1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 130 "struct pf_krule *", "struct mbuf *", "int"); 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 *, int); 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 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 278 struct pf_krule *, struct mbuf **); 279 static int pf_dummynet_route(struct pf_pdesc *, 280 struct pf_kstate *, struct pf_krule *, 281 struct ifnet *, struct sockaddr *, struct mbuf **); 282 static int pf_test_eth_rule(int, struct pfi_kkif *, 283 struct mbuf **); 284 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 285 struct pfi_kkif *, struct mbuf *, int, 286 struct pf_pdesc *, struct pf_krule **, 287 struct pf_kruleset **, struct inpcb *); 288 static int pf_create_state(struct pf_krule *, struct pf_krule *, 289 struct pf_krule *, struct pf_pdesc *, 290 struct pf_ksrc_node *, struct pf_state_key *, 291 struct pf_state_key *, struct mbuf *, int, 292 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 293 struct pf_kstate **, int, u_int16_t, u_int16_t, 294 int, struct pf_krule_slist *); 295 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 296 struct mbuf *, void *, struct pf_pdesc *, 297 struct pf_krule **, struct pf_kruleset **); 298 static int pf_tcp_track_full(struct pf_kstate **, 299 struct pfi_kkif *, struct mbuf *, int, 300 struct pf_pdesc *, u_short *, int *); 301 static int pf_tcp_track_sloppy(struct pf_kstate **, 302 struct pf_pdesc *, u_short *); 303 static int pf_test_state_tcp(struct pf_kstate **, 304 struct pfi_kkif *, struct mbuf *, int, 305 void *, struct pf_pdesc *, u_short *); 306 static int pf_test_state_udp(struct pf_kstate **, 307 struct pfi_kkif *, struct mbuf *, int, 308 void *, struct pf_pdesc *); 309 static int pf_test_state_icmp(struct pf_kstate **, 310 struct pfi_kkif *, struct mbuf *, int, 311 void *, struct pf_pdesc *, u_short *); 312 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 313 struct pfi_kkif *, struct pf_kstate *); 314 static int pf_test_state_sctp(struct pf_kstate **, 315 struct pfi_kkif *, struct mbuf *, int, 316 void *, struct pf_pdesc *, u_short *); 317 static int pf_test_state_other(struct pf_kstate **, 318 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 319 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 320 int, u_int16_t); 321 static int pf_check_proto_cksum(struct mbuf *, int, int, 322 u_int8_t, sa_family_t); 323 static void pf_print_state_parts(struct pf_kstate *, 324 struct pf_state_key *, struct pf_state_key *); 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 u_short 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 *, 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 *, 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) \ 361 do { \ 362 struct pf_state_key *nk; \ 363 if ((pd->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_MTAG_FLAG_PACKET_LOOPED) 372 373 #define STATE_LOOKUP(i, k, s, pd) \ 374 do { \ 375 (s) = pf_find_state((i), (k), (pd->dir)); \ 376 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), 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 struct pf_krule_item *mrm; \ 389 counter_u64_add(s->rule.ptr->states_cur, 1); \ 390 counter_u64_add(s->rule.ptr->states_tot, 1); \ 391 if (s->anchor.ptr != NULL) { \ 392 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 393 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 394 } \ 395 if (s->nat_rule.ptr != NULL) { \ 396 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 397 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 398 } \ 399 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 400 counter_u64_add(mrm->r->states_cur, 1); \ 401 counter_u64_add(mrm->r->states_tot, 1); \ 402 } \ 403 } while (0) 404 405 #define STATE_DEC_COUNTERS(s) \ 406 do { \ 407 struct pf_krule_item *mrm; \ 408 if (s->nat_rule.ptr != NULL) \ 409 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 410 if (s->anchor.ptr != NULL) \ 411 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 412 counter_u64_add(s->rule.ptr->states_cur, -1); \ 413 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 414 counter_u64_add(mrm->r->states_cur, -1); \ 415 } while (0) 416 417 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 418 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 419 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 420 VNET_DEFINE(struct pf_idhash *, pf_idhash); 421 VNET_DEFINE(struct pf_srchash *, pf_srchash); 422 423 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 424 "pf(4)"); 425 426 u_long pf_hashmask; 427 u_long pf_srchashmask; 428 static u_long pf_hashsize; 429 static u_long pf_srchashsize; 430 u_long pf_ioctl_maxcount = 65535; 431 432 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 433 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 434 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 435 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 436 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 437 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 438 439 VNET_DEFINE(void *, pf_swi_cookie); 440 VNET_DEFINE(struct intr_event *, pf_swi_ie); 441 442 VNET_DEFINE(uint32_t, pf_hashseed); 443 #define V_pf_hashseed VNET(pf_hashseed) 444 445 static void 446 pf_sctp_checksum(struct mbuf *m, int off) 447 { 448 uint32_t sum = 0; 449 450 /* Zero out the checksum, to enable recalculation. */ 451 m_copyback(m, off + offsetof(struct sctphdr, checksum), 452 sizeof(sum), (caddr_t)&sum); 453 454 sum = sctp_calculate_cksum(m, off); 455 456 m_copyback(m, off + offsetof(struct sctphdr, checksum), 457 sizeof(sum), (caddr_t)&sum); 458 } 459 460 int 461 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 462 { 463 464 switch (af) { 465 #ifdef INET 466 case AF_INET: 467 if (a->addr32[0] > b->addr32[0]) 468 return (1); 469 if (a->addr32[0] < b->addr32[0]) 470 return (-1); 471 break; 472 #endif /* INET */ 473 #ifdef INET6 474 case AF_INET6: 475 if (a->addr32[3] > b->addr32[3]) 476 return (1); 477 if (a->addr32[3] < b->addr32[3]) 478 return (-1); 479 if (a->addr32[2] > b->addr32[2]) 480 return (1); 481 if (a->addr32[2] < b->addr32[2]) 482 return (-1); 483 if (a->addr32[1] > b->addr32[1]) 484 return (1); 485 if (a->addr32[1] < b->addr32[1]) 486 return (-1); 487 if (a->addr32[0] > b->addr32[0]) 488 return (1); 489 if (a->addr32[0] < b->addr32[0]) 490 return (-1); 491 break; 492 #endif /* INET6 */ 493 default: 494 panic("%s: unknown address family %u", __func__, af); 495 } 496 return (0); 497 } 498 499 static void 500 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 501 struct pf_state_key *nk) 502 { 503 504 switch (pd->proto) { 505 case IPPROTO_TCP: { 506 struct tcphdr *th = &pd->hdr.tcp; 507 508 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 509 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 510 &th->th_sum, &nk->addr[pd->sidx], 511 nk->port[pd->sidx], 0, pd->af); 512 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 513 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 514 &th->th_sum, &nk->addr[pd->didx], 515 nk->port[pd->didx], 0, pd->af); 516 m_copyback(m, off, sizeof(*th), (caddr_t)th); 517 break; 518 } 519 case IPPROTO_UDP: { 520 struct udphdr *uh = &pd->hdr.udp; 521 522 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 523 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 524 &uh->uh_sum, &nk->addr[pd->sidx], 525 nk->port[pd->sidx], 1, pd->af); 526 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 527 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 528 &uh->uh_sum, &nk->addr[pd->didx], 529 nk->port[pd->didx], 1, pd->af); 530 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 531 break; 532 } 533 case IPPROTO_SCTP: { 534 struct sctphdr *sh = &pd->hdr.sctp; 535 uint16_t checksum = 0; 536 537 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 538 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 539 &checksum, &nk->addr[pd->sidx], 540 nk->port[pd->sidx], 1, pd->af); 541 } 542 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 543 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 544 &checksum, &nk->addr[pd->didx], 545 nk->port[pd->didx], 1, pd->af); 546 } 547 548 break; 549 } 550 case IPPROTO_ICMP: { 551 struct icmp *ih = &pd->hdr.icmp; 552 553 if (nk->port[pd->sidx] != ih->icmp_id) { 554 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 555 ih->icmp_cksum, ih->icmp_id, 556 nk->port[pd->sidx], 0); 557 ih->icmp_id = nk->port[pd->sidx]; 558 pd->sport = &ih->icmp_id; 559 560 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 561 } 562 /* FALLTHROUGH */ 563 } 564 default: 565 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 566 switch (pd->af) { 567 case AF_INET: 568 pf_change_a(&pd->src->v4.s_addr, 569 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 570 0); 571 break; 572 case AF_INET6: 573 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 574 break; 575 } 576 } 577 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 578 switch (pd->af) { 579 case AF_INET: 580 pf_change_a(&pd->dst->v4.s_addr, 581 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 582 0); 583 break; 584 case AF_INET6: 585 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 586 break; 587 } 588 } 589 break; 590 } 591 } 592 593 static __inline uint32_t 594 pf_hashkey(struct pf_state_key *sk) 595 { 596 uint32_t h; 597 598 h = murmur3_32_hash32((uint32_t *)sk, 599 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 600 V_pf_hashseed); 601 602 return (h & pf_hashmask); 603 } 604 605 static __inline uint32_t 606 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 607 { 608 uint32_t h; 609 610 switch (af) { 611 case AF_INET: 612 h = murmur3_32_hash32((uint32_t *)&addr->v4, 613 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 614 break; 615 case AF_INET6: 616 h = murmur3_32_hash32((uint32_t *)&addr->v6, 617 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 618 break; 619 default: 620 panic("%s: unknown address family %u", __func__, af); 621 } 622 623 return (h & pf_srchashmask); 624 } 625 626 #ifdef ALTQ 627 static int 628 pf_state_hash(struct pf_kstate *s) 629 { 630 u_int32_t hv = (intptr_t)s / sizeof(*s); 631 632 hv ^= crc32(&s->src, sizeof(s->src)); 633 hv ^= crc32(&s->dst, sizeof(s->dst)); 634 if (hv == 0) 635 hv = 1; 636 return (hv); 637 } 638 #endif 639 640 static __inline void 641 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 642 { 643 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 644 s->dst.state = newstate; 645 if (which == PF_PEER_DST) 646 return; 647 if (s->src.state == newstate) 648 return; 649 if (s->creatorid == V_pf_status.hostid && 650 s->key[PF_SK_STACK] != NULL && 651 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 652 !(TCPS_HAVEESTABLISHED(s->src.state) || 653 s->src.state == TCPS_CLOSED) && 654 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 655 atomic_add_32(&V_pf_status.states_halfopen, -1); 656 657 s->src.state = newstate; 658 } 659 660 #ifdef INET6 661 void 662 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 663 { 664 switch (af) { 665 #ifdef INET 666 case AF_INET: 667 dst->addr32[0] = src->addr32[0]; 668 break; 669 #endif /* INET */ 670 case AF_INET6: 671 dst->addr32[0] = src->addr32[0]; 672 dst->addr32[1] = src->addr32[1]; 673 dst->addr32[2] = src->addr32[2]; 674 dst->addr32[3] = src->addr32[3]; 675 break; 676 } 677 } 678 #endif /* INET6 */ 679 680 static void 681 pf_init_threshold(struct pf_threshold *threshold, 682 u_int32_t limit, u_int32_t seconds) 683 { 684 threshold->limit = limit * PF_THRESHOLD_MULT; 685 threshold->seconds = seconds; 686 threshold->count = 0; 687 threshold->last = time_uptime; 688 } 689 690 static void 691 pf_add_threshold(struct pf_threshold *threshold) 692 { 693 u_int32_t t = time_uptime, diff = t - threshold->last; 694 695 if (diff >= threshold->seconds) 696 threshold->count = 0; 697 else 698 threshold->count -= threshold->count * diff / 699 threshold->seconds; 700 threshold->count += PF_THRESHOLD_MULT; 701 threshold->last = t; 702 } 703 704 static int 705 pf_check_threshold(struct pf_threshold *threshold) 706 { 707 return (threshold->count > threshold->limit); 708 } 709 710 static int 711 pf_src_connlimit(struct pf_kstate **state) 712 { 713 struct pf_overload_entry *pfoe; 714 int bad = 0; 715 716 PF_STATE_LOCK_ASSERT(*state); 717 /* 718 * XXXKS: The src node is accessed unlocked! 719 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 720 */ 721 722 (*state)->src_node->conn++; 723 (*state)->src.tcp_est = 1; 724 pf_add_threshold(&(*state)->src_node->conn_rate); 725 726 if ((*state)->rule.ptr->max_src_conn && 727 (*state)->rule.ptr->max_src_conn < 728 (*state)->src_node->conn) { 729 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 730 bad++; 731 } 732 733 if ((*state)->rule.ptr->max_src_conn_rate.limit && 734 pf_check_threshold(&(*state)->src_node->conn_rate)) { 735 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 736 bad++; 737 } 738 739 if (!bad) 740 return (0); 741 742 /* Kill this state. */ 743 (*state)->timeout = PFTM_PURGE; 744 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 745 746 if ((*state)->rule.ptr->overload_tbl == NULL) 747 return (1); 748 749 /* Schedule overloading and flushing task. */ 750 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 751 if (pfoe == NULL) 752 return (1); /* too bad :( */ 753 754 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 755 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 756 pfoe->rule = (*state)->rule.ptr; 757 pfoe->dir = (*state)->direction; 758 PF_OVERLOADQ_LOCK(); 759 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 760 PF_OVERLOADQ_UNLOCK(); 761 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 762 763 return (1); 764 } 765 766 static void 767 pf_overload_task(void *v, int pending) 768 { 769 struct pf_overload_head queue; 770 struct pfr_addr p; 771 struct pf_overload_entry *pfoe, *pfoe1; 772 uint32_t killed = 0; 773 774 CURVNET_SET((struct vnet *)v); 775 776 PF_OVERLOADQ_LOCK(); 777 queue = V_pf_overloadqueue; 778 SLIST_INIT(&V_pf_overloadqueue); 779 PF_OVERLOADQ_UNLOCK(); 780 781 bzero(&p, sizeof(p)); 782 SLIST_FOREACH(pfoe, &queue, next) { 783 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 784 if (V_pf_status.debug >= PF_DEBUG_MISC) { 785 printf("%s: blocking address ", __func__); 786 pf_print_host(&pfoe->addr, 0, pfoe->af); 787 printf("\n"); 788 } 789 790 p.pfra_af = pfoe->af; 791 switch (pfoe->af) { 792 #ifdef INET 793 case AF_INET: 794 p.pfra_net = 32; 795 p.pfra_ip4addr = pfoe->addr.v4; 796 break; 797 #endif 798 #ifdef INET6 799 case AF_INET6: 800 p.pfra_net = 128; 801 p.pfra_ip6addr = pfoe->addr.v6; 802 break; 803 #endif 804 } 805 806 PF_RULES_WLOCK(); 807 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 808 PF_RULES_WUNLOCK(); 809 } 810 811 /* 812 * Remove those entries, that don't need flushing. 813 */ 814 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 815 if (pfoe->rule->flush == 0) { 816 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 817 free(pfoe, M_PFTEMP); 818 } else 819 counter_u64_add( 820 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 821 822 /* If nothing to flush, return. */ 823 if (SLIST_EMPTY(&queue)) { 824 CURVNET_RESTORE(); 825 return; 826 } 827 828 for (int i = 0; i <= pf_hashmask; i++) { 829 struct pf_idhash *ih = &V_pf_idhash[i]; 830 struct pf_state_key *sk; 831 struct pf_kstate *s; 832 833 PF_HASHROW_LOCK(ih); 834 LIST_FOREACH(s, &ih->states, entry) { 835 sk = s->key[PF_SK_WIRE]; 836 SLIST_FOREACH(pfoe, &queue, next) 837 if (sk->af == pfoe->af && 838 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 839 pfoe->rule == s->rule.ptr) && 840 ((pfoe->dir == PF_OUT && 841 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 842 (pfoe->dir == PF_IN && 843 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 844 s->timeout = PFTM_PURGE; 845 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 846 killed++; 847 } 848 } 849 PF_HASHROW_UNLOCK(ih); 850 } 851 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 852 free(pfoe, M_PFTEMP); 853 if (V_pf_status.debug >= PF_DEBUG_MISC) 854 printf("%s: %u states killed", __func__, killed); 855 856 CURVNET_RESTORE(); 857 } 858 859 /* 860 * Can return locked on failure, so that we can consistently 861 * allocate and insert a new one. 862 */ 863 struct pf_ksrc_node * 864 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 865 struct pf_srchash **sh, bool returnlocked) 866 { 867 struct pf_ksrc_node *n; 868 869 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 870 871 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 872 PF_HASHROW_LOCK(*sh); 873 LIST_FOREACH(n, &(*sh)->nodes, entry) 874 if (n->rule.ptr == rule && n->af == af && 875 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 876 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 877 break; 878 879 if (n != NULL) { 880 n->states++; 881 PF_HASHROW_UNLOCK(*sh); 882 } else if (returnlocked == false) 883 PF_HASHROW_UNLOCK(*sh); 884 885 return (n); 886 } 887 888 static void 889 pf_free_src_node(struct pf_ksrc_node *sn) 890 { 891 892 for (int i = 0; i < 2; i++) { 893 counter_u64_free(sn->bytes[i]); 894 counter_u64_free(sn->packets[i]); 895 } 896 uma_zfree(V_pf_sources_z, sn); 897 } 898 899 static u_short 900 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 901 struct pf_addr *src, sa_family_t af) 902 { 903 u_short reason = 0; 904 struct pf_srchash *sh = NULL; 905 906 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 907 rule->rpool.opts & PF_POOL_STICKYADDR), 908 ("%s for non-tracking rule %p", __func__, rule)); 909 910 if (*sn == NULL) 911 *sn = pf_find_src_node(src, rule, af, &sh, true); 912 913 if (*sn == NULL) { 914 PF_HASHROW_ASSERT(sh); 915 916 if (rule->max_src_nodes && 917 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 918 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 919 PF_HASHROW_UNLOCK(sh); 920 reason = PFRES_SRCLIMIT; 921 goto done; 922 } 923 924 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 925 if ((*sn) == NULL) { 926 PF_HASHROW_UNLOCK(sh); 927 reason = PFRES_MEMORY; 928 goto done; 929 } 930 931 for (int i = 0; i < 2; i++) { 932 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 933 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 934 935 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 936 pf_free_src_node(*sn); 937 PF_HASHROW_UNLOCK(sh); 938 reason = PFRES_MEMORY; 939 goto done; 940 } 941 } 942 943 pf_init_threshold(&(*sn)->conn_rate, 944 rule->max_src_conn_rate.limit, 945 rule->max_src_conn_rate.seconds); 946 947 MPASS((*sn)->lock == NULL); 948 (*sn)->lock = &sh->lock; 949 950 (*sn)->af = af; 951 (*sn)->rule.ptr = rule; 952 PF_ACPY(&(*sn)->addr, src, af); 953 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 954 (*sn)->creation = time_uptime; 955 (*sn)->ruletype = rule->action; 956 (*sn)->states = 1; 957 if ((*sn)->rule.ptr != NULL) 958 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 959 PF_HASHROW_UNLOCK(sh); 960 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 961 } else { 962 if (rule->max_src_states && 963 (*sn)->states >= rule->max_src_states) { 964 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 965 1); 966 reason = PFRES_SRCLIMIT; 967 goto done; 968 } 969 } 970 done: 971 return (reason); 972 } 973 974 void 975 pf_unlink_src_node(struct pf_ksrc_node *src) 976 { 977 PF_SRC_NODE_LOCK_ASSERT(src); 978 979 LIST_REMOVE(src, entry); 980 if (src->rule.ptr) 981 counter_u64_add(src->rule.ptr->src_nodes, -1); 982 } 983 984 u_int 985 pf_free_src_nodes(struct pf_ksrc_node_list *head) 986 { 987 struct pf_ksrc_node *sn, *tmp; 988 u_int count = 0; 989 990 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 991 pf_free_src_node(sn); 992 count++; 993 } 994 995 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 996 997 return (count); 998 } 999 1000 void 1001 pf_mtag_initialize(void) 1002 { 1003 1004 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1005 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1006 UMA_ALIGN_PTR, 0); 1007 } 1008 1009 /* Per-vnet data storage structures initialization. */ 1010 void 1011 pf_initialize(void) 1012 { 1013 struct pf_keyhash *kh; 1014 struct pf_idhash *ih; 1015 struct pf_srchash *sh; 1016 u_int i; 1017 1018 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 1019 pf_hashsize = PF_HASHSIZ; 1020 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 1021 pf_srchashsize = PF_SRCHASHSIZ; 1022 1023 V_pf_hashseed = arc4random(); 1024 1025 /* States and state keys storage. */ 1026 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1027 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1028 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1029 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1030 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1031 1032 V_pf_state_key_z = uma_zcreate("pf state keys", 1033 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1034 UMA_ALIGN_PTR, 0); 1035 1036 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 1037 M_PFHASH, M_NOWAIT | M_ZERO); 1038 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1039 M_PFHASH, M_NOWAIT | M_ZERO); 1040 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1041 printf("pf: Unable to allocate memory for " 1042 "state_hashsize %lu.\n", pf_hashsize); 1043 1044 free(V_pf_keyhash, M_PFHASH); 1045 free(V_pf_idhash, M_PFHASH); 1046 1047 pf_hashsize = PF_HASHSIZ; 1048 V_pf_keyhash = mallocarray(pf_hashsize, 1049 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1050 V_pf_idhash = mallocarray(pf_hashsize, 1051 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1052 } 1053 1054 pf_hashmask = pf_hashsize - 1; 1055 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1056 i++, kh++, ih++) { 1057 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1058 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1059 } 1060 1061 /* Source nodes. */ 1062 V_pf_sources_z = uma_zcreate("pf source nodes", 1063 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1064 0); 1065 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1066 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1067 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1068 1069 V_pf_srchash = mallocarray(pf_srchashsize, 1070 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1071 if (V_pf_srchash == NULL) { 1072 printf("pf: Unable to allocate memory for " 1073 "source_hashsize %lu.\n", pf_srchashsize); 1074 1075 pf_srchashsize = PF_SRCHASHSIZ; 1076 V_pf_srchash = mallocarray(pf_srchashsize, 1077 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1078 } 1079 1080 pf_srchashmask = pf_srchashsize - 1; 1081 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1082 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1083 1084 /* ALTQ */ 1085 TAILQ_INIT(&V_pf_altqs[0]); 1086 TAILQ_INIT(&V_pf_altqs[1]); 1087 TAILQ_INIT(&V_pf_altqs[2]); 1088 TAILQ_INIT(&V_pf_altqs[3]); 1089 TAILQ_INIT(&V_pf_pabuf); 1090 V_pf_altqs_active = &V_pf_altqs[0]; 1091 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1092 V_pf_altqs_inactive = &V_pf_altqs[2]; 1093 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1094 1095 /* Send & overload+flush queues. */ 1096 STAILQ_INIT(&V_pf_sendqueue); 1097 SLIST_INIT(&V_pf_overloadqueue); 1098 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1099 1100 /* Unlinked, but may be referenced rules. */ 1101 TAILQ_INIT(&V_pf_unlinked_rules); 1102 } 1103 1104 void 1105 pf_mtag_cleanup(void) 1106 { 1107 1108 uma_zdestroy(pf_mtag_z); 1109 } 1110 1111 void 1112 pf_cleanup(void) 1113 { 1114 struct pf_keyhash *kh; 1115 struct pf_idhash *ih; 1116 struct pf_srchash *sh; 1117 struct pf_send_entry *pfse, *next; 1118 u_int i; 1119 1120 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1121 i++, kh++, ih++) { 1122 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1123 __func__)); 1124 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1125 __func__)); 1126 mtx_destroy(&kh->lock); 1127 mtx_destroy(&ih->lock); 1128 } 1129 free(V_pf_keyhash, M_PFHASH); 1130 free(V_pf_idhash, M_PFHASH); 1131 1132 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1133 KASSERT(LIST_EMPTY(&sh->nodes), 1134 ("%s: source node hash not empty", __func__)); 1135 mtx_destroy(&sh->lock); 1136 } 1137 free(V_pf_srchash, M_PFHASH); 1138 1139 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1140 m_freem(pfse->pfse_m); 1141 free(pfse, M_PFTEMP); 1142 } 1143 1144 uma_zdestroy(V_pf_sources_z); 1145 uma_zdestroy(V_pf_state_z); 1146 uma_zdestroy(V_pf_state_key_z); 1147 } 1148 1149 static int 1150 pf_mtag_uminit(void *mem, int size, int how) 1151 { 1152 struct m_tag *t; 1153 1154 t = (struct m_tag *)mem; 1155 t->m_tag_cookie = MTAG_ABI_COMPAT; 1156 t->m_tag_id = PACKET_TAG_PF; 1157 t->m_tag_len = sizeof(struct pf_mtag); 1158 t->m_tag_free = pf_mtag_free; 1159 1160 return (0); 1161 } 1162 1163 static void 1164 pf_mtag_free(struct m_tag *t) 1165 { 1166 1167 uma_zfree(pf_mtag_z, t); 1168 } 1169 1170 struct pf_mtag * 1171 pf_get_mtag(struct mbuf *m) 1172 { 1173 struct m_tag *mtag; 1174 1175 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1176 return ((struct pf_mtag *)(mtag + 1)); 1177 1178 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1179 if (mtag == NULL) 1180 return (NULL); 1181 bzero(mtag + 1, sizeof(struct pf_mtag)); 1182 m_tag_prepend(m, mtag); 1183 1184 return ((struct pf_mtag *)(mtag + 1)); 1185 } 1186 1187 static int 1188 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1189 struct pf_kstate *s) 1190 { 1191 struct pf_keyhash *khs, *khw, *kh; 1192 struct pf_state_key *sk, *cur; 1193 struct pf_kstate *si, *olds = NULL; 1194 int idx; 1195 1196 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1197 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1198 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1199 1200 /* 1201 * We need to lock hash slots of both keys. To avoid deadlock 1202 * we always lock the slot with lower address first. Unlock order 1203 * isn't important. 1204 * 1205 * We also need to lock ID hash slot before dropping key 1206 * locks. On success we return with ID hash slot locked. 1207 */ 1208 1209 if (skw == sks) { 1210 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1211 PF_HASHROW_LOCK(khs); 1212 } else { 1213 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1214 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1215 if (khs == khw) { 1216 PF_HASHROW_LOCK(khs); 1217 } else if (khs < khw) { 1218 PF_HASHROW_LOCK(khs); 1219 PF_HASHROW_LOCK(khw); 1220 } else { 1221 PF_HASHROW_LOCK(khw); 1222 PF_HASHROW_LOCK(khs); 1223 } 1224 } 1225 1226 #define KEYS_UNLOCK() do { \ 1227 if (khs != khw) { \ 1228 PF_HASHROW_UNLOCK(khs); \ 1229 PF_HASHROW_UNLOCK(khw); \ 1230 } else \ 1231 PF_HASHROW_UNLOCK(khs); \ 1232 } while (0) 1233 1234 /* 1235 * First run: start with wire key. 1236 */ 1237 sk = skw; 1238 kh = khw; 1239 idx = PF_SK_WIRE; 1240 1241 MPASS(s->lock == NULL); 1242 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1243 1244 keyattach: 1245 LIST_FOREACH(cur, &kh->keys, entry) 1246 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1247 break; 1248 1249 if (cur != NULL) { 1250 /* Key exists. Check for same kif, if none, add to key. */ 1251 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1252 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1253 1254 PF_HASHROW_LOCK(ih); 1255 if (si->kif == s->kif && 1256 si->direction == s->direction) { 1257 if (sk->proto == IPPROTO_TCP && 1258 si->src.state >= TCPS_FIN_WAIT_2 && 1259 si->dst.state >= TCPS_FIN_WAIT_2) { 1260 /* 1261 * New state matches an old >FIN_WAIT_2 1262 * state. We can't drop key hash locks, 1263 * thus we can't unlink it properly. 1264 * 1265 * As a workaround we drop it into 1266 * TCPS_CLOSED state, schedule purge 1267 * ASAP and push it into the very end 1268 * of the slot TAILQ, so that it won't 1269 * conflict with our new state. 1270 */ 1271 pf_set_protostate(si, PF_PEER_BOTH, 1272 TCPS_CLOSED); 1273 si->timeout = PFTM_PURGE; 1274 olds = si; 1275 } else { 1276 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1277 printf("pf: %s key attach " 1278 "failed on %s: ", 1279 (idx == PF_SK_WIRE) ? 1280 "wire" : "stack", 1281 s->kif->pfik_name); 1282 pf_print_state_parts(s, 1283 (idx == PF_SK_WIRE) ? 1284 sk : NULL, 1285 (idx == PF_SK_STACK) ? 1286 sk : NULL); 1287 printf(", existing: "); 1288 pf_print_state_parts(si, 1289 (idx == PF_SK_WIRE) ? 1290 sk : NULL, 1291 (idx == PF_SK_STACK) ? 