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 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 141 142 /* 143 * Global variables 144 */ 145 146 /* state tables */ 147 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 148 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 149 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 150 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 151 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 152 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 153 VNET_DEFINE(struct pf_kstatus, pf_status); 154 155 VNET_DEFINE(u_int32_t, ticket_altqs_active); 156 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 157 VNET_DEFINE(int, altqs_inactive_open); 158 VNET_DEFINE(u_int32_t, ticket_pabuf); 159 160 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 161 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 162 VNET_DEFINE(u_char, pf_tcp_secret[16]); 163 #define V_pf_tcp_secret VNET(pf_tcp_secret) 164 VNET_DEFINE(int, pf_tcp_secret_init); 165 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 166 VNET_DEFINE(int, pf_tcp_iss_off); 167 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 168 VNET_DECLARE(int, pf_vnet_active); 169 #define V_pf_vnet_active VNET(pf_vnet_active) 170 171 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 172 #define V_pf_purge_idx VNET(pf_purge_idx) 173 174 #ifdef PF_WANT_32_TO_64_COUNTER 175 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 176 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 177 178 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 179 VNET_DEFINE(size_t, pf_allrulecount); 180 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 181 #endif 182 183 struct pf_sctp_endpoint; 184 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 185 struct pf_sctp_source { 186 sa_family_t af; 187 struct pf_addr addr; 188 TAILQ_ENTRY(pf_sctp_source) entry; 189 }; 190 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 191 struct pf_sctp_endpoint 192 { 193 uint32_t v_tag; 194 struct pf_sctp_sources sources; 195 RB_ENTRY(pf_sctp_endpoint) entry; 196 }; 197 static int 198 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 199 { 200 return (a->v_tag - b->v_tag); 201 } 202 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 203 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 204 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 205 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 206 static struct mtx_padalign pf_sctp_endpoints_mtx; 207 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 208 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 209 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 210 211 /* 212 * Queue for pf_intr() sends. 213 */ 214 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 215 struct pf_send_entry { 216 STAILQ_ENTRY(pf_send_entry) pfse_next; 217 struct mbuf *pfse_m; 218 enum { 219 PFSE_IP, 220 PFSE_IP6, 221 PFSE_ICMP, 222 PFSE_ICMP6, 223 } pfse_type; 224 struct { 225 int type; 226 int code; 227 int mtu; 228 } icmpopts; 229 }; 230 231 STAILQ_HEAD(pf_send_head, pf_send_entry); 232 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 233 #define V_pf_sendqueue VNET(pf_sendqueue) 234 235 static struct mtx_padalign pf_sendqueue_mtx; 236 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 237 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 238 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 239 240 /* 241 * Queue for pf_overload_task() tasks. 