endif /* INET6 */ 2725 } 2726 } 2727 } 2728 2729 /* 2730 * Need to modulate the sequence numbers in the TCP SACK option 2731 * (credits to Krzysztof Pfaff for report and patch) 2732 */ 2733 static int 2734 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2735 struct tcphdr *th, struct pf_state_peer *dst) 2736 { 2737 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2738 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2739 int copyback = 0, i, olen; 2740 struct sackblk sack; 2741 2742 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2743 if (hlen < TCPOLEN_SACKLEN || 2744 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2745 return 0; 2746 2747 while (hlen >= TCPOLEN_SACKLEN) { 2748 size_t startoff = opt - opts; 2749 olen = opt[1]; 2750 switch (*opt) { 2751 case TCPOPT_EOL: /* FALLTHROUGH */ 2752 case TCPOPT_NOP: 2753 opt++; 2754 hlen--; 2755 break; 2756 case TCPOPT_SACK: 2757 if (olen > hlen) 2758 olen = hlen; 2759 if (olen >= TCPOLEN_SACKLEN) { 2760 for (i = 2; i + TCPOLEN_SACK <= olen; 2761 i += TCPOLEN_SACK) { 2762 memcpy(&sack, &opt[i], sizeof(sack)); 2763 pf_patch_32_unaligned(m, 2764 &th->th_sum, &sack.start, 2765 htonl(ntohl(sack.start) - dst->seqdiff), 2766 PF_ALGNMNT(startoff), 2767 0); 2768 pf_patch_32_unaligned(m, &th->th_sum, 2769 &sack.end, 2770 htonl(ntohl(sack.end) - dst->seqdiff), 2771 PF_ALGNMNT(startoff), 2772 0); 2773 memcpy(&opt[i], &sack, sizeof(sack)); 2774 } 2775 copyback = 1; 2776 } 2777 /* FALLTHROUGH */ 2778 default: 2779 if (olen < 2) 2780 olen = 2; 2781 hlen -= olen; 2782 opt += olen; 2783 } 2784 } 2785 2786 if (copyback) 2787 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2788 return (copyback); 2789 } 2790 2791 struct mbuf * 2792 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2793 const struct pf_addr *saddr, const struct pf_addr *daddr, 2794 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2795 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2796 u_int16_t rtag, int rtableid) 2797 { 2798 struct mbuf *m; 2799 int len, tlen; 2800 #ifdef INET 2801 struct ip *h = NULL; 2802 #endif /* INET */ 2803 #ifdef INET6 2804 struct ip6_hdr *h6 = NULL; 2805 #endif /* INET6 */ 2806 struct tcphdr *th; 2807 char *opt; 2808 struct pf_mtag *pf_mtag; 2809 2810 len = 0; 2811 th = NULL; 2812 2813 /* maximum segment size tcp option */ 2814 tlen = sizeof(struct tcphdr); 2815 if (mss) 2816 tlen += 4; 2817 2818 switch (af) { 2819 #ifdef INET 2820 case AF_INET: 2821 len = sizeof(struct ip) + tlen; 2822 break; 2823 #endif /* INET */ 2824 #ifdef INET6 2825 case AF_INET6: 2826 len = sizeof(struct ip6_hdr) + tlen; 2827 break; 2828 #endif /* INET6 */ 2829 default: 2830 panic("%s: unsupported af %d", __func__, af); 2831 } 2832 2833 m = m_gethdr(M_NOWAIT, MT_DATA); 2834 if (m == NULL) 2835 return (NULL); 2836 2837 #ifdef MAC 2838 mac_netinet_firewall_send(m); 2839 #endif 2840 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2841 m_freem(m); 2842 return (NULL); 2843 } 2844 if (tag) 2845 m->m_flags |= M_SKIP_FIREWALL; 2846 pf_mtag->tag = rtag; 2847 2848 if (rtableid >= 0) 2849 M_SETFIB(m, rtableid); 2850 2851 #ifdef ALTQ 2852 if (r != NULL && r->qid) { 2853 pf_mtag->qid = r->qid; 2854 2855 /* add hints for ecn */ 2856 pf_mtag->hdr = mtod(m, struct ip *); 2857 } 2858 #endif /* ALTQ */ 2859 m->m_data += max_linkhdr; 2860 m->m_pkthdr.len = m->m_len = len; 2861 /* The rest of the stack assumes a rcvif, so provide one. 2862 * This is a locally generated packet, so .. close enough. */ 2863 m->m_pkthdr.rcvif = V_loif; 2864 bzero(m->m_data, len); 2865 switch (af) { 2866 #ifdef INET 2867 case AF_INET: 2868 h = mtod(m, struct ip *); 2869 2870 /* IP header fields included in the TCP checksum */ 2871 h->ip_p = IPPROTO_TCP; 2872 h->ip_len = htons(tlen); 2873 h->ip_src.s_addr = saddr->v4.s_addr; 2874 h->ip_dst.s_addr = daddr->v4.s_addr; 2875 2876 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2877 break; 2878 #endif /* INET */ 2879 #ifdef INET6 2880 case AF_INET6: 2881 h6 = mtod(m, struct ip6_hdr *); 2882 2883 /* IP header fields included in the TCP checksum */ 2884 h6->ip6_nxt = IPPROTO_TCP; 2885 h6->ip6_plen = htons(tlen); 2886 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2887 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2888 2889 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2890 break; 2891 #endif /* INET6 */ 2892 } 2893 2894 /* TCP header */ 2895 th->th_sport = sport; 2896 th->th_dport = dport; 2897 th->th_seq = htonl(seq); 2898 th->th_ack = htonl(ack); 2899 th->th_off = tlen >> 2; 2900 th->th_flags = flags; 2901 th->th_win = htons(win); 2902 2903 if (mss) { 2904 opt = (char *)(th + 1); 2905 opt[0] = TCPOPT_MAXSEG; 2906 opt[1] = 4; 2907 HTONS(mss); 2908 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2909 } 2910 2911 switch (af) { 2912 #ifdef INET 2913 case AF_INET: 2914 /* TCP checksum */ 2915 th->th_sum = in_cksum(m, len); 2916 2917 /* Finish the IP header */ 2918 h->ip_v = 4; 2919 h->ip_hl = sizeof(*h) >> 2; 2920 h->ip_tos = IPTOS_LOWDELAY; 2921 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2922 h->ip_len = htons(len); 2923 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2924 h->ip_sum = 0; 2925 break; 2926 #endif /* INET */ 2927 #ifdef INET6 2928 case AF_INET6: 2929 /* TCP checksum */ 2930 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2931 sizeof(struct ip6_hdr), tlen); 2932 2933 h6->ip6_vfc |= IPV6_VERSION; 2934 h6->ip6_hlim = IPV6_DEFHLIM; 2935 break; 2936 #endif /* INET6 */ 2937 } 2938 2939 return (m); 2940 } 2941 2942 void 2943 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 2944 const struct pf_addr *saddr, const struct pf_addr *daddr, 2945 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2946 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2947 u_int16_t rtag, int rtableid) 2948 { 2949 struct pf_send_entry *pfse; 2950 struct mbuf *m; 2951 2952 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, flags, 2953 win, mss, ttl, tag, rtag, rtableid); 2954 if (m == NULL) 2955 return; 2956 2957 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2958 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2959 if (pfse == NULL) { 2960 m_freem(m); 2961 return; 2962 } 2963 2964 switch (af) { 2965 #ifdef INET 2966 case AF_INET: 2967 pfse->pfse_type = PFSE_IP; 2968 break; 2969 #endif /* INET */ 2970 #ifdef INET6 2971 case AF_INET6: 2972 pfse->pfse_type = PFSE_IP6; 2973 break; 2974 #endif /* INET6 */ 2975 } 2976 2977 pfse->pfse_m = m; 2978 pf_send(pfse); 2979 } 2980 2981 static void 2982 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 2983 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 2984 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 2985 u_short *reason, int rtableid) 2986 { 2987 struct pf_addr * const saddr = pd->src; 2988 struct pf_addr * const daddr = pd->dst; 2989 sa_family_t af = pd->af; 2990 2991 /* undo NAT changes, if they have taken place */ 2992 if (nr != NULL) { 2993 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 2994 PF_ACPY(daddr, &sk->addr[pd->didx], af); 2995 if (pd->sport) 2996 *pd->sport = sk->port[pd->sidx]; 2997 if (pd->dport) 2998 *pd->dport = sk->port[pd->didx]; 2999 if (pd->proto_sum) 3000 *pd->proto_sum = bproto_sum; 3001 if (pd->ip_sum) 3002 *pd->ip_sum = bip_sum; 3003 m_copyback(m, off, hdrlen, pd->hdr.any); 3004 } 3005 if (pd->proto == IPPROTO_TCP && 3006 ((r->rule_flag & PFRULE_RETURNRST) || 3007 (r->rule_flag & PFRULE_RETURN)) && 3008 !(th->th_flags & TH_RST)) { 3009 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3010 int len = 0; 3011 #ifdef INET 3012 struct ip *h4; 3013 #endif 3014 #ifdef INET6 3015 struct ip6_hdr *h6; 3016 #endif 3017 3018 switch (af) { 3019 #ifdef INET 3020 case AF_INET: 3021 h4 = mtod(m, struct ip *); 3022 len = ntohs(h4->ip_len) - off; 3023 break; 3024 #endif 3025 #ifdef INET6 3026 case AF_INET6: 3027 h6 = mtod(m, struct ip6_hdr *); 3028 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3029 break; 3030 #endif 3031 } 3032 3033 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3034 REASON_SET(reason, PFRES_PROTCKSUM); 3035 else { 3036 if (th->th_flags & TH_SYN) 3037 ack++; 3038 if (th->th_flags & TH_FIN) 3039 ack++; 3040 pf_send_tcp(r, af, pd->dst, 3041 pd->src, th->th_dport, th->th_sport, 3042 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3043 r->return_ttl, 1, 0, rtableid); 3044 } 3045 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3046 r->return_icmp) 3047 pf_send_icmp(m, r->return_icmp >> 8, 3048 r->return_icmp & 255, af, r, rtableid); 3049 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3050 r->return_icmp6) 3051 pf_send_icmp(m, r->return_icmp6 >> 8, 3052 r->return_icmp6 & 255, af, r, rtableid); 3053 } 3054 3055 static int 3056 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3057 { 3058 struct m_tag *mtag; 3059 u_int8_t mpcp; 3060 3061 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3062 if (mtag == NULL) 3063 return (0); 3064 3065 if (prio == PF_PRIO_ZERO) 3066 prio = 0; 3067 3068 mpcp = *(uint8_t *)(mtag + 1); 3069 3070 return (mpcp == prio); 3071 } 3072 3073 static int 3074 pf_icmp_to_bandlim(uint8_t type) 3075 { 3076 switch (type) { 3077 case ICMP_ECHO: 3078 case ICMP_ECHOREPLY: 3079 return (BANDLIM_ICMP_ECHO); 3080 case ICMP_TSTAMP: 3081 case ICMP_TSTAMPREPLY: 3082 return (BANDLIM_ICMP_TSTAMP); 3083 case ICMP_UNREACH: 3084 default: 3085 return (BANDLIM_ICMP_UNREACH); 3086 } 3087 } 3088 3089 static void 3090 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3091 struct pf_krule *r, int rtableid) 3092 { 3093 struct pf_send_entry *pfse; 3094 struct mbuf *m0; 3095 struct pf_mtag *pf_mtag; 3096 3097 /* ICMP packet rate limitation. */ 3098 #ifdef INET6 3099 if (af == AF_INET6) { 3100 if (icmp6_ratelimit(NULL, type, code)) 3101 return; 3102 } 3103 #endif 3104 #ifdef INET 3105 if (af == AF_INET) { 3106 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3107 return; 3108 } 3109 #endif 3110 3111 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3112 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3113 if (pfse == NULL) 3114 return; 3115 3116 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3117 free(pfse, M_PFTEMP); 3118 return; 3119 } 3120 3121 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3122 free(pfse, M_PFTEMP); 3123 return; 3124 } 3125 /* XXX: revisit */ 3126 m0->m_flags |= M_SKIP_FIREWALL; 3127 3128 if (rtableid >= 0) 3129 M_SETFIB(m0, rtableid); 3130 3131 #ifdef ALTQ 3132 if (r->qid) { 3133 pf_mtag->qid = r->qid; 3134 /* add hints for ecn */ 3135 pf_mtag->hdr = mtod(m0, struct ip *); 3136 } 3137 #endif /* ALTQ */ 3138 3139 switch (af) { 3140 #ifdef INET 3141 case AF_INET: 3142 pfse->pfse_type = PFSE_ICMP; 3143 break; 3144 #endif /* INET */ 3145 #ifdef INET6 3146 case AF_INET6: 3147 pfse->pfse_type = PFSE_ICMP6; 3148 break; 3149 #endif /* INET6 */ 3150 } 3151 pfse->pfse_m = m0; 3152 pfse->icmpopts.type = type; 3153 pfse->icmpopts.code = code; 3154 pf_send(pfse); 3155 } 3156 3157 /* 3158 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3159 * If n is 0, they match if they are equal. If n is != 0, they match if they 3160 * are different. 3161 */ 3162 int 3163 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3164 struct pf_addr *b, sa_family_t af) 3165 { 3166 int match = 0; 3167 3168 switch (af) { 3169 #ifdef INET 3170 case AF_INET: 3171 if ((a->addr32[0] & m->addr32[0]) == 3172 (b->addr32[0] & m->addr32[0])) 3173 match++; 3174 break; 3175 #endif /* INET */ 3176 #ifdef INET6 3177 case AF_INET6: 3178 if (((a->addr32[0] & m->addr32[0]) == 3179 (b->addr32[0] & m->addr32[0])) && 3180 ((a->addr32[1] & m->addr32[1]) == 3181 (b->addr32[1] & m->addr32[1])) && 3182 ((a->addr32[2] & m->addr32[2]) == 3183 (b->addr32[2] & m->addr32[2])) && 3184 ((a->addr32[3] & m->addr32[3]) == 3185 (b->addr32[3] & m->addr32[3]))) 3186 match++; 3187 break; 3188 #endif /* INET6 */ 3189 } 3190 if (match) { 3191 if (n) 3192 return (0); 3193 else 3194 return (1); 3195 } else { 3196 if (n) 3197 return (1); 3198 else 3199 return (0); 3200 } 3201 } 3202 3203 /* 3204 * Return 1 if b <= a <= e, otherwise return 0. 3205 */ 3206 int 3207 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3208 struct pf_addr *a, sa_family_t af) 3209 { 3210 switch (af) { 3211 #ifdef INET 3212 case AF_INET: 3213 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3214 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3215 return (0); 3216 break; 3217 #endif /* INET */ 3218 #ifdef INET6 3219 case AF_INET6: { 3220 int i; 3221 3222 /* check a >= b */ 3223 for (i = 0; i < 4; ++i) 3224 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3225 break; 3226 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3227 return (0); 3228 /* check a <= e */ 3229 for (i = 0; i < 4; ++i) 3230 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3231 break; 3232 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3233 return (0); 3234 break; 3235 } 3236 #endif /* INET6 */ 3237 } 3238 return (1); 3239 } 3240 3241 static int 3242 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3243 { 3244 switch (op) { 3245 case PF_OP_IRG: 3246 return ((p > a1) && (p < a2)); 3247 case PF_OP_XRG: 3248 return ((p < a1) || (p > a2)); 3249 case PF_OP_RRG: 3250 return ((p >= a1) && (p <= a2)); 3251 case PF_OP_EQ: 3252 return (p == a1); 3253 case PF_OP_NE: 3254 return (p != a1); 3255 case PF_OP_LT: 3256 return (p < a1); 3257 case PF_OP_LE: 3258 return (p <= a1); 3259 case PF_OP_GT: 3260 return (p > a1); 3261 case PF_OP_GE: 3262 return (p >= a1); 3263 } 3264 return (0); /* never reached */ 3265 } 3266 3267 int 3268 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3269 { 3270 NTOHS(a1); 3271 NTOHS(a2); 3272 NTOHS(p); 3273 return (pf_match(op, a1, a2, p)); 3274 } 3275 3276 static int 3277 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3278 { 3279 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3280 return (0); 3281 return (pf_match(op, a1, a2, u)); 3282 } 3283 3284 static int 3285 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3286 { 3287 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3288 return (0); 3289 return (pf_match(op, a1, a2, g)); 3290 } 3291 3292 int 3293 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3294 { 3295 if (*tag == -1) 3296 *tag = mtag; 3297 3298 return ((!r->match_tag_not && r->match_tag == *tag) || 3299 (r->match_tag_not && r->match_tag != *tag)); 3300 } 3301 3302 int 3303 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3304 { 3305 3306 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3307 3308 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3309 return (ENOMEM); 3310 3311 pd->pf_mtag->tag = tag; 3312 3313 return (0); 3314 } 3315 3316 #define PF_ANCHOR_STACKSIZE 32 3317 struct pf_kanchor_stackframe { 3318 struct pf_kruleset *rs; 3319 struct pf_krule *r; /* XXX: + match bit */ 3320 struct pf_kanchor *child; 3321 }; 3322 3323 /* 3324 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3325 */ 3326 #define PF_ANCHORSTACK_MATCH 0x00000001 3327 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3328 3329 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3330 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3331 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3332 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3333 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3334 } while (0) 3335 3336 void 3337 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3338 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3339 int *match) 3340 { 3341 struct pf_kanchor_stackframe *f; 3342 3343 PF_RULES_RASSERT(); 3344 3345 if (match) 3346 *match = 0; 3347 if (*depth >= PF_ANCHOR_STACKSIZE) { 3348 printf("%s: anchor stack overflow on %s\n", 3349 __func__, (*r)->anchor->name); 3350 *r = TAILQ_NEXT(*r, entries); 3351 return; 3352 } else if (*depth == 0 && a != NULL) 3353 *a = *r; 3354 f = stack + (*depth)++; 3355 f->rs = *rs; 3356 f->r = *r; 3357 if ((*r)->anchor_wildcard) { 3358 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3359 3360 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3361 *r = NULL; 3362 return; 3363 } 3364 *rs = &f->child->ruleset; 3365 } else { 3366 f->child = NULL; 3367 *rs = &(*r)->anchor->ruleset; 3368 } 3369 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3370 } 3371 3372 int 3373 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3374 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3375 int *match) 3376 { 3377 struct pf_kanchor_stackframe *f; 3378 struct pf_krule *fr; 3379 int quick = 0; 3380 3381 PF_RULES_RASSERT(); 3382 3383 do { 3384 if (*depth <= 0) 3385 break; 3386 f = stack + *depth - 1; 3387 fr = PF_ANCHOR_RULE(f); 3388 if (f->child != NULL) { 3389 /* 3390 * This block traverses through 3391 * a wildcard anchor. 3392 */ 3393 if (match != NULL && *match) { 3394 /* 3395 * If any of "*" matched, then 3396 * "foo/ *" matched, mark frame 3397 * appropriately. 3398 */ 3399 PF_ANCHOR_SET_MATCH(f); 3400 *match = 0; 3401 } 3402 f->child = RB_NEXT(pf_kanchor_node, 3403 &fr->anchor->children, f->child); 3404 if (f->child != NULL) { 3405 *rs = &f->child->ruleset; 3406 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3407 if (*r == NULL) 3408 continue; 3409 else 3410 break; 3411 } 3412 } 3413 (*depth)--; 3414 if (*depth == 0 && a != NULL) 3415 *a = NULL; 3416 *rs = f->rs; 3417 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3418 quick = fr->quick; 3419 *r = TAILQ_NEXT(fr, entries); 3420 } while (*r == NULL); 3421 3422 return (quick); 3423 } 3424 3425 struct pf_keth_anchor_stackframe { 3426 struct pf_keth_ruleset *rs; 3427 struct pf_keth_rule *r; /* XXX: + match bit */ 3428 struct pf_keth_anchor *child; 3429 }; 3430 3431 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3432 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3433 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3434 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3435 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3436 } while (0) 3437 3438 void 3439 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3440 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3441 struct pf_keth_rule **a, int *match) 3442 { 3443 struct pf_keth_anchor_stackframe *f; 3444 3445 NET_EPOCH_ASSERT(); 3446 3447 if (match) 3448 *match = 0; 3449 if (*depth >= PF_ANCHOR_STACKSIZE) { 3450 printf("%s: anchor stack overflow on %s\n", 3451 __func__, (*r)->anchor->name); 3452 *r = TAILQ_NEXT(*r, entries); 3453 return; 3454 } else if (*depth == 0 && a != NULL) 3455 *a = *r; 3456 f = stack + (*depth)++; 3457 f->rs = *rs; 3458 f->r = *r; 3459 if ((*r)->anchor_wildcard) { 3460 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3461 3462 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3463 *r = NULL; 3464 return; 3465 } 3466 *rs = &f->child->ruleset; 3467 } else { 3468 f->child = NULL; 3469 *rs = &(*r)->anchor->ruleset; 3470 } 3471 *r = TAILQ_FIRST((*rs)->active.rules); 3472 } 3473 3474 int 3475 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3476 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3477 struct pf_keth_rule **a, int *match) 3478 { 3479 struct pf_keth_anchor_stackframe *f; 3480 struct pf_keth_rule *fr; 3481 int quick = 0; 3482 3483 NET_EPOCH_ASSERT(); 3484 3485 do { 3486 if (*depth <= 0) 3487 break; 3488 f = stack + *depth - 1; 3489 fr = PF_ETH_ANCHOR_RULE(f); 3490 if (f->child != NULL) { 3491 /* 3492 * This block traverses through 3493 * a wildcard anchor. 