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_types.h> 71 #include <net/if_vlan_var.h> 72 #include <net/route.h> 73 #include <net/route/nhop.h> 74 #include <net/vnet.h> 75 76 #include <net/pfil.h> 77 #include <net/pfvar.h> 78 #include <net/if_pflog.h> 79 #include <net/if_pfsync.h> 80 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_var.h> 83 #include <netinet/in_fib.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_fw.h> 86 #include <netinet/ip_icmp.h> 87 #include <netinet/icmp_var.h> 88 #include <netinet/ip_var.h> 89 #include <netinet/tcp.h> 90 #include <netinet/tcp_fsm.h> 91 #include <netinet/tcp_seq.h> 92 #include <netinet/tcp_timer.h> 93 #include <netinet/tcp_var.h> 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 97 /* dummynet */ 98 #include <netinet/ip_dummynet.h> 99 #include <netinet/ip_fw.h> 100 #include <netpfil/ipfw/dn_heap.h> 101 #include <netpfil/ipfw/ip_fw_private.h> 102 #include <netpfil/ipfw/ip_dn_private.h> 103 104 #ifdef INET6 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet6/in6_fib.h> 111 #include <netinet6/scope6_var.h> 112 #endif /* INET6 */ 113 114 #if defined(SCTP) || defined(SCTP_SUPPORT) 115 #include <netinet/sctp_crc32.h> 116 #endif 117 118 #include <machine/in_cksum.h> 119 #include <security/mac/mac_framework.h> 120 121 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 122 123 SDT_PROVIDER_DEFINE(pf); 124 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 127 "struct pf_kstate *"); 128 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 129 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 130 "struct pf_kstate *"); 131 132 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 133 "struct mbuf *"); 134 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 135 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 136 "int", "struct pf_keth_rule *", "char *"); 137 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 139 "int", "struct pf_keth_rule *"); 140 141 /* 142 * Global variables 143 */ 144 145 /* state tables */ 146 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 147 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 148 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 149 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 150 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 151 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 152 VNET_DEFINE(struct pf_kstatus, pf_status); 153 154 VNET_DEFINE(u_int32_t, ticket_altqs_active); 155 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 156 VNET_DEFINE(int, altqs_inactive_open); 157 VNET_DEFINE(u_int32_t, ticket_pabuf); 158 159 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 160 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 161 VNET_DEFINE(u_char, pf_tcp_secret[16]); 162 #define V_pf_tcp_secret VNET(pf_tcp_secret) 163 VNET_DEFINE(int, pf_tcp_secret_init); 164 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 165 VNET_DEFINE(int, pf_tcp_iss_off); 166 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 167 VNET_DECLARE(int, pf_vnet_active); 168 #define V_pf_vnet_active VNET(pf_vnet_active) 169 170 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 171 #define V_pf_purge_idx VNET(pf_purge_idx) 172 173 #ifdef PF_WANT_32_TO_64_COUNTER 174 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 175 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 176 177 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 178 VNET_DEFINE(size_t, pf_allrulecount); 179 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 180 #endif 181 182 /* 183 * Queue for pf_intr() sends. 184 */ 185 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 186 struct pf_send_entry { 187 STAILQ_ENTRY(pf_send_entry) pfse_next; 188 struct mbuf *pfse_m; 189 enum { 190 PFSE_IP, 191 PFSE_IP6, 192 PFSE_ICMP, 193 PFSE_ICMP6, 194 } pfse_type; 195 struct { 196 int type; 197 int code; 198 int mtu; 199 } icmpopts; 200 }; 201 202 STAILQ_HEAD(pf_send_head, pf_send_entry); 203 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 204 #define V_pf_sendqueue VNET(pf_sendqueue) 205 206 static struct mtx_padalign pf_sendqueue_mtx; 207 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 208 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 209 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 210 211 /* 212 * Queue for pf_overload_task() tasks. 213 */ 214 struct pf_overload_entry { 215 SLIST_ENTRY(pf_overload_entry) next; 216 struct pf_addr addr; 217 sa_family_t af; 218 uint8_t dir; 219 struct pf_krule *rule; 220 }; 221 222 SLIST_HEAD(pf_overload_head, pf_overload_entry); 223 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 224 #define V_pf_overloadqueue VNET(pf_overloadqueue) 225 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 226 #define V_pf_overloadtask VNET(pf_overloadtask) 227 228 static struct mtx_padalign pf_overloadqueue_mtx; 229 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 230 "pf overload/flush queue", MTX_DEF); 231 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 232 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 233 234 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 235 struct mtx_padalign pf_unlnkdrules_mtx; 236 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 237 MTX_DEF); 238 239 struct sx pf_config_lock; 240 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 241 242 struct mtx_padalign pf_table_stats_lock; 243 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 244 MTX_DEF); 245 246 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 247 #define V_pf_sources_z VNET(pf_sources_z) 248 uma_zone_t pf_mtag_z; 249 VNET_DEFINE(uma_zone_t, pf_state_z); 250 VNET_DEFINE(uma_zone_t, pf_state_key_z); 251 252 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 253 #define PFID_CPUBITS 8 254 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 255 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 256 #define PFID_MAXID (~PFID_CPUMASK) 257 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); 258 259 static void pf_src_tree_remove_state(struct pf_kstate *); 260 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 261 u_int32_t); 262 static void pf_add_threshold(struct pf_threshold *); 263 static int pf_check_threshold(struct pf_threshold *); 264 265 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 266 u_int16_t *, u_int16_t *, struct pf_addr *, 267 u_int16_t, u_int8_t, sa_family_t); 268 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 269 struct tcphdr *, struct pf_state_peer *); 270 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 271 struct pf_addr *, struct pf_addr *, u_int16_t, 272 u_int16_t *, u_int16_t *, u_int16_t *, 273 u_int16_t *, u_int8_t, sa_family_t); 274 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 275 sa_family_t, struct pf_krule *); 276 static void pf_detach_state(struct pf_kstate *); 277 static int pf_state_key_attach(struct pf_state_key *, 278 struct pf_state_key *, struct pf_kstate *); 279 static void pf_state_key_detach(struct pf_kstate *, int); 280 static int pf_state_key_ctor(void *, int, void *, int); 281 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 282 void pf_rule_to_actions(struct pf_krule *, 283 struct pf_rule_actions *); 284 static int pf_dummynet(struct pf_pdesc *, int, struct pf_kstate *, 285 struct pf_krule *, struct mbuf **); 286 static int pf_dummynet_route(struct pf_pdesc *, int, 287 struct pf_kstate *, struct pf_krule *, 288 struct ifnet *, struct sockaddr *, struct mbuf **); 289 static int pf_test_eth_rule(int, struct pfi_kkif *, 290 struct mbuf **); 291 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 292 int, struct pfi_kkif *, struct mbuf *, int, 293 struct pf_pdesc *, struct pf_krule **, 294 struct pf_kruleset **, struct inpcb *); 295 static int pf_create_state(struct pf_krule *, struct pf_krule *, 296 struct pf_krule *, struct pf_pdesc *, 297 struct pf_ksrc_node *, struct pf_state_key *, 298 struct pf_state_key *, struct mbuf *, int, 299 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 300 struct pf_kstate **, int, u_int16_t, u_int16_t, 301 int); 302 static int pf_test_fragment(struct pf_krule **, int, 303 struct pfi_kkif *, struct mbuf *, void *, 304 struct pf_pdesc *, struct pf_krule **, 305 struct pf_kruleset **); 306 static int pf_tcp_track_full(struct pf_kstate **, 307 struct pfi_kkif *, struct mbuf *, int, 308 struct pf_pdesc *, u_short *, int *); 309 static int pf_tcp_track_sloppy(struct pf_kstate **, 310 struct pf_pdesc *, u_short *); 311 static int pf_test_state_tcp(struct pf_kstate **, int, 312 struct pfi_kkif *, struct mbuf *, int, 313 void *, struct pf_pdesc *, u_short *); 314 static int pf_test_state_udp(struct pf_kstate **, int, 315 struct pfi_kkif *, struct mbuf *, int, 316 void *, struct pf_pdesc *); 317 static int pf_test_state_icmp(struct pf_kstate **, int, 318 struct pfi_kkif *, struct mbuf *, int, 319 void *, struct pf_pdesc *, u_short *); 320 static int pf_test_state_other(struct pf_kstate **, int, 321 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 322 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 323 int, u_int16_t); 324 static int pf_check_proto_cksum(struct mbuf *, int, int, 325 u_int8_t, sa_family_t); 326 static void pf_print_state_parts(struct pf_kstate *, 327 struct pf_state_key *, struct pf_state_key *); 328 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 329 struct pf_addr_wrap *); 330 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 331 bool, u_int8_t); 332 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 333 struct pf_state_key_cmp *, u_int); 334 static int pf_src_connlimit(struct pf_kstate **); 335 static void pf_overload_task(void *v, int pending); 336 static int pf_insert_src_node(struct pf_ksrc_node **, 337 struct pf_krule *, struct pf_addr *, sa_family_t); 338 static u_int pf_purge_expired_states(u_int, int); 339 static void pf_purge_unlinked_rules(void); 340 static int pf_mtag_uminit(void *, int, int); 341 static void pf_mtag_free(struct m_tag *); 342 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 343 int, struct pf_state_key *); 344 #ifdef INET 345 static void pf_route(struct mbuf **, struct pf_krule *, int, 346 struct ifnet *, struct pf_kstate *, 347 struct pf_pdesc *, struct inpcb *); 348 #endif /* INET */ 349 #ifdef INET6 350 static void pf_change_a6(struct pf_addr *, u_int16_t *, 351 struct pf_addr *, u_int8_t); 352 static void pf_route6(struct mbuf **, struct pf_krule *, int, 353 struct ifnet *, struct pf_kstate *, 354 struct pf_pdesc *, struct inpcb *); 355 #endif /* INET6 */ 356 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 357 358 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 359 360 extern int pf_end_threads; 361 extern struct proc *pf_purge_proc; 362 363 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 364 365 #define PACKET_UNDO_NAT(_m, _pd, _off, _s, _dir) \ 366 do { \ 367 struct pf_state_key *nk; \ 368 if ((_dir) == PF_OUT) \ 369 nk = (_s)->key[PF_SK_STACK]; \ 370 else \ 371 nk = (_s)->key[PF_SK_WIRE]; \ 372 pf_packet_rework_nat(_m, _pd, _off, nk); \ 373 } while (0) 374 375 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 376 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 377 378 #define STATE_LOOKUP(i, k, d, s, pd) \ 379 do { \ 380 (s) = pf_find_state((i), (k), (d)); \ 381 SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s)); \ 382 if ((s) == NULL) \ 383 return (PF_DROP); \ 384 if (PACKET_LOOPED(pd)) \ 385 return (PF_PASS); \ 386 } while (0) 387 388 #define BOUND_IFACE(r, k) \ 389 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 390 391 #define STATE_INC_COUNTERS(s) \ 392 do { \ 393 counter_u64_add(s->rule.ptr->states_cur, 1); \ 394 counter_u64_add(s->rule.ptr->states_tot, 1); \ 395 if (s->anchor.ptr != NULL) { \ 396 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 397 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 398 } \ 399 if (s->nat_rule.ptr != NULL) { \ 400 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 401 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 402 } \ 403 } while (0) 404 405 #define STATE_DEC_COUNTERS(s) \ 406 do { \ 407 if (s->nat_rule.ptr != NULL) \ 408 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 409 if (s->anchor.ptr != NULL) \ 410 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 411 counter_u64_add(s->rule.ptr->states_cur, -1); \ 412 } while (0) 413 414 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 415 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 416 VNET_DEFINE(struct pf_idhash *, pf_idhash); 417 VNET_DEFINE(struct pf_srchash *, pf_srchash); 418 419 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 420 "pf(4)"); 421 422 u_long pf_hashmask; 423 u_long pf_srchashmask; 424 static u_long pf_hashsize; 425 static u_long pf_srchashsize; 426 u_long pf_ioctl_maxcount = 65535; 427 428 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 429 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 430 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 431 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 432 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 433 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 434 435 VNET_DEFINE(void *, pf_swi_cookie); 436 VNET_DEFINE(struct intr_event *, pf_swi_ie); 437 438 VNET_DEFINE(uint32_t, pf_hashseed); 439 #define V_pf_hashseed VNET(pf_hashseed) 440 441 #ifdef __LP64__ 442 static int 443 pf_bcmp_state_key(struct pf_state_key *k1_orig, struct pf_state_key_cmp *k2_orig) 444 { 445 unsigned long *k1 = (unsigned long *)k1_orig; 446 unsigned long *k2 = (unsigned long *)k2_orig; 447 448 if (k1[0] != k2[0]) 449 return (1); 450 451 if (k1[1] != k2[1]) 452 return (1); 453 454 if (k1[2] != k2[2]) 455 return (1); 456 457 if (k1[3] != k2[3]) 458 return (1); 459 460 if (k1[4] != k2[4]) 461 return (1); 462 463 return (0); 464 } 465 _Static_assert(sizeof(struct pf_state_key_cmp) == 40, "bad size of pf_state_key_cmp"); 466 #else 467 static inline int 468 pf_bcmp_state_key(struct pf_state_key *k1_orig, struct pf_state_key_cmp *k2_orig) 469 { 470 471 return (bcmp(k1_orig, k2_orig, sizeof(struct pf_state_key_cmp))); 472 } 473 #endif 474 475 int 476 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 477 { 478 479 switch (af) { 480 #ifdef INET 481 case AF_INET: 482 if (a->addr32[0] > b->addr32[0]) 483 return (1); 484 if (a->addr32[0] < b->addr32[0]) 485 return (-1); 486 break; 487 #endif /* INET */ 488 #ifdef INET6 489 case AF_INET6: 490 if (a->addr32[3] > b->addr32[3]) 491 return (1); 492 if (a->addr32[3] < b->addr32[3]) 493 return (-1); 494 if (a->addr32[2] > b->addr32[2]) 495 return (1); 496 if (a->addr32[2] < b->addr32[2]) 497 return (-1); 498 if (a->addr32[1] > b->addr32[1]) 499 return (1); 500 if (a->addr32[1] < b->addr32[1]) 501 return (-1); 502 if (a->addr32[0] > b->addr32[0]) 503 return (1); 504 if (a->addr32[0] < b->addr32[0]) 505 return (-1); 506 break; 507 #endif /* INET6 */ 508 default: 509 panic("%s: unknown address family %u", __func__, af); 510 } 511 return (0); 512 } 513 514 static void 515 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 516 struct pf_state_key *nk) 517 { 518 519 switch (pd->proto) { 520 case IPPROTO_TCP: { 521 struct tcphdr *th = &pd->hdr.tcp; 522 523 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 524 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 525 &th->th_sum, &nk->addr[pd->sidx], 526 nk->port[pd->sidx], 0, pd->af); 527 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 528 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 529 &th->th_sum, &nk->addr[pd->didx], 530 nk->port[pd->didx], 0, pd->af); 531 m_copyback(m, off, sizeof(*th), (caddr_t)th); 532 break; 533 } 534 case IPPROTO_UDP: { 535 struct udphdr *uh = &pd->hdr.udp; 536 537 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 538 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 539 &uh->uh_sum, &nk->addr[pd->sidx], 540 nk->port[pd->sidx], 1, pd->af); 541 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 542 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 543 &uh->uh_sum, &nk->addr[pd->didx], 544 nk->port[pd->didx], 1, pd->af); 545 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 546 break; 547 } 548 case IPPROTO_ICMP: { 549 struct icmp *ih = &pd->hdr.icmp; 550 551 if (nk->port[pd->sidx] != ih->icmp_id) { 552 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 553 ih->icmp_cksum, ih->icmp_id, 554 nk->port[pd->sidx], 0); 555 ih->icmp_id = nk->port[pd->sidx]; 556 pd->sport = &ih->icmp_id; 557 558 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 559 } 560 /* FALLTHROUGH */ 561 } 562 default: 563 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 564 switch (pd->af) { 565 case AF_INET: 566 pf_change_a(&pd->src->v4.s_addr, 567 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 568 0); 569 break; 570 case AF_INET6: 571 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 572 break; 573 } 574 } 575 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 576 switch (pd->af) { 577 case AF_INET: 578 pf_change_a(&pd->dst->v4.s_addr, 579 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 580 0); 581 break; 582 case AF_INET6: 583 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 584 break; 585 } 586 } 587 break; 588 } 589 } 590 591 static __inline uint32_t 592 pf_hashkey(struct pf_state_key *sk) 593 { 594 uint32_t h; 595 596 h = murmur3_32_hash32((uint32_t *)sk, 597 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 598 V_pf_hashseed); 599 600 return (h & pf_hashmask); 601 } 602 603 static __inline uint32_t 604 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 605 { 606 uint32_t h; 607 608 switch (af) { 609 case AF_INET: 610 h = murmur3_32_hash32((uint32_t *)&addr->v4, 611 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 612 break; 613 case AF_INET6: 614 h = murmur3_32_hash32((uint32_t *)&addr->v6, 615 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 616 break; 617 default: 618 panic("%s: unknown address family %u", __func__, af); 619 } 620 621 return (h & pf_srchashmask); 622 } 623 624 #ifdef ALTQ 625 static int 626 pf_state_hash(struct pf_kstate *s) 627 { 628 u_int32_t hv = (intptr_t)s / sizeof(*s); 629 630 hv ^= crc32(&s->src, sizeof(s->src)); 631 hv ^= crc32(&s->dst, sizeof(s->dst)); 632 if (hv == 0) 633 hv = 1; 634 return (hv); 635 } 636 #endif 637 638 static __inline void 639 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 640 { 641 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 642 s->dst.state = newstate; 643 if (which == PF_PEER_DST) 644 return; 645 if (s->src.state == newstate) 646 return; 647 if (s->creatorid == V_pf_status.hostid && 648 s->key[PF_SK_STACK] != NULL && 649 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 650 !(TCPS_HAVEESTABLISHED(s->src.state) || 651 s->src.state == TCPS_CLOSED) && 652 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 653 atomic_add_32(&V_pf_status.states_halfopen, -1); 654 655 s->src.state = newstate; 656 } 657 658 #ifdef INET6 659 void 660 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 661 { 662 switch (af) { 663 #ifdef INET 664 case AF_INET: 665 dst->addr32[0] = src->addr32[0]; 666 break; 667 #endif /* INET */ 668 case AF_INET6: 669 dst->addr32[0] = src->addr32[0]; 670 dst->addr32[1] = src->addr32[1]; 671 dst->addr32[2] = src->addr32[2]; 672 dst->addr32[3] = src->addr32[3]; 673 break; 674 } 675 } 676 #endif /* INET6 */ 677 678 static void 679 pf_init_threshold(struct pf_threshold *threshold, 680 u_int32_t limit, u_int32_t seconds) 681 { 682 threshold->limit = limit * PF_THRESHOLD_MULT; 683 threshold->seconds = seconds; 684 threshold->count = 0; 685 threshold->last = time_uptime; 686 } 687 688 static void 689 pf_add_threshold(struct pf_threshold *threshold) 690 { 691 u_int32_t t = time_uptime, diff = t - threshold->last; 692 693 if (diff >= threshold->seconds) 694 threshold->count = 0; 695 else 696 threshold->count -= threshold->count * diff / 697 threshold->seconds; 698 threshold->count += PF_THRESHOLD_MULT; 699 threshold->last = t; 700 } 701 702 static int 703 pf_check_threshold(struct pf_threshold *threshold) 704 { 705 return (threshold->count > threshold->limit); 706 } 707 708 static int 709 pf_src_connlimit(struct pf_kstate **state) 710 { 711 struct pf_overload_entry *pfoe; 712 int bad = 0; 713 714 PF_STATE_LOCK_ASSERT(*state); 715 716 (*state)->src_node->conn++; 717 (*state)->src.tcp_est = 1; 718 pf_add_threshold(&(*state)->src_node->conn_rate); 719 720 if ((*state)->rule.ptr->max_src_conn && 721 (*state)->rule.ptr->max_src_conn < 722 (*state)->src_node->conn) { 723 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 724 bad++; 725 } 726 727 if ((*state)->rule.ptr->max_src_conn_rate.limit && 728 pf_check_threshold(&(*state)->src_node->conn_rate)) { 729 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 730 bad++; 731 } 732 733 if (!bad) 734 return (0); 735 736 /* Kill this state. */ 737 (*state)->timeout = PFTM_PURGE; 738 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 739 740 if ((*state)->rule.ptr->overload_tbl == NULL) 741 return (1); 742 743 /* Schedule overloading and flushing task. */ 744 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 745 if (pfoe == NULL) 746 return (1); /* too bad :( */ 747 748 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 749 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 750 pfoe->rule = (*state)->rule.ptr; 751 pfoe->dir = (*state)->direction; 752 PF_OVERLOADQ_LOCK(); 753 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 754 PF_OVERLOADQ_UNLOCK(); 755 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 756 757 return (1); 758 } 759 760 static void 761 pf_overload_task(void *v, int pending) 762 { 763 struct pf_overload_head queue; 764 struct pfr_addr p; 765 struct pf_overload_entry *pfoe, *pfoe1; 766 uint32_t killed = 0; 767 768 CURVNET_SET((struct vnet *)v); 769 770 PF_OVERLOADQ_LOCK(); 771 queue = V_pf_overloadqueue; 772 SLIST_INIT(&V_pf_overloadqueue); 773 PF_OVERLOADQ_UNLOCK(); 774 775 bzero(&p, sizeof(p)); 776 SLIST_FOREACH(pfoe, &queue, next) { 777 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 778 if (V_pf_status.debug >= PF_DEBUG_MISC) { 779 printf("%s: blocking address ", __func__); 780 pf_print_host(&pfoe->addr, 0, pfoe->af); 781 printf("\n"); 782 } 783 784 p.pfra_af = pfoe->af; 785 switch (pfoe->af) { 786 #ifdef INET 787 case AF_INET: 788 p.pfra_net = 32; 789 p.pfra_ip4addr = pfoe->addr.v4; 790 break; 791 #endif 792 #ifdef INET6 793 case AF_INET6: 794 p.pfra_net = 128; 795 p.pfra_ip6addr = pfoe->addr.v6; 796 break; 797 #endif 798 } 799 800 PF_RULES_WLOCK(); 801 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 802 PF_RULES_WUNLOCK(); 803 } 804 805 /* 806 * Remove those entries, that don't need flushing. 807 */ 808 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 809 if (pfoe->rule->flush == 0) { 810 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 811 free(pfoe, M_PFTEMP); 812 } else 813 counter_u64_add( 814 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 815 816 /* If nothing to flush, return. */ 817 if (SLIST_EMPTY(&queue)) { 818 CURVNET_RESTORE(); 819 return; 820 } 821 822 for (int i = 0; i <= pf_hashmask; i++) { 823 struct pf_idhash *ih = &V_pf_idhash[i]; 824 struct pf_state_key *sk; 825 struct pf_kstate *s; 826 827 PF_HASHROW_LOCK(ih); 828 LIST_FOREACH(s, &ih->states, entry) { 829 sk = s->key[PF_SK_WIRE]; 830 SLIST_FOREACH(pfoe, &queue, next) 831 if (sk->af == pfoe->af && 832 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 833 pfoe->rule == s->rule.ptr) && 834 ((pfoe->dir == PF_OUT && 835 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 836 (pfoe->dir == PF_IN && 837 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 838 s->timeout = PFTM_PURGE; 839 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 840 killed++; 841 } 842 } 843 PF_HASHROW_UNLOCK(ih); 844 } 845 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 846 free(pfoe, M_PFTEMP); 847 if (V_pf_status.debug >= PF_DEBUG_MISC) 848 printf("%s: %u states killed", __func__, killed); 849 850 CURVNET_RESTORE(); 851 } 852 853 /* 854 * Can return locked on failure, so that we can consistently 855 * allocate and insert a new one. 856 */ 857 struct pf_ksrc_node * 858 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 859 int returnlocked) 860 { 861 struct pf_srchash *sh; 862 struct pf_ksrc_node *n; 863 864 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 865 866 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 867 PF_HASHROW_LOCK(sh); 868 LIST_FOREACH(n, &sh->nodes, entry) 869 if (n->rule.ptr == rule && n->af == af && 870 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 871 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 872 break; 873 if (n != NULL) { 874 n->states++; 875 PF_HASHROW_UNLOCK(sh); 876 } else if (returnlocked == 0) 877 PF_HASHROW_UNLOCK(sh); 878 879 return (n); 880 } 881 882 static void 883 pf_free_src_node(struct pf_ksrc_node *sn) 884 { 885 886 for (int i = 0; i < 2; i++) { 887 counter_u64_free(sn->bytes[i]); 888 counter_u64_free(sn->packets[i]); 889 } 890 uma_zfree(V_pf_sources_z, sn); 891 } 892 893 static int 894 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 895 struct pf_addr *src, sa_family_t af) 896 { 897 898 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 899 rule->rpool.opts & PF_POOL_STICKYADDR), 900 ("%s for non-tracking rule %p", __func__, rule)); 901 902 if (*sn == NULL) 903 *sn = pf_find_src_node(src, rule, af, 1); 904 905 if (*sn == NULL) { 906 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 907 908 PF_HASHROW_ASSERT(sh); 909 910 if (!rule->max_src_nodes || 911 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 912 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 913 else 914 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 915 1); 916 if ((*sn) == NULL) { 917 PF_HASHROW_UNLOCK(sh); 918 return (-1); 919 } 920 921 for (int i = 0; i < 2; i++) { 922 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 923 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 924 925 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 926 pf_free_src_node(*sn); 927 PF_HASHROW_UNLOCK(sh); 928 return (-1); 929 } 930 } 931 932 pf_init_threshold(&(*sn)->conn_rate, 933 rule->max_src_conn_rate.limit, 934 rule->max_src_conn_rate.seconds); 935 936 (*sn)->af = af; 937 (*sn)->rule.ptr = rule; 938 PF_ACPY(&(*sn)->addr, src, af); 939 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 940 (*sn)->creation = time_uptime; 941 (*sn)->ruletype = rule->action; 942 (*sn)->states = 1; 943 if ((*sn)->rule.ptr != NULL) 944 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 945 PF_HASHROW_UNLOCK(sh); 946 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 947 } else { 948 if (rule->max_src_states && 949 (*sn)->states >= rule->max_src_states) { 950 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 951 1); 952 return (-1); 953 } 954 } 955 return (0); 956 } 957 958 void 959 pf_unlink_src_node(struct pf_ksrc_node *src) 960 { 961 962 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); 963 LIST_REMOVE(src, entry); 964 if (src->rule.ptr) 965 counter_u64_add(src->rule.ptr->src_nodes, -1); 966 } 967 968 u_int 969 pf_free_src_nodes(struct pf_ksrc_node_list *head) 970 { 971 struct pf_ksrc_node *sn, *tmp; 972 u_int count = 0; 973 974 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 975 pf_free_src_node(sn); 976 count++; 977 } 978 979 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 980 981 return (count); 982 } 983 984 void 985 pf_mtag_initialize(void) 986 { 987 988 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 989 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 990 UMA_ALIGN_PTR, 0); 991 } 992 993 /* Per-vnet data storage structures initialization. */ 994 void 995 pf_initialize(void) 996 { 997 struct pf_keyhash *kh; 998 struct pf_idhash *ih; 999 struct pf_srchash *sh; 1000 u_int i; 1001 1002 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 1003 pf_hashsize = PF_HASHSIZ; 1004 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 1005 pf_srchashsize = PF_SRCHASHSIZ; 1006 1007 V_pf_hashseed = arc4random(); 1008 1009 /* States and state keys storage. */ 1010 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1011 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1012 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1013 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1014 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1015 1016 V_pf_state_key_z = uma_zcreate("pf state keys", 1017 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1018 UMA_ALIGN_PTR, 0); 1019 1020 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 1021 M_PFHASH, M_NOWAIT | M_ZERO); 1022 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1023 M_PFHASH, M_NOWAIT | M_ZERO); 1024 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1025 printf("pf: Unable to allocate memory for " 1026 "state_hashsize %lu.