242 */ 243 struct pf_overload_entry { 244 SLIST_ENTRY(pf_overload_entry) next; 245 struct pf_addr addr; 246 sa_family_t af; 247 uint8_t dir; 248 struct pf_krule *rule; 249 }; 250 251 SLIST_HEAD(pf_overload_head, pf_overload_entry); 252 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 253 #define V_pf_overloadqueue VNET(pf_overloadqueue) 254 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 255 #define V_pf_overloadtask VNET(pf_overloadtask) 256 257 static struct mtx_padalign pf_overloadqueue_mtx; 258 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 259 "pf overload/flush queue", MTX_DEF); 260 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 261 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 262 263 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 264 struct mtx_padalign pf_unlnkdrules_mtx; 265 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 266 MTX_DEF); 267 268 struct sx pf_config_lock; 269 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 270 271 struct mtx_padalign pf_table_stats_lock; 272 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 273 MTX_DEF); 274 275 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 276 #define V_pf_sources_z VNET(pf_sources_z) 277 uma_zone_t pf_mtag_z; 278 VNET_DEFINE(uma_zone_t, pf_state_z); 279 VNET_DEFINE(uma_zone_t, pf_state_key_z); 280 281 VNET_DEFINE(struct unrhdr64, pf_stateid); 282 283 static void pf_src_tree_remove_state(struct pf_kstate *); 284 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 285 u_int32_t); 286 static void pf_add_threshold(struct pf_threshold *); 287 static int pf_check_threshold(struct pf_threshold *); 288 289 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 290 u_int16_t *, u_int16_t *, struct pf_addr *, 291 u_int16_t, u_int8_t, sa_family_t); 292 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 293 struct tcphdr *, struct pf_state_peer *); 294 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 295 struct pf_addr *, struct pf_addr *, u_int16_t, 296 u_int16_t *, u_int16_t *, u_int16_t *, 297 u_int16_t *, u_int8_t, sa_family_t); 298 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 299 sa_family_t, struct pf_krule *, int); 300 static void pf_detach_state(struct pf_kstate *); 301 static int pf_state_key_attach(struct pf_state_key *, 302 struct pf_state_key *, struct pf_kstate *); 303 static void pf_state_key_detach(struct pf_kstate *, int); 304 static int pf_state_key_ctor(void *, int, void *, int); 305 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 306 static __inline void pf_dummynet_flag_remove(struct mbuf *m, 307 struct pf_mtag *pf_mtag); 308 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 309 struct pf_krule *, struct mbuf **); 310 static int pf_dummynet_route(struct pf_pdesc *, 311 struct pf_kstate *, struct pf_krule *, 312 struct ifnet *, struct sockaddr *, struct mbuf **); 313 static int pf_test_eth_rule(int, struct pfi_kkif *, 314 struct mbuf **); 315 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 316 struct pfi_kkif *, struct mbuf *, int, 317 struct pf_pdesc *, struct pf_krule **, 318 struct pf_kruleset **, struct inpcb *); 319 static int pf_create_state(struct pf_krule *, struct pf_krule *, 320 struct pf_krule *, struct pf_pdesc *, 321 struct pf_ksrc_node *, struct pf_state_key *, 322 struct pf_state_key *, struct mbuf *, int, 323 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 324 struct pf_kstate **, int, u_int16_t, u_int16_t, 325 int, struct pf_krule_slist *); 326 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 327 struct mbuf *, void *, struct pf_pdesc *, 328 struct pf_krule **, struct pf_kruleset **); 329 static int pf_tcp_track_full(struct pf_kstate **, 330 struct pfi_kkif *, struct mbuf *, int, 331 struct pf_pdesc *, u_short *, int *); 332 static int pf_tcp_track_sloppy(struct pf_kstate **, 333 struct pf_pdesc *, u_short *); 334 static int pf_test_state_tcp(struct pf_kstate **, 335 struct pfi_kkif *, struct mbuf *, int, 336 void *, struct pf_pdesc *, u_short *); 337 static int pf_test_state_udp(struct pf_kstate **, 338 struct pfi_kkif *, struct mbuf *, int, 339 void *, struct pf_pdesc *); 340 static int pf_test_state_icmp(struct pf_kstate **, 341 struct pfi_kkif *, struct mbuf *, int, 342 void *, struct pf_pdesc *, u_short *); 343 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 344 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 345 struct pfi_kkif *, struct