1292 sk : NULL); 1293 printf("\n"); 1294 } 1295 PF_HASHROW_UNLOCK(ih); 1296 KEYS_UNLOCK(); 1297 uma_zfree(V_pf_state_key_z, sk); 1298 if (idx == PF_SK_STACK) 1299 pf_detach_state(s); 1300 return (EEXIST); /* collision! */ 1301 } 1302 } 1303 PF_HASHROW_UNLOCK(ih); 1304 } 1305 uma_zfree(V_pf_state_key_z, sk); 1306 s->key[idx] = cur; 1307 } else { 1308 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1309 s->key[idx] = sk; 1310 } 1311 1312 stateattach: 1313 /* List is sorted, if-bound states before floating. */ 1314 if (s->kif == V_pfi_all) 1315 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1316 else 1317 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1318 1319 if (olds) { 1320 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1321 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1322 key_list[idx]); 1323 olds = NULL; 1324 } 1325 1326 /* 1327 * Attach done. See how should we (or should not?) 1328 * attach a second key. 1329 */ 1330 if (sks == skw) { 1331 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1332 idx = PF_SK_STACK; 1333 sks = NULL; 1334 goto stateattach; 1335 } else if (sks != NULL) { 1336 /* 1337 * Continue attaching with stack key. 1338 */ 1339 sk = sks; 1340 kh = khs; 1341 idx = PF_SK_STACK; 1342 sks = NULL; 1343 goto keyattach; 1344 } 1345 1346 PF_STATE_LOCK(s); 1347 KEYS_UNLOCK(); 1348 1349 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1350 ("%s failure", __func__)); 1351 1352 return (0); 1353 #undef KEYS_UNLOCK 1354 } 1355 1356 static void 1357 pf_detach_state(struct pf_kstate *s) 1358 { 1359 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1360 struct pf_keyhash *kh; 1361 1362 if (sks != NULL) { 1363 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1364 PF_HASHROW_LOCK(kh); 1365 if (s->key[PF_SK_STACK] != NULL) 1366 pf_state_key_detach(s, PF_SK_STACK); 1367 /* 1368 * If both point to same key, then we are done. 1369 */ 1370 if (sks == s->key[PF_SK_WIRE]) { 1371 pf_state_key_detach(s, PF_SK_WIRE); 1372 PF_HASHROW_UNLOCK(kh); 1373 return; 1374 } 1375 PF_HASHROW_UNLOCK(kh); 1376 } 1377 1378 if (s->key[PF_SK_WIRE] != NULL) { 1379 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1380 PF_HASHROW_LOCK(kh); 1381 if (s->key[PF_SK_WIRE] != NULL) 1382 pf_state_key_detach(s, PF_SK_WIRE); 1383 PF_HASHROW_UNLOCK(kh); 1384 } 1385 } 1386 1387 static void 1388 pf_state_key_detach(struct pf_kstate *s, int idx) 1389 { 1390 struct pf_state_key *sk = s->key[idx]; 1391 #ifdef INVARIANTS 1392 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1393 1394 PF_HASHROW_ASSERT(kh); 1395 #endif 1396 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1397 s->key[idx] = NULL; 1398 1399 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1400 LIST_REMOVE(sk, entry); 1401 uma_zfree(V_pf_state_key_z, sk); 1402 } 1403 } 1404 1405 static int 1406 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1407 { 1408 struct pf_state_key *sk = mem; 1409 1410 bzero(sk, sizeof(struct pf_state_key_cmp)); 1411 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1412 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1413 1414 return (0); 1415 } 1416 1417 struct pf_state_key * 1418 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1419 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1420 { 1421 struct pf_state_key *sk; 1422 1423 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1424 if (sk == NULL) 1425 return (NULL); 1426 1427 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1428 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1429 sk->port[pd->sidx] = sport; 1430 sk->port[pd->didx] = dport; 1431 sk->proto = pd->proto; 1432 sk->af = pd->af; 1433 1434 return (sk); 1435 } 1436 1437 struct pf_state_key * 1438 pf_state_key_clone(struct pf_state_key *orig) 1439 { 1440 struct pf_state_key *sk; 1441 1442 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1443 if (sk == NULL) 1444 return (NULL); 1445 1446 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1447 1448 return (sk); 1449 } 1450 1451 int 1452 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1453 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1454 { 1455 struct pf_idhash *ih; 1456 struct pf_kstate *cur; 1457 int error; 1458 1459 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1460 ("%s: sks not pristine", __func__)); 1461 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1462 ("%s: skw not pristine", __func__)); 1463 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1464 1465 s->kif = kif; 1466 s->orig_kif = orig_kif; 1467 1468 if (s->id == 0 && s->creatorid == 0) { 1469 s->id = alloc_unr64(&V_pf_stateid); 1470 s->id = htobe64(s->id); 1471 s->creatorid = V_pf_status.hostid; 1472 } 1473 1474 /* Returns with ID locked on success. */ 1475 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1476 return (error); 1477 1478 ih = &V_pf_idhash[PF_IDHASH(s)]; 1479 PF_HASHROW_ASSERT(ih); 1480 LIST_FOREACH(cur, &ih->states, entry) 1481 if (cur->id == s->id && cur->creatorid == s->creatorid) 1482 break; 1483 1484 if (cur != NULL) { 1485 PF_HASHROW_UNLOCK(ih); 1486 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1487 printf("pf: state ID collision: " 1488 "id: %016llx creatorid: %08x\n", 1489 (unsigned long long)be64toh(s->id), 1490 ntohl(s->creatorid)); 1491 } 1492 pf_detach_state(s); 1493 return (EEXIST); 1494 } 1495 LIST_INSERT_HEAD(&ih->states, s, entry); 1496 /* One for keys, one for ID hash. */ 1497 refcount_init(&s->refs, 2); 1498 1499 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1500 if (V_pfsync_insert_state_ptr != NULL) 1501 V_pfsync_insert_state_ptr(s); 1502 1503 /* Returns locked. */ 1504 return (0); 1505 } 1506 1507 /* 1508 * Find state by ID: returns with locked row on success. 1509 */ 1510 struct pf_kstate * 1511 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1512 { 1513 struct pf_idhash *ih; 1514 struct pf_kstate *s; 1515 1516 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1517 1518 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1519 1520 PF_HASHROW_LOCK(ih); 1521 LIST_FOREACH(s, &ih->states, entry) 1522 if (s->id == id && s->creatorid == creatorid) 1523 break; 1524 1525 if (s == NULL) 1526 PF_HASHROW_UNLOCK(ih); 1527 1528 return (s); 1529 } 1530 1531 /* 1532 * Find state by key. 1533 * Returns with ID hash slot locked on success. 1534 */ 1535 static struct pf_kstate * 1536 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1537 { 1538 struct pf_keyhash *kh; 1539 struct pf_state_key *sk; 1540 struct pf_kstate *s; 1541 int idx; 1542 1543 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1544 1545 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1546 1547 PF_HASHROW_LOCK(kh); 1548 LIST_FOREACH(sk, &kh->keys, entry) 1549 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1550 break; 1551 if (sk == NULL) { 1552 PF_HASHROW_UNLOCK(kh); 1553 return (NULL); 1554 } 1555 1556 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1557 1558 /* List is sorted, if-bound states before floating ones. */ 1559 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1560 if (s->kif == V_pfi_all || s->kif == kif) { 1561 PF_STATE_LOCK(s); 1562 PF_HASHROW_UNLOCK(kh); 1563 if (__predict_false(s->timeout >= PFTM_MAX)) { 1564 /* 1565 * State is either being processed by 1566 * pf_unlink_state() in an other thread, or 1567 * is scheduled for immediate expiry. 1568 */ 1569 PF_STATE_UNLOCK(s); 1570 return (NULL); 1571 } 1572 return (s); 1573 } 1574 PF_HASHROW_UNLOCK(kh); 1575 1576 return (NULL); 1577 } 1578 1579 /* 1580 * Returns with ID hash slot locked on success. 1581 */ 1582 struct pf_kstate * 1583 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1584 { 1585 struct pf_keyhash *kh; 1586 struct pf_state_key *sk; 1587 struct pf_kstate *s, *ret = NULL; 1588 int idx, inout = 0; 1589 1590 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1591 1592 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1593 1594 PF_HASHROW_LOCK(kh); 1595 LIST_FOREACH(sk, &kh->keys, entry) 1596 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1597 break; 1598 if (sk == NULL) { 1599 PF_HASHROW_UNLOCK(kh); 1600 return (NULL); 1601 } 1602 switch (dir) { 1603 case PF_IN: 1604 idx = PF_SK_WIRE; 1605 break; 1606 case PF_OUT: 1607 idx = PF_SK_STACK; 1608 break; 1609 case PF_INOUT: 1610 idx = PF_SK_WIRE; 1611 inout = 1; 1612 break; 1613 default: 1614 panic("%s: dir %u", __func__, dir); 1615 } 1616 second_run: 1617 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1618 if (more == NULL) { 1619 PF_STATE_LOCK(s); 1620 PF_HASHROW_UNLOCK(kh); 1621 return (s); 1622 } 1623 1624 if (ret) 1625 (*more)++; 1626 else { 1627 ret = s; 1628 PF_STATE_LOCK(s); 1629 } 1630 } 1631 if (inout == 1) { 1632 inout = 0; 1633 idx = PF_SK_STACK; 1634 goto second_run; 1635 } 1636 PF_HASHROW_UNLOCK(kh); 1637 1638 return (ret); 1639 } 1640 1641 /* 1642 * FIXME 1643 * This routine is inefficient -- locks the state only to unlock immediately on 1644 * return. 1645 * It is racy -- after the state is unlocked nothing stops other threads from 1646 * removing it. 1647 */ 1648 bool 1649 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1650 { 1651 struct pf_kstate *s; 1652 1653 s = pf_find_state_all(key, dir, NULL); 1654 if (s != NULL) { 1655 PF_STATE_UNLOCK(s); 1656 return (true); 1657 } 1658 return (false); 1659 } 1660 1661 /* END state table stuff */ 1662 1663 static void 1664 pf_send(struct pf_send_entry *pfse) 1665 { 1666 1667 PF_SENDQ_LOCK(); 1668 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1669 PF_SENDQ_UNLOCK(); 1670 swi_sched(V_pf_swi_cookie, 0); 1671 } 1672 1673 static bool 1674 pf_isforlocal(struct mbuf *m, int af) 1675 { 1676 switch (af) { 1677 #ifdef INET 1678 case AF_INET: { 1679 struct ip *ip = mtod(m, struct ip *); 1680 1681 return (in_localip(ip->ip_dst)); 1682 } 1683 #endif 1684 #ifdef INET6 1685 case AF_INET6: { 1686 struct ip6_hdr *ip6; 1687 struct in6_ifaddr *ia; 1688 ip6 = mtod(m, struct ip6_hdr *); 1689 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1690 if (ia == NULL) 1691 return (false); 1692 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1693 } 1694 #endif 1695 default: 1696 panic("Unsupported af %d", af); 1697 } 1698 1699 return (false); 1700 } 1701 1702 void 1703 pf_intr(void *v) 1704 { 1705 struct epoch_tracker et; 1706 struct pf_send_head queue; 1707 struct pf_send_entry *pfse, *next; 1708 1709 CURVNET_SET((struct vnet *)v); 1710 1711 PF_SENDQ_LOCK(); 1712 queue = V_pf_sendqueue; 1713 STAILQ_INIT(&V_pf_sendqueue); 1714 PF_SENDQ_UNLOCK(); 1715 1716 NET_EPOCH_ENTER(et); 1717 1718 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1719 switch (pfse->pfse_type) { 1720 #ifdef INET 1721 case PFSE_IP: { 1722 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1723 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1724 pfse->pfse_m->m_pkthdr.csum_flags |= 1725 CSUM_IP_VALID | CSUM_IP_CHECKED; 1726 ip_input(pfse->pfse_m); 1727 } else { 1728 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1729 NULL); 1730 } 1731 break; 1732 } 1733 case PFSE_ICMP: 1734 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1735 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1736 break; 1737 #endif /* INET */ 1738 #ifdef INET6 1739 case PFSE_IP6: 1740 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1741 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1742 ip6_input(pfse->pfse_m); 1743 } else { 1744 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1745 NULL, NULL); 1746 } 1747 break; 1748 case PFSE_ICMP6: 1749 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1750 pfse->icmpopts.code, pfse->icmpopts.mtu); 1751 break; 1752 #endif /* INET6 */ 1753 default: 1754 panic("%s: unknown type", __func__); 1755 } 1756 free(pfse, M_PFTEMP); 1757 } 1758 NET_EPOCH_EXIT(et); 1759 CURVNET_RESTORE(); 1760 } 1761 1762 #define pf_purge_thread_period (hz / 10) 1763 1764 #ifdef PF_WANT_32_TO_64_COUNTER 1765 static void 1766 pf_status_counter_u64_periodic(void) 1767 { 1768 1769 PF_RULES_RASSERT(); 1770 1771 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1772 return; 1773 } 1774 1775 for (int i = 0; i < FCNT_MAX; i++) { 1776 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1777 } 1778 } 1779 1780 static void 1781 pf_kif_counter_u64_periodic(void) 1782 { 1783 struct pfi_kkif *kif; 1784 size_t r, run; 1785 1786 PF_RULES_RASSERT(); 1787 1788 if (__predict_false(V_pf_allkifcount == 0)) { 1789 return; 1790 } 1791 1792 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1793 return; 1794 } 1795 1796 run = V_pf_allkifcount / 10; 1797 if (run < 5) 1798 run = 5; 1799 1800 for (r = 0; r < run; r++) { 1801 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1802 if (kif == NULL) { 1803 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1804 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1805 break; 1806 } 1807 1808 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1809 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1810 1811 for (int i = 0; i < 2; i++) { 1812 for (int j = 0; j < 2; j++) { 1813 for (int k = 0; k < 2; k++) { 1814 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1815 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1816 } 1817 } 1818 } 1819 } 1820 } 1821 1822 static void 1823 pf_rule_counter_u64_periodic(void) 1824 { 1825 struct pf_krule *rule; 1826 size_t r, run; 1827 1828 PF_RULES_RASSERT(); 1829 1830 if (__predict_false(V_pf_allrulecount == 0)) { 1831 return; 1832 } 1833 1834 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1835 return; 1836 } 1837 1838 run = V_pf_allrulecount / 10; 1839 if (run < 5) 1840 run = 5; 1841 1842 for (r = 0; r < run; r++) { 1843 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1844 if (rule == NULL) { 1845 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1846 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1847 break; 1848 } 1849 1850 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1851 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1852 1853 pf_counter_u64_periodic(&rule->evaluations); 1854 for (int i = 0; i < 2; i++) { 1855 pf_counter_u64_periodic(&rule->packets[i]); 1856 pf_counter_u64_periodic(&rule->bytes[i]); 1857 } 1858 } 1859 } 1860 1861 static void 1862 pf_counter_u64_periodic_main(void) 1863 { 1864 PF_RULES_RLOCK_TRACKER; 1865 1866 V_pf_counter_periodic_iter++; 1867 1868 PF_RULES_RLOCK(); 1869 pf_counter_u64_critical_enter(); 1870 pf_status_counter_u64_periodic(); 1871 pf_kif_counter_u64_periodic(); 1872 pf_rule_counter_u64_periodic(); 1873 pf_counter_u64_critical_exit(); 1874 PF_RULES_RUNLOCK(); 1875 } 1876 #else 1877 #define pf_counter_u64_periodic_main() do { } while (0) 1878 #endif 1879 1880 void 1881 pf_purge_thread(void *unused __unused) 1882 { 1883 VNET_ITERATOR_DECL(vnet_iter); 1884 1885 sx_xlock(&pf_end_lock); 1886 while (pf_end_threads == 0) { 1887 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1888 1889 VNET_LIST_RLOCK(); 1890 VNET_FOREACH(vnet_iter) { 1891 CURVNET_SET(vnet_iter); 1892 1893 /* Wait until V_pf_default_rule is initialized. */ 1894 if (V_pf_vnet_active == 0) { 1895 CURVNET_RESTORE(); 1896 continue; 1897 } 1898 1899 pf_counter_u64_periodic_main(); 1900 1901 /* 1902 * Process 1/interval fraction of the state 1903 * table every run. 1904 */ 1905 V_pf_purge_idx = 1906 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1907 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1908 1909 /* 1910 * Purge other expired types every 1911 * PFTM_INTERVAL seconds. 1912 */ 1913 if (V_pf_purge_idx == 0) { 1914 /* 1915 * Order is important: 1916 * - states and src nodes reference rules 1917 * - states and rules reference kifs 1918 */ 1919 pf_purge_expired_fragments(); 1920 pf_purge_expired_src_nodes(); 1921 pf_purge_unlinked_rules(); 1922 pfi_kkif_purge(); 1923 } 1924 CURVNET_RESTORE(); 1925 } 1926 VNET_LIST_RUNLOCK(); 1927 } 1928 1929 pf_end_threads++; 1930 sx_xunlock(&pf_end_lock); 1931 kproc_exit(0); 1932 } 1933 1934 void 1935 pf_unload_vnet_purge(void) 1936 { 1937 1938 /* 1939 * To cleanse up all kifs and rules we need 1940 * two runs: first one clears reference flags, 1941 * then pf_purge_expired_states() doesn't 1942 * raise them, and then second run frees. 1943 */ 1944 pf_purge_unlinked_rules(); 1945 pfi_kkif_purge(); 1946 1947 /* 1948 * Now purge everything. 1949 */ 1950 pf_purge_expired_states(0, pf_hashmask); 1951 pf_purge_fragments(UINT_MAX); 1952 pf_purge_expired_src_nodes(); 1953 1954 /* 1955 * Now all kifs & rules should be unreferenced, 1956 * thus should be successfully freed. 1957 */ 1958 pf_purge_unlinked_rules(); 1959 pfi_kkif_purge(); 1960 } 1961 1962 u_int32_t 1963 pf_state_expires(const struct pf_kstate *state) 1964 { 1965 u_int32_t timeout; 1966 u_int32_t start; 1967 u_int32_t end; 1968 u_int32_t states; 1969 1970 /* handle all PFTM_* > PFTM_MAX here */ 1971 if (state->timeout == PFTM_PURGE) 1972 return (time_uptime); 1973 KASSERT(state->timeout != PFTM_UNLINKED, 1974 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1975 KASSERT((state->timeout < PFTM_MAX), 1976 ("pf_state_expires: timeout > PFTM_MAX")); 1977 timeout = state->rule.ptr->timeout[state->timeout]; 1978 if (!timeout) 1979 timeout = V_pf_default_rule.timeout[state->timeout]; 1980 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1981 if (start && state->rule.ptr != &V_pf_default_rule) { 1982 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1983 states = counter_u64_fetch(state->rule.ptr->states_cur); 1984 } else { 1985 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1986 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1987 states = V_pf_status.states; 1988 } 1989 if (end && states > start && start < end) { 1990 if (states < end) { 1991 timeout = (u_int64_t)timeout * (end - states) / 1992 (end - start); 1993 return (state->expire + timeout); 1994 } 1995 else 1996 return (time_uptime); 1997 } 1998 return (state->expire + timeout); 1999 } 2000 2001 void 2002 pf_purge_expired_src_nodes(void) 2003 { 2004 struct pf_ksrc_node_list freelist; 2005 struct pf_srchash *sh; 2006 struct pf_ksrc_node *cur, *next; 2007 int i; 2008 2009 LIST_INIT(&freelist); 2010 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 2011 PF_HASHROW_LOCK(sh); 2012 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2013 if (cur->states == 0 && cur->expire <= time_uptime) { 2014 pf_unlink_src_node(cur); 2015 LIST_INSERT_HEAD(&freelist, cur, entry); 2016 } else if (cur->rule.ptr != NULL) 2017 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2018 PF_HASHROW_UNLOCK(sh); 2019 } 2020 2021 pf_free_src_nodes(&freelist); 2022 2023 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2024 } 2025 2026 static void 2027 pf_src_tree_remove_state(struct pf_kstate *s) 2028 { 2029 struct pf_ksrc_node *sn; 2030 uint32_t timeout; 2031 2032 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2033 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2034 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2035 2036 if (s->src_node != NULL) { 2037 sn = s->src_node; 2038 PF_SRC_NODE_LOCK(sn); 2039 if (s->src.tcp_est) 2040 --sn->conn; 2041 if (--sn->states == 0) 2042 sn->expire = time_uptime + timeout; 2043 PF_SRC_NODE_UNLOCK(sn); 2044 } 2045 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2046 sn = s->nat_src_node; 2047 PF_SRC_NODE_LOCK(sn); 2048 if (--sn->states == 0) 2049 sn->expire = time_uptime + timeout; 2050 PF_SRC_NODE_UNLOCK(sn); 2051 } 2052 s->src_node = s->nat_src_node = NULL; 2053 } 2054 2055 /* 2056 * Unlink and potentilly free a state. Function may be 2057 * called with ID hash row locked, but always returns 2058 * unlocked, since it needs to go through key hash locking. 2059 */ 2060 int 2061 pf_unlink_state(struct pf_kstate *s) 2062 { 2063 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2064 2065 PF_HASHROW_ASSERT(ih); 2066 2067 if (s->timeout == PFTM_UNLINKED) { 2068 /* 2069 * State is being processed 2070 * by pf_unlink_state() in 2071 * an other thread. 2072 */ 2073 PF_HASHROW_UNLOCK(ih); 2074 return (0); /* XXXGL: undefined actually */ 2075 } 2076 2077 if (s->src.state == PF_TCPS_PROXY_DST) { 2078 /* XXX wire key the right one? */ 2079 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2080 &s->key[PF_SK_WIRE]->addr[1], 2081 &s->key[PF_SK_WIRE]->addr[0], 2082 s->key[PF_SK_WIRE]->port[1], 2083 s->key[PF_SK_WIRE]->port[0], 2084 s->src.seqhi, s->src.seqlo + 1, 2085 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2086 } 2087 2088 LIST_REMOVE(s, entry); 2089 pf_src_tree_remove_state(s); 2090 2091 if (V_pfsync_delete_state_ptr != NULL) 2092 V_pfsync_delete_state_ptr(s); 2093 2094 STATE_DEC_COUNTERS(s); 2095 2096 s->timeout = PFTM_UNLINKED; 2097 2098 /* Ensure we remove it from the list of halfopen states, if needed. */ 2099 if (s->key[PF_SK_STACK] != NULL && 2100 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2101 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2102 2103 PF_HASHROW_UNLOCK(ih); 2104 2105 pf_detach_state(s); 2106 /* pf_state_insert() initialises refs to 2 */ 2107 return (pf_release_staten(s, 2)); 2108 } 2109 2110 struct pf_kstate * 2111 pf_alloc_state(int flags) 2112 { 2113 2114 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2115 } 2116 2117 void 2118 pf_free_state(struct pf_kstate *cur) 2119 { 2120 struct pf_krule_item *ri; 2121 2122 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2123 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2124 cur->timeout)); 2125 2126 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2127 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2128 free(ri, M_PF_RULE_ITEM); 2129 } 2130 2131 pf_normalize_tcp_cleanup(cur); 2132 uma_zfree(V_pf_state_z, cur); 2133 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2134 } 2135 2136 /* 2137 * Called only from pf_purge_thread(), thus serialized. 2138 */ 2139 static u_int 2140 pf_purge_expired_states(u_int i, int maxcheck) 2141 { 2142 struct pf_idhash *ih; 2143 struct pf_kstate *s; 2144 struct pf_krule_item *mrm; 2145 2146 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2147 2148 /* 2149 * Go through hash and unlink states that expire now. 2150 */ 2151 while (maxcheck > 0) { 2152 ih = &V_pf_idhash[i]; 2153 2154 /* only take the lock if we expect to do work */ 2155 if (!LIST_EMPTY(&ih->states)) { 2156 relock: 2157 PF_HASHROW_LOCK(ih); 2158 LIST_FOREACH(s, &ih->states, entry) { 2159 if (pf_state_expires(s) <= time_uptime) { 2160 V_pf_status.states -= 2161 pf_unlink_state(s); 2162 goto relock; 2163 } 2164 s->rule.ptr->rule_ref |= PFRULE_REFS; 2165 if (s->nat_rule.ptr != NULL) 2166 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2167 if (s->anchor.ptr != NULL) 2168 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2169 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2170 SLIST_FOREACH(mrm, &s->match_rules, entry) 2171 mrm->r->rule_ref |= PFRULE_REFS; 2172 if (s->rt_kif) 2173 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2174 } 2175 PF_HASHROW_UNLOCK(ih); 2176 } 2177 2178 /* Return when we hit end of hash. */ 2179 if (++i > pf_hashmask) { 2180 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2181 return (0); 2182 } 2183 2184 maxcheck--; 2185 } 2186 2187 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2188 2189 return (i); 2190 } 2191 2192 static void 2193 pf_purge_unlinked_rules(void) 2194 { 2195 struct pf_krulequeue tmpq; 2196 struct pf_krule *r, *r1; 2197 2198 /* 2199 * If we have overloading task pending, then we'd 2200 * better skip purging this time. There is a tiny 2201 * probability that overloading task references 2202 * an already unlinked rule. 2203 */ 2204 PF_OVERLOADQ_LOCK(); 2205 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2206 PF_OVERLOADQ_UNLOCK(); 2207 return; 2208 } 2209 PF_OVERLOADQ_UNLOCK(); 2210 2211 /* 2212 * Do naive mark-and-sweep garbage collecting of old rules. 2213 * Reference flag is raised by pf_purge_expired_states() 2214 * and pf_purge_expired_src_nodes(). 2215 * 2216 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2217 * use a temporary queue. 2218 */ 2219 TAILQ_INIT(&tmpq); 2220 PF_UNLNKDRULES_LOCK(); 2221 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2222 if (!(r->rule_ref & PFRULE_REFS)) { 2223 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2224 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2225 } else 2226 r->rule_ref &= ~PFRULE_REFS; 2227 } 2228 PF_UNLNKDRULES_UNLOCK(); 2229 2230 if (!TAILQ_EMPTY(&tmpq)) { 2231 PF_CONFIG_LOCK(); 2232 PF_RULES_WLOCK(); 2233 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2234 TAILQ_REMOVE(&tmpq, r, entries); 2235 pf_free_rule(r); 2236 } 2237 PF_RULES_WUNLOCK(); 2238 PF_CONFIG_UNLOCK(); 2239 } 2240 } 2241 2242 void 2243 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2244 { 2245 switch (af) { 2246 #ifdef INET 2247 case AF_INET: { 2248 u_int32_t a = ntohl(addr->addr32[0]); 2249 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2250 (a>>8)&255, a&255); 2251 if (p) { 2252 p = ntohs(p); 2253 printf(":%u", p); 2254 } 2255 break; 2256 } 2257 #endif /* INET */ 2258 #ifdef INET6 2259 case AF_INET6: { 2260 u_int16_t b; 2261 u_int8_t i, curstart, curend, maxstart, maxend; 2262 curstart = curend = maxstart = maxend = 255; 2263 for (i = 0; i < 8; i++) { 2264 if (!addr->addr16[i]) { 2265 if (curstart == 255) 2266 curstart = i; 2267 curend = i; 2268 } else { 2269 if ((curend - curstart) > 2270 (maxend - maxstart)) { 2271 maxstart = curstart; 2272 maxend = curend; 2273 } 2274 curstart = curend = 255; 2275 } 2276 } 2277 if ((curend - curstart) > 2278 (maxend - maxstart)) { 2279 maxstart = curstart; 2280 maxend = curend; 2281 } 2282 for (i = 0; i < 8; i++) { 2283 if (i >= maxstart && i <= maxend) { 2284 if (i == 0) 2285 printf(":"); 2286 if (i == maxend) 2287 printf(":"); 2288 } else { 2289 b = ntohs(addr->addr16[i]); 2290 printf("%x", b); 2291 if (i < 7) 2292 printf(":"); 2293 } 2294 } 2295 if (p) { 2296 p = ntohs(p); 2297 printf("[%u]", p); 2298 } 2299 break; 2300 } 2301 #endif /* INET6 */ 2302 } 2303 } 2304 2305 void 2306 pf_print_state(struct pf_kstate *s) 2307 { 2308 pf_print_state_parts(s, NULL, NULL); 2309 } 2310 2311 static void 2312 pf_print_state_parts(struct pf_kstate *s, 2313 struct pf_state_key *skwp, struct pf_state_key *sksp) 2314 { 2315 struct pf_state_key *skw, *sks; 2316 u_int8_t proto, dir; 2317 2318 /* Do our best to fill these, but they're skipped if NULL */ 2319 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2320 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2321 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2322 dir = s ? s->direction : 0; 2323 2324 switch (proto) { 2325 case IPPROTO_IPV4: 2326 printf("IPv4"); 2327 break; 2328 case IPPROTO_IPV6: 2329 printf("IPv6"); 2330 break; 2331 case IPPROTO_TCP: 2332 printf("TCP"); 2333 break; 2334 case IPPROTO_UDP: 2335 printf("UDP"); 2336 break; 2337 case IPPROTO_ICMP: 2338 printf("ICMP"); 2339 break; 2340 case IPPROTO_ICMPV6: 2341 printf("ICMPv6"); 2342 break; 2343 default: 2344 printf("%u", proto); 2345 break; 2346 } 2347 switch (dir) { 2348 case PF_IN: 2349 printf(" in"); 2350 break; 2351 case PF_OUT: 2352 printf(" out"); 2353 break; 2354 } 2355 if (skw) { 2356 printf(" wire: "); 2357 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2358 printf(" "); 2359 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2360 } 2361 if (sks) { 2362 printf(" stack: "); 2363 if (sks != skw) { 2364 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2365 printf(" "); 2366 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2367 } else 2368 printf("-"); 2369 } 2370 if (s) { 2371 if (proto == IPPROTO_TCP) { 2372 printf(" [lo=%u high=%u win=%u modulator=%u", 2373 s->src.seqlo, s->src.seqhi, 2374 s->src.max_win, s->src.seqdiff); 2375 if (s->src.wscale && s->dst.wscale) 2376 printf(" wscale=%u", 2377 s->src.wscale & PF_WSCALE_MASK); 2378 printf("]"); 2379 printf(" [lo=%u high=%u win=%u modulator=%u", 2380 s->dst.seqlo, s->dst.seqhi, 2381 s->dst.max_win, s->dst.seqdiff); 2382 if (s->src.wscale && s->dst.wscale) 2383 printf(" wscale=%u", 2384 s->dst.wscale & PF_WSCALE_MASK); 2385 printf("]"); 2386 } 2387 printf(" %u:%u", s->src.state, s->dst.state); 2388 } 2389 } 2390 2391 void 2392 pf_print_flags(u_int8_t f) 2393 { 2394 if (f) 2395 printf(" "); 2396 if (f & TH_FIN) 2397 printf("F"); 2398 if (f & TH_SYN) 2399 printf("S"); 2400 if (f & TH_RST) 2401 printf("R"); 2402 if (f & TH_PUSH) 2403 printf("P"); 2404 if (f & TH_ACK) 2405 printf("A"); 2406 if (f & TH_URG) 2407 printf("U"); 2408 if (f & TH_ECE) 2409 printf("E"); 2410 if (f & TH_CWR) 2411 printf("W"); 2412 } 2413 2414 #define PF_SET_SKIP_STEPS(i) \ 2415 do { \ 2416 while (head[i] != cur) { \ 2417 head[i]->skip[i].ptr = cur; \ 2418 head[i] = TAILQ_NEXT(head[i], entries); \ 2419 } \ 2420 } while (0) 2421 2422 void 2423 pf_calc_skip_steps(struct pf_krulequeue *rules) 2424 { 2425 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2426 int i; 2427 2428 cur = TAILQ_FIRST(rules); 2429 prev = cur; 2430 for (i = 0; i < PF_SKIP_COUNT; ++i) 2431 head[i] = cur; 2432 while (cur != NULL) { 2433 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2434 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2435 if (cur->direction != prev->direction) 2436 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2437 if (cur->af != prev->af) 2438 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2439 if (cur->proto != prev->proto) 2440 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2441 if (cur->src.neg != prev->src.neg || 2442 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2443 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2444 if (cur->src.port[0] != prev->src.port[0] || 2445 cur->src.port[1] != prev->src.port[1] || 2446 cur->src.port_op != prev->src.port_op) 2447 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2448 if (cur->dst.neg != prev->dst.neg || 2449 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2450 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2451 if (cur->dst.port[0] != prev->dst.port[0] || 2452 cur->dst.port[1] != prev->dst.port[1] || 2453 cur->dst.port_op != prev->dst.port_op) 2454 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2455 2456 prev = cur; 2457 cur = TAILQ_NEXT(cur, entries); 2458 } 2459 for (i = 0; i < PF_SKIP_COUNT; ++i) 2460 PF_SET_SKIP_STEPS(i); 2461 } 2462 2463 int 2464 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2465 { 2466 if (aw1->type != aw2->type) 2467 return (1); 2468 switch (aw1->type) { 2469 case PF_ADDR_ADDRMASK: 2470 case PF_ADDR_RANGE: 2471 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2472 return (1); 2473 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2474 return (1); 2475 return (0); 2476 case PF_ADDR_DYNIFTL: 2477 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2478 case PF_ADDR_NOROUTE: 2479 case PF_ADDR_URPFFAILED: 2480 return (0); 2481 case PF_ADDR_TABLE: 2482 return (aw1->p.tbl != aw2->p.tbl); 2483 default: 2484 printf("invalid address type: %d\n", aw1->type); 2485 return (1); 2486 } 2487 } 2488 2489 /** 2490 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2491 * header isn't always a full checksum. In some cases (i.e. output) it's a 2492 * pseudo-header checksum, which is a partial checksum over src/dst IP 2493 * addresses, protocol number and length. 