\n", pf_hashsize); 1027 1028 free(V_pf_keyhash, M_PFHASH); 1029 free(V_pf_idhash, M_PFHASH); 1030 1031 pf_hashsize = PF_HASHSIZ; 1032 V_pf_keyhash = mallocarray(pf_hashsize, 1033 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1034 V_pf_idhash = mallocarray(pf_hashsize, 1035 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1036 } 1037 1038 pf_hashmask = pf_hashsize - 1; 1039 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1040 i++, kh++, ih++) { 1041 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1042 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1043 } 1044 1045 /* Source nodes. */ 1046 V_pf_sources_z = uma_zcreate("pf source nodes", 1047 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1048 0); 1049 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1050 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1051 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1052 1053 V_pf_srchash = mallocarray(pf_srchashsize, 1054 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1055 if (V_pf_srchash == NULL) { 1056 printf("pf: Unable to allocate memory for " 1057 "source_hashsize %lu.\n", pf_srchashsize); 1058 1059 pf_srchashsize = PF_SRCHASHSIZ; 1060 V_pf_srchash = mallocarray(pf_srchashsize, 1061 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1062 } 1063 1064 pf_srchashmask = pf_srchashsize - 1; 1065 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1066 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1067 1068 /* ALTQ */ 1069 TAILQ_INIT(&V_pf_altqs[0]); 1070 TAILQ_INIT(&V_pf_altqs[1]); 1071 TAILQ_INIT(&V_pf_altqs[2]); 1072 TAILQ_INIT(&V_pf_altqs[3]); 1073 TAILQ_INIT(&V_pf_pabuf); 1074 V_pf_altqs_active = &V_pf_altqs[0]; 1075 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1076 V_pf_altqs_inactive = &V_pf_altqs[2]; 1077 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1078 1079 /* Send & overload+flush queues. */ 1080 STAILQ_INIT(&V_pf_sendqueue); 1081 SLIST_INIT(&V_pf_overloadqueue); 1082 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1083 1084 /* Unlinked, but may be referenced rules. */ 1085 TAILQ_INIT(&V_pf_unlinked_rules); 1086 } 1087 1088 void 1089 pf_mtag_cleanup(void) 1090 { 1091 1092 uma_zdestroy(pf_mtag_z); 1093 } 1094 1095 void 1096 pf_cleanup(void) 1097 { 1098 struct pf_keyhash *kh; 1099 struct pf_idhash *ih; 1100 struct pf_srchash *sh; 1101 struct pf_send_entry *pfse, *next; 1102 u_int i; 1103 1104 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1105 i++, kh++, ih++) { 1106 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1107 __func__)); 1108 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1109 __func__)); 1110 mtx_destroy(&kh->lock); 1111 mtx_destroy(&ih->lock); 1112 } 1113 free(V_pf_keyhash, M_PFHASH); 1114 free(V_pf_idhash, M_PFHASH); 1115 1116 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1117 KASSERT(LIST_EMPTY(&sh->nodes), 1118 ("%s: source node hash not empty", __func__)); 1119 mtx_destroy(&sh->lock); 1120 } 1121 free(V_pf_srchash, M_PFHASH); 1122 1123 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1124 m_freem(pfse->pfse_m); 1125 free(pfse, M_PFTEMP); 1126 } 1127 1128 uma_zdestroy(V_pf_sources_z); 1129 uma_zdestroy(V_pf_state_z); 1130 uma_zdestroy(V_pf_state_key_z); 1131 } 1132 1133 static int 1134 pf_mtag_uminit(void *mem, int size, int how) 1135 { 1136 struct m_tag *t; 1137 1138 t = (struct m_tag *)mem; 1139 t->m_tag_cookie = MTAG_ABI_COMPAT; 1140 t->m_tag_id = PACKET_TAG_PF; 1141 t->m_tag_len = sizeof(struct pf_mtag); 1142 t->m_tag_free = pf_mtag_free; 1143 1144 return (0); 1145 } 1146 1147 static void 1148 pf_mtag_free(struct m_tag *t) 1149 { 1150 1151 uma_zfree(pf_mtag_z, t); 1152 } 1153 1154 struct pf_mtag * 1155 pf_get_mtag(struct mbuf *m) 1156 { 1157 struct m_tag *mtag; 1158 1159 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1160 return ((struct pf_mtag *)(mtag + 1)); 1161 1162 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1163 if (mtag == NULL) 1164 return (NULL); 1165 bzero(mtag + 1, sizeof(struct pf_mtag)); 1166 m_tag_prepend(m, mtag); 1167 1168 return ((struct pf_mtag *)(mtag + 1)); 1169 } 1170 1171 static int 1172 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1173 struct pf_kstate *s) 1174 { 1175 struct pf_keyhash *khs, *khw, *kh; 1176 struct pf_state_key *sk, *cur; 1177 struct pf_kstate *si, *olds = NULL; 1178 int idx; 1179 1180 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1181 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1182 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1183 1184 /* 1185 * We need to lock hash slots of both keys. To avoid deadlock 1186 * we always lock the slot with lower address first. Unlock order 1187 * isn't important. 1188 * 1189 * We also need to lock ID hash slot before dropping key 1190 * locks. On success we return with ID hash slot locked. 1191 */ 1192 1193 if (skw == sks) { 1194 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1195 PF_HASHROW_LOCK(khs); 1196 } else { 1197 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1198 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1199 if (khs == khw) { 1200 PF_HASHROW_LOCK(khs); 1201 } else if (khs < khw) { 1202 PF_HASHROW_LOCK(khs); 1203 PF_HASHROW_LOCK(khw); 1204 } else { 1205 PF_HASHROW_LOCK(khw); 1206 PF_HASHROW_LOCK(khs); 1207 } 1208 } 1209 1210 #define KEYS_UNLOCK() do { \ 1211 if (khs != khw) { \ 1212 PF_HASHROW_UNLOCK(khs); \ 1213 PF_HASHROW_UNLOCK(khw); \ 1214 } else \ 1215 PF_HASHROW_UNLOCK(khs); \ 1216 } while (0) 1217 1218 /* 1219 * First run: start with wire key. 1220 */ 1221 sk = skw; 1222 kh = khw; 1223 idx = PF_SK_WIRE; 1224 1225 MPASS(s->lock == NULL); 1226 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1227 1228 keyattach: 1229 LIST_FOREACH(cur, &kh->keys, entry) 1230 if (pf_bcmp_state_key(cur, (struct pf_state_key_cmp *)sk) == 0) 1231 break; 1232 1233 if (cur != NULL) { 1234 /* Key exists. Check for same kif, if none, add to key. */ 1235 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1236 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1237 1238 PF_HASHROW_LOCK(ih); 1239 if (si->kif == s->kif && 1240 si->direction == s->direction) { 1241 if (sk->proto == IPPROTO_TCP && 1242 si->src.state >= TCPS_FIN_WAIT_2 && 1243 si->dst.state >= TCPS_FIN_WAIT_2) { 1244 /* 1245 * New state matches an old >FIN_WAIT_2 1246 * state. We can't drop key hash locks, 1247 * thus we can't unlink it properly. 1248 * 1249 * As a workaround we drop it into 1250 * TCPS_CLOSED state, schedule purge 1251 * ASAP and push it into the very end 1252 * of the slot TAILQ, so that it won't 1253 * conflict with our new state. 1254 */ 1255 pf_set_protostate(si, PF_PEER_BOTH, 1256 TCPS_CLOSED); 1257 si->timeout = PFTM_PURGE; 1258 olds = si; 1259 } else { 1260 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1261 printf("pf: %s key attach " 1262 "failed on %s: ", 1263 (idx == PF_SK_WIRE) ? 1264 "wire" : "stack", 1265 s->kif->pfik_name); 1266 pf_print_state_parts(s, 1267 (idx == PF_SK_WIRE) ? 1268 sk : NULL, 1269 (idx == PF_SK_STACK) ? 1270 sk : NULL); 1271 printf(", existing: "); 1272 pf_print_state_parts(si, 1273 (idx == PF_SK_WIRE) ? 1274 sk : NULL, 1275 (idx == PF_SK_STACK) ? 1276 sk : NULL); 1277 printf("\n"); 1278 } 1279 PF_HASHROW_UNLOCK(ih); 1280 KEYS_UNLOCK(); 1281 uma_zfree(V_pf_state_key_z, sk); 1282 if (idx == PF_SK_STACK) 1283 pf_detach_state(s); 1284 return (EEXIST); /* collision! */ 1285 } 1286 } 1287 PF_HASHROW_UNLOCK(ih); 1288 } 1289 uma_zfree(V_pf_state_key_z, sk); 1290 s->key[idx] = cur; 1291 } else { 1292 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1293 s->key[idx] = sk; 1294 } 1295 1296 stateattach: 1297 /* List is sorted, if-bound states before floating. */ 1298 if (s->kif == V_pfi_all) 1299 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1300 else 1301 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1302 1303 if (olds) { 1304 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1305 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1306 key_list[idx]); 1307 olds = NULL; 1308 } 1309 1310 /* 1311 * Attach done. See how should we (or should not?) 1312 * attach a second key. 1313 */ 1314 if (sks == skw) { 1315 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1316 idx = PF_SK_STACK; 1317 sks = NULL; 1318 goto stateattach; 1319 } else if (sks != NULL) { 1320 /* 1321 * Continue attaching with stack key. 1322 */ 1323 sk = sks; 1324 kh = khs; 1325 idx = PF_SK_STACK; 1326 sks = NULL; 1327 goto keyattach; 1328 } 1329 1330 PF_STATE_LOCK(s); 1331 KEYS_UNLOCK(); 1332 1333 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1334 ("%s failure", __func__)); 1335 1336 return (0); 1337 #undef KEYS_UNLOCK 1338 } 1339 1340 static void 1341 pf_detach_state(struct pf_kstate *s) 1342 { 1343 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1344 struct pf_keyhash *kh; 1345 1346 if (sks != NULL) { 1347 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1348 PF_HASHROW_LOCK(kh); 1349 if (s->key[PF_SK_STACK] != NULL) 1350 pf_state_key_detach(s, PF_SK_STACK); 1351 /* 1352 * If both point to same key, then we are done. 1353 */ 1354 if (sks == s->key[PF_SK_WIRE]) { 1355 pf_state_key_detach(s, PF_SK_WIRE); 1356 PF_HASHROW_UNLOCK(kh); 1357 return; 1358 } 1359 PF_HASHROW_UNLOCK(kh); 1360 } 1361 1362 if (s->key[PF_SK_WIRE] != NULL) { 1363 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1364 PF_HASHROW_LOCK(kh); 1365 if (s->key[PF_SK_WIRE] != NULL) 1366 pf_state_key_detach(s, PF_SK_WIRE); 1367 PF_HASHROW_UNLOCK(kh); 1368 } 1369 } 1370 1371 static void 1372 pf_state_key_detach(struct pf_kstate *s, int idx) 1373 { 1374 struct pf_state_key *sk = s->key[idx]; 1375 #ifdef INVARIANTS 1376 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1377 1378 PF_HASHROW_ASSERT(kh); 1379 #endif 1380 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1381 s->key[idx] = NULL; 1382 1383 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1384 LIST_REMOVE(sk, entry); 1385 uma_zfree(V_pf_state_key_z, sk); 1386 } 1387 } 1388 1389 static int 1390 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1391 { 1392 struct pf_state_key *sk = mem; 1393 1394 bzero(sk, sizeof(struct pf_state_key_cmp)); 1395 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1396 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1397 1398 return (0); 1399 } 1400 1401 struct pf_state_key * 1402 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1403 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1404 { 1405 struct pf_state_key *sk; 1406 1407 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1408 if (sk == NULL) 1409 return (NULL); 1410 1411 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1412 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1413 sk->port[pd->sidx] = sport; 1414 sk->port[pd->didx] = dport; 1415 sk->proto = pd->proto; 1416 sk->af = pd->af; 1417 1418 return (sk); 1419 } 1420 1421 struct pf_state_key * 1422 pf_state_key_clone(struct pf_state_key *orig) 1423 { 1424 struct pf_state_key *sk; 1425 1426 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1427 if (sk == NULL) 1428 return (NULL); 1429 1430 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1431 1432 return (sk); 1433 } 1434 1435 int 1436 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1437 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1438 { 1439 struct pf_idhash *ih; 1440 struct pf_kstate *cur; 1441 int error; 1442 1443 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1444 ("%s: sks not pristine", __func__)); 1445 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1446 ("%s: skw not pristine", __func__)); 1447 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1448 1449 s->kif = kif; 1450 s->orig_kif = orig_kif; 1451 1452 if (s->id == 0 && s->creatorid == 0) { 1453 /* XXX: should be atomic, but probability of collision low */ 1454 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1455 V_pf_stateid[curcpu] = 1; 1456 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1457 s->id = htobe64(s->id); 1458 s->creatorid = V_pf_status.hostid; 1459 } 1460 1461 /* Returns with ID locked on success. */ 1462 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1463 return (error); 1464 1465 ih = &V_pf_idhash[PF_IDHASH(s)]; 1466 PF_HASHROW_ASSERT(ih); 1467 LIST_FOREACH(cur, &ih->states, entry) 1468 if (cur->id == s->id && cur->creatorid == s->creatorid) 1469 break; 1470 1471 if (cur != NULL) { 1472 PF_HASHROW_UNLOCK(ih); 1473 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1474 printf("pf: state ID collision: " 1475 "id: %016llx creatorid: %08x\n", 1476 (unsigned long long)be64toh(s->id), 1477 ntohl(s->creatorid)); 1478 } 1479 pf_detach_state(s); 1480 return (EEXIST); 1481 } 1482 LIST_INSERT_HEAD(&ih->states, s, entry); 1483 /* One for keys, one for ID hash. */ 1484 refcount_init(&s->refs, 2); 1485 1486 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1487 if (V_pfsync_insert_state_ptr != NULL) 1488 V_pfsync_insert_state_ptr(s); 1489 1490 /* Returns locked. */ 1491 return (0); 1492 } 1493 1494 /* 1495 * Find state by ID: returns with locked row on success. 1496 */ 1497 struct pf_kstate * 1498 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1499 { 1500 struct pf_idhash *ih; 1501 struct pf_kstate *s; 1502 1503 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1504 1505 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1506 1507 PF_HASHROW_LOCK(ih); 1508 LIST_FOREACH(s, &ih->states, entry) 1509 if (s->id == id && s->creatorid == creatorid) 1510 break; 1511 1512 if (s == NULL) 1513 PF_HASHROW_UNLOCK(ih); 1514 1515 return (s); 1516 } 1517 1518 /* 1519 * Find state by key. 1520 * Returns with ID hash slot locked on success. 1521 */ 1522 static struct pf_kstate * 1523 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1524 { 1525 struct pf_keyhash *kh; 1526 struct pf_state_key *sk; 1527 struct pf_kstate *s; 1528 int idx; 1529 1530 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1531 1532 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1533 1534 PF_HASHROW_LOCK(kh); 1535 LIST_FOREACH(sk, &kh->keys, entry) 1536 if (pf_bcmp_state_key(sk, key) == 0) 1537 break; 1538 if (sk == NULL) { 1539 PF_HASHROW_UNLOCK(kh); 1540 return (NULL); 1541 } 1542 1543 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1544 1545 /* List is sorted, if-bound states before floating ones. */ 1546 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1547 if (s->kif == V_pfi_all || s->kif == kif) { 1548 PF_STATE_LOCK(s); 1549 PF_HASHROW_UNLOCK(kh); 1550 if (__predict_false(s->timeout >= PFTM_MAX)) { 1551 /* 1552 * State is either being processed by 1553 * pf_unlink_state() in an other thread, or 1554 * is scheduled for immediate expiry. 1555 */ 1556 PF_STATE_UNLOCK(s); 1557 return (NULL); 1558 } 1559 return (s); 1560 } 1561 PF_HASHROW_UNLOCK(kh); 1562 1563 return (NULL); 1564 } 1565 1566 /* 1567 * Returns with ID hash slot locked on success. 1568 */ 1569 struct pf_kstate * 1570 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1571 { 1572 struct pf_keyhash *kh; 1573 struct pf_state_key *sk; 1574 struct pf_kstate *s, *ret = NULL; 1575 int idx, inout = 0; 1576 1577 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1578 1579 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1580 1581 PF_HASHROW_LOCK(kh); 1582 LIST_FOREACH(sk, &kh->keys, entry) 1583 if (pf_bcmp_state_key(sk, key) == 0) 1584 break; 1585 if (sk == NULL) { 1586 PF_HASHROW_UNLOCK(kh); 1587 return (NULL); 1588 } 1589 switch (dir) { 1590 case PF_IN: 1591 idx = PF_SK_WIRE; 1592 break; 1593 case PF_OUT: 1594 idx = PF_SK_STACK; 1595 break; 1596 case PF_INOUT: 1597 idx = PF_SK_WIRE; 1598 inout = 1; 1599 break; 1600 default: 1601 panic("%s: dir %u", __func__, dir); 1602 } 1603 second_run: 1604 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1605 if (more == NULL) { 1606 PF_STATE_LOCK(s); 1607 PF_HASHROW_UNLOCK(kh); 1608 return (s); 1609 } 1610 1611 if (ret) 1612 (*more)++; 1613 else { 1614 ret = s; 1615 PF_STATE_LOCK(s); 1616 } 1617 } 1618 if (inout == 1) { 1619 inout = 0; 1620 idx = PF_SK_STACK; 1621 goto second_run; 1622 } 1623 PF_HASHROW_UNLOCK(kh); 1624 1625 return (ret); 1626 } 1627 1628 /* 1629 * FIXME 1630 * This routine is inefficient -- locks the state only to unlock immediately on 1631 * return. 1632 * It is racy -- after the state is unlocked nothing stops other threads from 1633 * removing it. 1634 */ 1635 bool 1636 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1637 { 1638 struct pf_kstate *s; 1639 1640 s = pf_find_state_all(key, dir, NULL); 1641 if (s != NULL) { 1642 PF_STATE_UNLOCK(s); 1643 return (true); 1644 } 1645 return (false); 1646 } 1647 1648 /* END state table stuff */ 1649 1650 static void 1651 pf_send(struct pf_send_entry *pfse) 1652 { 1653 1654 PF_SENDQ_LOCK(); 1655 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1656 PF_SENDQ_UNLOCK(); 1657 swi_sched(V_pf_swi_cookie, 0); 1658 } 1659 1660 static bool 1661 pf_isforlocal(struct mbuf *m, int af) 1662 { 1663 switch (af) { 1664 #ifdef INET 1665 case AF_INET: { 1666 struct ip *ip = mtod(m, struct ip *); 1667 1668 return (in_localip(ip->ip_dst)); 1669 } 1670 #endif 1671 #ifdef INET6 1672 case AF_INET6: { 1673 struct ip6_hdr *ip6; 1674 struct in6_ifaddr *ia; 1675 ip6 = mtod(m, struct ip6_hdr *); 1676 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1677 if (ia == NULL) 1678 return (false); 1679 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1680 } 1681 #endif 1682 default: 1683 panic("Unsupported af %d", af); 1684 } 1685 1686 return (false); 1687 } 1688 1689 void 1690 pf_intr(void *v) 1691 { 1692 struct epoch_tracker et; 1693 struct pf_send_head queue; 1694 struct pf_send_entry *pfse, *next; 1695 1696 CURVNET_SET((struct vnet *)v); 1697 1698 PF_SENDQ_LOCK(); 1699 queue = V_pf_sendqueue; 1700 STAILQ_INIT(&V_pf_sendqueue); 1701 PF_SENDQ_UNLOCK(); 1702 1703 NET_EPOCH_ENTER(et); 1704 1705 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1706 switch (pfse->pfse_type) { 1707 #ifdef INET 1708 case PFSE_IP: { 1709 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1710 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1711 pfse->pfse_m->m_pkthdr.csum_flags |= 1712 CSUM_IP_VALID | CSUM_IP_CHECKED; 1713 ip_input(pfse->pfse_m); 1714 } else { 1715 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1716 NULL); 1717 } 1718 break; 1719 } 1720 case PFSE_ICMP: 1721 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1722 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1723 break; 1724 #endif /* INET */ 1725 #ifdef INET6 1726 case PFSE_IP6: 1727 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1728 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1729 ip6_input(pfse->pfse_m); 1730 } else { 1731 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1732 NULL, NULL); 1733 } 1734 break; 1735 case PFSE_ICMP6: 1736 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1737 pfse->icmpopts.code, pfse->icmpopts.mtu); 1738 break; 1739 #endif /* INET6 */ 1740 default: 1741 panic("%s: unknown type", __func__); 1742 } 1743 free(pfse, M_PFTEMP); 1744 } 1745 NET_EPOCH_EXIT(et); 1746 CURVNET_RESTORE(); 1747 } 1748 1749 #define pf_purge_thread_period (hz / 10) 1750 1751 #ifdef PF_WANT_32_TO_64_COUNTER 1752 static void 1753 pf_status_counter_u64_periodic(void) 1754 { 1755 1756 PF_RULES_RASSERT(); 1757 1758 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1759 return; 1760 } 1761 1762 for (int i = 0; i < FCNT_MAX; i++) { 1763 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1764 } 1765 } 1766 1767 static void 1768 pf_kif_counter_u64_periodic(void) 1769 { 1770 struct pfi_kkif *kif; 1771 size_t r, run; 1772 1773 PF_RULES_RASSERT(); 1774 1775 if (__predict_false(V_pf_allkifcount == 0)) { 1776 return; 1777 } 1778 1779 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1780 return; 1781 } 1782 1783 run = V_pf_allkifcount / 10; 1784 if (run < 5) 1785 run = 5; 1786 1787 for (r = 0; r < run; r++) { 1788 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1789 if (kif == NULL) { 1790 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1791 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1792 break; 1793 } 1794 1795 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1796 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1797 1798 for (int i = 0; i < 2; i++) { 1799 for (int j = 0; j < 2; j++) { 1800 for (int k = 0; k < 2; k++) { 1801 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1802 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1803 } 1804 } 1805 } 1806 } 1807 } 1808 1809 static void 1810 pf_rule_counter_u64_periodic(void) 1811 { 1812 struct pf_krule *rule; 1813 size_t r, run; 1814 1815 PF_RULES_RASSERT(); 1816 1817 if (__predict_false(V_pf_allrulecount == 0)) { 1818 return; 1819 } 1820 1821 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1822 return; 1823 } 1824 1825 run = V_pf_allrulecount / 10; 1826 if (run < 5) 1827 run = 5; 1828 1829 for (r = 0; r < run; r++) { 1830 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1831 if (rule == NULL) { 1832 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1833 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1834 break; 1835 } 1836 1837 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1838 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1839 1840 pf_counter_u64_periodic(&rule->evaluations); 1841 for (int i = 0; i < 2; i++) { 1842 pf_counter_u64_periodic(&rule->packets[i]); 1843 pf_counter_u64_periodic(&rule->bytes[i]); 1844 } 1845 } 1846 } 1847 1848 static void 1849 pf_counter_u64_periodic_main(void) 1850 { 1851 PF_RULES_RLOCK_TRACKER; 1852 1853 V_pf_counter_periodic_iter++; 1854 1855 PF_RULES_RLOCK(); 1856 pf_counter_u64_critical_enter(); 1857 pf_status_counter_u64_periodic(); 1858 pf_kif_counter_u64_periodic(); 1859 pf_rule_counter_u64_periodic(); 1860 pf_counter_u64_critical_exit(); 1861 PF_RULES_RUNLOCK(); 1862 } 1863 #else 1864 #define pf_counter_u64_periodic_main() do { } while (0) 1865 #endif 1866 1867 void 1868 pf_purge_thread(void *unused __unused) 1869 { 1870 VNET_ITERATOR_DECL(vnet_iter); 1871 1872 sx_xlock(&pf_end_lock); 1873 while (pf_end_threads == 0) { 1874 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1875 1876 VNET_LIST_RLOCK(); 1877 VNET_FOREACH(vnet_iter) { 1878 CURVNET_SET(vnet_iter); 1879 1880 /* Wait until V_pf_default_rule is initialized. */ 1881 if (V_pf_vnet_active == 0) { 1882 CURVNET_RESTORE(); 1883 continue; 1884 } 1885 1886 pf_counter_u64_periodic_main(); 1887 1888 /* 1889 * Process 1/interval fraction of the state 1890 * table every run. 1891 */ 1892 V_pf_purge_idx = 1893 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1894 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1895 1896 /* 1897 * Purge other expired types every 1898 * PFTM_INTERVAL seconds. 1899 */ 1900 if (V_pf_purge_idx == 0) { 1901 /* 1902 * Order is important: 1903 * - states and src nodes reference rules 1904 * - states and rules reference kifs 1905 */ 1906 pf_purge_expired_fragments(); 1907 pf_purge_expired_src_nodes(); 1908 pf_purge_unlinked_rules(); 1909 pfi_kkif_purge(); 1910 } 1911 CURVNET_RESTORE(); 1912 } 1913 VNET_LIST_RUNLOCK(); 1914 } 1915 1916 pf_end_threads++; 1917 sx_xunlock(&pf_end_lock); 1918 kproc_exit(0); 1919 } 1920 1921 void 1922 pf_unload_vnet_purge(void) 1923 { 1924 1925 /* 1926 * To cleanse up all kifs and rules we need 1927 * two runs: first one clears reference flags, 1928 * then pf_purge_expired_states() doesn't 1929 * raise them, and then second run frees. 1930 */ 1931 pf_purge_unlinked_rules(); 1932 pfi_kkif_purge(); 1933 1934 /* 1935 * Now purge everything. 1936 */ 1937 pf_purge_expired_states(0, pf_hashmask); 1938 pf_purge_fragments(UINT_MAX); 1939 pf_purge_expired_src_nodes(); 1940 1941 /* 1942 * Now all kifs & rules should be unreferenced, 1943 * thus should be successfully freed. 1944 */ 1945 pf_purge_unlinked_rules(); 1946 pfi_kkif_purge(); 1947 } 1948 1949 u_int32_t 1950 pf_state_expires(const struct pf_kstate *state) 1951 { 1952 u_int32_t timeout; 1953 u_int32_t start; 1954 u_int32_t end; 1955 u_int32_t states; 1956 1957 /* handle all PFTM_* > PFTM_MAX here */ 1958 if (state->timeout == PFTM_PURGE) 1959 return (time_uptime); 1960 KASSERT(state->timeout != PFTM_UNLINKED, 1961 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1962 KASSERT((state->timeout < PFTM_MAX), 1963 ("pf_state_expires: timeout > PFTM_MAX")); 1964 timeout = state->rule.ptr->timeout[state->timeout]; 1965 if (!timeout) 1966 timeout = V_pf_default_rule.timeout[state->timeout]; 1967 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1968 if (start && state->rule.ptr != &V_pf_default_rule) { 1969 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1970 states = counter_u64_fetch(state->rule.ptr->states_cur); 1971 } else { 1972 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1973 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1974 states = V_pf_status.states; 1975 } 1976 if (end && states > start && start < end) { 1977 if (states < end) { 1978 timeout = (u_int64_t)timeout * (end - states) / 1979 (end - start); 1980 return (state->expire + timeout); 1981 } 1982 else 1983 return (time_uptime); 1984 } 1985 return (state->expire + timeout); 1986 } 1987 1988 void 1989 pf_purge_expired_src_nodes(void) 1990 { 1991 struct pf_ksrc_node_list freelist; 1992 struct pf_srchash *sh; 1993 struct pf_ksrc_node *cur, *next; 1994 int i; 1995 1996 LIST_INIT(&freelist); 1997 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1998 PF_HASHROW_LOCK(sh); 1999 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2000 if (cur->states == 0 && cur->expire <= time_uptime) { 2001 pf_unlink_src_node(cur); 2002 LIST_INSERT_HEAD(&freelist, cur, entry); 2003 } else if (cur->rule.ptr != NULL) 2004 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2005 PF_HASHROW_UNLOCK(sh); 2006 } 2007 2008 pf_free_src_nodes(&freelist); 2009 2010 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2011 } 2012 2013 static void 2014 pf_src_tree_remove_state(struct pf_kstate *s) 2015 { 2016 struct pf_ksrc_node *sn; 2017 struct pf_srchash *sh; 2018 uint32_t timeout; 2019 2020 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2021 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2022 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2023 2024 if (s->src_node != NULL) { 2025 sn = s->src_node; 2026 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 2027 PF_HASHROW_LOCK(sh); 2028 if (s->src.tcp_est) 2029 --sn->conn; 2030 if (--sn->states == 0) 2031 sn->expire = time_uptime + timeout; 2032 PF_HASHROW_UNLOCK(sh); 2033 } 2034 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2035 sn = s->nat_src_node; 2036 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 2037 PF_HASHROW_LOCK(sh); 2038 if (--sn->states == 0) 2039 sn->expire = time_uptime + timeout; 2040 PF_HASHROW_UNLOCK(sh); 2041 } 2042 s->src_node = s->nat_src_node = NULL; 2043 } 2044 2045 /* 2046 * Unlink and potentilly free a state. Function may be 2047 * called with ID hash row locked, but always returns 2048 * unlocked, since it needs to go through key hash locking. 