pf_kstate *, int); 346 static int pf_test_state_sctp(struct pf_kstate **, 347 struct pfi_kkif *, struct mbuf *, int, 348 void *, struct pf_pdesc *, u_short *); 349 static int pf_test_state_other(struct pf_kstate **, 350 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 351 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 352 int, u_int16_t); 353 static int pf_check_proto_cksum(struct mbuf *, int, int, 354 u_int8_t, sa_family_t); 355 static void pf_print_state_parts(struct pf_kstate *, 356 struct pf_state_key *, struct pf_state_key *); 357 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 358 bool, u_int8_t); 359 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 360 struct pf_state_key_cmp *, u_int); 361 static int pf_src_connlimit(struct pf_kstate **); 362 static void pf_overload_task(void *v, int pending); 363 static u_short pf_insert_src_node(struct pf_ksrc_node **, 364 struct pf_krule *, struct pf_addr *, sa_family_t); 365 static u_int pf_purge_expired_states(u_int, int); 366 static void pf_purge_unlinked_rules(void); 367 static int pf_mtag_uminit(void *, int, int); 368 static void pf_mtag_free(struct m_tag *); 369 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 370 int, struct pf_state_key *); 371 #ifdef INET 372 static void pf_route(struct mbuf **, struct pf_krule *, 373 struct ifnet *, struct pf_kstate *, 374 struct pf_pdesc *, struct inpcb *); 375 #endif /* INET */ 376 #ifdef INET6 377 static void pf_change_a6(struct pf_addr *, u_int16_t *, 378 struct pf_addr *, u_int8_t); 379 static void pf_route6(struct mbuf **, struct pf_krule *, 380 struct ifnet *, struct pf_kstate *, 381 struct pf_pdesc *, struct inpcb *); 382 #endif /* INET6 */ 383 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 384 385 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 386 387 extern int pf_end_threads; 388 extern struct proc *pf_purge_proc; 389 390 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 391 392 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 393 do { \ 394 struct pf_state_key *nk; \ 395 if ((pd->dir) == PF_OUT) \ 396 nk = (_s)->key[PF_SK_STACK]; \ 397 else \ 398 nk = (_s)->key[PF_SK_WIRE]; \ 399 pf_packet_rework_nat(_m, _pd, _off, nk); \ 400 } while (0) 401 402 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 403 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 404 405 #define STATE_LOOKUP(i, k, s, pd) \ 406 do { \ 407 (s) = pf_find_state((i), (k), (pd->dir)); \ 408 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 409 if ((s) == NULL) \ 410 return (PF_DROP); \ 411 if (PACKET_LOOPED(pd)) \ 412 return (PF_PASS); \ 413 } while (0) 414 415 #define BOUND_IFACE(r, k) \ 416 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 417 418 #define STATE_INC_COUNTERS(s) \ 419 do { \ 420 struct pf_krule_item *mrm; \ 421 counter_u64_add(s->rule.ptr->states_cur, 1); \ 422 counter_u64_add(s->rule.ptr->states_tot, 1); \ 423 if (s->anchor.ptr != NULL) { \ 424 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 425 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 426 } \ 427 if (s->nat_rule.ptr != NULL) { \ 428 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 429 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 430 } \ 431 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 432 counter_u64_add(mrm->r->states_cur, 1); \ 433 counter_u64_add(mrm->r->states_tot, 1); \ 434 } \ 435 } while (0) 436 437 #define STATE_DEC_COUNTERS(s) \ 438 do { \ 439 struct pf_krule_item *mrm; \ 440 if (s->nat_rule.ptr != NULL) \ 441 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 442 if (s->anchor.ptr != NULL) \ 443 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 444 counter_u64_add(s->rule.ptr->states_cur, -1); \ 445 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 446 counter_u64_add(mrm->r->states_cur, -1); \ 447 } while (0) 448 449 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 450 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 