2494 * 2495 * That means we have the following cases: 2496 * * Input or forwarding: we don't have TSO, the checksum fields are full 2497 * checksums, we need to update the checksum whenever we change anything. 2498 * * Output (i.e. the checksum is a pseudo-header checksum): 2499 * x The field being updated is src/dst address or affects the length of 2500 * the packet. We need to update the pseudo-header checksum (note that this 2501 * checksum is not ones' complement). 2502 * x Some other field is being modified (e.g. src/dst port numbers): We 2503 * don't have to update anything. 2504 **/ 2505 u_int16_t 2506 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2507 { 2508 u_int32_t x; 2509 2510 x = cksum + old - new; 2511 x = (x + (x >> 16)) & 0xffff; 2512 2513 /* optimise: eliminate a branch when not udp */ 2514 if (udp && cksum == 0x0000) 2515 return cksum; 2516 if (udp && x == 0x0000) 2517 x = 0xffff; 2518 2519 return (u_int16_t)(x); 2520 } 2521 2522 static void 2523 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2524 u_int8_t udp) 2525 { 2526 u_int16_t old = htons(hi ? (*f << 8) : *f); 2527 u_int16_t new = htons(hi ? ( v << 8) : v); 2528 2529 if (*f == v) 2530 return; 2531 2532 *f = v; 2533 2534 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2535 return; 2536 2537 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2538 } 2539 2540 void 2541 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2542 bool hi, u_int8_t udp) 2543 { 2544 u_int8_t *fb = (u_int8_t *)f; 2545 u_int8_t *vb = (u_int8_t *)&v; 2546 2547 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2548 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2549 } 2550 2551 void 2552 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2553 bool hi, u_int8_t udp) 2554 { 2555 u_int8_t *fb = (u_int8_t *)f; 2556 u_int8_t *vb = (u_int8_t *)&v; 2557 2558 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2559 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2560 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2561 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2562 } 2563 2564 u_int16_t 2565 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2566 u_int16_t new, u_int8_t udp) 2567 { 2568 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2569 return (cksum); 2570 2571 return (pf_cksum_fixup(cksum, old, new, udp)); 2572 } 2573 2574 static void 2575 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2576 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2577 sa_family_t af) 2578 { 2579 struct pf_addr ao; 2580 u_int16_t po = *p; 2581 2582 PF_ACPY(&ao, a, af); 2583 PF_ACPY(a, an, af); 2584 2585 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2586 *pc = ~*pc; 2587 2588 *p = pn; 2589 2590 switch (af) { 2591 #ifdef INET 2592 case AF_INET: 2593 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2594 ao.addr16[0], an->addr16[0], 0), 2595 ao.addr16[1], an->addr16[1], 0); 2596 *p = pn; 2597 2598 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2599 ao.addr16[0], an->addr16[0], u), 2600 ao.addr16[1], an->addr16[1], u); 2601 2602 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2603 break; 2604 #endif /* INET */ 2605 #ifdef INET6 2606 case AF_INET6: 2607 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2608 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2609 pf_cksum_fixup(pf_cksum_fixup(*pc, 2610 ao.addr16[0], an->addr16[0], u), 2611 ao.addr16[1], an->addr16[1], u), 2612 ao.addr16[2], an->addr16[2], u), 2613 ao.addr16[3], an->addr16[3], u), 2614 ao.addr16[4], an->addr16[4], u), 2615 ao.addr16[5], an->addr16[5], u), 2616 ao.addr16[6], an->addr16[6], u), 2617 ao.addr16[7], an->addr16[7], u); 2618 2619 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2620 break; 2621 #endif /* INET6 */ 2622 } 2623 2624 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2625 CSUM_DELAY_DATA_IPV6)) { 2626 *pc = ~*pc; 2627 if (! *pc) 2628 *pc = 0xffff; 2629 } 2630 } 2631 2632 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2633 void 2634 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2635 { 2636 u_int32_t ao; 2637 2638 memcpy(&ao, a, sizeof(ao)); 2639 memcpy(a, &an, sizeof(u_int32_t)); 2640 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2641 ao % 65536, an % 65536, u); 2642 } 2643 2644 void 2645 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2646 { 2647 u_int32_t ao; 2648 2649 memcpy(&ao, a, sizeof(ao)); 2650 memcpy(a, &an, sizeof(u_int32_t)); 2651 2652 *c = pf_proto_cksum_fixup(m, 2653 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2654 ao % 65536, an % 65536, udp); 2655 } 2656 2657 #ifdef INET6 2658 static void 2659 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2660 { 2661 struct pf_addr ao; 2662 2663 PF_ACPY(&ao, a, AF_INET6); 2664 PF_ACPY(a, an, AF_INET6); 2665 2666 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2667 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2668 pf_cksum_fixup(pf_cksum_fixup(*c, 2669 ao.addr16[0], an->addr16[0], u), 2670 ao.addr16[1], an->addr16[1], u), 2671 ao.addr16[2], an->addr16[2], u), 2672 ao.addr16[3], an->addr16[3], u), 2673 ao.addr16[4], an->addr16[4], u), 2674 ao.addr16[5], an->addr16[5], u), 2675 ao.addr16[6], an->addr16[6], u), 2676 ao.addr16[7], an->addr16[7], u); 2677 } 2678 #endif /* INET6 */ 2679 2680 static void 2681 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2682 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2683 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2684 { 2685 struct pf_addr oia, ooa; 2686 2687 PF_ACPY(&oia, ia, af); 2688 if (oa) 2689 PF_ACPY(&ooa, oa, af); 2690 2691 /* Change inner protocol port, fix inner protocol checksum. */ 2692 if (ip != NULL) { 2693 u_int16_t oip = *ip; 2694 u_int32_t opc; 2695 2696 if (pc != NULL) 2697 opc = *pc; 2698 *ip = np; 2699 if (pc != NULL) 2700 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2701 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2702 if (pc != NULL) 2703 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2704 } 2705 /* Change inner ip address, fix inner ip and icmp checksums. */ 2706 PF_ACPY(ia, na, af); 2707 switch (af) { 2708 #ifdef INET 2709 case AF_INET: { 2710 u_int32_t oh2c = *h2c; 2711 2712 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2713 oia.addr16[0], ia->addr16[0], 0), 2714 oia.addr16[1], ia->addr16[1], 0); 2715 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2716 oia.addr16[0], ia->addr16[0], 0), 2717 oia.addr16[1], ia->addr16[1], 0); 2718 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2719 break; 2720 } 2721 #endif /* INET */ 2722 #ifdef INET6 2723 case AF_INET6: 2724 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2725 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2726 pf_cksum_fixup(pf_cksum_fixup(*ic, 2727 oia.addr16[0], ia->addr16[0], u), 2728 oia.addr16[1], ia->addr16[1], u), 2729 oia.addr16[2], ia->addr16[2], u), 2730 oia.addr16[3], ia->addr16[3], u), 2731 oia.addr16[4], ia->addr16[4], u), 2732 oia.addr16[5], ia->addr16[5], u), 2733 oia.addr16[6], ia->addr16[6], u), 2734 oia.addr16[7], ia->addr16[7], u); 2735 break; 2736 #endif /* INET6 */ 2737 } 2738 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2739 if (oa) { 2740 PF_ACPY(oa, na, af); 2741 switch (af) { 2742 #ifdef INET 2743 case AF_INET: 2744 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2745 ooa.addr16[0], oa->addr16[0], 0), 2746 ooa.addr16[1], oa->addr16[1], 0); 2747 break; 2748 #endif /* INET */ 2749 #ifdef INET6 2750 case AF_INET6: 2751 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2752 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2753 pf_cksum_fixup(pf_cksum_fixup(*ic, 2754 ooa.addr16[0], oa->addr16[0], u), 2755 ooa.addr16[1], oa->addr16[1], u), 2756 ooa.addr16[2], oa->addr16[2], u), 2757 ooa.addr16[3], oa->addr16[3], u), 2758 ooa.addr16[4], oa->addr16[4], u), 2759 ooa.addr16[5], oa->addr16[5], u), 2760 ooa.addr16[6], oa->addr16[6], u), 2761 ooa.addr16[7], oa->addr16[7], u); 2762 break; 2763 #endif /* INET6 */ 2764 } 2765 } 2766 } 2767 2768 /* 2769 * Need to modulate the sequence numbers in the TCP SACK option 2770 * (credits to Krzysztof Pfaff for report and patch) 2771 */ 2772 static int 2773 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2774 struct tcphdr *th, struct pf_state_peer *dst) 2775 { 2776 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2777 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2778 int copyback = 0, i, olen; 2779 struct sackblk sack; 2780 2781 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2782 if (hlen < TCPOLEN_SACKLEN || 2783 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2784 return 0; 2785 2786 while (hlen >= TCPOLEN_SACKLEN) { 2787 size_t startoff = opt - opts; 2788 olen = opt[1]; 2789 switch (*opt) { 2790 case TCPOPT_EOL: /* FALLTHROUGH */ 2791 case TCPOPT_NOP: 2792 opt++; 2793 hlen--; 2794 break; 2795 case TCPOPT_SACK: 2796 if (olen > hlen) 2797 olen = hlen; 2798 if (olen >= TCPOLEN_SACKLEN) { 2799 for (i = 2; i + TCPOLEN_SACK <= olen; 2800 i += TCPOLEN_SACK) { 2801 memcpy(&sack, &opt[i], sizeof(sack)); 2802 pf_patch_32_unaligned(m, 2803 &th->th_sum, &sack.start, 2804 htonl(ntohl(sack.start) - dst->seqdiff), 2805 PF_ALGNMNT(startoff), 2806 0); 2807 pf_patch_32_unaligned(m, &th->th_sum, 2808 &sack.end, 2809 htonl(ntohl(sack.end) - dst->seqdiff), 2810 PF_ALGNMNT(startoff), 2811 0); 2812 memcpy(&opt[i], &sack, sizeof(sack)); 2813 } 2814 copyback = 1; 2815 } 2816 /* FALLTHROUGH */ 2817 default: 2818 if (olen < 2) 2819 olen = 2; 2820 hlen -= olen; 2821 opt += olen; 2822 } 2823 } 2824 2825 if (copyback) 2826 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2827 return (copyback); 2828 } 2829 2830 struct mbuf * 2831 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2832 const struct pf_addr *saddr, const struct pf_addr *daddr, 2833 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2834 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 2835 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 2836 { 2837 struct mbuf *m; 2838 int len, tlen; 2839 #ifdef INET 2840 struct ip *h = NULL; 2841 #endif /* INET */ 2842 #ifdef INET6 2843 struct ip6_hdr *h6 = NULL; 2844 #endif /* INET6 */ 2845 struct tcphdr *th; 2846 char *opt; 2847 struct pf_mtag *pf_mtag; 2848 2849 len = 0; 2850 th = NULL; 2851 2852 /* maximum segment size tcp option */ 2853 tlen = sizeof(struct tcphdr); 2854 if (mss) 2855 tlen += 4; 2856 2857 switch (af) { 2858 #ifdef INET 2859 case AF_INET: 2860 len = sizeof(struct ip) + tlen; 2861 break; 2862 #endif /* INET */ 2863 #ifdef INET6 2864 case AF_INET6: 2865 len = sizeof(struct ip6_hdr) + tlen; 2866 break; 2867 #endif /* INET6 */ 2868 default: 2869 panic("%s: unsupported af %d", __func__, af); 2870 } 2871 2872 m = m_gethdr(M_NOWAIT, MT_DATA); 2873 if (m == NULL) 2874 return (NULL); 2875 2876 #ifdef MAC 2877 mac_netinet_firewall_send(m); 2878 #endif 2879 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2880 m_freem(m); 2881 return (NULL); 2882 } 2883 if (skip_firewall) 2884 m->m_flags |= M_SKIP_FIREWALL; 2885 pf_mtag->tag = mtag_tag; 2886 pf_mtag->flags = mtag_flags; 2887 2888 if (rtableid >= 0) 2889 M_SETFIB(m, rtableid); 2890 2891 #ifdef ALTQ 2892 if (r != NULL && r->qid) { 2893 pf_mtag->qid = r->qid; 2894 2895 /* add hints for ecn */ 2896 pf_mtag->hdr = mtod(m, struct ip *); 2897 } 2898 #endif /* ALTQ */ 2899 m->m_data += max_linkhdr; 2900 m->m_pkthdr.len = m->m_len = len; 2901 /* The rest of the stack assumes a rcvif, so provide one. 2902 * This is a locally generated packet, so .. close enough. */ 2903 m->m_pkthdr.rcvif = V_loif; 2904 bzero(m->m_data, len); 2905 switch (af) { 2906 #ifdef INET 2907 case AF_INET: 2908 h = mtod(m, struct ip *); 2909 2910 /* IP header fields included in the TCP checksum */ 2911 h->ip_p = IPPROTO_TCP; 2912 h->ip_len = htons(tlen); 2913 h->ip_src.s_addr = saddr->v4.s_addr; 2914 h->ip_dst.s_addr = daddr->v4.s_addr; 2915 2916 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2917 break; 2918 #endif /* INET */ 2919 #ifdef INET6 2920 case AF_INET6: 2921 h6 = mtod(m, struct ip6_hdr *); 2922 2923 /* IP header fields included in the TCP checksum */ 2924 h6->ip6_nxt = IPPROTO_TCP; 2925 h6->ip6_plen = htons(tlen); 2926 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2927 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2928 2929 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2930 break; 2931 #endif /* INET6 */ 2932 } 2933 2934 /* TCP header */ 2935 th->th_sport = sport; 2936 th->th_dport = dport; 2937 th->th_seq = htonl(seq); 2938 th->th_ack = htonl(ack); 2939 th->th_off = tlen >> 2; 2940 th->th_flags = tcp_flags; 2941 th->th_win = htons(win); 2942 2943 if (mss) { 2944 opt = (char *)(th + 1); 2945 opt[0] = TCPOPT_MAXSEG; 2946 opt[1] = 4; 2947 HTONS(mss); 2948 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2949 } 2950 2951 switch (af) { 2952 #ifdef INET 2953 case AF_INET: 2954 /* TCP checksum */ 2955 th->th_sum = in_cksum(m, len); 2956 2957 /* Finish the IP header */ 2958 h->ip_v = 4; 2959 h->ip_hl = sizeof(*h) >> 2; 2960 h->ip_tos = IPTOS_LOWDELAY; 2961 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2962 h->ip_len = htons(len); 2963 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2964 h->ip_sum = 0; 2965 break; 2966 #endif /* INET */ 2967 #ifdef INET6 2968 case AF_INET6: 2969 /* TCP checksum */ 2970 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2971 sizeof(struct ip6_hdr), tlen); 2972 2973 h6->ip6_vfc |= IPV6_VERSION; 2974 h6->ip6_hlim = IPV6_DEFHLIM; 2975 break; 2976 #endif /* INET6 */ 2977 } 2978 2979 return (m); 2980 } 2981 2982 static void 2983 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 2984 uint8_t ttl, int rtableid) 2985 { 2986 struct mbuf *m; 2987 #ifdef INET 2988 struct ip *h = NULL; 2989 #endif /* INET */ 2990 #ifdef INET6 2991 struct ip6_hdr *h6 = NULL; 2992 #endif /* INET6 */ 2993 struct sctphdr *hdr; 2994 struct sctp_chunkhdr *chunk; 2995 struct pf_send_entry *pfse; 2996 int off = 0; 2997 2998 MPASS(af == pd->af); 2999 3000 m = m_gethdr(M_NOWAIT, MT_DATA); 3001 if (m == NULL) 3002 return; 3003 3004 m->m_data += max_linkhdr; 3005 m->m_flags |= M_SKIP_FIREWALL; 3006 /* The rest of the stack assumes a rcvif, so provide one. 3007 * This is a locally generated packet, so .. close enough. */ 3008 m->m_pkthdr.rcvif = V_loif; 3009 3010 /* IPv4|6 header */ 3011 switch (af) { 3012 #ifdef INET 3013 case AF_INET: 3014 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3015 3016 h = mtod(m, struct ip *); 3017 3018 /* IP header fields included in the TCP checksum */ 3019 3020 h->ip_p = IPPROTO_SCTP; 3021 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3022 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3023 h->ip_src = pd->dst->v4; 3024 h->ip_dst = pd->src->v4; 3025 3026 off += sizeof(struct ip); 3027 break; 3028 #endif /* INET */ 3029 #ifdef INET6 3030 case AF_INET6: 3031 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3032 3033 h6 = mtod(m, struct ip6_hdr *); 3034 3035 /* IP header fields included in the TCP checksum */ 3036 h6->ip6_vfc |= IPV6_VERSION; 3037 h6->ip6_nxt = IPPROTO_SCTP; 3038 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3039 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3040 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3041 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3042 3043 off += sizeof(struct ip6_hdr); 3044 break; 3045 #endif /* INET6 */ 3046 } 3047 3048 /* SCTP header */ 3049 hdr = mtodo(m, off); 3050 3051 hdr->src_port = pd->hdr.sctp.dest_port; 3052 hdr->dest_port = pd->hdr.sctp.src_port; 3053 hdr->v_tag = pd->sctp_initiate_tag; 3054 hdr->checksum = 0; 3055 3056 /* Abort chunk. */ 3057 off += sizeof(struct sctphdr); 3058 chunk = mtodo(m, off); 3059 3060 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3061 chunk->chunk_length = htons(sizeof(*chunk)); 3062 3063 /* SCTP checksum */ 3064 off += sizeof(*chunk); 3065 m->m_pkthdr.len = m->m_len = off; 3066 3067 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));; 3068 3069 if (rtableid >= 0) 3070 M_SETFIB(m, rtableid); 3071 3072 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3073 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3074 if (pfse == NULL) { 3075 m_freem(m); 3076 return; 3077 } 3078 3079 switch (af) { 3080 #ifdef INET 3081 case AF_INET: 3082 pfse->pfse_type = PFSE_IP; 3083 break; 3084 #endif /* INET */ 3085 #ifdef INET6 3086 case AF_INET6: 3087 pfse->pfse_type = PFSE_IP6; 3088 break; 3089 #endif /* INET6 */ 3090 } 3091 3092 pfse->pfse_m = m; 3093 pf_send(pfse); 3094 } 3095 3096 void 3097 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3098 const struct pf_addr *saddr, const struct pf_addr *daddr, 3099 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3100 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3101 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3102 { 3103 struct pf_send_entry *pfse; 3104 struct mbuf *m; 3105 3106 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3107 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3108 if (m == NULL) 3109 return; 3110 3111 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3112 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3113 if (pfse == NULL) { 3114 m_freem(m); 3115 return; 3116 } 3117 3118 switch (af) { 3119 #ifdef INET 3120 case AF_INET: 3121 pfse->pfse_type = PFSE_IP; 3122 break; 3123 #endif /* INET */ 3124 #ifdef INET6 3125 case AF_INET6: 3126 pfse->pfse_type = PFSE_IP6; 3127 break; 3128 #endif /* INET6 */ 3129 } 3130 3131 pfse->pfse_m = m; 3132 pf_send(pfse); 3133 } 3134 3135 static void 3136 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3137 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3138 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3139 u_short *reason, int rtableid) 3140 { 3141 struct pf_addr * const saddr = pd->src; 3142 struct pf_addr * const daddr = pd->dst; 3143 sa_family_t af = pd->af; 3144 3145 /* undo NAT changes, if they have taken place */ 3146 if (nr != NULL) { 3147 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3148 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3149 if (pd->sport) 3150 *pd->sport = sk->port[pd->sidx]; 3151 if (pd->dport) 3152 *pd->dport = sk->port[pd->didx]; 3153 if (pd->proto_sum) 3154 *pd->proto_sum = bproto_sum; 3155 if (pd->ip_sum) 3156 *pd->ip_sum = bip_sum; 3157 m_copyback(m, off, hdrlen, pd->hdr.any); 3158 } 3159 if (pd->proto == IPPROTO_TCP && 3160 ((r->rule_flag & PFRULE_RETURNRST) || 3161 (r->rule_flag & PFRULE_RETURN)) && 3162 !(th->th_flags & TH_RST)) { 3163 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3164 int len = 0; 3165 #ifdef INET 3166 struct ip *h4; 3167 #endif 3168 #ifdef INET6 3169 struct ip6_hdr *h6; 3170 #endif 3171 3172 switch (af) { 3173 #ifdef INET 3174 case AF_INET: 3175 h4 = mtod(m, struct ip *); 3176 len = ntohs(h4->ip_len) - off; 3177 break; 3178 #endif 3179 #ifdef INET6 3180 case AF_INET6: 3181 h6 = mtod(m, struct ip6_hdr *); 3182 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3183 break; 3184 #endif 3185 } 3186 3187 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3188 REASON_SET(reason, PFRES_PROTCKSUM); 3189 else { 3190 if (th->th_flags & TH_SYN) 3191 ack++; 3192 if (th->th_flags & TH_FIN) 3193 ack++; 3194 pf_send_tcp(r, af, pd->dst, 3195 pd->src, th->th_dport, th->th_sport, 3196 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3197 r->return_ttl, true, 0, 0, rtableid); 3198 } 3199 } else if (pd->proto == IPPROTO_SCTP && 3200 (r->rule_flag & PFRULE_RETURN)) { 3201 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3202 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3203 r->return_icmp) 3204 pf_send_icmp(m, r->return_icmp >> 8, 3205 r->return_icmp & 255, af, r, rtableid); 3206 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3207 r->return_icmp6) 3208 pf_send_icmp(m, r->return_icmp6 >> 8, 3209 r->return_icmp6 & 255, af, r, rtableid); 3210 } 3211 3212 static int 3213 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3214 { 3215 struct m_tag *mtag; 3216 u_int8_t mpcp; 3217 3218 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3219 if (mtag == NULL) 3220 return (0); 3221 3222 if (prio == PF_PRIO_ZERO) 3223 prio = 0; 3224 3225 mpcp = *(uint8_t *)(mtag + 1); 3226 3227 return (mpcp == prio); 3228 } 3229 3230 static int 3231 pf_icmp_to_bandlim(uint8_t type) 3232 { 3233 switch (type) { 3234 case ICMP_ECHO: 3235 case ICMP_ECHOREPLY: 3236 return (BANDLIM_ICMP_ECHO); 3237 case ICMP_TSTAMP: 3238 case ICMP_TSTAMPREPLY: 3239 return (BANDLIM_ICMP_TSTAMP); 3240 case ICMP_UNREACH: 3241 default: 3242 return (BANDLIM_ICMP_UNREACH); 3243 } 3244 } 3245 3246 static void 3247 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3248 struct pf_krule *r, int rtableid) 3249 { 3250 struct pf_send_entry *pfse; 3251 struct mbuf *m0; 3252 struct pf_mtag *pf_mtag; 3253 3254 /* ICMP packet rate limitation. */ 3255 #ifdef INET6 3256 if (af == AF_INET6) { 3257 if (icmp6_ratelimit(NULL, type, code)) 3258 return; 3259 } 3260 #endif 3261 #ifdef INET 3262 if (af == AF_INET) { 3263 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3264 return; 3265 } 3266 #endif 3267 3268 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3269 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3270 if (pfse == NULL) 3271 return; 3272 3273 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3274 free(pfse, M_PFTEMP); 3275 return; 3276 } 3277 3278 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3279 free(pfse, M_PFTEMP); 3280 return; 3281 } 3282 /* XXX: revisit */ 3283 m0->m_flags |= M_SKIP_FIREWALL; 3284 3285 if (rtableid >= 0) 3286 M_SETFIB(m0, rtableid); 3287 3288 #ifdef ALTQ 3289 if (r->qid) { 3290 pf_mtag->qid = r->qid; 3291 /* add hints for ecn */ 3292 pf_mtag->hdr = mtod(m0, struct ip *); 3293 } 3294 #endif /* ALTQ */ 3295 3296 switch (af) { 3297 #ifdef INET 3298 case AF_INET: 3299 pfse->pfse_type = PFSE_ICMP; 3300 break; 3301 #endif /* INET */ 3302 #ifdef INET6 3303 case AF_INET6: 3304 pfse->pfse_type = PFSE_ICMP6; 3305 break; 3306 #endif /* INET6 */ 3307 } 3308 pfse->pfse_m = m0; 3309 pfse->icmpopts.type = type; 3310 pfse->icmpopts.code = code; 3311 pf_send(pfse); 3312 } 3313 3314 /* 3315 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3316 * If n is 0, they match if they are equal. If n is != 0, they match if they 3317 * are different. 3318 */ 3319 int 3320 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3321 struct pf_addr *b, sa_family_t af) 3322 { 3323 int match = 0; 3324 3325 switch (af) { 3326 #ifdef INET 3327 case AF_INET: 3328 if ((a->addr32[0] & m->addr32[0]) == 3329 (b->addr32[0] & m->addr32[0])) 3330 match++; 3331 break; 3332 #endif /* INET */ 3333 #ifdef INET6 3334 case AF_INET6: 3335 if (((a->addr32[0] & m->addr32[0]) == 3336 (b->addr32[0] & m->addr32[0])) && 3337 ((a->addr32[1] & m->addr32[1]) == 3338 (b->addr32[1] & m->addr32[1])) && 3339 ((a->addr32[2] & m->addr32[2]) == 3340 (b->addr32[2] & m->addr32[2])) && 3341 ((a->addr32[3] & m->addr32[3]) == 3342 (b->addr32[3] & m->addr32[3]))) 3343 match++; 3344 break; 3345 #endif /* INET6 */ 3346 } 3347 if (match) { 3348 if (n) 3349 return (0); 3350 else 3351 return (1); 3352 } else { 3353 if (n) 3354 return (1); 3355 else 3356 return (0); 3357 } 3358 } 3359 3360 /* 3361 * Return 1 if b <= a <= e, otherwise return 0. 3362 */ 3363 int 3364 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3365 struct pf_addr *a, sa_family_t af) 3366 { 3367 switch (af) { 3368 #ifdef INET 3369 case AF_INET: 3370 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3371 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3372 return (0); 3373 break; 3374 #endif /* INET */ 3375 #ifdef INET6 3376 case AF_INET6: { 3377 int i; 3378 3379 /* check a >= b */ 3380 for (i = 0; i < 4; ++i) 3381 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3382 break; 3383 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3384 return (0); 3385 /* check a <= e */ 3386 for (i = 0; i < 4; ++i) 3387 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3388 break; 3389 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3390 return (0); 3391 break; 3392 } 3393 #endif /* INET6 */ 3394 } 3395 return (1); 3396 } 3397 3398 static int 3399 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3400 { 3401 switch (op) { 3402 case PF_OP_IRG: 3403 return ((p > a1) && (p < a2)); 3404 case PF_OP_XRG: 3405 return ((p < a1) || (p > a2)); 3406 case PF_OP_RRG: 3407 return ((p >= a1) && (p <= a2)); 3408 case PF_OP_EQ: 3409 return (p == a1); 3410 case PF_OP_NE: 3411 return (p != a1); 3412 case PF_OP_LT: 3413 return (p < a1); 3414 case PF_OP_LE: 3415 return (p <= a1); 3416 case PF_OP_GT: 3417 return (p > a1); 3418 case PF_OP_GE: 3419 return (p >= a1); 3420 } 3421 return (0); /* never reached */ 3422 } 3423 3424 int 3425 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3426 { 3427 NTOHS(a1); 3428 NTOHS(a2); 3429 NTOHS(p); 3430 return (pf_match(op, a1, a2, p)); 3431 } 3432 3433 static int 3434 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3435 { 3436 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3437 return (0); 3438 return (pf_match(op, a1, a2, u)); 3439 } 3440 3441 static int 3442 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3443 { 3444 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3445 return (0); 3446 return (pf_match(op, a1, a2, g)); 3447 } 3448 3449 int 3450 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3451 { 3452 if (*tag == -1) 3453 *tag = mtag; 3454 3455 return ((!r->match_tag_not && r->match_tag == *tag) || 3456 (r->match_tag_not && r->match_tag != *tag)); 3457 } 3458 3459 int 3460 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3461 { 3462 3463 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3464 3465 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3466 return (ENOMEM); 3467 3468 pd->pf_mtag->tag = tag; 3469 3470 return (0); 3471 } 3472 3473 #define PF_ANCHOR_STACKSIZE 32 3474 struct pf_kanchor_stackframe { 3475 struct pf_kruleset *rs; 3476 struct pf_krule *r; /* XXX: + match bit */ 3477 struct pf_kanchor *child; 3478 }; 3479 3480 /* 3481 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3482 */ 3483 #define PF_ANCHORSTACK_MATCH 0x00000001 3484 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3485 3486 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3487 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3488 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3489 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3490 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3491 } while (0) 3492 3493 void 3494 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3495 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3496 int *match) 3497 { 3498 struct pf_kanchor_stackframe *f; 3499 3500 PF_RULES_RASSERT(); 3501 3502 if (match) 3503 *match = 0; 3504 if (*depth >= PF_ANCHOR_STACKSIZE) { 3505 printf("%s: anchor stack overflow on %s\n", 3506 __func__, (*r)->anchor->name); 3507 *r = TAILQ_NEXT(*r, entries); 3508 return; 3509 } else if (*depth == 0 && a != NULL) 3510 *a = *r; 3511 f = stack + (*depth)++; 3512 f->rs = *rs; 3513 f->r = *r; 3514 if ((*r)->anchor_wildcard) { 3515 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3516 3517 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3518 *r = NULL; 3519 return; 3520 } 3521 *rs = &f->child->ruleset; 3522 } else { 3523 f->child = NULL; 3524 *rs = &(*r)->anchor->ruleset; 3525 } 3526 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3527 } 3528 3529 int 3530 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3531 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3532 int *match) 3533 { 3534 struct pf_kanchor_stackframe *f; 3535 struct pf_krule *fr; 3536 int quick = 0; 3537 3538 PF_RULES_RASSERT(); 3539 3540 do { 3541 if (*depth <= 0) 3542 break; 3543 f = stack + *depth - 1; 3544 fr = PF_ANCHOR_RULE(f); 3545 if (f->child != NULL) { 3546 /* 3547 * This block traverses through 3548 * a wildcard anchor. 3549 */ 3550 if (match != NULL && *match) { 3551 /* 3552 * If any of "*" matched, then 3553 * "foo/ *" matched, mark frame 3554 * appropriately. 3555 */ 3556 PF_ANCHOR_SET_MATCH(f); 3557 *match = 0; 3558 } 3559 f->child = RB_NEXT(pf_kanchor_node, 3560 &fr->anchor->children, f->child); 3561 if (f->child != NULL) { 3562 *rs = &f->child->ruleset; 3563 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3564 if (*r == NULL) 3565 continue; 3566 else 3567 break; 3568 } 3569 } 3570 (*depth)--; 3571 if (*depth == 0 && a != NULL) 3572 *a = NULL; 3573 *rs = f->rs; 3574 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3575 quick = fr->quick; 3576 *r = TAILQ_NEXT(fr, entries); 3577 } while (*r == NULL); 3578 3579 return (quick); 3580 } 3581 3582 struct pf_keth_anchor_stackframe { 3583 struct pf_keth_ruleset *rs; 3584 struct pf_keth_rule *r; /* XXX: + match bit */ 3585 struct pf_keth_anchor *child; 3586 }; 3587 3588 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3589 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3590 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3591 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3592 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3593 } while (0) 3594 3595 void 3596 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3597 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3598 struct pf_keth_rule **a, int *match) 3599 { 3600 struct pf_keth_anchor_stackframe *f; 3601 3602 NET_EPOCH_ASSERT(); 3603 3604 if (match) 3605 *match = 0; 3606 if (*depth >= PF_ANCHOR_STACKSIZE) { 3607 printf("%s: anchor stack overflow on %s\n", 3608 __func__, (*r)->anchor->name); 3609 *r = TAILQ_NEXT(*r, entries); 3610 return; 3611 } else if (*depth == 0 && a != NULL) 3612 *a = *r; 3613 f = stack + (*depth)++; 3614 f->rs = *rs; 3615 f->r = *r; 3616 if ((*r)->anchor_wildcard) { 3617 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3618 3619 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3620 *r = NULL; 3621 return; 3622 } 3623 *rs = &f->child->ruleset; 3624 } else { 3625 f->child = NULL; 3626 *rs = &(*r)->anchor->ruleset; 3627 } 3628 *r = TAILQ_FIRST((*rs)->active.rules); 3629 } 3630 3631 int 3632 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3633 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3634 struct pf_keth_rule **a, int *match) 3635 { 3636 struct pf_keth_anchor_stackframe *f; 3637 struct pf_keth_rule *fr; 3638 int quick = 0; 3639 3640 NET_EPOCH_ASSERT(); 3641 3642 do { 3643 if (*depth <= 0) 3644 break; 3645 f = stack + *depth - 1; 3646 fr = PF_ETH_ANCHOR_RULE(f); 3647 if (f->child != NULL) { 3648 /* 3649 * This block traverses through 3650 * a wildcard anchor. 