2049 */ 2050 int 2051 pf_unlink_state(struct pf_kstate *s) 2052 { 2053 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2054 2055 PF_HASHROW_ASSERT(ih); 2056 2057 if (s->timeout == PFTM_UNLINKED) { 2058 /* 2059 * State is being processed 2060 * by pf_unlink_state() in 2061 * an other thread. 2062 */ 2063 PF_HASHROW_UNLOCK(ih); 2064 return (0); /* XXXGL: undefined actually */ 2065 } 2066 2067 if (s->src.state == PF_TCPS_PROXY_DST) { 2068 /* XXX wire key the right one? */ 2069 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2070 &s->key[PF_SK_WIRE]->addr[1], 2071 &s->key[PF_SK_WIRE]->addr[0], 2072 s->key[PF_SK_WIRE]->port[1], 2073 s->key[PF_SK_WIRE]->port[0], 2074 s->src.seqhi, s->src.seqlo + 1, 2075 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag); 2076 } 2077 2078 LIST_REMOVE(s, entry); 2079 pf_src_tree_remove_state(s); 2080 2081 if (V_pfsync_delete_state_ptr != NULL) 2082 V_pfsync_delete_state_ptr(s); 2083 2084 STATE_DEC_COUNTERS(s); 2085 2086 s->timeout = PFTM_UNLINKED; 2087 2088 /* Ensure we remove it from the list of halfopen states, if needed. */ 2089 if (s->key[PF_SK_STACK] != NULL && 2090 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2091 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2092 2093 PF_HASHROW_UNLOCK(ih); 2094 2095 pf_detach_state(s); 2096 /* pf_state_insert() initialises refs to 2 */ 2097 return (pf_release_staten(s, 2)); 2098 } 2099 2100 struct pf_kstate * 2101 pf_alloc_state(int flags) 2102 { 2103 2104 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2105 } 2106 2107 void 2108 pf_free_state(struct pf_kstate *cur) 2109 { 2110 2111 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2112 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2113 cur->timeout)); 2114 2115 pf_normalize_tcp_cleanup(cur); 2116 uma_zfree(V_pf_state_z, cur); 2117 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2118 } 2119 2120 /* 2121 * Called only from pf_purge_thread(), thus serialized. 2122 */ 2123 static u_int 2124 pf_purge_expired_states(u_int i, int maxcheck) 2125 { 2126 struct pf_idhash *ih; 2127 struct pf_kstate *s; 2128 2129 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2130 2131 /* 2132 * Go through hash and unlink states that expire now. 2133 */ 2134 while (maxcheck > 0) { 2135 ih = &V_pf_idhash[i]; 2136 2137 /* only take the lock if we expect to do work */ 2138 if (!LIST_EMPTY(&ih->states)) { 2139 relock: 2140 PF_HASHROW_LOCK(ih); 2141 LIST_FOREACH(s, &ih->states, entry) { 2142 if (pf_state_expires(s) <= time_uptime) { 2143 V_pf_status.states -= 2144 pf_unlink_state(s); 2145 goto relock; 2146 } 2147 s->rule.ptr->rule_ref |= PFRULE_REFS; 2148 if (s->nat_rule.ptr != NULL) 2149 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2150 if (s->anchor.ptr != NULL) 2151 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2152 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2153 if (s->rt_kif) 2154 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2155 } 2156 PF_HASHROW_UNLOCK(ih); 2157 } 2158 2159 /* Return when we hit end of hash. */ 2160 if (++i > pf_hashmask) { 2161 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2162 return (0); 2163 } 2164 2165 maxcheck--; 2166 } 2167 2168 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2169 2170 return (i); 2171 } 2172 2173 static void 2174 pf_purge_unlinked_rules(void) 2175 { 2176 struct pf_krulequeue tmpq; 2177 struct pf_krule *r, *r1; 2178 2179 /* 2180 * If we have overloading task pending, then we'd 2181 * better skip purging this time. There is a tiny 2182 * probability that overloading task references 2183 * an already unlinked rule. 2184 */ 2185 PF_OVERLOADQ_LOCK(); 2186 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2187 PF_OVERLOADQ_UNLOCK(); 2188 return; 2189 } 2190 PF_OVERLOADQ_UNLOCK(); 2191 2192 /* 2193 * Do naive mark-and-sweep garbage collecting of old rules. 2194 * Reference flag is raised by pf_purge_expired_states() 2195 * and pf_purge_expired_src_nodes(). 2196 * 2197 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2198 * use a temporary queue. 2199 */ 2200 TAILQ_INIT(&tmpq); 2201 PF_UNLNKDRULES_LOCK(); 2202 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2203 if (!(r->rule_ref & PFRULE_REFS)) { 2204 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2205 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2206 } else 2207 r->rule_ref &= ~PFRULE_REFS; 2208 } 2209 PF_UNLNKDRULES_UNLOCK(); 2210 2211 if (!TAILQ_EMPTY(&tmpq)) { 2212 PF_CONFIG_LOCK(); 2213 PF_RULES_WLOCK(); 2214 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2215 TAILQ_REMOVE(&tmpq, r, entries); 2216 pf_free_rule(r); 2217 } 2218 PF_RULES_WUNLOCK(); 2219 PF_CONFIG_UNLOCK(); 2220 } 2221 } 2222 2223 void 2224 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2225 { 2226 switch (af) { 2227 #ifdef INET 2228 case AF_INET: { 2229 u_int32_t a = ntohl(addr->addr32[0]); 2230 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2231 (a>>8)&255, a&255); 2232 if (p) { 2233 p = ntohs(p); 2234 printf(":%u", p); 2235 } 2236 break; 2237 } 2238 #endif /* INET */ 2239 #ifdef INET6 2240 case AF_INET6: { 2241 u_int16_t b; 2242 u_int8_t i, curstart, curend, maxstart, maxend; 2243 curstart = curend = maxstart = maxend = 255; 2244 for (i = 0; i < 8; i++) { 2245 if (!addr->addr16[i]) { 2246 if (curstart == 255) 2247 curstart = i; 2248 curend = i; 2249 } else { 2250 if ((curend - curstart) > 2251 (maxend - maxstart)) { 2252 maxstart = curstart; 2253 maxend = curend; 2254 } 2255 curstart = curend = 255; 2256 } 2257 } 2258 if ((curend - curstart) > 2259 (maxend - maxstart)) { 2260 maxstart = curstart; 2261 maxend = curend; 2262 } 2263 for (i = 0; i < 8; i++) { 2264 if (i >= maxstart && i <= maxend) { 2265 if (i == 0) 2266 printf(":"); 2267 if (i == maxend) 2268 printf(":"); 2269 } else { 2270 b = ntohs(addr->addr16[i]); 2271 printf("%x", b); 2272 if (i < 7) 2273 printf(":"); 2274 } 2275 } 2276 if (p) { 2277 p = ntohs(p); 2278 printf("[%u]", p); 2279 } 2280 break; 2281 } 2282 #endif /* INET6 */ 2283 } 2284 } 2285 2286 void 2287 pf_print_state(struct pf_kstate *s) 2288 { 2289 pf_print_state_parts(s, NULL, NULL); 2290 } 2291 2292 static void 2293 pf_print_state_parts(struct pf_kstate *s, 2294 struct pf_state_key *skwp, struct pf_state_key *sksp) 2295 { 2296 struct pf_state_key *skw, *sks; 2297 u_int8_t proto, dir; 2298 2299 /* Do our best to fill these, but they're skipped if NULL */ 2300 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2301 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2302 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2303 dir = s ? s->direction : 0; 2304 2305 switch (proto) { 2306 case IPPROTO_IPV4: 2307 printf("IPv4"); 2308 break; 2309 case IPPROTO_IPV6: 2310 printf("IPv6"); 2311 break; 2312 case IPPROTO_TCP: 2313 printf("TCP"); 2314 break; 2315 case IPPROTO_UDP: 2316 printf("UDP"); 2317 break; 2318 case IPPROTO_ICMP: 2319 printf("ICMP"); 2320 break; 2321 case IPPROTO_ICMPV6: 2322 printf("ICMPv6"); 2323 break; 2324 default: 2325 printf("%u", proto); 2326 break; 2327 } 2328 switch (dir) { 2329 case PF_IN: 2330 printf(" in"); 2331 break; 2332 case PF_OUT: 2333 printf(" out"); 2334 break; 2335 } 2336 if (skw) { 2337 printf(" wire: "); 2338 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2339 printf(" "); 2340 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2341 } 2342 if (sks) { 2343 printf(" stack: "); 2344 if (sks != skw) { 2345 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2346 printf(" "); 2347 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2348 } else 2349 printf("-"); 2350 } 2351 if (s) { 2352 if (proto == IPPROTO_TCP) { 2353 printf(" [lo=%u high=%u win=%u modulator=%u", 2354 s->src.seqlo, s->src.seqhi, 2355 s->src.max_win, s->src.seqdiff); 2356 if (s->src.wscale && s->dst.wscale) 2357 printf(" wscale=%u", 2358 s->src.wscale & PF_WSCALE_MASK); 2359 printf("]"); 2360 printf(" [lo=%u high=%u win=%u modulator=%u", 2361 s->dst.seqlo, s->dst.seqhi, 2362 s->dst.max_win, s->dst.seqdiff); 2363 if (s->src.wscale && s->dst.wscale) 2364 printf(" wscale=%u", 2365 s->dst.wscale & PF_WSCALE_MASK); 2366 printf("]"); 2367 } 2368 printf(" %u:%u", s->src.state, s->dst.state); 2369 } 2370 } 2371 2372 void 2373 pf_print_flags(u_int8_t f) 2374 { 2375 if (f) 2376 printf(" "); 2377 if (f & TH_FIN) 2378 printf("F"); 2379 if (f & TH_SYN) 2380 printf("S"); 2381 if (f & TH_RST) 2382 printf("R"); 2383 if (f & TH_PUSH) 2384 printf("P"); 2385 if (f & TH_ACK) 2386 printf("A"); 2387 if (f & TH_URG) 2388 printf("U"); 2389 if (f & TH_ECE) 2390 printf("E"); 2391 if (f & TH_CWR) 2392 printf("W"); 2393 } 2394 2395 #define PF_SET_SKIP_STEPS(i) \ 2396 do { \ 2397 while (head[i] != cur) { \ 2398 head[i]->skip[i].ptr = cur; \ 2399 head[i] = TAILQ_NEXT(head[i], entries); \ 2400 } \ 2401 } while (0) 2402 2403 void 2404 pf_calc_skip_steps(struct pf_krulequeue *rules) 2405 { 2406 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2407 int i; 2408 2409 cur = TAILQ_FIRST(rules); 2410 prev = cur; 2411 for (i = 0; i < PF_SKIP_COUNT; ++i) 2412 head[i] = cur; 2413 while (cur != NULL) { 2414 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2415 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2416 if (cur->direction != prev->direction) 2417 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2418 if (cur->af != prev->af) 2419 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2420 if (cur->proto != prev->proto) 2421 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2422 if (cur->src.neg != prev->src.neg || 2423 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2424 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2425 if (cur->src.port[0] != prev->src.port[0] || 2426 cur->src.port[1] != prev->src.port[1] || 2427 cur->src.port_op != prev->src.port_op) 2428 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2429 if (cur->dst.neg != prev->dst.neg || 2430 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2431 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2432 if (cur->dst.port[0] != prev->dst.port[0] || 2433 cur->dst.port[1] != prev->dst.port[1] || 2434 cur->dst.port_op != prev->dst.port_op) 2435 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2436 2437 prev = cur; 2438 cur = TAILQ_NEXT(cur, entries); 2439 } 2440 for (i = 0; i < PF_SKIP_COUNT; ++i) 2441 PF_SET_SKIP_STEPS(i); 2442 } 2443 2444 static int 2445 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2446 { 2447 if (aw1->type != aw2->type) 2448 return (1); 2449 switch (aw1->type) { 2450 case PF_ADDR_ADDRMASK: 2451 case PF_ADDR_RANGE: 2452 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2453 return (1); 2454 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2455 return (1); 2456 return (0); 2457 case PF_ADDR_DYNIFTL: 2458 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2459 case PF_ADDR_NOROUTE: 2460 case PF_ADDR_URPFFAILED: 2461 return (0); 2462 case PF_ADDR_TABLE: 2463 return (aw1->p.tbl != aw2->p.tbl); 2464 default: 2465 printf("invalid address type: %d\n", aw1->type); 2466 return (1); 2467 } 2468 } 2469 2470 /** 2471 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2472 * header isn't always a full checksum. In some cases (i.e. output) it's a 2473 * pseudo-header checksum, which is a partial checksum over src/dst IP 2474 * addresses, protocol number and length. 2475 * 2476 * That means we have the following cases: 2477 * * Input or forwarding: we don't have TSO, the checksum fields are full 2478 * checksums, we need to update the checksum whenever we change anything. 2479 * * Output (i.e. the checksum is a pseudo-header checksum): 2480 * x The field being updated is src/dst address or affects the length of 2481 * the packet. We need to update the pseudo-header checksum (note that this 2482 * checksum is not ones' complement). 2483 * x Some other field is being modified (e.g. src/dst port numbers): We 2484 * don't have to update anything. 2485 **/ 2486 u_int16_t 2487 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2488 { 2489 u_int32_t x; 2490 2491 x = cksum + old - new; 2492 x = (x + (x >> 16)) & 0xffff; 2493 2494 /* optimise: eliminate a branch when not udp */ 2495 if (udp && cksum == 0x0000) 2496 return cksum; 2497 if (udp && x == 0x0000) 2498 x = 0xffff; 2499 2500 return (u_int16_t)(x); 2501 } 2502 2503 static void 2504 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2505 u_int8_t udp) 2506 { 2507 u_int16_t old = htons(hi ? (*f << 8) : *f); 2508 u_int16_t new = htons(hi ? ( v << 8) : v); 2509 2510 if (*f == v) 2511 return; 2512 2513 *f = v; 2514 2515 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2516 return; 2517 2518 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2519 } 2520 2521 void 2522 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2523 bool hi, u_int8_t udp) 2524 { 2525 u_int8_t *fb = (u_int8_t *)f; 2526 u_int8_t *vb = (u_int8_t *)&v; 2527 2528 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2529 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2530 } 2531 2532 void 2533 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2534 bool hi, u_int8_t udp) 2535 { 2536 u_int8_t *fb = (u_int8_t *)f; 2537 u_int8_t *vb = (u_int8_t *)&v; 2538 2539 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2540 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2541 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2542 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2543 } 2544 2545 u_int16_t 2546 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2547 u_int16_t new, u_int8_t udp) 2548 { 2549 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2550 return (cksum); 2551 2552 return (pf_cksum_fixup(cksum, old, new, udp)); 2553 } 2554 2555 static void 2556 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2557 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2558 sa_family_t af) 2559 { 2560 struct pf_addr ao; 2561 u_int16_t po = *p; 2562 2563 PF_ACPY(&ao, a, af); 2564 PF_ACPY(a, an, af); 2565 2566 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2567 *pc = ~*pc; 2568 2569 *p = pn; 2570 2571 switch (af) { 2572 #ifdef INET 2573 case AF_INET: 2574 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2575 ao.addr16[0], an->addr16[0], 0), 2576 ao.addr16[1], an->addr16[1], 0); 2577 *p = pn; 2578 2579 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2580 ao.addr16[0], an->addr16[0], u), 2581 ao.addr16[1], an->addr16[1], u); 2582 2583 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2584 break; 2585 #endif /* INET */ 2586 #ifdef INET6 2587 case AF_INET6: 2588 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2589 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2590 pf_cksum_fixup(pf_cksum_fixup(*pc, 2591 ao.addr16[0], an->addr16[0], u), 2592 ao.addr16[1], an->addr16[1], u), 2593 ao.addr16[2], an->addr16[2], u), 2594 ao.addr16[3], an->addr16[3], u), 2595 ao.addr16[4], an->addr16[4], u), 2596 ao.addr16[5], an->addr16[5], u), 2597 ao.addr16[6], an->addr16[6], u), 2598 ao.addr16[7], an->addr16[7], u); 2599 2600 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2601 break; 2602 #endif /* INET6 */ 2603 } 2604 2605 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2606 CSUM_DELAY_DATA_IPV6)) { 2607 *pc = ~*pc; 2608 if (! *pc) 2609 *pc = 0xffff; 2610 } 2611 } 2612 2613 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2614 void 2615 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2616 { 2617 u_int32_t ao; 2618 2619 memcpy(&ao, a, sizeof(ao)); 2620 memcpy(a, &an, sizeof(u_int32_t)); 2621 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2622 ao % 65536, an % 65536, u); 2623 } 2624 2625 void 2626 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2627 { 2628 u_int32_t ao; 2629 2630 memcpy(&ao, a, sizeof(ao)); 2631 memcpy(a, &an, sizeof(u_int32_t)); 2632 2633 *c = pf_proto_cksum_fixup(m, 2634 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2635 ao % 65536, an % 65536, udp); 2636 } 2637 2638 #ifdef INET6 2639 static void 2640 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2641 { 2642 struct pf_addr ao; 2643 2644 PF_ACPY(&ao, a, AF_INET6); 2645 PF_ACPY(a, an, AF_INET6); 2646 2647 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2648 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2649 pf_cksum_fixup(pf_cksum_fixup(*c, 2650 ao.addr16[0], an->addr16[0], u), 2651 ao.addr16[1], an->addr16[1], u), 2652 ao.addr16[2], an->addr16[2], u), 2653 ao.addr16[3], an->addr16[3], u), 2654 ao.addr16[4], an->addr16[4], u), 2655 ao.addr16[5], an->addr16[5], u), 2656 ao.addr16[6], an->addr16[6], u), 2657 ao.addr16[7], an->addr16[7], u); 2658 } 2659 #endif /* INET6 */ 2660 2661 static void 2662 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2663 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2664 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2665 { 2666 struct pf_addr oia, ooa; 2667 2668 PF_ACPY(&oia, ia, af); 2669 if (oa) 2670 PF_ACPY(&ooa, oa, af); 2671 2672 /* Change inner protocol port, fix inner protocol checksum. */ 2673 if (ip != NULL) { 2674 u_int16_t oip = *ip; 2675 u_int32_t opc; 2676 2677 if (pc != NULL) 2678 opc = *pc; 2679 *ip = np; 2680 if (pc != NULL) 2681 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2682 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2683 if (pc != NULL) 2684 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2685 } 2686 /* Change inner ip address, fix inner ip and icmp checksums. */ 2687 PF_ACPY(ia, na, af); 2688 switch (af) { 2689 #ifdef INET 2690 case AF_INET: { 2691 u_int32_t oh2c = *h2c; 2692 2693 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2694 oia.addr16[0], ia->addr16[0], 0), 2695 oia.addr16[1], ia->addr16[1], 0); 2696 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2697 oia.addr16[0], ia->addr16[0], 0), 2698 oia.addr16[1], ia->addr16[1], 0); 2699 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2700 break; 2701 } 2702 #endif /* INET */ 2703 #ifdef INET6 2704 case AF_INET6: 2705 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2706 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2707 pf_cksum_fixup(pf_cksum_fixup(*ic, 2708 oia.addr16[0], ia->addr16[0], u), 2709 oia.addr16[1], ia->addr16[1], u), 2710 oia.addr16[2], ia->addr16[2], u), 2711 oia.addr16[3], ia->addr16[3], u), 2712 oia.addr16[4], ia->addr16[4], u), 2713 oia.addr16[5], ia->addr16[5], u), 2714 oia.addr16[6], ia->addr16[6], u), 2715 oia.addr16[7], ia->addr16[7], u); 2716 break; 2717 #endif /* INET6 */ 2718 } 2719 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2720 if (oa) { 2721 PF_ACPY(oa, na, af); 2722 switch (af) { 2723 #ifdef INET 2724 case AF_INET: 2725 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2726 ooa.addr16[0], oa->addr16[0], 0), 2727 ooa.addr16[1], oa->addr16[1], 0); 2728 break; 2729 #endif /* INET */ 2730 #ifdef INET6 2731 case AF_INET6: 2732 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2733 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2734 pf_cksum_fixup(pf_cksum_fixup(*ic, 2735 ooa.addr16[0], oa->addr16[0], u), 2736 ooa.addr16[1], oa->addr16[1], u), 2737 ooa.addr16[2], oa->addr16[2], u), 2738 ooa.addr16[3], oa->addr16[3], u), 2739 ooa.addr16[4], oa->addr16[4], u), 2740 ooa.addr16[5], oa->addr16[5], u), 2741 ooa.addr16[6], oa->addr16[6], u), 2742 ooa.addr16[7], oa->addr16[7], u); 2743 break; 2744 #endif /* INET6 */ 2745 } 2746 } 2747 } 2748 2749 /* 2750 * Need to modulate the sequence numbers in the TCP SACK option 2751 * (credits to Krzysztof Pfaff for report and patch) 2752 */ 2753 static int 2754 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2755 struct tcphdr *th, struct pf_state_peer *dst) 2756 { 2757 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2758 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2759 int copyback = 0, i, olen; 2760 struct sackblk sack; 2761 2762 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2763 if (hlen < TCPOLEN_SACKLEN || 2764 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2765 return 0; 2766 2767 while (hlen >= TCPOLEN_SACKLEN) { 2768 size_t startoff = opt - opts; 2769 olen = opt[1]; 2770 switch (*opt) { 2771 case TCPOPT_EOL: /* FALLTHROUGH */ 2772 case TCPOPT_NOP: 2773 opt++; 2774 hlen--; 2775 break; 2776 case TCPOPT_SACK: 2777 if (olen > hlen) 2778 olen = hlen; 2779 if (olen >= TCPOLEN_SACKLEN) { 2780 for (i = 2; i + TCPOLEN_SACK <= olen; 2781 i += TCPOLEN_SACK) { 2782 memcpy(&sack, &opt[i], sizeof(sack)); 2783 pf_patch_32_unaligned(m, 2784 &th->th_sum, &sack.start, 2785 htonl(ntohl(sack.start) - dst->seqdiff), 2786 PF_ALGNMNT(startoff), 2787 0); 2788 pf_patch_32_unaligned(m, &th->th_sum, 2789 &sack.end, 2790 htonl(ntohl(sack.end) - dst->seqdiff), 2791 PF_ALGNMNT(startoff), 2792 0); 2793 memcpy(&opt[i], &sack, sizeof(sack)); 2794 } 2795 copyback = 1; 2796 } 2797 /* FALLTHROUGH */ 2798 default: 2799 if (olen < 2) 2800 olen = 2; 2801 hlen -= olen; 2802 opt += olen; 2803 } 2804 } 2805 2806 if (copyback) 2807 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2808 return (copyback); 2809 } 2810 2811 struct mbuf * 2812 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2813 const struct pf_addr *saddr, const struct pf_addr *daddr, 2814 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2815 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2816 u_int16_t rtag) 2817 { 2818 struct mbuf *m; 2819 int len, tlen; 2820 #ifdef INET 2821 struct ip *h = NULL; 2822 #endif /* INET */ 2823 #ifdef INET6 2824 struct ip6_hdr *h6 = NULL; 2825 #endif /* INET6 */ 2826 struct tcphdr *th; 2827 char *opt; 2828 struct pf_mtag *pf_mtag; 2829 2830 len = 0; 2831 th = NULL; 2832 2833 /* maximum segment size tcp option */ 2834 tlen = sizeof(struct tcphdr); 2835 if (mss) 2836 tlen += 4; 2837 2838 switch (af) { 2839 #ifdef INET 2840 case AF_INET: 2841 len = sizeof(struct ip) + tlen; 2842 break; 2843 #endif /* INET */ 2844 #ifdef INET6 2845 case AF_INET6: 2846 len = sizeof(struct ip6_hdr) + tlen; 2847 break; 2848 #endif /* INET6 */ 2849 default: 2850 panic("%s: unsupported af %d", __func__, af); 2851 } 2852 2853 m = m_gethdr(M_NOWAIT, MT_DATA); 2854 if (m == NULL) 2855 return (NULL); 2856 2857 #ifdef MAC 2858 mac_netinet_firewall_send(m); 2859 #endif 2860 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2861 m_freem(m); 2862 return (NULL); 2863 } 2864 if (tag) 2865 m->m_flags |= M_SKIP_FIREWALL; 2866 pf_mtag->tag = rtag; 2867 2868 if (r != NULL && r->rtableid >= 0) 2869 M_SETFIB(m, r->rtableid); 2870 2871 #ifdef ALTQ 2872 if (r != NULL && r->qid) { 2873 pf_mtag->qid = r->qid; 2874 2875 /* add hints for ecn */ 2876 pf_mtag->hdr = mtod(m, struct ip *); 2877 } 2878 #endif /* ALTQ */ 2879 m->m_data += max_linkhdr; 2880 m->m_pkthdr.len = m->m_len = len; 2881 /* The rest of the stack assumes a rcvif, so provide one. 2882 * This is a locally generated packet, so .. close enough. */ 2883 m->m_pkthdr.rcvif = V_loif; 2884 bzero(m->m_data, len); 2885 switch (af) { 2886 #ifdef INET 2887 case AF_INET: 2888 h = mtod(m, struct ip *); 2889 2890 /* IP header fields included in the TCP checksum */ 2891 h->ip_p = IPPROTO_TCP; 2892 h->ip_len = htons(tlen); 2893 h->ip_src.s_addr = saddr->v4.s_addr; 2894 h->ip_dst.s_addr = daddr->v4.s_addr; 2895 2896 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2897 break; 2898 #endif /* INET */ 2899 #ifdef INET6 2900 case AF_INET6: 2901 h6 = mtod(m, struct ip6_hdr *); 2902 2903 /* IP header fields included in the TCP checksum */ 2904 h6->ip6_nxt = IPPROTO_TCP; 2905 h6->ip6_plen = htons(tlen); 2906 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2907 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2908 2909 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2910 break; 2911 #endif /* INET6 */ 2912 } 2913 2914 /* TCP header */ 2915 th->th_sport = sport; 2916 th->th_dport = dport; 2917 th->th_seq = htonl(seq); 2918 th->th_ack = htonl(ack); 2919 th->th_off = tlen >> 2; 2920 th->th_flags = flags; 2921 th->th_win = htons(win); 2922 2923 if (mss) { 2924 opt = (char *)(th + 1); 2925 opt[0] = TCPOPT_MAXSEG; 2926 opt[1] = 4; 2927 HTONS(mss); 2928 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2929 } 2930 2931 switch (af) { 2932 #ifdef INET 2933 case AF_INET: 2934 /* TCP checksum */ 2935 th->th_sum = in_cksum(m, len); 2936 2937 /* Finish the IP header */ 2938 h->ip_v = 4; 2939 h->ip_hl = sizeof(*h) >> 2; 2940 h->ip_tos = IPTOS_LOWDELAY; 2941 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2942 h->ip_len = htons(len); 2943 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2944 h->ip_sum = 0; 2945 break; 2946 #endif /* INET */ 2947 #ifdef INET6 2948 case AF_INET6: 2949 /* TCP checksum */ 2950 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2951 sizeof(struct ip6_hdr), tlen); 2952 2953 h6->ip6_vfc |= IPV6_VERSION; 2954 h6->ip6_hlim = IPV6_DEFHLIM; 2955 break; 2956 #endif /* INET6 */ 2957 } 2958 2959 return (m); 2960 } 2961 2962 void 2963 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 2964 const struct pf_addr *saddr, const struct pf_addr *daddr, 2965 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2966 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2967 u_int16_t rtag) 2968 { 2969 struct pf_send_entry *pfse; 2970 struct mbuf *m; 2971 2972 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, flags, 2973 win, mss, ttl, tag, rtag); 2974 if (m == NULL) 2975 return; 2976 2977 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2978 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2979 if (pfse == NULL) { 2980 m_freem(m); 2981 return; 2982 } 2983 2984 switch (af) { 2985 #ifdef INET 2986 case AF_INET: 2987 pfse->pfse_type = PFSE_IP; 2988 break; 2989 #endif /* INET */ 2990 #ifdef INET6 2991 case AF_INET6: 2992 pfse->pfse_type = PFSE_IP6; 2993 break; 2994 #endif /* INET6 */ 2995 } 2996 2997 pfse->pfse_m = m; 2998 pf_send(pfse); 2999 } 3000 3001 static void 3002 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3003 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3004 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3005 u_short *reason) 3006 { 3007 struct pf_addr * const saddr = pd->src; 3008 struct pf_addr * const daddr = pd->dst; 3009 sa_family_t af = pd->af; 3010 3011 /* undo NAT changes, if they have taken place */ 3012 if (nr != NULL) { 3013 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3014 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3015 if (pd->sport) 3016 *pd->sport = sk->port[pd->sidx]; 3017 if (pd->dport) 3018 *pd->dport = sk->port[pd->didx]; 3019 if (pd->proto_sum) 3020 *pd->proto_sum = bproto_sum; 3021 if (pd->ip_sum) 3022 *pd->ip_sum = bip_sum; 3023 m_copyback(m, off, hdrlen, pd->hdr.any); 3024 } 3025 if (pd->proto == IPPROTO_TCP && 3026 ((r->rule_flag & PFRULE_RETURNRST) || 3027 (r->rule_flag & PFRULE_RETURN)) && 3028 !