451 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 452 VNET_DEFINE(struct pf_idhash *, pf_idhash); 453 VNET_DEFINE(struct pf_srchash *, pf_srchash); 454 455 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 456 "pf(4)"); 457 458 u_long pf_hashmask; 459 u_long pf_srchashmask; 460 static u_long pf_hashsize; 461 static u_long pf_srchashsize; 462 u_long pf_ioctl_maxcount = 65535; 463 464 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 465 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 466 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 467 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 468 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 469 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 470 471 VNET_DEFINE(void *, pf_swi_cookie); 472 VNET_DEFINE(struct intr_event *, pf_swi_ie); 473 474 VNET_DEFINE(uint32_t, pf_hashseed); 475 #define V_pf_hashseed VNET(pf_hashseed) 476 477 static void 478 pf_sctp_checksum(struct mbuf *m, int off) 479 { 480 uint32_t sum = 0; 481 482 /* Zero out the checksum, to enable recalculation. */ 483 m_copyback(m, off + offsetof(struct sctphdr, checksum), 484 sizeof(sum), (caddr_t)&sum); 485 486 sum = sctp_calculate_cksum(m, off); 487 488 m_copyback(m, off + offsetof(struct sctphdr, checksum), 489 sizeof(sum), (caddr_t)&sum); 490 } 491 492 int 493 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 494 { 495 496 switch (af) { 497 #ifdef INET 498 case AF_INET: 499 if (a->addr32[0] > b->addr32[0]) 500 return (1); 501 if (a->addr32[0] < b->addr32[0]) 502 return (-1); 503 break; 504 #endif /* INET */ 505 #ifdef INET6 506 case AF_INET6: 507 if (a->addr32[3] > b->addr32[3]) 508 return (1); 509 if (a->addr32[3] < b->addr32[3]) 510 return (-1); 511 if (a->addr32[2] > b->addr32[2]) 512 return (1); 513 if (a->addr32[2] < b->addr32[2]) 514 return (-1); 515 if (a->addr32[1] > b->addr32[1]) 516 return (1); 517 if (a->addr32[1] < b->addr32[1]) 518 return (-1); 519 if (a->addr32[0] > b->addr32[0]) 520 return (1); 521 if (a->addr32[0] < b->addr32[0]) 522 return (-1); 523 break; 524 #endif /* INET6 */ 525 default: 526 panic("%s: unknown address family %u", __func__, af); 527 } 528 return (0); 529 } 530 531 static void 532 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 533 struct pf_state_key *nk) 534 { 535 536 switch (pd->proto) { 537 case IPPROTO_TCP: { 538 struct tcphdr *th = &pd->hdr.tcp; 539 540 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 541 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 542 &th->th_sum, &nk->addr[pd->sidx], 543 nk->port[pd->sidx], 0, pd->af); 544 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 545 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 546 &th->th_sum, &nk->addr[pd->didx], 547 nk->port[pd->didx], 0, pd->af); 548 m_copyback(m, off, sizeof(*th), (caddr_t)th); 549 break; 550 } 551 case IPPROTO_UDP: { 552 struct udphdr *uh = &pd->hdr.udp; 553 554 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 555 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 556 &uh->uh_sum, &nk->addr[pd->sidx], 557 nk->port[pd->sidx], 1, pd->af); 558 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 559 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 560 &uh->uh_sum, &nk->addr[pd->didx], 561 nk->port[pd->didx], 1, pd->af); 562 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 563 break; 564 } 565 case IPPROTO_SCTP: { 566 struct sctphdr *sh = &pd->hdr.sctp; 567 uint16_t checksum = 0; 568 569 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 570 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 571 &checksum, &nk->addr[pd->sidx], 572 nk->port[pd->sidx], 1, pd->af); 573 } 574 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 575 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 576 &checksum, &nk->addr[pd->didx], 577 nk->port[pd->didx], 1, pd->af); 578 } 579 580 break; 581 } 582 case IPPROTO_ICMP: { 583 struct icmp *ih = &pd->hdr.icmp; 584 585 if (nk->port[pd->sidx] != ih->icmp_id) { 586 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 587 ih->icmp_cksum, ih->icmp_id, 588 nk->port[pd->sidx], 0); 589 ih->icmp_id = nk->port[pd->sidx]; 590 pd->sport = &ih->icmp_id; 591 592 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 593 } 594 /* FALLTHROUGH */ 595 } 596 default: 597 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 598 switch (pd->af) { 599 case AF_INET: 600 pf_change_a(&pd->src->v4.s_addr, 601 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 602 0); 603 break; 604 case AF_INET6: 605 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 606 break; 607 } 608 } 609 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 610 switch (pd->af) { 611 case AF_INET: 612 pf_change_a(&pd->dst->v4.s_addr, 613 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 614 0); 615 break; 616 case AF_INET6: 617 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 618 break; 619 } 620 } 621 break; 622 } 623 } 624 625 static __inline uint32_t 626 pf_hashkey(struct pf_state_key *sk) 627 { 628 uint32_t h; 629 630 h = murmur3_32_hash32((uint32_t *)sk, 631 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 632 V_pf_hashseed); 633 634 return (h & pf_hashmask); 635 } 636 637 static __inline uint32_t 638 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 639 { 640 uint32_t h; 641 642 switch (af) { 643 case AF_INET: 644 h = murmur3_32_hash32((uint32_t *)&addr->v4, 645 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 646 break; 647 case AF_INET6: 648 h = murmur3_32_hash32((uint32_t *)&addr->v6, 649 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 650 break; 651 default: 652 panic("%s: unknown address family %u", __func__, af); 653 } 654 655 return (h & pf_srchashmask); 656 } 657 658 #ifdef ALTQ 659 static int 660 pf_state_hash(struct pf_kstate *s) 661 { 662 u_int32_t hv = (intptr_t)s / sizeof(*s); 663 664 hv ^= crc32(&s->src, sizeof(s->src)); 665 hv ^= crc32(&s->dst, sizeof(s->dst)); 666 if (hv == 0) 667 hv = 1; 668 return (hv); 669 } 670 #endif 671 672 static __inline void 673 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 674 { 675 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 676 s->dst.state = newstate; 677 if (which == PF_PEER_DST) 678 return; 679 if (s->src.state == newstate) 680 return; 681 if (s->creatorid == V_pf_status.hostid && 682 s->key[PF_SK_STACK] != NULL && 683 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 684 !(TCPS_HAVEESTABLISHED(s->src.state) || 685 s->src.state == TCPS_CLOSED) && 686 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 687 atomic_add_32(&V_pf_status.states_halfopen, -1); 688 689 s->src.state = newstate; 690 } 691 692 #ifdef INET6 693 void 694 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 695 { 696 switch (af) { 697 #ifdef INET 698 case AF_INET: 699 dst->addr32[0] = src->addr32[0]; 700 break; 701 #endif /* INET */ 702 case AF_INET6: 703 dst->addr32[0] = src->addr32[0]; 704 dst->addr32[1] = src->addr32[1]; 705 dst->addr32[2] = src->addr32[2]; 706 dst->addr32[3] = src->addr32[3]; 707 break; 708 } 709 } 710 #endif /* INET6 */ 711 712 static void 713 pf_init_threshold(struct pf_threshold *threshold, 714 u_int32_t limit, u_int32_t seconds) 715 { 716 threshold->limit = limit * PF_THRESHOLD_MULT; 717 threshold->seconds = seconds; 718 threshold->count = 0; 719 threshold->last = time_uptime; 720 } 721 722 static void 723 pf_add_threshold(struct pf_threshold *threshold) 724 { 725 u_int32_t t = time_uptime, diff = t - threshold->last; 726 727 if (diff >= threshold->seconds) 728 threshold->count = 0; 729 else 730 threshold->count -= threshold->count * diff / 731 threshold->seconds; 732 threshold->count += PF_THRESHOLD_MULT; 733 threshold->last = t; 734 } 735 736 static int 737 pf_check_threshold(struct pf_threshold *threshold) 738 { 739 return (threshold->count > threshold->limit); 740 } 741 742 static int 743 pf_src_connlimit(struct pf_kstate **state) 744 { 745 struct pf_overload_entry *pfoe; 746 int bad = 0; 747 748 PF_STATE_LOCK_ASSERT(*state); 749 /* 750 * XXXKS: The src node is accessed unlocked! 