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