(th->th_flags & TH_RST)) { 3029 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3030 int len = 0; 3031 #ifdef INET 3032 struct ip *h4; 3033 #endif 3034 #ifdef INET6 3035 struct ip6_hdr *h6; 3036 #endif 3037 3038 switch (af) { 3039 #ifdef INET 3040 case AF_INET: 3041 h4 = mtod(m, struct ip *); 3042 len = ntohs(h4->ip_len) - off; 3043 break; 3044 #endif 3045 #ifdef INET6 3046 case AF_INET6: 3047 h6 = mtod(m, struct ip6_hdr *); 3048 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3049 break; 3050 #endif 3051 } 3052 3053 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3054 REASON_SET(reason, PFRES_PROTCKSUM); 3055 else { 3056 if (th->th_flags & TH_SYN) 3057 ack++; 3058 if (th->th_flags & TH_FIN) 3059 ack++; 3060 pf_send_tcp(r, af, pd->dst, 3061 pd->src, th->th_dport, th->th_sport, 3062 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3063 r->return_ttl, 1, 0); 3064 } 3065 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3066 r->return_icmp) 3067 pf_send_icmp(m, r->return_icmp >> 8, 3068 r->return_icmp & 255, af, r); 3069 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3070 r->return_icmp6) 3071 pf_send_icmp(m, r->return_icmp6 >> 8, 3072 r->return_icmp6 & 255, af, r); 3073 } 3074 3075 static int 3076 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3077 { 3078 struct m_tag *mtag; 3079 u_int8_t mpcp; 3080 3081 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3082 if (mtag == NULL) 3083 return (0); 3084 3085 if (prio == PF_PRIO_ZERO) 3086 prio = 0; 3087 3088 mpcp = *(uint8_t *)(mtag + 1); 3089 3090 return (mpcp == prio); 3091 } 3092 3093 static void 3094 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3095 struct pf_krule *r) 3096 { 3097 struct pf_send_entry *pfse; 3098 struct mbuf *m0; 3099 struct pf_mtag *pf_mtag; 3100 3101 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3102 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3103 if (pfse == NULL) 3104 return; 3105 3106 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3107 free(pfse, M_PFTEMP); 3108 return; 3109 } 3110 3111 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3112 free(pfse, M_PFTEMP); 3113 return; 3114 } 3115 /* XXX: revisit */ 3116 m0->m_flags |= M_SKIP_FIREWALL; 3117 3118 if (r->rtableid >= 0) 3119 M_SETFIB(m0, r->rtableid); 3120 3121 #ifdef ALTQ 3122 if (r->qid) { 3123 pf_mtag->qid = r->qid; 3124 /* add hints for ecn */ 3125 pf_mtag->hdr = mtod(m0, struct ip *); 3126 } 3127 #endif /* ALTQ */ 3128 3129 switch (af) { 3130 #ifdef INET 3131 case AF_INET: 3132 pfse->pfse_type = PFSE_ICMP; 3133 break; 3134 #endif /* INET */ 3135 #ifdef INET6 3136 case AF_INET6: 3137 pfse->pfse_type = PFSE_ICMP6; 3138 break; 3139 #endif /* INET6 */ 3140 } 3141 pfse->pfse_m = m0; 3142 pfse->icmpopts.type = type; 3143 pfse->icmpopts.code = code; 3144 pf_send(pfse); 3145 } 3146 3147 /* 3148 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3149 * If n is 0, they match if they are equal. If n is != 0, they match if they 3150 * are different. 3151 */ 3152 int 3153 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3154 struct pf_addr *b, sa_family_t af) 3155 { 3156 int match = 0; 3157 3158 switch (af) { 3159 #ifdef INET 3160 case AF_INET: 3161 if ((a->addr32[0] & m->addr32[0]) == 3162 (b->addr32[0] & m->addr32[0])) 3163 match++; 3164 break; 3165 #endif /* INET */ 3166 #ifdef INET6 3167 case AF_INET6: 3168 if (((a->addr32[0] & m->addr32[0]) == 3169 (b->addr32[0] & m->addr32[0])) && 3170 ((a->addr32[1] & m->addr32[1]) == 3171 (b->addr32[1] & m->addr32[1])) && 3172 ((a->addr32[2] & m->addr32[2]) == 3173 (b->addr32[2] & m->addr32[2])) && 3174 ((a->addr32[3] & m->addr32[3]) == 3175 (b->addr32[3] & m->addr32[3]))) 3176 match++; 3177 break; 3178 #endif /* INET6 */ 3179 } 3180 if (match) { 3181 if (n) 3182 return (0); 3183 else 3184 return (1); 3185 } else { 3186 if (n) 3187 return (1); 3188 else 3189 return (0); 3190 } 3191 } 3192 3193 /* 3194 * Return 1 if b <= a <= e, otherwise return 0. 3195 */ 3196 int 3197 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3198 struct pf_addr *a, sa_family_t af) 3199 { 3200 switch (af) { 3201 #ifdef INET 3202 case AF_INET: 3203 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3204 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3205 return (0); 3206 break; 3207 #endif /* INET */ 3208 #ifdef INET6 3209 case AF_INET6: { 3210 int i; 3211 3212 /* check a >= b */ 3213 for (i = 0; i < 4; ++i) 3214 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3215 break; 3216 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3217 return (0); 3218 /* check a <= e */ 3219 for (i = 0; i < 4; ++i) 3220 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3221 break; 3222 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3223 return (0); 3224 break; 3225 } 3226 #endif /* INET6 */ 3227 } 3228 return (1); 3229 } 3230 3231 static int 3232 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3233 { 3234 switch (op) { 3235 case PF_OP_IRG: 3236 return ((p > a1) && (p < a2)); 3237 case PF_OP_XRG: 3238 return ((p < a1) || (p > a2)); 3239 case PF_OP_RRG: 3240 return ((p >= a1) && (p <= a2)); 3241 case PF_OP_EQ: 3242 return (p == a1); 3243 case PF_OP_NE: 3244 return (p != a1); 3245 case PF_OP_LT: 3246 return (p < a1); 3247 case PF_OP_LE: 3248 return (p <= a1); 3249 case PF_OP_GT: 3250 return (p > a1); 3251 case PF_OP_GE: 3252 return (p >= a1); 3253 } 3254 return (0); /* never reached */ 3255 } 3256 3257 int 3258 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3259 { 3260 NTOHS(a1); 3261 NTOHS(a2); 3262 NTOHS(p); 3263 return (pf_match(op, a1, a2, p)); 3264 } 3265 3266 static int 3267 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3268 { 3269 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3270 return (0); 3271 return (pf_match(op, a1, a2, u)); 3272 } 3273 3274 static int 3275 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3276 { 3277 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3278 return (0); 3279 return (pf_match(op, a1, a2, g)); 3280 } 3281 3282 int 3283 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3284 { 3285 if (*tag == -1) 3286 *tag = mtag; 3287 3288 return ((!r->match_tag_not && r->match_tag == *tag) || 3289 (r->match_tag_not && r->match_tag != *tag)); 3290 } 3291 3292 int 3293 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3294 { 3295 3296 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3297 3298 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3299 return (ENOMEM); 3300 3301 pd->pf_mtag->tag = tag; 3302 3303 return (0); 3304 } 3305 3306 #define PF_ANCHOR_STACKSIZE 32 3307 struct pf_kanchor_stackframe { 3308 struct pf_kruleset *rs; 3309 struct pf_krule *r; /* XXX: + match bit */ 3310 struct pf_kanchor *child; 3311 }; 3312 3313 /* 3314 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3315 */ 3316 #define PF_ANCHORSTACK_MATCH 0x00000001 3317 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3318 3319 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3320 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3321 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3322 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3323 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3324 } while (0) 3325 3326 void 3327 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3328 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3329 int *match) 3330 { 3331 struct pf_kanchor_stackframe *f; 3332 3333 PF_RULES_RASSERT(); 3334 3335 if (match) 3336 *match = 0; 3337 if (*depth >= PF_ANCHOR_STACKSIZE) { 3338 printf("%s: anchor stack overflow on %s\n", 3339 __func__, (*r)->anchor->name); 3340 *r = TAILQ_NEXT(*r, entries); 3341 return; 3342 } else if (*depth == 0 && a != NULL) 3343 *a = *r; 3344 f = stack + (*depth)++; 3345 f->rs = *rs; 3346 f->r = *r; 3347 if ((*r)->anchor_wildcard) { 3348 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3349 3350 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3351 *r = NULL; 3352 return; 3353 } 3354 *rs = &f->child->ruleset; 3355 } else { 3356 f->child = NULL; 3357 *rs = &(*r)->anchor->ruleset; 3358 } 3359 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3360 } 3361 3362 int 3363 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3364 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3365 int *match) 3366 { 3367 struct pf_kanchor_stackframe *f; 3368 struct pf_krule *fr; 3369 int quick = 0; 3370 3371 PF_RULES_RASSERT(); 3372 3373 do { 3374 if (*depth <= 0) 3375 break; 3376 f = stack + *depth - 1; 3377 fr = PF_ANCHOR_RULE(f); 3378 if (f->child != NULL) { 3379 /* 3380 * This block traverses through 3381 * a wildcard anchor. 3382 */ 3383 if (match != NULL && *match) { 3384 /* 3385 * If any of "*" matched, then 3386 * "foo/ *" matched, mark frame 3387 * appropriately. 3388 */ 3389 PF_ANCHOR_SET_MATCH(f); 3390 *match = 0; 3391 } 3392 f->child = RB_NEXT(pf_kanchor_node, 3393 &fr->anchor->children, f->child); 3394 if (f->child != NULL) { 3395 *rs = &f->child->ruleset; 3396 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3397 if (*r == NULL) 3398 continue; 3399 else 3400 break; 3401 } 3402 } 3403 (*depth)--; 3404 if (*depth == 0 && a != NULL) 3405 *a = NULL; 3406 *rs = f->rs; 3407 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3408 quick = fr->quick; 3409 *r = TAILQ_NEXT(fr, entries); 3410 } while (*r == NULL); 3411 3412 return (quick); 3413 } 3414 3415 struct pf_keth_anchor_stackframe { 3416 struct pf_keth_ruleset *rs; 3417 struct pf_keth_rule *r; /* XXX: + match bit */ 3418 struct pf_keth_anchor *child; 3419 }; 3420 3421 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3422 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3423 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3424 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3425 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3426 } while (0) 3427 3428 void 3429 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3430 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3431 struct pf_keth_rule **a, int *match) 3432 { 3433 struct pf_keth_anchor_stackframe *f; 3434 3435 NET_EPOCH_ASSERT(); 3436 3437 if (match) 3438 *match = 0; 3439 if (*depth >= PF_ANCHOR_STACKSIZE) { 3440 printf("%s: anchor stack overflow on %s\n", 3441 __func__, (*r)->anchor->name); 3442 *r = TAILQ_NEXT(*r, entries); 3443 return; 3444 } else if (*depth == 0 && a != NULL) 3445 *a = *r; 3446 f = stack + (*depth)++; 3447 f->rs = *rs; 3448 f->r = *r; 3449 if ((*r)->anchor_wildcard) { 3450 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3451 3452 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3453 *r = NULL; 3454 return; 3455 } 3456 *rs = &f->child->ruleset; 3457 } else { 3458 f->child = NULL; 3459 *rs = &(*r)->anchor->ruleset; 3460 } 3461 *r = TAILQ_FIRST((*rs)->active.rules); 3462 } 3463 3464 int 3465 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3466 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3467 struct pf_keth_rule **a, int *match) 3468 { 3469 struct pf_keth_anchor_stackframe *f; 3470 struct pf_keth_rule *fr; 3471 int quick = 0; 3472 3473 NET_EPOCH_ASSERT(); 3474 3475 do { 3476 if (*depth <= 0) 3477 break; 3478 f = stack + *depth - 1; 3479 fr = PF_ETH_ANCHOR_RULE(f); 3480 if (f->child != NULL) { 3481 /* 3482 * This block traverses through 3483 * a wildcard anchor. 3484 */ 3485 if (match != NULL && *match) { 3486 /* 3487 * If any of "*" matched, then 3488 * "foo/ *" matched, mark frame 3489 * appropriately. 3490 */ 3491 PF_ETH_ANCHOR_SET_MATCH(f); 3492 *match = 0; 3493 } 3494 f->child = RB_NEXT(pf_keth_anchor_node, 3495 &fr->anchor->children, f->child); 3496 if (f->child != NULL) { 3497 *rs = &f->child->ruleset; 3498 *r = TAILQ_FIRST((*rs)->active.rules); 3499 if (*r == NULL) 3500 continue; 3501 else 3502 break; 3503 } 3504 } 3505 (*depth)--; 3506 if (*depth == 0 && a != NULL) 3507 *a = NULL; 3508 *rs = f->rs; 3509 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3510 quick = fr->quick; 3511 *r = TAILQ_NEXT(fr, entries); 3512 } while (*r == NULL); 3513 3514 return (quick); 3515 } 3516 3517 #ifdef INET6 3518 void 3519 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3520 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3521 { 3522 switch (af) { 3523 #ifdef INET 3524 case AF_INET: 3525 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3526 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3527 break; 3528 #endif /* INET */ 3529 case AF_INET6: 3530 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3531 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3532 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3533 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3534 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3535 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3536 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3537 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3538 break; 3539 } 3540 } 3541 3542 void 3543 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3544 { 3545 switch (af) { 3546 #ifdef INET 3547 case AF_INET: 3548 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3549 break; 3550 #endif /* INET */ 3551 case AF_INET6: 3552 if (addr->addr32[3] == 0xffffffff) { 3553 addr->addr32[3] = 0; 3554 if (addr->addr32[2] == 0xffffffff) { 3555 addr->addr32[2] = 0; 3556 if (addr->addr32[1] == 0xffffffff) { 3557 addr->addr32[1] = 0; 3558 addr->addr32[0] = 3559 htonl(ntohl(addr->addr32[0]) + 1); 3560 } else 3561 addr->addr32[1] = 3562 htonl(ntohl(addr->addr32[1]) + 1); 3563 } else 3564 addr->addr32[2] = 3565 htonl(ntohl(addr->addr32[2]) + 1); 3566 } else 3567 addr->addr32[3] = 3568 htonl(ntohl(addr->addr32[3]) + 1); 3569 break; 3570 } 3571 } 3572 #endif /* INET6 */ 3573 3574 void 3575 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3576 { 3577 if (r->qid) 3578 a->qid = r->qid; 3579 if (r->pqid) 3580 a->pqid = r->pqid; 3581 if (r->dnpipe) 3582 a->dnpipe = r->dnpipe; 3583 if (r->dnrpipe) 3584 a->dnrpipe = r->dnrpipe; 3585 if (r->dnpipe || r->dnrpipe) { 3586 if (r->free_flags & PFRULE_DN_IS_PIPE) 3587 a->flags |= PFRULE_DN_IS_PIPE; 3588 else 3589 a->flags &= ~PFRULE_DN_IS_PIPE; 3590 } 3591 } 3592 3593 int 3594 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 3595 { 3596 struct pf_addr *saddr, *daddr; 3597 u_int16_t sport, dport; 3598 struct inpcbinfo *pi; 3599 struct inpcb *inp; 3600 3601 pd->lookup.uid = UID_MAX; 3602 pd->lookup.gid = GID_MAX; 3603 3604 switch (pd->proto) { 3605 case IPPROTO_TCP: 3606 sport = pd->hdr.tcp.th_sport; 3607 dport = pd->hdr.tcp.th_dport; 3608 pi = &V_tcbinfo; 3609 break; 3610 case IPPROTO_UDP: 3611 sport = pd->hdr.udp.uh_sport; 3612 dport = pd->hdr.udp.uh_dport; 3613 pi = &V_udbinfo; 3614 break; 3615 default: 3616 return (-1); 3617 } 3618 if (direction == PF_IN) { 3619 saddr = pd->src; 3620 daddr = pd->dst; 3621 } else { 3622 u_int16_t p; 3623 3624 p = sport; 3625 sport = dport; 3626 dport = p; 3627 saddr = pd->dst; 3628 daddr = pd->src; 3629 } 3630 switch (pd->af) { 3631 #ifdef INET 3632 case AF_INET: 3633 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3634 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3635 if (inp == NULL) { 3636 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3637 daddr->v4, dport, INPLOOKUP_WILDCARD | 3638 INPLOOKUP_RLOCKPCB, NULL, m); 3639 if (inp == NULL) 3640 return (-1); 3641 } 3642 break; 3643 #endif /* INET */ 3644 #ifdef INET6 3645 case AF_INET6: 3646 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3647 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3648 if (inp == NULL) { 3649 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3650 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3651 INPLOOKUP_RLOCKPCB, NULL, m); 3652 if (inp == NULL) 3653 return (-1); 3654 } 3655 break; 3656 #endif /* INET6 */ 3657 3658 default: 3659 return (-1); 3660 } 3661 INP_RLOCK_ASSERT(inp); 3662 pd->lookup.uid = inp->inp_cred->cr_uid; 3663 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3664 INP_RUNLOCK(inp); 3665 3666 return (1); 3667 } 3668 3669 u_int8_t 3670 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3671 { 3672 int hlen; 3673 u_int8_t hdr[60]; 3674 u_int8_t *opt, optlen; 3675 u_int8_t wscale = 0; 3676 3677 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3678 if (hlen <= sizeof(struct tcphdr)) 3679 return (0); 3680 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3681 return (0); 3682 opt = hdr + sizeof(struct tcphdr); 3683 hlen -= sizeof(struct tcphdr); 3684 while (hlen >= 3) { 3685 switch (*opt) { 3686 case TCPOPT_EOL: 3687 case TCPOPT_NOP: 3688 ++opt; 3689 --hlen; 3690 break; 3691 case TCPOPT_WINDOW: 3692 wscale = opt[2]; 3693 if (wscale > TCP_MAX_WINSHIFT) 3694 wscale = TCP_MAX_WINSHIFT; 3695 wscale |= PF_WSCALE_FLAG; 3696 /* FALLTHROUGH */ 3697 default: 3698 optlen = opt[1]; 3699 if (optlen < 2) 3700 optlen = 2; 3701 hlen -= optlen; 3702 opt += optlen; 3703 break; 3704 } 3705 } 3706 return (wscale); 3707 } 3708 3709 u_int16_t 3710 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3711 { 3712 int hlen; 3713 u_int8_t hdr[60]; 3714 u_int8_t *opt, optlen; 3715 u_int16_t mss = V_tcp_mssdflt; 3716 3717 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3718 if (hlen <= sizeof(struct tcphdr)) 3719 return (0); 3720 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3721 return (0); 3722 opt = hdr + sizeof(struct tcphdr); 3723 hlen -= sizeof(struct tcphdr); 3724 while (hlen >= TCPOLEN_MAXSEG) { 3725 switch (*opt) { 3726 case TCPOPT_EOL: 3727 case TCPOPT_NOP: 3728 ++opt; 3729 --hlen; 3730 break; 3731 case TCPOPT_MAXSEG: 3732 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3733 NTOHS(mss); 3734 /* FALLTHROUGH */ 3735 default: 3736 optlen = opt[1]; 3737 if (optlen < 2) 3738 optlen = 2; 3739 hlen -= optlen; 3740 opt += optlen; 3741 break; 3742 } 3743 } 3744 return (mss); 3745 } 3746 3747 static u_int16_t 3748 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3749 { 3750 struct nhop_object *nh; 3751 #ifdef INET6 3752 struct in6_addr dst6; 3753 uint32_t scopeid; 3754 #endif /* INET6 */ 3755 int hlen = 0; 3756 uint16_t mss = 0; 3757 3758 NET_EPOCH_ASSERT(); 3759 3760 switch (af) { 3761 #ifdef INET 3762 case AF_INET: 3763 hlen = sizeof(struct ip); 3764 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3765 if (nh != NULL) 3766 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3767 break; 3768 #endif /* INET */ 3769 #ifdef INET6 3770 case AF_INET6: 3771 hlen = sizeof(struct ip6_hdr); 3772 in6_splitscope(&addr->v6, &dst6, &scopeid); 3773 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 3774 if (nh != NULL) 3775 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3776 break; 3777 #endif /* INET6 */ 3778 } 3779 3780 mss = max(V_tcp_mssdflt, mss); 3781 mss = min(mss, offer); 3782 mss = max(mss, 64); /* sanity - at least max opt space */ 3783 return (mss); 3784 } 3785 3786 static u_int32_t 3787 pf_tcp_iss(struct pf_pdesc *pd) 3788 { 3789 MD5_CTX ctx; 3790 u_int32_t digest[4]; 3791 3792 if (V_pf_tcp_secret_init == 0) { 3793 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3794 MD5Init(&V_pf_tcp_secret_ctx); 3795 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3796 sizeof(V_pf_tcp_secret)); 3797 V_pf_tcp_secret_init = 1; 3798 } 3799 3800 ctx = V_pf_tcp_secret_ctx; 3801 3802 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 3803 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 3804 if (pd->af == AF_INET6) { 3805 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3806 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3807 } else { 3808 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3809 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3810 } 3811 MD5Final((u_char *)digest, &ctx); 3812 V_pf_tcp_iss_off += 4096; 3813 #define ISN_RANDOM_INCREMENT (4096 - 1) 3814 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3815 V_pf_tcp_iss_off); 3816 #undef ISN_RANDOM_INCREMENT 3817 } 3818 3819 static bool 3820 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 3821 { 3822 bool match = true; 3823 3824 /* Always matches if not set */ 3825 if (! r->isset) 3826 return (!r->neg); 3827 3828 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 3829 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 3830 match = false; 3831 break; 3832 } 3833 } 3834 3835 return (match ^ r->neg); 3836 } 3837 3838 static int 3839 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 3840 { 3841 if (*tag == -1) 3842 *tag = mtag; 3843 3844 return ((!r->match_tag_not && r->match_tag == *tag) || 3845 (r->match_tag_not && r->match_tag != *tag)); 3846 } 3847 3848 static int 3849 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 3850 { 3851 #ifdef INET 3852 struct ip ip; 3853 #endif 3854 #ifdef INET6 3855 struct ip6_hdr ip6; 3856 #endif 3857 struct mbuf *m = *m0; 3858 struct ether_header *e; 3859 struct pf_keth_rule *r, *rm, *a = NULL; 3860 struct pf_keth_ruleset *ruleset = NULL; 3861 struct pf_mtag *mtag; 3862 struct pf_keth_ruleq *rules; 3863 struct pf_addr *src = NULL, *dst = NULL; 3864 sa_family_t af = 0; 3865 uint16_t proto; 3866 int asd = 0, match = 0; 3867 int tag = -1; 3868 uint8_t action; 3869 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3870 3871 MPASS(kif->pfik_ifp->if_vnet == curvnet); 3872 NET_EPOCH_ASSERT(); 3873 3874 PF_RULES_RLOCK_TRACKER; 3875 3876 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 3877 3878 mtag = pf_find_mtag(m); 3879 if (mtag != NULL && mtag->flags & PF_TAG_DUMMYNET) { 3880 /* Dummynet re-injects packets after they've 3881 * completed their delay. We've already 3882 * processed them, so pass unconditionally. */ 3883 3884 /* But only once. We may see the packet multiple times (e.g. 3885 * PFIL_IN/PFIL_OUT). */ 3886 mtag->flags &= ~PF_TAG_DUMMYNET; 3887 3888 return (PF_PASS); 3889 } 3890 3891 ruleset = V_pf_keth; 3892 rules = ck_pr_load_ptr(&ruleset->active.rules); 3893 r = TAILQ_FIRST(rules); 3894 rm = NULL; 3895 3896 e = mtod(m, struct ether_header *); 3897 proto = ntohs(e->ether_type); 3898 3899 switch (proto) { 3900 #ifdef INET 3901 case ETHERTYPE_IP: { 3902 if (m_length(m, NULL) < (sizeof(struct ether_header) + 3903 sizeof(ip))) 3904 return (PF_DROP); 3905 3906 af = AF_INET; 3907 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 3908 (caddr_t)&ip); 3909 src = (struct pf_addr *)&ip.ip_src; 3910 dst = (struct pf_addr *)&ip.ip_dst; 3911 break; 3912 } 3913 #endif /* INET */ 3914 #ifdef INET6 3915 case ETHERTYPE_IPV6: { 3916 if (m_length(m, NULL) < (sizeof(struct ether_header) + 3917 sizeof(ip6))) 3918 return (PF_DROP); 3919 3920 af = AF_INET6; 3921 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 3922 (caddr_t)&ip6); 3923 src = (struct pf_addr *)&ip6.ip6_src; 3924 dst = (struct pf_addr *)&ip6.ip6_dst; 3925 break; 3926 } 3927 #endif /* INET6 */ 3928 } 3929 3930 PF_RULES_RLOCK(); 3931 3932 while (r != NULL) { 3933 counter_u64_add(r->evaluations, 1); 3934 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 3935 3936 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 3937 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3938 "kif"); 3939 r = r->skip[PFE_SKIP_IFP].ptr; 3940 } 3941 else if (r->direction && r->direction != dir) { 3942 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3943 "dir"); 3944 r = r->skip[PFE_SKIP_DIR].ptr; 3945 } 3946 else if (r->proto && r->proto != proto) { 3947 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3948 "proto"); 3949 r = r->skip[PFE_SKIP_PROTO].ptr; 3950 } 3951 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 3952 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3953 "src"); 3954 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 3955 } 3956 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 3957 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3958 "dst"); 3959 r = TAILQ_NEXT(r, entries); 3960 } 3961 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 3962 r->ipsrc.neg, kif, M_GETFIB(m))) { 3963 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3964 "ip_src"); 3965 r = TAILQ_NEXT(r, entries); 3966 } 3967 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 3968 r->ipdst.neg, kif, M_GETFIB(m))) { 3969 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3970 "ip_dst"); 3971 r = TAILQ_NEXT(r, entries); 3972 } 3973 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 3974 mtag ? mtag->tag : 0)) { 3975 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 3976 "match_tag"); 3977 r = TAILQ_NEXT(r, entries); 3978 } 3979 else { 3980 if (r->tag) 3981 tag = r->tag; 3982 if (r->anchor == NULL) { 3983 /* Rule matches */ 3984 rm = r; 3985 3986 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 3987 3988 if (r->quick) 3989 break; 3990 3991 r = TAILQ_NEXT(r, entries); 3992 } else { 3993 pf_step_into_keth_anchor(anchor_stack, &asd, 3994 &ruleset, &r, &a, &match); 3995 } 3996 } 3997 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 3998 &ruleset, &r, &a, &match)) 3999 break; 4000 } 4001 4002 r = rm; 4003 4004 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4005 4006 /* Default to pass. */ 4007 if (r == NULL) { 4008 PF_RULES_RUNLOCK(); 4009 return (PF_PASS); 4010 } 4011 4012 /* Execute action. */ 4013 counter_u64_add(r->packets[dir == PF_OUT], 1); 4014 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4015 pf_update_timestamp(r); 4016 4017 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4018 if (r->action == PF_DROP) { 4019 PF_RULES_RUNLOCK(); 4020 return (PF_DROP); 4021 } 4022 4023 if (tag > 0) { 4024 if (mtag == NULL) 4025 mtag = pf_get_mtag(m); 4026 if (mtag == NULL) { 4027 PF_RULES_RUNLOCK(); 4028 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4029 return (PF_DROP); 4030 } 4031 mtag->tag = tag; 4032 } 4033 4034 if (r->qid != 0) { 4035 if (mtag == NULL) 4036 mtag = pf_get_mtag(m); 4037 if (mtag == NULL) { 4038 PF_RULES_RUNLOCK(); 4039 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4040 return (PF_DROP); 4041 } 4042 mtag->qid = r->qid; 4043 } 4044 4045 /* Dummynet */ 4046 if (r->dnpipe) { 4047 struct ip_fw_args dnflow; 4048 4049 /* Drop packet if dummynet is not loaded. */ 4050 if (ip_dn_io_ptr == NULL) { 4051 PF_RULES_RUNLOCK(); 4052 m_freem(m); 4053 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4054 return (PF_DROP); 4055 } 4056 if (mtag == NULL) 4057 mtag = pf_get_mtag(m); 4058 if (mtag == NULL) { 4059 PF_RULES_RUNLOCK(); 4060 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4061 return (PF_DROP); 4062 } 4063 4064 bzero(&dnflow, sizeof(dnflow)); 4065 4066 /* We don't have port numbers here, so we set 0. That means 4067 * that we'll be somewhat limited in distinguishing flows (i.e. 4068 * only based on IP addresses, not based on port numbers), but 4069 * it's better than nothing. */ 4070 dnflow.f_id.dst_port = 0; 4071 dnflow.f_id.src_port = 0; 4072 dnflow.f_id.proto = 0; 4073 4074 dnflow.rule.info = r->dnpipe; 4075 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4076 if (r->dnflags & PFRULE_DN_IS_PIPE) 4077 dnflow.rule.info |= IPFW_IS_PIPE; 4078 4079 dnflow.f_id.extra = dnflow.rule.info; 4080 4081 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4082 dnflow.flags |= IPFW_ARGS_ETHER; 4083 dnflow.ifp = kif->pfik_ifp; 4084 4085 switch (af) { 4086 case AF_INET: 4087 dnflow.f_id.addr_type = 4; 4088 dnflow.f_id.src_ip = src->v4.s_addr; 4089 dnflow.f_id.dst_ip = dst->v4.s_addr; 4090 break; 4091 case AF_INET6: 4092 dnflow.flags |= IPFW_ARGS_IP6; 4093 dnflow.f_id.addr_type = 6; 4094 dnflow.f_id.src_ip6 = src->v6; 4095 dnflow.f_id.dst_ip6 = dst->v6; 4096 break; 4097 } 4098 4099 mtag->flags |= PF_TAG_DUMMYNET; 4100 ip_dn_io_ptr(m0, &dnflow); 4101 if (*m0 != NULL) 4102 mtag->flags &= ~PF_TAG_DUMMYNET; 4103 } 4104 4105 action = r->action; 4106 4107 PF_RULES_RUNLOCK(); 4108 4109 return (action); 4110 } 4111 4112 static int 4113 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, int direction, 4114 struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 4115 struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp) 4116 { 4117 struct pf_krule *nr = NULL; 4118 struct pf_addr * const saddr = pd->src; 4119 struct pf_addr * const daddr = pd->dst; 4120 sa_family_t af = pd->af; 4121 struct pf_krule *r, *a = NULL; 4122 struct pf_kruleset *ruleset = NULL; 4123 struct pf_ksrc_node *nsn = NULL; 4124 struct tcphdr *th = &pd->hdr.tcp; 4125 struct pf_state_key *sk = NULL, *nk = NULL; 4126 u_short reason; 4127 int rewrite = 0, hdrlen = 0; 4128 int tag = -1, rtableid = -1; 4129 int asd = 0; 4130 int match = 0; 4131 int state_icmp = 0; 4132 u_int16_t sport = 0, dport = 0; 4133 u_int16_t bproto_sum = 0, bip_sum = 0; 4134 u_int8_t icmptype = 0, icmpcode = 0; 4135 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4136 4137 PF_RULES_RASSERT(); 4138 4139 if (inp != NULL) { 4140 INP_LOCK_ASSERT(inp); 4141 pd->lookup.uid = inp->inp_cred->cr_uid; 4142 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4143 pd->lookup.done = 1; 4144 } 4145 4146 switch (pd->proto) { 4147 case IPPROTO_TCP: 4148 sport = th->th_sport; 4149 dport = th->th_dport; 4150 hdrlen = sizeof(*th); 4151 break; 4152 case IPPROTO_UDP: 4153 sport = pd->hdr.udp.uh_sport; 4154 dport = pd->hdr.udp.uh_dport; 4155 hdrlen = sizeof(pd->hdr.udp); 4156 break; 4157 #ifdef INET 4158 case IPPROTO_ICMP: 4159 if (pd->af != AF_INET) 4160 break; 4161 sport = dport = pd->hdr.icmp.icmp_id; 4162 hdrlen = sizeof(pd->hdr.icmp); 4163 icmptype = pd->hdr.icmp.icmp_type; 4164 icmpcode = pd->hdr.icmp.icmp_code; 4165 4166 if (icmptype == ICMP_UNREACH || 4167 icmptype == ICMP_SOURCEQUENCH || 4168 icmptype == ICMP_REDIRECT || 4169 icmptype == ICMP_TIMXCEED || 4170 icmptype == ICMP_PARAMPROB) 4171 state_icmp++; 4172 break; 4173 #endif /* INET */ 4174 #ifdef INET6 4175 case IPPROTO_ICMPV6: 4176 if (af != AF_INET6) 4177 break; 4178 sport = dport = pd->hdr.icmp6.icmp6_id; 4179 hdrlen = sizeof(pd->hdr.icmp6); 4180 icmptype = pd->hdr.icmp6.icmp6_type; 4181 icmpcode = pd->hdr.icmp6.icmp6_code; 4182 4183 if (icmptype == ICMP6_DST_UNREACH || 4184 icmptype == ICMP6_PACKET_TOO_BIG || 4185 icmptype == ICMP6_TIME_EXCEEDED || 4186 icmptype == ICMP6_PARAM_PROB) 4187 state_icmp++; 4188 break; 4189 #endif /* INET6 */ 4190 default: 4191 sport = dport = hdrlen = 0; 4192 break; 4193 } 4194 4195 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4196 4197 /* check packet for BINAT/NAT/RDR */ 4198 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 4199 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4200 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4201 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4202 4203 if (nr->log) { 4204 PFLOG_PACKET(kif, m, af, direction, PFRES_MATCH, nr, a, 4205 ruleset, pd, 1); 4206 } 4207 4208 if (pd->ip_sum) 4209 bip_sum = *pd->ip_sum; 4210 4211 switch (pd->proto) { 4212 case IPPROTO_TCP: 4213 bproto_sum = th->th_sum; 4214 pd->proto_sum = &th->th_sum; 4215 4216 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4217 nk->port[pd->sidx] != sport) { 4218 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4219 &th->th_sum, &nk->addr[pd->sidx], 4220 nk->port[pd->sidx], 0, af); 4221 pd->sport = &th->th_sport; 4222 sport = th->th_sport; 4223 } 4224 4225 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4226 nk->port[pd->didx] != dport) { 4227 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4228 &th->th_sum, &nk->addr[pd->didx], 4229 nk->port[pd->didx], 0, af); 4230 dport = th->th_dport; 4231 pd->dport = &th->th_dport; 4232 } 4233 rewrite++; 4234 break; 4235 case IPPROTO_UDP: 4236 bproto_sum = pd->hdr.udp.uh_sum; 4237 pd->proto_sum = &pd->hdr.udp.uh_sum; 4238 4239 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4240 nk->port[pd->sidx] != sport) { 4241 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4242 pd->ip_sum, &pd->hdr.udp.uh_sum, 4243 &nk->addr[pd->sidx], 4244 nk->port[pd->sidx], 1, af); 4245 sport = pd->hdr.udp.uh_sport; 4246 pd->sport = &pd->hdr.udp.uh_sport; 4247 } 4248 4249 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4250 nk->port[pd->didx] != dport) { 4251 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4252 pd->ip_sum, &pd->hdr.udp.uh_sum, 4253 &nk->addr[pd->didx], 4254 nk->port[pd->didx], 1, af); 4255 dport = pd->hdr.udp.uh_dport; 4256 pd->dport = &pd->hdr.udp.uh_dport; 4257 } 4258 rewrite++; 4259 break; 4260 #ifdef INET 4261 case IPPROTO_ICMP: 4262 nk->port[0] = nk->port[1]; 4263 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4264 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4265 nk->addr[pd->sidx].v4.s_addr, 0); 4266 4267 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4268 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4269 nk->addr[pd->didx].v4.s_addr, 0); 4270 4271 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4272 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4273 pd->hdr.icmp.icmp_cksum, sport, 4274 nk->port[1], 0); 4275 pd->hdr.icmp.icmp_id = nk->port[1]; 4276 pd->sport = &pd->hdr.icmp.icmp_id; 4277 } 4278 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4279 break; 4280 #endif /* INET */ 4281 #ifdef INET6 4282 case IPPROTO_ICMPV6: 4283 nk->port[0] = nk->port[1]; 4284 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4285 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4286 &nk->addr[pd->sidx], 0); 4287 4288 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4289 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4290 &nk->addr[pd->didx], 0); 4291 rewrite++; 4292 break; 4293 #endif /* INET */ 4294 default: 4295 switch (af) { 4296 #ifdef INET 4297 case AF_INET: 4298 if (PF_ANEQ(saddr, 4299 &nk->addr[pd->sidx], AF_INET)) 4300 pf_change_a(&saddr->v4.s_addr, 4301 pd->ip_sum, 4302 nk->addr[pd->sidx].v4.s_addr, 0); 4303 4304 if (PF_ANEQ(daddr, 4305 &nk->addr[pd->didx], AF_INET)) 4306 pf_change_a(&daddr->v4.s_addr, 4307 pd->ip_sum, 4308 nk->addr[pd->didx].v4.s_addr, 0); 4309 break; 4310 #endif /* INET */ 4311 #ifdef INET6 4312 case AF_INET6: 4313 if (PF_ANEQ(saddr, 4314 &nk->addr[pd->sidx], AF_INET6)) 4315 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4316 4317 if (PF_ANEQ(daddr, 4318 &nk->addr[pd->didx], AF_INET6)) 4319 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4320 break; 4321 #endif /* INET */ 4322 } 4323 break; 4324 } 4325 if (nr->natpass) 4326 r = NULL; 4327 pd->nat_rule = nr; 4328 } 4329 4330 while (r != NULL) { 4331 pf_counter_u64_add(&r->evaluations, 1); 4332 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4333 r = r->skip[PF_SKIP_IFP].ptr; 4334 else if (r->direction && r->direction != direction) 4335 r = r->skip[PF_SKIP_DIR].ptr; 4336 else if (r->af && r->af != af) 4337 r = r->skip[PF_SKIP_AF].ptr; 4338 else if (r->proto && r->proto != pd->proto) 4339 r = r->skip[PF_SKIP_PROTO].ptr; 4340 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4341 r->src.neg, kif, M_GETFIB(m))) 4342 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4343 /* tcp/udp only. port_op always 0 in other cases */ 4344 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4345 r->src.port[0], r->src.port[1], sport)) 4346 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4347 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4348 r->dst.neg, NULL, M_GETFIB(m))) 4349 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4350 /* tcp/udp only. port_op always 0 in other cases */ 4351 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4352 r->dst.port[0], r->dst.port[1], dport)) 4353 r = r->skip[PF_SKIP_DST_PORT].ptr; 4354 /* icmp only. type always 0 in other cases */ 4355 else if (r->type && r->type != icmptype + 1) 4356 r = TAILQ_NEXT(r, entries); 4357 /* icmp only. type always 0 in other cases */ 4358 else if (r->code && r->code != icmpcode + 1) 4359 r = TAILQ_NEXT(r, entries); 4360 else if (r->tos && !(r->tos == pd->tos)) 4361 r = TAILQ_NEXT(r, entries); 4362 else if (r->rule_flag & PFRULE_FRAGMENT) 4363 r = TAILQ_NEXT(r, entries); 4364 else if (pd->proto == IPPROTO_TCP && 4365 (r->flagset & th->th_flags) != r->flags) 4366 r = TAILQ_NEXT(r, entries); 4367 /* tcp/udp only. uid.op always 0 in other cases */ 4368 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4369 pf_socket_lookup(direction, pd, m), 1)) && 4370 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4371 pd->lookup.uid)) 4372 r = TAILQ_NEXT(r, entries); 4373 /* tcp/udp only. gid.op always 0 in other cases */ 4374 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4375 pf_socket_lookup(direction, pd, m), 1)) && 4376 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4377 pd->lookup.gid)) 4378 r = TAILQ_NEXT(r, entries); 4379 else if (r->prio && 4380 !pf_match_ieee8021q_pcp(r->prio, m)) 4381 r = TAILQ_NEXT(r, entries); 4382 else if (r->prob && 4383 r->prob <= arc4random()) 4384 r = TAILQ_NEXT(r, entries); 4385 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4386 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4387 r = TAILQ_NEXT(r, entries); 4388 else if (r->os_fingerprint != PF_OSFP_ANY && 4389 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4390 pf_osfp_fingerprint(pd, m, off, th), 4391 r->os_fingerprint))) 4392 r = TAILQ_NEXT(r, entries); 4393 else { 4394 if (r->tag) 4395 tag = r->tag; 4396 if (r->rtableid >= 0) 4397 rtableid = r->rtableid; 4398 if (r->anchor == NULL) { 4399 if (r->action == PF_MATCH) { 4400 pf_counter_u64_critical_enter(); 4401 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 4402 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 4403 pf_counter_u64_critical_exit(); 4404 pf_rule_to_actions(r, &pd->act); 4405 if (r->log) 4406 PFLOG_PACKET(kif, m, af, 4407 direction, PFRES_MATCH, r, 4408 a, ruleset, pd, 1); 4409 } else { 4410 match = 1; 4411 *rm = r; 4412 *am = a; 4413 *rsm = ruleset; 4414 } 4415 if ((*rm)->quick) 4416 break; 4417 r = TAILQ_NEXT(r, entries); 4418 } else 4419 pf_step_into_anchor(anchor_stack, &asd, 4420 &ruleset, PF_RULESET_FILTER, &r, &a, 4421 &match); 4422 } 4423 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4424 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4425 break; 4426 } 4427 r = *rm; 4428 a = *am; 4429 ruleset = *rsm; 4430 4431 REASON_SET(&reason, PFRES_MATCH); 4432 4433 /* apply actions for last matching pass/block rule */ 4434 pf_rule_to_actions(r, &pd->act); 4435 4436 if (r->log) { 4437 if (rewrite) 4438 m_copyback(m, off, hdrlen, pd->hdr.any); 4439 PFLOG_PACKET(kif, m, af, direction, reason, r, a, 4440 ruleset, pd, 1); 4441 } 4442 4443 if ((r->action == PF_DROP) && 4444 ((r->rule_flag & PFRULE_RETURNRST) || 4445 (r->rule_flag & PFRULE_RETURNICMP) || 4446 (r->rule_flag & PFRULE_RETURN))) { 4447 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4448 bip_sum, hdrlen, &reason); 4449 } 4450 4451 if (r->action == PF_DROP) 4452 goto cleanup; 4453 4454 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4455 REASON_SET(&reason, PFRES_MEMORY); 4456 goto cleanup; 4457 } 4458 if (rtableid >= 0) 4459 M_SETFIB(m, rtableid); 4460 4461 if (!state_icmp && (r->keep_state || nr != NULL || 4462 (pd->flags & PFDESC_TCP_NORM))) { 4463 int action; 4464 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4465 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4466 hdrlen); 4467 if (action != PF_PASS) { 4468 if (action == PF_DROP && 4469 (r->rule_flag & PFRULE_RETURN)) 4470 pf_return(r, nr, pd, sk, off, m, th, kif, 4471 bproto_sum, bip_sum, hdrlen, &reason); 4472 return (action); 4473 } 4474 } else { 4475 if (sk != NULL) 4476 uma_zfree(V_pf_state_key_z, sk); 4477 if (nk != NULL) 4478 uma_zfree(V_pf_state_key_z, nk); 4479 } 4480 4481 /* copy back packet headers if we performed NAT operations */ 4482 if (rewrite) 4483 m_copyback(m, off, hdrlen, pd->hdr.any); 4484 4485 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4486 direction == PF_OUT && 4487 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4488 /* 4489 * We want the state created, but we dont 4490 * want to send this in case a partner 4491 * firewall has to know about it to allow 4492 * replies through it. 4493 */ 4494 return (PF_DEFER); 4495 4496 return (PF_PASS); 4497 4498 cleanup: 4499 if (sk != NULL) 4500 uma_zfree(V_pf_state_key_z, sk); 4501 if (nk != NULL) 4502 uma_zfree(V_pf_state_key_z, nk); 4503 return (PF_DROP); 4504 } 4505 4506 static int 4507 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4508 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4509 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4510 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4511 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 4512 { 4513 struct pf_kstate *s = NULL; 4514 struct pf_ksrc_node *sn = NULL; 4515 struct tcphdr *th = &pd->hdr.tcp; 4516 u_int16_t mss = V_tcp_mssdflt; 4517 u_short reason; 4518 4519 /* check maximums */ 4520 if (r->max_states && 4521 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4522 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4523 REASON_SET(&reason, PFRES_MAXSTATES); 4524 goto csfailed; 4525 } 4526 /* src node for filter rule */ 4527 if ((r->rule_flag & PFRULE_SRCTRACK || 4528 r->rpool.opts & PF_POOL_STICKYADDR) && 4529 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 4530 REASON_SET(&reason, PFRES_SRCLIMIT); 4531 goto csfailed; 4532 } 4533 /* src node for translation rule */ 4534 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4535 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 4536 REASON_SET(&reason, PFRES_SRCLIMIT); 4537 goto csfailed; 4538 } 4539 s = pf_alloc_state(M_NOWAIT); 4540 if (s == NULL) { 4541 REASON_SET(&reason, PFRES_MEMORY); 4542 goto csfailed; 4543 } 4544 s->rule.ptr = r; 4545 s->nat_rule.ptr = nr; 4546 s->anchor.ptr = a; 4547 STATE_INC_COUNTERS(s); 4548 if (r->allow_opts) 4549 s->state_flags |= PFSTATE_ALLOWOPTS; 4550 if (r->rule_flag & PFRULE_STATESLOPPY) 4551 s->state_flags |= PFSTATE_SLOPPY; 4552 s->log = r->log & PF_LOG_ALL; 4553 s->sync_state = PFSYNC_S_NONE; 4554 s->qid = pd->act.qid; 4555 s->pqid = pd->act.pqid; 4556 s->dnpipe = pd->act.dnpipe; 4557 s->dnrpipe = pd->act.dnrpipe; 4558 s->state_flags |= pd->act.flags; 4559 if (nr != NULL) 4560 s->log |= nr->log & PF_LOG_ALL; 4561 switch (pd->proto) { 4562 case IPPROTO_TCP: 4563 s->src.seqlo = ntohl(th->th_seq); 4564 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4565 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4566 r->keep_state == PF_STATE_MODULATE) { 4567 /* Generate sequence number modulator */ 4568 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4569 0) 4570 s->src.seqdiff = 1; 4571 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4572 htonl(s->src.seqlo + s->src.seqdiff), 0); 4573 *rewrite = 1; 4574 } else 4575 s->src.seqdiff = 0; 4576 if (th->th_flags & TH_SYN) { 4577 s->src.seqhi++; 4578 s->src.wscale = pf_get_wscale(m, off, 4579 th->th_off, pd->af); 4580 } 4581 s->src.max_win = MAX(ntohs(th->th_win), 1); 4582 if (s->src.wscale & PF_WSCALE_MASK) { 4583 /* Remove scale factor from initial window */ 4584 int win = s->src.max_win; 4585 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4586 s->src.max_win = (win - 1) >> 4587 (s->src.wscale & PF_WSCALE_MASK); 4588 } 4589 if (th->th_flags & TH_FIN) 4590 s->src.seqhi++; 4591 s->dst.seqhi = 1; 4592 s->dst.max_win = 1; 4593 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4594 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4595 s->timeout = PFTM_TCP_FIRST_PACKET; 4596 atomic_add_32(&V_pf_status.states_halfopen, 1); 4597 break; 4598 case IPPROTO_UDP: 4599 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4600 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4601 s->timeout = PFTM_UDP_FIRST_PACKET; 4602 break; 4603 case IPPROTO_ICMP: 4604 #ifdef INET6 4605 case IPPROTO_ICMPV6: 4606 #endif 4607 s->timeout = PFTM_ICMP_FIRST_PACKET; 4608 break; 4609 default: 4610 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4611 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4612 s->timeout = PFTM_OTHER_FIRST_PACKET; 4613 } 4614 4615 if (r->rt) { 4616 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 4617 REASON_SET(&reason, PFRES_MAPFAILED); 4618 pf_src_tree_remove_state(s); 4619 s->timeout = PFTM_UNLINKED; 4620 STATE_DEC_COUNTERS(s); 4621 pf_free_state(s); 4622 goto csfailed; 4623 } 4624 s->rt_kif = r->rpool.cur->kif; 4625 } 4626 4627 s->creation = time_uptime; 4628 s->expire = time_uptime; 4629 4630 if (sn != NULL) 4631 s->src_node = sn; 4632 if (nsn != NULL) { 4633 /* XXX We only modify one side for now. */ 4634 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4635 s->nat_src_node = nsn; 4636 } 4637 if (pd->proto == IPPROTO_TCP) { 4638 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 4639 off, pd, th, &s->src, &s->dst)) { 4640 REASON_SET(&reason, PFRES_MEMORY); 4641 pf_src_tree_remove_state(s); 4642 s->timeout = PFTM_UNLINKED; 4643 STATE_DEC_COUNTERS(s); 4644 pf_free_state(s); 4645 return (PF_DROP); 4646 } 4647 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 4648 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4649 &s->src, &s->dst, rewrite)) { 4650 /* This really shouldn't happen!!! */ 4651 DPFPRINTF(PF_DEBUG_URGENT, 4652 ("pf_normalize_tcp_stateful failed on first " 4653 "pkt\n")); 4654 pf_src_tree_remove_state(s); 4655 s->timeout = PFTM_UNLINKED; 4656 STATE_DEC_COUNTERS(s); 4657 pf_free_state(s); 4658 return (PF_DROP); 4659 } 4660 } 4661 s->direction = pd->dir; 4662 4663 /* 4664 * sk/nk could already been setup by pf_get_translation(). 4665 */ 4666 if (nr == NULL) { 4667 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4668 __func__, nr, sk, nk)); 4669 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4670 if (sk == NULL) 4671 goto csfailed; 4672 nk = sk; 4673 } else 4674 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4675 __func__, nr, sk, nk)); 4676 4677 /* Swap sk/nk for PF_OUT. */ 4678 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4679 (pd->dir == PF_IN) ? sk : nk, 4680 (pd->dir == PF_IN) ? nk : sk, s)) { 4681 REASON_SET(&reason, PFRES_STATEINS); 4682 pf_src_tree_remove_state(s); 4683 s->timeout = PFTM_UNLINKED; 4684 STATE_DEC_COUNTERS(s); 4685 pf_free_state(s); 4686 return (PF_DROP); 4687 } else 4688 *sm = s; 4689 4690 if (tag > 0) 4691 s->tag = tag; 4692 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4693 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4694 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4695 /* undo NAT changes, if they have taken place */ 4696 if (nr != NULL) { 4697 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4698 if (pd->dir == PF_OUT) 4699 skt = s->key[PF_SK_STACK]; 4700 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4701 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4702 if (pd->sport) 4703 *pd->sport = skt->port[pd->sidx]; 4704 if (pd->dport) 4705 *pd->dport = skt->port[pd->didx]; 4706 if (pd->proto_sum) 4707 *pd->proto_sum = bproto_sum; 4708 if (pd->ip_sum) 4709 *pd->ip_sum = bip_sum; 4710 m_copyback(m, off, hdrlen, pd->hdr.any); 4711 } 4712 s->src.seqhi = htonl(arc4random()); 4713 /* Find mss option */ 4714 int rtid = M_GETFIB(m); 4715 mss = pf_get_mss(m, off, th->th_off, pd->af); 4716 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4717 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4718 s->src.mss = mss; 4719 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4720 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4721 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0); 4722 REASON_SET(&reason, PFRES_SYNPROXY); 4723 return (PF_SYNPROXY_DROP); 4724 } 4725 4726 return (PF_PASS); 4727 4728 csfailed: 4729 if (sk != NULL) 4730 uma_zfree(V_pf_state_key_z, sk); 4731 if (nk != NULL) 4732 uma_zfree(V_pf_state_key_z, nk); 4733 4734 if (sn != NULL) { 4735 struct pf_srchash *sh; 4736 4737 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 4738 PF_HASHROW_LOCK(sh); 4739 if (--sn->states == 0 && sn->expire == 0) { 4740 pf_unlink_src_node(sn); 4741 uma_zfree(V_pf_sources_z, sn); 4742 counter_u64_add( 4743 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4744 } 4745 PF_HASHROW_UNLOCK(sh); 4746 } 4747 4748 if (nsn != sn && nsn != NULL) { 4749 struct pf_srchash *sh; 4750 4751 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; 4752 PF_HASHROW_LOCK(sh); 4753 if (--nsn->states == 0 && nsn->expire == 0) { 4754 pf_unlink_src_node(nsn); 4755 uma_zfree(V_pf_sources_z, nsn); 4756 counter_u64_add( 4757 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4758 } 4759 PF_HASHROW_UNLOCK(sh); 4760 } 4761 4762 return (PF_DROP); 4763 } 4764 4765 static int 4766 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif, 4767 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 4768 struct pf_kruleset **rsm) 4769 { 4770 struct pf_krule *r, *a = NULL; 4771 struct pf_kruleset *ruleset = NULL; 4772 sa_family_t af = pd->af; 4773 u_short reason; 4774 int tag = -1; 4775 int asd = 0; 4776 int match = 0; 4777 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4778 4779 PF_RULES_RASSERT(); 4780 4781 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4782 while (r != NULL) { 4783 pf_counter_u64_add(&r->evaluations, 1); 4784 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4785 r = r->skip[PF_SKIP_IFP].ptr; 4786 else if (r->direction && r->direction != direction) 4787 r = r->skip[PF_SKIP_DIR].ptr; 4788 else if (r->af && r->af != af) 4789 r = r->skip[PF_SKIP_AF].ptr; 4790 else if (r->proto && r->proto != pd->proto) 4791 r = r->skip[PF_SKIP_PROTO].ptr; 4792 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 4793 r->src.neg, kif, M_GETFIB(m))) 4794 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4795 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 4796 r->dst.neg, NULL, M_GETFIB(m))) 4797 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4798 else if (r->tos && !(r->tos == pd->tos)) 4799 r = TAILQ_NEXT(r, entries); 4800 else if (r->os_fingerprint != PF_OSFP_ANY) 4801 r = TAILQ_NEXT(r, entries); 4802 else if (pd->proto == IPPROTO_UDP && 4803 (r->src.port_op || r->dst.port_op)) 4804 r = TAILQ_NEXT(r, entries); 4805 else if (pd->proto == IPPROTO_TCP && 4806 (r->src.port_op || r->dst.port_op || r->flagset)) 4807 r = TAILQ_NEXT(r, entries); 4808 else if ((pd->proto == IPPROTO_ICMP || 4809 pd->proto == IPPROTO_ICMPV6) && 4810 (r->type || r->code)) 4811 r = TAILQ_NEXT(r, entries); 4812 else if (r->prio && 4813 !pf_match_ieee8021q_pcp(r->prio, m)) 4814 r = TAILQ_NEXT(r, entries); 4815 else if (r->prob && r->prob <= 4816 (arc4random() % (UINT_MAX - 1) + 1)) 4817 r = TAILQ_NEXT(r, entries); 4818 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4819 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4820 r = TAILQ_NEXT(r, entries); 4821 else { 4822 if (r->anchor == NULL) { 4823 if (r->action == PF_MATCH) { 4824 pf_counter_u64_critical_enter(); 4825 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 4826 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 4827 pf_counter_u64_critical_exit(); 4828 pf_rule_to_actions(r, &pd->act); 4829 if (r->log) 4830 PFLOG_PACKET(kif, m, af, 4831 direction, PFRES_MATCH, r, 4832 a, ruleset, pd, 1); 4833 } else { 4834 match = 1; 4835 *rm = r; 4836 *am = a; 4837 *rsm = ruleset; 4838 } 4839 if ((*rm)->quick) 4840 break; 4841 r = TAILQ_NEXT(r, entries); 4842 } else 4843 pf_step_into_anchor(anchor_stack, &asd, 4844 &ruleset, PF_RULESET_FILTER, &r, &a, 4845 &match); 4846 } 4847 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4848 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4849 break; 4850 } 4851 r = *rm; 4852 a = *am; 4853 ruleset = *rsm; 4854 4855 REASON_SET(&reason, PFRES_MATCH); 4856 4857 /* apply actions for last matching pass/block rule */ 4858 pf_rule_to_actions(r, &pd->act); 4859 4860 if (r->log) 4861 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 4862 1); 4863 4864 if (r->action != PF_PASS) 4865 return (PF_DROP); 4866 4867 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4868 REASON_SET(&reason, PFRES_MEMORY); 4869 return (PF_DROP); 4870 } 4871 4872 return (PF_PASS); 4873 } 4874 4875 static int 4876 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 4877 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 4878 int *copyback) 4879 { 4880 struct tcphdr *th = &pd->hdr.tcp; 4881 struct pf_state_peer *src, *dst; 4882 u_int16_t win = ntohs(th->th_win); 4883 u_int32_t ack, end, seq, orig_seq; 4884 u_int8_t sws, dws, psrc, pdst; 4885 int ackskew; 4886 4887 if (pd->dir == (*state)->direction) { 4888 src = &(*state)->src; 4889 dst = &(*state)->dst; 4890 psrc = PF_PEER_SRC; 4891 pdst = PF_PEER_DST; 4892 } else { 4893 src = &(*state)->dst; 4894 dst = &(*state)->src; 4895 psrc = PF_PEER_DST; 4896 pdst = PF_PEER_SRC; 4897 } 4898 4899 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4900 sws = src->wscale & PF_WSCALE_MASK; 4901 dws = dst->wscale & PF_WSCALE_MASK; 4902 } else 4903 sws = dws = 0; 4904 4905 /* 4906 * Sequence tracking algorithm from Guido van Rooij's paper: 4907 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4908 * tcp_filtering.ps 4909 */ 4910 4911 orig_seq = seq = ntohl(th->th_seq); 4912 if (src->seqlo == 0) { 4913 /* First packet from this end. Set its state */ 4914 4915 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 4916 src->scrub == NULL) { 4917 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 4918 REASON_SET(reason, PFRES_MEMORY); 4919 return (PF_DROP); 4920 } 4921 } 4922 4923 /* Deferred generation of sequence number modulator */ 4924 if (dst->seqdiff && !src->seqdiff) { 4925 /* use random iss for the TCP server */ 4926 while ((src->seqdiff = arc4random() - seq) == 0) 4927 ; 4928 ack = ntohl(th->th_ack) - dst->seqdiff; 4929 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4930 src->seqdiff), 0); 4931 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4932 *copyback = 1; 4933 } else { 4934 ack = ntohl(th->th_ack); 4935 } 4936 4937 end = seq + pd->p_len; 4938 if (th->th_flags & TH_SYN) { 4939 end++; 4940 if (dst->wscale & PF_WSCALE_FLAG) { 4941 src->wscale = pf_get_wscale(m, off, th->th_off, 4942 pd->af); 4943 if (src->wscale & PF_WSCALE_FLAG) { 4944 /* Remove scale factor from initial 4945 * window */ 4946 sws = src->wscale & PF_WSCALE_MASK; 4947 win = ((u_int32_t)win + (1 << sws) - 1) 4948 >> sws; 4949 dws = dst->wscale & PF_WSCALE_MASK; 4950 } else { 4951 /* fixup other window */ 4952 dst->max_win <<= dst->wscale & 4953 PF_WSCALE_MASK; 4954 /* in case of a retrans SYN|ACK */ 4955 dst->wscale = 0; 4956 } 4957 } 4958 } 4959 if (th->th_flags & TH_FIN) 4960 end++; 4961 4962 src->seqlo = seq; 4963 if (src->state < TCPS_SYN_SENT) 4964 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 4965 4966 /* 4967 * May need to slide the window (seqhi may have been set by 4968 * the crappy stack check or if we picked up the connection 4969 * after establishment) 4970 */ 4971 if (src->seqhi == 1 || 4972 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4973 src->seqhi = end + MAX(1, dst->max_win << dws); 4974 if (win > src->max_win) 4975 src->max_win = win; 4976 4977 } else { 4978 ack = ntohl(th->th_ack) - dst->seqdiff; 4979 if (src->seqdiff) { 4980 /* Modulate sequence numbers */ 4981 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4982 src->seqdiff), 0); 4983 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4984 *copyback = 1; 4985 } 4986 end = seq + pd->p_len; 4987 if (th->th_flags & TH_SYN) 4988 end++; 4989 if (th->th_flags & TH_FIN) 4990 end++; 4991 } 4992 4993 if ((th->th_flags & TH_ACK) == 0) { 4994 /* Let it pass through the ack skew check */ 4995 ack = dst->seqlo; 4996 } else if ((ack == 0 && 4997 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4998 /* broken tcp stacks do not set ack */ 4999 (dst->state < TCPS_SYN_SENT)) { 5000 /* 5001 * Many stacks (ours included) will set the ACK number in an 5002 * FIN|ACK if the SYN times out -- no sequence to ACK. 5003 */ 5004 ack = dst->seqlo; 5005 } 5006 5007 if (seq == end) { 5008 /* Ease sequencing restrictions on no data packets */ 5009 seq = src->seqlo; 5010 end = seq; 5011 } 5012 5013 ackskew = dst->seqlo - ack; 5014 5015 /* 5016 * Need to demodulate the sequence numbers in any TCP SACK options 5017 * (Selective ACK). We could optionally validate the SACK values 5018 * against the current ACK window, either forwards or backwards, but 5019 * I'm not confident that SACK has been implemented properly 5020 * everywhere. It wouldn't surprise me if several stacks accidentally 5021 * SACK too far backwards of previously ACKed data. There really aren't 5022 * any security implications of bad SACKing unless the target stack 5023 * doesn't validate the option length correctly. Someone trying to 5024 * spoof into a TCP connection won't bother blindly sending SACK 5025 * options anyway. 