751 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 752 */ 753 754 (*state)->src_node->conn++; 755 (*state)->src.tcp_est = 1; 756 pf_add_threshold(&(*state)->src_node->conn_rate); 757 758 if ((*state)->rule.ptr->max_src_conn && 759 (*state)->rule.ptr->max_src_conn < 760 (*state)->src_node->conn) { 761 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 762 bad++; 763 } 764 765 if ((*state)->rule.ptr->max_src_conn_rate.limit && 766 pf_check_threshold(&(*state)->src_node->conn_rate)) { 767 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 768 bad++; 769 } 770 771 if (!bad) 772 return (0); 773 774 /* Kill this state. */ 775 (*state)->timeout = PFTM_PURGE; 776 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 777 778 if ((*state)->rule.ptr->overload_tbl == NULL) 779 return (1); 780 781 /* Schedule overloading and flushing task. */ 782 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 783 if (pfoe == NULL) 784 return (1); /* too bad :( */ 785 786 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 787 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 788 pfoe->rule = (*state)->rule.ptr; 789 pfoe->dir = (*state)->direction; 790 PF_OVERLOADQ_LOCK(); 791 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 792 PF_OVERLOADQ_UNLOCK(); 793 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 794 795 return (1); 796 } 797 798 static void 799 pf_overload_task(void *v, int pending) 800 { 801 struct pf_overload_head queue; 802 struct pfr_addr p; 803 struct pf_overload_entry *pfoe, *pfoe1; 804 uint32_t killed = 0; 805 806 CURVNET_SET((struct vnet *)v); 807 808 PF_OVERLOADQ_LOCK(); 809 queue = V_pf_overloadqueue; 810 SLIST_INIT(&V_pf_overloadqueue); 811 PF_OVERLOADQ_UNLOCK(); 812 813 bzero(&p, sizeof(p)); 814 SLIST_FOREACH(pfoe, &queue, next) { 815 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 816 if (V_pf_status.debug >= PF_DEBUG_MISC) { 817 printf("%s: blocking address ", __func__); 818 pf_print_host(&pfoe->addr, 0, pfoe->af); 819 printf("\n"); 820 } 821 822 p.pfra_af = pfoe->af; 823 switch (pfoe->af) { 824 #ifdef INET 825 case AF_INET: 826 p.pfra_net = 32; 827 p.pfra_ip4addr = pfoe->addr.v4; 828 break; 829 #endif 830 #ifdef INET6 831 case AF_INET6: 832 p.pfra_net = 128; 833 p.pfra_ip6addr = pfoe->addr.v6; 834 break; 835 #endif 836 } 837 838 PF_RULES_WLOCK(); 839 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 840 PF_RULES_WUNLOCK(); 841 } 842 843 /* 844 * Remove those entries, that don't need flushing. 845 */ 846 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 847 if (pfoe->rule->flush == 0) { 848 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 849 free(pfoe, M_PFTEMP); 850 } else 851 counter_u64_add( 852 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 853 854 /* If nothing to flush, return. */ 855 if (SLIST_EMPTY(&queue)) { 856 CURVNET_RESTORE(); 857 return; 858 } 859 860 for (int i = 0; i <= pf_hashmask; i++) { 861 struct pf_idhash *ih = &V_pf_idhash[i]; 862 struct pf_state_key *sk; 863 struct pf_kstate *s; 864 865 PF_HASHROW_LOCK(ih); 866 LIST_FOREACH(s, &ih->states, entry) { 867 sk = s->key[PF_SK_WIRE]; 868 SLIST_FOREACH(pfoe, &queue, next) 869 if (sk->af == pfoe->af && 870 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 871 pfoe->rule == s->rule.ptr) && 872 ((pfoe->dir == PF_OUT && 873 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 874 (pfoe->dir == PF_IN && 875 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 876 s->timeout = PFTM_PURGE; 877 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 878 killed++; 879 } 880 } 881 PF_HASHROW_UNLOCK(ih); 882 } 883 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 884 free(pfoe, M_PFTEMP); 885 if (V_pf_status.debug >= PF_DEBUG_MISC) 886 printf("%s: %u states killed", __func__, killed); 887 888 CURVNET_RESTORE(); 889 } 890 891 /* 892 * Can return