3494 */ 3495 if (match != NULL && *match) { 3496 /* 3497 * If any of "*" matched, then 3498 * "foo/ *" matched, mark frame 3499 * appropriately. 3500 */ 3501 PF_ETH_ANCHOR_SET_MATCH(f); 3502 *match = 0; 3503 } 3504 f->child = RB_NEXT(pf_keth_anchor_node, 3505 &fr->anchor->children, f->child); 3506 if (f->child != NULL) { 3507 *rs = &f->child->ruleset; 3508 *r = TAILQ_FIRST((*rs)->active.rules); 3509 if (*r == NULL) 3510 continue; 3511 else 3512 break; 3513 } 3514 } 3515 (*depth)--; 3516 if (*depth == 0 && a != NULL) 3517 *a = NULL; 3518 *rs = f->rs; 3519 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3520 quick = fr->quick; 3521 *r = TAILQ_NEXT(fr, entries); 3522 } while (*r == NULL); 3523 3524 return (quick); 3525 } 3526 3527 #ifdef INET6 3528 void 3529 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3530 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3531 { 3532 switch (af) { 3533 #ifdef INET 3534 case AF_INET: 3535 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3536 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3537 break; 3538 #endif /* INET */ 3539 case AF_INET6: 3540 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3541 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3542 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3543 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3544 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3545 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3546 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3547 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3548 break; 3549 } 3550 } 3551 3552 void 3553 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3554 { 3555 switch (af) { 3556 #ifdef INET 3557 case AF_INET: 3558 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3559 break; 3560 #endif /* INET */ 3561 case AF_INET6: 3562 if (addr->addr32[3] == 0xffffffff) { 3563 addr->addr32[3] = 0; 3564 if (addr->addr32[2] == 0xffffffff) { 3565 addr->addr32[2] = 0; 3566 if (addr->addr32[1] == 0xffffffff) { 3567 addr->addr32[1] = 0; 3568 addr->addr32[0] = 3569 htonl(ntohl(addr->addr32[0]) + 1); 3570 } else 3571 addr->addr32[1] = 3572 htonl(ntohl(addr->addr32[1]) + 1); 3573 } else 3574 addr->addr32[2] = 3575 htonl(ntohl(addr->addr32[2]) + 1); 3576 } else 3577 addr->addr32[3] = 3578 htonl(ntohl(addr->addr32[3]) + 1); 3579 break; 3580 } 3581 } 3582 #endif /* INET6 */ 3583 3584 void 3585 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3586 { 3587 if (r->qid) 3588 a->qid = r->qid; 3589 if (r->pqid) 3590 a->pqid = r->pqid; 3591 if (r->rtableid >= 0) 3592 a->rtableid = r->rtableid; 3593 a->log |= r->log; 3594 if (r->scrub_flags & PFSTATE_SETTOS) 3595 a->set_tos = r->set_tos; 3596 if (r->min_ttl) 3597 a->min_ttl = r->min_ttl; 3598 if (r->max_mss) 3599 a->max_mss = r->max_mss; 3600 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3601 PFSTATE_SETTOS|PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3602 if (r->dnpipe) 3603 a->dnpipe = r->dnpipe; 3604 if (r->dnrpipe) 3605 a->dnrpipe = r->dnrpipe; 3606 if (r->dnpipe || r->dnrpipe) { 3607 if (r->free_flags & PFRULE_DN_IS_PIPE) 3608 a->flags |= PFSTATE_DN_IS_PIPE; 3609 else 3610 a->flags &= ~PFSTATE_DN_IS_PIPE; 3611 } 3612 } 3613 3614 int 3615 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 3616 { 3617 struct pf_addr *saddr, *daddr; 3618 u_int16_t sport, dport; 3619 struct inpcbinfo *pi; 3620 struct inpcb *inp; 3621 3622 pd->lookup.uid = UID_MAX; 3623 pd->lookup.gid = GID_MAX; 3624 3625 switch (pd->proto) { 3626 case IPPROTO_TCP: 3627 sport = pd->hdr.tcp.th_sport; 3628 dport = pd->hdr.tcp.th_dport; 3629 pi = &V_tcbinfo; 3630 break; 3631 case IPPROTO_UDP: 3632 sport = pd->hdr.udp.uh_sport; 3633 dport = pd->hdr.udp.uh_dport; 3634 pi = &V_udbinfo; 3635 break; 3636 default: 3637 return (-1); 3638 } 3639 if (direction == PF_IN) { 3640 saddr = pd->src; 3641 daddr = pd->dst; 3642 } else { 3643 u_int16_t p; 3644 3645 p = sport; 3646 sport = dport; 3647 dport = p; 3648 saddr = pd->dst; 3649 daddr = pd->src; 3650 } 3651 switch (pd->af) { 3652 #ifdef INET 3653 case AF_INET: 3654 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3655 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3656 if (inp == NULL) { 3657 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3658 daddr->v4, dport, INPLOOKUP_WILDCARD | 3659 INPLOOKUP_RLOCKPCB, NULL, m); 3660 if (inp == NULL) 3661 return (-1); 3662 } 3663 break; 3664 #endif /* INET */ 3665 #ifdef INET6 3666 case AF_INET6: 3667 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3668 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3669 if (inp == NULL) { 3670 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3671 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3672 INPLOOKUP_RLOCKPCB, NULL, m); 3673 if (inp == NULL) 3674 return (-1); 3675 } 3676 break; 3677 #endif /* INET6 */ 3678 3679 default: 3680 return (-1); 3681 } 3682 INP_RLOCK_ASSERT(inp); 3683 pd->lookup.uid = inp->inp_cred->cr_uid; 3684 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3685 INP_RUNLOCK(inp); 3686 3687 return (1); 3688 } 3689 3690 u_int8_t 3691 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3692 { 3693 int hlen; 3694 u_int8_t hdr[60]; 3695 u_int8_t *opt, optlen; 3696 u_int8_t wscale = 0; 3697 3698 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3699 if (hlen <= sizeof(struct tcphdr)) 3700 return (0); 3701 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3702 return (0); 3703 opt = hdr + sizeof(struct tcphdr); 3704 hlen -= sizeof(struct tcphdr); 3705 while (hlen >= 3) { 3706 switch (*opt) { 3707 case TCPOPT_EOL: 3708 case TCPOPT_NOP: 3709 ++opt; 3710 --hlen; 3711 break; 3712 case TCPOPT_WINDOW: 3713 wscale = opt[2]; 3714 if (wscale > TCP_MAX_WINSHIFT) 3715 wscale = TCP_MAX_WINSHIFT; 3716 wscale |= PF_WSCALE_FLAG; 3717 /* FALLTHROUGH */ 3718 default: 3719 optlen = opt[1]; 3720 if (optlen < 2) 3721 optlen = 2; 3722 hlen -= optlen; 3723 opt += optlen; 3724 break; 3725 } 3726 } 3727 return (wscale); 3728 } 3729 3730 u_int16_t 3731 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3732 { 3733 int hlen; 3734 u_int8_t hdr[60]; 3735 u_int8_t *opt, optlen; 3736 u_int16_t mss = V_tcp_mssdflt; 3737 3738 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3739 if (hlen <= sizeof(struct tcphdr)) 3740 return (0); 3741 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3742 return (0); 3743 opt = hdr + sizeof(struct tcphdr); 3744 hlen -= sizeof(struct tcphdr); 3745 while (hlen >= TCPOLEN_MAXSEG) { 3746 switch (*opt) { 3747 case TCPOPT_EOL: 3748 case TCPOPT_NOP: 3749 ++opt; 3750 --hlen; 3751 break; 3752 case TCPOPT_MAXSEG: 3753 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3754 NTOHS(mss); 3755 /* FALLTHROUGH */ 3756 default: 3757 optlen = opt[1]; 3758 if (optlen < 2) 3759 optlen = 2; 3760 hlen -= optlen; 3761 opt += optlen; 3762 break; 3763 } 3764 } 3765 return (mss); 3766 } 3767 3768 static u_int16_t 3769 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3770 { 3771 struct nhop_object *nh; 3772 #ifdef INET6 3773 struct in6_addr dst6; 3774 uint32_t scopeid; 3775 #endif /* INET6 */ 3776 int hlen = 0; 3777 uint16_t mss = 0; 3778 3779 NET_EPOCH_ASSERT(); 3780 3781 switch (af) { 3782 #ifdef INET 3783 case AF_INET: 3784 hlen = sizeof(struct ip); 3785 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3786 if (nh != NULL) 3787 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3788 break; 3789 #endif /* INET */ 3790 #ifdef INET6 3791 case AF_INET6: 3792 hlen = sizeof(struct ip6_hdr); 3793 in6_splitscope(&addr->v6, &dst6, &scopeid); 3794 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 3795 if (nh != NULL) 3796 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3797 break; 3798 #endif /* INET6 */ 3799 } 3800 3801 mss = max(V_tcp_mssdflt, mss); 3802 mss = min(mss, offer); 3803 mss = max(mss, 64); /* sanity - at least max opt space */ 3804 return (mss); 3805 } 3806 3807 static u_int32_t 3808 pf_tcp_iss(struct pf_pdesc *pd) 3809 { 3810 MD5_CTX ctx; 3811 u_int32_t digest[4]; 3812 3813 if (V_pf_tcp_secret_init == 0) { 3814 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3815 MD5Init(&V_pf_tcp_secret_ctx); 3816 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3817 sizeof(V_pf_tcp_secret)); 3818 V_pf_tcp_secret_init = 1; 3819 } 3820 3821 ctx = V_pf_tcp_secret_ctx; 3822 3823 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 3824 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 3825 if (pd->af == AF_INET6) { 3826 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3827 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3828 } else { 3829 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3830 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3831 } 3832 MD5Final((u_char *)digest, &ctx); 3833 V_pf_tcp_iss_off += 4096; 3834 #define ISN_RANDOM_INCREMENT (4096 - 1) 3835 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3836 V_pf_tcp_iss_off); 3837 #undef ISN_RANDOM_INCREMENT 3838 } 3839 3840 static bool 3841 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 3842 { 3843 bool match = true; 3844 3845 /* Always matches if not set */ 3846 if (! r->isset) 3847 return (!r->neg); 3848 3849 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 3850 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 3851 match = false; 3852 break; 3853 } 3854 } 3855 3856 return (match ^ r->neg); 3857 } 3858 3859 static int 3860 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 3861 { 3862 if (*tag == -1) 3863 *tag = mtag; 3864 3865 return ((!r->match_tag_not && r->match_tag == *tag) || 3866 (r->match_tag_not && r->match_tag != *tag)); 3867 } 3868 3869 static void 3870 pf_bridge_to(struct pfi_kkif *kif, struct mbuf *m) 3871 { 3872 struct ifnet *ifp = kif->pfik_ifp; 3873 3874 /* If we don't have the interface drop the packet. */ 3875 if (ifp == NULL) { 3876 m_freem(m); 3877 return; 3878 } 3879 3880 switch (ifp->if_type) { 3881 case IFT_ETHER: 3882 case IFT_XETHER: 3883 case IFT_L2VLAN: 3884 case IFT_BRIDGE: 3885 case IFT_IEEE8023ADLAG: 3886 break; 3887 default: 3888 m_freem(m); 3889 return; 3890 } 3891 3892 kif->pfik_ifp->if_transmit(kif->pfik_ifp, m); 3893 } 3894 3895 static int 3896 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 3897 { 3898 #ifdef INET 3899 struct ip ip; 3900 #endif 3901 #ifdef INET6 3902 struct ip6_hdr ip6; 3903 #endif 3904 struct mbuf *m = *m0; 3905 struct ether_header *e; 3906 struct pf_keth_rule *r, *rm, *a = NULL; 3907 struct pf_keth_ruleset *ruleset = NULL; 3908 struct pf_mtag *mtag; 3909 struct pf_keth_ruleq *rules; 3910 struct pf_addr *src = NULL, *dst = NULL; 3911 sa_family_t af = 0; 3912 uint16_t proto; 3913 int asd = 0, match = 0; 3914 int tag = -1; 3915 uint8_t action; 3916 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3917 3918 MPASS(kif->pfik_ifp->if_vnet == curvnet); 3919 NET_EPOCH_ASSERT(); 3920 3921 PF_RULES_RLOCK_TRACKER; 3922 3923 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 3924 3925 mtag = pf_find_mtag(m); 3926 if (mtag != NULL && mtag->flags & PF_TAG_DUMMYNET) { 3927 /* Dummynet re-injects packets after they've 3928 * completed their delay. We've already 3929 * processed them, so pass unconditionally. */ 3930 3931 /* But only once. We may see the packet multiple times (e.g. 3932 * PFIL_IN/PFIL_OUT). */ 3933 mtag->flags &= ~PF_TAG_DUMMYNET; 3934 3935 return (PF_PASS); 3936 } 3937 3938 ruleset = V_pf_keth; 3939 rules = ck_pr_load_ptr(&ruleset->active.rules); 3940 r = TAILQ_FIRST(rules); 3941 rm = NULL; 3942 3943 e = mtod(m, struct ether_header *); 3944 proto = ntohs(e->ether_type); 3945 3946 switch (proto) { 3947 #ifdef INET 3948 case ETHERTYPE_IP: { 3949 if (m_length(m, NULL) < (sizeof(struct ether_header) + 3950 sizeof(ip))) 3951 return (PF_DROP); 3952 3953 af = AF_INET; 3954 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 3955 (caddr_t)&ip); 3956 src = (struct pf_addr *)&ip.ip_src; 3957 dst = (struct pf_addr *)&ip.ip_dst; 3958 break; 3959 } 3960 #endif /* INET */ 3961 #ifdef INET6 3962 case ETHERTYPE_IPV6: { 3963 if (m_length(m, NULL) < (sizeof(struct ether_header) + 3964 sizeof(ip6))) 3965 return (PF_DROP); 3966 3967 af = AF_INET6; 3968 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 3969 (caddr_t)&ip6); 3970 src = (struct pf_addr *)&ip6.ip6_src; 3971 dst = (struct pf_addr *)&ip6.ip6_dst; 3972 break; 3973 } 3974 #endif /* INET6 */ 3975 } 3976 3977 PF_RULES_RLOCK(); 3978 3979 while (r != NULL) { 3980 counter_u64_add(r->evaluations, 1); 3981 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 3982 3983 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 3984 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3985 "kif"); 3986 r = r->skip[PFE_SKIP_IFP].ptr; 3987 } 3988 else if (r->direction && r->direction != dir) { 3989 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3990 "dir"); 3991 r = r->skip[PFE_SKIP_DIR].ptr; 3992 } 3993 else if (r->proto && r->proto != proto) { 3994 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3995 "proto"); 3996 r = r->skip[PFE_SKIP_PROTO].ptr; 3997 } 3998 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 3999 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4000 "src"); 4001 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4002 } 4003 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4004 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4005 "dst"); 4006 r = TAILQ_NEXT(r, entries); 4007 } 4008 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4009 r->ipsrc.neg, kif, M_GETFIB(m))) { 4010 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4011 "ip_src"); 4012 r = TAILQ_NEXT(r, entries); 4013 } 4014 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4015 r->ipdst.neg, kif, M_GETFIB(m))) { 4016 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4017 "ip_dst"); 4018 r = TAILQ_NEXT(r, entries); 4019 } 4020 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4021 mtag ? mtag->tag : 0)) { 4022 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4023 "match_tag"); 4024 r = TAILQ_NEXT(r, entries); 4025 } 4026 else { 4027 if (r->tag) 4028 tag = r->tag; 4029 if (r->anchor == NULL) { 4030 /* Rule matches */ 4031 rm = r; 4032 4033 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4034 4035 if (r->quick) 4036 break; 4037 4038 r = TAILQ_NEXT(r, entries); 4039 } else { 4040 pf_step_into_keth_anchor(anchor_stack, &asd, 4041 &ruleset, &r, &a, &match); 4042 } 4043 } 4044 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4045 &ruleset, &r, &a, &match)) 4046 break; 4047 } 4048 4049 r = rm; 4050 4051 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4052 4053 /* Default to pass. */ 4054 if (r == NULL) { 4055 PF_RULES_RUNLOCK(); 4056 return (PF_PASS); 4057 } 4058 4059 /* Execute action. */ 4060 counter_u64_add(r->packets[dir == PF_OUT], 1); 4061 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4062 pf_update_timestamp(r); 4063 4064 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4065 if (r->action == PF_DROP) { 4066 PF_RULES_RUNLOCK(); 4067 return (PF_DROP); 4068 } 4069 4070 if (tag > 0) { 4071 if (mtag == NULL) 4072 mtag = pf_get_mtag(m); 4073 if (mtag == NULL) { 4074 PF_RULES_RUNLOCK(); 4075 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4076 return (PF_DROP); 4077 } 4078 mtag->tag = tag; 4079 } 4080 4081 if (r->qid != 0) { 4082 if (mtag == NULL) 4083 mtag = pf_get_mtag(m); 4084 if (mtag == NULL) { 4085 PF_RULES_RUNLOCK(); 4086 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4087 return (PF_DROP); 4088 } 4089 mtag->qid = r->qid; 4090 } 4091 4092 /* Dummynet */ 4093 if (r->dnpipe) { 4094 struct ip_fw_args dnflow; 4095 4096 /* Drop packet if dummynet is not loaded. */ 4097 if (ip_dn_io_ptr == NULL) { 4098 PF_RULES_RUNLOCK(); 4099 m_freem(m); 4100 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4101 return (PF_DROP); 4102 } 4103 if (mtag == NULL) 4104 mtag = pf_get_mtag(m); 4105 if (mtag == NULL) { 4106 PF_RULES_RUNLOCK(); 4107 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4108 return (PF_DROP); 4109 } 4110 4111 bzero(&dnflow, sizeof(dnflow)); 4112 4113 /* We don't have port numbers here, so we set 0. That means 4114 * that we'll be somewhat limited in distinguishing flows (i.e. 4115 * only based on IP addresses, not based on port numbers), but 4116 * it's better than nothing. */ 4117 dnflow.f_id.dst_port = 0; 4118 dnflow.f_id.src_port = 0; 4119 dnflow.f_id.proto = 0; 4120 4121 dnflow.rule.info = r->dnpipe; 4122 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4123 if (r->dnflags & PFRULE_DN_IS_PIPE) 4124 dnflow.rule.info |= IPFW_IS_PIPE; 4125 4126 dnflow.f_id.extra = dnflow.rule.info; 4127 4128 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4129 dnflow.flags |= IPFW_ARGS_ETHER; 4130 dnflow.ifp = kif->pfik_ifp; 4131 4132 switch (af) { 4133 case AF_INET: 4134 dnflow.f_id.addr_type = 4; 4135 dnflow.f_id.src_ip = src->v4.s_addr; 4136 dnflow.f_id.dst_ip = dst->v4.s_addr; 4137 break; 4138 case AF_INET6: 4139 dnflow.flags |= IPFW_ARGS_IP6; 4140 dnflow.f_id.addr_type = 6; 4141 dnflow.f_id.src_ip6 = src->v6; 4142 dnflow.f_id.dst_ip6 = dst->v6; 4143 break; 4144 } 4145 4146 mtag->flags |= PF_TAG_DUMMYNET; 4147 ip_dn_io_ptr(m0, &dnflow); 4148 if (*m0 != NULL) 4149 mtag->flags &= ~PF_TAG_DUMMYNET; 4150 } 4151 4152 action = r->action; 4153 4154 if (action == PF_PASS && r->bridge_to) { 4155 pf_bridge_to(r->bridge_to, *m0); 4156 *m0 = NULL; /* We've eaten the packet. */ 4157 } 4158 4159 PF_RULES_RUNLOCK(); 4160 4161 return (action); 4162 } 4163 4164 static int 4165 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, int direction, 4166 struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 4167 struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp) 4168 { 4169 struct pf_krule *nr = NULL; 4170 struct pf_addr * const saddr = pd->src; 4171 struct pf_addr * const daddr = pd->dst; 4172 sa_family_t af = pd->af; 4173 struct pf_krule *r, *a = NULL; 4174 struct pf_kruleset *ruleset = NULL; 4175 struct pf_krule_slist match_rules; 4176 struct pf_krule_item *ri; 4177 struct pf_ksrc_node *nsn = NULL; 4178 struct tcphdr *th = &pd->hdr.tcp; 4179 struct pf_state_key *sk = NULL, *nk = NULL; 4180 u_short reason; 4181 int rewrite = 0, hdrlen = 0; 4182 int tag = -1; 4183 int asd = 0; 4184 int match = 0; 4185 int state_icmp = 0; 4186 u_int16_t sport = 0, dport = 0; 4187 u_int16_t bproto_sum = 0, bip_sum = 0; 4188 u_int8_t icmptype = 0, icmpcode = 0; 4189 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4190 4191 PF_RULES_RASSERT(); 4192 4193 if (inp != NULL) { 4194 INP_LOCK_ASSERT(inp); 4195 pd->lookup.uid = inp->inp_cred->cr_uid; 4196 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4197 pd->lookup.done = 1; 4198 } 4199 4200 switch (pd->proto) { 4201 case IPPROTO_TCP: 4202 sport = th->th_sport; 4203 dport = th->th_dport; 4204 hdrlen = sizeof(*th); 4205 break; 4206 case IPPROTO_UDP: 4207 sport = pd->hdr.udp.uh_sport; 4208 dport = pd->hdr.udp.uh_dport; 4209 hdrlen = sizeof(pd->hdr.udp); 4210 break; 4211 #ifdef INET 4212 case IPPROTO_ICMP: 4213 if (pd->af != AF_INET) 4214 break; 4215 sport = dport = pd->hdr.icmp.icmp_id; 4216 hdrlen = sizeof(pd->hdr.icmp); 4217 icmptype = pd->hdr.icmp.icmp_type; 4218 icmpcode = pd->hdr.icmp.icmp_code; 4219 4220 if (icmptype == ICMP_UNREACH || 4221 icmptype == ICMP_SOURCEQUENCH || 4222 icmptype == ICMP_REDIRECT || 4223 icmptype == ICMP_TIMXCEED || 4224 icmptype == ICMP_PARAMPROB) 4225 state_icmp++; 4226 break; 4227 #endif /* INET */ 4228 #ifdef INET6 4229 case IPPROTO_ICMPV6: 4230 if (af != AF_INET6) 4231 break; 4232 sport = dport = pd->hdr.icmp6.icmp6_id; 4233 hdrlen = sizeof(pd->hdr.icmp6); 4234 icmptype = pd->hdr.icmp6.icmp6_type; 4235 icmpcode = pd->hdr.icmp6.icmp6_code; 4236 4237 if (icmptype == ICMP6_DST_UNREACH || 4238 icmptype == ICMP6_PACKET_TOO_BIG || 4239 icmptype == ICMP6_TIME_EXCEEDED || 4240 icmptype == ICMP6_PARAM_PROB) 4241 state_icmp++; 4242 break; 4243 #endif /* INET6 */ 4244 default: 4245 sport = dport = hdrlen = 0; 4246 break; 4247 } 4248 4249 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4250 4251 /* check packet for BINAT/NAT/RDR */ 4252 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 4253 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4254 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4255 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4256 4257 if (nr->log) { 4258 PFLOG_PACKET(kif, m, af, direction, PFRES_MATCH, nr, a, 4259 ruleset, pd, 1); 4260 } 4261 4262 if (pd->ip_sum) 4263 bip_sum = *pd->ip_sum; 4264 4265 switch (pd->proto) { 4266 case IPPROTO_TCP: 4267 bproto_sum = th->th_sum; 4268 pd->proto_sum = &th->th_sum; 4269 4270 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4271 nk->port[pd->sidx] != sport) { 4272 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4273 &th->th_sum, &nk->addr[pd->sidx], 4274 nk->port[pd->sidx], 0, af); 4275 pd->sport = &th->th_sport; 4276 sport = th->th_sport; 4277 } 4278 4279 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4280 nk->port[pd->didx] != dport) { 4281 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4282 &th->th_sum, &nk->addr[pd->didx], 4283 nk->port[pd->didx], 0, af); 4284 dport = th->th_dport; 4285 pd->dport = &th->th_dport; 4286 } 4287 rewrite++; 4288 break; 4289 case IPPROTO_UDP: 4290 bproto_sum = pd->hdr.udp.uh_sum; 4291 pd->proto_sum = &pd->hdr.udp.uh_sum; 4292 4293 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4294 nk->port[pd->sidx] != sport) { 4295 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4296 pd->ip_sum, &pd->hdr.udp.uh_sum, 4297 &nk->addr[pd->sidx], 4298 nk->port[pd->sidx], 1, af); 4299 sport = pd->hdr.udp.uh_sport; 4300 pd->sport = &pd->hdr.udp.uh_sport; 4301 } 4302 4303 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4304 nk->port[pd->didx] != dport) { 4305 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4306 pd->ip_sum, &pd->hdr.udp.uh_sum, 4307 &nk->addr[pd->didx], 4308 nk->port[pd->didx], 1, af); 4309 dport = pd->hdr.udp.uh_dport; 4310 pd->dport = &pd->hdr.udp.uh_dport; 4311 } 4312 rewrite++; 4313 break; 4314 #ifdef INET 4315 case IPPROTO_ICMP: 4316 nk->port[0] = nk->port[1]; 4317 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4318 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4319 nk->addr[pd->sidx].v4.s_addr, 0); 4320 4321 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4322 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4323 nk->addr[pd->didx].v4.s_addr, 0); 4324 4325 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4326 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4327 pd->hdr.icmp.icmp_cksum, sport, 4328 nk->port[1], 0); 4329 pd->hdr.icmp.icmp_id = nk->port[1]; 4330 pd->sport = &pd->hdr.icmp.icmp_id; 4331 } 4332 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4333 break; 4334 #endif /* INET */ 4335 #ifdef INET6 4336 case IPPROTO_ICMPV6: 4337 nk->port[0] = nk->port[1]; 4338 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4339 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4340 &nk->addr[pd->sidx], 0); 4341 4342 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4343 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4344 &nk->addr[pd->didx], 0); 4345 rewrite++; 4346 break; 4347 #endif /* INET */ 4348 default: 4349 switch (af) { 4350 #ifdef INET 4351 case AF_INET: 4352 if (PF_ANEQ(saddr, 4353 &nk->addr[pd->sidx], AF_INET)) 4354 pf_change_a(&saddr->v4.s_addr, 4355 pd->ip_sum, 4356 nk->addr[pd->sidx].v4.s_addr, 0); 4357 4358 if (PF_ANEQ(daddr, 4359 &nk->addr[pd->didx], AF_INET)) 4360 pf_change_a(&daddr->v4.s_addr, 4361 pd->ip_sum, 4362 nk->addr[pd->didx].v4.s_addr, 0); 4363 break; 4364 #endif /* INET */ 4365 #ifdef INET6 4366 case AF_INET6: 4367 if (PF_ANEQ(saddr, 4368 &nk->addr[pd->sidx], AF_INET6)) 4369 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4370 4371 if (PF_ANEQ(daddr, 4372 &nk->addr[pd->didx], AF_INET6)) 4373 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4374 break; 4375 #endif /* INET */ 4376 } 4377 break; 4378 } 4379 if (nr->natpass) 4380 r = NULL; 4381 pd->nat_rule = nr; 4382 } 4383 4384 SLIST_INIT(&match_rules); 4385 while (r != NULL) { 4386 pf_counter_u64_add(&r->evaluations, 1); 4387 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4388 r = r->skip[PF_SKIP_IFP].ptr; 4389 else if (r->direction && r->direction != direction) 4390 r = r->skip[PF_SKIP_DIR].ptr; 4391 else if (r->af && r->af != af) 4392 r = r->skip[PF_SKIP_AF].ptr; 4393 else if (r->proto && r->proto != pd->proto) 4394 r = r->skip[PF_SKIP_PROTO].ptr; 4395 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4396 r->src.neg, kif, M_GETFIB(m))) 4397 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4398 /* tcp/udp only. port_op always 0 in other cases */ 4399 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4400 r->src.port[0], r->src.port[1], sport)) 4401 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4402 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4403 r->dst.neg, NULL, M_GETFIB(m))) 4404 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4405 /* tcp/udp only. port_op always 0 in other cases */ 4406 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4407 r->dst.port[0], r->dst.port[1], dport)) 4408 r = r->skip[PF_SKIP_DST_PORT].ptr; 4409 /* icmp only. type always 0 in other cases */ 4410 else if (r->type && r->type != icmptype + 1) 4411 r = TAILQ_NEXT(r, entries); 4412 /* icmp only. type always 0 in other cases */ 4413 else if (r->code && r->code != icmpcode + 1) 4414 r = TAILQ_NEXT(r, entries); 4415 else if (r->tos && !(r->tos == pd->tos)) 4416 r = TAILQ_NEXT(r, entries); 4417 else if (r->rule_flag & PFRULE_FRAGMENT) 4418 r = TAILQ_NEXT(r, entries); 4419 else if (pd->proto == IPPROTO_TCP && 4420 (r->flagset & th->th_flags) != r->flags) 4421 r = TAILQ_NEXT(r, entries); 4422 /* tcp/udp only. uid.op always 0 in other cases */ 4423 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4424 pf_socket_lookup(direction, pd, m), 1)) && 4425 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4426 pd->lookup.uid)) 4427 r = TAILQ_NEXT(r, entries); 4428 /* tcp/udp only. gid.op always 0 in other cases */ 4429 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4430 pf_socket_lookup(direction, pd, m), 1)) && 4431 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4432 pd->lookup.gid)) 4433 r = TAILQ_NEXT(r, entries); 4434 else if (r->prio && 4435 !pf_match_ieee8021q_pcp(r->prio, m)) 4436 r = TAILQ_NEXT(r, entries); 4437 else if (r->prob && 4438 r->prob <= arc4random()) 4439 r = TAILQ_NEXT(r, entries); 4440 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4441 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4442 r = TAILQ_NEXT(r, entries); 4443 else if (r->os_fingerprint != PF_OSFP_ANY && 4444 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4445 pf_osfp_fingerprint(pd, m, off, th), 4446 r->os_fingerprint))) 4447 r = TAILQ_NEXT(r, entries); 4448 else { 4449 if (r->tag) 4450 tag = r->tag; 4451 if (r->anchor == NULL) { 4452 if (r->action == PF_MATCH) { 4453 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4454 if (ri == NULL) { 4455 REASON_SET(&reason, PFRES_MEMORY); 4456 goto cleanup; 4457 } 4458 ri->r = r; 4459 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4460 pf_counter_u64_critical_enter(); 4461 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 4462 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 4463 pf_counter_u64_critical_exit(); 4464 pf_rule_to_actions(r, &pd->act); 4465 if (r->log) 4466 PFLOG_PACKET(kif, m, af, 4467 direction, PFRES_MATCH, r, 4468 a, ruleset, pd, 1); 4469 } else { 4470 match = 1; 4471 *rm = r; 4472 *am = a; 4473 *rsm = ruleset; 4474 } 4475 if ((*rm)->quick) 4476 break; 4477 r = TAILQ_NEXT(r, entries); 4478 } else 4479 pf_step_into_anchor(anchor_stack, &asd, 4480 &ruleset, PF_RULESET_FILTER, &r, &a, 4481 &match); 4482 } 4483 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4484 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4485 break; 4486 } 4487 r = *rm; 4488 a = *am; 4489 ruleset = *rsm; 4490 4491 REASON_SET(&reason, PFRES_MATCH); 4492 4493 /* apply actions for last matching pass/block rule */ 4494 pf_rule_to_actions(r, &pd->act); 4495 4496 if (r->log) { 4497 if (rewrite) 4498 m_copyback(m, off, hdrlen, pd->hdr.any); 4499 PFLOG_PACKET(kif, m, af, direction, reason, r, a, 4500 ruleset, pd, 1); 4501 } 4502 4503 if ((r->action == PF_DROP) && 4504 ((r->rule_flag & PFRULE_RETURNRST) || 4505 (r->rule_flag & PFRULE_RETURNICMP) || 4506 (r->rule_flag & PFRULE_RETURN))) { 4507 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4508 bip_sum, hdrlen, &reason, r->rtableid); 4509 } 4510 4511 if (r->action == PF_DROP) 4512 goto cleanup; 4513 4514 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4515 REASON_SET(&reason, PFRES_MEMORY); 4516 goto cleanup; 4517 } 4518 if (pd->act.rtableid >= 0) 4519 M_SETFIB(m, pd->act.rtableid); 4520 4521 if (!state_icmp && (r->keep_state || nr != NULL || 4522 (pd->flags & PFDESC_TCP_NORM))) { 4523 int action; 4524 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4525 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4526 hdrlen, &match_rules); 4527 if (action != PF_PASS) { 4528 if (action == PF_DROP && 4529 (r->rule_flag & PFRULE_RETURN)) 4530 pf_return(r, nr, pd, sk, off, m, th, kif, 4531 bproto_sum, bip_sum, hdrlen, &reason, 4532 pd->act.rtableid); 4533 return (action); 4534 } 4535 } else { 4536 if (sk != NULL) 4537 uma_zfree(V_pf_state_key_z, sk); 4538 if (nk != NULL) 4539 uma_zfree(V_pf_state_key_z, nk); 4540 } 4541 4542 /* copy back packet headers if we performed NAT operations */ 4543 if (rewrite) 4544 m_copyback(m, off, hdrlen, pd->hdr.any); 4545 4546 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4547 direction == PF_OUT && 4548 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4549 /* 4550 * We want the state created, but we dont 4551 * want to send this in case a partner 4552 * firewall has to know about it to allow 4553 * replies through it. 4554 */ 4555 return (PF_DEFER); 4556 4557 return (PF_PASS); 4558 4559 cleanup: 4560 while ((ri = SLIST_FIRST(&match_rules))) { 4561 SLIST_REMOVE_HEAD(&match_rules, entry); 4562 free(ri, M_PF_RULE_ITEM); 4563 } 4564 4565 if (sk != NULL) 4566 uma_zfree(V_pf_state_key_z, sk); 4567 if (nk != NULL) 4568 uma_zfree(V_pf_state_key_z, nk); 4569 return (PF_DROP); 4570 } 4571 4572 static int 4573 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4574 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4575 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4576 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4577 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 4578 struct pf_krule_slist *match_rules) 4579 { 4580 struct pf_kstate *s = NULL; 4581 struct pf_ksrc_node *sn = NULL; 4582 struct tcphdr *th = &pd->hdr.tcp; 4583 u_int16_t mss = V_tcp_mssdflt; 4584 u_short reason; 4585 4586 /* check maximums */ 4587 if (r->max_states && 4588 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4589 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4590 REASON_SET(&reason, PFRES_MAXSTATES); 4591 goto csfailed; 4592 } 4593 /* src node for filter rule */ 4594 if ((r->rule_flag & PFRULE_SRCTRACK || 4595 r->rpool.opts & PF_POOL_STICKYADDR) && 4596 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 4597 REASON_SET(&reason, PFRES_SRCLIMIT); 4598 goto csfailed; 4599 } 4600 /* src node for translation rule */ 4601 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4602 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 4603 REASON_SET(&reason, PFRES_SRCLIMIT); 4604 goto csfailed; 4605 } 4606 s = pf_alloc_state(M_NOWAIT); 4607 if (s == NULL) { 4608 REASON_SET(&reason, PFRES_MEMORY); 4609 goto csfailed; 4610 } 4611 s->rule.ptr = r; 4612 s->nat_rule.ptr = nr; 4613 s->anchor.ptr = a; 4614 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 4615 STATE_INC_COUNTERS(s); 4616 if (r->allow_opts) 4617 s->state_flags |= PFSTATE_ALLOWOPTS; 4618 if (r->rule_flag & PFRULE_STATESLOPPY) 4619 s->state_flags |= PFSTATE_SLOPPY; 4620 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 4621 s->state_flags |= PFSTATE_SCRUB_TCP; 4622 s->log = pd->act.log & PF_LOG_ALL; 4623 s->qid = pd->act.qid; 4624 s->pqid = pd->act.pqid; 4625 s->rtableid = pd->act.rtableid; 4626 s->min_ttl = pd->act.min_ttl; 4627 s->set_tos = pd->act.set_tos; 4628 s->max_mss = pd->act.max_mss; 4629 s->sync_state = PFSYNC_S_NONE; 4630 s->qid = pd->act.qid; 4631 s->pqid = pd->act.pqid; 4632 s->dnpipe = pd->act.dnpipe; 4633 s->dnrpipe = pd->act.dnrpipe; 4634 s->state_flags |= pd->act.flags; 4635 if (nr != NULL) 4636 s->log |= nr->log & PF_LOG_ALL; 4637 switch (pd->proto) { 4638 case IPPROTO_TCP: 4639 s->src.seqlo = ntohl(th->th_seq); 4640 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4641 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4642 r->keep_state == PF_STATE_MODULATE) { 4643 /* Generate sequence number modulator */ 4644 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4645 0) 4646 s->src.seqdiff = 1; 4647 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4648 htonl(s->src.seqlo + s->src.seqdiff), 0); 4649 *rewrite = 1; 4650 } else 4651 s->src.seqdiff = 0; 4652 if (th->th_flags & TH_SYN) { 4653 s->src.seqhi++; 4654 s->src.wscale = pf_get_wscale(m, off, 4655 th->th_off, pd->af); 4656 } 4657 s->src.max_win = MAX(ntohs(th->th_win), 1); 4658 if (s->src.wscale & PF_WSCALE_MASK) { 4659 /* Remove scale factor from initial window */ 4660 int win = s->src.max_win; 4661 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4662 s->src.max_win = (win - 1) >> 4663 (s->src.wscale & PF_WSCALE_MASK); 4664 } 4665 if (th->th_flags & TH_FIN) 4666 s->src.seqhi++; 4667 s->dst.seqhi = 1; 4668 s->dst.max_win = 1; 4669 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4670 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4671 s->timeout = PFTM_TCP_FIRST_PACKET; 4672 atomic_add_32(&V_pf_status.states_halfopen, 1); 4673 break; 4674 case IPPROTO_UDP: 4675 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4676 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4677 s->timeout = PFTM_UDP_FIRST_PACKET; 4678 break; 4679 case IPPROTO_ICMP: 4680 #ifdef INET6 4681 case IPPROTO_ICMPV6: 4682 #endif 4683 s->timeout = PFTM_ICMP_FIRST_PACKET; 4684 break; 4685 default: 4686 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4687 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4688 s->timeout = PFTM_OTHER_FIRST_PACKET; 4689 } 4690 4691 if (r->rt) { 4692 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 4693 REASON_SET(&reason, PFRES_MAPFAILED); 4694 pf_src_tree_remove_state(s); 4695 s->timeout = PFTM_UNLINKED; 4696 STATE_DEC_COUNTERS(s); 4697 pf_free_state(s); 4698 goto csfailed; 4699 } 4700 s->rt_kif = r->rpool.cur->kif; 4701 } 4702 4703 s->creation = time_uptime; 4704 s->expire = time_uptime; 4705 4706 if (sn != NULL) 4707 s->src_node = sn; 4708 if (nsn != NULL) { 4709 /* XXX We only modify one side for now. */ 4710 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4711 s->nat_src_node = nsn; 4712 } 4713 if (pd->proto == IPPROTO_TCP) { 4714 if (s->state_flags & PFSTATE_SCRUB_TCP && 4715 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 4716 REASON_SET(&reason, PFRES_MEMORY); 4717 pf_src_tree_remove_state(s); 4718 s->timeout = PFTM_UNLINKED; 4719 STATE_DEC_COUNTERS(s); 4720 pf_free_state(s); 4721 return (PF_DROP); 4722 } 4723 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 4724 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4725 &s->src, &s->dst, rewrite)) { 4726 /* This really shouldn't happen!!! */ 4727 DPFPRINTF(PF_DEBUG_URGENT, 4728 ("pf_normalize_tcp_stateful failed on first " 4729 "pkt\n")); 4730 pf_src_tree_remove_state(s); 4731 s->timeout = PFTM_UNLINKED; 4732 STATE_DEC_COUNTERS(s); 4733 pf_free_state(s); 4734 return (PF_DROP); 4735 } 4736 } 4737 s->direction = pd->dir; 4738 4739 /* 4740 * sk/nk could already been setup by pf_get_translation(). 4741 */ 4742 if (nr == NULL) { 4743 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4744 __func__, nr, sk, nk)); 4745 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4746 if (sk == NULL) 4747 goto csfailed; 4748 nk = sk; 4749 } else 4750 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4751 __func__, nr, sk, nk)); 4752 4753 /* Swap sk/nk for PF_OUT. */ 4754 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4755 (pd->dir == PF_IN) ? sk : nk, 4756 (pd->dir == PF_IN) ? nk : sk, s)) { 4757 REASON_SET(&reason, PFRES_STATEINS); 4758 pf_src_tree_remove_state(s); 4759 s->timeout = PFTM_UNLINKED; 4760 STATE_DEC_COUNTERS(s); 4761 pf_free_state(s); 4762 return (PF_DROP); 4763 } else 4764 *sm = s; 4765 4766 if (tag > 0) 4767 s->tag = tag; 4768 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4769 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4770 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4771 /* undo NAT changes, if they have taken place */ 4772 if (nr != NULL) { 4773 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4774 if (pd->dir == PF_OUT) 4775 skt = s->key[PF_SK_STACK]; 4776 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4777 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4778 if (pd->sport) 4779 *pd->sport = skt->port[pd->sidx]; 4780 if (pd->dport) 4781 *pd->dport = skt->port[pd->didx]; 4782 if (pd->proto_sum) 4783 *pd->proto_sum = bproto_sum; 4784 if (pd->ip_sum) 4785 *pd->ip_sum = bip_sum; 4786 m_copyback(m, off, hdrlen, pd->hdr.any); 4787 } 4788 s->src.seqhi = htonl(arc4random()); 4789 /* Find mss option */ 4790 int rtid = M_GETFIB(m); 4791 mss = pf_get_mss(m, off, th->th_off, pd->af); 4792 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4793 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4794 s->src.mss = mss; 4795 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4796 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4797 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, pd->act.rtableid); 4798 REASON_SET(&reason, PFRES_SYNPROXY); 4799 return (PF_SYNPROXY_DROP); 4800 } 4801 4802 return (PF_PASS); 4803 4804 csfailed: 4805 if (sk != NULL) 4806 uma_zfree(V_pf_state_key_z, sk); 4807 if (nk != NULL) 4808 uma_zfree(V_pf_state_key_z, nk); 4809 4810 if (sn != NULL) { 4811 PF_SRC_NODE_LOCK(sn); 4812 if (--sn->states == 0 && sn->expire == 0) { 4813 pf_unlink_src_node(sn); 4814 uma_zfree(V_pf_sources_z, sn); 4815 counter_u64_add( 4816 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4817 } 4818 PF_SRC_NODE_UNLOCK(sn); 4819 } 4820 4821 if (nsn != sn && nsn != NULL) { 4822 PF_SRC_NODE_LOCK(nsn); 4823 if (--nsn->states == 0 && nsn->expire == 0) { 4824 pf_unlink_src_node(nsn); 4825 uma_zfree(V_pf_sources_z, nsn); 4826 counter_u64_add( 4827 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4828 } 4829 PF_SRC_NODE_UNLOCK(nsn); 4830 } 4831 4832 return (PF_DROP); 4833 } 4834 4835 static int 4836 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif, 4837 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 4838 struct pf_kruleset **rsm) 4839 { 4840 struct pf_krule *r, *a = NULL; 4841 struct pf_kruleset *ruleset = NULL; 4842 struct pf_krule_slist match_rules; 4843 struct pf_krule_item *ri; 4844 sa_family_t af = pd->af; 4845 u_short reason; 4846 int tag = -1; 4847 int asd = 0; 4848 int match = 0; 4849 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4850 4851 PF_RULES_RASSERT(); 4852 4853 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4854 SLIST_INIT(&match_rules); 4855 while (r != NULL) { 4856 pf_counter_u64_add(&r->evaluations, 1); 4857 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4858 r = r->skip[PF_SKIP_IFP].ptr; 4859 else if (r->direction && r->direction != direction) 4860 r = r->skip[PF_SKIP_DIR].ptr; 4861 else if (r->af && r->af != af) 4862 r = r->skip[PF_SKIP_AF].ptr; 4863 else if (r->proto && r->proto != pd->proto) 4864 r = r->skip[PF_SKIP_PROTO].ptr; 4865 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 4866 r->src.neg, kif, M_GETFIB(m))) 4867 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4868 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 4869 r->dst.neg, NULL, M_GETFIB(m))) 4870 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4871 else if (r->tos && !