5026 */ 5027 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5028 if (pf_modulate_sack(m, off, pd, th, dst)) 5029 *copyback = 1; 5030 } 5031 5032 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5033 if (SEQ_GEQ(src->seqhi, end) && 5034 /* Last octet inside other's window space */ 5035 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5036 /* Retrans: not more than one window back */ 5037 (ackskew >= -MAXACKWINDOW) && 5038 /* Acking not more than one reassembled fragment backwards */ 5039 (ackskew <= (MAXACKWINDOW << sws)) && 5040 /* Acking not more than one window forward */ 5041 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5042 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 5043 (pd->flags & PFDESC_IP_REAS) == 0)) { 5044 /* Require an exact/+1 sequence match on resets when possible */ 5045 5046 if (dst->scrub || src->scrub) { 5047 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5048 *state, src, dst, copyback)) 5049 return (PF_DROP); 5050 } 5051 5052 /* update max window */ 5053 if (src->max_win < win) 5054 src->max_win = win; 5055 /* synchronize sequencing */ 5056 if (SEQ_GT(end, src->seqlo)) 5057 src->seqlo = end; 5058 /* slide the window of what the other end can send */ 5059 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5060 dst->seqhi = ack + MAX((win << sws), 1); 5061 5062 /* update states */ 5063 if (th->th_flags & TH_SYN) 5064 if (src->state < TCPS_SYN_SENT) 5065 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5066 if (th->th_flags & TH_FIN) 5067 if (src->state < TCPS_CLOSING) 5068 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5069 if (th->th_flags & TH_ACK) { 5070 if (dst->state == TCPS_SYN_SENT) { 5071 pf_set_protostate(*state, pdst, 5072 TCPS_ESTABLISHED); 5073 if (src->state == TCPS_ESTABLISHED && 5074 (*state)->src_node != NULL && 5075 pf_src_connlimit(state)) { 5076 REASON_SET(reason, PFRES_SRCLIMIT); 5077 return (PF_DROP); 5078 } 5079 } else if (dst->state == TCPS_CLOSING) 5080 pf_set_protostate(*state, pdst, 5081 TCPS_FIN_WAIT_2); 5082 } 5083 if (th->th_flags & TH_RST) 5084 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5085 5086 /* update expire time */ 5087 (*state)->expire = time_uptime; 5088 if (src->state >= TCPS_FIN_WAIT_2 && 5089 dst->state >= TCPS_FIN_WAIT_2) 5090 (*state)->timeout = PFTM_TCP_CLOSED; 5091 else if (src->state >= TCPS_CLOSING && 5092 dst->state >= TCPS_CLOSING) 5093 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5094 else if (src->state < TCPS_ESTABLISHED || 5095 dst->state < TCPS_ESTABLISHED) 5096 (*state)->timeout = PFTM_TCP_OPENING; 5097 else if (src->state >= TCPS_CLOSING || 5098 dst->state >= TCPS_CLOSING) 5099 (*state)->timeout = PFTM_TCP_CLOSING; 5100 else 5101 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5102 5103 /* Fall through to PASS packet */ 5104 5105 } else if ((dst->state < TCPS_SYN_SENT || 5106 dst->state >= TCPS_FIN_WAIT_2 || 5107 src->state >= TCPS_FIN_WAIT_2) && 5108 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5109 /* Within a window forward of the originating packet */ 5110 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5111 /* Within a window backward of the originating packet */ 5112 5113 /* 5114 * This currently handles three situations: 5115 * 1) Stupid stacks will shotgun SYNs before their peer 5116 * replies. 5117 * 2) When PF catches an already established stream (the 5118 * firewall rebooted, the state table was flushed, routes 5119 * changed...) 5120 * 3) Packets get funky immediately after the connection 5121 * closes (this should catch Solaris spurious ACK|FINs 5122 * that web servers like to spew after a close) 5123 * 5124 * This must be a little more careful than the above code 5125 * since packet floods will also be caught here. We don't 5126 * update the TTL here to mitigate the damage of a packet 5127 * flood and so the same code can handle awkward establishment 5128 * and a loosened connection close. 5129 * In the establishment case, a correct peer response will 5130 * validate the connection, go through the normal state code 5131 * and keep updating the state TTL. 5132 */ 5133 5134 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5135 printf("pf: loose state match: "); 5136 pf_print_state(*state); 5137 pf_print_flags(th->th_flags); 5138 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5139 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5140 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5141 (unsigned long long)(*state)->packets[1], 5142 pd->dir == PF_IN ? "in" : "out", 5143 pd->dir == (*state)->direction ? "fwd" : "rev"); 5144 } 5145 5146 if (dst->scrub || src->scrub) { 5147 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5148 *state, src, dst, copyback)) 5149 return (PF_DROP); 5150 } 5151 5152 /* update max window */ 5153 if (src->max_win < win) 5154 src->max_win = win; 5155 /* synchronize sequencing */ 5156 if (SEQ_GT(end, src->seqlo)) 5157 src->seqlo = end; 5158 /* slide the window of what the other end can send */ 5159 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5160 dst->seqhi = ack + MAX((win << sws), 1); 5161 5162 /* 5163 * Cannot set dst->seqhi here since this could be a shotgunned 5164 * SYN and not an already established connection. 5165 */ 5166 5167 if (th->th_flags & TH_FIN) 5168 if (src->state < TCPS_CLOSING) 5169 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5170 if (th->th_flags & TH_RST) 5171 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5172 5173 /* Fall through to PASS packet */ 5174 5175 } else { 5176 if ((*state)->dst.state == TCPS_SYN_SENT && 5177 (*state)->src.state == TCPS_SYN_SENT) { 5178 /* Send RST for state mismatches during handshake */ 5179 if (!(th->th_flags & TH_RST)) 5180 pf_send_tcp((*state)->rule.ptr, pd->af, 5181 pd->dst, pd->src, th->th_dport, 5182 th->th_sport, ntohl(th->th_ack), 0, 5183 TH_RST, 0, 0, 5184 (*state)->rule.ptr->return_ttl, 1, 0); 5185 src->seqlo = 0; 5186 src->seqhi = 1; 5187 src->max_win = 1; 5188 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5189 printf("pf: BAD state: "); 5190 pf_print_state(*state); 5191 pf_print_flags(th->th_flags); 5192 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5193 "pkts=%llu:%llu dir=%s,%s\n", 5194 seq, orig_seq, ack, pd->p_len, ackskew, 5195 (unsigned long long)(*state)->packets[0], 5196 (unsigned long long)(*state)->packets[1], 5197 pd->dir == PF_IN ? "in" : "out", 5198 pd->dir == (*state)->direction ? "fwd" : "rev"); 5199 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5200 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5201 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5202 ' ': '2', 5203 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5204 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5205 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5206 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5207 } 5208 REASON_SET(reason, PFRES_BADSTATE); 5209 return (PF_DROP); 5210 } 5211 5212 return (PF_PASS); 5213 } 5214 5215 static int 5216 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5217 { 5218 struct tcphdr *th = &pd->hdr.tcp; 5219 struct pf_state_peer *src, *dst; 5220 u_int8_t psrc, pdst; 5221 5222 if (pd->dir == (*state)->direction) { 5223 src = &(*state)->src; 5224 dst = &(*state)->dst; 5225 psrc = PF_PEER_SRC; 5226 pdst = PF_PEER_DST; 5227 } else { 5228 src = &(*state)->dst; 5229 dst = &(*state)->src; 5230 psrc = PF_PEER_DST; 5231 pdst = PF_PEER_SRC; 5232 } 5233 5234 if (th->th_flags & TH_SYN) 5235 if (src->state < TCPS_SYN_SENT) 5236 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5237 if (th->th_flags & TH_FIN) 5238 if (src->state < TCPS_CLOSING) 5239 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5240 if (th->th_flags & TH_ACK) { 5241 if (dst->state == TCPS_SYN_SENT) { 5242 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5243 if (src->state == TCPS_ESTABLISHED && 5244 (*state)->src_node != NULL && 5245 pf_src_connlimit(state)) { 5246 REASON_SET(reason, PFRES_SRCLIMIT); 5247 return (PF_DROP); 5248 } 5249 } else if (dst->state == TCPS_CLOSING) { 5250 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5251 } else if (src->state == TCPS_SYN_SENT && 5252 dst->state < TCPS_SYN_SENT) { 5253 /* 5254 * Handle a special sloppy case where we only see one 5255 * half of the connection. If there is a ACK after 5256 * the initial SYN without ever seeing a packet from 5257 * the destination, set the connection to established. 5258 */ 5259 pf_set_protostate(*state, PF_PEER_BOTH, 5260 TCPS_ESTABLISHED); 5261 dst->state = src->state = TCPS_ESTABLISHED; 5262 if ((*state)->src_node != NULL && 5263 pf_src_connlimit(state)) { 5264 REASON_SET(reason, PFRES_SRCLIMIT); 5265 return (PF_DROP); 5266 } 5267 } else if (src->state == TCPS_CLOSING && 5268 dst->state == TCPS_ESTABLISHED && 5269 dst->seqlo == 0) { 5270 /* 5271 * Handle the closing of half connections where we 5272 * don't see the full bidirectional FIN/ACK+ACK 5273 * handshake. 5274 */ 5275 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5276 } 5277 } 5278 if (th->th_flags & TH_RST) 5279 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5280 5281 /* update expire time */ 5282 (*state)->expire = time_uptime; 5283 if (src->state >= TCPS_FIN_WAIT_2 && 5284 dst->state >= TCPS_FIN_WAIT_2) 5285 (*state)->timeout = PFTM_TCP_CLOSED; 5286 else if (src->state >= TCPS_CLOSING && 5287 dst->state >= TCPS_CLOSING) 5288 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5289 else if (src->state < TCPS_ESTABLISHED || 5290 dst->state < TCPS_ESTABLISHED) 5291 (*state)->timeout = PFTM_TCP_OPENING; 5292 else if (src->state >= TCPS_CLOSING || 5293 dst->state >= TCPS_CLOSING) 5294 (*state)->timeout = PFTM_TCP_CLOSING; 5295 else 5296 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5297 5298 return (PF_PASS); 5299 } 5300 5301 static int 5302 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5303 { 5304 struct pf_state_key *sk = (*state)->key[pd->didx]; 5305 struct tcphdr *th = &pd->hdr.tcp; 5306 5307 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5308 if (pd->dir != (*state)->direction) { 5309 REASON_SET(reason, PFRES_SYNPROXY); 5310 return (PF_SYNPROXY_DROP); 5311 } 5312 if (th->th_flags & TH_SYN) { 5313 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5314 REASON_SET(reason, PFRES_SYNPROXY); 5315 return (PF_DROP); 5316 } 5317 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5318 pd->src, th->th_dport, th->th_sport, 5319 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5320 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0); 5321 REASON_SET(reason, PFRES_SYNPROXY); 5322 return (PF_SYNPROXY_DROP); 5323 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5324 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5325 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5326 REASON_SET(reason, PFRES_SYNPROXY); 5327 return (PF_DROP); 5328 } else if ((*state)->src_node != NULL && 5329 pf_src_connlimit(state)) { 5330 REASON_SET(reason, PFRES_SRCLIMIT); 5331 return (PF_DROP); 5332 } else 5333 pf_set_protostate(*state, PF_PEER_SRC, 5334 PF_TCPS_PROXY_DST); 5335 } 5336 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5337 if (pd->dir == (*state)->direction) { 5338 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5339 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5340 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5341 REASON_SET(reason, PFRES_SYNPROXY); 5342 return (PF_DROP); 5343 } 5344 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5345 if ((*state)->dst.seqhi == 1) 5346 (*state)->dst.seqhi = htonl(arc4random()); 5347 pf_send_tcp((*state)->rule.ptr, pd->af, 5348 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5349 sk->port[pd->sidx], sk->port[pd->didx], 5350 (*state)->dst.seqhi, 0, TH_SYN, 0, 5351 (*state)->src.mss, 0, 0, (*state)->tag); 5352 REASON_SET(reason, PFRES_SYNPROXY); 5353 return (PF_SYNPROXY_DROP); 5354 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5355 (TH_SYN|TH_ACK)) || 5356 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5357 REASON_SET(reason, PFRES_SYNPROXY); 5358 return (PF_DROP); 5359 } else { 5360 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5361 (*state)->dst.seqlo = ntohl(th->th_seq); 5362 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5363 pd->src, th->th_dport, th->th_sport, 5364 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5365 TH_ACK, (*state)->src.max_win, 0, 0, 0, 5366 (*state)->tag); 5367 pf_send_tcp((*state)->rule.ptr, pd->af, 5368 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5369 sk->port[pd->sidx], sk->port[pd->didx], 5370 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5371 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0); 5372 (*state)->src.seqdiff = (*state)->dst.seqhi - 5373 (*state)->src.seqlo; 5374 (*state)->dst.seqdiff = (*state)->src.seqhi - 5375 (*state)->dst.seqlo; 5376 (*state)->src.seqhi = (*state)->src.seqlo + 5377 (*state)->dst.max_win; 5378 (*state)->dst.seqhi = (*state)->dst.seqlo + 5379 (*state)->src.max_win; 5380 (*state)->src.wscale = (*state)->dst.wscale = 0; 5381 pf_set_protostate(*state, PF_PEER_BOTH, 5382 TCPS_ESTABLISHED); 5383 REASON_SET(reason, PFRES_SYNPROXY); 5384 return (PF_SYNPROXY_DROP); 5385 } 5386 } 5387 5388 return (PF_PASS); 5389 } 5390 5391 static int 5392 pf_test_state_tcp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5393 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5394 u_short *reason) 5395 { 5396 struct pf_state_key_cmp key; 5397 struct tcphdr *th = &pd->hdr.tcp; 5398 int copyback = 0; 5399 int action; 5400 struct pf_state_peer *src, *dst; 5401 5402 bzero(&key, sizeof(key)); 5403 key.af = pd->af; 5404 key.proto = IPPROTO_TCP; 5405 if (direction == PF_IN) { /* wire side, straight */ 5406 PF_ACPY(&key.addr[0], pd->src, key.af); 5407 PF_ACPY(&key.addr[1], pd->dst, key.af); 5408 key.port[0] = th->th_sport; 5409 key.port[1] = th->th_dport; 5410 } else { /* stack side, reverse */ 5411 PF_ACPY(&key.addr[1], pd->src, key.af); 5412 PF_ACPY(&key.addr[0], pd->dst, key.af); 5413 key.port[1] = th->th_sport; 5414 key.port[0] = th->th_dport; 5415 } 5416 5417 STATE_LOOKUP(kif, &key, direction, *state, pd); 5418 5419 if (direction == (*state)->direction) { 5420 src = &(*state)->src; 5421 dst = &(*state)->dst; 5422 } else { 5423 src = &(*state)->dst; 5424 dst = &(*state)->src; 5425 } 5426 5427 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5428 return (action); 5429 5430 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 5431 dst->state >= TCPS_FIN_WAIT_2 && 5432 src->state >= TCPS_FIN_WAIT_2) { 5433 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5434 printf("pf: state reuse "); 5435 pf_print_state(*state); 5436 pf_print_flags(th->th_flags); 5437 printf("\n"); 5438 } 5439 /* XXX make sure it's the same direction ?? */ 5440 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5441 pf_unlink_state(*state); 5442 *state = NULL; 5443 return (PF_DROP); 5444 } 5445 5446 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5447 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5448 return (PF_DROP); 5449 } else { 5450 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5451 ©back) == PF_DROP) 5452 return (PF_DROP); 5453 } 5454 5455 /* translate source/destination address, if necessary */ 5456 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5457 struct pf_state_key *nk = (*state)->key[pd->didx]; 5458 5459 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5460 nk->port[pd->sidx] != th->th_sport) 5461 pf_change_ap(m, pd->src, &th->th_sport, 5462 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5463 nk->port[pd->sidx], 0, pd->af); 5464 5465 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5466 nk->port[pd->didx] != th->th_dport) 5467 pf_change_ap(m, pd->dst, &th->th_dport, 5468 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5469 nk->port[pd->didx], 0, pd->af); 5470 copyback = 1; 5471 } 5472 5473 /* Copyback sequence modulation or stateful scrub changes if needed */ 5474 if (copyback) 5475 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5476 5477 return (PF_PASS); 5478 } 5479 5480 static int 5481 pf_test_state_udp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5482 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5483 { 5484 struct pf_state_peer *src, *dst; 5485 struct pf_state_key_cmp key; 5486 struct udphdr *uh = &pd->hdr.udp; 5487 uint8_t psrc, pdst; 5488 5489 bzero(&key, sizeof(key)); 5490 key.af = pd->af; 5491 key.proto = IPPROTO_UDP; 5492 if (direction == PF_IN) { /* wire side, straight */ 5493 PF_ACPY(&key.addr[0], pd->src, key.af); 5494 PF_ACPY(&key.addr[1], pd->dst, key.af); 5495 key.port[0] = uh->uh_sport; 5496 key.port[1] = uh->uh_dport; 5497 } else { /* stack side, reverse */ 5498 PF_ACPY(&key.addr[1], pd->src, key.af); 5499 PF_ACPY(&key.addr[0], pd->dst, key.af); 5500 key.port[1] = uh->uh_sport; 5501 key.port[0] = uh->uh_dport; 5502 } 5503 5504 STATE_LOOKUP(kif, &key, direction, *state, pd); 5505 5506 if (direction == (*state)->direction) { 5507 src = &(*state)->src; 5508 dst = &(*state)->dst; 5509 psrc = PF_PEER_SRC; 5510 pdst = PF_PEER_DST; 5511 } else { 5512 src = &(*state)->dst; 5513 dst = &(*state)->src; 5514 psrc = PF_PEER_DST; 5515 pdst = PF_PEER_SRC; 5516 } 5517 5518 /* update states */ 5519 if (src->state < PFUDPS_SINGLE) 5520 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5521 if (dst->state == PFUDPS_SINGLE) 5522 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5523 5524 /* update expire time */ 5525 (*state)->expire = time_uptime; 5526 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5527 (*state)->timeout = PFTM_UDP_MULTIPLE; 5528 else 5529 (*state)->timeout = PFTM_UDP_SINGLE; 5530 5531 /* translate source/destination address, if necessary */ 5532 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5533 struct pf_state_key *nk = (*state)->key[pd->didx]; 5534 5535 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5536 nk->port[pd->sidx] != uh->uh_sport) 5537 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5538 &uh->uh_sum, &nk->addr[pd->sidx], 5539 nk->port[pd->sidx], 1, pd->af); 5540 5541 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5542 nk->port[pd->didx] != uh->uh_dport) 5543 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5544 &uh->uh_sum, &nk->addr[pd->didx], 5545 nk->port[pd->didx], 1, pd->af); 5546 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5547 } 5548 5549 return (PF_PASS); 5550 } 5551 5552 static int 5553 pf_test_state_icmp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5554 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5555 { 5556 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 5557 u_int16_t icmpid = 0, *icmpsum; 5558 u_int8_t icmptype, icmpcode; 5559 int state_icmp = 0; 5560 struct pf_state_key_cmp key; 5561 5562 bzero(&key, sizeof(key)); 5563 switch (pd->proto) { 5564 #ifdef INET 5565 case IPPROTO_ICMP: 5566 icmptype = pd->hdr.icmp.icmp_type; 5567 icmpcode = pd->hdr.icmp.icmp_code; 5568 icmpid = pd->hdr.icmp.icmp_id; 5569 icmpsum = &pd->hdr.icmp.icmp_cksum; 5570 5571 if (icmptype == ICMP_UNREACH || 5572 icmptype == ICMP_SOURCEQUENCH || 5573 icmptype == ICMP_REDIRECT || 5574 icmptype == ICMP_TIMXCEED || 5575 icmptype == ICMP_PARAMPROB) 5576 state_icmp++; 5577 break; 5578 #endif /* INET */ 5579 #ifdef INET6 5580 case IPPROTO_ICMPV6: 5581 icmptype = pd->hdr.icmp6.icmp6_type; 5582 icmpcode = pd->hdr.icmp6.icmp6_code; 5583 icmpid = pd->hdr.icmp6.icmp6_id; 5584 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 5585 5586 if (icmptype == ICMP6_DST_UNREACH || 5587 icmptype == ICMP6_PACKET_TOO_BIG || 5588 icmptype == ICMP6_TIME_EXCEEDED || 5589 icmptype == ICMP6_PARAM_PROB) 5590 state_icmp++; 5591 break; 5592 #endif /* INET6 */ 5593 } 5594 5595 if (!state_icmp) { 5596 /* 5597 * ICMP query/reply message not related to a TCP/UDP packet. 5598 * Search for an ICMP state. 5599 */ 5600 key.af = pd->af; 5601 key.proto = pd->proto; 5602 key.port[0] = key.port[1] = icmpid; 5603 if (direction == PF_IN) { /* wire side, straight */ 5604 PF_ACPY(&key.addr[0], pd->src, key.af); 5605 PF_ACPY(&key.addr[1], pd->dst, key.af); 5606 } else { /* stack side, reverse */ 5607 PF_ACPY(&key.addr[1], pd->src, key.af); 5608 PF_ACPY(&key.addr[0], pd->dst, key.af); 5609 } 5610 5611 STATE_LOOKUP(kif, &key, direction, *state, pd); 5612 5613 (*state)->expire = time_uptime; 5614 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 5615 5616 /* translate source/destination address, if necessary */ 5617 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5618 struct pf_state_key *nk = (*state)->key[pd->didx]; 5619 5620 switch (pd->af) { 5621 #ifdef INET 5622 case AF_INET: 5623 if (PF_ANEQ(pd->src, 5624 &nk->addr[pd->sidx], AF_INET)) 5625 pf_change_a(&saddr->v4.s_addr, 5626 pd->ip_sum, 5627 nk->addr[pd->sidx].v4.s_addr, 0); 5628 5629 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 5630 AF_INET)) 5631 pf_change_a(&daddr->v4.s_addr, 5632 pd->ip_sum, 5633 nk->addr[pd->didx].v4.s_addr, 0); 5634 5635 if (nk->port[0] != 5636 pd->hdr.icmp.icmp_id) { 5637 pd->hdr.icmp.icmp_cksum = 5638 pf_cksum_fixup( 5639 pd->hdr.icmp.icmp_cksum, icmpid, 5640 nk->port[pd->sidx], 0); 5641 pd->hdr.icmp.icmp_id = 5642 nk->port[pd->sidx]; 5643 } 5644 5645 m_copyback(m, off, ICMP_MINLEN, 5646 (caddr_t )&pd->hdr.icmp); 5647 break; 5648 #endif /* INET */ 5649 #ifdef INET6 5650 case AF_INET6: 5651 if (PF_ANEQ(pd->src, 5652 &nk->addr[pd->sidx], AF_INET6)) 5653 pf_change_a6(saddr, 5654 &pd->hdr.icmp6.icmp6_cksum, 5655 &nk->addr[pd->sidx], 0); 5656 5657 if (PF_ANEQ(pd->dst, 5658 &nk->addr[pd->didx], AF_INET6)) 5659 pf_change_a6(daddr, 5660 &pd->hdr.icmp6.icmp6_cksum, 5661 &nk->addr[pd->didx], 0); 5662 5663 m_copyback(m, off, sizeof(struct icmp6_hdr), 5664 (caddr_t )&pd->hdr.icmp6); 5665 break; 5666 #endif /* INET6 */ 5667 } 5668 } 5669 return (PF_PASS); 5670 5671 } else { 5672 /* 5673 * ICMP error message in response to a TCP/UDP packet. 5674 * Extract the inner TCP/UDP header and search for that state. 5675 */ 5676 5677 struct pf_pdesc pd2; 5678 bzero(&pd2, sizeof pd2); 5679 #ifdef INET 5680 struct ip h2; 5681 #endif /* INET */ 5682 #ifdef INET6 5683 struct ip6_hdr h2_6; 5684 int terminal = 0; 5685 #endif /* INET6 */ 5686 int ipoff2 = 0; 5687 int off2 = 0; 5688 5689 pd2.af = pd->af; 5690 /* Payload packet is from the opposite direction. */ 5691 pd2.sidx = (direction == PF_IN) ? 1 : 0; 5692 pd2.didx = (direction == PF_IN) ? 0 : 1; 5693 switch (pd->af) { 5694 #ifdef INET 5695 case AF_INET: 5696 /* offset of h2 in mbuf chain */ 5697 ipoff2 = off + ICMP_MINLEN; 5698 5699 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 5700 NULL, reason, pd2.af)) { 5701 DPFPRINTF(PF_DEBUG_MISC, 5702 ("pf: ICMP error message too short " 5703 "(ip)\n")); 5704 return (PF_DROP); 5705 } 5706 /* 5707 * ICMP error messages don't refer to non-first 5708 * fragments 5709 */ 5710 if (h2.ip_off & htons(IP_OFFMASK)) { 5711 REASON_SET(reason, PFRES_FRAG); 5712 return (PF_DROP); 5713 } 5714 5715 /* offset of protocol header that follows h2 */ 5716 off2 = ipoff2 + (h2.ip_hl << 2); 5717 5718 pd2.proto = h2.ip_p; 5719 pd2.src = (struct pf_addr *)&h2.ip_src; 5720 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5721 pd2.ip_sum = &h2.ip_sum; 5722 break; 5723 #endif /* INET */ 5724 #ifdef INET6 5725 case AF_INET6: 5726 ipoff2 = off + sizeof(struct icmp6_hdr); 5727 5728 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 5729 NULL, reason, pd2.af)) { 5730 DPFPRINTF(PF_DEBUG_MISC, 5731 ("pf: ICMP error message too short " 5732 "(ip6)\n")); 5733 return (PF_DROP); 5734 } 5735 pd2.proto = h2_6.ip6_nxt; 5736 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5737 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5738 pd2.ip_sum = NULL; 5739 off2 = ipoff2 + sizeof(h2_6); 5740 do { 5741 switch (pd2.proto) { 5742 case IPPROTO_FRAGMENT: 5743 /* 5744 * ICMPv6 error messages for 5745 * non-first fragments 5746 */ 5747 REASON_SET(reason, PFRES_FRAG); 5748 return (PF_DROP); 5749 case IPPROTO_AH: 5750 case IPPROTO_HOPOPTS: 5751 case IPPROTO_ROUTING: 5752 case IPPROTO_DSTOPTS: { 5753 /* get next header and header length */ 5754 struct ip6_ext opt6; 5755 5756 if (!pf_pull_hdr(m, off2, &opt6, 5757 sizeof(opt6), NULL, reason, 5758 pd2.af)) { 5759 DPFPRINTF(PF_DEBUG_MISC, 5760 ("pf: ICMPv6 short opt\n")); 5761 return (PF_DROP); 5762 } 5763 if (pd2.proto == IPPROTO_AH) 5764 off2 += (opt6.ip6e_len + 2) * 4; 5765 else 5766 off2 += (opt6.ip6e_len + 1) * 8; 5767 pd2.proto = opt6.ip6e_nxt; 5768 /* goto the next header */ 5769 break; 5770 } 5771 default: 5772 terminal++; 5773 break; 5774 } 5775 } while (!terminal); 5776 break; 5777 #endif /* INET6 */ 5778 } 5779 5780 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5781 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5782 printf("pf: BAD ICMP %d:%d outer dst: ", 5783 icmptype, icmpcode); 5784 pf_print_host(pd->src, 0, pd->af); 5785 printf(" -> "); 5786 pf_print_host(pd->dst, 0, pd->af); 5787 printf(" inner src: "); 5788 pf_print_host(pd2.src, 0, pd2.af); 5789 printf(" -> "); 5790 pf_print_host(pd2.dst, 0, pd2.af); 5791 printf("\n"); 5792 } 5793 REASON_SET(reason, PFRES_BADSTATE); 5794 return (PF_DROP); 5795 } 5796 5797 switch (pd2.proto) { 5798 case IPPROTO_TCP: { 5799 struct tcphdr th; 5800 u_int32_t seq; 5801 struct pf_state_peer *src, *dst; 5802 u_int8_t dws; 5803 int copyback = 0; 5804 5805 /* 5806 * Only the first 8 bytes of the TCP header can be 5807 * expected. Don't access any TCP header fields after 5808 * th_seq, an ackskew test is not possible. 