locked on failure, so that we can consistently 893 * allocate and insert a new one. 894 */ 895 struct pf_ksrc_node * 896 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 897 struct pf_srchash **sh, bool returnlocked) 898 { 899 struct pf_ksrc_node *n; 900 901 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 902 903 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 904 PF_HASHROW_LOCK(*sh); 905 LIST_FOREACH(n, &(*sh)->nodes, entry) 906 if (n->rule.ptr == rule && n->af == af && 907 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 908 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 909 break; 910 911 if (n != NULL) { 912 n->states++; 913 PF_HASHROW_UNLOCK(*sh); 914 } else if (returnlocked == false) 915 PF_HASHROW_UNLOCK(*sh); 916 917 return (n); 918 } 919 920 static void 921 pf_free_src_node(struct pf_ksrc_node *sn) 922 { 923 924 for (int i = 0; i < 2; i++) { 925 counter_u64_free(sn->bytes[i]); 926 counter_u64_free(sn->packets[i]); 927 } 928 uma_zfree(V_pf_sources_z, sn); 929 } 930 931 static u_short 932 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 933 struct pf_addr *src, sa_family_t af) 934 { 935 u_short reason = 0; 936 struct pf_srchash *sh = NULL; 937 938 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 939 rule->rpool.opts & PF_POOL_STICKYADDR), 940 ("%s for non-tracking rule %p", __func__, rule)); 941 942 if (*sn == NULL) 943 *sn = pf_find_src_node(src, rule, af, &sh, true); 944 945 if (*sn == NULL) { 946 PF_HASHROW_ASSERT(sh); 947 948 if (rule->max_src_nodes && 949 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 950 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 951 PF_HASHROW_UNLOCK(sh); 952 reason = PFRES_SRCLIMIT; 953 goto done; 954 } 955 956 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 957 if ((*sn) == NULL) { 958 PF_HASHROW_UNLOCK(sh); 959 reason = PFRES_MEMORY; 960 goto done; 961 } 962 963 for (int i = 0; i < 2; i++) { 964 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 965 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 966 967 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 968 pf_free_src_node(*sn); 969 PF_HASHROW_UNLOCK(sh); 970 reason = PFRES_MEMORY; 971 goto done; 972 } 973 } 974 975 pf_init_threshold(&(*sn)->conn_rate, 976 rule->max_src_conn_rate.limit, 977 rule->max_src_conn_rate.seconds); 978 979 MPASS((*sn)->lock == NULL); 980 (*sn)->lock = &sh->lock; 981 982 (*sn)->af = af; 983 (*sn)->rule.ptr = rule; 984 PF_ACPY(&(*sn)->addr, src, af); 985 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 986 (*sn)->creation = time_uptime; 987 (*sn)->ruletype = rule->action; 988 (*sn)->states = 1; 989 if ((*sn)->rule.ptr != NULL) 990 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 991 PF_HASHROW_UNLOCK(sh); 992 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 993 } else { 994 if (rule->max_src_states && 995 (*sn)->states >= rule->max_src_states) { 996 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 997 1); 998 reason = PFRES_SRCLIMIT; 999 goto done; 1000 } 1001 } 1002 done: 1003 return (reason); 1004 } 1005 1006 void 1007 pf_unlink_src_node(struct pf_ksrc_node *src) 1008 { 1009 PF_SRC_NODE_LOCK_ASSERT(src); 1010 1011 LIST_REMOVE(src, entry); 1012 if (src->rule.ptr) 1013 counter_u64_add(src->rule.ptr->src_nodes, -1); 1014 } 1015 1016 u_int 1017 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1018 { 1019 struct pf_ksrc_node *sn, *tmp; 1020 u_int count = 0; 1021 1022 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1023 pf_free_src_node(sn); 1024 count++; 1025 } 1026 1027 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1028 1029 return (count); 1030 } 1031 1032 void 1033 pf_mtag_initialize(void) 1034 { 1035 1036 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1037 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1038 UMA_ALIGN_PTR, 0); 1039 } 1040 1041 /* Per-vnet data storage structures initialization. */ 1042 void 1043 pf_initialize(void) 1044 { 1045 struct