3651 */ 3652 if (match != NULL && *match) { 3653 /* 3654 * If any of "*" matched, then 3655 * "foo/ *" matched, mark frame 3656 * appropriately. 3657 */ 3658 PF_ETH_ANCHOR_SET_MATCH(f); 3659 *match = 0; 3660 } 3661 f->child = RB_NEXT(pf_keth_anchor_node, 3662 &fr->anchor->children, f->child); 3663 if (f->child != NULL) { 3664 *rs = &f->child->ruleset; 3665 *r = TAILQ_FIRST((*rs)->active.rules); 3666 if (*r == NULL) 3667 continue; 3668 else 3669 break; 3670 } 3671 } 3672 (*depth)--; 3673 if (*depth == 0 && a != NULL) 3674 *a = NULL; 3675 *rs = f->rs; 3676 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3677 quick = fr->quick; 3678 *r = TAILQ_NEXT(fr, entries); 3679 } while (*r == NULL); 3680 3681 return (quick); 3682 } 3683 3684 #ifdef INET6 3685 void 3686 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3687 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3688 { 3689 switch (af) { 3690 #ifdef INET 3691 case AF_INET: 3692 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3693 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3694 break; 3695 #endif /* INET */ 3696 case AF_INET6: 3697 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3698 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3699 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3700 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3701 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3702 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3703 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3704 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3705 break; 3706 } 3707 } 3708 3709 void 3710 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3711 { 3712 switch (af) { 3713 #ifdef INET 3714 case AF_INET: 3715 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3716 break; 3717 #endif /* INET */ 3718 case AF_INET6: 3719 if (addr->addr32[3] == 0xffffffff) { 3720 addr->addr32[3] = 0; 3721 if (addr->addr32[2] == 0xffffffff) { 3722 addr->addr32[2] = 0; 3723 if (addr->addr32[1] == 0xffffffff) { 3724 addr->addr32[1] = 0; 3725 addr->addr32[0] = 3726 htonl(ntohl(addr->addr32[0]) + 1); 3727 } else 3728 addr->addr32[1] = 3729 htonl(ntohl(addr->addr32[1]) + 1); 3730 } else 3731 addr->addr32[2] = 3732 htonl(ntohl(addr->addr32[2]) + 1); 3733 } else 3734 addr->addr32[3] = 3735 htonl(ntohl(addr->addr32[3]) + 1); 3736 break; 3737 } 3738 } 3739 #endif /* INET6 */ 3740 3741 void 3742 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3743 { 3744 /* 3745 * Modern rules use the same flags in rules as they do in states. 3746 */ 3747 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3748 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3749 3750 /* 3751 * Old-style scrub rules have different flags which need to be translated. 3752 */ 3753 if (r->rule_flag & PFRULE_RANDOMID) 3754 a->flags |= PFSTATE_RANDOMID; 3755 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 3756 a->flags |= PFSTATE_SETTOS; 3757 a->set_tos = r->set_tos; 3758 } 3759 3760 if (r->qid) 3761 a->qid = r->qid; 3762 if (r->pqid) 3763 a->pqid = r->pqid; 3764 if (r->rtableid >= 0) 3765 a->rtableid = r->rtableid; 3766 a->log |= r->log; 3767 if (r->min_ttl) 3768 a->min_ttl = r->min_ttl; 3769 if (r->max_mss) 3770 a->max_mss = r->max_mss; 3771 if (r->dnpipe) 3772 a->dnpipe = r->dnpipe; 3773 if (r->dnrpipe) 3774 a->dnrpipe = r->dnrpipe; 3775 if (r->dnpipe || r->dnrpipe) { 3776 if (r->free_flags & PFRULE_DN_IS_PIPE) 3777 a->flags |= PFSTATE_DN_IS_PIPE; 3778 else 3779 a->flags &= ~PFSTATE_DN_IS_PIPE; 3780 } 3781 if (r->scrub_flags & PFSTATE_SETPRIO) { 3782 a->set_prio[0] = r->set_prio[0]; 3783 a->set_prio[1] = r->set_prio[1]; 3784 } 3785 } 3786 3787 int 3788 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 3789 { 3790 struct pf_addr *saddr, *daddr; 3791 u_int16_t sport, dport; 3792 struct inpcbinfo *pi; 3793 struct inpcb *inp; 3794 3795 pd->lookup.uid = UID_MAX; 3796 pd->lookup.gid = GID_MAX; 3797 3798 switch (pd->proto) { 3799 case IPPROTO_TCP: 3800 sport = pd->hdr.tcp.th_sport; 3801 dport = pd->hdr.tcp.th_dport; 3802 pi = &V_tcbinfo; 3803 break; 3804 case IPPROTO_UDP: 3805 sport = pd->hdr.udp.uh_sport; 3806 dport = pd->hdr.udp.uh_dport; 3807 pi = &V_udbinfo; 3808 break; 3809 default: 3810 return (-1); 3811 } 3812 if (pd->dir == PF_IN) { 3813 saddr = pd->src; 3814 daddr = pd->dst; 3815 } else { 3816 u_int16_t p; 3817 3818 p = sport; 3819 sport = dport; 3820 dport = p; 3821 saddr = pd->dst; 3822 daddr = pd->src; 3823 } 3824 switch (pd->af) { 3825 #ifdef INET 3826 case AF_INET: 3827 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3828 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3829 if (inp == NULL) { 3830 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3831 daddr->v4, dport, INPLOOKUP_WILDCARD | 3832 INPLOOKUP_RLOCKPCB, NULL, m); 3833 if (inp == NULL) 3834 return (-1); 3835 } 3836 break; 3837 #endif /* INET */ 3838 #ifdef INET6 3839 case AF_INET6: 3840 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3841 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3842 if (inp == NULL) { 3843 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3844 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3845 INPLOOKUP_RLOCKPCB, NULL, m); 3846 if (inp == NULL) 3847 return (-1); 3848 } 3849 break; 3850 #endif /* INET6 */ 3851 3852 default: 3853 return (-1); 3854 } 3855 INP_RLOCK_ASSERT(inp); 3856 pd->lookup.uid = inp->inp_cred->cr_uid; 3857 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3858 INP_RUNLOCK(inp); 3859 3860 return (1); 3861 } 3862 3863 u_int8_t 3864 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3865 { 3866 int hlen; 3867 u_int8_t hdr[60]; 3868 u_int8_t *opt, optlen; 3869 u_int8_t wscale = 0; 3870 3871 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3872 if (hlen <= sizeof(struct tcphdr)) 3873 return (0); 3874 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3875 return (0); 3876 opt = hdr + sizeof(struct tcphdr); 3877 hlen -= sizeof(struct tcphdr); 3878 while (hlen >= 3) { 3879 switch (*opt) { 3880 case TCPOPT_EOL: 3881 case TCPOPT_NOP: 3882 ++opt; 3883 --hlen; 3884 break; 3885 case TCPOPT_WINDOW: 3886 wscale = opt[2]; 3887 if (wscale > TCP_MAX_WINSHIFT) 3888 wscale = TCP_MAX_WINSHIFT; 3889 wscale |= PF_WSCALE_FLAG; 3890 /* FALLTHROUGH */ 3891 default: 3892 optlen = opt[1]; 3893 if (optlen < 2) 3894 optlen = 2; 3895 hlen -= optlen; 3896 opt += optlen; 3897 break; 3898 } 3899 } 3900 return (wscale); 3901 } 3902 3903 u_int16_t 3904 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3905 { 3906 int hlen; 3907 u_int8_t hdr[60]; 3908 u_int8_t *opt, optlen; 3909 u_int16_t mss = V_tcp_mssdflt; 3910 3911 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3912 if (hlen <= sizeof(struct tcphdr)) 3913 return (0); 3914 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3915 return (0); 3916 opt = hdr + sizeof(struct tcphdr); 3917 hlen -= sizeof(struct tcphdr); 3918 while (hlen >= TCPOLEN_MAXSEG) { 3919 switch (*opt) { 3920 case TCPOPT_EOL: 3921 case TCPOPT_NOP: 3922 ++opt; 3923 --hlen; 3924 break; 3925 case TCPOPT_MAXSEG: 3926 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3927 NTOHS(mss); 3928 /* FALLTHROUGH */ 3929 default: 3930 optlen = opt[1]; 3931 if (optlen < 2) 3932 optlen = 2; 3933 hlen -= optlen; 3934 opt += optlen; 3935 break; 3936 } 3937 } 3938 return (mss); 3939 } 3940 3941 static u_int16_t 3942 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3943 { 3944 struct nhop_object *nh; 3945 #ifdef INET6 3946 struct in6_addr dst6; 3947 uint32_t scopeid; 3948 #endif /* INET6 */ 3949 int hlen = 0; 3950 uint16_t mss = 0; 3951 3952 NET_EPOCH_ASSERT(); 3953 3954 switch (af) { 3955 #ifdef INET 3956 case AF_INET: 3957 hlen = sizeof(struct ip); 3958 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3959 if (nh != NULL) 3960 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3961 break; 3962 #endif /* INET */ 3963 #ifdef INET6 3964 case AF_INET6: 3965 hlen = sizeof(struct ip6_hdr); 3966 in6_splitscope(&addr->v6, &dst6, &scopeid); 3967 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 3968 if (nh != NULL) 3969 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3970 break; 3971 #endif /* INET6 */ 3972 } 3973 3974 mss = max(V_tcp_mssdflt, mss); 3975 mss = min(mss, offer); 3976 mss = max(mss, 64); /* sanity - at least max opt space */ 3977 return (mss); 3978 } 3979 3980 static u_int32_t 3981 pf_tcp_iss(struct pf_pdesc *pd) 3982 { 3983 MD5_CTX ctx; 3984 u_int32_t digest[4]; 3985 3986 if (V_pf_tcp_secret_init == 0) { 3987 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3988 MD5Init(&V_pf_tcp_secret_ctx); 3989 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3990 sizeof(V_pf_tcp_secret)); 3991 V_pf_tcp_secret_init = 1; 3992 } 3993 3994 ctx = V_pf_tcp_secret_ctx; 3995 3996 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 3997 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 3998 if (pd->af == AF_INET6) { 3999 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4000 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4001 } else { 4002 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4003 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4004 } 4005 MD5Final((u_char *)digest, &ctx); 4006 V_pf_tcp_iss_off += 4096; 4007 #define ISN_RANDOM_INCREMENT (4096 - 1) 4008 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4009 V_pf_tcp_iss_off); 4010 #undef ISN_RANDOM_INCREMENT 4011 } 4012 4013 static bool 4014 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4015 { 4016 bool match = true; 4017 4018 /* Always matches if not set */ 4019 if (! r->isset) 4020 return (!r->neg); 4021 4022 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4023 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4024 match = false; 4025 break; 4026 } 4027 } 4028 4029 return (match ^ r->neg); 4030 } 4031 4032 static int 4033 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4034 { 4035 if (*tag == -1) 4036 *tag = mtag; 4037 4038 return ((!r->match_tag_not && r->match_tag == *tag) || 4039 (r->match_tag_not && r->match_tag != *tag)); 4040 } 4041 4042 static void 4043 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4044 { 4045 /* If we don't have the interface drop the packet. */ 4046 if (ifp == NULL) { 4047 m_freem(m); 4048 return; 4049 } 4050 4051 switch (ifp->if_type) { 4052 case IFT_ETHER: 4053 case IFT_XETHER: 4054 case IFT_L2VLAN: 4055 case IFT_BRIDGE: 4056 case IFT_IEEE8023ADLAG: 4057 break; 4058 default: 4059 m_freem(m); 4060 return; 4061 } 4062 4063 ifp->if_transmit(ifp, m); 4064 } 4065 4066 static int 4067 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4068 { 4069 #ifdef INET 4070 struct ip ip; 4071 #endif 4072 #ifdef INET6 4073 struct ip6_hdr ip6; 4074 #endif 4075 struct mbuf *m = *m0; 4076 struct ether_header *e; 4077 struct pf_keth_rule *r, *rm, *a = NULL; 4078 struct pf_keth_ruleset *ruleset = NULL; 4079 struct pf_mtag *mtag; 4080 struct pf_keth_ruleq *rules; 4081 struct pf_addr *src = NULL, *dst = NULL; 4082 struct pfi_kkif *bridge_to; 4083 sa_family_t af = 0; 4084 uint16_t proto; 4085 int asd = 0, match = 0; 4086 int tag = -1; 4087 uint8_t action; 4088 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4089 4090 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4091 NET_EPOCH_ASSERT(); 4092 4093 PF_RULES_RLOCK_TRACKER; 4094 4095 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4096 4097 mtag = pf_find_mtag(m); 4098 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4099 /* Dummynet re-injects packets after they've 4100 * completed their delay. We've already 4101 * processed them, so pass unconditionally. */ 4102 4103 /* But only once. We may see the packet multiple times (e.g. 4104 * PFIL_IN/PFIL_OUT). */ 4105 mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 4106 4107 return (PF_PASS); 4108 } 4109 4110 ruleset = V_pf_keth; 4111 rules = ck_pr_load_ptr(&ruleset->active.rules); 4112 r = TAILQ_FIRST(rules); 4113 rm = NULL; 4114 4115 e = mtod(m, struct ether_header *); 4116 proto = ntohs(e->ether_type); 4117 4118 switch (proto) { 4119 #ifdef INET 4120 case ETHERTYPE_IP: { 4121 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4122 sizeof(ip))) 4123 return (PF_DROP); 4124 4125 af = AF_INET; 4126 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4127 (caddr_t)&ip); 4128 src = (struct pf_addr *)&ip.ip_src; 4129 dst = (struct pf_addr *)&ip.ip_dst; 4130 break; 4131 } 4132 #endif /* INET */ 4133 #ifdef INET6 4134 case ETHERTYPE_IPV6: { 4135 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4136 sizeof(ip6))) 4137 return (PF_DROP); 4138 4139 af = AF_INET6; 4140 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4141 (caddr_t)&ip6); 4142 src = (struct pf_addr *)&ip6.ip6_src; 4143 dst = (struct pf_addr *)&ip6.ip6_dst; 4144 break; 4145 } 4146 #endif /* INET6 */ 4147 } 4148 4149 PF_RULES_RLOCK(); 4150 4151 while (r != NULL) { 4152 counter_u64_add(r->evaluations, 1); 4153 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4154 4155 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4156 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4157 "kif"); 4158 r = r->skip[PFE_SKIP_IFP].ptr; 4159 } 4160 else if (r->direction && r->direction != dir) { 4161 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4162 "dir"); 4163 r = r->skip[PFE_SKIP_DIR].ptr; 4164 } 4165 else if (r->proto && r->proto != proto) { 4166 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4167 "proto"); 4168 r = r->skip[PFE_SKIP_PROTO].ptr; 4169 } 4170 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4171 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4172 "src"); 4173 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4174 } 4175 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4176 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4177 "dst"); 4178 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4179 } 4180 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4181 r->ipsrc.neg, kif, M_GETFIB(m))) { 4182 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4183 "ip_src"); 4184 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4185 } 4186 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4187 r->ipdst.neg, kif, M_GETFIB(m))) { 4188 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4189 "ip_dst"); 4190 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4191 } 4192 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4193 mtag ? mtag->tag : 0)) { 4194 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4195 "match_tag"); 4196 r = TAILQ_NEXT(r, entries); 4197 } 4198 else { 4199 if (r->tag) 4200 tag = r->tag; 4201 if (r->anchor == NULL) { 4202 /* Rule matches */ 4203 rm = r; 4204 4205 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4206 4207 if (r->quick) 4208 break; 4209 4210 r = TAILQ_NEXT(r, entries); 4211 } else { 4212 pf_step_into_keth_anchor(anchor_stack, &asd, 4213 &ruleset, &r, &a, &match); 4214 } 4215 } 4216 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4217 &ruleset, &r, &a, &match)) 4218 break; 4219 } 4220 4221 r = rm; 4222 4223 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4224 4225 /* Default to pass. */ 4226 if (r == NULL) { 4227 PF_RULES_RUNLOCK(); 4228 return (PF_PASS); 4229 } 4230 4231 /* Execute action. */ 4232 counter_u64_add(r->packets[dir == PF_OUT], 1); 4233 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4234 pf_update_timestamp(r); 4235 4236 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4237 if (r->action == PF_DROP) { 4238 PF_RULES_RUNLOCK(); 4239 return (PF_DROP); 4240 } 4241 4242 if (tag > 0) { 4243 if (mtag == NULL) 4244 mtag = pf_get_mtag(m); 4245 if (mtag == NULL) { 4246 PF_RULES_RUNLOCK(); 4247 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4248 return (PF_DROP); 4249 } 4250 mtag->tag = tag; 4251 } 4252 4253 if (r->qid != 0) { 4254 if (mtag == NULL) 4255 mtag = pf_get_mtag(m); 4256 if (mtag == NULL) { 4257 PF_RULES_RUNLOCK(); 4258 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4259 return (PF_DROP); 4260 } 4261 mtag->qid = r->qid; 4262 } 4263 4264 action = r->action; 4265 bridge_to = r->bridge_to; 4266 4267 /* Dummynet */ 4268 if (r->dnpipe) { 4269 struct ip_fw_args dnflow; 4270 4271 /* Drop packet if dummynet is not loaded. */ 4272 if (ip_dn_io_ptr == NULL) { 4273 PF_RULES_RUNLOCK(); 4274 m_freem(m); 4275 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4276 return (PF_DROP); 4277 } 4278 if (mtag == NULL) 4279 mtag = pf_get_mtag(m); 4280 if (mtag == NULL) { 4281 PF_RULES_RUNLOCK(); 4282 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4283 return (PF_DROP); 4284 } 4285 4286 bzero(&dnflow, sizeof(dnflow)); 4287 4288 /* We don't have port numbers here, so we set 0. That means 4289 * that we'll be somewhat limited in distinguishing flows (i.e. 4290 * only based on IP addresses, not based on port numbers), but 4291 * it's better than nothing. */ 4292 dnflow.f_id.dst_port = 0; 4293 dnflow.f_id.src_port = 0; 4294 dnflow.f_id.proto = 0; 4295 4296 dnflow.rule.info = r->dnpipe; 4297 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4298 if (r->dnflags & PFRULE_DN_IS_PIPE) 4299 dnflow.rule.info |= IPFW_IS_PIPE; 4300 4301 dnflow.f_id.extra = dnflow.rule.info; 4302 4303 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4304 dnflow.flags |= IPFW_ARGS_ETHER; 4305 dnflow.ifp = kif->pfik_ifp; 4306 4307 switch (af) { 4308 case AF_INET: 4309 dnflow.f_id.addr_type = 4; 4310 dnflow.f_id.src_ip = src->v4.s_addr; 4311 dnflow.f_id.dst_ip = dst->v4.s_addr; 4312 break; 4313 case AF_INET6: 4314 dnflow.flags |= IPFW_ARGS_IP6; 4315 dnflow.f_id.addr_type = 6; 4316 dnflow.f_id.src_ip6 = src->v6; 4317 dnflow.f_id.dst_ip6 = dst->v6; 4318 break; 4319 } 4320 4321 PF_RULES_RUNLOCK(); 4322 4323 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4324 ip_dn_io_ptr(m0, &dnflow); 4325 if (*m0 != NULL) 4326 mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 4327 } else { 4328 PF_RULES_RUNLOCK(); 4329 } 4330 4331 if (action == PF_PASS && bridge_to) { 4332 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4333 *m0 = NULL; /* We've eaten the packet. */ 4334 } 4335 4336 return (action); 4337 } 4338 4339 static int 4340 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4341 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4342 struct pf_kruleset **rsm, struct inpcb *inp) 4343 { 4344 struct pf_krule *nr = NULL; 4345 struct pf_addr * const saddr = pd->src; 4346 struct pf_addr * const daddr = pd->dst; 4347 sa_family_t af = pd->af; 4348 struct pf_krule *r, *a = NULL; 4349 struct pf_kruleset *ruleset = NULL; 4350 struct pf_krule_slist match_rules; 4351 struct pf_krule_item *ri; 4352 struct pf_ksrc_node *nsn = NULL; 4353 struct tcphdr *th = &pd->hdr.tcp; 4354 struct pf_state_key *sk = NULL, *nk = NULL; 4355 u_short reason; 4356 int rewrite = 0, hdrlen = 0; 4357 int tag = -1; 4358 int asd = 0; 4359 int match = 0; 4360 int state_icmp = 0; 4361 u_int16_t sport = 0, dport = 0; 4362 u_int16_t bproto_sum = 0, bip_sum = 0; 4363 u_int8_t icmptype = 0, icmpcode = 0; 4364 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4365 4366 PF_RULES_RASSERT(); 4367 4368 if (inp != NULL) { 4369 INP_LOCK_ASSERT(inp); 4370 pd->lookup.uid = inp->inp_cred->cr_uid; 4371 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4372 pd->lookup.done = 1; 4373 } 4374 4375 switch (pd->proto) { 4376 case IPPROTO_TCP: 4377 sport = th->th_sport; 4378 dport = th->th_dport; 4379 hdrlen = sizeof(*th); 4380 break; 4381 case IPPROTO_UDP: 4382 sport = pd->hdr.udp.uh_sport; 4383 dport = pd->hdr.udp.uh_dport; 4384 hdrlen = sizeof(pd->hdr.udp); 4385 break; 4386 case IPPROTO_SCTP: 4387 sport = pd->hdr.sctp.src_port; 4388 dport = pd->hdr.sctp.dest_port; 4389 hdrlen = sizeof(pd->hdr.sctp); 4390 break; 4391 #ifdef INET 4392 case IPPROTO_ICMP: 4393 if (pd->af != AF_INET) 4394 break; 4395 sport = dport = pd->hdr.icmp.icmp_id; 4396 hdrlen = sizeof(pd->hdr.icmp); 4397 icmptype = pd->hdr.icmp.icmp_type; 4398 icmpcode = pd->hdr.icmp.icmp_code; 4399 4400 if (icmptype == ICMP_UNREACH || 4401 icmptype == ICMP_SOURCEQUENCH || 4402 icmptype == ICMP_REDIRECT || 4403 icmptype == ICMP_TIMXCEED || 4404 icmptype == ICMP_PARAMPROB) 4405 state_icmp++; 4406 break; 4407 #endif /* INET */ 4408 #ifdef INET6 4409 case IPPROTO_ICMPV6: 4410 if (af != AF_INET6) 4411 break; 4412 sport = dport = pd->hdr.icmp6.icmp6_id; 4413 hdrlen = sizeof(pd->hdr.icmp6); 4414 icmptype = pd->hdr.icmp6.icmp6_type; 4415 icmpcode = pd->hdr.icmp6.icmp6_code; 4416 4417 if (icmptype == ICMP6_DST_UNREACH || 4418 icmptype == ICMP6_PACKET_TOO_BIG || 4419 icmptype == ICMP6_TIME_EXCEEDED || 4420 icmptype == ICMP6_PARAM_PROB) 4421 state_icmp++; 4422 break; 4423 #endif /* INET6 */ 4424 default: 4425 sport = dport = hdrlen = 0; 4426 break; 4427 } 4428 4429 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4430 4431 /* check packet for BINAT/NAT/RDR */ 4432 if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4433 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4434 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4435 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4436 4437 if (nr->log) { 4438 PFLOG_PACKET(kif, m, af, PFRES_MATCH, nr, a, 4439 ruleset, pd, 1); 4440 } 4441 4442 if (pd->ip_sum) 4443 bip_sum = *pd->ip_sum; 4444 4445 switch (pd->proto) { 4446 case IPPROTO_TCP: 4447 bproto_sum = th->th_sum; 4448 pd->proto_sum = &th->th_sum; 4449 4450 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4451 nk->port[pd->sidx] != sport) { 4452 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4453 &th->th_sum, &nk->addr[pd->sidx], 4454 nk->port[pd->sidx], 0, af); 4455 pd->sport = &th->th_sport; 4456 sport = th->th_sport; 4457 } 4458 4459 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4460 nk->port[pd->didx] != dport) { 4461 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4462 &th->th_sum, &nk->addr[pd->didx], 4463 nk->port[pd->didx], 0, af); 4464 dport = th->th_dport; 4465 pd->dport = &th->th_dport; 4466 } 4467 rewrite++; 4468 break; 4469 case IPPROTO_UDP: 4470 bproto_sum = pd->hdr.udp.uh_sum; 4471 pd->proto_sum = &pd->hdr.udp.uh_sum; 4472 4473 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4474 nk->port[pd->sidx] != sport) { 4475 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4476 pd->ip_sum, &pd->hdr.udp.uh_sum, 4477 &nk->addr[pd->sidx], 4478 nk->port[pd->sidx], 1, af); 4479 sport = pd->hdr.udp.uh_sport; 4480 pd->sport = &pd->hdr.udp.uh_sport; 4481 } 4482 4483 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4484 nk->port[pd->didx] != dport) { 4485 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4486 pd->ip_sum, &pd->hdr.udp.uh_sum, 4487 &nk->addr[pd->didx], 4488 nk->port[pd->didx], 1, af); 4489 dport = pd->hdr.udp.uh_dport; 4490 pd->dport = &pd->hdr.udp.uh_dport; 4491 } 4492 rewrite++; 4493 break; 4494 case IPPROTO_SCTP: { 4495 uint16_t checksum = 0; 4496 4497 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4498 nk->port[pd->sidx] != sport) { 4499 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 4500 pd->ip_sum, &checksum, 4501 &nk->addr[pd->sidx], 4502 nk->port[pd->sidx], 1, af); 4503 } 4504 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4505 nk->port[pd->didx] != dport) { 4506 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 4507 pd->ip_sum, &checksum, 4508 &nk->addr[pd->didx], 4509 nk->port[pd->didx], 1, af); 4510 } 4511 break; 4512 } 4513 #ifdef INET 4514 case IPPROTO_ICMP: 4515 nk->port[0] = nk->port[1]; 4516 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4517 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4518 nk->addr[pd->sidx].v4.s_addr, 0); 4519 4520 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4521 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4522 nk->addr[pd->didx].v4.s_addr, 0); 4523 4524 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4525 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4526 pd->hdr.icmp.icmp_cksum, sport, 4527 nk->port[1], 0); 4528 pd->hdr.icmp.icmp_id = nk->port[1]; 4529 pd->sport = &pd->hdr.icmp.icmp_id; 4530 } 4531 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4532 break; 4533 #endif /* INET */ 4534 #ifdef INET6 4535 case IPPROTO_ICMPV6: 4536 nk->port[0] = nk->port[1]; 4537 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4538 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4539 &nk->addr[pd->sidx], 0); 4540 4541 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4542 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4543 &nk->addr[pd->didx], 0); 4544 rewrite++; 4545 break; 4546 #endif /* INET */ 4547 default: 4548 switch (af) { 4549 #ifdef INET 4550 case AF_INET: 4551 if (PF_ANEQ(saddr, 4552 &nk->addr[pd->sidx], AF_INET)) 4553 pf_change_a(&saddr->v4.s_addr, 4554 pd->ip_sum, 4555 nk->addr[pd->sidx].v4.s_addr, 0); 4556 4557 if (PF_ANEQ(daddr, 4558 &nk->addr[pd->didx], AF_INET)) 4559 pf_change_a(&daddr->v4.s_addr, 4560 pd->ip_sum, 4561 nk->addr[pd->didx].v4.s_addr, 0); 4562 break; 4563 #endif /* INET */ 4564 #ifdef INET6 4565 case AF_INET6: 4566 if (PF_ANEQ(saddr, 4567 &nk->addr[pd->sidx], AF_INET6)) 4568 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4569 4570 if (PF_ANEQ(daddr, 4571 &nk->addr[pd->didx], AF_INET6)) 4572 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4573 break; 4574 #endif /* INET */ 4575 } 4576 break; 4577 } 4578 if (nr->natpass) 4579 r = NULL; 4580 pd->nat_rule = nr; 4581 } 4582 4583 SLIST_INIT(&match_rules); 4584 while (r != NULL) { 4585 pf_counter_u64_add(&r->evaluations, 1); 4586 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4587 r = r->skip[PF_SKIP_IFP].ptr; 4588 else if (r->direction && r->direction != pd->dir) 4589 r = r->skip[PF_SKIP_DIR].ptr; 4590 else if (r->af && r->af != af) 4591 r = r->skip[PF_SKIP_AF].ptr; 4592 else if (r->proto && r->proto != pd->proto) 4593 r = r->skip[PF_SKIP_PROTO].ptr; 4594 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4595 r->src.neg, kif, M_GETFIB(m))) 4596 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4597 /* tcp/udp only. port_op always 0 in other cases */ 4598 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4599 r->src.port[0], r->src.port[1], sport)) 4600 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4601 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4602 r->dst.neg, NULL, M_GETFIB(m))) 4603 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4604 /* tcp/udp only. port_op always 0 in other cases */ 4605 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4606 r->dst.port[0], r->dst.port[1], dport)) 4607 r = r->skip[PF_SKIP_DST_PORT].ptr; 4608 /* icmp only. type always 0 in other cases */ 4609 else if (r->type && r->type != icmptype + 1) 4610 r = TAILQ_NEXT(r, entries); 4611 /* icmp only. type always 0 in other cases */ 4612 else if (r->code && r->code != icmpcode + 1) 4613 r = TAILQ_NEXT(r, entries); 4614 else if (r->tos && !