5809 */ 5810 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 5811 pd2.af)) { 5812 DPFPRINTF(PF_DEBUG_MISC, 5813 ("pf: ICMP error message too short " 5814 "(tcp)\n")); 5815 return (PF_DROP); 5816 } 5817 5818 key.af = pd2.af; 5819 key.proto = IPPROTO_TCP; 5820 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5821 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5822 key.port[pd2.sidx] = th.th_sport; 5823 key.port[pd2.didx] = th.th_dport; 5824 5825 STATE_LOOKUP(kif, &key, direction, *state, pd); 5826 5827 if (direction == (*state)->direction) { 5828 src = &(*state)->dst; 5829 dst = &(*state)->src; 5830 } else { 5831 src = &(*state)->src; 5832 dst = &(*state)->dst; 5833 } 5834 5835 if (src->wscale && dst->wscale) 5836 dws = dst->wscale & PF_WSCALE_MASK; 5837 else 5838 dws = 0; 5839 5840 /* Demodulate sequence number */ 5841 seq = ntohl(th.th_seq) - src->seqdiff; 5842 if (src->seqdiff) { 5843 pf_change_a(&th.th_seq, icmpsum, 5844 htonl(seq), 0); 5845 copyback = 1; 5846 } 5847 5848 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 5849 (!SEQ_GEQ(src->seqhi, seq) || 5850 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 5851 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5852 printf("pf: BAD ICMP %d:%d ", 5853 icmptype, icmpcode); 5854 pf_print_host(pd->src, 0, pd->af); 5855 printf(" -> "); 5856 pf_print_host(pd->dst, 0, pd->af); 5857 printf(" state: "); 5858 pf_print_state(*state); 5859 printf(" seq=%u\n", seq); 5860 } 5861 REASON_SET(reason, PFRES_BADSTATE); 5862 return (PF_DROP); 5863 } else { 5864 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5865 printf("pf: OK ICMP %d:%d ", 5866 icmptype, icmpcode); 5867 pf_print_host(pd->src, 0, pd->af); 5868 printf(" -> "); 5869 pf_print_host(pd->dst, 0, pd->af); 5870 printf(" state: "); 5871 pf_print_state(*state); 5872 printf(" seq=%u\n", seq); 5873 } 5874 } 5875 5876 /* translate source/destination address, if necessary */ 5877 if ((*state)->key[PF_SK_WIRE] != 5878 (*state)->key[PF_SK_STACK]) { 5879 struct pf_state_key *nk = 5880 (*state)->key[pd->didx]; 5881 5882 if (PF_ANEQ(pd2.src, 5883 &nk->addr[pd2.sidx], pd2.af) || 5884 nk->port[pd2.sidx] != th.th_sport) 5885 pf_change_icmp(pd2.src, &th.th_sport, 5886 daddr, &nk->addr[pd2.sidx], 5887 nk->port[pd2.sidx], NULL, 5888 pd2.ip_sum, icmpsum, 5889 pd->ip_sum, 0, pd2.af); 5890 5891 if (PF_ANEQ(pd2.dst, 5892 &nk->addr[pd2.didx], pd2.af) || 5893 nk->port[pd2.didx] != th.th_dport) 5894 pf_change_icmp(pd2.dst, &th.th_dport, 5895 saddr, &nk->addr[pd2.didx], 5896 nk->port[pd2.didx], NULL, 5897 pd2.ip_sum, icmpsum, 5898 pd->ip_sum, 0, pd2.af); 5899 copyback = 1; 5900 } 5901 5902 if (copyback) { 5903 switch (pd2.af) { 5904 #ifdef INET 5905 case AF_INET: 5906 m_copyback(m, off, ICMP_MINLEN, 5907 (caddr_t )&pd->hdr.icmp); 5908 m_copyback(m, ipoff2, sizeof(h2), 5909 (caddr_t )&h2); 5910 break; 5911 #endif /* INET */ 5912 #ifdef INET6 5913 case AF_INET6: 5914 m_copyback(m, off, 5915 sizeof(struct icmp6_hdr), 5916 (caddr_t )&pd->hdr.icmp6); 5917 m_copyback(m, ipoff2, sizeof(h2_6), 5918 (caddr_t )&h2_6); 5919 break; 5920 #endif /* INET6 */ 5921 } 5922 m_copyback(m, off2, 8, (caddr_t)&th); 5923 } 5924 5925 return (PF_PASS); 5926 break; 5927 } 5928 case IPPROTO_UDP: { 5929 struct udphdr uh; 5930 5931 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 5932 NULL, reason, pd2.af)) { 5933 DPFPRINTF(PF_DEBUG_MISC, 5934 ("pf: ICMP error message too short " 5935 "(udp)\n")); 5936 return (PF_DROP); 5937 } 5938 5939 key.af = pd2.af; 5940 key.proto = IPPROTO_UDP; 5941 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5942 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5943 key.port[pd2.sidx] = uh.uh_sport; 5944 key.port[pd2.didx] = uh.uh_dport; 5945 5946 STATE_LOOKUP(kif, &key, direction, *state, pd); 5947 5948 /* translate source/destination address, if necessary */ 5949 if ((*state)->key[PF_SK_WIRE] != 5950 (*state)->key[PF_SK_STACK]) { 5951 struct pf_state_key *nk = 5952 (*state)->key[pd->didx]; 5953 5954 if (PF_ANEQ(pd2.src, 5955 &nk->addr[pd2.sidx], pd2.af) || 5956 nk->port[pd2.sidx] != uh.uh_sport) 5957 pf_change_icmp(pd2.src, &uh.uh_sport, 5958 daddr, &nk->addr[pd2.sidx], 5959 nk->port[pd2.sidx], &uh.uh_sum, 5960 pd2.ip_sum, icmpsum, 5961 pd->ip_sum, 1, pd2.af); 5962 5963 if (PF_ANEQ(pd2.dst, 5964 &nk->addr[pd2.didx], pd2.af) || 5965 nk->port[pd2.didx] != uh.uh_dport) 5966 pf_change_icmp(pd2.dst, &uh.uh_dport, 5967 saddr, &nk->addr[pd2.didx], 5968 nk->port[pd2.didx], &uh.uh_sum, 5969 pd2.ip_sum, icmpsum, 5970 pd->ip_sum, 1, pd2.af); 5971 5972 switch (pd2.af) { 5973 #ifdef INET 5974 case AF_INET: 5975 m_copyback(m, off, ICMP_MINLEN, 5976 (caddr_t )&pd->hdr.icmp); 5977 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5978 break; 5979 #endif /* INET */ 5980 #ifdef INET6 5981 case AF_INET6: 5982 m_copyback(m, off, 5983 sizeof(struct icmp6_hdr), 5984 (caddr_t )&pd->hdr.icmp6); 5985 m_copyback(m, ipoff2, sizeof(h2_6), 5986 (caddr_t )&h2_6); 5987 break; 5988 #endif /* INET6 */ 5989 } 5990 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 5991 } 5992 return (PF_PASS); 5993 break; 5994 } 5995 #ifdef INET 5996 case IPPROTO_ICMP: { 5997 struct icmp iih; 5998 5999 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6000 NULL, reason, pd2.af)) { 6001 DPFPRINTF(PF_DEBUG_MISC, 6002 ("pf: ICMP error message too short i" 6003 "(icmp)\n")); 6004 return (PF_DROP); 6005 } 6006 6007 key.af = pd2.af; 6008 key.proto = IPPROTO_ICMP; 6009 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6010 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6011 key.port[0] = key.port[1] = iih.icmp_id; 6012 6013 STATE_LOOKUP(kif, &key, direction, *state, pd); 6014 6015 /* translate source/destination address, if necessary */ 6016 if ((*state)->key[PF_SK_WIRE] != 6017 (*state)->key[PF_SK_STACK]) { 6018 struct pf_state_key *nk = 6019 (*state)->key[pd->didx]; 6020 6021 if (PF_ANEQ(pd2.src, 6022 &nk->addr[pd2.sidx], pd2.af) || 6023 nk->port[pd2.sidx] != iih.icmp_id) 6024 pf_change_icmp(pd2.src, &iih.icmp_id, 6025 daddr, &nk->addr[pd2.sidx], 6026 nk->port[pd2.sidx], NULL, 6027 pd2.ip_sum, icmpsum, 6028 pd->ip_sum, 0, AF_INET); 6029 6030 if (PF_ANEQ(pd2.dst, 6031 &nk->addr[pd2.didx], pd2.af) || 6032 nk->port[pd2.didx] != iih.icmp_id) 6033 pf_change_icmp(pd2.dst, &iih.icmp_id, 6034 saddr, &nk->addr[pd2.didx], 6035 nk->port[pd2.didx], NULL, 6036 pd2.ip_sum, icmpsum, 6037 pd->ip_sum, 0, AF_INET); 6038 6039 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6040 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6041 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6042 } 6043 return (PF_PASS); 6044 break; 6045 } 6046 #endif /* INET */ 6047 #ifdef INET6 6048 case IPPROTO_ICMPV6: { 6049 struct icmp6_hdr iih; 6050 6051 if (!pf_pull_hdr(m, off2, &iih, 6052 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6053 DPFPRINTF(PF_DEBUG_MISC, 6054 ("pf: ICMP error message too short " 6055 "(icmp6)\n")); 6056 return (PF_DROP); 6057 } 6058 6059 key.af = pd2.af; 6060 key.proto = IPPROTO_ICMPV6; 6061 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6062 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6063 key.port[0] = key.port[1] = iih.icmp6_id; 6064 6065 STATE_LOOKUP(kif, &key, direction, *state, pd); 6066 6067 /* translate source/destination address, if necessary */ 6068 if ((*state)->key[PF_SK_WIRE] != 6069 (*state)->key[PF_SK_STACK]) { 6070 struct pf_state_key *nk = 6071 (*state)->key[pd->didx]; 6072 6073 if (PF_ANEQ(pd2.src, 6074 &nk->addr[pd2.sidx], pd2.af) || 6075 nk->port[pd2.sidx] != iih.icmp6_id) 6076 pf_change_icmp(pd2.src, &iih.icmp6_id, 6077 daddr, &nk->addr[pd2.sidx], 6078 nk->port[pd2.sidx], NULL, 6079 pd2.ip_sum, icmpsum, 6080 pd->ip_sum, 0, AF_INET6); 6081 6082 if (PF_ANEQ(pd2.dst, 6083 &nk->addr[pd2.didx], pd2.af) || 6084 nk->port[pd2.didx] != iih.icmp6_id) 6085 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6086 saddr, &nk->addr[pd2.didx], 6087 nk->port[pd2.didx], NULL, 6088 pd2.ip_sum, icmpsum, 6089 pd->ip_sum, 0, AF_INET6); 6090 6091 m_copyback(m, off, sizeof(struct icmp6_hdr), 6092 (caddr_t)&pd->hdr.icmp6); 6093 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6094 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6095 (caddr_t)&iih); 6096 } 6097 return (PF_PASS); 6098 break; 6099 } 6100 #endif /* INET6 */ 6101 default: { 6102 key.af = pd2.af; 6103 key.proto = pd2.proto; 6104 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6105 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6106 key.port[0] = key.port[1] = 0; 6107 6108 STATE_LOOKUP(kif, &key, direction, *state, pd); 6109 6110 /* translate source/destination address, if necessary */ 6111 if ((*state)->key[PF_SK_WIRE] != 6112 (*state)->key[PF_SK_STACK]) { 6113 struct pf_state_key *nk = 6114 (*state)->key[pd->didx]; 6115 6116 if (PF_ANEQ(pd2.src, 6117 &nk->addr[pd2.sidx], pd2.af)) 6118 pf_change_icmp(pd2.src, NULL, daddr, 6119 &nk->addr[pd2.sidx], 0, NULL, 6120 pd2.ip_sum, icmpsum, 6121 pd->ip_sum, 0, pd2.af); 6122 6123 if (PF_ANEQ(pd2.dst, 6124 &nk->addr[pd2.didx], pd2.af)) 6125 pf_change_icmp(pd2.dst, NULL, saddr, 6126 &nk->addr[pd2.didx], 0, NULL, 6127 pd2.ip_sum, icmpsum, 6128 pd->ip_sum, 0, pd2.af); 6129 6130 switch (pd2.af) { 6131 #ifdef INET 6132 case AF_INET: 6133 m_copyback(m, off, ICMP_MINLEN, 6134 (caddr_t)&pd->hdr.icmp); 6135 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6136 break; 6137 #endif /* INET */ 6138 #ifdef INET6 6139 case AF_INET6: 6140 m_copyback(m, off, 6141 sizeof(struct icmp6_hdr), 6142 (caddr_t )&pd->hdr.icmp6); 6143 m_copyback(m, ipoff2, sizeof(h2_6), 6144 (caddr_t )&h2_6); 6145 break; 6146 #endif /* INET6 */ 6147 } 6148 } 6149 return (PF_PASS); 6150 break; 6151 } 6152 } 6153 } 6154 } 6155 6156 static int 6157 pf_test_state_other(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 6158 struct mbuf *m, struct pf_pdesc *pd) 6159 { 6160 struct pf_state_peer *src, *dst; 6161 struct pf_state_key_cmp key; 6162 uint8_t psrc, pdst; 6163 6164 bzero(&key, sizeof(key)); 6165 key.af = pd->af; 6166 key.proto = pd->proto; 6167 if (direction == PF_IN) { 6168 PF_ACPY(&key.addr[0], pd->src, key.af); 6169 PF_ACPY(&key.addr[1], pd->dst, key.af); 6170 key.port[0] = key.port[1] = 0; 6171 } else { 6172 PF_ACPY(&key.addr[1], pd->src, key.af); 6173 PF_ACPY(&key.addr[0], pd->dst, key.af); 6174 key.port[1] = key.port[0] = 0; 6175 } 6176 6177 STATE_LOOKUP(kif, &key, direction, *state, pd); 6178 6179 if (direction == (*state)->direction) { 6180 src = &(*state)->src; 6181 dst = &(*state)->dst; 6182 psrc = PF_PEER_SRC; 6183 pdst = PF_PEER_DST; 6184 } else { 6185 src = &(*state)->dst; 6186 dst = &(*state)->src; 6187 psrc = PF_PEER_DST; 6188 pdst = PF_PEER_SRC; 6189 } 6190 6191 /* update states */ 6192 if (src->state < PFOTHERS_SINGLE) 6193 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 6194 if (dst->state == PFOTHERS_SINGLE) 6195 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 6196 6197 /* update expire time */ 6198 (*state)->expire = time_uptime; 6199 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 6200 (*state)->timeout = PFTM_OTHER_MULTIPLE; 6201 else 6202 (*state)->timeout = PFTM_OTHER_SINGLE; 6203 6204 /* translate source/destination address, if necessary */ 6205 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6206 struct pf_state_key *nk = (*state)->key[pd->didx]; 6207 6208 KASSERT(nk, ("%s: nk is null", __func__)); 6209 KASSERT(pd, ("%s: pd is null", __func__)); 6210 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 6211 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 6212 switch (pd->af) { 6213 #ifdef INET 6214 case AF_INET: 6215 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6216 pf_change_a(&pd->src->v4.s_addr, 6217 pd->ip_sum, 6218 nk->addr[pd->sidx].v4.s_addr, 6219 0); 6220 6221 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6222 pf_change_a(&pd->dst->v4.s_addr, 6223 pd->ip_sum, 6224 nk->addr[pd->didx].v4.s_addr, 6225 0); 6226 6227 break; 6228 #endif /* INET */ 6229 #ifdef INET6 6230 case AF_INET6: 6231 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6232 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 6233 6234 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6235 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 6236 #endif /* INET6 */ 6237 } 6238 } 6239 return (PF_PASS); 6240 } 6241 6242 /* 6243 * ipoff and off are measured from the start of the mbuf chain. 6244 * h must be at "ipoff" on the mbuf chain. 6245 */ 6246 void * 6247 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6248 u_short *actionp, u_short *reasonp, sa_family_t af) 6249 { 6250 switch (af) { 6251 #ifdef INET 6252 case AF_INET: { 6253 struct ip *h = mtod(m, struct ip *); 6254 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6255 6256 if (fragoff) { 6257 if (fragoff >= len) 6258 ACTION_SET(actionp, PF_PASS); 6259 else { 6260 ACTION_SET(actionp, PF_DROP); 6261 REASON_SET(reasonp, PFRES_FRAG); 6262 } 6263 return (NULL); 6264 } 6265 if (m->m_pkthdr.len < off + len || 6266 ntohs(h->ip_len) < off + len) { 6267 ACTION_SET(actionp, PF_DROP); 6268 REASON_SET(reasonp, PFRES_SHORT); 6269 return (NULL); 6270 } 6271 break; 6272 } 6273 #endif /* INET */ 6274 #ifdef INET6 6275 case AF_INET6: { 6276 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6277 6278 if (m->m_pkthdr.len < off + len || 6279 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 6280 (unsigned)(off + len)) { 6281 ACTION_SET(actionp, PF_DROP); 6282 REASON_SET(reasonp, PFRES_SHORT); 6283 return (NULL); 6284 } 6285 break; 6286 } 6287 #endif /* INET6 */ 6288 } 6289 m_copydata(m, off, len, p); 6290 return (p); 6291 } 6292 6293 int 6294 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 6295 int rtableid) 6296 { 6297 struct ifnet *ifp; 6298 6299 /* 6300 * Skip check for addresses with embedded interface scope, 6301 * as they would always match anyway. 6302 */ 6303 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 6304 return (1); 6305 6306 if (af != AF_INET && af != AF_INET6) 6307 return (0); 6308 6309 /* Skip checks for ipsec interfaces */ 6310 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6311 return (1); 6312 6313 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 6314 6315 switch (af) { 6316 #ifdef INET6 6317 case AF_INET6: 6318 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 6319 ifp)); 6320 #endif 6321 #ifdef INET 6322 case AF_INET: 6323 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 6324 ifp)); 6325 #endif 6326 } 6327 6328 return (0); 6329 } 6330 6331 #ifdef INET 6332 static void 6333 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 6334 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 6335 { 6336 struct mbuf *m0, *m1, *md; 6337 struct sockaddr_in dst; 6338 struct ip *ip; 6339 struct ifnet *ifp = NULL; 6340 struct pf_addr naddr; 6341 struct pf_ksrc_node *sn = NULL; 6342 int error = 0; 6343 uint16_t ip_len, ip_off; 6344 6345 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 6346 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 6347 __func__)); 6348 6349 if ((pd->pf_mtag == NULL && 6350 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 6351 pd->pf_mtag->routed++ > 3) { 6352 m0 = *m; 6353 *m = NULL; 6354 goto bad_locked; 6355 } 6356 6357 if (r->rt == PF_DUPTO) { 6358 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 6359 if (s == NULL) { 6360 ifp = r->rpool.cur->kif ? 6361 r->rpool.cur->kif->pfik_ifp : NULL; 6362 } else { 6363 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6364 /* If pfsync'd */ 6365 if (ifp == NULL) 6366 ifp = r->rpool.cur->kif ? 6367 r->rpool.cur->kif->pfik_ifp : NULL; 6368 PF_STATE_UNLOCK(s); 6369 } 6370 if (ifp == oifp) { 6371 /* When the 2nd interface is not skipped */ 6372 return; 6373 } else { 6374 m0 = *m; 6375 *m = NULL; 6376 goto bad; 6377 } 6378 } else { 6379 pd->pf_mtag->flags |= PF_DUPLICATED; 6380 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6381 if (s) 6382 PF_STATE_UNLOCK(s); 6383 return; 6384 } 6385 } 6386 } else { 6387 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6388 pf_dummynet(pd, dir, s, r, m); 6389 if (s) 6390 PF_STATE_UNLOCK(s); 6391 return; 6392 } 6393 m0 = *m; 6394 } 6395 6396 ip = mtod(m0, struct ip *); 6397 6398 bzero(&dst, sizeof(dst)); 6399 dst.sin_family = AF_INET; 6400 dst.sin_len = sizeof(dst); 6401 dst.sin_addr = ip->ip_dst; 6402 6403 bzero(&naddr, sizeof(naddr)); 6404 6405 if (s == NULL) { 6406 if (TAILQ_EMPTY(&r->rpool.list)) { 6407 DPFPRINTF(PF_DEBUG_URGENT, 6408 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 6409 goto bad_locked; 6410 } 6411 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 6412 &naddr, NULL, &sn); 6413 if (!PF_AZERO(&naddr, AF_INET)) 6414 dst.sin_addr.s_addr = naddr.v4.s_addr; 6415 ifp = r->rpool.cur->kif ? 6416 r->rpool.cur->kif->pfik_ifp : NULL; 6417 } else { 6418 if (!PF_AZERO(&s->rt_addr, AF_INET)) 6419 dst.sin_addr.s_addr = 6420 s->rt_addr.v4.s_addr; 6421 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6422 PF_STATE_UNLOCK(s); 6423 } 6424 /* If pfsync'd */ 6425 if (ifp == NULL) 6426 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6427 if (ifp == NULL) 6428 goto bad; 6429 6430 if (dir == PF_IN) { 6431 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) 6432 goto bad; 6433 else if (m0 == NULL) 6434 goto done; 6435 if (m0->m_len < sizeof(struct ip)) { 6436 DPFPRINTF(PF_DEBUG_URGENT, 6437 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 6438 goto bad; 6439 } 6440 ip = mtod(m0, struct ip *); 6441 } 6442 6443 if (ifp->if_flags & IFF_LOOPBACK) 6444 m0->m_flags |= M_SKIP_FIREWALL; 6445 6446 ip_len = ntohs(ip->ip_len); 6447 ip_off = ntohs(ip->ip_off); 6448 6449 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 6450 m0->m_pkthdr.csum_flags |= CSUM_IP; 6451 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 6452 in_delayed_cksum(m0); 6453 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 6454 } 6455 #if defined(SCTP) || defined(SCTP_SUPPORT) 6456 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 6457 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2)); 6458 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 6459 } 6460 #endif 6461 6462 /* 6463 * If small enough for interface, or the interface will take 6464 * care of the fragmentation for us, we can just send directly. 6465 */ 6466 if (ip_len <= ifp->if_mtu || 6467 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 6468 ip->ip_sum = 0; 6469 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 6470 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 6471 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 6472 } 6473 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 6474 6475 md = m0; 6476 error = pf_dummynet_route(pd, dir, s, r, ifp, sintosa(&dst), &md); 6477 if (md != NULL) 6478 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 6479 goto done; 6480 } 6481 6482 /* Balk when DF bit is set or the interface didn't support TSO. */ 6483 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 6484 error = EMSGSIZE; 6485 KMOD_IPSTAT_INC(ips_cantfrag); 6486 if (r->rt != PF_DUPTO) { 6487 if (s && pd->nat_rule != NULL) 6488 PACKET_UNDO_NAT(m0, pd, 6489 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 6490 s, dir); 6491 6492 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 6493 ifp->if_mtu); 6494 goto done; 6495 } else 6496 goto bad; 6497 } 6498 6499 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 6500 if (error) 6501 goto bad; 6502 6503 for (; m0; m0 = m1) { 6504 m1 = m0->m_nextpkt; 6505 m0->m_nextpkt = NULL; 6506 if (error == 0) { 6507 m_clrprotoflags(m0); 6508 md = m0; 6509 error = pf_dummynet_route(pd, dir, s, r, ifp, 6510 sintosa(&dst), &md); 6511 if (md != NULL) 6512 error = (*ifp->if_output)(ifp, md, 6513 sintosa(&dst), NULL); 6514 } else 6515 m_freem(m0); 6516 } 6517 6518 if (error == 0) 6519 KMOD_IPSTAT_INC(ips_fragmented); 6520 6521 done: 6522 if (r->rt != PF_DUPTO) 6523 *m = NULL; 6524 return; 6525 6526 bad_locked: 6527 if (s) 6528 PF_STATE_UNLOCK(s); 6529 bad: 6530 m_freem(m0); 6531 goto done; 6532 } 6533 #endif /* INET */ 6534 6535 #ifdef INET6 6536 static void 6537 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 6538 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 6539 { 6540 struct mbuf *m0, *md; 6541 struct sockaddr_in6 dst; 6542 struct ip6_hdr *ip6; 6543 struct ifnet *ifp = NULL; 6544 struct pf_addr naddr; 6545 struct pf_ksrc_node *sn = NULL; 6546 6547 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 6548 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 6549 __func__)); 6550 6551 if ((pd->pf_mtag == NULL && 6552 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 6553 pd->pf_mtag->routed++ > 3) { 6554 m0 = *m; 6555 *m = NULL; 6556 goto bad_locked; 6557 } 6558 6559 if (r->rt == PF_DUPTO) { 6560 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 6561 if (s == NULL) { 6562 ifp = r->rpool.cur->kif ? 6563 r->rpool.cur->kif->pfik_ifp : NULL; 6564 } else { 6565 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6566 /* If pfsync'd */ 6567 if (ifp == NULL) 6568 ifp = r->rpool.cur->kif ? 6569 r->rpool.cur->kif->pfik_ifp : NULL; 6570 PF_STATE_UNLOCK(s); 6571 } 6572 if (ifp == oifp) { 6573 /* When the 2nd interface is not skipped */ 6574 return; 6575 } else { 6576 m0 = *m; 6577 *m = NULL; 6578 goto bad; 6579 } 6580 } else { 6581 pd->pf_mtag->flags |= PF_DUPLICATED; 6582 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6583 if (s) 6584 PF_STATE_UNLOCK(s); 6585 return; 6586 } 6587 } 6588 } else { 6589 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6590 pf_dummynet(pd, dir, s, r, m); 6591 if (s) 6592 PF_STATE_UNLOCK(s); 6593 return; 6594 } 6595 m0 = *m; 6596 } 6597 6598 ip6 = mtod(m0, struct ip6_hdr *); 6599 6600 bzero(&dst, sizeof(dst)); 6601 dst.sin6_family = AF_INET6; 6602 dst.sin6_len = sizeof(dst); 6603 dst.sin6_addr = ip6->ip6_dst; 6604 6605 bzero(&naddr, sizeof(naddr)); 6606 6607 if (s == NULL) { 6608 if (TAILQ_EMPTY(&r->rpool.list)) { 6609 DPFPRINTF(PF_DEBUG_URGENT, 6610 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 6611 goto bad_locked; 6612 } 6613 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 6614 &naddr, NULL, &sn); 6615 if (!PF_AZERO(&naddr, AF_INET6)) 6616 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6617 &naddr, AF_INET6); 6618 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6619 } else { 6620 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 6621 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6622 &s->rt_addr, AF_INET6); 6623 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6624 } 6625 6626 if (s) 6627 PF_STATE_UNLOCK(s); 6628 6629 /* If pfsync'd */ 6630 if (ifp == NULL) 6631 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6632 if (ifp == NULL) 6633 goto bad; 6634 6635 if (dir == PF_IN) { 6636 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS) 6637 goto bad; 6638 else if (m0 == NULL) 6639 goto done; 6640 if (m0->m_len < sizeof(struct ip6_hdr)) { 6641 DPFPRINTF(PF_DEBUG_URGENT, 6642 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 6643 __func__)); 6644 goto bad; 6645 } 6646 ip6 = mtod(m0, struct ip6_hdr *); 6647 } 6648 6649 if (ifp->if_flags & IFF_LOOPBACK) 6650 m0->m_flags |= M_SKIP_FIREWALL; 6651 6652 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 6653 ~ifp->if_hwassist) { 6654 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 6655 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 6656 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 6657 } 6658 6659 /* 6660 * If the packet is too large for the outgoing interface, 6661 * send back an icmp6 error. 6662 */ 6663 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 6664 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 6665 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 6666 md = m0; 6667 pf_dummynet_route(pd, dir, s, r, ifp, sintosa(&dst), &md); 6668 if (md != NULL) 6669 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 6670 } 6671 else { 6672 in6_ifstat_inc(ifp, ifs6_in_toobig); 6673 if (r->rt != PF_DUPTO) { 6674 if (s && pd->nat_rule != NULL) 6675 PACKET_UNDO_NAT(m0, pd, 6676 ((caddr_t)ip6 - m0->m_data) + 6677 sizeof(struct ip6_hdr), s, dir); 6678 6679 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 6680 } else 6681 goto bad; 6682 } 6683 6684 done: 6685 if (r->rt != PF_DUPTO) 6686 *m = NULL; 6687 return; 6688 6689 bad_locked: 6690 if (s) 6691 PF_STATE_UNLOCK(s); 6692 bad: 6693 m_freem(m0); 6694 goto done; 6695 } 6696 #endif /* INET6 */ 6697 6698 /* 6699 * FreeBSD supports cksum offloads for the following drivers. 6700 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 6701 * 6702 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 6703 * network driver performed cksum including pseudo header, need to verify 6704 * csum_data 6705 * CSUM_DATA_VALID : 6706 * network driver performed cksum, needs to additional pseudo header 6707 * cksum computation with partial csum_data(i.e. lack of H/W support for 6708 * pseudo header, for instance sk(4) and possibly gem(4)) 6709 * 6710 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 6711 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 6712 * TCP/UDP layer. 6713 * Also, set csum_data to 0xffff to force cksum validation. 6714 */ 6715 static int 6716 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 6717 { 6718 u_int16_t sum = 0; 6719 int hw_assist = 0; 6720 struct ip *ip; 6721 6722 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 6723 return (1); 6724 if (m->m_pkthdr.len < off + len) 6725 return (1); 6726 6727 switch (p) { 6728 case IPPROTO_TCP: 6729 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6730 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6731 sum = m->m_pkthdr.csum_data; 6732 } else { 6733 ip = mtod(m, struct ip *); 6734 sum = in_pseudo(ip->ip_src.s_addr, 6735 ip->ip_dst.s_addr, htonl((u_short)len + 6736 m->m_pkthdr.csum_data + IPPROTO_TCP)); 6737 } 6738 sum ^= 0xffff; 6739 ++hw_assist; 6740 } 6741 break; 6742 case IPPROTO_UDP: 6743 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6744 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6745 sum = m->m_pkthdr.csum_data; 6746 } else { 6747 ip = mtod(m, struct ip *); 6748 sum = in_pseudo(ip->ip_src.s_addr, 6749 ip->ip_dst.s_addr, htonl((u_short)len + 6750 m->m_pkthdr.csum_data + IPPROTO_UDP)); 6751 } 6752 sum ^= 0xffff; 6753 ++hw_assist; 6754 } 6755 break; 6756 case IPPROTO_ICMP: 6757 #ifdef INET6 6758 case IPPROTO_ICMPV6: 6759 #endif /* INET6 */ 6760 break; 6761 default: 6762 return (1); 6763 } 6764 6765 if (!hw_assist) { 6766 switch (af) { 6767 case AF_INET: 6768 if (p == IPPROTO_ICMP) { 6769 if (m->m_len < off) 6770 return (1); 6771 m->m_data += off; 6772 m->m_len -= off; 6773 sum = in_cksum(m, len); 6774 m->m_data -= off; 6775 m->m_len += off; 6776 } else { 6777 if (m->m_len < sizeof(struct ip)) 6778 return (1); 6779 sum = in4_cksum(m, p, off, len); 6780 } 6781 break; 6782 #ifdef INET6 6783 case AF_INET6: 6784 if (m->m_len < sizeof(struct ip6_hdr)) 6785 return (1); 6786 sum = in6_cksum(m, p, off, len); 6787 break; 6788 #endif /* INET6 */ 6789 default: 6790 return (1); 6791 } 6792 } 6793 if (sum) { 6794 switch (p) { 6795 case IPPROTO_TCP: 6796 { 6797 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 6798 break; 6799 } 6800 case IPPROTO_UDP: 6801 { 6802 KMOD_UDPSTAT_INC(udps_badsum); 6803 break; 6804 } 6805 #ifdef INET 6806 case IPPROTO_ICMP: 6807 { 6808 KMOD_ICMPSTAT_INC(icps_checksum); 6809 break; 6810 } 6811 #endif 6812 #ifdef INET6 6813 case IPPROTO_ICMPV6: 6814 { 6815 KMOD_ICMP6STAT_INC(icp6s_checksum); 6816 break; 6817 } 6818 #endif /* INET6 */ 6819 } 6820 return (1); 6821 } else { 6822 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 6823 m->m_pkthdr.csum_flags |= 6824 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 6825 m->m_pkthdr.csum_data = 0xffff; 6826 } 6827 } 6828 return (0); 6829 } 6830 6831 static bool 6832 pf_pdesc_to_dnflow(int dir, const struct pf_pdesc *pd, 6833 const struct pf_krule *r, const struct pf_kstate *s, 6834 struct ip_fw_args *dnflow) 6835 { 6836 int dndir = r->direction; 6837 6838 if (s && dndir == PF_INOUT) { 6839 dndir = s->direction; 6840 } else if (dndir == PF_INOUT) { 6841 /* Assume primary direction. Happens when we've set dnpipe in 6842 * the ethernet level code. */ 6843 dndir = dir; 6844 } 6845 6846 memset(dnflow, 0, sizeof(*dnflow)); 6847 6848 if (pd->dport != NULL) 6849 dnflow->f_id.dst_port = ntohs(*pd->dport); 6850 if (pd->sport != NULL) 6851 dnflow->f_id.src_port = ntohs(*pd->sport); 6852 6853 if (dir == PF_IN) 6854 dnflow->flags |= IPFW_ARGS_IN; 6855 else 6856 dnflow->flags |= IPFW_ARGS_OUT; 6857 6858 if (dir != dndir && pd->act.dnrpipe) { 6859 dnflow->rule.info = pd->act.dnrpipe; 6860 } 6861 else if (dir == dndir && pd->act.dnpipe) { 6862 dnflow->rule.info = pd->act.dnpipe; 6863 } 6864 else { 6865 return (false); 6866 } 6867 6868 dnflow->rule.info |= IPFW_IS_DUMMYNET; 6869 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFRULE_DN_IS_PIPE) 6870 dnflow->rule.info |= IPFW_IS_PIPE; 6871 6872 dnflow->f_id.proto = pd->proto; 6873 dnflow->f_id.extra = dnflow->rule.info; 6874 switch (pd->af) { 6875 case AF_INET: 6876 dnflow->f_id.addr_type = 4; 6877 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 6878 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 6879 break; 6880 case AF_INET6: 6881 dnflow->flags |= IPFW_ARGS_IP6; 6882 dnflow->f_id.addr_type = 6; 6883 dnflow->f_id.src_ip6 = pd->src->v6; 6884 dnflow->f_id.dst_ip6 = pd->dst->v6; 6885 break; 6886 default: 6887 panic("Invalid AF"); 6888 break; 6889 } 6890 6891 return (true); 6892 } 6893 6894 int 6895 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 6896 struct inpcb *inp) 6897 { 6898 struct pfi_kkif *kif; 6899 struct mbuf *m = *m0; 6900 6901 M_ASSERTPKTHDR(m); 6902 MPASS(ifp->if_vnet == curvnet); 6903 NET_EPOCH_ASSERT(); 6904 6905 if (!V_pf_status.running) 6906 return (PF_PASS); 6907 6908 kif = (struct pfi_kkif *)ifp->if_pf_kif; 6909 6910 if (kif == NULL) { 6911 DPFPRINTF(PF_DEBUG_URGENT, 6912 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 6913 return (PF_DROP); 6914 } 6915 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6916 return (PF_PASS); 6917 6918 if (m->m_flags & M_SKIP_FIREWALL) 6919 return (PF_PASS); 6920 6921 /* Stateless! */ 6922 return (pf_test_eth_rule(dir, kif, m0)); 6923 } 6924 6925 static int 6926 pf_dummynet(struct pf_pdesc *pd, int dir, struct pf_kstate *s, 6927 struct pf_krule *r, struct mbuf **m0) 6928 { 6929 return (pf_dummynet_route(pd, dir, s, r, NULL, NULL, m0)); 6930 } 6931 6932 static int 6933 pf_dummynet_route(struct pf_pdesc *pd, int dir, struct pf_kstate *s, 6934 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 6935 struct mbuf **m0) 6936 { 6937 NET_EPOCH_ASSERT(); 6938 6939 if (s && (s->dnpipe || s->dnrpipe)) { 6940 pd->act.dnpipe = s->dnpipe; 6941 pd->act.dnrpipe = s->dnrpipe; 6942 pd->act.flags = s->state_flags; 6943 } else if (r->dnpipe || r->dnrpipe) { 6944 pd->act.dnpipe = r->dnpipe; 6945 pd->act.dnrpipe = r->dnrpipe; 6946 pd->act.flags = r->free_flags; 6947 } 6948 if (pd->act.dnpipe || pd->act.dnrpipe) { 6949 struct ip_fw_args dnflow; 6950 if (ip_dn_io_ptr == NULL) { 6951 m_freem(*m0); 6952 *m0 = NULL; 6953 return (ENOMEM); 6954 } 6955 6956 if (pd->pf_mtag == NULL && 6957 