(r->tos == pd->tos)) 4872 r = TAILQ_NEXT(r, entries); 4873 else if (r->os_fingerprint != PF_OSFP_ANY) 4874 r = TAILQ_NEXT(r, entries); 4875 else if (pd->proto == IPPROTO_UDP && 4876 (r->src.port_op || r->dst.port_op)) 4877 r = TAILQ_NEXT(r, entries); 4878 else if (pd->proto == IPPROTO_TCP && 4879 (r->src.port_op || r->dst.port_op || r->flagset)) 4880 r = TAILQ_NEXT(r, entries); 4881 else if ((pd->proto == IPPROTO_ICMP || 4882 pd->proto == IPPROTO_ICMPV6) && 4883 (r->type || r->code)) 4884 r = TAILQ_NEXT(r, entries); 4885 else if (r->prio && 4886 !pf_match_ieee8021q_pcp(r->prio, m)) 4887 r = TAILQ_NEXT(r, entries); 4888 else if (r->prob && r->prob <= 4889 (arc4random() % (UINT_MAX - 1) + 1)) 4890 r = TAILQ_NEXT(r, entries); 4891 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4892 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4893 r = TAILQ_NEXT(r, entries); 4894 else { 4895 if (r->anchor == NULL) { 4896 if (r->action == PF_MATCH) { 4897 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4898 if (ri == NULL) { 4899 REASON_SET(&reason, PFRES_MEMORY); 4900 goto cleanup; 4901 } 4902 ri->r = r; 4903 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4904 pf_counter_u64_critical_enter(); 4905 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 4906 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 4907 pf_counter_u64_critical_exit(); 4908 pf_rule_to_actions(r, &pd->act); 4909 if (r->log) 4910 PFLOG_PACKET(kif, m, af, 4911 direction, PFRES_MATCH, r, 4912 a, ruleset, pd, 1); 4913 } else { 4914 match = 1; 4915 *rm = r; 4916 *am = a; 4917 *rsm = ruleset; 4918 } 4919 if ((*rm)->quick) 4920 break; 4921 r = TAILQ_NEXT(r, entries); 4922 } else 4923 pf_step_into_anchor(anchor_stack, &asd, 4924 &ruleset, PF_RULESET_FILTER, &r, &a, 4925 &match); 4926 } 4927 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4928 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4929 break; 4930 } 4931 r = *rm; 4932 a = *am; 4933 ruleset = *rsm; 4934 4935 REASON_SET(&reason, PFRES_MATCH); 4936 4937 /* apply actions for last matching pass/block rule */ 4938 pf_rule_to_actions(r, &pd->act); 4939 4940 if (r->log) 4941 PFLOG_PACKET(kif, m, af, direction, reason, r, a, 4942 ruleset, pd, 1); 4943 4944 if (r->action != PF_PASS) 4945 return (PF_DROP); 4946 4947 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4948 REASON_SET(&reason, PFRES_MEMORY); 4949 goto cleanup; 4950 } 4951 4952 return (PF_PASS); 4953 4954 cleanup: 4955 while ((ri = SLIST_FIRST(&match_rules))) { 4956 SLIST_REMOVE_HEAD(&match_rules, entry); 4957 free(ri, M_PF_RULE_ITEM); 4958 } 4959 4960 return (PF_DROP); 4961 } 4962 4963 static int 4964 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 4965 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 4966 int *copyback) 4967 { 4968 struct tcphdr *th = &pd->hdr.tcp; 4969 struct pf_state_peer *src, *dst; 4970 u_int16_t win = ntohs(th->th_win); 4971 u_int32_t ack, end, seq, orig_seq; 4972 u_int8_t sws, dws, psrc, pdst; 4973 int ackskew; 4974 4975 if (pd->dir == (*state)->direction) { 4976 src = &(*state)->src; 4977 dst = &(*state)->dst; 4978 psrc = PF_PEER_SRC; 4979 pdst = PF_PEER_DST; 4980 } else { 4981 src = &(*state)->dst; 4982 dst = &(*state)->src; 4983 psrc = PF_PEER_DST; 4984 pdst = PF_PEER_SRC; 4985 } 4986 4987 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4988 sws = src->wscale & PF_WSCALE_MASK; 4989 dws = dst->wscale & PF_WSCALE_MASK; 4990 } else 4991 sws = dws = 0; 4992 4993 /* 4994 * Sequence tracking algorithm from Guido van Rooij's paper: 4995 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4996 * tcp_filtering.ps 4997 */ 4998 4999 orig_seq = seq = ntohl(th->th_seq); 5000 if (src->seqlo == 0) { 5001 /* First packet from this end. Set its state */ 5002 5003 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5004 src->scrub == NULL) { 5005 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5006 REASON_SET(reason, PFRES_MEMORY); 5007 return (PF_DROP); 5008 } 5009 } 5010 5011 /* Deferred generation of sequence number modulator */ 5012 if (dst->seqdiff && !src->seqdiff) { 5013 /* use random iss for the TCP server */ 5014 while ((src->seqdiff = arc4random() - seq) == 0) 5015 ; 5016 ack = ntohl(th->th_ack) - dst->seqdiff; 5017 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5018 src->seqdiff), 0); 5019 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5020 *copyback = 1; 5021 } else { 5022 ack = ntohl(th->th_ack); 5023 } 5024 5025 end = seq + pd->p_len; 5026 if (th->th_flags & TH_SYN) { 5027 end++; 5028 if (dst->wscale & PF_WSCALE_FLAG) { 5029 src->wscale = pf_get_wscale(m, off, th->th_off, 5030 pd->af); 5031 if (src->wscale & PF_WSCALE_FLAG) { 5032 /* Remove scale factor from initial 5033 * window */ 5034 sws = src->wscale & PF_WSCALE_MASK; 5035 win = ((u_int32_t)win + (1 << sws) - 1) 5036 >> sws; 5037 dws = dst->wscale & PF_WSCALE_MASK; 5038 } else { 5039 /* fixup other window */ 5040 dst->max_win <<= dst->wscale & 5041 PF_WSCALE_MASK; 5042 /* in case of a retrans SYN|ACK */ 5043 dst->wscale = 0; 5044 } 5045 } 5046 } 5047 if (th->th_flags & TH_FIN) 5048 end++; 5049 5050 src->seqlo = seq; 5051 if (src->state < TCPS_SYN_SENT) 5052 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5053 5054 /* 5055 * May need to slide the window (seqhi may have been set by 5056 * the crappy stack check or if we picked up the connection 5057 * after establishment) 5058 */ 5059 if (src->seqhi == 1 || 5060 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5061 src->seqhi = end + MAX(1, dst->max_win << dws); 5062 if (win > src->max_win) 5063 src->max_win = win; 5064 5065 } else { 5066 ack = ntohl(th->th_ack) - dst->seqdiff; 5067 if (src->seqdiff) { 5068 /* Modulate sequence numbers */ 5069 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5070 src->seqdiff), 0); 5071 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5072 *copyback = 1; 5073 } 5074 end = seq + pd->p_len; 5075 if (th->th_flags & TH_SYN) 5076 end++; 5077 if (th->th_flags & TH_FIN) 5078 end++; 5079 } 5080 5081 if ((th->th_flags & TH_ACK) == 0) { 5082 /* Let it pass through the ack skew check */ 5083 ack = dst->seqlo; 5084 } else if ((ack == 0 && 5085 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5086 /* broken tcp stacks do not set ack */ 5087 (dst->state < TCPS_SYN_SENT)) { 5088 /* 5089 * Many stacks (ours included) will set the ACK number in an 5090 * FIN|ACK if the SYN times out -- no sequence to ACK. 5091 */ 5092 ack = dst->seqlo; 5093 } 5094 5095 if (seq == end) { 5096 /* Ease sequencing restrictions on no data packets */ 5097 seq = src->seqlo; 5098 end = seq; 5099 } 5100 5101 ackskew = dst->seqlo - ack; 5102 5103 /* 5104 * Need to demodulate the sequence numbers in any TCP SACK options 5105 * (Selective ACK). We could optionally validate the SACK values 5106 * against the current ACK window, either forwards or backwards, but 5107 * I'm not confident that SACK has been implemented properly 5108 * everywhere. It wouldn't surprise me if several stacks accidentally 5109 * SACK too far backwards of previously ACKed data. There really aren't 5110 * any security implications of bad SACKing unless the target stack 5111 * doesn't validate the option length correctly. Someone trying to 5112 * spoof into a TCP connection won't bother blindly sending SACK 5113 * options anyway. 5114 */ 5115 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5116 if (pf_modulate_sack(m, off, pd, th, dst)) 5117 *copyback = 1; 5118 } 5119 5120 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5121 if (SEQ_GEQ(src->seqhi, end) && 5122 /* Last octet inside other's window space */ 5123 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5124 /* Retrans: not more than one window back */ 5125 (ackskew >= -MAXACKWINDOW) && 5126 /* Acking not more than one reassembled fragment backwards */ 5127 (ackskew <= (MAXACKWINDOW << sws)) && 5128 /* Acking not more than one window forward */ 5129 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5130 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 5131 (pd->flags & PFDESC_IP_REAS) == 0)) { 5132 /* Require an exact/+1 sequence match on resets when possible */ 5133 5134 if (dst->scrub || src->scrub) { 5135 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5136 *state, src, dst, copyback)) 5137 return (PF_DROP); 5138 } 5139 5140 /* update max window */ 5141 if (src->max_win < win) 5142 src->max_win = win; 5143 /* synchronize sequencing */ 5144 if (SEQ_GT(end, src->seqlo)) 5145 src->seqlo = end; 5146 /* slide the window of what the other end can send */ 5147 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5148 dst->seqhi = ack + MAX((win << sws), 1); 5149 5150 /* update states */ 5151 if (th->th_flags & TH_SYN) 5152 if (src->state < TCPS_SYN_SENT) 5153 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5154 if (th->th_flags & TH_FIN) 5155 if (src->state < TCPS_CLOSING) 5156 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5157 if (th->th_flags & TH_ACK) { 5158 if (dst->state == TCPS_SYN_SENT) { 5159 pf_set_protostate(*state, pdst, 5160 TCPS_ESTABLISHED); 5161 if (src->state == TCPS_ESTABLISHED && 5162 (*state)->src_node != NULL && 5163 pf_src_connlimit(state)) { 5164 REASON_SET(reason, PFRES_SRCLIMIT); 5165 return (PF_DROP); 5166 } 5167 } else if (dst->state == TCPS_CLOSING) 5168 pf_set_protostate(*state, pdst, 5169 TCPS_FIN_WAIT_2); 5170 } 5171 if (th->th_flags & TH_RST) 5172 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5173 5174 /* update expire time */ 5175 (*state)->expire = time_uptime; 5176 if (src->state >= TCPS_FIN_WAIT_2 && 5177 dst->state >= TCPS_FIN_WAIT_2) 5178 (*state)->timeout = PFTM_TCP_CLOSED; 5179 else if (src->state >= TCPS_CLOSING && 5180 dst->state >= TCPS_CLOSING) 5181 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5182 else if (src->state < TCPS_ESTABLISHED || 5183 dst->state < TCPS_ESTABLISHED) 5184 (*state)->timeout = PFTM_TCP_OPENING; 5185 else if (src->state >= TCPS_CLOSING || 5186 dst->state >= TCPS_CLOSING) 5187 (*state)->timeout = PFTM_TCP_CLOSING; 5188 else 5189 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5190 5191 /* Fall through to PASS packet */ 5192 5193 } else if ((dst->state < TCPS_SYN_SENT || 5194 dst->state >= TCPS_FIN_WAIT_2 || 5195 src->state >= TCPS_FIN_WAIT_2) && 5196 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5197 /* Within a window forward of the originating packet */ 5198 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5199 /* Within a window backward of the originating packet */ 5200 5201 /* 5202 * This currently handles three situations: 5203 * 1) Stupid stacks will shotgun SYNs before their peer 5204 * replies. 5205 * 2) When PF catches an already established stream (the 5206 * firewall rebooted, the state table was flushed, routes 5207 * changed...) 5208 * 3) Packets get funky immediately after the connection 5209 * closes (this should catch Solaris spurious ACK|FINs 5210 * that web servers like to spew after a close) 5211 * 5212 * This must be a little more careful than the above code 5213 * since packet floods will also be caught here. We don't 5214 * update the TTL here to mitigate the damage of a packet 5215 * flood and so the same code can handle awkward establishment 5216 * and a loosened connection close. 5217 * In the establishment case, a correct peer response will 5218 * validate the connection, go through the normal state code 5219 * and keep updating the state TTL. 5220 */ 5221 5222 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5223 printf("pf: loose state match: "); 5224 pf_print_state(*state); 5225 pf_print_flags(th->th_flags); 5226 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5227 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5228 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5229 (unsigned long long)(*state)->packets[1], 5230 pd->dir == PF_IN ? "in" : "out", 5231 pd->dir == (*state)->direction ? "fwd" : "rev"); 5232 } 5233 5234 if (dst->scrub || src->scrub) { 5235 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5236 *state, src, dst, copyback)) 5237 return (PF_DROP); 5238 } 5239 5240 /* update max window */ 5241 if (src->max_win < win) 5242 src->max_win = win; 5243 /* synchronize sequencing */ 5244 if (SEQ_GT(end, src->seqlo)) 5245 src->seqlo = end; 5246 /* slide the window of what the other end can send */ 5247 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5248 dst->seqhi = ack + MAX((win << sws), 1); 5249 5250 /* 5251 * Cannot set dst->seqhi here since this could be a shotgunned 5252 * SYN and not an already established connection. 5253 */ 5254 5255 if (th->th_flags & TH_FIN) 5256 if (src->state < TCPS_CLOSING) 5257 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5258 if (th->th_flags & TH_RST) 5259 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5260 5261 /* Fall through to PASS packet */ 5262 5263 } else { 5264 if ((*state)->dst.state == TCPS_SYN_SENT && 5265 (*state)->src.state == TCPS_SYN_SENT) { 5266 /* Send RST for state mismatches during handshake */ 5267 if (!(th->th_flags & TH_RST)) 5268 pf_send_tcp((*state)->rule.ptr, pd->af, 5269 pd->dst, pd->src, th->th_dport, 5270 th->th_sport, ntohl(th->th_ack), 0, 5271 TH_RST, 0, 0, 5272 (*state)->rule.ptr->return_ttl, 1, 0, 5273 (*state)->rtableid); 5274 src->seqlo = 0; 5275 src->seqhi = 1; 5276 src->max_win = 1; 5277 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5278 printf("pf: BAD state: "); 5279 pf_print_state(*state); 5280 pf_print_flags(th->th_flags); 5281 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5282 "pkts=%llu:%llu dir=%s,%s\n", 5283 seq, orig_seq, ack, pd->p_len, ackskew, 5284 (unsigned long long)(*state)->packets[0], 5285 (unsigned long long)(*state)->packets[1], 5286 pd->dir == PF_IN ? "in" : "out", 5287 pd->dir == (*state)->direction ? "fwd" : "rev"); 5288 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5289 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5290 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5291 ' ': '2', 5292 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5293 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5294 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5295 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5296 } 5297 REASON_SET(reason, PFRES_BADSTATE); 5298 return (PF_DROP); 5299 } 5300 5301 return (PF_PASS); 5302 } 5303 5304 static int 5305 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5306 { 5307 struct tcphdr *th = &pd->hdr.tcp; 5308 struct pf_state_peer *src, *dst; 5309 u_int8_t psrc, pdst; 5310 5311 if (pd->dir == (*state)->direction) { 5312 src = &(*state)->src; 5313 dst = &(*state)->dst; 5314 psrc = PF_PEER_SRC; 5315 pdst = PF_PEER_DST; 5316 } else { 5317 src = &(*state)->dst; 5318 dst = &(*state)->src; 5319 psrc = PF_PEER_DST; 5320 pdst = PF_PEER_SRC; 5321 } 5322 5323 if (th->th_flags & TH_SYN) 5324 if (src->state < TCPS_SYN_SENT) 5325 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5326 if (th->th_flags & TH_FIN) 5327 if (src->state < TCPS_CLOSING) 5328 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5329 if (th->th_flags & TH_ACK) { 5330 if (dst->state == TCPS_SYN_SENT) { 5331 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5332 if (src->state == TCPS_ESTABLISHED && 5333 (*state)->src_node != NULL && 5334 pf_src_connlimit(state)) { 5335 REASON_SET(reason, PFRES_SRCLIMIT); 5336 return (PF_DROP); 5337 } 5338 } else if (dst->state == TCPS_CLOSING) { 5339 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5340 } else if (src->state == TCPS_SYN_SENT && 5341 dst->state < TCPS_SYN_SENT) { 5342 /* 5343 * Handle a special sloppy case where we only see one 5344 * half of the connection. If there is a ACK after 5345 * the initial SYN without ever seeing a packet from 5346 * the destination, set the connection to established. 5347 */ 5348 pf_set_protostate(*state, PF_PEER_BOTH, 5349 TCPS_ESTABLISHED); 5350 dst->state = src->state = TCPS_ESTABLISHED; 5351 if ((*state)->src_node != NULL && 5352 pf_src_connlimit(state)) { 5353 REASON_SET(reason, PFRES_SRCLIMIT); 5354 return (PF_DROP); 5355 } 5356 } else if (src->state == TCPS_CLOSING && 5357 dst->state == TCPS_ESTABLISHED && 5358 dst->seqlo == 0) { 5359 /* 5360 * Handle the closing of half connections where we 5361 * don't see the full bidirectional FIN/ACK+ACK 5362 * handshake. 