(r->tos == pd->tos)) 4615 r = TAILQ_NEXT(r, entries); 4616 else if (r->rule_flag & PFRULE_FRAGMENT) 4617 r = TAILQ_NEXT(r, entries); 4618 else if (pd->proto == IPPROTO_TCP && 4619 (r->flagset & th->th_flags) != r->flags) 4620 r = TAILQ_NEXT(r, entries); 4621 /* tcp/udp only. uid.op always 0 in other cases */ 4622 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4623 pf_socket_lookup(pd, m), 1)) && 4624 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4625 pd->lookup.uid)) 4626 r = TAILQ_NEXT(r, entries); 4627 /* tcp/udp only. gid.op always 0 in other cases */ 4628 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4629 pf_socket_lookup(pd, m), 1)) && 4630 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4631 pd->lookup.gid)) 4632 r = TAILQ_NEXT(r, entries); 4633 else if (r->prio && 4634 !pf_match_ieee8021q_pcp(r->prio, m)) 4635 r = TAILQ_NEXT(r, entries); 4636 else if (r->prob && 4637 r->prob <= arc4random()) 4638 r = TAILQ_NEXT(r, entries); 4639 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4640 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4641 r = TAILQ_NEXT(r, entries); 4642 else if (r->os_fingerprint != PF_OSFP_ANY && 4643 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4644 pf_osfp_fingerprint(pd, m, off, th), 4645 r->os_fingerprint))) 4646 r = TAILQ_NEXT(r, entries); 4647 else { 4648 if (r->tag) 4649 tag = r->tag; 4650 if (r->anchor == NULL) { 4651 if (r->action == PF_MATCH) { 4652 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4653 if (ri == NULL) { 4654 REASON_SET(&reason, PFRES_MEMORY); 4655 goto cleanup; 4656 } 4657 ri->r = r; 4658 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4659 pf_counter_u64_critical_enter(); 4660 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 4661 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 4662 pf_counter_u64_critical_exit(); 4663 pf_rule_to_actions(r, &pd->act); 4664 if (r->log) 4665 PFLOG_PACKET(kif, m, af, 4666 PFRES_MATCH, r, 4667 a, ruleset, pd, 1); 4668 } else { 4669 match = 1; 4670 *rm = r; 4671 *am = a; 4672 *rsm = ruleset; 4673 } 4674 if ((*rm)->quick) 4675 break; 4676 r = TAILQ_NEXT(r, entries); 4677 } else 4678 pf_step_into_anchor(anchor_stack, &asd, 4679 &ruleset, PF_RULESET_FILTER, &r, &a, 4680 &match); 4681 } 4682 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4683 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4684 break; 4685 } 4686 r = *rm; 4687 a = *am; 4688 ruleset = *rsm; 4689 4690 REASON_SET(&reason, PFRES_MATCH); 4691 4692 /* apply actions for last matching pass/block rule */ 4693 pf_rule_to_actions(r, &pd->act); 4694 4695 if (r->log) { 4696 if (rewrite) 4697 m_copyback(m, off, hdrlen, pd->hdr.any); 4698 PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1); 4699 } 4700 4701 if ((r->action == PF_DROP) && 4702 ((r->rule_flag & PFRULE_RETURNRST) || 4703 (r->rule_flag & PFRULE_RETURNICMP) || 4704 (r->rule_flag & PFRULE_RETURN))) { 4705 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4706 bip_sum, hdrlen, &reason, r->rtableid); 4707 } 4708 4709 if (r->action == PF_DROP) 4710 goto cleanup; 4711 4712 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4713 REASON_SET(&reason, PFRES_MEMORY); 4714 goto cleanup; 4715 } 4716 if (pd->act.rtableid >= 0) 4717 M_SETFIB(m, pd->act.rtableid); 4718 4719 if (!state_icmp && (r->keep_state || nr != NULL || 4720 (pd->flags & PFDESC_TCP_NORM))) { 4721 int action; 4722 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4723 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4724 hdrlen, &match_rules); 4725 if (action != PF_PASS) { 4726 if (action == PF_DROP && 4727 (r->rule_flag & PFRULE_RETURN)) 4728 pf_return(r, nr, pd, sk, off, m, th, kif, 4729 bproto_sum, bip_sum, hdrlen, &reason, 4730 pd->act.rtableid); 4731 return (action); 4732 } 4733 } else { 4734 uma_zfree(V_pf_state_key_z, sk); 4735 uma_zfree(V_pf_state_key_z, nk); 4736 } 4737 4738 /* copy back packet headers if we performed NAT operations */ 4739 if (rewrite) 4740 m_copyback(m, off, hdrlen, pd->hdr.any); 4741 4742 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4743 pd->dir == PF_OUT && 4744 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4745 /* 4746 * We want the state created, but we dont 4747 * want to send this in case a partner 4748 * firewall has to know about it to allow 4749 * replies through it. 4750 */ 4751 return (PF_DEFER); 4752 4753 return (PF_PASS); 4754 4755 cleanup: 4756 while ((ri = SLIST_FIRST(&match_rules))) { 4757 SLIST_REMOVE_HEAD(&match_rules, entry); 4758 free(ri, M_PF_RULE_ITEM); 4759 } 4760 4761 uma_zfree(V_pf_state_key_z, sk); 4762 uma_zfree(V_pf_state_key_z, nk); 4763 return (PF_DROP); 4764 } 4765 4766 static int 4767 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4768 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4769 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4770 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4771 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 4772 struct pf_krule_slist *match_rules) 4773 { 4774 struct pf_kstate *s = NULL; 4775 struct pf_ksrc_node *sn = NULL; 4776 struct tcphdr *th = &pd->hdr.tcp; 4777 u_int16_t mss = V_tcp_mssdflt; 4778 u_short reason, sn_reason; 4779 4780 /* check maximums */ 4781 if (r->max_states && 4782 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4783 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4784 REASON_SET(&reason, PFRES_MAXSTATES); 4785 goto csfailed; 4786 } 4787 /* src node for filter rule */ 4788 if ((r->rule_flag & PFRULE_SRCTRACK || 4789 r->rpool.opts & PF_POOL_STICKYADDR) && 4790 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 4791 REASON_SET(&reason, sn_reason); 4792 goto csfailed; 4793 } 4794 /* src node for translation rule */ 4795 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4796 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 4797 pd->af)) != 0 ) { 4798 REASON_SET(&reason, sn_reason); 4799 goto csfailed; 4800 } 4801 s = pf_alloc_state(M_NOWAIT); 4802 if (s == NULL) { 4803 REASON_SET(&reason, PFRES_MEMORY); 4804 goto csfailed; 4805 } 4806 s->rule.ptr = r; 4807 s->nat_rule.ptr = nr; 4808 s->anchor.ptr = a; 4809 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 4810 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 4811 4812 STATE_INC_COUNTERS(s); 4813 if (r->allow_opts) 4814 s->state_flags |= PFSTATE_ALLOWOPTS; 4815 if (r->rule_flag & PFRULE_STATESLOPPY) 4816 s->state_flags |= PFSTATE_SLOPPY; 4817 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 4818 s->state_flags |= PFSTATE_SCRUB_TCP; 4819 4820 s->act.log = pd->act.log & PF_LOG_ALL; 4821 s->sync_state = PFSYNC_S_NONE; 4822 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 4823 4824 if (nr != NULL) 4825 s->act.log |= nr->log & PF_LOG_ALL; 4826 switch (pd->proto) { 4827 case IPPROTO_TCP: 4828 s->src.seqlo = ntohl(th->th_seq); 4829 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4830 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4831 r->keep_state == PF_STATE_MODULATE) { 4832 /* Generate sequence number modulator */ 4833 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4834 0) 4835 s->src.seqdiff = 1; 4836 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4837 htonl(s->src.seqlo + s->src.seqdiff), 0); 4838 *rewrite = 1; 4839 } else 4840 s->src.seqdiff = 0; 4841 if (th->th_flags & TH_SYN) { 4842 s->src.seqhi++; 4843 s->src.wscale = pf_get_wscale(m, off, 4844 th->th_off, pd->af); 4845 } 4846 s->src.max_win = MAX(ntohs(th->th_win), 1); 4847 if (s->src.wscale & PF_WSCALE_MASK) { 4848 /* Remove scale factor from initial window */ 4849 int win = s->src.max_win; 4850 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4851 s->src.max_win = (win - 1) >> 4852 (s->src.wscale & PF_WSCALE_MASK); 4853 } 4854 if (th->th_flags & TH_FIN) 4855 s->src.seqhi++; 4856 s->dst.seqhi = 1; 4857 s->dst.max_win = 1; 4858 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4859 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4860 s->timeout = PFTM_TCP_FIRST_PACKET; 4861 atomic_add_32(&V_pf_status.states_halfopen, 1); 4862 break; 4863 case IPPROTO_UDP: 4864 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4865 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4866 s->timeout = PFTM_UDP_FIRST_PACKET; 4867 break; 4868 case IPPROTO_SCTP: 4869 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 4870 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 4871 s->timeout = PFTM_TCP_FIRST_PACKET; 4872 break; 4873 case IPPROTO_ICMP: 4874 #ifdef INET6 4875 case IPPROTO_ICMPV6: 4876 #endif 4877 s->timeout = PFTM_ICMP_FIRST_PACKET; 4878 break; 4879 default: 4880 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4881 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4882 s->timeout = PFTM_OTHER_FIRST_PACKET; 4883 } 4884 4885 if (r->rt) { 4886 /* pf_map_addr increases the reason counters */ 4887 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 4888 &s->rt_kif, NULL, &sn)) != 0) 4889 goto csfailed; 4890 s->rt = r->rt; 4891 } 4892 4893 s->creation = time_uptime; 4894 s->expire = time_uptime; 4895 4896 if (sn != NULL) 4897 s->src_node = sn; 4898 if (nsn != NULL) { 4899 /* XXX We only modify one side for now. */ 4900 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4901 s->nat_src_node = nsn; 4902 } 4903 if (pd->proto == IPPROTO_TCP) { 4904 if (s->state_flags & PFSTATE_SCRUB_TCP && 4905 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 4906 REASON_SET(&reason, PFRES_MEMORY); 4907 goto drop; 4908 } 4909 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 4910 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4911 &s->src, &s->dst, rewrite)) { 4912 /* This really shouldn't happen!!! */ 4913 DPFPRINTF(PF_DEBUG_URGENT, 4914 ("pf_normalize_tcp_stateful failed on first " 4915 "pkt\n")); 4916 goto drop; 4917 } 4918 } else if (pd->proto == IPPROTO_SCTP) { 4919 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 4920 goto drop; 4921 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 4922 goto drop; 4923 } 4924 s->direction = pd->dir; 4925 4926 /* 4927 * sk/nk could already been setup by pf_get_translation(). 4928 */ 4929 if (nr == NULL) { 4930 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4931 __func__, nr, sk, nk)); 4932 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4933 if (sk == NULL) 4934 goto csfailed; 4935 nk = sk; 4936 } else 4937 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4938 __func__, nr, sk, nk)); 4939 4940 /* Swap sk/nk for PF_OUT. */ 4941 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4942 (pd->dir == PF_IN) ? sk : nk, 4943 (pd->dir == PF_IN) ? nk : sk, s)) { 4944 REASON_SET(&reason, PFRES_STATEINS); 4945 goto drop; 4946 } else 4947 *sm = s; 4948 4949 if (tag > 0) 4950 s->tag = tag; 4951 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4952 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4953 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4954 /* undo NAT changes, if they have taken place */ 4955 if (nr != NULL) { 4956 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4957 if (pd->dir == PF_OUT) 4958 skt = s->key[PF_SK_STACK]; 4959 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4960 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4961 if (pd->sport) 4962 *pd->sport = skt->port[pd->sidx]; 4963 if (pd->dport) 4964 *pd->dport = skt->port[pd->didx]; 4965 if (pd->proto_sum) 4966 *pd->proto_sum = bproto_sum; 4967 if (pd->ip_sum) 4968 *pd->ip_sum = bip_sum; 4969 m_copyback(m, off, hdrlen, pd->hdr.any); 4970 } 4971 s->src.seqhi = htonl(arc4random()); 4972 /* Find mss option */ 4973 int rtid = M_GETFIB(m); 4974 mss = pf_get_mss(m, off, th->th_off, pd->af); 4975 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4976 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4977 s->src.mss = mss; 4978 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4979 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4980 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 4981 pd->act.rtableid); 4982 REASON_SET(&reason, PFRES_SYNPROXY); 4983 return (PF_SYNPROXY_DROP); 4984 } 4985 4986 return (PF_PASS); 4987 4988 csfailed: 4989 uma_zfree(V_pf_state_key_z, sk); 4990 uma_zfree(V_pf_state_key_z, nk); 4991 4992 if (sn != NULL) { 4993 PF_SRC_NODE_LOCK(sn); 4994 if (--sn->states == 0 && sn->expire == 0) { 4995 pf_unlink_src_node(sn); 4996 uma_zfree(V_pf_sources_z, sn); 4997 counter_u64_add( 4998 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4999 } 5000 PF_SRC_NODE_UNLOCK(sn); 5001 } 5002 5003 if (nsn != sn && nsn != NULL) { 5004 PF_SRC_NODE_LOCK(nsn); 5005 if (--nsn->states == 0 && nsn->expire == 0) { 5006 pf_unlink_src_node(nsn); 5007 uma_zfree(V_pf_sources_z, nsn); 5008 counter_u64_add( 5009 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5010 } 5011 PF_SRC_NODE_UNLOCK(nsn); 5012 } 5013 5014 drop: 5015 if (s != NULL) { 5016 pf_src_tree_remove_state(s); 5017 s->timeout = PFTM_UNLINKED; 5018 STATE_DEC_COUNTERS(s); 5019 pf_free_state(s); 5020 } 5021 5022 return (PF_DROP); 5023 } 5024 5025 static int 5026 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5027 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5028 struct pf_kruleset **rsm) 5029 { 5030 struct pf_krule *r, *a = NULL; 5031 struct pf_kruleset *ruleset = NULL; 5032 struct pf_krule_slist match_rules; 5033 struct pf_krule_item *ri; 5034 sa_family_t af = pd->af; 5035 u_short reason; 5036 int tag = -1; 5037 int asd = 0; 5038 int match = 0; 5039 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5040 5041 PF_RULES_RASSERT(); 5042 5043 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5044 SLIST_INIT(&match_rules); 5045 while (r != NULL) { 5046 pf_counter_u64_add(&r->evaluations, 1); 5047 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5048 r = r->skip[PF_SKIP_IFP].ptr; 5049 else if (r->direction && r->direction != pd->dir) 5050 r = r->skip[PF_SKIP_DIR].ptr; 5051 else if (r->af && r->af != af) 5052 r = r->skip[PF_SKIP_AF].ptr; 5053 else if (r->proto && r->proto != pd->proto) 5054 r = r->skip[PF_SKIP_PROTO].ptr; 5055 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5056 r->src.neg, kif, M_GETFIB(m))) 5057 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5058 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5059 r->dst.neg, NULL, M_GETFIB(m))) 5060 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5061 else if (r->tos && !(r->tos == pd->tos)) 5062 r = TAILQ_NEXT(r, entries); 5063 else if (r->os_fingerprint != PF_OSFP_ANY) 5064 r = TAILQ_NEXT(r, entries); 5065 else if (pd->proto == IPPROTO_UDP && 5066 (r->src.port_op || r->dst.port_op)) 5067 r = TAILQ_NEXT(r, entries); 5068 else if (pd->proto == IPPROTO_TCP && 5069 (r->src.port_op || r->dst.port_op || r->flagset)) 5070 r = TAILQ_NEXT(r, entries); 5071 else if ((pd->proto == IPPROTO_ICMP || 5072 pd->proto == IPPROTO_ICMPV6) && 5073 (r->type || r->code)) 5074 r = TAILQ_NEXT(r, entries); 5075 else if (r->prio && 5076 !pf_match_ieee8021q_pcp(r->prio, m)) 5077 r = TAILQ_NEXT(r, entries); 5078 else if (r->prob && r->prob <= 5079 (arc4random() % (UINT_MAX - 1) + 1)) 5080 r = TAILQ_NEXT(r, entries); 5081 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5082 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5083 r = TAILQ_NEXT(r, entries); 5084 else { 5085 if (r->anchor == NULL) { 5086 if (r->action == PF_MATCH) { 5087 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5088 if (ri == NULL) { 5089 REASON_SET(&reason, PFRES_MEMORY); 5090 goto cleanup; 5091 } 5092 ri->r = r; 5093 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5094 pf_counter_u64_critical_enter(); 5095 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5096 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5097 pf_counter_u64_critical_exit(); 5098 pf_rule_to_actions(r, &pd->act); 5099 if (r->log) 5100 PFLOG_PACKET(kif, m, af, 5101 PFRES_MATCH, r, 5102 a, ruleset, pd, 1); 5103 } else { 5104 match = 1; 5105 *rm = r; 5106 *am = a; 5107 *rsm = ruleset; 5108 } 5109 if ((*rm)->quick) 5110 break; 5111 r = TAILQ_NEXT(r, entries); 5112 } else 5113 pf_step_into_anchor(anchor_stack, &asd, 5114 &ruleset, PF_RULESET_FILTER, &r, &a, 5115 &match); 5116 } 5117 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5118 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5119 break; 5120 } 5121 r = *rm; 5122 a = *am; 5123 ruleset = *rsm; 5124 5125 REASON_SET(&reason, PFRES_MATCH); 5126 5127 /* apply actions for last matching pass/block rule */ 5128 pf_rule_to_actions(r, &pd->act); 5129 5130 if (r->log) 5131 PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1); 5132 5133 if (r->action != PF_PASS) 5134 return (PF_DROP); 5135 5136 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5137 REASON_SET(&reason, PFRES_MEMORY); 5138 goto cleanup; 5139 } 5140 5141 return (PF_PASS); 5142 5143 cleanup: 5144 while ((ri = SLIST_FIRST(&match_rules))) { 5145 SLIST_REMOVE_HEAD(&match_rules, entry); 5146 free(ri, M_PF_RULE_ITEM); 5147 } 5148 5149 return (PF_DROP); 5150 } 5151 5152 static int 5153 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5154 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5155 int *copyback) 5156 { 5157 struct tcphdr *th = &pd->hdr.tcp; 5158 struct pf_state_peer *src, *dst; 5159 u_int16_t win = ntohs(th->th_win); 5160 u_int32_t ack, end, seq, orig_seq; 5161 u_int8_t sws, dws, psrc, pdst; 5162 int ackskew; 5163 5164 if (pd->dir == (*state)->direction) { 5165 src = &(*state)->src; 5166 dst = &(*state)->dst; 5167 psrc = PF_PEER_SRC; 5168 pdst = PF_PEER_DST; 5169 } else { 5170 src = &(*state)->dst; 5171 dst = &(*state)->src; 5172 psrc = PF_PEER_DST; 5173 pdst = PF_PEER_SRC; 5174 } 5175 5176 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5177 sws = src->wscale & PF_WSCALE_MASK; 5178 dws = dst->wscale & PF_WSCALE_MASK; 5179 } else 5180 sws = dws = 0; 5181 5182 /* 5183 * Sequence tracking algorithm from Guido van Rooij's paper: 5184 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5185 * tcp_filtering.ps 5186 */ 5187 5188 orig_seq = seq = ntohl(th->th_seq); 5189 if (src->seqlo == 0) { 5190 /* First packet from this end. Set its state */ 5191 5192 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5193 src->scrub == NULL) { 5194 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5195 REASON_SET(reason, PFRES_MEMORY); 5196 return (PF_DROP); 5197 } 5198 } 5199 5200 /* Deferred generation of sequence number modulator */ 5201 if (dst->seqdiff && !src->seqdiff) { 5202 /* use random iss for the TCP server */ 5203 while ((src->seqdiff = arc4random() - seq) == 0) 5204 ; 5205 ack = ntohl(th->th_ack) - dst->seqdiff; 5206 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5207 src->seqdiff), 0); 5208 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5209 *copyback = 1; 5210 } else { 5211 ack = ntohl(th->th_ack); 5212 } 5213 5214 end = seq + pd->p_len; 5215 if (th->th_flags & TH_SYN) { 5216 end++; 5217 if (dst->wscale & PF_WSCALE_FLAG) { 5218 src->wscale = pf_get_wscale(m, off, th->th_off, 5219 pd->af); 5220 if (src->wscale & PF_WSCALE_FLAG) { 5221 /* Remove scale factor from initial 5222 * window */ 5223 sws = src->wscale & PF_WSCALE_MASK; 5224 win = ((u_int32_t)win + (1 << sws) - 1) 5225 >> sws; 5226 dws = dst->wscale & PF_WSCALE_MASK; 5227 } else { 5228 /* fixup other window */ 5229 dst->max_win <<= dst->wscale & 5230 PF_WSCALE_MASK; 5231 /* in case of a retrans SYN|ACK */ 5232 dst->wscale = 0; 5233 } 5234 } 5235 } 5236 if (th->th_flags & TH_FIN) 5237 end++; 5238 5239 src->seqlo = seq; 5240 if (src->state < TCPS_SYN_SENT) 5241 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5242 5243 /* 5244 * May need to slide the window (seqhi may have been set by 5245 * the crappy stack check or if we picked up the connection 5246 * after establishment) 5247 */ 5248 if (src->seqhi == 1 || 5249 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5250 src->seqhi = end + MAX(1, dst->max_win << dws); 5251 if (win > src->max_win) 5252 src->max_win = win; 5253 5254 } else { 5255 ack = ntohl(th->th_ack) - dst->seqdiff; 5256 if (src->seqdiff) { 5257 /* Modulate sequence numbers */ 5258 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5259 src->seqdiff), 0); 5260 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5261 *copyback = 1; 5262 } 5263 end = seq + pd->p_len; 5264 if (th->th_flags & TH_SYN) 5265 end++; 5266 if (th->th_flags & TH_FIN) 5267 end++; 5268 } 5269 5270 if ((th->th_flags & TH_ACK) == 0) { 5271 /* Let it pass through the ack skew check */ 5272 ack = dst->seqlo; 5273 } else if ((ack == 0 && 5274 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5275 /* broken tcp stacks do not set ack */ 5276 (dst->state < TCPS_SYN_SENT)) { 5277 /* 5278 * Many stacks (ours included) will set the ACK number in an 5279 * FIN|ACK if the SYN times out -- no sequence to ACK. 5280 */ 5281 ack = dst->seqlo; 5282 } 5283 5284 if (seq == end) { 5285 /* Ease sequencing restrictions on no data packets */ 5286 seq = src->seqlo; 5287 end = seq; 5288 } 5289 5290 ackskew = dst->seqlo - ack; 5291 5292 /* 5293 * Need to demodulate the sequence numbers in any TCP SACK options 5294 * (Selective ACK). We could optionally validate the SACK values 5295 * against the current ACK window, either forwards or backwards, but 5296 * I'm not confident that SACK has been implemented properly 5297 * everywhere. It wouldn't surprise me if several stacks accidentally 5298 * SACK too far backwards of previously ACKed data. There really aren't 5299 * any security implications of bad SACKing unless the target stack 5300 * doesn't validate the option length correctly. Someone trying to 5301 * spoof into a TCP connection won't bother blindly sending SACK 5302 * options anyway. 5303 */ 5304 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5305 if (pf_modulate_sack(m, off, pd, th, dst)) 5306 *copyback = 1; 5307 } 5308 5309 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5310 if (SEQ_GEQ(src->seqhi, end) && 5311 /* Last octet inside other's window space */ 5312 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5313 /* Retrans: not more than one window back */ 5314 (ackskew >= -MAXACKWINDOW) && 5315 /* Acking not more than one reassembled fragment backwards */ 5316 (ackskew <= (MAXACKWINDOW << sws)) && 5317 /* Acking not more than one window forward */ 5318 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5319 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 5320 (pd->flags & PFDESC_IP_REAS) == 0)) { 5321 /* Require an exact/+1 sequence match on resets when possible */ 5322 5323 if (dst->scrub || src->scrub) { 5324 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5325 *state, src, dst, copyback)) 5326 return (PF_DROP); 5327 } 5328 5329 /* update max window */ 5330 if (src->max_win < win) 5331 src->max_win = win; 5332 /* synchronize sequencing */ 5333 if (SEQ_GT(end, src->seqlo)) 5334 src->seqlo = end; 5335 /* slide the window of what the other end can send */ 5336 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5337 dst->seqhi = ack + MAX((win << sws), 1); 5338 5339 /* update states */ 5340 if (th->th_flags & TH_SYN) 5341 if (src->state < TCPS_SYN_SENT) 5342 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5343 if (th->th_flags & TH_FIN) 5344 if (src->state < TCPS_CLOSING) 5345 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5346 if (th->th_flags & TH_ACK) { 5347 if (dst->state == TCPS_SYN_SENT) { 5348 pf_set_protostate(*state, pdst, 5349 TCPS_ESTABLISHED); 5350 if (src->state == TCPS_ESTABLISHED && 5351 (*state)->src_node != NULL && 5352 pf_src_connlimit(state)) { 5353 REASON_SET(reason, PFRES_SRCLIMIT); 5354 return (PF_DROP); 5355 } 5356 } else if (dst->state == TCPS_CLOSING) 5357 pf_set_protostate(*state, pdst, 5358 TCPS_FIN_WAIT_2); 5359 } 5360 if (th->th_flags & TH_RST) 5361 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5362 5363 /* update expire time */ 5364 (*state)->expire = time_uptime; 5365 if (src->state >= TCPS_FIN_WAIT_2 && 5366 dst->state >= TCPS_FIN_WAIT_2) 5367 (*state)->timeout = PFTM_TCP_CLOSED; 5368 else if (src->state >= TCPS_CLOSING && 5369 dst->state >= TCPS_CLOSING) 5370 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5371 else if (src->state < TCPS_ESTABLISHED || 5372 dst->state < TCPS_ESTABLISHED) 5373 (*state)->timeout = PFTM_TCP_OPENING; 5374 else if (src->state >= TCPS_CLOSING || 5375 dst->state >= TCPS_CLOSING) 5376 (*state)->timeout = PFTM_TCP_CLOSING; 5377 else 5378 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5379 5380 /* Fall through to PASS packet */ 5381 5382 } else if ((dst->state < TCPS_SYN_SENT || 5383 dst->state >= TCPS_FIN_WAIT_2 || 5384 src->state >= TCPS_FIN_WAIT_2) && 5385 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5386 /* Within a window forward of the originating packet */ 5387 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5388 /* Within a window backward of the originating packet */ 5389 5390 /* 5391 * This currently handles three situations: 5392 * 1) Stupid stacks will shotgun SYNs before their peer 5393 * replies. 5394 * 2) When PF catches an already established stream (the 5395 * firewall rebooted, the state table was flushed, routes 5396 * changed...) 5397 * 3) Packets get funky immediately after the connection 5398 * closes (this should catch Solaris spurious ACK|FINs 5399 * that web servers like to spew after a close) 5400 * 5401 * This must be a little more careful than the above code 5402 * since packet floods will also be caught here. We don't 5403 * update the TTL here to mitigate the damage of a packet 5404 * flood and so the same code can handle awkward establishment 5405 * and a loosened connection close. 5406 * In the establishment case, a correct peer response will 5407 * validate the connection, go through the normal state code 5408 * and keep updating the state TTL. 5409 */ 5410 5411 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5412 printf("pf: loose state match: "); 5413 pf_print_state(*state); 5414 pf_print_flags(th->th_flags); 5415 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5416 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5417 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5418 (unsigned long long)(*state)->packets[1], 5419 pd->dir == PF_IN ? "in" : "out", 5420 pd->dir == (*state)->direction ? "fwd" : "rev"); 5421 } 5422 5423 if (dst->scrub || src->scrub) { 5424 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5425 *state, src, dst, copyback)) 5426 return (PF_DROP); 5427 } 5428 5429 /* update max window */ 5430 if (src->max_win < win) 5431 src->max_win = win; 5432 /* synchronize sequencing */ 5433 if (SEQ_GT(end, src->seqlo)) 5434 src->seqlo = end; 5435 /* slide the window of what the other end can send */ 5436 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5437 dst->seqhi = ack + MAX((win << sws), 1); 5438 5439 /* 5440 * Cannot set dst->seqhi here since this could be a shotgunned 5441 * SYN and not an already established connection. 5442 */ 5443 5444 if (th->th_flags & TH_FIN) 5445 if (src->state < TCPS_CLOSING) 5446 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5447 if (th->th_flags & TH_RST) 5448 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5449 5450 /* Fall through to PASS packet */ 5451 5452 } else { 5453 if ((*state)->dst.state == TCPS_SYN_SENT && 5454 (*state)->src.state == TCPS_SYN_SENT) { 5455 /* Send RST for state mismatches during handshake */ 5456 if (!(th->th_flags & TH_RST)) 5457 pf_send_tcp((*state)->rule.ptr, pd->af, 5458 pd->dst, pd->src, th->th_dport, 5459 th->th_sport, ntohl(th->th_ack), 0, 5460 TH_RST, 0, 0, 5461 (*state)->rule.ptr->return_ttl, true, 0, 0, 5462 (*state)->act.rtableid); 5463 src->seqlo = 0; 5464 src->seqhi = 1; 5465 src->max_win = 1; 5466 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5467 printf("pf: BAD state: "); 5468 pf_print_state(*state); 5469 pf_print_flags(th->th_flags); 5470 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5471 "pkts=%llu:%llu dir=%s,%s\n", 5472 seq, orig_seq, ack, pd->p_len, ackskew, 5473 (unsigned long long)(*state)->packets[0], 5474 (unsigned long long)(*state)->packets[1], 5475 pd->dir == PF_IN ? "in" : "out", 5476 pd->dir == (*state)->direction ? "fwd" : "rev"); 5477 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5478 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5479 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5480 ' ': '2', 5481 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5482 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5483 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5484 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5485 } 5486 REASON_SET(reason, PFRES_BADSTATE); 5487 return (PF_DROP); 5488 } 5489 5490 return (PF_PASS); 5491 } 5492 5493 static int 5494 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5495 { 5496 struct tcphdr *th = &pd->hdr.tcp; 5497 struct pf_state_peer *src, *dst; 5498 u_int8_t psrc, pdst; 5499 5500 if (pd->dir == (*state)->direction) { 5501 src = &(*state)->src; 5502 dst = &(*state)->dst; 5503 psrc = PF_PEER_SRC; 5504 pdst = PF_PEER_DST; 5505 } else { 5506 src = &(*state)->dst; 5507 dst = &(*state)->src; 5508 psrc = PF_PEER_DST; 5509 pdst = PF_PEER_SRC; 5510 } 5511 5512 if (th->th_flags & TH_SYN) 5513 if (src->state < TCPS_SYN_SENT) 5514 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5515 if (th->th_flags & TH_FIN) 5516 if (src->state < TCPS_CLOSING) 5517 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5518 if (th->th_flags & TH_ACK) { 5519 if (dst->state == TCPS_SYN_SENT) { 5520 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5521 if (src->state == TCPS_ESTABLISHED && 5522 (*state)->src_node != NULL && 5523 pf_src_connlimit(state)) { 5524 REASON_SET(reason, PFRES_SRCLIMIT); 5525 return (PF_DROP); 5526 } 5527 } else if (dst->state == TCPS_CLOSING) { 5528 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5529 } else if (src->state == TCPS_SYN_SENT && 5530 dst->state < TCPS_SYN_SENT) { 5531 /* 5532 * Handle a special sloppy case where we only see one 5533 * half of the connection. If there is a ACK after 5534 * the initial SYN without ever seeing a packet from 5535 * the destination, set the connection to established. 5536 */ 5537 pf_set_protostate(*state, PF_PEER_BOTH, 5538 TCPS_ESTABLISHED); 5539 dst->state = src->state = TCPS_ESTABLISHED; 5540 if ((*state)->src_node != NULL && 5541 pf_src_connlimit(state)) { 5542 REASON_SET(reason, PFRES_SRCLIMIT); 5543 return (PF_DROP); 5544 } 5545 } else if (src->state == TCPS_CLOSING && 5546 dst->state == TCPS_ESTABLISHED && 5547 dst->seqlo == 0) { 5548 /* 5549 * Handle the closing of half connections where we 5550 * don't see the full bidirectional FIN/ACK+ACK 5551 * handshake. 