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 6958 m_freem(*m0); 6959 *m0 = NULL; 6960 return (ENOMEM); 6961 } 6962 6963 if (ifp != NULL) { 6964 pd->pf_mtag->flags |= PF_TAG_ROUTE_TO; 6965 6966 pd->pf_mtag->if_index = ifp->if_index; 6967 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 6968 6969 MPASS(sa != NULL); 6970 6971 if (pd->af == AF_INET) 6972 memcpy(&pd->pf_mtag->dst, sa, 6973 sizeof(struct sockaddr_in)); 6974 else 6975 memcpy(&pd->pf_mtag->dst, sa, 6976 sizeof(struct sockaddr_in6)); 6977 } 6978 6979 if (pf_pdesc_to_dnflow(dir, pd, r, s, &dnflow)) { 6980 pd->pf_mtag->flags |= PF_TAG_DUMMYNET; 6981 ip_dn_io_ptr(m0, &dnflow); 6982 if (*m0 != NULL) 6983 pd->pf_mtag->flags &= ~PF_TAG_DUMMYNET; 6984 } 6985 } 6986 6987 return (0); 6988 } 6989 6990 #ifdef INET 6991 int 6992 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6993 { 6994 struct pfi_kkif *kif; 6995 u_short action, reason = 0, log = 0; 6996 struct mbuf *m = *m0; 6997 struct ip *h = NULL; 6998 struct m_tag *ipfwtag; 6999 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7000 struct pf_kstate *s = NULL; 7001 struct pf_kruleset *ruleset = NULL; 7002 struct pf_pdesc pd; 7003 int off, dirndx, pqid = 0; 7004 7005 PF_RULES_RLOCK_TRACKER; 7006 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7007 M_ASSERTPKTHDR(m); 7008 7009 if (!V_pf_status.running) 7010 return (PF_PASS); 7011 7012 PF_RULES_RLOCK(); 7013 7014 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7015 7016 if (__predict_false(kif == NULL)) { 7017 DPFPRINTF(PF_DEBUG_URGENT, 7018 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7019 PF_RULES_RUNLOCK(); 7020 return (PF_DROP); 7021 } 7022 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7023 PF_RULES_RUNLOCK(); 7024 return (PF_PASS); 7025 } 7026 7027 if (m->m_flags & M_SKIP_FIREWALL) { 7028 PF_RULES_RUNLOCK(); 7029 return (PF_PASS); 7030 } 7031 7032 memset(&pd, 0, sizeof(pd)); 7033 pd.pf_mtag = pf_find_mtag(m); 7034 7035 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_TAG_ROUTE_TO)) { 7036 pd.pf_mtag->flags &= ~PF_TAG_ROUTE_TO; 7037 7038 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7039 pd.pf_mtag->if_idxgen); 7040 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7041 PF_RULES_RUNLOCK(); 7042 m_freem(*m0); 7043 *m0 = NULL; 7044 return (PF_PASS); 7045 } 7046 PF_RULES_RUNLOCK(); 7047 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 7048 *m0 = NULL; 7049 return (PF_PASS); 7050 } 7051 7052 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7053 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7054 pd.act.flags = pd.pf_mtag->dnflags; 7055 } 7056 7057 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7058 pd.pf_mtag->flags & PF_TAG_DUMMYNET) { 7059 /* Dummynet re-injects packets after they've 7060 * completed their delay. We've already 7061 * processed them, so pass unconditionally. */ 7062 7063 /* But only once. We may see the packet multiple times (e.g. 7064 * PFIL_IN/PFIL_OUT). */ 7065 pd.pf_mtag->flags &= ~PF_TAG_DUMMYNET; 7066 PF_RULES_RUNLOCK(); 7067 7068 return (PF_PASS); 7069 } 7070 7071 if (__predict_false(ip_divert_ptr != NULL) && 7072 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 7073 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 7074 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 7075 if (pd.pf_mtag == NULL && 7076 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7077 action = PF_DROP; 7078 goto done; 7079 } 7080 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 7081 m_tag_delete(m, ipfwtag); 7082 } 7083 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 7084 m->m_flags |= M_FASTFWD_OURS; 7085 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 7086 } 7087 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 7088 /* We do IP header normalization and packet reassembly here */ 7089 action = PF_DROP; 7090 goto done; 7091 } 7092 m = *m0; /* pf_normalize messes with m0 */ 7093 h = mtod(m, struct ip *); 7094 7095 off = h->ip_hl << 2; 7096 if (off < (int)sizeof(struct ip)) { 7097 action = PF_DROP; 7098 REASON_SET(&reason, PFRES_SHORT); 7099 log = 1; 7100 goto done; 7101 } 7102 7103 pd.src = (struct pf_addr *)&h->ip_src; 7104 pd.dst = (struct pf_addr *)&h->ip_dst; 7105 pd.sport = pd.dport = NULL; 7106 pd.ip_sum = &h->ip_sum; 7107 pd.proto_sum = NULL; 7108 pd.proto = h->ip_p; 7109 pd.dir = dir; 7110 pd.sidx = (dir == PF_IN) ? 0 : 1; 7111 pd.didx = (dir == PF_IN) ? 1 : 0; 7112 pd.af = AF_INET; 7113 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 7114 pd.tot_len = ntohs(h->ip_len); 7115 7116 /* handle fragments that didn't get reassembled by normalization */ 7117 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 7118 action = pf_test_fragment(&r, dir, kif, m, h, 7119 &pd, &a, &ruleset); 7120 goto done; 7121 } 7122 7123 switch (h->ip_p) { 7124 case IPPROTO_TCP: { 7125 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 7126 &action, &reason, AF_INET)) { 7127 log = action != PF_PASS; 7128 goto done; 7129 } 7130 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 7131 7132 pd.sport = &pd.hdr.tcp.th_sport; 7133 pd.dport = &pd.hdr.tcp.th_dport; 7134 7135 /* Respond to SYN with a syncookie. */ 7136 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 7137 pd.dir == PF_IN && pf_synflood_check(&pd)) { 7138 pf_syncookie_send(m, off, &pd); 7139 action = PF_DROP; 7140 break; 7141 } 7142 7143 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7144 pqid = 1; 7145 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 7146 if (action == PF_DROP) 7147 goto done; 7148 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 7149 &reason); 7150 if (action == PF_PASS) { 7151 if (V_pfsync_update_state_ptr != NULL) 7152 V_pfsync_update_state_ptr(s); 7153 r = s->rule.ptr; 7154 a = s->anchor.ptr; 7155 log = s->log; 7156 } else if (s == NULL) { 7157 /* Validate remote SYN|ACK, re-create original SYN if 7158 * valid. */ 7159 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 7160 TH_ACK && pf_syncookie_validate(&pd) && 7161 pd.dir == PF_IN) { 7162 struct mbuf *msyn; 7163 7164 msyn = pf_syncookie_recreate_syn(h->ip_ttl, 7165 off,&pd); 7166 if (msyn == NULL) { 7167 action = PF_DROP; 7168 break; 7169 } 7170 7171 action = pf_test(dir, pflags, ifp, &msyn, inp); 7172 m_freem(msyn); 7173 7174 if (action == PF_PASS) { 7175 action = pf_test_state_tcp(&s, dir, 7176 kif, m, off, h, &pd, &reason); 7177 if (action != PF_PASS || s == NULL) { 7178 action = PF_DROP; 7179 break; 7180 } 7181 7182 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) 7183 - 1; 7184 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) 7185 - 1; 7186 pf_set_protostate(s, PF_PEER_SRC, 7187 PF_TCPS_PROXY_DST); 7188 7189 action = pf_synproxy(&pd, &s, &reason); 7190 if (action != PF_PASS) 7191 break; 7192 } 7193 break; 7194 } 7195 else { 7196 action = pf_test_rule(&r, &s, dir, kif, m, off, 7197 &pd, &a, &ruleset, inp); 7198 } 7199 } 7200 break; 7201 } 7202 7203 case IPPROTO_UDP: { 7204 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 7205 &action, &reason, AF_INET)) { 7206 log = action != PF_PASS; 7207 goto done; 7208 } 7209 pd.sport = &pd.hdr.udp.uh_sport; 7210 pd.dport = &pd.hdr.udp.uh_dport; 7211 if (pd.hdr.udp.uh_dport == 0 || 7212 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 7213 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 7214 action = PF_DROP; 7215 REASON_SET(&reason, PFRES_SHORT); 7216 goto done; 7217 } 7218 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7219 if (action == PF_PASS) { 7220 if (V_pfsync_update_state_ptr != NULL) 7221 V_pfsync_update_state_ptr(s); 7222 r = s->rule.ptr; 7223 a = s->anchor.ptr; 7224 log = s->log; 7225 } else if (s == NULL) 7226 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7227 &a, &ruleset, inp); 7228 break; 7229 } 7230 7231 case IPPROTO_ICMP: { 7232 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 7233 &action, &reason, AF_INET)) { 7234 log = action != PF_PASS; 7235 goto done; 7236 } 7237 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 7238 &reason); 7239 if (action == PF_PASS) { 7240 if (V_pfsync_update_state_ptr != NULL) 7241 V_pfsync_update_state_ptr(s); 7242 r = s->rule.ptr; 7243 a = s->anchor.ptr; 7244 log = s->log; 7245 } else if (s == NULL) 7246 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7247 &a, &ruleset, inp); 7248 break; 7249 } 7250 7251 #ifdef INET6 7252 case IPPROTO_ICMPV6: { 7253 action = PF_DROP; 7254 DPFPRINTF(PF_DEBUG_MISC, 7255 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 7256 goto done; 7257 } 7258 #endif 7259 7260 default: 7261 action = pf_test_state_other(&s, dir, kif, m, &pd); 7262 if (action == PF_PASS) { 7263 if (V_pfsync_update_state_ptr != NULL) 7264 V_pfsync_update_state_ptr(s); 7265 r = s->rule.ptr; 7266 a = s->anchor.ptr; 7267 log = s->log; 7268 } else if (s == NULL) 7269 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7270 &a, &ruleset, inp); 7271 break; 7272 } 7273 7274 done: 7275 PF_RULES_RUNLOCK(); 7276 if (action == PF_PASS && h->ip_hl > 5 && 7277 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7278 action = PF_DROP; 7279 REASON_SET(&reason, PFRES_IPOPTIONS); 7280 log = r->log; 7281 DPFPRINTF(PF_DEBUG_MISC, 7282 ("pf: dropping packet with ip options\n")); 7283 } 7284 7285 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 7286 action = PF_DROP; 7287 REASON_SET(&reason, PFRES_MEMORY); 7288 } 7289 if (r->rtableid >= 0) 7290 M_SETFIB(m, r->rtableid); 7291 7292 if (r->scrub_flags & PFSTATE_SETPRIO) { 7293 if (pd.tos & IPTOS_LOWDELAY) 7294 pqid = 1; 7295 if (vlan_set_pcp(m, r->set_prio[pqid])) { 7296 action = PF_DROP; 7297 REASON_SET(&reason, PFRES_MEMORY); 7298 log = 1; 7299 DPFPRINTF(PF_DEBUG_MISC, 7300 ("pf: failed to allocate 802.1q mtag\n")); 7301 } 7302 } 7303 7304 #ifdef ALTQ 7305 if (s && s->qid) { 7306 pd.act.pqid = s->pqid; 7307 pd.act.qid = s->qid; 7308 } else if (r->qid) { 7309 pd.act.pqid = r->pqid; 7310 pd.act.qid = r->qid; 7311 } 7312 if (action == PF_PASS && pd.act.qid) { 7313 if (pd.pf_mtag == NULL && 7314 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7315 action = PF_DROP; 7316 REASON_SET(&reason, PFRES_MEMORY); 7317 } else { 7318 if (s != NULL) 7319 pd.pf_mtag->qid_hash = pf_state_hash(s); 7320 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 7321 pd.pf_mtag->qid = pd.act.pqid; 7322 else 7323 pd.pf_mtag->qid = pd.act.qid; 7324 /* Add hints for ecn. */ 7325 pd.pf_mtag->hdr = h; 7326 } 7327 } 7328 #endif /* ALTQ */ 7329 7330 /* 7331 * connections redirected to loopback should not match sockets 7332 * bound specifically to loopback due to security implications, 7333 * see tcp_input() and in_pcblookup_listen(). 7334 */ 7335 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7336 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7337 (s->nat_rule.ptr->action == PF_RDR || 7338 s->nat_rule.ptr->action == PF_BINAT) && 7339 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 7340 m->m_flags |= M_SKIP_FIREWALL; 7341 7342 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 7343 r->divert.port && !PACKET_LOOPED(&pd)) { 7344 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 7345 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 7346 if (ipfwtag != NULL) { 7347 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 7348 ntohs(r->divert.port); 7349 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 7350 7351 if (s) 7352 PF_STATE_UNLOCK(s); 7353 7354 m_tag_prepend(m, ipfwtag); 7355 if (m->m_flags & M_FASTFWD_OURS) { 7356 if (pd.pf_mtag == NULL && 7357 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7358 action = PF_DROP; 7359 REASON_SET(&reason, PFRES_MEMORY); 7360 log = 1; 7361 DPFPRINTF(PF_DEBUG_MISC, 7362 ("pf: failed to allocate tag\n")); 7363 } else { 7364 pd.pf_mtag->flags |= 7365 PF_FASTFWD_OURS_PRESENT; 7366 m->m_flags &= ~M_FASTFWD_OURS; 7367 } 7368 } 7369 ip_divert_ptr(*m0, dir == PF_IN); 7370 *m0 = NULL; 7371 7372 return (action); 7373 } else { 7374 /* XXX: ipfw has the same behaviour! */ 7375 action = PF_DROP; 7376 REASON_SET(&reason, PFRES_MEMORY); 7377 log = 1; 7378 DPFPRINTF(PF_DEBUG_MISC, 7379 ("pf: failed to allocate divert tag\n")); 7380 } 7381 } 7382 7383 if (log) { 7384 struct pf_krule *lr; 7385 7386 if (s != NULL && s->nat_rule.ptr != NULL && 7387 s->nat_rule.ptr->log & PF_LOG_ALL) 7388 lr = s->nat_rule.ptr; 7389 else 7390 lr = r; 7391 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 7392 (s == NULL)); 7393 } 7394 7395 pf_counter_u64_critical_enter(); 7396 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 7397 pd.tot_len); 7398 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 7399 1); 7400 7401 if (action == PF_PASS || r->action == PF_DROP) { 7402 dirndx = (dir == PF_OUT); 7403 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 7404 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 7405 pf_update_timestamp(r); 7406 7407 if (a != NULL) { 7408 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 7409 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 7410 } 7411 if (s != NULL) { 7412 if (s->nat_rule.ptr != NULL) { 7413 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 7414 1); 7415 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 7416 pd.tot_len); 7417 } 7418 if (s->src_node != NULL) { 7419 counter_u64_add(s->src_node->packets[dirndx], 7420 1); 7421 counter_u64_add(s->src_node->bytes[dirndx], 7422 pd.tot_len); 7423 } 7424 if (s->nat_src_node != NULL) { 7425 counter_u64_add(s->nat_src_node->packets[dirndx], 7426 1); 7427 counter_u64_add(s->nat_src_node->bytes[dirndx], 7428 pd.tot_len); 7429 } 7430 dirndx = (dir == s->direction) ? 0 : 1; 7431 s->packets[dirndx]++; 7432 s->bytes[dirndx] += pd.tot_len; 7433 } 7434 tr = r; 7435 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7436 if (nr != NULL && r == &V_pf_default_rule) 7437 tr = nr; 7438 if (tr->src.addr.type == PF_ADDR_TABLE) 7439 pfr_update_stats(tr->src.addr.p.tbl, 7440 (s == NULL) ? pd.src : 7441 &s->key[(s->direction == PF_IN)]-> 7442 addr[(s->direction == PF_OUT)], 7443 pd.af, pd.tot_len, dir == PF_OUT, 7444 r->action == PF_PASS, tr->src.neg); 7445 if (tr->dst.addr.type == PF_ADDR_TABLE) 7446 pfr_update_stats(tr->dst.addr.p.tbl, 7447 (s == NULL) ? pd.dst : 7448 &s->key[(s->direction == PF_IN)]-> 7449 addr[(s->direction == PF_IN)], 7450 pd.af, pd.tot_len, dir == PF_OUT, 7451 r->action == PF_PASS, tr->dst.neg); 7452 } 7453 pf_counter_u64_critical_exit(); 7454 7455 switch (action) { 7456 case PF_SYNPROXY_DROP: 7457 m_freem(*m0); 7458 case PF_DEFER: 7459 *m0 = NULL; 7460 action = PF_PASS; 7461 break; 7462 case PF_DROP: 7463 m_freem(*m0); 7464 *m0 = NULL; 7465 break; 7466 default: 7467 /* pf_route() returns unlocked. */ 7468 if (r->rt) { 7469 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 7470 return (action); 7471 } 7472 if (pf_dummynet(&pd, dir, s, r, m0) != 0) { 7473 action = PF_DROP; 7474 REASON_SET(&reason, PFRES_MEMORY); 7475 } 7476 break; 7477 } 7478 7479 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 7480 7481 if (s) 7482 PF_STATE_UNLOCK(s); 7483 7484 return (action); 7485 } 7486 #endif /* INET */ 7487 7488 #ifdef INET6 7489 int 7490 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 7491 { 7492 struct pfi_kkif *kif; 7493 u_short action, reason = 0, log = 0; 7494 struct mbuf *m = *m0, *n = NULL; 7495 struct m_tag *mtag; 7496 struct ip6_hdr *h = NULL; 7497 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7498 struct pf_kstate *s = NULL; 7499 struct pf_kruleset *ruleset = NULL; 7500 struct pf_pdesc pd; 7501 int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; 7502 7503 PF_RULES_RLOCK_TRACKER; 7504 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7505 M_ASSERTPKTHDR(m); 7506 7507 if (!V_pf_status.running) 7508 return (PF_PASS); 7509 7510 PF_RULES_RLOCK(); 7511 7512 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7513 if (__predict_false(kif == NULL)) { 7514 DPFPRINTF(PF_DEBUG_URGENT, 7515 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 7516 PF_RULES_RUNLOCK(); 7517 return (PF_DROP); 7518 } 7519 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7520 PF_RULES_RUNLOCK(); 7521 return (PF_PASS); 7522 } 7523 7524 if (m->m_flags & M_SKIP_FIREWALL) { 7525 PF_RULES_RUNLOCK(); 7526 return (PF_PASS); 7527 } 7528 7529 memset(&pd, 0, sizeof(pd)); 7530 pd.pf_mtag = pf_find_mtag(m); 7531 7532 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_TAG_ROUTE_TO)) { 7533 pd.pf_mtag->flags &= ~PF_TAG_ROUTE_TO; 7534 7535 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7536 pd.pf_mtag->if_idxgen); 7537 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7538 PF_RULES_RUNLOCK(); 7539 m_freem(*m0); 7540 *m0 = NULL; 7541 return (PF_PASS); 7542 } 7543 PF_RULES_RUNLOCK(); 7544 nd6_output_ifp(ifp, ifp, m, 7545 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 7546 *m0 = NULL; 7547 return (PF_PASS); 7548 } 7549 7550 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7551 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7552 pd.act.flags = pd.pf_mtag->dnflags; 7553 } 7554 7555 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7556 pd.pf_mtag->flags & PF_TAG_DUMMYNET) { 7557 pd.pf_mtag->flags &= ~PF_TAG_DUMMYNET; 7558 /* Dummynet re-injects packets after they've 7559 * completed their delay. We've already 7560 * processed them, so pass unconditionally. */ 7561 PF_RULES_RUNLOCK(); 7562 return (PF_PASS); 7563 } 7564 7565 /* We do IP header normalization and packet reassembly here */ 7566 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 7567 action = PF_DROP; 7568 goto done; 7569 } 7570 m = *m0; /* pf_normalize messes with m0 */ 7571 h = mtod(m, struct ip6_hdr *); 7572 7573 /* 7574 * we do not support jumbogram. if we keep going, zero ip6_plen 7575 * will do something bad, so drop the packet for now. 7576 */ 7577 if (htons(h->ip6_plen) == 0) { 7578 action = PF_DROP; 7579 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 7580 goto done; 7581 } 7582 7583 pd.src = (struct pf_addr *)&h->ip6_src; 7584 pd.dst = (struct pf_addr *)&h->ip6_dst; 7585 pd.sport = pd.dport = NULL; 7586 pd.ip_sum = NULL; 7587 pd.proto_sum = NULL; 7588 pd.dir = dir; 7589 pd.sidx = (dir == PF_IN) ? 0 : 1; 7590 pd.didx = (dir == PF_IN) ? 1 : 0; 7591 pd.af = AF_INET6; 7592 pd.tos = IPV6_DSCP(h); 7593 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 7594 7595 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 7596 pd.proto = h->ip6_nxt; 7597 do { 7598 switch (pd.proto) { 7599 case IPPROTO_FRAGMENT: 7600 action = pf_test_fragment(&r, dir, kif, m, h, 7601 &pd, &a, &ruleset); 7602 if (action == PF_DROP) 7603 REASON_SET(&reason, PFRES_FRAG); 7604 goto done; 7605 case IPPROTO_ROUTING: { 7606 struct ip6_rthdr rthdr; 7607 7608 if (rh_cnt++) { 7609 DPFPRINTF(PF_DEBUG_MISC, 7610 ("pf: IPv6 more than one rthdr\n")); 7611 action = PF_DROP; 7612 REASON_SET(&reason, PFRES_IPOPTIONS); 7613 log = 1; 7614 goto done; 7615 } 7616 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 7617 &reason, pd.af)) { 7618 DPFPRINTF(PF_DEBUG_MISC, 7619 ("pf: IPv6 short rthdr\n")); 7620 action = PF_DROP; 7621 REASON_SET(&reason, PFRES_SHORT); 7622 log = 1; 7623 goto done; 7624 } 7625 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 7626 DPFPRINTF(PF_DEBUG_MISC, 7627 ("pf: IPv6 rthdr0\n")); 7628 action = PF_DROP; 7629 REASON_SET(&reason, PFRES_IPOPTIONS); 7630 log = 1; 7631 goto done; 7632 } 7633 /* FALLTHROUGH */ 7634 } 7635 case IPPROTO_AH: 7636 case IPPROTO_HOPOPTS: 7637 case IPPROTO_DSTOPTS: { 7638 /* get next header and header length */ 7639 struct ip6_ext opt6; 7640 7641 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 7642 NULL, &reason, pd.af)) { 7643 DPFPRINTF(PF_DEBUG_MISC, 7644 ("pf: IPv6 short opt\n")); 7645 action = PF_DROP; 7646 log = 1; 7647 goto done; 7648 } 7649 if (pd.proto == IPPROTO_AH) 7650 off += (opt6.ip6e_len + 2) * 4; 7651 else 7652 off += (opt6.ip6e_len + 1) * 8; 7653 pd.proto = opt6.ip6e_nxt; 7654 /* goto the next header */ 7655 break; 7656 } 7657 default: 7658 terminal++; 7659 break; 7660 } 7661 } while (!terminal); 7662 7663 /* if there's no routing header, use unmodified mbuf for checksumming */ 7664 if (!n) 7665 n = m; 7666 7667 switch (pd.proto) { 7668 case IPPROTO_TCP: { 7669 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 7670 &action, &reason, AF_INET6)) { 7671 log = action != PF_PASS; 7672 goto done; 7673 } 7674 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 7675 pd.sport = &pd.hdr.tcp.th_sport; 7676 pd.dport = &pd.hdr.tcp.th_dport; 7677 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 7678 if (action == PF_DROP) 7679 goto done; 7680 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 7681 &reason); 7682 if (action == PF_PASS) { 7683 if (V_pfsync_update_state_ptr != NULL) 7684 V_pfsync_update_state_ptr(s); 7685 r = s->rule.ptr; 7686 a = s->anchor.ptr; 7687 log = s->log; 7688 } else if (s == NULL) 7689 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7690 &a, &ruleset, inp); 7691 break; 7692 } 7693 7694 case IPPROTO_UDP: { 7695 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 7696 &action, &reason, AF_INET6)) { 7697 log = action != PF_PASS; 7698 goto done; 7699 } 7700 pd.sport = &pd.hdr.udp.uh_sport; 7701 pd.dport = &pd.hdr.udp.uh_dport; 7702 if (pd.hdr.udp.uh_dport == 0 || 7703 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 7704 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 7705 action = PF_DROP; 7706 REASON_SET(&reason, PFRES_SHORT); 7707 goto done; 7708 } 7709 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 7710 if (action == PF_PASS) { 7711 if (V_pfsync_update_state_ptr != NULL) 7712 V_pfsync_update_state_ptr(s); 7713 r = s->rule.ptr; 7714 a = s->anchor.ptr; 7715 log = s->log; 7716 } else if (s == NULL) 7717 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7718 &a, &ruleset, inp); 7719 break; 7720 } 7721 7722 case IPPROTO_ICMP: { 7723 action = PF_DROP; 7724 DPFPRINTF(PF_DEBUG_MISC, 7725 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 7726 goto done; 7727 } 7728 7729 case IPPROTO_ICMPV6: { 7730 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 7731 &action, &reason, AF_INET6)) { 7732 log = action != PF_PASS; 7733 goto done; 7734 } 7735 action = pf_test_state_icmp(&s, dir, kif, 7736 m, off, h, &pd, &reason); 7737 if (action == PF_PASS) { 7738 if (V_pfsync_update_state_ptr != NULL) 7739 V_pfsync_update_state_ptr(s); 7740 r = s->rule.ptr; 7741 a = s->anchor.ptr; 7742 log = s->log; 7743 } else if (s == NULL) 7744 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7745 &a, &ruleset, inp); 7746 break; 7747 } 7748 7749 default: 7750 action = pf_test_state_other(&s, dir, kif, m, &pd); 7751 if (action == PF_PASS) { 7752 if (V_pfsync_update_state_ptr != NULL) 7753 V_pfsync_update_state_ptr(s); 7754 r = s->rule.ptr; 7755 a = s->anchor.ptr; 7756 log = s->log; 7757 } else if (s == NULL) 7758 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7759 &a, &ruleset, inp); 7760 break; 7761 } 7762 7763 done: 7764 PF_RULES_RUNLOCK(); 7765 if (n != m) { 7766 m_freem(n); 7767 n = NULL; 7768 } 7769 7770 /* handle dangerous IPv6 extension headers. */ 7771 if (action == PF_PASS && rh_cnt && 7772 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7773 action = PF_DROP; 7774 REASON_SET(&reason, PFRES_IPOPTIONS); 7775 log = r->log; 7776 DPFPRINTF(PF_DEBUG_MISC, 7777 ("pf: dropping packet with dangerous v6 headers\n")); 7778 } 7779 7780 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 7781 action = PF_DROP; 7782 REASON_SET(&reason, PFRES_MEMORY); 7783 } 7784 if (r->rtableid >= 0) 7785 M_SETFIB(m, r->rtableid); 7786 7787 if (r->scrub_flags & PFSTATE_SETPRIO) { 7788 if (pd.tos & IPTOS_LOWDELAY) 7789 pqid = 1; 7790 if (vlan_set_pcp(m, r->set_prio[pqid])) { 7791 action = PF_DROP; 7792 REASON_SET(&reason, PFRES_MEMORY); 7793 log = 1; 7794 DPFPRINTF(PF_DEBUG_MISC, 7795 ("pf: failed to allocate 802.1q mtag\n")); 7796 } 7797 } 7798 7799 #ifdef ALTQ 7800 if (s && s->qid) { 7801 pd.act.pqid = s->pqid; 7802 pd.act.qid = s->qid; 7803 } else if (r->qid) { 7804 pd.act.pqid = r->pqid; 7805 pd.act.qid = r->qid; 7806 } 7807 if (action == PF_PASS && pd.act.qid) { 7808 if (pd.pf_mtag == NULL && 7809 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7810 action = PF_DROP; 7811 REASON_SET(&reason, PFRES_MEMORY); 7812 } else { 7813 if (s != NULL) 7814 pd.pf_mtag->qid_hash = pf_state_hash(s); 7815 if (pd.tos & IPTOS_LOWDELAY) 7816 pd.pf_mtag->qid = pd.act.pqid; 7817 else 7818 pd.pf_mtag->qid = pd.act.qid; 7819 /* Add hints for ecn. */ 7820 pd.pf_mtag->hdr = h; 7821 } 7822 } 7823 #endif /* ALTQ */ 7824 7825 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7826 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7827 (s->nat_rule.ptr->action == PF_RDR || 7828 s->nat_rule.ptr->action == PF_BINAT) && 7829 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 7830 m->m_flags |= M_SKIP_FIREWALL; 7831 7832 /* XXX: Anybody working on it?! */ 7833 if (r->divert.port) 7834 printf("pf: divert(9) is not supported for IPv6\n"); 7835 7836 if (log) { 7837 struct pf_krule *lr; 7838 7839 if (s != NULL && s->nat_rule.ptr != NULL && 7840 s->nat_rule.ptr->log & PF_LOG_ALL) 7841 lr = s->nat_rule.ptr; 7842 else 7843 lr = r; 7844 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 7845 &pd, (s == NULL)); 7846 } 7847 7848 pf_counter_u64_critical_enter(); 7849 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 7850 pd.tot_len); 7851 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 7852 1); 7853 7854 if (action == PF_PASS || r->action == PF_DROP) { 7855 dirndx = (dir == PF_OUT); 7856 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 7857 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 7858 if (a != NULL) { 7859 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 7860 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 7861 } 7862 if (s != NULL) { 7863 if (s->nat_rule.ptr != NULL) { 7864 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 7865 1); 7866 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 7867 pd.tot_len); 7868 } 7869 if (s->src_node != NULL) { 7870 counter_u64_add(s->src_node->packets[dirndx], 7871 1); 7872 counter_u64_add(s->src_node->bytes[dirndx], 7873 pd.tot_len); 7874 } 7875 if (s->nat_src_node != NULL) { 7876 counter_u64_add(s->nat_src_node->packets[dirndx], 7877 1); 7878 counter_u64_add(s->nat_src_node->bytes[dirndx], 7879 pd.tot_len); 7880 } 7881 dirndx = (dir == s->direction) ? 0 : 1; 7882 s->packets[dirndx]++; 7883 s->bytes[dirndx] += pd.tot_len; 7884 } 7885 tr = r; 7886 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7887 if (nr != NULL && r == &V_pf_default_rule) 7888 tr = nr; 7889 if (tr->src.addr.type == PF_ADDR_TABLE) 7890 pfr_update_stats(tr->src.addr.p.tbl, 7891 (s == NULL) ? pd.src : 7892 &s->key[(s->direction == PF_IN)]->addr[0], 7893 pd.af, pd.tot_len, dir == PF_OUT, 7894 r->action == PF_PASS, tr->src.neg); 7895 if (tr->dst.addr.type == PF_ADDR_TABLE) 7896 pfr_update_stats(tr->dst.addr.p.tbl, 7897 (s == NULL) ? pd.dst : 7898 &s->key[(s->direction == PF_IN)]->addr[1], 7899 pd.af, pd.tot_len, dir == PF_OUT, 7900 r->action == PF_PASS, tr->dst.neg); 7901 } 7902 pf_counter_u64_critical_exit(); 7903 7904 switch (action) { 7905 case PF_SYNPROXY_DROP: 7906 m_freem(*m0); 7907 case PF_DEFER: 7908 *m0 = NULL; 7909 action = PF_PASS; 7910 break; 7911 case PF_DROP: 7912 m_freem(*m0); 7913 *m0 = NULL; 7914 break; 7915 default: 7916 /* pf_route6() returns unlocked. */ 7917 if (r->rt) { 7918 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 7919 return (action); 7920 } 7921 if (pf_dummynet(&pd, dir, s, r, m0) != 0) { 7922 action = PF_DROP; 7923 REASON_SET(&reason, PFRES_MEMORY); 7924 } 7925 break; 7926 } 7927 7928 if (s) 7929 PF_STATE_UNLOCK(s); 7930 7931 /* If reassembled packet passed, create new fragments. */ 7932 if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) && 7933 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 7934 action = pf_refragment6(ifp, m0, mtag); 7935 7936 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 7937 7938 return (action); 7939 } 7940 #endif /* INET6 */ 7941