5363 */ 5364 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5365 } 5366 } 5367 if (th->th_flags & TH_RST) 5368 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5369 5370 /* update expire time */ 5371 (*state)->expire = time_uptime; 5372 if (src->state >= TCPS_FIN_WAIT_2 && 5373 dst->state >= TCPS_FIN_WAIT_2) 5374 (*state)->timeout = PFTM_TCP_CLOSED; 5375 else if (src->state >= TCPS_CLOSING && 5376 dst->state >= TCPS_CLOSING) 5377 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5378 else if (src->state < TCPS_ESTABLISHED || 5379 dst->state < TCPS_ESTABLISHED) 5380 (*state)->timeout = PFTM_TCP_OPENING; 5381 else if (src->state >= TCPS_CLOSING || 5382 dst->state >= TCPS_CLOSING) 5383 (*state)->timeout = PFTM_TCP_CLOSING; 5384 else 5385 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5386 5387 return (PF_PASS); 5388 } 5389 5390 static int 5391 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5392 { 5393 struct pf_state_key *sk = (*state)->key[pd->didx]; 5394 struct tcphdr *th = &pd->hdr.tcp; 5395 5396 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5397 if (pd->dir != (*state)->direction) { 5398 REASON_SET(reason, PFRES_SYNPROXY); 5399 return (PF_SYNPROXY_DROP); 5400 } 5401 if (th->th_flags & TH_SYN) { 5402 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5403 REASON_SET(reason, PFRES_SYNPROXY); 5404 return (PF_DROP); 5405 } 5406 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5407 pd->src, th->th_dport, th->th_sport, 5408 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5409 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, 5410 (*state)->rtableid); 5411 REASON_SET(reason, PFRES_SYNPROXY); 5412 return (PF_SYNPROXY_DROP); 5413 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5414 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5415 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5416 REASON_SET(reason, PFRES_SYNPROXY); 5417 return (PF_DROP); 5418 } else if ((*state)->src_node != NULL && 5419 pf_src_connlimit(state)) { 5420 REASON_SET(reason, PFRES_SRCLIMIT); 5421 return (PF_DROP); 5422 } else 5423 pf_set_protostate(*state, PF_PEER_SRC, 5424 PF_TCPS_PROXY_DST); 5425 } 5426 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5427 if (pd->dir == (*state)->direction) { 5428 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5429 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5430 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5431 REASON_SET(reason, PFRES_SYNPROXY); 5432 return (PF_DROP); 5433 } 5434 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5435 if ((*state)->dst.seqhi == 1) 5436 (*state)->dst.seqhi = htonl(arc4random()); 5437 pf_send_tcp((*state)->rule.ptr, pd->af, 5438 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5439 sk->port[pd->sidx], sk->port[pd->didx], 5440 (*state)->dst.seqhi, 0, TH_SYN, 0, 5441 (*state)->src.mss, 0, 0, (*state)->tag, 5442 (*state)->rtableid); 5443 REASON_SET(reason, PFRES_SYNPROXY); 5444 return (PF_SYNPROXY_DROP); 5445 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5446 (TH_SYN|TH_ACK)) || 5447 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5448 REASON_SET(reason, PFRES_SYNPROXY); 5449 return (PF_DROP); 5450 } else { 5451 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5452 (*state)->dst.seqlo = ntohl(th->th_seq); 5453 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5454 pd->src, th->th_dport, th->th_sport, 5455 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5456 TH_ACK, (*state)->src.max_win, 0, 0, 0, 5457 (*state)->tag, (*state)->rtableid); 5458 pf_send_tcp((*state)->rule.ptr, pd->af, 5459 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5460 sk->port[pd->sidx], sk->port[pd->didx], 5461 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5462 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, 5463 (*state)->rtableid); 5464 (*state)->src.seqdiff = (*state)->dst.seqhi - 5465 (*state)->src.seqlo; 5466 (*state)->dst.seqdiff = (*state)->src.seqhi - 5467 (*state)->dst.seqlo; 5468 (*state)->src.seqhi = (*state)->src.seqlo + 5469 (*state)->dst.max_win; 5470 (*state)->dst.seqhi = (*state)->dst.seqlo + 5471 (*state)->src.max_win; 5472 (*state)->src.wscale = (*state)->dst.wscale = 0; 5473 pf_set_protostate(*state, PF_PEER_BOTH, 5474 TCPS_ESTABLISHED); 5475 REASON_SET(reason, PFRES_SYNPROXY); 5476 return (PF_SYNPROXY_DROP); 5477 } 5478 } 5479 5480 return (PF_PASS); 5481 } 5482 5483 static int 5484 pf_test_state_tcp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5485 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5486 u_short *reason) 5487 { 5488 struct pf_state_key_cmp key; 5489 struct tcphdr *th = &pd->hdr.tcp; 5490 int copyback = 0; 5491 int action; 5492 struct pf_state_peer *src, *dst; 5493 5494 bzero(&key, sizeof(key)); 5495 key.af = pd->af; 5496 key.proto = IPPROTO_TCP; 5497 if (direction == PF_IN) { /* wire side, straight */ 5498 PF_ACPY(&key.addr[0], pd->src, key.af); 5499 PF_ACPY(&key.addr[1], pd->dst, key.af); 5500 key.port[0] = th->th_sport; 5501 key.port[1] = th->th_dport; 5502 } else { /* stack side, reverse */ 5503 PF_ACPY(&key.addr[1], pd->src, key.af); 5504 PF_ACPY(&key.addr[0], pd->dst, key.af); 5505 key.port[1] = th->th_sport; 5506 key.port[0] = th->th_dport; 5507 } 5508 5509 STATE_LOOKUP(kif, &key, direction, *state, pd); 5510 5511 if (direction == (*state)->direction) { 5512 src = &(*state)->src; 5513 dst = &(*state)->dst; 5514 } else { 5515 src = &(*state)->dst; 5516 dst = &(*state)->src; 5517 } 5518 5519 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5520 return (action); 5521 5522 if (dst->state >= TCPS_FIN_WAIT_2 && 5523 src->state >= TCPS_FIN_WAIT_2 && 5524 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 5525 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 5526 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 5527 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5528 printf("pf: state reuse "); 5529 pf_print_state(*state); 5530 pf_print_flags(th->th_flags); 5531 printf("\n"); 5532 } 5533 /* XXX make sure it's the same direction ?? */ 5534 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5535 pf_unlink_state(*state); 5536 *state = NULL; 5537 return (PF_DROP); 5538 } 5539 5540 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5541 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5542 return (PF_DROP); 5543 } else { 5544 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5545 ©back) == PF_DROP) 5546 return (PF_DROP); 5547 } 5548 5549 /* translate source/destination address, if necessary */ 5550 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5551 struct pf_state_key *nk = (*state)->key[pd->didx]; 5552 5553 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5554 nk->port[pd->sidx] != th->th_sport) 5555 pf_change_ap(m, pd->src, &th->th_sport, 5556 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5557 nk->port[pd->sidx], 0, pd->af); 5558 5559 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5560 nk->port[pd->didx] != th->th_dport) 5561 pf_change_ap(m, pd->dst, &th->th_dport, 5562 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5563 nk->port[pd->didx], 0, pd->af); 5564 copyback = 1; 5565 } 5566 5567 /* Copyback sequence modulation or stateful scrub changes if needed */ 5568 if (copyback) 5569 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5570 5571 return (PF_PASS); 5572 } 5573 5574 static int 5575 pf_test_state_udp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5576 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5577 { 5578 struct pf_state_peer *src, *dst; 5579 struct pf_state_key_cmp key; 5580 struct udphdr *uh = &pd->hdr.udp; 5581 uint8_t psrc, pdst; 5582 5583 bzero(&key, sizeof(key)); 5584 key.af = pd->af; 5585 key.proto = IPPROTO_UDP; 5586 if (direction == PF_IN) { /* wire side, straight */ 5587 PF_ACPY(&key.addr[0], pd->src, key.af); 5588 PF_ACPY(&key.addr[1], pd->dst, key.af); 5589 key.port[0] = uh->uh_sport; 5590 key.port[1] = uh->uh_dport; 5591 } else { /* stack side, reverse */ 5592 PF_ACPY(&key.addr[1], pd->src, key.af); 5593 PF_ACPY(&key.addr[0], pd->dst, key.af); 5594 key.port[1] = uh->uh_sport; 5595 key.port[0] = uh->uh_dport; 5596 } 5597 5598 STATE_LOOKUP(kif, &key, direction, *state, pd); 5599 5600 if (direction == (*state)->direction) { 5601 src = &(*state)->src; 5602 dst = &(*state)->dst; 5603 psrc = PF_PEER_SRC; 5604 pdst = PF_PEER_DST; 5605 } else { 5606 src = &(*state)->dst; 5607 dst = &(*state)->src; 5608 psrc = PF_PEER_DST; 5609 pdst = PF_PEER_SRC; 5610 } 5611 5612 /* update states */ 5613 if (src->state < PFUDPS_SINGLE) 5614 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5615 if (dst->state == PFUDPS_SINGLE) 5616 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5617 5618 /* update expire time */ 5619 (*state)->expire = time_uptime; 5620 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5621 (*state)->timeout = PFTM_UDP_MULTIPLE; 5622 else 5623 (*state)->timeout = PFTM_UDP_SINGLE; 5624 5625 /* translate source/destination address, if necessary */ 5626 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5627 struct pf_state_key *nk = (*state)->key[pd->didx]; 5628 5629 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5630 nk->port[pd->sidx] != uh->uh_sport) 5631 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5632 &uh->uh_sum, &nk->addr[pd->sidx], 5633 nk->port[pd->sidx], 1, pd->af); 5634 5635 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5636 nk->port[pd->didx] != uh->uh_dport) 5637 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5638 &uh->uh_sum, &nk->addr[pd->didx], 5639 nk->port[pd->didx], 1, pd->af); 5640 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5641 } 5642 5643 return (PF_PASS); 5644 } 5645 5646 static int 5647 pf_test_state_icmp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5648 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5649 { 5650 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 5651 u_int16_t icmpid = 0, *icmpsum; 5652 u_int8_t icmptype, icmpcode; 5653 int state_icmp = 0; 5654 struct pf_state_key_cmp key; 5655 5656 bzero(&key, sizeof(key)); 5657 switch (pd->proto) { 5658 #ifdef INET 5659 case IPPROTO_ICMP: 5660 icmptype = pd->hdr.icmp.icmp_type; 5661 icmpcode = pd->hdr.icmp.icmp_code; 5662 icmpid = pd->hdr.icmp.icmp_id; 5663 icmpsum = &pd->hdr.icmp.icmp_cksum; 5664 5665 if (icmptype == ICMP_UNREACH || 5666 icmptype == ICMP_SOURCEQUENCH || 5667 icmptype == ICMP_REDIRECT || 5668 icmptype == ICMP_TIMXCEED || 5669 icmptype == ICMP_PARAMPROB) 5670 state_icmp++; 5671 break; 5672 #endif /* INET */ 5673 #ifdef INET6 5674 case IPPROTO_ICMPV6: 5675 icmptype = pd->hdr.icmp6.icmp6_type; 5676 icmpcode = pd->hdr.icmp6.icmp6_code; 5677 icmpid = pd->hdr.icmp6.icmp6_id; 5678 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 5679 5680 if (icmptype == ICMP6_DST_UNREACH || 5681 icmptype == ICMP6_PACKET_TOO_BIG || 5682 icmptype == ICMP6_TIME_EXCEEDED || 5683 icmptype == ICMP6_PARAM_PROB) 5684 state_icmp++; 5685 break; 5686 #endif /* INET6 */ 5687 } 5688 5689 if (!state_icmp) { 5690 /* 5691 * ICMP query/reply message not related to a TCP/UDP packet. 5692 * Search for an ICMP state. 5693 */ 5694 key.af = pd->af; 5695 key.proto = pd->proto; 5696 key.port[0] = key.port[1] = icmpid; 5697 if (direction == PF_IN) { /* wire side, straight */ 5698 PF_ACPY(&key.addr[0], pd->src, key.af); 5699 PF_ACPY(&key.addr[1], pd->dst, key.af); 5700 } else { /* stack side, reverse */ 5701 PF_ACPY(&key.addr[1], pd->src, key.af); 5702 PF_ACPY(&key.addr[0], pd->dst, key.af); 5703 } 5704 5705 STATE_LOOKUP(kif, &key, direction, *state, pd); 5706 5707 (*state)->expire = time_uptime; 5708 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 5709 5710 /* translate source/destination address, if necessary */ 5711 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5712 struct pf_state_key *nk = (*state)->key[pd->didx]; 5713 5714 switch (pd->af) { 5715 #ifdef INET 5716 case AF_INET: 5717 if (PF_ANEQ(pd->src, 5718 &nk->addr[pd->sidx], AF_INET)) 5719 pf_change_a(&saddr->v4.s_addr, 5720 pd->ip_sum, 5721 nk->addr[pd->sidx].v4.s_addr, 0); 5722 5723 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 5724 AF_INET)) 5725 pf_change_a(&daddr->v4.s_addr, 5726 pd->ip_sum, 5727 nk->addr[pd->didx].v4.s_addr, 0); 5728 5729 if (nk->port[0] != 5730 pd->hdr.icmp.icmp_id) { 5731 pd->hdr.icmp.icmp_cksum = 5732 pf_cksum_fixup( 5733 pd->hdr.icmp.icmp_cksum, icmpid, 5734 nk->port[pd->sidx], 0); 5735 pd->hdr.icmp.icmp_id = 5736 nk->port[pd->sidx]; 5737 } 5738 5739 m_copyback(m, off, ICMP_MINLEN, 5740 (caddr_t )&pd->hdr.icmp); 5741 break; 5742 #endif /* INET */ 5743 #ifdef INET6 5744 case AF_INET6: 5745 if (PF_ANEQ(pd->src, 5746 &nk->addr[pd->sidx], AF_INET6)) 5747 pf_change_a6(saddr, 5748 &pd->hdr.icmp6.icmp6_cksum, 5749 &nk->addr[pd->sidx], 0); 5750 5751 if (PF_ANEQ(pd->dst, 5752 &nk->addr[pd->didx], AF_INET6)) 5753 pf_change_a6(daddr, 5754 &pd->hdr.icmp6.icmp6_cksum, 5755 &nk->addr[pd->didx], 0); 5756 5757 m_copyback(m, off, sizeof(struct icmp6_hdr), 5758 (caddr_t )&pd->hdr.icmp6); 5759 break; 5760 #endif /* INET6 */ 5761 } 5762 } 5763 return (PF_PASS); 5764 5765 } else { 5766 /* 5767 * ICMP error message in response to a TCP/UDP packet. 5768 * Extract the inner TCP/UDP header and search for that state. 5769 */ 5770 5771 struct pf_pdesc pd2; 5772 bzero(&pd2, sizeof pd2); 5773 #ifdef INET 5774 struct ip h2; 5775 #endif /* INET */ 5776 #ifdef INET6 5777 struct ip6_hdr h2_6; 5778 int terminal = 0; 5779 #endif /* INET6 */ 5780 int ipoff2 = 0; 5781 int off2 = 0; 5782 5783 pd2.af = pd->af; 5784 /* Payload packet is from the opposite direction. */ 5785 pd2.sidx = (direction == PF_IN) ? 1 : 0; 5786 pd2.didx = (direction == PF_IN) ? 0 : 1; 5787 switch (pd->af) { 5788 #ifdef INET 5789 case AF_INET: 5790 /* offset of h2 in mbuf chain */ 5791 ipoff2 = off + ICMP_MINLEN; 5792 5793 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 5794 NULL, reason, pd2.af)) { 5795 DPFPRINTF(PF_DEBUG_MISC, 5796 ("pf: ICMP error message too short " 5797 "(ip)\n")); 5798 return (PF_DROP); 5799 } 5800 /* 5801 * ICMP error messages don't refer to non-first 5802 * fragments 5803 */ 5804 if (h2.ip_off & htons(IP_OFFMASK)) { 5805 REASON_SET(reason, PFRES_FRAG); 5806 return (PF_DROP); 5807 } 5808 5809 /* offset of protocol header that follows h2 */ 5810 off2 = ipoff2 + (h2.ip_hl << 2); 5811 5812 pd2.proto = h2.ip_p; 5813 pd2.src = (struct pf_addr *)&h2.ip_src; 5814 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5815 pd2.ip_sum = &h2.ip_sum; 5816 break; 5817 #endif /* INET */ 5818 #ifdef INET6 5819 case AF_INET6: 5820 ipoff2 = off + sizeof(struct icmp6_hdr); 5821 5822 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 5823 NULL, reason, pd2.af)) { 5824 DPFPRINTF(PF_DEBUG_MISC, 5825 ("pf: ICMP error message too short " 5826 "(ip6)\n")); 5827 return (PF_DROP); 5828 } 5829 pd2.proto = h2_6.ip6_nxt; 5830 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5831 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5832 pd2.ip_sum = NULL; 5833 off2 = ipoff2 + sizeof(h2_6); 5834 do { 5835 switch (pd2.proto) { 5836 case IPPROTO_FRAGMENT: 5837 /* 5838 * ICMPv6 error messages for 5839 * non-first fragments 5840 */ 5841 REASON_SET(reason, PFRES_FRAG); 5842 return (PF_DROP); 5843 case IPPROTO_AH: 5844 case IPPROTO_HOPOPTS: 5845 case IPPROTO_ROUTING: 5846 case IPPROTO_DSTOPTS: { 5847 /* get next header and header length */ 5848 struct ip6_ext opt6; 5849 5850 if (!pf_pull_hdr(m, off2, &opt6, 5851 sizeof(opt6), NULL, reason, 5852 pd2.af)) { 5853 DPFPRINTF(PF_DEBUG_MISC, 5854 ("pf: ICMPv6 short opt\n")); 5855 return (PF_DROP); 5856 } 5857 if (pd2.proto == IPPROTO_AH) 5858 off2 += (opt6.ip6e_len + 2) * 4; 5859 else 5860 off2 += (opt6.ip6e_len + 1) * 8; 5861 pd2.proto = opt6.ip6e_nxt; 5862 /* goto the next header */ 5863 break; 5864 } 5865 default: 5866 terminal++; 5867 break; 5868 } 5869 } while (!terminal); 5870 break; 5871 #endif /* INET6 */ 5872 } 5873 5874 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5875 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5876 printf("pf: BAD ICMP %d:%d outer dst: ", 5877 icmptype, icmpcode); 5878 pf_print_host(pd->src, 0, pd->af); 5879 printf(" -> "); 5880 pf_print_host(pd->dst, 0, pd->af); 5881 printf(" inner src: "); 5882 pf_print_host(pd2.src, 0, pd2.af); 5883 printf(" -> "); 5884 pf_print_host(pd2.dst, 0, pd2.af); 5885 printf("\n"); 5886 } 5887 REASON_SET(reason, PFRES_BADSTATE); 5888 return (PF_DROP); 5889 } 5890 5891 switch (pd2.proto) { 5892 case IPPROTO_TCP: { 5893 struct tcphdr th; 5894 u_int32_t seq; 5895 struct pf_state_peer *src, *dst; 5896 u_int8_t dws; 5897 int copyback = 0; 5898 5899 /* 5900 * Only the first 8 bytes of the TCP header can be 5901 * expected. Don't access any TCP header fields after 5902 * th_seq, an ackskew test is not possible. 5903 */ 5904 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 5905 pd2.af)) { 5906 DPFPRINTF(PF_DEBUG_MISC, 5907 ("pf: ICMP error message too short " 5908 "(tcp)\n")); 5909 return (PF_DROP); 5910 } 5911 5912 key.af = pd2.af; 5913 key.proto = IPPROTO_TCP; 5914 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5915 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5916 key.port[pd2.sidx] = th.th_sport; 5917 key.port[pd2.didx] = th.th_dport; 5918 5919 STATE_LOOKUP(kif, &key, direction, *state, pd); 5920 5921 if (direction == (*state)->direction) { 5922 src = &(*state)->dst; 5923 dst = &(*state)->src; 5924 } else { 5925 src = &(*state)->src; 5926 dst = &(*state)->dst; 5927 } 5928 5929 if (src->wscale && dst->wscale) 5930 dws = dst->wscale & PF_WSCALE_MASK; 5931 else 5932 dws = 0; 5933 5934 /* Demodulate sequence number */ 5935 seq = ntohl(th.th_seq) - src->seqdiff; 5936 if (src->seqdiff) { 5937 pf_change_a(&th.th_seq, icmpsum, 5938 htonl(seq), 0); 5939 copyback = 1; 5940 } 5941 5942 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 5943 (!SEQ_GEQ(src->seqhi, seq) || 5944 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 5945 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5946 printf("pf: BAD ICMP %d:%d ", 5947 icmptype, icmpcode); 5948 pf_print_host(pd->src, 0, pd->af); 5949 printf(" -> "); 5950 pf_print_host(pd->dst, 0, pd->af); 5951 printf(" state: "); 5952 pf_print_state(*state); 5953 printf(" seq=%u\n", seq); 5954 } 5955 REASON_SET(reason, PFRES_BADSTATE); 5956 return (PF_DROP); 5957 } else { 5958 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5959 printf("pf: OK ICMP %d:%d ", 5960 icmptype, icmpcode); 5961 pf_print_host(pd->src, 0, pd->af); 5962 printf(" -> "); 5963 pf_print_host(pd->dst, 0, pd->af); 5964 printf(" state: "); 5965 pf_print_state(*state); 5966 printf(" seq=%u\n", seq); 5967 } 5968 } 5969 5970 /* translate source/destination address, if necessary */ 5971 if ((*state)->key[PF_SK_WIRE] != 5972 (*state)->key[PF_SK_STACK]) { 5973 struct pf_state_key *nk = 5974 (*state)->key[pd->didx]; 5975 5976 if (PF_ANEQ(pd2.src, 5977 &nk->addr[pd2.sidx], pd2.af) || 5978 nk->port[pd2.sidx] != th.th_sport) 5979 pf_change_icmp(pd2.src, &th.th_sport, 5980 daddr, &nk->addr[pd2.sidx], 5981 nk->port[pd2.sidx], NULL, 5982 pd2.ip_sum, icmpsum, 5983 pd->ip_sum, 0, pd2.af); 5984 5985 if (PF_ANEQ(pd2.dst, 5986 &nk->addr[pd2.didx], pd2.af) || 5987 nk->port[pd2.didx] != th.th_dport) 5988 pf_change_icmp(pd2.dst, &th.th_dport, 5989 saddr, &nk->addr[pd2.didx], 5990 nk->port[pd2.didx], NULL, 5991 pd2.ip_sum, icmpsum, 5992 pd->ip_sum, 0, pd2.af); 5993 copyback = 1; 5994 } 5995 5996 if (copyback) { 5997 switch (pd2.af) { 5998 #ifdef INET 5999 case AF_INET: 6000 m_copyback(m, off, ICMP_MINLEN, 6001 (caddr_t )&pd->hdr.icmp); 6002 m_copyback(m, ipoff2, sizeof(h2), 6003 (caddr_t )&h2); 6004 break; 6005 #endif /* INET */ 6006 #ifdef INET6 6007 case AF_INET6: 6008 m_copyback(m, off, 6009 sizeof(struct icmp6_hdr), 6010 (caddr_t )&pd->hdr.icmp6); 6011 m_copyback(m, ipoff2, sizeof(h2_6), 6012 (caddr_t )&h2_6); 6013 break; 6014 #endif /* INET6 */ 6015 } 6016 m_copyback(m, off2, 8, (caddr_t)&th); 6017 } 6018 6019 return (PF_PASS); 6020 break; 6021 } 6022 case IPPROTO_UDP: { 6023 struct udphdr uh; 6024 6025 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 6026 NULL, reason, pd2.af)) { 6027 DPFPRINTF(PF_DEBUG_MISC, 6028 ("pf: ICMP error message too short " 6029 "(udp)\n")); 6030 return (PF_DROP); 6031 } 6032 6033 key.af = pd2.af; 6034 key.proto = IPPROTO_UDP; 6035 