5552 */ 5553 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5554 } 5555 } 5556 if (th->th_flags & TH_RST) 5557 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5558 5559 /* update expire time */ 5560 (*state)->expire = time_uptime; 5561 if (src->state >= TCPS_FIN_WAIT_2 && 5562 dst->state >= TCPS_FIN_WAIT_2) 5563 (*state)->timeout = PFTM_TCP_CLOSED; 5564 else if (src->state >= TCPS_CLOSING && 5565 dst->state >= TCPS_CLOSING) 5566 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5567 else if (src->state < TCPS_ESTABLISHED || 5568 dst->state < TCPS_ESTABLISHED) 5569 (*state)->timeout = PFTM_TCP_OPENING; 5570 else if (src->state >= TCPS_CLOSING || 5571 dst->state >= TCPS_CLOSING) 5572 (*state)->timeout = PFTM_TCP_CLOSING; 5573 else 5574 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5575 5576 return (PF_PASS); 5577 } 5578 5579 static int 5580 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5581 { 5582 struct pf_state_key *sk = (*state)->key[pd->didx]; 5583 struct tcphdr *th = &pd->hdr.tcp; 5584 5585 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5586 if (pd->dir != (*state)->direction) { 5587 REASON_SET(reason, PFRES_SYNPROXY); 5588 return (PF_SYNPROXY_DROP); 5589 } 5590 if (th->th_flags & TH_SYN) { 5591 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5592 REASON_SET(reason, PFRES_SYNPROXY); 5593 return (PF_DROP); 5594 } 5595 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5596 pd->src, th->th_dport, th->th_sport, 5597 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5598 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 5599 (*state)->act.rtableid); 5600 REASON_SET(reason, PFRES_SYNPROXY); 5601 return (PF_SYNPROXY_DROP); 5602 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5603 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5604 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5605 REASON_SET(reason, PFRES_SYNPROXY); 5606 return (PF_DROP); 5607 } else if ((*state)->src_node != NULL && 5608 pf_src_connlimit(state)) { 5609 REASON_SET(reason, PFRES_SRCLIMIT); 5610 return (PF_DROP); 5611 } else 5612 pf_set_protostate(*state, PF_PEER_SRC, 5613 PF_TCPS_PROXY_DST); 5614 } 5615 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5616 if (pd->dir == (*state)->direction) { 5617 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5618 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5619 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5620 REASON_SET(reason, PFRES_SYNPROXY); 5621 return (PF_DROP); 5622 } 5623 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5624 if ((*state)->dst.seqhi == 1) 5625 (*state)->dst.seqhi = htonl(arc4random()); 5626 pf_send_tcp((*state)->rule.ptr, pd->af, 5627 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5628 sk->port[pd->sidx], sk->port[pd->didx], 5629 (*state)->dst.seqhi, 0, TH_SYN, 0, 5630 (*state)->src.mss, 0, false, (*state)->tag, 0, 5631 (*state)->act.rtableid); 5632 REASON_SET(reason, PFRES_SYNPROXY); 5633 return (PF_SYNPROXY_DROP); 5634 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5635 (TH_SYN|TH_ACK)) || 5636 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5637 REASON_SET(reason, PFRES_SYNPROXY); 5638 return (PF_DROP); 5639 } else { 5640 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5641 (*state)->dst.seqlo = ntohl(th->th_seq); 5642 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5643 pd->src, th->th_dport, th->th_sport, 5644 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5645 TH_ACK, (*state)->src.max_win, 0, 0, false, 5646 (*state)->tag, 0, (*state)->act.rtableid); 5647 pf_send_tcp((*state)->rule.ptr, pd->af, 5648 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5649 sk->port[pd->sidx], sk->port[pd->didx], 5650 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5651 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 5652 (*state)->act.rtableid); 5653 (*state)->src.seqdiff = (*state)->dst.seqhi - 5654 (*state)->src.seqlo; 5655 (*state)->dst.seqdiff = (*state)->src.seqhi - 5656 (*state)->dst.seqlo; 5657 (*state)->src.seqhi = (*state)->src.seqlo + 5658 (*state)->dst.max_win; 5659 (*state)->dst.seqhi = (*state)->dst.seqlo + 5660 (*state)->src.max_win; 5661 (*state)->src.wscale = (*state)->dst.wscale = 0; 5662 pf_set_protostate(*state, PF_PEER_BOTH, 5663 TCPS_ESTABLISHED); 5664 REASON_SET(reason, PFRES_SYNPROXY); 5665 return (PF_SYNPROXY_DROP); 5666 } 5667 } 5668 5669 return (PF_PASS); 5670 } 5671 5672 static int 5673 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 5674 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5675 u_short *reason) 5676 { 5677 struct pf_state_key_cmp key; 5678 struct tcphdr *th = &pd->hdr.tcp; 5679 int copyback = 0; 5680 int action; 5681 struct pf_state_peer *src, *dst; 5682 5683 bzero(&key, sizeof(key)); 5684 key.af = pd->af; 5685 key.proto = IPPROTO_TCP; 5686 if (pd->dir == PF_IN) { /* wire side, straight */ 5687 PF_ACPY(&key.addr[0], pd->src, key.af); 5688 PF_ACPY(&key.addr[1], pd->dst, key.af); 5689 key.port[0] = th->th_sport; 5690 key.port[1] = th->th_dport; 5691 } else { /* stack side, reverse */ 5692 PF_ACPY(&key.addr[1], pd->src, key.af); 5693 PF_ACPY(&key.addr[0], pd->dst, key.af); 5694 key.port[1] = th->th_sport; 5695 key.port[0] = th->th_dport; 5696 } 5697 5698 STATE_LOOKUP(kif, &key, *state, pd); 5699 5700 if (pd->dir == (*state)->direction) { 5701 src = &(*state)->src; 5702 dst = &(*state)->dst; 5703 } else { 5704 src = &(*state)->dst; 5705 dst = &(*state)->src; 5706 } 5707 5708 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5709 return (action); 5710 5711 if (dst->state >= TCPS_FIN_WAIT_2 && 5712 src->state >= TCPS_FIN_WAIT_2 && 5713 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 5714 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 5715 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 5716 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5717 printf("pf: state reuse "); 5718 pf_print_state(*state); 5719 pf_print_flags(th->th_flags); 5720 printf("\n"); 5721 } 5722 /* XXX make sure it's the same direction ?? */ 5723 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5724 pf_unlink_state(*state); 5725 *state = NULL; 5726 return (PF_DROP); 5727 } 5728 5729 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5730 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5731 return (PF_DROP); 5732 } else { 5733 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5734 ©back) == PF_DROP) 5735 return (PF_DROP); 5736 } 5737 5738 /* translate source/destination address, if necessary */ 5739 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5740 struct pf_state_key *nk = (*state)->key[pd->didx]; 5741 5742 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5743 nk->port[pd->sidx] != th->th_sport) 5744 pf_change_ap(m, pd->src, &th->th_sport, 5745 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5746 nk->port[pd->sidx], 0, pd->af); 5747 5748 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5749 nk->port[pd->didx] != th->th_dport) 5750 pf_change_ap(m, pd->dst, &th->th_dport, 5751 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5752 nk->port[pd->didx], 0, pd->af); 5753 copyback = 1; 5754 } 5755 5756 /* Copyback sequence modulation or stateful scrub changes if needed */ 5757 if (copyback) 5758 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5759 5760 return (PF_PASS); 5761 } 5762 5763 static int 5764 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 5765 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5766 { 5767 struct pf_state_peer *src, *dst; 5768 struct pf_state_key_cmp key; 5769 struct udphdr *uh = &pd->hdr.udp; 5770 uint8_t psrc, pdst; 5771 5772 bzero(&key, sizeof(key)); 5773 key.af = pd->af; 5774 key.proto = IPPROTO_UDP; 5775 if (pd->dir == PF_IN) { /* wire side, straight */ 5776 PF_ACPY(&key.addr[0], pd->src, key.af); 5777 PF_ACPY(&key.addr[1], pd->dst, key.af); 5778 key.port[0] = uh->uh_sport; 5779 key.port[1] = uh->uh_dport; 5780 } else { /* stack side, reverse */ 5781 PF_ACPY(&key.addr[1], pd->src, key.af); 5782 PF_ACPY(&key.addr[0], pd->dst, key.af); 5783 key.port[1] = uh->uh_sport; 5784 key.port[0] = uh->uh_dport; 5785 } 5786 5787 STATE_LOOKUP(kif, &key, *state, pd); 5788 5789 if (pd->dir == (*state)->direction) { 5790 src = &(*state)->src; 5791 dst = &(*state)->dst; 5792 psrc = PF_PEER_SRC; 5793 pdst = PF_PEER_DST; 5794 } else { 5795 src = &(*state)->dst; 5796 dst = &(*state)->src; 5797 psrc = PF_PEER_DST; 5798 pdst = PF_PEER_SRC; 5799 } 5800 5801 /* update states */ 5802 if (src->state < PFUDPS_SINGLE) 5803 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5804 if (dst->state == PFUDPS_SINGLE) 5805 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5806 5807 /* update expire time */ 5808 (*state)->expire = time_uptime; 5809 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5810 (*state)->timeout = PFTM_UDP_MULTIPLE; 5811 else 5812 (*state)->timeout = PFTM_UDP_SINGLE; 5813 5814 /* translate source/destination address, if necessary */ 5815 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5816 struct pf_state_key *nk = (*state)->key[pd->didx]; 5817 5818 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5819 nk->port[pd->sidx] != uh->uh_sport) 5820 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5821 &uh->uh_sum, &nk->addr[pd->sidx], 5822 nk->port[pd->sidx], 1, pd->af); 5823 5824 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5825 nk->port[pd->didx] != uh->uh_dport) 5826 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5827 &uh->uh_sum, &nk->addr[pd->didx], 5828 nk->port[pd->didx], 1, pd->af); 5829 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5830 } 5831 5832 return (PF_PASS); 5833 } 5834 5835 static int 5836 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 5837 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5838 { 5839 struct pf_state_key_cmp key; 5840 struct pf_state_peer *src; //, *dst; 5841 struct sctphdr *sh = &pd->hdr.sctp; 5842 u_int8_t psrc; //, pdst; 5843 5844 bzero(&key, sizeof(key)); 5845 key.af = pd->af; 5846 key.proto = IPPROTO_SCTP; 5847 if (pd->dir == PF_IN) { /* wire side, straight */ 5848 PF_ACPY(&key.addr[0], pd->src, key.af); 5849 PF_ACPY(&key.addr[1], pd->dst, key.af); 5850 key.port[0] = sh->src_port; 5851 key.port[1] = sh->dest_port; 5852 } else { /* stack side, reverse */ 5853 PF_ACPY(&key.addr[1], pd->src, key.af); 5854 PF_ACPY(&key.addr[0], pd->dst, key.af); 5855 key.port[1] = sh->src_port; 5856 key.port[0] = sh->dest_port; 5857 } 5858 5859 STATE_LOOKUP(kif, &key, *state, pd); 5860 5861 if (pd->dir == (*state)->direction) { 5862 src = &(*state)->src; 5863 psrc = PF_PEER_SRC; 5864 } else { 5865 src = &(*state)->dst; 5866 psrc = PF_PEER_DST; 5867 } 5868 5869 /* Track state. */ 5870 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 5871 if (src->state < SCTP_COOKIE_WAIT) { 5872 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 5873 (*state)->timeout = PFTM_TCP_OPENING; 5874 } 5875 } 5876 if (pd->sctp_flags & PFDESC_SCTP_COOKIE) { 5877 if (src->state < SCTP_ESTABLISHED) { 5878 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 5879 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5880 } 5881 } 5882 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 5883 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5884 if (src->state < SCTP_SHUTDOWN_PENDING) { 5885 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 5886 (*state)->timeout = PFTM_TCP_CLOSING; 5887 } 5888 } 5889 5890 if (src->scrub != NULL) { 5891 if (src->scrub->pfss_v_tag == 0) { 5892 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 5893 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 5894 return (PF_DROP); 5895 } 5896 5897 (*state)->expire = time_uptime; 5898 5899 /* translate source/destination address, if necessary */ 5900 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5901 uint16_t checksum = 0; 5902 struct pf_state_key *nk = (*state)->key[pd->didx]; 5903 5904 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5905 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 5906 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 5907 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 5908 nk->port[pd->sidx], 1, pd->af); 5909 } 5910 5911 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5912 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 5913 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 5914 pd->ip_sum, &checksum, &nk->addr[pd->didx], 5915 nk->port[pd->didx], 1, pd->af); 5916 } 5917 } 5918 5919 return (PF_PASS); 5920 } 5921 5922 static void 5923 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 5924 struct pf_kstate *s) 5925 { 5926 struct pf_sctp_multihome_job *j, *tmp; 5927 int action __unused; 5928 struct pf_kstate *sm = NULL; 5929 struct pf_krule *ra = NULL; 5930 struct pf_krule *r = &V_pf_default_rule; 5931 struct pf_kruleset *rs = NULL; 5932 5933 PF_RULES_RLOCK_TRACKER; 5934 5935 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 5936 switch (j->op) { 5937 case SCTP_ADD_IP_ADDRESS: { 5938 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 5939 PF_RULES_RLOCK(); 5940 action = pf_test_rule(&r, &sm, kif, 5941 j->m, off, &j->pd, &ra, &rs, NULL); 5942 PF_RULES_RUNLOCK(); 5943 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, action); 5944 if (sm) { 5945 /* Inherit v_tag values. */ 5946 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_flags; 5947 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_flags; 5948 PF_STATE_UNLOCK(sm); 5949 } 5950 break; 5951 } 5952 case SCTP_DEL_IP_ADDRESS: { 5953 struct pf_state_key_cmp key; 5954 uint8_t psrc; 5955 5956 bzero(&key, sizeof(key)); 5957 key.af = j->pd.af; 5958 key.proto = IPPROTO_SCTP; 5959 if (j->pd.dir == PF_IN) { /* wire side, straight */ 5960 PF_ACPY(&key.addr[0], j->pd.src, key.af); 5961 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 5962 key.port[0] = j->pd.hdr.sctp.src_port; 5963 key.port[1] = j->pd.hdr.sctp.dest_port; 5964 } else { /* stack side, reverse */ 5965 PF_ACPY(&key.addr[1], j->pd.src, key.af); 5966 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 5967 key.port[1] = j->pd.hdr.sctp.src_port; 5968 key.port[0] = j->pd.hdr.sctp.dest_port; 5969 } 5970 5971 sm = pf_find_state(kif, &key, j->pd.dir); 5972 if (sm != NULL) { 5973 PF_STATE_LOCK_ASSERT(sm); 5974 if (j->pd.dir == sm->direction) { 5975 psrc = PF_PEER_SRC; 5976 } else { 5977 psrc = PF_PEER_DST; 5978 } 5979 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 5980 sm->timeout = PFTM_TCP_CLOSING; 5981 PF_STATE_UNLOCK(sm); 5982 } 5983 break; 5984 default: 5985 panic("Unknown op %#x", j->op); 5986 } 5987 } 5988 5989 free(j, M_PFTEMP); 5990 } 5991 } 5992 5993 static int 5994 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 5995 struct pfi_kkif *kif, int op) 5996 { 5997 int off = 0; 5998 struct pf_sctp_multihome_job *job; 5999 6000 while (off < len) { 6001 struct sctp_paramhdr h; 6002 6003 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6004 pd->af)) 6005 return (PF_DROP); 6006 6007 /* Parameters are at least 4 bytes. */ 6008 if (ntohs(h.param_length) < 4) 6009 return (PF_DROP); 6010 6011 switch (ntohs(h.param_type)) { 6012 case SCTP_IPV4_ADDRESS: { 6013 struct in_addr t; 6014 6015 if (ntohs(h.param_length) != 6016 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6017 return (PF_DROP); 6018 6019 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6020 NULL, NULL, pd->af)) 6021 return (PF_DROP); 6022 6023 /* 6024 * Avoid duplicating states. We'll already have 6025 * created a state based on the source address of 6026 * the packet, but SCTP endpoints may also list this 6027 * address again in the INIT(_ACK) parameters. 6028 */ 6029 if (t.s_addr == pd->src->v4.s_addr) 6030 break; 6031 6032 if (in_nullhost(t)) 6033 t.s_addr = pd->src->v4.s_addr; 6034 6035 /* 6036 * We hold the state lock (idhash) here, which means 6037 * that we can't acquire the keyhash, or we'll get a 6038 * LOR (and potentially double-lock things too). We also 6039 * can't release the state lock here, so instead we'll 6040 * enqueue this for async handling. 6041 * There's a relatively small race here, in that a 6042 * packet using the new addresses could arrive already, 6043 * but that's just though luck for it. 6044 */ 6045 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6046 if (! job) 6047 return (PF_DROP); 6048 6049 memcpy(&job->pd, pd, sizeof(*pd)); 6050 6051 // New source address! 6052 memcpy(&job->src, &t, sizeof(t)); 6053 job->pd.src = &job->src; 6054 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6055 job->pd.dst = &job->dst; 6056 job->m = m; 6057 job->op = op; 6058 6059 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6060 break; 6061 } 6062 #ifdef INET6 6063 case SCTP_IPV6_ADDRESS: { 6064 struct in6_addr t; 6065 6066 if (ntohs(h.param_length) != 6067 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6068 return (PF_DROP); 6069 6070 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6071 NULL, NULL, pd->af)) 6072 return (PF_DROP); 6073 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6074 break; 6075 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6076 memcpy(&t, &pd->src->v6, sizeof(t)); 6077 6078 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6079 if (! job) 6080 return (PF_DROP); 6081 6082 memcpy(&job->pd, pd, sizeof(*pd)); 6083 memcpy(&job->src, &t, sizeof(t)); 6084 job->pd.src = &job->src; 6085 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6086 job->pd.dst = &job->dst; 6087 job->m = m; 6088 job->op = op; 6089 6090 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6091 break; 6092 } 6093 #endif 6094 case SCTP_ADD_IP_ADDRESS: { 6095 int ret; 6096 struct sctp_asconf_paramhdr ah; 6097 6098 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6099 NULL, NULL, pd->af)) 6100 return (PF_DROP); 6101 6102 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6103 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6104 SCTP_ADD_IP_ADDRESS); 6105 if (ret != PF_PASS) 6106 return (ret); 6107 break; 6108 } 6109 case SCTP_DEL_IP_ADDRESS: { 6110 int ret; 6111 struct sctp_asconf_paramhdr ah; 6112 6113 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6114 NULL, NULL, pd->af)) 6115 return (PF_DROP); 6116 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6117 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6118 SCTP_DEL_IP_ADDRESS); 6119 if (ret != PF_PASS) 6120 return (ret); 6121 break; 6122 } 6123 default: 6124 break; 6125 } 6126 6127 off += roundup(ntohs(h.param_length), 4); 6128 } 6129 6130 return (PF_PASS); 6131 } 6132 int 6133 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6134 struct pfi_kkif *kif) 6135 { 6136 start += sizeof(struct sctp_init_chunk); 6137 len -= sizeof(struct sctp_init_chunk); 6138 6139 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6140 } 6141 6142 int 6143 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6144 struct pf_pdesc *pd, struct pfi_kkif *kif) 6145 { 6146 start += sizeof(struct sctp_asconf_chunk); 6147 len -= sizeof(struct sctp_asconf_chunk); 6148 6149 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6150 } 6151 6152 static int 6153 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6154 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6155 { 6156 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6157 u_int16_t icmpid = 0, *icmpsum; 6158 u_int8_t icmptype, icmpcode; 6159 int state_icmp = 0; 6160 struct pf_state_key_cmp key; 6161 6162 bzero(&key, sizeof(key)); 6163 switch (pd->proto) { 6164 #ifdef INET 6165 case IPPROTO_ICMP: 6166 icmptype = pd->hdr.icmp.icmp_type; 6167 icmpcode = pd->hdr.icmp.icmp_code; 6168 icmpid = pd->hdr.icmp.icmp_id; 6169 icmpsum = &pd->hdr.icmp.icmp_cksum; 6170 6171 if (icmptype == ICMP_UNREACH || 6172 icmptype == ICMP_SOURCEQUENCH || 6173 icmptype == ICMP_REDIRECT || 6174 icmptype == ICMP_TIMXCEED || 6175 icmptype == ICMP_PARAMPROB) 6176 state_icmp++; 6177 break; 6178 #endif /* INET */ 6179 #ifdef INET6 6180 case IPPROTO_ICMPV6: 6181 icmptype = pd->hdr.icmp6.icmp6_type; 6182 icmpcode = pd->hdr.icmp6.icmp6_code; 6183 icmpid = pd->hdr.icmp6.icmp6_id; 6184 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6185 6186 if (icmptype == ICMP6_DST_UNREACH || 6187 icmptype == ICMP6_PACKET_TOO_BIG || 6188 icmptype == ICMP6_TIME_EXCEEDED || 6189 icmptype == ICMP6_PARAM_PROB) 6190 state_icmp++; 6191 break; 6192 #endif /* INET6 */ 6193 } 6194 6195 if (!state_icmp) { 6196 /* 6197 * ICMP query/reply message not related to a TCP/UDP packet. 6198 * Search for an ICMP state. 6199 */ 6200 key.af = pd->af; 6201 key.proto = pd->proto; 6202 key.port[0] = key.port[1] = icmpid; 6203 if (pd->dir == PF_IN) { /* wire side, straight */ 6204 PF_ACPY(&key.addr[0], pd->src, key.af); 6205 PF_ACPY(&key.addr[1], pd->dst, key.af); 6206 } else { /* stack side, reverse */ 6207 PF_ACPY(&key.addr[1], pd->src, key.af); 6208 PF_ACPY(&key.addr[0], pd->dst, key.af); 6209 } 6210 6211 STATE_LOOKUP(kif, &key, *state, pd); 6212 6213 (*state)->expire = time_uptime; 6214 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6215 6216 /* translate source/destination address, if necessary */ 6217 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6218 struct pf_state_key *nk = (*state)->key[pd->didx]; 6219 6220 switch (pd->af) { 6221 #ifdef INET 6222 case AF_INET: 6223 if (PF_ANEQ(pd->src, 6224 &nk->addr[pd->sidx], AF_INET)) 6225 pf_change_a(&saddr->v4.s_addr, 6226 pd->ip_sum, 6227 nk->addr[pd->sidx].v4.s_addr, 0); 6228 6229 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6230 AF_INET)) 6231 pf_change_a(&daddr->v4.s_addr, 6232 pd->ip_sum, 6233 nk->addr[pd->didx].v4.s_addr, 0); 6234 6235 if (nk->port[0] != 6236 pd->hdr.icmp.icmp_id) { 6237 pd->hdr.icmp.icmp_cksum = 6238 pf_cksum_fixup( 6239 pd->hdr.icmp.icmp_cksum, icmpid, 6240 nk->port[pd->sidx], 0); 6241 pd->hdr.icmp.icmp_id = 6242 nk->port[pd->sidx]; 6243 } 6244 6245 m_copyback(m, off, ICMP_MINLEN, 6246 (caddr_t )&pd->hdr.icmp); 6247 break; 6248 #endif /* INET */ 6249 #ifdef INET6 6250 case AF_INET6: 6251 if (PF_ANEQ(pd->src, 6252 &nk->addr[pd->sidx], AF_INET6)) 6253 pf_change_a6(saddr, 6254 &pd->hdr.icmp6.icmp6_cksum, 6255 &nk->addr[pd->sidx], 0); 6256 6257 if (PF_ANEQ(pd->dst, 6258 &nk->addr[pd->didx], AF_INET6)) 6259 pf_change_a6(daddr, 6260 &pd->hdr.icmp6.icmp6_cksum, 6261 &nk->addr[pd->didx], 0); 6262 6263 m_copyback(m, off, sizeof(struct icmp6_hdr), 6264 (caddr_t )&pd->hdr.icmp6); 6265 break; 6266 #endif /* INET6 */ 6267 } 6268 } 6269 return (PF_PASS); 6270 6271 } else { 6272 /* 6273 * ICMP error message in response to a TCP/UDP packet. 6274 * Extract the inner TCP/UDP header and search for that state. 6275 */ 6276 6277 struct pf_pdesc pd2; 6278 bzero(&pd2, sizeof pd2); 6279 #ifdef INET 6280 struct ip h2; 6281 #endif /* INET */ 6282 #ifdef INET6 6283 struct ip6_hdr h2_6; 6284 int terminal = 0; 6285 #endif /* INET6 */ 6286 int ipoff2 = 0; 6287 int off2 = 0; 6288 6289 pd2.af = pd->af; 6290 /* Payload packet is from the opposite direction. */ 6291 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 6292 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 6293 switch (pd->af) { 6294 #ifdef INET 6295 case AF_INET: 6296 /* offset of h2 in mbuf chain */ 6297 ipoff2 = off + ICMP_MINLEN; 6298 6299 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 6300 NULL, reason, pd2.af)) { 6301 DPFPRINTF(PF_DEBUG_MISC, 6302 ("pf: ICMP error message too short " 6303 "(ip)\n")); 6304 return (PF_DROP); 6305 } 6306 /* 6307 * ICMP error messages don't refer to non-first 6308 * fragments 6309 */ 6310 if (h2.ip_off & htons(IP_OFFMASK)) { 6311 REASON_SET(reason, PFRES_FRAG); 6312 return (PF_DROP); 6313 } 6314 6315 /* offset of protocol header that follows h2 */ 6316 off2 = ipoff2 + (h2.ip_hl << 2); 6317 6318 pd2.proto = h2.ip_p; 6319 pd2.src = (struct pf_addr *)&h2.ip_src; 6320 pd2.dst = (struct pf_addr *)&h2.ip_dst; 6321 pd2.ip_sum = &h2.ip_sum; 6322 break; 6323 #endif /* INET */ 6324 #ifdef INET6 6325 case AF_INET6: 6326 ipoff2 = off + sizeof(struct icmp6_hdr); 6327 6328 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 6329 NULL, reason, pd2.af)) { 6330 DPFPRINTF(PF_DEBUG_MISC, 6331 ("pf: ICMP error message too short " 6332 "(ip6)\n")); 6333 return (PF_DROP); 6334 } 6335 pd2.proto = h2_6.ip6_nxt; 6336 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 6337 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 6338 pd2.ip_sum = NULL; 6339 off2 = ipoff2 + sizeof(h2_6); 6340 do { 6341 switch (pd2.proto) { 6342 case IPPROTO_FRAGMENT: 6343 /* 6344 * ICMPv6 error messages for 6345 * non-first fragments 6346 */ 6347 REASON_SET(reason, PFRES_FRAG); 6348 return (PF_DROP); 6349 case IPPROTO_AH: 6350 case IPPROTO_HOPOPTS: 6351 case IPPROTO_ROUTING: 6352 case IPPROTO_DSTOPTS: { 6353 /* get next header and header length */ 6354 struct ip6_ext opt6; 6355 6356 if (!pf_pull_hdr(m, off2, &opt6, 6357 sizeof(opt6), NULL, reason, 6358 pd2.af)) { 6359 DPFPRINTF(PF_DEBUG_MISC, 6360 ("pf: ICMPv6 short opt\n")); 6361 return (PF_DROP); 6362 } 6363 if (pd2.proto == IPPROTO_AH) 6364 off2 += (opt6.ip6e_len + 2) * 4; 6365 else 6366 off2 += (opt6.ip6e_len + 1) * 8; 6367 pd2.proto = opt6.ip6e_nxt; 6368 /* goto the next header */ 6369 break; 6370 } 6371 default: 6372 terminal++; 6373 break; 6374 } 6375 } while (!terminal); 6376 break; 6377 #endif /* INET6 */ 6378 } 6379 6380 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 6381 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6382 printf("pf: BAD ICMP %d:%d outer dst: ", 6383 icmptype, icmpcode); 6384 pf_print_host(pd->src, 0, pd->af); 6385 printf(" -> "); 6386 pf_print_host(pd->dst, 0, pd->af); 6387 printf(" inner src: "); 6388 pf_print_host(pd2.src, 0, pd2.af); 6389 printf(" -> "); 6390 pf_print_host(pd2.dst, 0, pd2.af); 6391 printf("\n"); 6392 } 6393 REASON_SET(reason, PFRES_BADSTATE); 6394 return (PF_DROP); 6395 } 6396 6397 switch (pd2.proto) { 6398 case IPPROTO_TCP: { 6399 struct tcphdr th; 6400 u_int32_t seq; 6401 struct pf_state_peer *src, *dst; 6402 u_int8_t dws; 6403 int copyback = 0; 6404 6405 /* 6406 * Only the first 8 bytes of the TCP header can be 6407 * expected. Don't access any TCP header fields after 6408 * th_seq, an ackskew test is not possible. 6409 */ 6410 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 6411 pd2.af)) { 6412 DPFPRINTF(PF_DEBUG_MISC, 6413 ("pf: ICMP error message too short " 6414 "(tcp)\n")); 6415 return (PF_DROP); 6416 } 6417 6418 key.af = pd2.af; 6419 key.proto = IPPROTO_TCP; 6420 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6421 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6422 key.port[pd2.sidx] = th.th_sport; 6423 key.port[pd2.didx] = th.th_dport; 6424 6425 STATE_LOOKUP(kif, &key, *state, pd); 6426 6427 if (pd->dir == (*state)->direction) { 6428 src = &(*state)->dst; 6429 dst = &(*state)->src; 6430 } else { 6431 src = &(*state)->src; 6432 dst = &(*state)->dst; 6433 } 6434 6435 if (src->wscale && dst->wscale) 6436 dws = dst->wscale & PF_WSCALE_MASK; 6437 else 6438 dws = 0; 6439 6440 /* Demodulate sequence number */ 6441 seq = ntohl(th.th_seq) - src->seqdiff; 6442 if (src->seqdiff) { 6443 pf_change_a(&th.th_seq, icmpsum, 6444 htonl(seq), 0); 6445 copyback = 1; 6446 } 6447 6448 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 6449 (!SEQ_GEQ(src->seqhi, seq) || 6450 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 6451 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6452 printf("pf: BAD ICMP %d:%d ", 6453 icmptype, icmpcode); 6454 pf_print_host(pd->src, 0, pd->af); 6455 printf(" -> "); 6456 pf_print_host(pd->dst, 0, pd->af); 6457 printf(" state: "); 6458 pf_print_state(*state); 6459 printf(" seq=%u\n", seq); 6460 } 6461 REASON_SET(reason, PFRES_BADSTATE); 6462 return (PF_DROP); 6463 } else { 6464 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6465 printf("pf: OK ICMP %d:%d ", 6466 icmptype, icmpcode); 6467 pf_print_host(pd->src, 0, pd->af); 6468 printf(" -> "); 6469 pf_print_host(pd->dst, 0, pd->af); 6470 printf(" state: "); 6471 pf_print_state(*state); 6472 printf(" seq=%u\n", seq); 6473 } 6474 } 6475 6476 /* translate source/destination address, if necessary */ 6477 if ((*state)->key[PF_SK_WIRE] != 6478 (*state)->key[PF_SK_STACK]) { 6479 struct pf_state_key *nk = 6480 (*state)->key[pd->didx]; 6481 6482 if (PF_ANEQ(pd2.src, 6483 &nk->addr[pd2.sidx], pd2.af) || 6484 nk->port[pd2.sidx] != th.th_sport) 6485 pf_change_icmp(pd2.src, &th.th_sport, 6486 daddr, &nk->addr[pd2.sidx], 6487 nk->port[pd2.sidx], NULL, 6488 pd2.ip_sum, icmpsum, 6489 pd->ip_sum, 0, pd2.af); 6490 6491 if (PF_ANEQ(pd2.dst, 6492 &nk->addr[pd2.didx], pd2.af) || 6493 nk->port[pd2.didx] != th.th_dport) 6494 pf_change_icmp(pd2.dst, &th.th_dport, 6495 saddr, &nk->addr[pd2.didx], 6496 nk->port[pd2.didx], NULL, 6497 pd2.ip_sum, icmpsum, 6498 pd->ip_sum, 0, pd2.af); 6499 copyback = 1; 6500 } 6501 6502 if (copyback) { 6503 switch (pd2.af) { 6504 #ifdef INET 6505 case AF_INET: 6506 m_copyback(m, off, ICMP_MINLEN, 6507 (caddr_t )&pd->hdr.icmp); 6508 m_copyback(m, ipoff2, sizeof(h2), 6509 (caddr_t )&h2); 6510 break; 6511 #endif /* INET */ 6512 #ifdef INET6 6513 case AF_INET6: 6514 m_copyback(m, off, 6515 sizeof(struct icmp6_hdr), 6516 (caddr_t )&pd->hdr.icmp6); 6517 m_copyback(m, ipoff2, sizeof(h2_6), 6518 (caddr_t )&h2_6); 6519 break; 6520 #endif /* INET6 */ 6521 } 6522 m_copyback(m, off2, 8, (caddr_t)&th); 6523 } 6524 6525 return (PF_PASS); 6526 break; 6527 } 6528 case IPPROTO_UDP: { 6529 struct udphdr uh; 6530 6531 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 6532 NULL, reason, pd2.af)) { 6533 DPFPRINTF(PF_DEBUG_MISC, 6534 ("pf: ICMP error message too short " 6535 "(udp)\n")); 6536 return (PF_DROP); 6537 } 6538 6539 key.af = pd2.af; 6540 key.proto = IPPROTO_UDP; 6541 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6542 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6543 key.port[pd2.sidx] = uh.uh_sport; 6544 key.port[pd2.didx] = uh.uh_dport; 6545 6546 STATE_LOOKUP(kif, &key, *state, pd); 6547 6548 /* translate source/destination address, if necessary */ 6549 if ((*state)->key[PF_SK_WIRE] != 6550 (*state)->key[PF_SK_STACK]) { 6551 struct pf_state_key *nk = 6552 (*state)->key[pd->didx]; 6553 6554 if (PF_ANEQ(pd2.src, 6555 &nk->addr[pd2.sidx], pd2.af) || 6556 nk->port[pd2.sidx] != uh.uh_sport) 6557 pf_change_icmp(pd2.src, &uh.uh_sport, 6558 daddr, &nk->addr[pd2.sidx], 6559 nk->port[pd2.sidx], &uh.uh_sum, 6560 pd2.ip_sum, icmpsum, 6561 pd->ip_sum, 1, pd2.af); 6562 6563 if (PF_ANEQ(pd2.dst, 6564 &nk->addr[pd2.didx], pd2.af) || 6565 nk->port[pd2.didx] != uh.uh_dport) 6566 pf_change_icmp(pd2.dst, &uh.uh_dport, 6567 saddr, &nk->addr[pd2.didx], 6568 nk->port[pd2.didx], &uh.uh_sum, 6569 pd2.ip_sum, icmpsum, 6570 pd->ip_sum, 1, pd2.af); 6571 6572 switch (pd2.af) { 6573 #ifdef INET 6574 case AF_INET: 6575 m_copyback(m, off, ICMP_MINLEN, 6576 (caddr_t )&pd->hdr.icmp); 