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6036 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6037 key.port[pd2.sidx] = uh.uh_sport; 6038 key.port[pd2.didx] = uh.uh_dport; 6039 6040 STATE_LOOKUP(kif, &key, direction, *state, pd); 6041 6042 /* translate source/destination address, if necessary */ 6043 if ((*state)->key[PF_SK_WIRE] != 6044 (*state)->key[PF_SK_STACK]) { 6045 struct pf_state_key *nk = 6046 (*state)->key[pd->didx]; 6047 6048 if (PF_ANEQ(pd2.src, 6049 &nk->addr[pd2.sidx], pd2.af) || 6050 nk->port[pd2.sidx] != uh.uh_sport) 6051 pf_change_icmp(pd2.src, &uh.uh_sport, 6052 daddr, &nk->addr[pd2.sidx], 6053 nk->port[pd2.sidx], &uh.uh_sum, 6054 pd2.ip_sum, icmpsum, 6055 pd->ip_sum, 1, pd2.af); 6056 6057 if (PF_ANEQ(pd2.dst, 6058 &nk->addr[pd2.didx], pd2.af) || 6059 nk->port[pd2.didx] != uh.uh_dport) 6060 pf_change_icmp(pd2.dst, &uh.uh_dport, 6061 saddr, &nk->addr[pd2.didx], 6062 nk->port[pd2.didx], &uh.uh_sum, 6063 pd2.ip_sum, icmpsum, 6064 pd->ip_sum, 1, pd2.af); 6065 6066 switch (pd2.af) { 6067 #ifdef INET 6068 case AF_INET: 6069 m_copyback(m, off, ICMP_MINLEN, 6070 (caddr_t )&pd->hdr.icmp); 6071 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6072 break; 6073 #endif /* INET */ 6074 #ifdef INET6 6075 case AF_INET6: 6076 m_copyback(m, off, 6077 sizeof(struct icmp6_hdr), 6078 (caddr_t )&pd->hdr.icmp6); 6079 m_copyback(m, ipoff2, sizeof(h2_6), 6080 (caddr_t )&h2_6); 6081 break; 6082 #endif /* INET6 */ 6083 } 6084 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 6085 } 6086 return (PF_PASS); 6087 break; 6088 } 6089 #ifdef INET 6090 case IPPROTO_ICMP: { 6091 struct icmp iih; 6092 6093 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6094 NULL, reason, pd2.af)) { 6095 DPFPRINTF(PF_DEBUG_MISC, 6096 ("pf: ICMP error message too short i" 6097 "(icmp)\n")); 6098 return (PF_DROP); 6099 } 6100 6101 key.af = pd2.af; 6102 key.proto = IPPROTO_ICMP; 6103 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6104 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6105 key.port[0] = key.port[1] = iih.icmp_id; 6106 6107 STATE_LOOKUP(kif, &key, direction, *state, pd); 6108 6109 /* translate source/destination address, if necessary */ 6110 if ((*state)->key[PF_SK_WIRE] != 6111 (*state)->key[PF_SK_STACK]) { 6112 struct pf_state_key *nk = 6113 (*state)->key[pd->didx]; 6114 6115 if (PF_ANEQ(pd2.src, 6116 &nk->addr[pd2.sidx], pd2.af) || 6117 nk->port[pd2.sidx] != iih.icmp_id) 6118 pf_change_icmp(pd2.src, &iih.icmp_id, 6119 daddr, &nk->addr[pd2.sidx], 6120 nk->port[pd2.sidx], NULL, 6121 pd2.ip_sum, icmpsum, 6122 pd->ip_sum, 0, AF_INET); 6123 6124 if (PF_ANEQ(pd2.dst, 6125 &nk->addr[pd2.didx], pd2.af) || 6126 nk->port[pd2.didx] != iih.icmp_id) 6127 pf_change_icmp(pd2.dst, &iih.icmp_id, 6128 saddr, &nk->addr[pd2.didx], 6129 nk->port[pd2.didx], NULL, 6130 pd2.ip_sum, icmpsum, 6131 pd->ip_sum, 0, AF_INET); 6132 6133 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6134 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6135 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6136 } 6137 return (PF_PASS); 6138 break; 6139 } 6140 #endif /* INET */ 6141 #ifdef INET6 6142 case IPPROTO_ICMPV6: { 6143 struct icmp6_hdr iih; 6144 6145 if (!pf_pull_hdr(m, off2, &iih, 6146 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6147 DPFPRINTF(PF_DEBUG_MISC, 6148 ("pf: ICMP error message too short " 6149 "(icmp6)\n")); 6150 return (PF_DROP); 6151 } 6152 6153 key.af = pd2.af; 6154 key.proto = IPPROTO_ICMPV6; 6155 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6156 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6157 key.port[0] = key.port[1] = iih.icmp6_id; 6158 6159 STATE_LOOKUP(kif, &key, direction, *state, pd); 6160 6161 /* translate source/destination address, if necessary */ 6162 if ((*state)->key[PF_SK_WIRE] != 6163 (*state)->key[PF_SK_STACK]) { 6164 struct pf_state_key *nk = 6165 (*state)->key[pd->didx]; 6166 6167 if (PF_ANEQ(pd2.src, 6168 &nk->addr[pd2.sidx], pd2.af) || 6169 nk->port[pd2.sidx] != iih.icmp6_id) 6170 pf_change_icmp(pd2.src, &iih.icmp6_id, 6171 daddr, &nk->addr[pd2.sidx], 6172 nk->port[pd2.sidx], NULL, 6173 pd2.ip_sum, icmpsum, 6174 pd->ip_sum, 0, AF_INET6); 6175 6176 if (PF_ANEQ(pd2.dst, 6177 &nk->addr[pd2.didx], pd2.af) || 6178 nk->port[pd2.didx] != iih.icmp6_id) 6179 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6180 saddr, &nk->addr[pd2.didx], 6181 nk->port[pd2.didx], NULL, 6182 pd2.ip_sum, icmpsum, 6183 pd->ip_sum, 0, AF_INET6); 6184 6185 m_copyback(m, off, sizeof(struct icmp6_hdr), 6186 (caddr_t)&pd->hdr.icmp6); 6187 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6188 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6189 (caddr_t)&iih); 6190 } 6191 return (PF_PASS); 6192 break; 6193 } 6194 #endif /* INET6 */ 6195 default: { 6196 key.af = pd2.af; 6197 key.proto = pd2.proto; 6198 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6199 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6200 key.port[0] = key.port[1] = 0; 6201 6202 STATE_LOOKUP(kif, &key, direction, *state, pd); 6203 6204 /* translate source/destination address, if necessary */ 6205 if ((*state)->key[PF_SK_WIRE] != 6206 (*state)->key[PF_SK_STACK]) { 6207 struct pf_state_key *nk = 6208 (*state)->key[pd->didx]; 6209 6210 if (PF_ANEQ(pd2.src, 6211 &nk->addr[pd2.sidx], pd2.af)) 6212 pf_change_icmp(pd2.src, NULL, daddr, 6213 &nk->addr[pd2.sidx], 0, NULL, 6214 pd2.ip_sum, icmpsum, 6215 pd->ip_sum, 0, pd2.af); 6216 6217 if (PF_ANEQ(pd2.dst, 6218 &nk->addr[pd2.didx], pd2.af)) 6219 pf_change_icmp(pd2.dst, NULL, saddr, 6220 &nk->addr[pd2.didx], 0, NULL, 6221 pd2.ip_sum, icmpsum, 6222 pd->ip_sum, 0, pd2.af); 6223 6224 switch (pd2.af) { 6225 #ifdef INET 6226 case AF_INET: 6227 m_copyback(m, off, ICMP_MINLEN, 6228 (caddr_t)&pd->hdr.icmp); 6229 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6230 break; 6231 #endif /* INET */ 6232 #ifdef INET6 6233 case AF_INET6: 6234 m_copyback(m, off, 6235 sizeof(struct icmp6_hdr), 6236 (caddr_t )&pd->hdr.icmp6); 6237 m_copyback(m, ipoff2, sizeof(h2_6), 6238 (caddr_t )&h2_6); 6239 break; 6240 #endif /* INET6 */ 6241 } 6242 } 6243 return (PF_PASS); 6244 break; 6245 } 6246 } 6247 } 6248 } 6249 6250 static int 6251 pf_test_state_other(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 6252 struct mbuf *m, struct pf_pdesc *pd) 6253 { 6254 struct pf_state_peer *src, *dst; 6255 struct pf_state_key_cmp key; 6256 uint8_t psrc, pdst; 6257 6258 bzero(&key, sizeof(key)); 6259 key.af = pd->af; 6260 key.proto = pd->proto; 6261 if (direction == PF_IN) { 6262 PF_ACPY(&key.addr[0], pd->src, key.af); 6263 PF_ACPY(&key.addr[1], pd->dst, key.af); 6264 key.port[0] = key.port[1] = 0; 6265 } else { 6266 PF_ACPY(&key.addr[1], pd->src, key.af); 6267 PF_ACPY(&key.addr[0], pd->dst, key.af); 6268 key.port[1] = key.port[0] = 0; 6269 } 6270 6271 STATE_LOOKUP(kif, &key, direction, *state, pd); 6272 6273 if (direction == (*state)->direction) { 6274 src = &(*state)->src; 6275 dst = &(*state)->dst; 6276 psrc = PF_PEER_SRC; 6277 pdst = PF_PEER_DST; 6278 } else { 6279 src = &(*state)->dst; 6280 dst = &(*state)->src; 6281 psrc = PF_PEER_DST; 6282 pdst = PF_PEER_SRC; 6283 } 6284 6285 /* update states */ 6286 if (src->state < PFOTHERS_SINGLE) 6287 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 6288 if (dst->state == PFOTHERS_SINGLE) 6289 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 6290 6291 /* update expire time */ 6292 (*state)->expire = time_uptime; 6293 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 6294 (*state)->timeout = PFTM_OTHER_MULTIPLE; 6295 else 6296 (*state)->timeout = PFTM_OTHER_SINGLE; 6297 6298 /* translate source/destination address, if necessary */ 6299 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6300 struct pf_state_key *nk = (*state)->key[pd->didx]; 6301 6302 KASSERT(nk, ("%s: nk is null", __func__)); 6303 KASSERT(pd, ("%s: pd is null", __func__)); 6304 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 6305 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 6306 switch (pd->af) { 6307 #ifdef INET 6308 case AF_INET: 6309 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6310 pf_change_a(&pd->src->v4.s_addr, 6311 pd->ip_sum, 6312 nk->addr[pd->sidx].v4.s_addr, 6313 0); 6314 6315 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6316 pf_change_a(&pd->dst->v4.s_addr, 6317 pd->ip_sum, 6318 nk->addr[pd->didx].v4.s_addr, 6319 0); 6320 6321 break; 6322 #endif /* INET */ 6323 #ifdef INET6 6324 case AF_INET6: 6325 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6326 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 6327 6328 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6329 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 6330 #endif /* INET6 */ 6331 } 6332 } 6333 return (PF_PASS); 6334 } 6335 6336 /* 6337 * ipoff and off are measured from the start of the mbuf chain. 6338 * h must be at "ipoff" on the mbuf chain. 6339 */ 6340 void * 6341 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6342 u_short *actionp, u_short *reasonp, sa_family_t af) 6343 { 6344 switch (af) { 6345 #ifdef INET 6346 case AF_INET: { 6347 struct ip *h = mtod(m, struct ip *); 6348 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6349 6350 if (fragoff) { 6351 if (fragoff >= len) 6352 ACTION_SET(actionp, PF_PASS); 6353 else { 6354 ACTION_SET(actionp, PF_DROP); 6355 REASON_SET(reasonp, PFRES_FRAG); 6356 } 6357 return (NULL); 6358 } 6359 if (m->m_pkthdr.len < off + len || 6360 ntohs(h->ip_len) < off + len) { 6361 ACTION_SET(actionp, PF_DROP); 6362 REASON_SET(reasonp, PFRES_SHORT); 6363 return (NULL); 6364 } 6365 break; 6366 } 6367 #endif /* INET */ 6368 #ifdef INET6 6369 case AF_INET6: { 6370 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6371 6372 if (m->m_pkthdr.len < off + len || 6373 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 6374 (unsigned)(off + len)) { 6375 ACTION_SET(actionp, PF_DROP); 6376 REASON_SET(reasonp, PFRES_SHORT); 6377 return (NULL); 6378 } 6379 break; 6380 } 6381 #endif /* INET6 */ 6382 } 6383 m_copydata(m, off, len, p); 6384 return (p); 6385 } 6386 6387 int 6388 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 6389 int rtableid) 6390 { 6391 struct ifnet *ifp; 6392 6393 /* 6394 * Skip check for addresses with embedded interface scope, 6395 * as they would always match anyway. 6396 */ 6397 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 6398 return (1); 6399 6400 if (af != AF_INET && af != AF_INET6) 6401 return (0); 6402 6403 /* Skip checks for ipsec interfaces */ 6404 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6405 return (1); 6406 6407 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 6408 6409 switch (af) { 6410 #ifdef INET6 6411 case AF_INET6: 6412 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 6413 ifp)); 6414 #endif 6415 #ifdef INET 6416 case AF_INET: 6417 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 6418 ifp)); 6419 #endif 6420 } 6421 6422 return (0); 6423 } 6424 6425 #ifdef INET 6426 static void 6427 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 6428 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 6429 { 6430 struct mbuf *m0, *m1, *md; 6431 struct sockaddr_in dst; 6432 struct ip *ip; 6433 struct ifnet *ifp = NULL; 6434 struct pf_addr naddr; 6435 struct pf_ksrc_node *sn = NULL; 6436 int error = 0; 6437 uint16_t ip_len, ip_off; 6438 6439 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 6440 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 6441 __func__)); 6442 6443 if ((pd->pf_mtag == NULL && 6444 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 6445 pd->pf_mtag->routed++ > 3) { 6446 m0 = *m; 6447 *m = NULL; 6448 goto bad_locked; 6449 } 6450 6451 if (r->rt == PF_DUPTO) { 6452 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 6453 if (s == NULL) { 6454 ifp = r->rpool.cur->kif ? 6455 r->rpool.cur->kif->pfik_ifp : NULL; 6456 } else { 6457 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6458 /* If pfsync'd */ 6459 if (ifp == NULL) 6460 ifp = r->rpool.cur->kif ? 6461 r->rpool.cur->kif->pfik_ifp : NULL; 6462 PF_STATE_UNLOCK(s); 6463 } 6464 if (ifp == oifp) { 6465 /* When the 2nd interface is not skipped */ 6466 return; 6467 } else { 6468 m0 = *m; 6469 *m = NULL; 6470 goto bad; 6471 } 6472 } else { 6473 pd->pf_mtag->flags |= PF_DUPLICATED; 6474 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6475 if (s) 6476 PF_STATE_UNLOCK(s); 6477 return; 6478 } 6479 } 6480 } else { 6481 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6482 pf_dummynet(pd, dir, s, r, m); 6483 if (s) 6484 PF_STATE_UNLOCK(s); 6485 return; 6486 } 6487 m0 = *m; 6488 } 6489 6490 ip = mtod(m0, struct ip *); 6491 6492 bzero(&dst, sizeof(dst)); 6493 dst.sin_family = AF_INET; 6494 dst.sin_len = sizeof(dst); 6495 dst.sin_addr = ip->ip_dst; 6496 6497 bzero(&naddr, sizeof(naddr)); 6498 6499 if (s == NULL) { 6500 if (TAILQ_EMPTY(&r->rpool.list)) { 6501 DPFPRINTF(PF_DEBUG_URGENT, 6502 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 6503 goto bad_locked; 6504 } 6505 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 6506 &naddr, NULL, &sn); 6507 if (!PF_AZERO(&naddr, AF_INET)) 6508 dst.sin_addr.s_addr = naddr.v4.s_addr; 6509 ifp = r->rpool.cur->kif ? 6510 r->rpool.cur->kif->pfik_ifp : NULL; 6511 } else { 6512 if (!PF_AZERO(&s->rt_addr, AF_INET)) 6513 dst.sin_addr.s_addr = 6514 s->rt_addr.v4.s_addr; 6515 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6516 PF_STATE_UNLOCK(s); 6517 } 6518 /* If pfsync'd */ 6519 if (ifp == NULL) 6520 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6521 if (ifp == NULL) 6522 goto bad; 6523 6524 if (dir == PF_IN) { 6525 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) 6526 goto bad; 6527 else if (m0 == NULL) 6528 goto done; 6529 if (m0->m_len < sizeof(struct ip)) { 6530 DPFPRINTF(PF_DEBUG_URGENT, 6531 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 6532 goto bad; 6533 } 6534 ip = mtod(m0, struct ip *); 6535 } 6536 6537 if (ifp->if_flags & IFF_LOOPBACK) 6538 m0->m_flags |= M_SKIP_FIREWALL; 6539 6540 ip_len = ntohs(ip->ip_len); 6541 ip_off = ntohs(ip->ip_off); 6542 6543 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 6544 m0->m_pkthdr.csum_flags |= CSUM_IP; 6545 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 6546 in_delayed_cksum(m0); 6547 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 6548 } 6549 #if defined(SCTP) || defined(SCTP_SUPPORT) 6550 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 6551 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2)); 6552 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 6553 } 6554 #endif 6555 6556 /* 6557 * If small enough for interface, or the interface will take 6558 * care of the fragmentation for us, we can just send directly. 6559 */ 6560 if (ip_len <= ifp->if_mtu || 6561 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 6562 ip->ip_sum = 0; 6563 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 6564 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 6565 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 6566 } 6567 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 6568 6569 md = m0; 6570 error = pf_dummynet_route(pd, dir, s, r, ifp, sintosa(&dst), &md); 6571 if (md != NULL) 6572 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 6573 goto done; 6574 } 6575 6576 /* Balk when DF bit is set or the interface didn't support TSO. */ 6577 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 6578 error = EMSGSIZE; 6579 KMOD_IPSTAT_INC(ips_cantfrag); 6580 if (r->rt != PF_DUPTO) { 6581 if (s && pd->nat_rule != NULL) 6582 PACKET_UNDO_NAT(m0, pd, 6583 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 6584 s, dir); 6585 6586 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 6587 ifp->if_mtu); 6588 goto done; 6589 } else 6590 goto bad; 6591 } 6592 6593 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 6594 if (error) 6595 goto bad; 6596 6597 for (; m0; m0 = m1) { 6598 m1 = m0->m_nextpkt; 6599 m0->m_nextpkt = NULL; 6600 if (error == 0) { 6601 m_clrprotoflags(m0); 6602 md = m0; 6603 error = pf_dummynet_route(pd, dir, s, r, ifp, 6604 sintosa(&dst), &md); 6605 if (md != NULL) 6606 error = (*ifp->if_output)(ifp, md, 6607 sintosa(&dst), NULL); 6608 } else 6609 m_freem(m0); 6610 } 6611 6612 if (error == 0) 6613 KMOD_IPSTAT_INC(ips_fragmented); 6614 6615 done: 6616 if (r->rt != PF_DUPTO) 6617 *m = NULL; 6618 return; 6619 6620 bad_locked: 6621 if (s) 6622 PF_STATE_UNLOCK(s); 6623 bad: 6624 m_freem(m0); 6625 goto done; 6626 } 6627 #endif /* INET */ 6628 6629 #ifdef INET6 6630 static void 6631 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 6632 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 6633 { 6634 struct mbuf *m0, *md; 6635 struct sockaddr_in6 dst; 6636 struct ip6_hdr *ip6; 6637 struct ifnet *ifp = NULL; 6638 struct pf_addr naddr; 6639 struct pf_ksrc_node *sn = NULL; 6640 6641 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 6642 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 6643 __func__)); 6644 6645 if ((pd->pf_mtag == NULL && 6646 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 6647 pd->pf_mtag->routed++ > 3) { 6648 m0 = *m; 6649 *m = NULL; 6650 goto bad_locked; 6651 } 6652 6653 if (r->rt == PF_DUPTO) { 6654 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 6655 if (s == NULL) { 6656 ifp = r->rpool.cur->kif ? 6657 r->rpool.cur->kif->pfik_ifp : NULL; 6658 } else { 6659 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6660 /* If pfsync'd */ 6661 if (ifp == NULL) 6662 ifp = r->rpool.cur->kif ? 6663 r->rpool.cur->kif->pfik_ifp : NULL; 6664 PF_STATE_UNLOCK(s); 6665 } 6666 if (ifp == oifp) { 6667 /* When the 2nd interface is not skipped */ 6668 return; 6669 } else { 6670 m0 = *m; 6671 *m = NULL; 6672 goto bad; 6673 } 6674 } else { 6675 pd->pf_mtag->flags |= PF_DUPLICATED; 6676 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6677 if (s) 6678 PF_STATE_UNLOCK(s); 6679 return; 6680 } 6681 } 6682 } else { 6683 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6684 pf_dummynet(pd, dir, s, r, m); 6685 if (s) 6686 PF_STATE_UNLOCK(s); 6687 return; 6688 } 6689 m0 = *m; 6690 } 6691 6692 ip6 = mtod(m0, struct ip6_hdr *); 6693 6694 bzero(&dst, sizeof(dst)); 6695 dst.sin6_family = AF_INET6; 6696 dst.sin6_len = sizeof(dst); 6697 dst.sin6_addr = ip6->ip6_dst; 6698 6699 bzero(&naddr, sizeof(naddr)); 6700 6701 if (s == NULL) { 6702 if (TAILQ_EMPTY(&r->rpool.list)) { 6703 DPFPRINTF(PF_DEBUG_URGENT, 6704 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 6705 goto bad_locked; 6706 } 6707 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 6708 &naddr, NULL, &sn); 6709 if (!PF_AZERO(&naddr, AF_INET6)) 6710 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6711 &naddr, AF_INET6); 6712 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6713 } else { 6714 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 6715 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6716 &s->rt_addr, AF_INET6); 6717 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6718 } 6719 6720 if (s) 6721 PF_STATE_UNLOCK(s); 6722 6723 /* If pfsync'd */ 6724 if (ifp == NULL) 6725 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6726 if (ifp == NULL) 6727 goto bad; 6728 6729 if (dir == PF_IN) { 6730 if (pf_test6(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) 6731 goto bad; 6732 else if (m0 == NULL) 6733 goto done; 6734 if (m0->m_len < sizeof(struct ip6_hdr)) { 6735 DPFPRINTF(PF_DEBUG_URGENT, 6736 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 6737 __func__)); 6738 goto bad; 6739 } 6740 ip6 = mtod(m0, struct ip6_hdr *); 6741 } 6742 6743 if (ifp->if_flags & IFF_LOOPBACK) 6744 m0->m_flags |= M_SKIP_FIREWALL; 6745 6746 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 6747 ~ifp->if_hwassist) { 6748 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 6749 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 6750 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 6751 } 6752 6753 /* 6754 * If the packet is too large for the outgoing interface, 6755 * send back an icmp6 error. 