6577 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6578 break; 6579 #endif /* INET */ 6580 #ifdef INET6 6581 case AF_INET6: 6582 m_copyback(m, off, 6583 sizeof(struct icmp6_hdr), 6584 (caddr_t )&pd->hdr.icmp6); 6585 m_copyback(m, ipoff2, sizeof(h2_6), 6586 (caddr_t )&h2_6); 6587 break; 6588 #endif /* INET6 */ 6589 } 6590 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 6591 } 6592 return (PF_PASS); 6593 break; 6594 } 6595 #ifdef INET 6596 case IPPROTO_ICMP: { 6597 struct icmp iih; 6598 6599 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6600 NULL, reason, pd2.af)) { 6601 DPFPRINTF(PF_DEBUG_MISC, 6602 ("pf: ICMP error message too short i" 6603 "(icmp)\n")); 6604 return (PF_DROP); 6605 } 6606 6607 key.af = pd2.af; 6608 key.proto = IPPROTO_ICMP; 6609 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6610 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6611 key.port[0] = key.port[1] = iih.icmp_id; 6612 6613 STATE_LOOKUP(kif, &key, *state, pd); 6614 6615 /* translate source/destination address, if necessary */ 6616 if ((*state)->key[PF_SK_WIRE] != 6617 (*state)->key[PF_SK_STACK]) { 6618 struct pf_state_key *nk = 6619 (*state)->key[pd->didx]; 6620 6621 if (PF_ANEQ(pd2.src, 6622 &nk->addr[pd2.sidx], pd2.af) || 6623 nk->port[pd2.sidx] != iih.icmp_id) 6624 pf_change_icmp(pd2.src, &iih.icmp_id, 6625 daddr, &nk->addr[pd2.sidx], 6626 nk->port[pd2.sidx], NULL, 6627 pd2.ip_sum, icmpsum, 6628 pd->ip_sum, 0, AF_INET); 6629 6630 if (PF_ANEQ(pd2.dst, 6631 &nk->addr[pd2.didx], pd2.af) || 6632 nk->port[pd2.didx] != iih.icmp_id) 6633 pf_change_icmp(pd2.dst, &iih.icmp_id, 6634 saddr, &nk->addr[pd2.didx], 6635 nk->port[pd2.didx], NULL, 6636 pd2.ip_sum, icmpsum, 6637 pd->ip_sum, 0, AF_INET); 6638 6639 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6640 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6641 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6642 } 6643 return (PF_PASS); 6644 break; 6645 } 6646 #endif /* INET */ 6647 #ifdef INET6 6648 case IPPROTO_ICMPV6: { 6649 struct icmp6_hdr iih; 6650 6651 if (!pf_pull_hdr(m, off2, &iih, 6652 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6653 DPFPRINTF(PF_DEBUG_MISC, 6654 ("pf: ICMP error message too short " 6655 "(icmp6)\n")); 6656 return (PF_DROP); 6657 } 6658 6659 key.af = pd2.af; 6660 key.proto = IPPROTO_ICMPV6; 6661 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6662 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6663 key.port[0] = key.port[1] = iih.icmp6_id; 6664 6665 STATE_LOOKUP(kif, &key, *state, pd); 6666 6667 /* translate source/destination address, if necessary */ 6668 if ((*state)->key[PF_SK_WIRE] != 6669 (*state)->key[PF_SK_STACK]) { 6670 struct pf_state_key *nk = 6671 (*state)->key[pd->didx]; 6672 6673 if (PF_ANEQ(pd2.src, 6674 &nk->addr[pd2.sidx], pd2.af) || 6675 nk->port[pd2.sidx] != iih.icmp6_id) 6676 pf_change_icmp(pd2.src, &iih.icmp6_id, 6677 daddr, &nk->addr[pd2.sidx], 6678 nk->port[pd2.sidx], NULL, 6679 pd2.ip_sum, icmpsum, 6680 pd->ip_sum, 0, AF_INET6); 6681 6682 if (PF_ANEQ(pd2.dst, 6683 &nk->addr[pd2.didx], pd2.af) || 6684 nk->port[pd2.didx] != iih.icmp6_id) 6685 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6686 saddr, &nk->addr[pd2.didx], 6687 nk->port[pd2.didx], NULL, 6688 pd2.ip_sum, icmpsum, 6689 pd->ip_sum, 0, AF_INET6); 6690 6691 m_copyback(m, off, sizeof(struct icmp6_hdr), 6692 (caddr_t)&pd->hdr.icmp6); 6693 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6694 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6695 (caddr_t)&iih); 6696 } 6697 return (PF_PASS); 6698 break; 6699 } 6700 #endif /* INET6 */ 6701 default: { 6702 key.af = pd2.af; 6703 key.proto = pd2.proto; 6704 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6705 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6706 key.port[0] = key.port[1] = 0; 6707 6708 STATE_LOOKUP(kif, &key, *state, pd); 6709 6710 /* translate source/destination address, if necessary */ 6711 if ((*state)->key[PF_SK_WIRE] != 6712 (*state)->key[PF_SK_STACK]) { 6713 struct pf_state_key *nk = 6714 (*state)->key[pd->didx]; 6715 6716 if (PF_ANEQ(pd2.src, 6717 &nk->addr[pd2.sidx], pd2.af)) 6718 pf_change_icmp(pd2.src, NULL, daddr, 6719 &nk->addr[pd2.sidx], 0, NULL, 6720 pd2.ip_sum, icmpsum, 6721 pd->ip_sum, 0, pd2.af); 6722 6723 if (PF_ANEQ(pd2.dst, 6724 &nk->addr[pd2.didx], pd2.af)) 6725 pf_change_icmp(pd2.dst, NULL, saddr, 6726 &nk->addr[pd2.didx], 0, NULL, 6727 pd2.ip_sum, icmpsum, 6728 pd->ip_sum, 0, pd2.af); 6729 6730 switch (pd2.af) { 6731 #ifdef INET 6732 case AF_INET: 6733 m_copyback(m, off, ICMP_MINLEN, 6734 (caddr_t)&pd->hdr.icmp); 6735 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6736 break; 6737 #endif /* INET */ 6738 #ifdef INET6 6739 case AF_INET6: 6740 m_copyback(m, off, 6741 sizeof(struct icmp6_hdr), 6742 (caddr_t )&pd->hdr.icmp6); 6743 m_copyback(m, ipoff2, sizeof(h2_6), 6744 (caddr_t )&h2_6); 6745 break; 6746 #endif /* INET6 */ 6747 } 6748 } 6749 return (PF_PASS); 6750 break; 6751 } 6752 } 6753 } 6754 } 6755 6756 static int 6757 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 6758 struct mbuf *m, struct pf_pdesc *pd) 6759 { 6760 struct pf_state_peer *src, *dst; 6761 struct pf_state_key_cmp key; 6762 uint8_t psrc, pdst; 6763 6764 bzero(&key, sizeof(key)); 6765 key.af = pd->af; 6766 key.proto = pd->proto; 6767 if (pd->dir == PF_IN) { 6768 PF_ACPY(&key.addr[0], pd->src, key.af); 6769 PF_ACPY(&key.addr[1], pd->dst, key.af); 6770 key.port[0] = key.port[1] = 0; 6771 } else { 6772 PF_ACPY(&key.addr[1], pd->src, key.af); 6773 PF_ACPY(&key.addr[0], pd->dst, key.af); 6774 key.port[1] = key.port[0] = 0; 6775 } 6776 6777 STATE_LOOKUP(kif, &key, *state, pd); 6778 6779 if (pd->dir == (*state)->direction) { 6780 src = &(*state)->src; 6781 dst = &(*state)->dst; 6782 psrc = PF_PEER_SRC; 6783 pdst = PF_PEER_DST; 6784 } else { 6785 src = &(*state)->dst; 6786 dst = &(*state)->src; 6787 psrc = PF_PEER_DST; 6788 pdst = PF_PEER_SRC; 6789 } 6790 6791 /* update states */ 6792 if (src->state < PFOTHERS_SINGLE) 6793 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 6794 if (dst->state == PFOTHERS_SINGLE) 6795 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 6796 6797 /* update expire time */ 6798 (*state)->expire = time_uptime; 6799 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 6800 (*state)->timeout = PFTM_OTHER_MULTIPLE; 6801 else 6802 (*state)->timeout = PFTM_OTHER_SINGLE; 6803 6804 /* translate source/destination address, if necessary */ 6805 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6806 struct pf_state_key *nk = (*state)->key[pd->didx]; 6807 6808 KASSERT(nk, ("%s: nk is null", __func__)); 6809 KASSERT(pd, ("%s: pd is null", __func__)); 6810 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 6811 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 6812 switch (pd->af) { 6813 #ifdef INET 6814 case AF_INET: 6815 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6816 pf_change_a(&pd->src->v4.s_addr, 6817 pd->ip_sum, 6818 nk->addr[pd->sidx].v4.s_addr, 6819 0); 6820 6821 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6822 pf_change_a(&pd->dst->v4.s_addr, 6823 pd->ip_sum, 6824 nk->addr[pd->didx].v4.s_addr, 6825 0); 6826 6827 break; 6828 #endif /* INET */ 6829 #ifdef INET6 6830 case AF_INET6: 6831 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 6832 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 6833 6834 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 6835 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 6836 #endif /* INET6 */ 6837 } 6838 } 6839 return (PF_PASS); 6840 } 6841 6842 /* 6843 * ipoff and off are measured from the start of the mbuf chain. 6844 * h must be at "ipoff" on the mbuf chain. 6845 */ 6846 void * 6847 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 6848 u_short *actionp, u_short *reasonp, sa_family_t af) 6849 { 6850 switch (af) { 6851 #ifdef INET 6852 case AF_INET: { 6853 struct ip *h = mtod(m, struct ip *); 6854 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 6855 6856 if (fragoff) { 6857 if (fragoff >= len) 6858 ACTION_SET(actionp, PF_PASS); 6859 else { 6860 ACTION_SET(actionp, PF_DROP); 6861 REASON_SET(reasonp, PFRES_FRAG); 6862 } 6863 return (NULL); 6864 } 6865 if (m->m_pkthdr.len < off + len || 6866 ntohs(h->ip_len) < off + len) { 6867 ACTION_SET(actionp, PF_DROP); 6868 REASON_SET(reasonp, PFRES_SHORT); 6869 return (NULL); 6870 } 6871 break; 6872 } 6873 #endif /* INET */ 6874 #ifdef INET6 6875 case AF_INET6: { 6876 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 6877 6878 if (m->m_pkthdr.len < off + len || 6879 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 6880 (unsigned)(off + len)) { 6881 ACTION_SET(actionp, PF_DROP); 6882 REASON_SET(reasonp, PFRES_SHORT); 6883 return (NULL); 6884 } 6885 break; 6886 } 6887 #endif /* INET6 */ 6888 } 6889 m_copydata(m, off, len, p); 6890 return (p); 6891 } 6892 6893 int 6894 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 6895 int rtableid) 6896 { 6897 struct ifnet *ifp; 6898 6899 /* 6900 * Skip check for addresses with embedded interface scope, 6901 * as they would always match anyway. 6902 */ 6903 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 6904 return (1); 6905 6906 if (af != AF_INET && af != AF_INET6) 6907 return (0); 6908 6909 /* Skip checks for ipsec interfaces */ 6910 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 6911 return (1); 6912 6913 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 6914 6915 switch (af) { 6916 #ifdef INET6 6917 case AF_INET6: 6918 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 6919 ifp)); 6920 #endif 6921 #ifdef INET 6922 case AF_INET: 6923 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 6924 ifp)); 6925 #endif 6926 } 6927 6928 return (0); 6929 } 6930 6931 #ifdef INET 6932 static void 6933 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 6934 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 6935 { 6936 struct mbuf *m0, *m1, *md; 6937 struct sockaddr_in dst; 6938 struct ip *ip; 6939 struct pfi_kkif *nkif = NULL; 6940 struct ifnet *ifp = NULL; 6941 struct pf_addr naddr; 6942 struct pf_ksrc_node *sn = NULL; 6943 int error = 0; 6944 uint16_t ip_len, ip_off; 6945 int r_rt, r_dir; 6946 6947 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 6948 6949 if (s) { 6950 r_rt = s->rt; 6951 r_dir = s->direction; 6952 } else { 6953 r_rt = r->rt; 6954 r_dir = r->direction; 6955 } 6956 6957 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 6958 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 6959 __func__)); 6960 6961 if ((pd->pf_mtag == NULL && 6962 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 6963 pd->pf_mtag->routed++ > 3) { 6964 m0 = *m; 6965 *m = NULL; 6966 goto bad_locked; 6967 } 6968 6969 if (r_rt == PF_DUPTO) { 6970 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 6971 if (s == NULL) { 6972 ifp = r->rpool.cur->kif ? 6973 r->rpool.cur->kif->pfik_ifp : NULL; 6974 } else { 6975 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6976 /* If pfsync'd */ 6977 if (ifp == NULL) 6978 ifp = r->rpool.cur->kif ? 6979 r->rpool.cur->kif->pfik_ifp : NULL; 6980 PF_STATE_UNLOCK(s); 6981 } 6982 if (ifp == oifp) { 6983 /* When the 2nd interface is not skipped */ 6984 return; 6985 } else { 6986 m0 = *m; 6987 *m = NULL; 6988 goto bad; 6989 } 6990 } else { 6991 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 6992 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6993 if (s) 6994 PF_STATE_UNLOCK(s); 6995 return; 6996 } 6997 } 6998 } else { 6999 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7000 pf_dummynet(pd, s, r, m); 7001 if (s) 7002 PF_STATE_UNLOCK(s); 7003 return; 7004 } 7005 m0 = *m; 7006 } 7007 7008 ip = mtod(m0, struct ip *); 7009 7010 bzero(&dst, sizeof(dst)); 7011 dst.sin_family = AF_INET; 7012 dst.sin_len = sizeof(dst); 7013 dst.sin_addr = ip->ip_dst; 7014 7015 bzero(&naddr, sizeof(naddr)); 7016 7017 if (s == NULL) { 7018 if (TAILQ_EMPTY(&r->rpool.list)) { 7019 DPFPRINTF(PF_DEBUG_URGENT, 7020 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7021 goto bad_locked; 7022 } 7023 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7024 &naddr, &nkif, NULL, &sn); 7025 if (!PF_AZERO(&naddr, AF_INET)) 7026 dst.sin_addr.s_addr = naddr.v4.s_addr; 7027 ifp = nkif ? nkif->pfik_ifp : NULL; 7028 } else { 7029 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7030 dst.sin_addr.s_addr = 7031 s->rt_addr.v4.s_addr; 7032 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7033 /* If pfsync'd */ 7034 if (ifp == NULL) 7035 ifp = r->rpool.cur->kif ? 7036 r->rpool.cur->kif->pfik_ifp : NULL; 7037 PF_STATE_UNLOCK(s); 7038 } 7039 7040 if (ifp == NULL) 7041 goto bad; 7042 7043 if (pd->dir == PF_IN) { 7044 if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7045 goto bad; 7046 else if (m0 == NULL) 7047 goto done; 7048 if (m0->m_len < sizeof(struct ip)) { 7049 DPFPRINTF(PF_DEBUG_URGENT, 7050 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7051 goto bad; 7052 } 7053 ip = mtod(m0, struct ip *); 7054 } 7055 7056 if (ifp->if_flags & IFF_LOOPBACK) 7057 m0->m_flags |= M_SKIP_FIREWALL; 7058 7059 ip_len = ntohs(ip->ip_len); 7060 ip_off = ntohs(ip->ip_off); 7061 7062 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7063 m0->m_pkthdr.csum_flags |= CSUM_IP; 7064 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7065 in_delayed_cksum(m0); 7066 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7067 } 7068 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7069 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7070 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7071 } 7072 7073 /* 7074 * If small enough for interface, or the interface will take 7075 * care of the fragmentation for us, we can just send directly. 7076 */ 7077 if (ip_len <= ifp->if_mtu || 7078 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7079 ip->ip_sum = 0; 7080 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7081 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7082 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7083 } 7084 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7085 7086 md = m0; 7087 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7088 if (md != NULL) 7089 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7090 goto done; 7091 } 7092 7093 /* Balk when DF bit is set or the interface didn't support TSO. */ 7094 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7095 error = EMSGSIZE; 7096 KMOD_IPSTAT_INC(ips_cantfrag); 7097 if (r_rt != PF_DUPTO) { 7098 if (s && pd->nat_rule != NULL) 7099 PACKET_UNDO_NAT(m0, pd, 7100 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7101 s); 7102 7103 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7104 ifp->if_mtu); 7105 goto done; 7106 } else 7107 goto bad; 7108 } 7109 7110 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7111 if (error) 7112 goto bad; 7113 7114 for (; m0; m0 = m1) { 7115 m1 = m0->m_nextpkt; 7116 m0->m_nextpkt = NULL; 7117 if (error == 0) { 7118 m_clrprotoflags(m0); 7119 md = m0; 7120 error = pf_dummynet_route(pd, s, r, ifp, 7121 sintosa(&dst), &md); 7122 if (md != NULL) 7123 error = (*ifp->if_output)(ifp, md, 7124 sintosa(&dst), NULL); 7125 } else 7126 m_freem(m0); 7127 } 7128 7129 if (error == 0) 7130 KMOD_IPSTAT_INC(ips_fragmented); 7131 7132 done: 7133 if (r_rt != PF_DUPTO) 7134 *m = NULL; 7135 return; 7136 7137 bad_locked: 7138 if (s) 7139 PF_STATE_UNLOCK(s); 7140 bad: 7141 m_freem(m0); 7142 goto done; 7143 } 7144 #endif /* INET */ 7145 7146 #ifdef INET6 7147 static void 7148 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7149 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7150 { 7151 struct mbuf *m0, *md; 7152 struct sockaddr_in6 dst; 7153 struct ip6_hdr *ip6; 7154 struct pfi_kkif *nkif = NULL; 7155 struct ifnet *ifp = NULL; 7156 struct pf_addr naddr; 7157 struct pf_ksrc_node *sn = NULL; 7158 int r_rt, r_dir; 7159 7160 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7161 7162 if (s) { 7163 r_rt = s->rt; 7164 r_dir = s->direction; 7165 } else { 7166 r_rt = r->rt; 7167 r_dir = r->direction; 7168 } 7169 7170 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7171 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7172 __func__)); 7173 7174 if ((pd->pf_mtag == NULL && 7175 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7176 pd->pf_mtag->routed++ > 3) { 7177 m0 = *m; 7178 *m = NULL; 7179 goto bad_locked; 7180 } 7181 7182 if (r_rt == PF_DUPTO) { 7183 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7184 if (s == NULL) { 7185 ifp = r->rpool.cur->kif ? 7186 r->rpool.cur->kif->pfik_ifp : NULL; 7187 } else { 7188 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7189 /* If pfsync'd */ 7190 if (ifp == NULL) 7191 ifp = r->rpool.cur->kif ? 7192 r->rpool.cur->kif->pfik_ifp : NULL; 7193 PF_STATE_UNLOCK(s); 7194 } 7195 if (ifp == oifp) { 7196 /* When the 2nd interface is not skipped */ 7197 return; 7198 } else { 7199 m0 = *m; 7200 *m = NULL; 7201 goto bad; 7202 } 7203 } else { 7204 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7205 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7206 if (s) 7207 PF_STATE_UNLOCK(s); 7208 return; 7209 } 7210 } 7211 } else { 7212 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7213 pf_dummynet(pd, s, r, m); 7214 if (s) 7215 PF_STATE_UNLOCK(s); 7216 return; 7217 } 7218 m0 = *m; 7219 } 7220 7221 ip6 = mtod(m0, struct ip6_hdr *); 7222 7223 bzero(&dst, sizeof(dst)); 7224 dst.sin6_family = AF_INET6; 7225 dst.sin6_len = sizeof(dst); 7226 dst.sin6_addr = ip6->ip6_dst; 7227 7228 bzero(&naddr, sizeof(naddr)); 7229 7230 if (s == NULL) { 7231 if (TAILQ_EMPTY(&r->rpool.list)) { 7232 DPFPRINTF(PF_DEBUG_URGENT, 7233 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7234 goto bad_locked; 7235 } 7236 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7237 &naddr, &nkif, NULL, &sn); 7238 if (!PF_AZERO(&naddr, AF_INET6)) 7239 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7240 &naddr, AF_INET6); 7241 ifp = nkif ? nkif->pfik_ifp : NULL; 7242 } else { 7243 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7244 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7245 &s->rt_addr, AF_INET6); 7246 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7247 /* If pfsync'd */ 7248 if (ifp == NULL) 7249 ifp = r->rpool.cur->kif ? 7250 r->rpool.cur->kif->pfik_ifp : NULL; 7251 } 7252 7253 if (s) 7254 PF_STATE_UNLOCK(s); 7255 7256 if (ifp == NULL) 7257 goto bad; 7258 7259 if (pd->dir == PF_IN) { 7260 if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7261 goto bad; 7262 else if (m0 == NULL) 7263 goto done; 7264 if (m0->m_len < sizeof(struct ip6_hdr)) { 7265 DPFPRINTF(PF_DEBUG_URGENT, 7266 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 7267 __func__)); 7268 goto bad; 7269 } 7270 ip6 = mtod(m0, struct ip6_hdr *); 7271 } 7272 7273 if (ifp->if_flags & IFF_LOOPBACK) 7274 m0->m_flags |= M_SKIP_FIREWALL; 7275 7276 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 7277 ~ifp->if_hwassist) { 7278 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 7279 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 7280 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 7281 } 7282 7283 /* 7284 * If the packet is too large for the outgoing interface, 7285 * send back an icmp6 error. 7286 */ 7287 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 7288 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 7289 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 7290 md = m0; 7291 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7292 if (md != NULL) 7293 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 7294 } 7295 else { 7296 in6_ifstat_inc(ifp, ifs6_in_toobig); 7297 if (r_rt != PF_DUPTO) { 7298 if (s && pd->nat_rule != NULL) 7299 PACKET_UNDO_NAT(m0, pd, 7300 ((caddr_t)ip6 - m0->m_data) + 7301 sizeof(struct ip6_hdr), s); 7302 7303 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 7304 } else 7305 goto bad; 7306 } 7307 7308 done: 7309 if (r_rt != PF_DUPTO) 7310 *m = NULL; 7311 return; 7312 7313 bad_locked: 7314 if (s) 7315 PF_STATE_UNLOCK(s); 7316 bad: 7317 m_freem(m0); 7318 goto done; 7319 } 7320 #endif /* INET6 */ 7321 7322 /* 7323 * FreeBSD supports cksum offloads for the following drivers. 7324 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 7325 * 7326 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 7327 * network driver performed cksum including pseudo header, need to verify 7328 * csum_data 7329 * CSUM_DATA_VALID : 7330 * network driver performed cksum, needs to additional pseudo header 7331 * cksum computation with partial csum_data(i.e. lack of H/W support for 7332 * pseudo header, for instance sk(4) and possibly gem(4)) 7333 * 7334 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 7335 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 7336 * TCP/UDP layer. 7337 * Also, set csum_data to 0xffff to force cksum validation. 7338 */ 7339 static int 7340 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 7341 { 7342 u_int16_t sum = 0; 7343 int hw_assist = 0; 7344 struct ip *ip; 7345 7346 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 7347 return (1); 7348 if (m->m_pkthdr.len < off + len) 7349 return (1); 7350 7351 switch (p) { 7352 case IPPROTO_TCP: 7353 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7354 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7355 sum = m->m_pkthdr.csum_data; 7356 } else { 7357 ip = mtod(m, struct ip *); 7358 sum = in_pseudo(ip->ip_src.s_addr, 7359 ip->ip_dst.s_addr, htonl((u_short)len + 7360 m->m_pkthdr.csum_data + IPPROTO_TCP)); 7361 } 7362 sum ^= 0xffff; 7363 ++hw_assist; 7364 } 7365 break; 7366 case IPPROTO_UDP: 7367 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7368 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7369 sum = m->m_pkthdr.csum_data; 7370 } else { 7371 ip = mtod(m, struct ip *); 7372 sum = in_pseudo(ip->ip_src.s_addr, 7373 ip->ip_dst.s_addr, htonl((u_short)len + 7374 m->m_pkthdr.csum_data + IPPROTO_UDP)); 7375 } 7376 sum ^= 0xffff; 7377 ++hw_assist; 7378 } 7379 break; 7380 case IPPROTO_ICMP: 7381 #ifdef INET6 7382 case IPPROTO_ICMPV6: 7383 #endif /* INET6 */ 7384 break; 7385 default: 7386 return (1); 7387 } 7388 7389 if (!hw_assist) { 7390 switch (af) { 7391 case AF_INET: 7392 if (p == IPPROTO_ICMP) { 7393 if (m->m_len < off) 7394 return (1); 7395 m->m_data += off; 7396 m->m_len -= off; 7397 sum = in_cksum(m, len); 7398 m->m_data -= off; 7399 m->m_len += off; 7400 } else { 7401 if (m->m_len < sizeof(struct ip)) 7402 return (1); 7403 sum = in4_cksum(m, p, off, len); 7404 } 7405 break; 7406 #ifdef INET6 7407 case AF_INET6: 7408 if (m->m_len < sizeof(struct ip6_hdr)) 7409 return (1); 7410 sum = in6_cksum(m, p, off, len); 7411 break; 7412 #endif /* INET6 */ 7413 default: 7414 return (1); 7415 } 7416 } 7417 if (sum) { 7418 switch (p) { 7419 case IPPROTO_TCP: 7420 { 7421 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 7422 break; 7423 } 7424 case IPPROTO_UDP: 7425 { 7426 KMOD_UDPSTAT_INC(udps_badsum); 7427 break; 7428 } 7429 #ifdef INET 7430 case IPPROTO_ICMP: 7431 { 7432 KMOD_ICMPSTAT_INC(icps_checksum); 7433 break; 7434 } 7435 #endif 7436 #ifdef INET6 7437 case IPPROTO_ICMPV6: 7438 { 7439 KMOD_ICMP6STAT_INC(icp6s_checksum); 7440 break; 7441 } 7442 #endif /* INET6 */ 7443 } 7444 return (1); 7445 } else { 7446 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 7447 m->m_pkthdr.csum_flags |= 7448 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 7449 m->m_pkthdr.csum_data = 0xffff; 7450 } 7451 } 7452 return (0); 7453 } 7454 7455 static bool 7456 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 7457 const struct pf_kstate *s, struct ip_fw_args *dnflow) 7458 { 7459 int dndir = r->direction; 7460 7461 if (s && dndir == PF_INOUT) { 7462 dndir = s->direction; 7463 } else if (dndir == PF_INOUT) { 7464 /* Assume primary direction. Happens when we've set dnpipe in 7465 * the ethernet level code. */ 7466 dndir = pd->dir; 7467 } 7468 7469 memset(dnflow, 0, sizeof(*dnflow)); 7470 7471 if (pd->dport != NULL) 7472 dnflow->f_id.dst_port = ntohs(*pd->dport); 7473 if (pd->sport != NULL) 7474 dnflow->f_id.src_port = ntohs(*pd->sport); 7475 7476 if (pd->dir == PF_IN) 7477 dnflow->flags |= IPFW_ARGS_IN; 7478 else 7479 dnflow->flags |= IPFW_ARGS_OUT; 7480 7481 if (pd->dir != dndir && pd->act.dnrpipe) { 7482 dnflow->rule.info = pd->act.dnrpipe; 7483 } 7484 else if (pd->dir == dndir && pd->act.dnpipe) { 7485 dnflow->rule.info = pd->act.dnpipe; 7486 } 7487 else { 7488 return (false); 7489 } 7490 7491 dnflow->rule.info |= IPFW_IS_DUMMYNET; 7492 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 7493 dnflow->rule.info |= IPFW_IS_PIPE; 7494 7495 dnflow->f_id.proto = pd->proto; 7496 dnflow->f_id.extra = dnflow->rule.info; 7497 switch (pd->af) { 7498 case AF_INET: 7499 dnflow->f_id.addr_type = 4; 7500 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 7501 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 7502 break; 7503 case AF_INET6: 7504 dnflow->flags |= IPFW_ARGS_IP6; 7505 dnflow->f_id.addr_type = 6; 7506 dnflow->f_id.src_ip6 = pd->src->v6; 7507 dnflow->f_id.dst_ip6 = pd->dst->v6; 7508 break; 7509 default: 7510 panic("Invalid AF"); 7511 break; 7512 } 7513 7514 return (true); 7515 } 7516 7517 int 7518 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7519 struct inpcb *inp) 7520 { 7521 struct pfi_kkif *kif; 7522 struct mbuf *m = *m0; 7523 7524 M_ASSERTPKTHDR(m); 7525 MPASS(ifp->if_vnet == curvnet); 7526 NET_EPOCH_ASSERT(); 7527 7528 if (!V_pf_status.running) 7529 return (PF_PASS); 7530 7531 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7532 7533 if (kif == NULL) { 7534 DPFPRINTF(PF_DEBUG_URGENT, 7535 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 7536 return (PF_DROP); 7537 } 7538 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7539 return (PF_PASS); 7540 7541 if (m->m_flags & M_SKIP_FIREWALL) 7542 return (PF_PASS); 7543 7544 /* Stateless! */ 7545 return (pf_test_eth_rule(dir, kif, m0)); 7546 } 7547 7548 static int 7549 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 7550 struct pf_krule *r, struct mbuf **m0) 7551 { 7552 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 7553 } 7554 7555 static int 7556 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 7557 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 7558 struct mbuf **m0) 7559 { 7560 NET_EPOCH_ASSERT(); 7561 7562 if (pd->act.dnpipe || pd->act.dnrpipe) { 7563 struct ip_fw_args dnflow; 7564 if (ip_dn_io_ptr == NULL) { 7565 m_freem(*m0); 7566 *m0 = NULL; 7567 return (ENOMEM); 7568 } 7569 7570 if (pd->pf_mtag == NULL && 7571 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 7572 m_freem(*m0); 7573 *m0 = NULL; 7574 return (ENOMEM); 7575 } 7576 7577 if (ifp != NULL) { 7578 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 7579 7580 pd->pf_mtag->if_index = ifp->if_index; 7581 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 7582 7583 MPASS(sa != NULL); 7584 7585 if (pd->af == AF_INET) 7586 memcpy(&pd->pf_mtag->dst, sa, 7587 sizeof(struct sockaddr_in)); 7588 else 7589 memcpy(&pd->pf_mtag->dst, sa, 7590 sizeof(struct sockaddr_in6)); 7591 } 7592 7593 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 7594 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 7595 ip_dn_io_ptr(m0, &dnflow); 7596 if (*m0 != NULL) { 7597 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7598 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7599 } 7600 } 7601 } 7602 7603 return (0); 7604 } 7605 7606 #ifdef INET 7607 int 7608 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7609 struct inpcb *inp, struct pf_rule_actions *default_actions) 7610 { 7611 struct pfi_kkif *kif; 7612 u_short action, reason = 0; 7613 struct mbuf *m = *m0; 7614 struct ip *h = NULL; 7615 struct m_tag *ipfwtag; 7616 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7617 struct pf_kstate *s = NULL; 7618 struct pf_kruleset *ruleset = NULL; 7619 struct pf_pdesc pd; 7620 int off, dirndx, use_2nd_queue = 0; 7621 uint16_t tag; 7622 uint8_t rt; 7623 7624 PF_RULES_RLOCK_TRACKER; 7625 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7626 M_ASSERTPKTHDR(m); 7627 7628 if (!V_pf_status.running) 7629 return (PF_PASS); 7630 7631 PF_RULES_RLOCK(); 7632 7633 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7634 7635 if (__predict_false(kif == NULL)) { 7636 DPFPRINTF(PF_DEBUG_URGENT, 7637 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7638 PF_RULES_RUNLOCK(); 7639 return (PF_DROP); 7640 } 7641 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7642 PF_RULES_RUNLOCK(); 7643 return (PF_PASS); 7644 } 7645 7646 if (m->m_flags & M_SKIP_FIREWALL) { 7647 PF_RULES_RUNLOCK(); 7648 return (PF_PASS); 7649 } 7650 7651 memset(&pd, 0, sizeof(pd)); 7652 TAILQ_INIT(&pd.sctp_multihome_jobs); 7653 if (default_actions != NULL) 7654 memcpy(&pd.act, default_actions, sizeof(pd.act)); 7655 pd.pf_mtag = pf_find_mtag(m); 7656 7657 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 7658 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7659 7660 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7661 pd.pf_mtag->if_idxgen); 7662 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7663 PF_RULES_RUNLOCK(); 7664 m_freem(*m0); 7665 *m0 = NULL; 7666 return (PF_PASS); 7667 } 7668 PF_RULES_RUNLOCK(); 7669 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 7670 *m0 = NULL; 7671 return (PF_PASS); 7672 } 7673 7674 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7675 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7676 pd.act.flags = pd.pf_mtag->dnflags; 7677 } 7678 7679 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7680 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 7681 /* Dummynet re-injects packets after they've 7682 * completed their delay. We've already 7683 * processed them, so pass unconditionally. */ 7684 7685 /* But only once. We may see the packet multiple times (e.g. 7686 * PFIL_IN/PFIL_OUT). */ 7687 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7688 PF_RULES_RUNLOCK(); 7689 7690 return (PF_PASS); 7691 } 7692 7693 pd.sport = pd.dport = NULL; 7694 pd.proto_sum = NULL; 7695 pd.dir = dir; 7696 pd.sidx = (dir == PF_IN) ? 