6756 */ 6757 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 6758 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 6759 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 6760 md = m0; 6761 pf_dummynet_route(pd, dir, s, r, ifp, sintosa(&dst), &md); 6762 if (md != NULL) 6763 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 6764 } 6765 else { 6766 in6_ifstat_inc(ifp, ifs6_in_toobig); 6767 if (r->rt != PF_DUPTO) { 6768 if (s && pd->nat_rule != NULL) 6769 PACKET_UNDO_NAT(m0, pd, 6770 ((caddr_t)ip6 - m0->m_data) + 6771 sizeof(struct ip6_hdr), s, dir); 6772 6773 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 6774 } else 6775 goto bad; 6776 } 6777 6778 done: 6779 if (r->rt != PF_DUPTO) 6780 *m = NULL; 6781 return; 6782 6783 bad_locked: 6784 if (s) 6785 PF_STATE_UNLOCK(s); 6786 bad: 6787 m_freem(m0); 6788 goto done; 6789 } 6790 #endif /* INET6 */ 6791 6792 /* 6793 * FreeBSD supports cksum offloads for the following drivers. 6794 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 6795 * 6796 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 6797 * network driver performed cksum including pseudo header, need to verify 6798 * csum_data 6799 * CSUM_DATA_VALID : 6800 * network driver performed cksum, needs to additional pseudo header 6801 * cksum computation with partial csum_data(i.e. lack of H/W support for 6802 * pseudo header, for instance sk(4) and possibly gem(4)) 6803 * 6804 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 6805 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 6806 * TCP/UDP layer. 6807 * Also, set csum_data to 0xffff to force cksum validation. 6808 */ 6809 static int 6810 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 6811 { 6812 u_int16_t sum = 0; 6813 int hw_assist = 0; 6814 struct ip *ip; 6815 6816 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 6817 return (1); 6818 if (m->m_pkthdr.len < off + len) 6819 return (1); 6820 6821 switch (p) { 6822 case IPPROTO_TCP: 6823 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6824 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6825 sum = m->m_pkthdr.csum_data; 6826 } else { 6827 ip = mtod(m, struct ip *); 6828 sum = in_pseudo(ip->ip_src.s_addr, 6829 ip->ip_dst.s_addr, htonl((u_short)len + 6830 m->m_pkthdr.csum_data + IPPROTO_TCP)); 6831 } 6832 sum ^= 0xffff; 6833 ++hw_assist; 6834 } 6835 break; 6836 case IPPROTO_UDP: 6837 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6838 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6839 sum = m->m_pkthdr.csum_data; 6840 } else { 6841 ip = mtod(m, struct ip *); 6842 sum = in_pseudo(ip->ip_src.s_addr, 6843 ip->ip_dst.s_addr, htonl((u_short)len + 6844 m->m_pkthdr.csum_data + IPPROTO_UDP)); 6845 } 6846 sum ^= 0xffff; 6847 ++hw_assist; 6848 } 6849 break; 6850 case IPPROTO_ICMP: 6851 #ifdef INET6 6852 case IPPROTO_ICMPV6: 6853 #endif /* INET6 */ 6854 break; 6855 default: 6856 return (1); 6857 } 6858 6859 if (!hw_assist) { 6860 switch (af) { 6861 case AF_INET: 6862 if (p == IPPROTO_ICMP) { 6863 if (m->m_len < off) 6864 return (1); 6865 m->m_data += off; 6866 m->m_len -= off; 6867 sum = in_cksum(m, len); 6868 m->m_data -= off; 6869 m->m_len += off; 6870 } else { 6871 if (m->m_len < sizeof(struct ip)) 6872 return (1); 6873 sum = in4_cksum(m, p, off, len); 6874 } 6875 break; 6876 #ifdef INET6 6877 case AF_INET6: 6878 if (m->m_len < sizeof(struct ip6_hdr)) 6879 return (1); 6880 sum = in6_cksum(m, p, off, len); 6881 break; 6882 #endif /* INET6 */ 6883 default: 6884 return (1); 6885 } 6886 } 6887 if (sum) { 6888 switch (p) { 6889 case IPPROTO_TCP: 6890 { 6891 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 6892 break; 6893 } 6894 case IPPROTO_UDP: 6895 { 6896 KMOD_UDPSTAT_INC(udps_badsum); 6897 break; 6898 } 6899 #ifdef INET 6900 case IPPROTO_ICMP: 6901 { 6902 KMOD_ICMPSTAT_INC(icps_checksum); 6903 break; 6904 } 6905 #endif 6906 #ifdef INET6 6907 case IPPROTO_ICMPV6: 6908 { 6909 KMOD_ICMP6STAT_INC(icp6s_checksum); 6910 break; 6911 } 6912 #endif /* INET6 */ 6913 } 6914 return (1); 6915 } else { 6916 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 6917 m->m_pkthdr.csum_flags |= 6918 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 6919 m->m_pkthdr.csum_data = 0xffff; 6920 } 6921 } 6922 return (0); 6923 } 6924 6925 static bool 6926 pf_pdesc_to_dnflow(int dir, const struct pf_pdesc *pd, 6927 const struct pf_krule *r, const struct pf_kstate *s, 6928 struct ip_fw_args *dnflow) 6929 { 6930 int dndir = r->direction; 6931 6932 if (s && dndir == PF_INOUT) { 6933 dndir = s->direction; 6934 } else if (dndir == PF_INOUT) { 6935 /* Assume primary direction. Happens when we've set dnpipe in 6936 * the ethernet level code. */ 6937 dndir = dir; 6938 } 6939 6940 memset(dnflow, 0, sizeof(*dnflow)); 6941 6942 if (pd->dport != NULL) 6943 dnflow->f_id.dst_port = ntohs(*pd->dport); 6944 if (pd->sport != NULL) 6945 dnflow->f_id.src_port = ntohs(*pd->sport); 6946 6947 if (dir == PF_IN) 6948 dnflow->flags |= IPFW_ARGS_IN; 6949 else 6950 dnflow->flags |= IPFW_ARGS_OUT; 6951 6952 if (dir != dndir && pd->act.dnrpipe) { 6953 dnflow->rule.info = pd->act.dnrpipe; 6954 } 6955 else if (dir == dndir && pd->act.dnpipe) { 6956 dnflow->rule.info = pd->act.dnpipe; 6957 } 6958 else { 6959 return (false); 6960 } 6961 6962 dnflow->rule.info |= IPFW_IS_DUMMYNET; 6963 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 6964 dnflow->rule.info |= IPFW_IS_PIPE; 6965 6966 dnflow->f_id.proto = pd->proto; 6967 dnflow->f_id.extra = dnflow->rule.info; 6968 switch (pd->af) { 6969 case AF_INET: 6970 dnflow->f_id.addr_type = 4; 6971 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 6972 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 6973 break; 6974 case AF_INET6: 6975 dnflow->flags |= IPFW_ARGS_IP6; 6976 dnflow->f_id.addr_type = 6; 6977 dnflow->f_id.src_ip6 = pd->src->v6; 6978 dnflow->f_id.dst_ip6 = pd->dst->v6; 6979 break; 6980 default: 6981 panic("Invalid AF"); 6982 break; 6983 } 6984 6985 return (true); 6986 } 6987 6988 int 6989 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 6990 struct inpcb *inp) 6991 { 6992 struct pfi_kkif *kif; 6993 struct mbuf *m = *m0; 6994 6995 M_ASSERTPKTHDR(m); 6996 MPASS(ifp->if_vnet == curvnet); 6997 NET_EPOCH_ASSERT(); 6998 6999 if (!V_pf_status.running) 7000 return (PF_PASS); 7001 7002 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7003 7004 if (kif == NULL) { 7005 DPFPRINTF(PF_DEBUG_URGENT, 7006 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7007 return (PF_DROP); 7008 } 7009 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7010 return (PF_PASS); 7011 7012 if (m->m_flags & M_SKIP_FIREWALL) 7013 return (PF_PASS); 7014 7015 /* Stateless! */ 7016 return (pf_test_eth_rule(dir, kif, m0)); 7017 } 7018 7019 static int 7020 pf_dummynet(struct pf_pdesc *pd, int dir, struct pf_kstate *s, 7021 struct pf_krule *r, struct mbuf **m0) 7022 { 7023 return (pf_dummynet_route(pd, dir, s, r, NULL, NULL, m0)); 7024 } 7025 7026 static int 7027 pf_dummynet_route(struct pf_pdesc *pd, int dir, struct pf_kstate *s, 7028 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 7029 struct mbuf **m0) 7030 { 7031 NET_EPOCH_ASSERT(); 7032 7033 if (s && (s->dnpipe || s->dnrpipe)) { 7034 pd->act.dnpipe = s->dnpipe; 7035 pd->act.dnrpipe = s->dnrpipe; 7036 pd->act.flags = s->state_flags; 7037 } else if (r->dnpipe || r->dnrpipe) { 7038 pd->act.dnpipe = r->dnpipe; 7039 pd->act.dnrpipe = r->dnrpipe; 7040 pd->act.flags = r->free_flags; 7041 } 7042 if (pd->act.dnpipe || pd->act.dnrpipe) { 7043 struct ip_fw_args dnflow; 7044 if (ip_dn_io_ptr == NULL) { 7045 m_freem(*m0); 7046 *m0 = NULL; 7047 return (ENOMEM); 7048 } 7049 7050 if (pd->pf_mtag == NULL && 7051 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 7052 m_freem(*m0); 7053 *m0 = NULL; 7054 return (ENOMEM); 7055 } 7056 7057 if (ifp != NULL) { 7058 pd->pf_mtag->flags |= PF_TAG_ROUTE_TO; 7059 7060 pd->pf_mtag->if_index = ifp->if_index; 7061 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 7062 7063 MPASS(sa != NULL); 7064 7065 if (pd->af == AF_INET) 7066 memcpy(&pd->pf_mtag->dst, sa, 7067 sizeof(struct sockaddr_in)); 7068 else 7069 memcpy(&pd->pf_mtag->dst, sa, 7070 sizeof(struct sockaddr_in6)); 7071 } 7072 7073 if (pf_pdesc_to_dnflow(dir, pd, r, s, &dnflow)) { 7074 pd->pf_mtag->flags |= PF_TAG_DUMMYNET; 7075 ip_dn_io_ptr(m0, &dnflow); 7076 if (*m0 != NULL) { 7077 pd->pf_mtag->flags &= ~PF_TAG_ROUTE_TO; 7078 pd->pf_mtag->flags &= ~PF_TAG_DUMMYNET; 7079 } 7080 } 7081 } 7082 7083 return (0); 7084 } 7085 7086 #ifdef INET 7087 int 7088 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 7089 { 7090 struct pfi_kkif *kif; 7091 u_short action, reason = 0, log = 0; 7092 struct mbuf *m = *m0; 7093 struct ip *h = NULL; 7094 struct m_tag *ipfwtag; 7095 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7096 struct pf_kstate *s = NULL; 7097 struct pf_kruleset *ruleset = NULL; 7098 struct pf_pdesc pd; 7099 int off, dirndx, pqid = 0; 7100 7101 PF_RULES_RLOCK_TRACKER; 7102 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7103 M_ASSERTPKTHDR(m); 7104 7105 if (!V_pf_status.running) 7106 return (PF_PASS); 7107 7108 PF_RULES_RLOCK(); 7109 7110 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7111 7112 if (__predict_false(kif == NULL)) { 7113 DPFPRINTF(PF_DEBUG_URGENT, 7114 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7115 PF_RULES_RUNLOCK(); 7116 return (PF_DROP); 7117 } 7118 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7119 PF_RULES_RUNLOCK(); 7120 return (PF_PASS); 7121 } 7122 7123 if (m->m_flags & M_SKIP_FIREWALL) { 7124 PF_RULES_RUNLOCK(); 7125 return (PF_PASS); 7126 } 7127 7128 memset(&pd, 0, sizeof(pd)); 7129 pd.pf_mtag = pf_find_mtag(m); 7130 7131 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_TAG_ROUTE_TO)) { 7132 pd.pf_mtag->flags &= ~PF_TAG_ROUTE_TO; 7133 7134 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7135 pd.pf_mtag->if_idxgen); 7136 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7137 PF_RULES_RUNLOCK(); 7138 m_freem(*m0); 7139 *m0 = NULL; 7140 return (PF_PASS); 7141 } 7142 PF_RULES_RUNLOCK(); 7143 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 7144 *m0 = NULL; 7145 return (PF_PASS); 7146 } 7147 7148 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7149 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7150 pd.act.flags = pd.pf_mtag->dnflags; 7151 } 7152 7153 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7154 pd.pf_mtag->flags & PF_TAG_DUMMYNET) { 7155 /* Dummynet re-injects packets after they've 7156 * completed their delay. We've already 7157 * processed them, so pass unconditionally. */ 7158 7159 /* But only once. We may see the packet multiple times (e.g. 7160 * PFIL_IN/PFIL_OUT). */ 7161 pd.pf_mtag->flags &= ~PF_TAG_DUMMYNET; 7162 PF_RULES_RUNLOCK(); 7163 7164 return (PF_PASS); 7165 } 7166 7167 if (__predict_false(ip_divert_ptr != NULL) && 7168 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 7169 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 7170 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 7171 if (pd.pf_mtag == NULL && 7172 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7173 action = PF_DROP; 7174 goto done; 7175 } 7176 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 7177 m_tag_delete(m, ipfwtag); 7178 } 7179 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 7180 m->m_flags |= M_FASTFWD_OURS; 7181 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 7182 } 7183 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 7184 /* We do IP header normalization and packet reassembly here */ 7185 action = PF_DROP; 7186 goto done; 7187 } 7188 m = *m0; /* pf_normalize messes with m0 */ 7189 h = mtod(m, struct ip *); 7190 7191 off = h->ip_hl << 2; 7192 if (off < (int)sizeof(struct ip)) { 7193 action = PF_DROP; 7194 REASON_SET(&reason, PFRES_SHORT); 7195 log = PF_LOG_FORCE; 7196 goto done; 7197 } 7198 7199 pd.src = (struct pf_addr *)&h->ip_src; 7200 pd.dst = (struct pf_addr *)&h->ip_dst; 7201 pd.sport = pd.dport = NULL; 7202 pd.ip_sum = &h->ip_sum; 7203 pd.proto_sum = NULL; 7204 pd.proto = h->ip_p; 7205 pd.dir = dir; 7206 pd.sidx = (dir == PF_IN) ? 0 : 1; 7207 pd.didx = (dir == PF_IN) ? 1 : 0; 7208 pd.af = AF_INET; 7209 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 7210 pd.tot_len = ntohs(h->ip_len); 7211 pd.act.rtableid = -1; 7212 7213 /* handle fragments that didn't get reassembled by normalization */ 7214 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 7215 action = pf_test_fragment(&r, dir, kif, m, h, 7216 &pd, &a, &ruleset); 7217 goto done; 7218 } 7219 7220 switch (h->ip_p) { 7221 case IPPROTO_TCP: { 7222 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 7223 &action, &reason, AF_INET)) { 7224 if (action != PF_PASS) 7225 log = PF_LOG_FORCE; 7226 goto done; 7227 } 7228 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 7229 7230 pd.sport = &pd.hdr.tcp.th_sport; 7231 pd.dport = &pd.hdr.tcp.th_dport; 7232 7233 /* Respond to SYN with a syncookie. */ 7234 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 7235 pd.dir == PF_IN && pf_synflood_check(&pd)) { 7236 pf_syncookie_send(m, off, &pd); 7237 action = PF_DROP; 7238 break; 7239 } 7240 7241 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7242 pqid = 1; 7243 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 7244 if (action == PF_DROP) 7245 goto done; 7246 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 7247 &reason); 7248 if (action == PF_PASS) { 7249 if (V_pfsync_update_state_ptr != NULL) 7250 V_pfsync_update_state_ptr(s); 7251 r = s->rule.ptr; 7252 a = s->anchor.ptr; 7253 log = s->log; 7254 } else if (s == NULL) { 7255 /* Validate remote SYN|ACK, re-create original SYN if 7256 * valid. */ 7257 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 7258 TH_ACK && pf_syncookie_validate(&pd) && 7259 pd.dir == PF_IN) { 7260 struct mbuf *msyn; 7261 7262 msyn = pf_syncookie_recreate_syn(h->ip_ttl, 7263 off,&pd); 7264 if (msyn == NULL) { 7265 action = PF_DROP; 7266 break; 7267 } 7268 7269 action = pf_test(dir, pflags, ifp, &msyn, inp); 7270 m_freem(msyn); 7271 7272 if (action == PF_PASS) { 7273 action = pf_test_state_tcp(&s, dir, 7274 kif, m, off, h, &pd, &reason); 7275 if (action != PF_PASS || s == NULL) { 7276 action = PF_DROP; 7277 break; 7278 } 7279 7280 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) 7281 - 1; 7282 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) 7283 - 1; 7284 pf_set_protostate(s, PF_PEER_SRC, 7285 PF_TCPS_PROXY_DST); 7286 7287 action = pf_synproxy(&pd, &s, &reason); 7288 if (action != PF_PASS) 7289 break; 7290 } 7291 break; 7292 } 7293 else { 7294 action = pf_test_rule(&r, &s, dir, kif, m, off, 7295 &pd, &a, &ruleset, inp); 7296 } 7297 } 7298 if (s) { 7299 if (s->max_mss) 7300 pf_normalize_mss(m, off, &pd, s->max_mss); 7301 } else if (r->max_mss) 7302 pf_normalize_mss(m, off, &pd, r->max_mss); 7303 break; 7304 } 7305 7306 case IPPROTO_UDP: { 7307 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 7308 &action, &reason, AF_INET)) { 7309 if (action != PF_PASS) 7310 log = PF_LOG_FORCE; 7311 goto done; 7312 } 7313 pd.sport = &pd.hdr.udp.uh_sport; 7314 pd.dport = &pd.hdr.udp.uh_dport; 7315 if (pd.hdr.udp.uh_dport == 0 || 7316 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 7317 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 7318 action = PF_DROP; 7319 REASON_SET(&reason, PFRES_SHORT); 7320 goto done; 7321 } 7322 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7323 if (action == PF_PASS) { 7324 if (V_pfsync_update_state_ptr != NULL) 7325 V_pfsync_update_state_ptr(s); 7326 r = s->rule.ptr; 7327 a = s->anchor.ptr; 7328 log = s->log; 7329 } else if (s == NULL) 7330 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7331 &a, &ruleset, inp); 7332 break; 7333 } 7334 7335 case IPPROTO_ICMP: { 7336 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 7337 &action, &reason, AF_INET)) { 7338 if (action != PF_PASS) 7339 log = PF_LOG_FORCE; 7340 goto done; 7341 } 7342 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 7343 &reason); 7344 if (action == PF_PASS) { 7345 if (V_pfsync_update_state_ptr != NULL) 7346 V_pfsync_update_state_ptr(s); 7347 r = s->rule.ptr; 7348 a = s->anchor.ptr; 7349 log = s->log; 7350 } else if (s == NULL) 7351 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7352 &a, &ruleset, inp); 7353 break; 7354 } 7355 7356 #ifdef INET6 7357 case IPPROTO_ICMPV6: { 7358 action = PF_DROP; 7359 DPFPRINTF(PF_DEBUG_MISC, 7360 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 7361 goto done; 7362 } 7363 #endif 7364 7365 default: 7366 action = pf_test_state_other(&s, dir, kif, m, &pd); 7367 if (action == PF_PASS) { 7368 if (V_pfsync_update_state_ptr != NULL) 7369 V_pfsync_update_state_ptr(s); 7370 r = s->rule.ptr; 7371 a = s->anchor.ptr; 7372 log = s->log; 7373 } else if (s == NULL) 7374 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7375 &a, &ruleset, inp); 7376 break; 7377 } 7378 7379 done: 7380 PF_RULES_RUNLOCK(); 7381 if (action == PF_PASS && h->ip_hl > 5 && 7382 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7383 action = PF_DROP; 7384 REASON_SET(&reason, PFRES_IPOPTIONS); 7385 log = PF_LOG_FORCE; 7386 DPFPRINTF(PF_DEBUG_MISC, 7387 ("pf: dropping packet with ip options\n")); 7388 } 7389 7390 if (s) { 7391 pf_scrub_ip(&m, s->state_flags, s->min_ttl, s->set_tos); 7392 if (s->rtableid >= 0) 7393 M_SETFIB(m, s->rtableid); 7394 #ifdef ALTQ 7395 if (s->qid) { 7396 pd.act.pqid = s->pqid; 7397 pd.act.qid = s->qid; 7398 } 7399 #endif 7400 } else { 7401 pf_scrub_ip(&m, r->scrub_flags, r->min_ttl, r->set_tos); 7402 if (r->rtableid >= 0) 7403 M_SETFIB(m, r->rtableid); 7404 #ifdef ALTQ 7405 if (r->qid) { 7406 pd.act.pqid = r->pqid; 7407 pd.act.qid = r->qid; 7408 } 7409 #endif 7410 } 7411 7412 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 7413 action = PF_DROP; 7414 REASON_SET(&reason, PFRES_MEMORY); 7415 } 7416 7417 if (r->scrub_flags & PFSTATE_SETPRIO) { 7418 if (pd.tos & IPTOS_LOWDELAY) 7419 pqid = 1; 7420 if (vlan_set_pcp(m, r->set_prio[pqid])) { 7421 action = PF_DROP; 7422 REASON_SET(&reason, PFRES_MEMORY); 7423 log = PF_LOG_FORCE; 7424 DPFPRINTF(PF_DEBUG_MISC, 7425 ("pf: failed to allocate 802.1q mtag\n")); 7426 } 7427 } 7428 7429 #ifdef ALTQ 7430 if (action == PF_PASS && pd.act.qid) { 7431 if (pd.pf_mtag == NULL && 7432 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7433 action = PF_DROP; 7434 REASON_SET(&reason, PFRES_MEMORY); 7435 } else { 7436 if (s != NULL) 7437 pd.pf_mtag->qid_hash = pf_state_hash(s); 7438 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 7439 pd.pf_mtag->qid = pd.act.pqid; 7440 else 7441 pd.pf_mtag->qid = pd.act.qid; 7442 /* Add hints for ecn. */ 7443 pd.pf_mtag->hdr = h; 7444 } 7445 } 7446 #endif /* ALTQ */ 7447 7448 /* 7449 * connections redirected to loopback should not match sockets 7450 * bound specifically to loopback due to security implications, 7451 * see tcp_input() and in_pcblookup_listen(). 