0 : 1; 7697 pd.didx = (dir == PF_IN) ? 1 : 0; 7698 pd.af = AF_INET; 7699 pd.act.rtableid = -1; 7700 7701 if (__predict_false(ip_divert_ptr != NULL) && 7702 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 7703 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 7704 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 7705 if (pd.pf_mtag == NULL && 7706 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7707 action = PF_DROP; 7708 goto done; 7709 } 7710 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 7711 m_tag_delete(m, ipfwtag); 7712 } 7713 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 7714 m->m_flags |= M_FASTFWD_OURS; 7715 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 7716 } 7717 } else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 7718 /* We do IP header normalization and packet reassembly here */ 7719 action = PF_DROP; 7720 goto done; 7721 } 7722 m = *m0; /* pf_normalize messes with m0 */ 7723 h = mtod(m, struct ip *); 7724 7725 off = h->ip_hl << 2; 7726 if (off < (int)sizeof(struct ip)) { 7727 action = PF_DROP; 7728 REASON_SET(&reason, PFRES_SHORT); 7729 pd.act.log = PF_LOG_FORCE; 7730 goto done; 7731 } 7732 7733 pd.src = (struct pf_addr *)&h->ip_src; 7734 pd.dst = (struct pf_addr *)&h->ip_dst; 7735 pd.ip_sum = &h->ip_sum; 7736 pd.proto = h->ip_p; 7737 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 7738 pd.tot_len = ntohs(h->ip_len); 7739 7740 /* handle fragments that didn't get reassembled by normalization */ 7741 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 7742 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 7743 goto done; 7744 } 7745 7746 switch (h->ip_p) { 7747 case IPPROTO_TCP: { 7748 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 7749 &action, &reason, AF_INET)) { 7750 if (action != PF_PASS) 7751 pd.act.log = PF_LOG_FORCE; 7752 goto done; 7753 } 7754 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 7755 7756 pd.sport = &pd.hdr.tcp.th_sport; 7757 pd.dport = &pd.hdr.tcp.th_dport; 7758 7759 /* Respond to SYN with a syncookie. */ 7760 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 7761 pd.dir == PF_IN && pf_synflood_check(&pd)) { 7762 pf_syncookie_send(m, off, &pd); 7763 action = PF_DROP; 7764 break; 7765 } 7766 7767 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 7768 use_2nd_queue = 1; 7769 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 7770 if (action == PF_DROP) 7771 goto done; 7772 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 7773 if (action == PF_PASS) { 7774 if (V_pfsync_update_state_ptr != NULL) 7775 V_pfsync_update_state_ptr(s); 7776 r = s->rule.ptr; 7777 a = s->anchor.ptr; 7778 } else if (s == NULL) { 7779 /* Validate remote SYN|ACK, re-create original SYN if 7780 * valid. */ 7781 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 7782 TH_ACK && pf_syncookie_validate(&pd) && 7783 pd.dir == PF_IN) { 7784 struct mbuf *msyn; 7785 7786 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 7787 &pd); 7788 if (msyn == NULL) { 7789 action = PF_DROP; 7790 break; 7791 } 7792 7793 action = pf_test(dir, pflags, ifp, &msyn, inp, 7794 &pd.act); 7795 m_freem(msyn); 7796 if (action != PF_PASS) 7797 break; 7798 7799 action = pf_test_state_tcp(&s, kif, m, off, h, 7800 &pd, &reason); 7801 if (action != PF_PASS || s == NULL) { 7802 action = PF_DROP; 7803 break; 7804 } 7805 7806 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 7807 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 7808 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 7809 action = pf_synproxy(&pd, &s, &reason); 7810 break; 7811 } else { 7812 action = pf_test_rule(&r, &s, kif, m, off, &pd, 7813 &a, &ruleset, inp); 7814 } 7815 } 7816 break; 7817 } 7818 7819 case IPPROTO_UDP: { 7820 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 7821 &action, &reason, AF_INET)) { 7822 if (action != PF_PASS) 7823 pd.act.log = PF_LOG_FORCE; 7824 goto done; 7825 } 7826 pd.sport = &pd.hdr.udp.uh_sport; 7827 pd.dport = &pd.hdr.udp.uh_dport; 7828 if (pd.hdr.udp.uh_dport == 0 || 7829 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 7830 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 7831 action = PF_DROP; 7832 REASON_SET(&reason, PFRES_SHORT); 7833 goto done; 7834 } 7835 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 7836 if (action == PF_PASS) { 7837 if (V_pfsync_update_state_ptr != NULL) 7838 V_pfsync_update_state_ptr(s); 7839 r = s->rule.ptr; 7840 a = s->anchor.ptr; 7841 } else if (s == NULL) 7842 action = pf_test_rule(&r, &s, kif, m, off, &pd, 7843 &a, &ruleset, inp); 7844 break; 7845 } 7846 7847 case IPPROTO_SCTP: { 7848 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 7849 &action, &reason, AF_INET)) { 7850 if (action != PF_PASS) 7851 pd.act.log |= PF_LOG_FORCE; 7852 goto done; 7853 } 7854 pd.p_len = pd.tot_len - off; 7855 7856 pd.sport = &pd.hdr.sctp.src_port; 7857 pd.dport = &pd.hdr.sctp.dest_port; 7858 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 7859 action = PF_DROP; 7860 REASON_SET(&reason, PFRES_SHORT); 7861 goto done; 7862 } 7863 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 7864 if (action == PF_DROP) 7865 goto done; 7866 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 7867 &reason); 7868 if (action == PF_PASS) { 7869 if (V_pfsync_update_state_ptr != NULL) 7870 V_pfsync_update_state_ptr(s); 7871 r = s->rule.ptr; 7872 a = s->anchor.ptr; 7873 } else { 7874 action = pf_test_rule(&r, &s, kif, m, off, 7875 &pd, &a, &ruleset, inp); 7876 } 7877 break; 7878 } 7879 7880 case IPPROTO_ICMP: { 7881 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 7882 &action, &reason, AF_INET)) { 7883 if (action != PF_PASS) 7884 pd.act.log = PF_LOG_FORCE; 7885 goto done; 7886 } 7887 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 7888 if (action == PF_PASS) { 7889 if (V_pfsync_update_state_ptr != NULL) 7890 V_pfsync_update_state_ptr(s); 7891 r = s->rule.ptr; 7892 a = s->anchor.ptr; 7893 } else if (s == NULL) 7894 action = pf_test_rule(&r, &s, kif, m, off, &pd, 7895 &a, &ruleset, inp); 7896 break; 7897 } 7898 7899 #ifdef INET6 7900 case IPPROTO_ICMPV6: { 7901 action = PF_DROP; 7902 DPFPRINTF(PF_DEBUG_MISC, 7903 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 7904 goto done; 7905 } 7906 #endif 7907 7908 default: 7909 action = pf_test_state_other(&s, kif, m, &pd); 7910 if (action == PF_PASS) { 7911 if (V_pfsync_update_state_ptr != NULL) 7912 V_pfsync_update_state_ptr(s); 7913 r = s->rule.ptr; 7914 a = s->anchor.ptr; 7915 } else if (s == NULL) 7916 action = pf_test_rule(&r, &s, kif, m, off, &pd, 7917 &a, &ruleset, inp); 7918 break; 7919 } 7920 7921 done: 7922 PF_RULES_RUNLOCK(); 7923 if (action == PF_PASS && h->ip_hl > 5 && 7924 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7925 action = PF_DROP; 7926 REASON_SET(&reason, PFRES_IPOPTIONS); 7927 pd.act.log = PF_LOG_FORCE; 7928 DPFPRINTF(PF_DEBUG_MISC, 7929 ("pf: dropping packet with ip options\n")); 7930 } 7931 7932 if (s) { 7933 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 7934 tag = s->tag; 7935 rt = s->rt; 7936 } else { 7937 tag = r->tag; 7938 rt = r->rt; 7939 } 7940 7941 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 7942 action = PF_DROP; 7943 REASON_SET(&reason, PFRES_MEMORY); 7944 } 7945 7946 pf_scrub_ip(&m, &pd); 7947 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 7948 pf_normalize_mss(m, off, &pd); 7949 7950 if (pd.act.rtableid >= 0) 7951 M_SETFIB(m, pd.act.rtableid); 7952 7953 if (pd.act.flags & PFSTATE_SETPRIO) { 7954 if (pd.tos & IPTOS_LOWDELAY) 7955 use_2nd_queue = 1; 7956 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 7957 action = PF_DROP; 7958 REASON_SET(&reason, PFRES_MEMORY); 7959 pd.act.log = PF_LOG_FORCE; 7960 DPFPRINTF(PF_DEBUG_MISC, 7961 ("pf: failed to allocate 802.1q mtag\n")); 7962 } 7963 } 7964 7965 #ifdef ALTQ 7966 if (action == PF_PASS && pd.act.qid) { 7967 if (pd.pf_mtag == NULL && 7968 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7969 action = PF_DROP; 7970 REASON_SET(&reason, PFRES_MEMORY); 7971 } else { 7972 if (s != NULL) 7973 pd.pf_mtag->qid_hash = pf_state_hash(s); 7974 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 7975 pd.pf_mtag->qid = pd.act.pqid; 7976 else 7977 pd.pf_mtag->qid = pd.act.qid; 7978 /* Add hints for ecn. */ 7979 pd.pf_mtag->hdr = h; 7980 } 7981 } 7982 #endif /* ALTQ */ 7983 7984 /* 7985 * connections redirected to loopback should not match sockets 7986 * bound specifically to loopback due to security implications, 7987 * see tcp_input() and in_pcblookup_listen(). 7988 */ 7989 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7990 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7991 (s->nat_rule.ptr->action == PF_RDR || 7992 s->nat_rule.ptr->action == PF_BINAT) && 7993 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 7994 m->m_flags |= M_SKIP_FIREWALL; 7995 7996 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 7997 r->divert.port && !PACKET_LOOPED(&pd)) { 7998 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 7999 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 8000 if (ipfwtag != NULL) { 8001 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 8002 ntohs(r->divert.port); 8003 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 8004 8005 if (s) 8006 PF_STATE_UNLOCK(s); 8007 8008 m_tag_prepend(m, ipfwtag); 8009 if (m->m_flags & M_FASTFWD_OURS) { 8010 if (pd.pf_mtag == NULL && 8011 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8012 action = PF_DROP; 8013 REASON_SET(&reason, PFRES_MEMORY); 8014 pd.act.log = PF_LOG_FORCE; 8015 DPFPRINTF(PF_DEBUG_MISC, 8016 ("pf: failed to allocate tag\n")); 8017 } else { 8018 pd.pf_mtag->flags |= 8019 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8020 m->m_flags &= ~M_FASTFWD_OURS; 8021 } 8022 } 8023 ip_divert_ptr(*m0, dir == PF_IN); 8024 *m0 = NULL; 8025 8026 return (action); 8027 } else { 8028 /* XXX: ipfw has the same behaviour! */ 8029 action = PF_DROP; 8030 REASON_SET(&reason, PFRES_MEMORY); 8031 pd.act.log = PF_LOG_FORCE; 8032 DPFPRINTF(PF_DEBUG_MISC, 8033 ("pf: failed to allocate divert tag\n")); 8034 } 8035 } 8036 8037 if (pd.act.log) { 8038 struct pf_krule *lr; 8039 struct pf_krule_item *ri; 8040 8041 if (s != NULL && s->nat_rule.ptr != NULL && 8042 s->nat_rule.ptr->log & PF_LOG_ALL) 8043 lr = s->nat_rule.ptr; 8044 else 8045 lr = r; 8046 8047 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8048 PFLOG_PACKET(kif, m, AF_INET, reason, lr, a, ruleset, 8049 &pd, (s == NULL)); 8050 if (s) { 8051 SLIST_FOREACH(ri, &s->match_rules, entry) 8052 if (ri->r->log & PF_LOG_ALL) 8053 PFLOG_PACKET(kif, m, AF_INET, reason, 8054 ri->r, a, ruleset, &pd, 0); 8055 } 8056 } 8057 8058 pf_counter_u64_critical_enter(); 8059 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 8060 pd.tot_len); 8061 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 8062 1); 8063 8064 if (action == PF_PASS || r->action == PF_DROP) { 8065 dirndx = (dir == PF_OUT); 8066 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8067 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8068 pf_update_timestamp(r); 8069 8070 if (a != NULL) { 8071 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8072 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8073 } 8074 if (s != NULL) { 8075 struct pf_krule_item *ri; 8076 8077 if (s->nat_rule.ptr != NULL) { 8078 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8079 1); 8080 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8081 pd.tot_len); 8082 } 8083 if (s->src_node != NULL) { 8084 counter_u64_add(s->src_node->packets[dirndx], 8085 1); 8086 counter_u64_add(s->src_node->bytes[dirndx], 8087 pd.tot_len); 8088 } 8089 if (s->nat_src_node != NULL) { 8090 counter_u64_add(s->nat_src_node->packets[dirndx], 8091 1); 8092 counter_u64_add(s->nat_src_node->bytes[dirndx], 8093 pd.tot_len); 8094 } 8095 dirndx = (dir == s->direction) ? 0 : 1; 8096 s->packets[dirndx]++; 8097 s->bytes[dirndx] += pd.tot_len; 8098 SLIST_FOREACH(ri, &s->match_rules, entry) { 8099 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8100 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 8101 } 8102 } 8103 tr = r; 8104 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8105 if (nr != NULL && r == &V_pf_default_rule) 8106 tr = nr; 8107 if (tr->src.addr.type == PF_ADDR_TABLE) 8108 pfr_update_stats(tr->src.addr.p.tbl, 8109 (s == NULL) ? pd.src : 8110 &s->key[(s->direction == PF_IN)]-> 8111 addr[(s->direction == PF_OUT)], 8112 pd.af, pd.tot_len, dir == PF_OUT, 8113 r->action == PF_PASS, tr->src.neg); 8114 if (tr->dst.addr.type == PF_ADDR_TABLE) 8115 pfr_update_stats(tr->dst.addr.p.tbl, 8116 (s == NULL) ? pd.dst : 8117 &s->key[(s->direction == PF_IN)]-> 8118 addr[(s->direction == PF_IN)], 8119 pd.af, pd.tot_len, dir == PF_OUT, 8120 r->action == PF_PASS, tr->dst.neg); 8121 } 8122 pf_counter_u64_critical_exit(); 8123 8124 switch (action) { 8125 case PF_SYNPROXY_DROP: 8126 m_freem(*m0); 8127 case PF_DEFER: 8128 *m0 = NULL; 8129 action = PF_PASS; 8130 break; 8131 case PF_DROP: 8132 m_freem(*m0); 8133 *m0 = NULL; 8134 break; 8135 default: 8136 /* pf_route() returns unlocked. */ 8137 if (rt) { 8138 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 8139 goto out; 8140 } 8141 if (pf_dummynet(&pd, s, r, m0) != 0) { 8142 action = PF_DROP; 8143 REASON_SET(&reason, PFRES_MEMORY); 8144 } 8145 break; 8146 } 8147 8148 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 8149 8150 if (s) 8151 PF_STATE_UNLOCK(s); 8152 8153 out: 8154 pf_sctp_multihome_delayed(&pd, off, kif, s); 8155 8156 return (action); 8157 } 8158 #endif /* INET */ 8159 8160 #ifdef INET6 8161 int 8162 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 8163 struct pf_rule_actions *default_actions) 8164 { 8165 struct pfi_kkif *kif; 8166 u_short action, reason = 0; 8167 struct mbuf *m = *m0, *n = NULL; 8168 struct m_tag *mtag; 8169 struct ip6_hdr *h = NULL; 8170 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8171 struct pf_kstate *s = NULL; 8172 struct pf_kruleset *ruleset = NULL; 8173 struct pf_pdesc pd; 8174 int off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0; 8175 uint16_t tag; 8176 uint8_t rt; 8177 8178 PF_RULES_RLOCK_TRACKER; 8179 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8180 M_ASSERTPKTHDR(m); 8181 8182 if (!V_pf_status.running) 8183 return (PF_PASS); 8184 8185 PF_RULES_RLOCK(); 8186 8187 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8188 if (__predict_false(kif == NULL)) { 8189 DPFPRINTF(PF_DEBUG_URGENT, 8190 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 8191 PF_RULES_RUNLOCK(); 8192 return (PF_DROP); 8193 } 8194 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8195 PF_RULES_RUNLOCK(); 8196 return (PF_PASS); 8197 } 8198 8199 if (m->m_flags & M_SKIP_FIREWALL) { 8200 PF_RULES_RUNLOCK(); 8201 return (PF_PASS); 8202 } 8203 8204 memset(&pd, 0, sizeof(pd)); 8205 TAILQ_INIT(&pd.sctp_multihome_jobs); 8206 if (default_actions != NULL) 8207 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8208 pd.pf_mtag = pf_find_mtag(m); 8209 8210 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8211 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8212 8213 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8214 pd.pf_mtag->if_idxgen); 8215 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8216 PF_RULES_RUNLOCK(); 8217 m_freem(*m0); 8218 *m0 = NULL; 8219 return (PF_PASS); 8220 } 8221 PF_RULES_RUNLOCK(); 8222 nd6_output_ifp(ifp, ifp, m, 8223 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 8224 *m0 = NULL; 8225 return (PF_PASS); 8226 } 8227 8228 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8229 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8230 pd.act.flags = pd.pf_mtag->dnflags; 8231 } 8232 8233 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8234 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8235 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8236 /* Dummynet re-injects packets after they've 8237 * completed their delay. We've already 8238 * processed them, so pass unconditionally. */ 8239 PF_RULES_RUNLOCK(); 8240 return (PF_PASS); 8241 } 8242 8243 pd.sport = pd.dport = NULL; 8244 pd.ip_sum = NULL; 8245 pd.proto_sum = NULL; 8246 pd.dir = dir; 8247 pd.sidx = (dir == PF_IN) ? 0 : 1; 8248 pd.didx = (dir == PF_IN) ? 1 : 0; 8249 pd.af = AF_INET6; 8250 pd.act.rtableid = -1; 8251 8252 /* We do IP header normalization and packet reassembly here */ 8253 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 8254 action = PF_DROP; 8255 goto done; 8256 } 8257 m = *m0; /* pf_normalize messes with m0 */ 8258 h = mtod(m, struct ip6_hdr *); 8259 8260 /* 8261 * we do not support jumbogram. if we keep going, zero ip6_plen 8262 * will do something bad, so drop the packet for now. 8263 */ 8264 if (htons(h->ip6_plen) == 0) { 8265 action = PF_DROP; 8266 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 8267 goto done; 8268 } 8269 8270 pd.src = (struct pf_addr *)&h->ip6_src; 8271 pd.dst = (struct pf_addr *)&h->ip6_dst; 8272 pd.tos = IPV6_DSCP(h); 8273 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8274 8275 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8276 pd.proto = h->ip6_nxt; 8277 do { 8278 switch (pd.proto) { 8279 case IPPROTO_FRAGMENT: 8280 action = pf_test_fragment(&r, kif, m, h, &pd, &a, 8281 &ruleset); 8282 if (action == PF_DROP) 8283 REASON_SET(&reason, PFRES_FRAG); 8284 goto done; 8285 case IPPROTO_ROUTING: { 8286 struct ip6_rthdr rthdr; 8287 8288 if (rh_cnt++) { 8289 DPFPRINTF(PF_DEBUG_MISC, 8290 ("pf: IPv6 more than one rthdr\n")); 8291 action = PF_DROP; 8292 REASON_SET(&reason, PFRES_IPOPTIONS); 8293 pd.act.log = PF_LOG_FORCE; 8294 goto done; 8295 } 8296 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 8297 &reason, pd.af)) { 8298 DPFPRINTF(PF_DEBUG_MISC, 8299 ("pf: IPv6 short rthdr\n")); 8300 action = PF_DROP; 8301 REASON_SET(&reason, PFRES_SHORT); 8302 pd.act.log = PF_LOG_FORCE; 8303 goto done; 8304 } 8305 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8306 DPFPRINTF(PF_DEBUG_MISC, 8307 ("pf: IPv6 rthdr0\n")); 8308 action = PF_DROP; 8309 REASON_SET(&reason, PFRES_IPOPTIONS); 8310 pd.act.log = PF_LOG_FORCE; 8311 goto done; 8312 } 8313 /* FALLTHROUGH */ 8314 } 8315 case IPPROTO_AH: 8316 case IPPROTO_HOPOPTS: 8317 case IPPROTO_DSTOPTS: { 8318 /* get next header and header length */ 8319 struct ip6_ext opt6; 8320 8321 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 8322 NULL, &reason, pd.af)) { 8323 DPFPRINTF(PF_DEBUG_MISC, 8324 ("pf: IPv6 short opt\n")); 8325 action = PF_DROP; 8326 pd.act.log = PF_LOG_FORCE; 8327 goto done; 8328 } 8329 if (pd.proto == IPPROTO_AH) 8330 off += (opt6.ip6e_len + 2) * 4; 8331 else 8332 off += (opt6.ip6e_len + 1) * 8; 8333 pd.proto = opt6.ip6e_nxt; 8334 /* goto the next header */ 8335 break; 8336 } 8337 default: 8338 terminal++; 8339 break; 8340 } 8341 } while (!terminal); 8342 8343 /* if there's no routing header, use unmodified mbuf for checksumming */ 8344 if (!n) 8345 n = m; 8346 8347 switch (pd.proto) { 8348 case IPPROTO_TCP: { 8349 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8350 &action, &reason, AF_INET6)) { 8351 if (action != PF_PASS) 8352 pd.act.log |= PF_LOG_FORCE; 8353 goto done; 8354 } 8355 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8356 pd.sport = &pd.hdr.tcp.th_sport; 8357 pd.dport = &pd.hdr.tcp.th_dport; 8358 8359 /* Respond to SYN with a syncookie. */ 8360 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8361 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8362 pf_syncookie_send(m, off, &pd); 8363 action = PF_DROP; 8364 break; 8365 } 8366 8367 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8368 if (action == PF_DROP) 8369 goto done; 8370 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8371 if (action == PF_PASS) { 8372 if (V_pfsync_update_state_ptr != NULL) 8373 V_pfsync_update_state_ptr(s); 8374 r = s->rule.ptr; 8375 a = s->anchor.ptr; 8376 } else if (s == NULL) { 8377 /* Validate remote SYN|ACK, re-create original SYN if 8378 * valid. */ 8379 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8380 TH_ACK && pf_syncookie_validate(&pd) && 8381 pd.dir == PF_IN) { 8382 struct mbuf *msyn; 8383 8384 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 8385 off, &pd); 8386 if (msyn == NULL) { 8387 action = PF_DROP; 8388 break; 8389 } 8390 8391 action = pf_test6(dir, pflags, ifp, &msyn, inp, 8392 &pd.act); 8393 m_freem(msyn); 8394 if (action != PF_PASS) 8395 break; 8396 8397 action = pf_test_state_tcp(&s, kif, m, off, h, 8398 &pd, &reason); 8399 if (action != PF_PASS || s == NULL) { 8400 action = PF_DROP; 8401 break; 8402 } 8403 8404 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8405 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8406 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8407 8408 action = pf_synproxy(&pd, &s, &reason); 8409 break; 8410 } else { 8411 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8412 &a, &ruleset, inp); 8413 } 8414 } 8415 break; 8416 } 8417 8418 case IPPROTO_UDP: { 8419 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8420 &action, &reason, AF_INET6)) { 8421 if (action != PF_PASS) 8422 pd.act.log |= PF_LOG_FORCE; 8423 goto done; 8424 } 8425 pd.sport = &pd.hdr.udp.uh_sport; 8426 pd.dport = &pd.hdr.udp.uh_dport; 8427 if (pd.hdr.udp.uh_dport == 0 || 8428 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8429 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8430 action = PF_DROP; 8431 REASON_SET(&reason, PFRES_SHORT); 8432 goto done; 8433 } 8434 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8435 if (action == PF_PASS) { 8436 if (V_pfsync_update_state_ptr != NULL) 8437 V_pfsync_update_state_ptr(s); 8438 r = s->rule.ptr; 8439 a = s->anchor.ptr; 8440 } else if (s == NULL) 8441 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8442 &a, &ruleset, inp); 8443 break; 8444 } 8445 8446 case IPPROTO_SCTP: { 8447 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8448 &action, &reason, AF_INET6)) { 8449 if (action != PF_PASS) 8450 pd.act.log |= PF_LOG_FORCE; 8451 goto done; 8452 } 8453 pd.sport = &pd.hdr.sctp.src_port; 8454 pd.dport = &pd.hdr.sctp.dest_port; 8455 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8456 action = PF_DROP; 8457 REASON_SET(&reason, PFRES_SHORT); 8458 goto done; 8459 } 8460 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8461 if (action == PF_DROP) 8462 goto done; 8463 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8464 &reason); 8465 if (action == PF_PASS) { 8466 if (V_pfsync_update_state_ptr != NULL) 8467 V_pfsync_update_state_ptr(s); 8468 r = s->rule.ptr; 8469 a = s->anchor.ptr; 8470 } else { 8471 action = pf_test_rule(&r, &s, kif, m, off, 8472 &pd, &a, &ruleset, inp); 8473 } 8474 break; 8475 } 8476 8477 case IPPROTO_ICMP: { 8478 action = PF_DROP; 8479 DPFPRINTF(PF_DEBUG_MISC, 8480 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 8481 goto done; 8482 } 8483 8484 case IPPROTO_ICMPV6: { 8485 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 8486 &action, &reason, AF_INET6)) { 8487 if (action != PF_PASS) 8488 pd.act.log |= PF_LOG_FORCE; 8489 goto done; 8490 } 8491 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8492 if (action == PF_PASS) { 8493 if (V_pfsync_update_state_ptr != NULL) 8494 V_pfsync_update_state_ptr(s); 8495 r = s->rule.ptr; 8496 a = s->anchor.ptr; 8497 } else if (s == NULL) 8498 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8499 &a, &ruleset, inp); 8500 break; 8501 } 8502 8503 default: 8504 action = pf_test_state_other(&s, kif, m, &pd); 8505 if (action == PF_PASS) { 8506 if (V_pfsync_update_state_ptr != NULL) 8507 V_pfsync_update_state_ptr(s); 8508 r = s->rule.ptr; 8509 a = s->anchor.ptr; 8510 } else if (s == NULL) 8511 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8512 &a, &ruleset, inp); 8513 break; 8514 } 8515 8516 done: 8517 PF_RULES_RUNLOCK(); 8518 if (n != m) { 8519 m_freem(n); 8520 n = NULL; 8521 } 8522 8523 /* handle dangerous IPv6 extension headers. */ 8524 if (action == PF_PASS && rh_cnt && 8525 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8526 action = PF_DROP; 8527 REASON_SET(&reason, PFRES_IPOPTIONS); 8528 pd.act.log = r->log; 8529 DPFPRINTF(PF_DEBUG_MISC, 8530 ("pf: dropping packet with dangerous v6 headers\n")); 8531 } 8532 8533 if (s) { 8534 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8535 tag = s->tag; 8536 rt = s->rt; 8537 } else { 8538 tag = r->tag; 8539 rt = r->rt; 8540 } 8541 8542 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8543 action = PF_DROP; 8544 REASON_SET(&reason, PFRES_MEMORY); 8545 } 8546 8547 pf_scrub_ip6(&m, &pd); 8548 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8549 pf_normalize_mss(m, off, &pd); 8550 8551 if (pd.act.rtableid >= 0) 8552 M_SETFIB(m, pd.act.rtableid); 8553 8554 if (pd.act.flags & PFSTATE_SETPRIO) { 8555 if (pd.tos & IPTOS_LOWDELAY) 8556 use_2nd_queue = 1; 8557 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8558 action = PF_DROP; 8559 REASON_SET(&reason, PFRES_MEMORY); 8560 pd.act.log = PF_LOG_FORCE; 8561 DPFPRINTF(PF_DEBUG_MISC, 8562 ("pf: failed to allocate 802.1q mtag\n")); 8563 } 8564 } 8565 8566 #ifdef ALTQ 8567 if (action == PF_PASS && pd.act.qid) { 8568 if (pd.pf_mtag == NULL && 8569 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8570 action = PF_DROP; 8571 REASON_SET(&reason, PFRES_MEMORY); 8572 } else { 8573 if (s != NULL) 8574 pd.pf_mtag->qid_hash = pf_state_hash(s); 8575 if (pd.tos & IPTOS_LOWDELAY) 8576 pd.pf_mtag->qid = pd.act.pqid; 8577 else 8578 pd.pf_mtag->qid = pd.act.qid; 8579 /* Add hints for ecn. */ 8580 pd.pf_mtag->hdr = h; 8581 } 8582 } 8583 #endif /* ALTQ */ 8584 8585 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8586 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8587 (s->nat_rule.ptr->action == PF_RDR || 8588 s->nat_rule.ptr->action == PF_BINAT) && 8589 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 8590 m->m_flags |= M_SKIP_FIREWALL; 8591 8592 /* XXX: Anybody working on it?! */ 8593 if (r->divert.port) 8594 printf("pf: divert(9) is not supported for IPv6\n"); 8595 8596 if (pd.act.log) { 8597 struct pf_krule *lr; 8598 struct pf_krule_item *ri; 8599 8600 if (s != NULL && s->nat_rule.ptr != NULL && 8601 s->nat_rule.ptr->log & PF_LOG_ALL) 8602 lr = s->nat_rule.ptr; 8603 else 8604 lr = r; 8605 8606 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8607 PFLOG_PACKET(kif, m, AF_INET6, reason, lr, a, ruleset, 8608 &pd, (s == NULL)); 8609 if (s) { 8610 SLIST_FOREACH(ri, &s->match_rules, entry) 8611 if (ri->r->log & PF_LOG_ALL) 8612 PFLOG_PACKET(kif, m, AF_INET6, reason, 8613 ri->r, a, ruleset, &pd, 0); 8614 } 8615 } 8616 8617 pf_counter_u64_critical_enter(); 8618 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 8619 pd.tot_len); 8620 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 8621 1); 8622 8623 if (action == PF_PASS || r->action == PF_DROP) { 8624 dirndx = (dir == PF_OUT); 8625 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8626 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8627 if (a != NULL) { 8628 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8629 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8630 } 8631 if (s != NULL) { 8632 if (s->nat_rule.ptr != NULL) { 8633 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8634 1); 8635 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8636 pd.tot_len); 8637 } 8638 if (s->src_node != NULL) { 8639 counter_u64_add(s->src_node->packets[dirndx], 8640 1); 8641 counter_u64_add(s->src_node->bytes[dirndx], 8642 pd.tot_len); 8643 } 8644 if (s->nat_src_node != NULL) { 8645 counter_u64_add(s->nat_src_node->packets[dirndx], 8646 1); 8647 counter_u64_add(s->nat_src_node->bytes[dirndx], 8648 pd.tot_len); 8649 } 8650 dirndx = (dir == s->direction) ? 0 : 1; 8651 s->packets[dirndx]++; 8652 s->bytes[dirndx] += pd.tot_len; 8653 } 8654 tr = r; 8655 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8656 if (nr != NULL && r == &V_pf_default_rule) 8657 tr = nr; 8658 if (tr->src.addr.type == PF_ADDR_TABLE) 8659 pfr_update_stats(tr->src.addr.p.tbl, 8660 (s == NULL) ? pd.src : 8661 &s->key[(s->direction == PF_IN)]->addr[0], 8662 pd.af, pd.tot_len, dir == PF_OUT, 8663 r->action == PF_PASS, tr->src.neg); 8664 if (tr->dst.addr.type == PF_ADDR_TABLE) 8665 pfr_update_stats(tr->dst.addr.p.tbl, 8666 (s == NULL) ? pd.dst : 8667 &s->key[(s->direction == PF_IN)]->addr[1], 8668 pd.af, pd.tot_len, dir == PF_OUT, 8669 r->action == PF_PASS, tr->dst.neg); 8670 } 8671 pf_counter_u64_critical_exit(); 8672 8673 switch (action) { 8674 case PF_SYNPROXY_DROP: 8675 m_freem(*m0); 8676 case PF_DEFER: 8677 *m0 = NULL; 8678 action = PF_PASS; 8679 break; 8680 case PF_DROP: 8681 m_freem(*m0); 8682 *m0 = NULL; 8683 break; 8684 default: 8685 /* pf_route6() returns unlocked. */ 8686 if (rt) { 8687 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 8688 goto out; 8689 } 8690 if (pf_dummynet(&pd, s, r, m0) != 0) { 8691 action = PF_DROP; 8692 REASON_SET(&reason, PFRES_MEMORY); 8693 } 8694 break; 8695 } 8696 8697 if (s) 8698 PF_STATE_UNLOCK(s); 8699 8700 /* If reassembled packet passed, create new fragments. */ 8701 if (action == PF_PASS && *m0 && dir == PF_OUT && 8702 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 8703 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 8704 8705 out: 8706 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 8707 8708 pf_sctp_multihome_delayed(&pd, off, kif, s); 8709 8710 return (action); 8711 } 8712 #endif /* INET6 */ 8713