7452 */ 7453 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7454 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7455 (s->nat_rule.ptr->action == PF_RDR || 7456 s->nat_rule.ptr->action == PF_BINAT) && 7457 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 7458 m->m_flags |= M_SKIP_FIREWALL; 7459 7460 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 7461 r->divert.port && !PACKET_LOOPED(&pd)) { 7462 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 7463 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 7464 if (ipfwtag != NULL) { 7465 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 7466 ntohs(r->divert.port); 7467 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 7468 7469 if (s) 7470 PF_STATE_UNLOCK(s); 7471 7472 m_tag_prepend(m, ipfwtag); 7473 if (m->m_flags & M_FASTFWD_OURS) { 7474 if (pd.pf_mtag == NULL && 7475 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7476 action = PF_DROP; 7477 REASON_SET(&reason, PFRES_MEMORY); 7478 log = PF_LOG_FORCE; 7479 DPFPRINTF(PF_DEBUG_MISC, 7480 ("pf: failed to allocate tag\n")); 7481 } else { 7482 pd.pf_mtag->flags |= 7483 PF_FASTFWD_OURS_PRESENT; 7484 m->m_flags &= ~M_FASTFWD_OURS; 7485 } 7486 } 7487 ip_divert_ptr(*m0, dir == PF_IN); 7488 *m0 = NULL; 7489 7490 return (action); 7491 } else { 7492 /* XXX: ipfw has the same behaviour! */ 7493 action = PF_DROP; 7494 REASON_SET(&reason, PFRES_MEMORY); 7495 log = PF_LOG_FORCE; 7496 DPFPRINTF(PF_DEBUG_MISC, 7497 ("pf: failed to allocate divert tag\n")); 7498 } 7499 } 7500 7501 if (log) { 7502 struct pf_krule *lr; 7503 struct pf_krule_item *ri; 7504 7505 if (s != NULL && s->nat_rule.ptr != NULL && 7506 s->nat_rule.ptr->log & PF_LOG_ALL) 7507 lr = s->nat_rule.ptr; 7508 else 7509 lr = r; 7510 7511 if (log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 7512 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, 7513 ruleset, &pd, (s == NULL)); 7514 if (s) { 7515 SLIST_FOREACH(ri, &s->match_rules, entry) 7516 if (ri->r->log & PF_LOG_ALL) 7517 PFLOG_PACKET(kif, m, AF_INET, dir, 7518 reason, ri->r, a, ruleset, &pd, 0); 7519 } 7520 } 7521 7522 pf_counter_u64_critical_enter(); 7523 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 7524 pd.tot_len); 7525 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 7526 1); 7527 7528 if (action == PF_PASS || r->action == PF_DROP) { 7529 dirndx = (dir == PF_OUT); 7530 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 7531 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 7532 pf_update_timestamp(r); 7533 7534 if (a != NULL) { 7535 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 7536 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 7537 } 7538 if (s != NULL) { 7539 struct pf_krule_item *ri; 7540 7541 if (s->nat_rule.ptr != NULL) { 7542 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 7543 1); 7544 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 7545 pd.tot_len); 7546 } 7547 if (s->src_node != NULL) { 7548 counter_u64_add(s->src_node->packets[dirndx], 7549 1); 7550 counter_u64_add(s->src_node->bytes[dirndx], 7551 pd.tot_len); 7552 } 7553 if (s->nat_src_node != NULL) { 7554 counter_u64_add(s->nat_src_node->packets[dirndx], 7555 1); 7556 counter_u64_add(s->nat_src_node->bytes[dirndx], 7557 pd.tot_len); 7558 } 7559 dirndx = (dir == s->direction) ? 0 : 1; 7560 s->packets[dirndx]++; 7561 s->bytes[dirndx] += pd.tot_len; 7562 SLIST_FOREACH(ri, &s->match_rules, entry) { 7563 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 7564 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 7565 } 7566 } 7567 tr = r; 7568 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7569 if (nr != NULL && r == &V_pf_default_rule) 7570 tr = nr; 7571 if (tr->src.addr.type == PF_ADDR_TABLE) 7572 pfr_update_stats(tr->src.addr.p.tbl, 7573 (s == NULL) ? pd.src : 7574 &s->key[(s->direction == PF_IN)]-> 7575 addr[(s->direction == PF_OUT)], 7576 pd.af, pd.tot_len, dir == PF_OUT, 7577 r->action == PF_PASS, tr->src.neg); 7578 if (tr->dst.addr.type == PF_ADDR_TABLE) 7579 pfr_update_stats(tr->dst.addr.p.tbl, 7580 (s == NULL) ? pd.dst : 7581 &s->key[(s->direction == PF_IN)]-> 7582 addr[(s->direction == PF_IN)], 7583 pd.af, pd.tot_len, dir == PF_OUT, 7584 r->action == PF_PASS, tr->dst.neg); 7585 } 7586 pf_counter_u64_critical_exit(); 7587 7588 switch (action) { 7589 case PF_SYNPROXY_DROP: 7590 m_freem(*m0); 7591 case PF_DEFER: 7592 *m0 = NULL; 7593 action = PF_PASS; 7594 break; 7595 case PF_DROP: 7596 m_freem(*m0); 7597 *m0 = NULL; 7598 break; 7599 default: 7600 /* pf_route() returns unlocked. */ 7601 if (r->rt) { 7602 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 7603 return (action); 7604 } 7605 if (pf_dummynet(&pd, dir, s, r, m0) != 0) { 7606 action = PF_DROP; 7607 REASON_SET(&reason, PFRES_MEMORY); 7608 } 7609 break; 7610 } 7611 7612 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 7613 7614 if (s) 7615 PF_STATE_UNLOCK(s); 7616 7617 return (action); 7618 } 7619 #endif /* INET */ 7620 7621 #ifdef INET6 7622 int 7623 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 7624 { 7625 struct pfi_kkif *kif; 7626 u_short action, reason = 0, log = 0; 7627 struct mbuf *m = *m0, *n = NULL; 7628 struct m_tag *mtag; 7629 struct ip6_hdr *h = NULL; 7630 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7631 struct pf_kstate *s = NULL; 7632 struct pf_kruleset *ruleset = NULL; 7633 struct pf_pdesc pd; 7634 int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; 7635 7636 PF_RULES_RLOCK_TRACKER; 7637 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7638 M_ASSERTPKTHDR(m); 7639 7640 if (!V_pf_status.running) 7641 return (PF_PASS); 7642 7643 PF_RULES_RLOCK(); 7644 7645 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7646 if (__predict_false(kif == NULL)) { 7647 DPFPRINTF(PF_DEBUG_URGENT, 7648 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 7649 PF_RULES_RUNLOCK(); 7650 return (PF_DROP); 7651 } 7652 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7653 PF_RULES_RUNLOCK(); 7654 return (PF_PASS); 7655 } 7656 7657 if (m->m_flags & M_SKIP_FIREWALL) { 7658 PF_RULES_RUNLOCK(); 7659 return (PF_PASS); 7660 } 7661 7662 memset(&pd, 0, sizeof(pd)); 7663 pd.pf_mtag = pf_find_mtag(m); 7664 7665 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_TAG_ROUTE_TO)) { 7666 pd.pf_mtag->flags &= ~PF_TAG_ROUTE_TO; 7667 7668 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7669 pd.pf_mtag->if_idxgen); 7670 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7671 PF_RULES_RUNLOCK(); 7672 m_freem(*m0); 7673 *m0 = NULL; 7674 return (PF_PASS); 7675 } 7676 PF_RULES_RUNLOCK(); 7677 nd6_output_ifp(ifp, ifp, m, 7678 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 7679 *m0 = NULL; 7680 return (PF_PASS); 7681 } 7682 7683 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7684 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7685 pd.act.flags = pd.pf_mtag->dnflags; 7686 } 7687 7688 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7689 pd.pf_mtag->flags & PF_TAG_DUMMYNET) { 7690 pd.pf_mtag->flags &= ~PF_TAG_DUMMYNET; 7691 /* Dummynet re-injects packets after they've 7692 * completed their delay. We've already 7693 * processed them, so pass unconditionally. */ 7694 PF_RULES_RUNLOCK(); 7695 return (PF_PASS); 7696 } 7697 7698 /* We do IP header normalization and packet reassembly here */ 7699 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 7700 action = PF_DROP; 7701 goto done; 7702 } 7703 m = *m0; /* pf_normalize messes with m0 */ 7704 h = mtod(m, struct ip6_hdr *); 7705 7706 /* 7707 * we do not support jumbogram. if we keep going, zero ip6_plen 7708 * will do something bad, so drop the packet for now. 7709 */ 7710 if (htons(h->ip6_plen) == 0) { 7711 action = PF_DROP; 7712 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 7713 goto done; 7714 } 7715 7716 pd.src = (struct pf_addr *)&h->ip6_src; 7717 pd.dst = (struct pf_addr *)&h->ip6_dst; 7718 pd.sport = pd.dport = NULL; 7719 pd.ip_sum = NULL; 7720 pd.proto_sum = NULL; 7721 pd.dir = dir; 7722 pd.sidx = (dir == PF_IN) ? 0 : 1; 7723 pd.didx = (dir == PF_IN) ? 1 : 0; 7724 pd.af = AF_INET6; 7725 pd.tos = IPV6_DSCP(h); 7726 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 7727 pd.act.rtableid = -1; 7728 7729 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 7730 pd.proto = h->ip6_nxt; 7731 do { 7732 switch (pd.proto) { 7733 case IPPROTO_FRAGMENT: 7734 action = pf_test_fragment(&r, dir, kif, m, h, 7735 &pd, &a, &ruleset); 7736 if (action == PF_DROP) 7737 REASON_SET(&reason, PFRES_FRAG); 7738 goto done; 7739 case IPPROTO_ROUTING: { 7740 struct ip6_rthdr rthdr; 7741 7742 if (rh_cnt++) { 7743 DPFPRINTF(PF_DEBUG_MISC, 7744 ("pf: IPv6 more than one rthdr\n")); 7745 action = PF_DROP; 7746 REASON_SET(&reason, PFRES_IPOPTIONS); 7747 log = PF_LOG_FORCE; 7748 goto done; 7749 } 7750 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 7751 &reason, pd.af)) { 7752 DPFPRINTF(PF_DEBUG_MISC, 7753 ("pf: IPv6 short rthdr\n")); 7754 action = PF_DROP; 7755 REASON_SET(&reason, PFRES_SHORT); 7756 log = PF_LOG_FORCE; 7757 goto done; 7758 } 7759 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 7760 DPFPRINTF(PF_DEBUG_MISC, 7761 ("pf: IPv6 rthdr0\n")); 7762 action = PF_DROP; 7763 REASON_SET(&reason, PFRES_IPOPTIONS); 7764 log = PF_LOG_FORCE; 7765 goto done; 7766 } 7767 /* FALLTHROUGH */ 7768 } 7769 case IPPROTO_AH: 7770 case IPPROTO_HOPOPTS: 7771 case IPPROTO_DSTOPTS: { 7772 /* get next header and header length */ 7773 struct ip6_ext opt6; 7774 7775 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 7776 NULL, &reason, pd.af)) { 7777 DPFPRINTF(PF_DEBUG_MISC, 7778 ("pf: IPv6 short opt\n")); 7779 action = PF_DROP; 7780 log = PF_LOG_FORCE; 7781 goto done; 7782 } 7783 if (pd.proto == IPPROTO_AH) 7784 off += (opt6.ip6e_len + 2) * 4; 7785 else 7786 off += (opt6.ip6e_len + 1) * 8; 7787 pd.proto = opt6.ip6e_nxt; 7788 /* goto the next header */ 7789 break; 7790 } 7791 default: 7792 terminal++; 7793 break; 7794 } 7795 } while (!terminal); 7796 7797 /* if there's no routing header, use unmodified mbuf for checksumming */ 7798 if (!n) 7799 n = m; 7800 7801 switch (pd.proto) { 7802 case IPPROTO_TCP: { 7803 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 7804 &action, &reason, AF_INET6)) { 7805 if (action != PF_PASS) 7806 log |= PF_LOG_FORCE; 7807 goto done; 7808 } 7809 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 7810 pd.sport = &pd.hdr.tcp.th_sport; 7811 pd.dport = &pd.hdr.tcp.th_dport; 7812 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 7813 if (action == PF_DROP) 7814 goto done; 7815 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 7816 &reason); 7817 if (action == PF_PASS) { 7818 if (V_pfsync_update_state_ptr != NULL) 7819 V_pfsync_update_state_ptr(s); 7820 r = s->rule.ptr; 7821 a = s->anchor.ptr; 7822 log = s->log; 7823 } else if (s == NULL) 7824 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7825 &a, &ruleset, inp); 7826 if (s) { 7827 if (s->max_mss) 7828 pf_normalize_mss(m, off, &pd, s->max_mss); 7829 } else if (r->max_mss) 7830 pf_normalize_mss(m, off, &pd, r->max_mss); 7831 break; 7832 } 7833 7834 case IPPROTO_UDP: { 7835 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 7836 &action, &reason, AF_INET6)) { 7837 if (action != PF_PASS) 7838 log |= PF_LOG_FORCE; 7839 goto done; 7840 } 7841 pd.sport = &pd.hdr.udp.uh_sport; 7842 pd.dport = &pd.hdr.udp.uh_dport; 7843 if (pd.hdr.udp.uh_dport == 0 || 7844 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 7845 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 7846 action = PF_DROP; 7847 REASON_SET(&reason, PFRES_SHORT); 7848 goto done; 7849 } 7850 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7851 if (action == PF_PASS) { 7852 if (V_pfsync_update_state_ptr != NULL) 7853 V_pfsync_update_state_ptr(s); 7854 r = s->rule.ptr; 7855 a = s->anchor.ptr; 7856 log = s->log; 7857 } else if (s == NULL) 7858 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7859 &a, &ruleset, inp); 7860 break; 7861 } 7862 7863 case IPPROTO_ICMP: { 7864 action = PF_DROP; 7865 DPFPRINTF(PF_DEBUG_MISC, 7866 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 7867 goto done; 7868 } 7869 7870 case IPPROTO_ICMPV6: { 7871 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 7872 &action, &reason, AF_INET6)) { 7873 if (action != PF_PASS) 7874 log |= PF_LOG_FORCE; 7875 goto done; 7876 } 7877 action = pf_test_state_icmp(&s, dir, kif, 7878 m, off, h, &pd, &reason); 7879 if (action == PF_PASS) { 7880 if (V_pfsync_update_state_ptr != NULL) 7881 V_pfsync_update_state_ptr(s); 7882 r = s->rule.ptr; 7883 a = s->anchor.ptr; 7884 log = s->log; 7885 } else if (s == NULL) 7886 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7887 &a, &ruleset, inp); 7888 break; 7889 } 7890 7891 default: 7892 action = pf_test_state_other(&s, dir, kif, m, &pd); 7893 if (action == PF_PASS) { 7894 if (V_pfsync_update_state_ptr != NULL) 7895 V_pfsync_update_state_ptr(s); 7896 r = s->rule.ptr; 7897 a = s->anchor.ptr; 7898 log = s->log; 7899 } else if (s == NULL) 7900 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7901 &a, &ruleset, inp); 7902 break; 7903 } 7904 7905 done: 7906 PF_RULES_RUNLOCK(); 7907 if (n != m) { 7908 m_freem(n); 7909 n = NULL; 7910 } 7911 7912 /* handle dangerous IPv6 extension headers. */ 7913 if (action == PF_PASS && rh_cnt && 7914 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7915 action = PF_DROP; 7916 REASON_SET(&reason, PFRES_IPOPTIONS); 7917 log = r->log; 7918 DPFPRINTF(PF_DEBUG_MISC, 7919 ("pf: dropping packet with dangerous v6 headers\n")); 7920 } 7921 7922 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 7923 action = PF_DROP; 7924 REASON_SET(&reason, PFRES_MEMORY); 7925 } 7926 7927 if (s) { 7928 pf_scrub_ip6(&m, s->state_flags, s->min_ttl, s->set_tos); 7929 if (s->rtableid >= 0) 7930 M_SETFIB(m, s->rtableid); 7931 #ifdef ALTQ 7932 if (s->qid) { 7933 pd.act.pqid = s->pqid; 7934 pd.act.qid = s->qid; 7935 } 7936 #endif 7937 } else { 7938 pf_scrub_ip6(&m, r->scrub_flags, r->min_ttl, r->set_tos); 7939 if (r->rtableid >= 0) 7940 M_SETFIB(m, r->rtableid); 7941 #ifdef ALTQ 7942 if (r->qid) { 7943 pd.act.pqid = r->pqid; 7944 pd.act.qid = r->qid; 7945 } 7946 #endif 7947 } 7948 7949 if (r->scrub_flags & PFSTATE_SETPRIO) { 7950 if (pd.tos & IPTOS_LOWDELAY) 7951 pqid = 1; 7952 if (vlan_set_pcp(m, r->set_prio[pqid])) { 7953 action = PF_DROP; 7954 REASON_SET(&reason, PFRES_MEMORY); 7955 log = PF_LOG_FORCE; 7956 DPFPRINTF(PF_DEBUG_MISC, 7957 ("pf: failed to allocate 802.1q mtag\n")); 7958 } 7959 } 7960 7961 #ifdef ALTQ 7962 if (action == PF_PASS && pd.act.qid) { 7963 if (pd.pf_mtag == NULL && 7964 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7965 action = PF_DROP; 7966 REASON_SET(&reason, PFRES_MEMORY); 7967 } else { 7968 if (s != NULL) 7969 pd.pf_mtag->qid_hash = pf_state_hash(s); 7970 if (pd.tos & IPTOS_LOWDELAY) 7971 pd.pf_mtag->qid = pd.act.pqid; 7972 else 7973 pd.pf_mtag->qid = pd.act.qid; 7974 /* Add hints for ecn. */ 7975 pd.pf_mtag->hdr = h; 7976 } 7977 } 7978 #endif /* ALTQ */ 7979 7980 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7981 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7982 (s->nat_rule.ptr->action == PF_RDR || 7983 s->nat_rule.ptr->action == PF_BINAT) && 7984 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 7985 m->m_flags |= M_SKIP_FIREWALL; 7986 7987 /* XXX: Anybody working on it?! */ 7988 if (r->divert.port) 7989 printf("pf: divert(9) is not supported for IPv6\n"); 7990 7991 if (log) { 7992 struct pf_krule *lr; 7993 struct pf_krule_item *ri; 7994 7995 if (s != NULL && s->nat_rule.ptr != NULL && 7996 s->nat_rule.ptr->log & PF_LOG_ALL) 7997 lr = s->nat_rule.ptr; 7998 else 7999 lr = r; 8000 8001 if (log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8002 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, 8003 ruleset, &pd, (s == NULL)); 8004 if (s) { 8005 SLIST_FOREACH(ri, &s->match_rules, entry) 8006 if (ri->r->log & PF_LOG_ALL) 8007 PFLOG_PACKET(kif, m, AF_INET6, dir, 8008 reason, ri->r, a, ruleset, &pd, 0); 8009 } 8010 } 8011 8012 pf_counter_u64_critical_enter(); 8013 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 8014 pd.tot_len); 8015 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 8016 1); 8017 8018 if (action == PF_PASS || r->action == PF_DROP) { 8019 dirndx = (dir == PF_OUT); 8020 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8021 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8022 if (a != NULL) { 8023 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8024 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8025 } 8026 if (s != NULL) { 8027 if (s->nat_rule.ptr != NULL) { 8028 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8029 1); 8030 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8031 pd.tot_len); 8032 } 8033 if (s->src_node != NULL) { 8034 counter_u64_add(s->src_node->packets[dirndx], 8035 1); 8036 counter_u64_add(s->src_node->bytes[dirndx], 8037 pd.tot_len); 8038 } 8039 if (s->nat_src_node != NULL) { 8040 counter_u64_add(s->nat_src_node->packets[dirndx], 8041 1); 8042 counter_u64_add(s->nat_src_node->bytes[dirndx], 8043 pd.tot_len); 8044 } 8045 dirndx = (dir == s->direction) ? 0 : 1; 8046 s->packets[dirndx]++; 8047 s->bytes[dirndx] += pd.tot_len; 8048 } 8049 tr = r; 8050 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8051 if (nr != NULL && r == &V_pf_default_rule) 8052 tr = nr; 8053 if (tr->src.addr.type == PF_ADDR_TABLE) 8054 pfr_update_stats(tr->src.addr.p.tbl, 8055 (s == NULL) ? pd.src : 8056 &s->key[(s->direction == PF_IN)]->addr[0], 8057 pd.af, pd.tot_len, dir == PF_OUT, 8058 r->action == PF_PASS, tr->src.neg); 8059 if (tr->dst.addr.type == PF_ADDR_TABLE) 8060 pfr_update_stats(tr->dst.addr.p.tbl, 8061 (s == NULL) ? pd.dst : 8062 &s->key[(s->direction == PF_IN)]->addr[1], 8063 pd.af, pd.tot_len, dir == PF_OUT, 8064 r->action == PF_PASS, tr->dst.neg); 8065 } 8066 pf_counter_u64_critical_exit(); 8067 8068 switch (action) { 8069 case PF_SYNPROXY_DROP: 8070 m_freem(*m0); 8071 case PF_DEFER: 8072 *m0 = NULL; 8073 action = PF_PASS; 8074 break; 8075 case PF_DROP: 8076 m_freem(*m0); 8077 *m0 = NULL; 8078 break; 8079 default: 8080 /* pf_route6() returns unlocked. */ 8081 if (r->rt) { 8082 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 8083 return (action); 8084 } 8085 if (pf_dummynet(&pd, dir, s, r, m0) != 0) { 8086 action = PF_DROP; 8087 REASON_SET(&reason, PFRES_MEMORY); 8088 } 8089 break; 8090 } 8091 8092 if (s) 8093 PF_STATE_UNLOCK(s); 8094 8095 /* If reassembled packet passed, create new fragments. */ 8096 if (action == PF_PASS && *m0 && dir == PF_OUT && 8097 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 8098 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 8099 8100 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 8101 8102 return (action); 8103 } 8104 #endif /* INET6 */ 8105