pf_keyhash *kh; 1046 struct pf_idhash *ih; 1047 struct pf_srchash *sh; 1048 u_int i; 1049 1050 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 1051 pf_hashsize = PF_HASHSIZ; 1052 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 1053 pf_srchashsize = PF_SRCHASHSIZ; 1054 1055 V_pf_hashseed = arc4random(); 1056 1057 /* States and state keys storage. */ 1058 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1059 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1060 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1061 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1062 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1063 1064 V_pf_state_key_z = uma_zcreate("pf state keys", 1065 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1066 UMA_ALIGN_PTR, 0); 1067 1068 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 1069 M_PFHASH, M_NOWAIT | M_ZERO); 1070 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1071 M_PFHASH, M_NOWAIT | M_ZERO); 1072 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1073 printf("pf: Unable to allocate memory for " 1074 "state_hashsize %lu.\n", pf_hashsize); 1075 1076 free(V_pf_keyhash, M_PFHASH); 1077 free(V_pf_idhash, M_PFHASH); 1078 1079 pf_hashsize = PF_HASHSIZ; 1080 V_pf_keyhash = mallocarray(pf_hashsize, 1081 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1082 V_pf_idhash = mallocarray(pf_hashsize, 1083 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1084 } 1085 1086 pf_hashmask = pf_hashsize - 1; 1087 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1088 i++, kh++, ih++) { 1089 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1090 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1091 } 1092 1093 /* Source nodes. */ 1094 V_pf_sources_z = uma_zcreate("pf source nodes", 1095 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1096 0); 1097 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1098 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1099 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1100 1101 V_pf_srchash = mallocarray(pf_srchashsize, 1102 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1103 if (V_pf_srchash == NULL) { 1104 printf("pf: Unable to allocate memory for " 1105 "source_hashsize %lu.\n", pf_srchashsize); 1106 1107 pf_srchashsize = PF_SRCHASHSIZ; 1108 V_pf_srchash = mallocarray(pf_srchashsize, 1109 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1110 } 1111 1112 pf_srchashmask = pf_srchashsize - 1; 1113 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1114 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1115 1116 /* ALTQ */ 1117 TAILQ_INIT(&V_pf_altqs[0]); 1118 TAILQ_INIT(&V_pf_altqs[1]); 1119 TAILQ_INIT(&V_pf_altqs[2]); 1120 TAILQ_INIT(&V_pf_altqs[3]); 1121 TAILQ_INIT(&V_pf_pabuf); 1122 V_pf_altqs_active = &V_pf_altqs[0]; 1123 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1124 V_pf_altqs_inactive = &V_pf_altqs[2]; 1125 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1126 1127 /* Send & overload+flush queues. */ 1128 STAILQ_INIT(&V_pf_sendqueue); 1129 SLIST_INIT(&V_pf_overloadqueue); 1130 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1131 1132 /* Unlinked, but may be referenced rules. */ 1133 TAILQ_INIT(&V_pf_unlinked_rules); 1134 } 1135 1136 void 1137 pf_mtag_cleanup(void) 1138 { 1139 1140 uma_zdestroy(pf_mtag_z); 1141 } 1142 1143 void 1144 pf_cleanup(void) 1145 { 1146 struct pf_keyhash *kh; 1147 struct pf_idhash *ih; 1148 struct pf_srchash *sh; 1149 struct pf_send_entry *pfse, *next; 1150 u_int i; 1151 1152 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1153 i++, kh++, ih++) { 1154 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1155 __func__)); 1156 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1157 __func__)); 1158 mtx_destroy(&kh->lock); 1159 mtx_destroy(&ih->lock); 1160 } 1161 free(V_pf_keyhash, M_PFHASH); 1162 free(V_pf_idhash, M_PFHASH); 1163 1164 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1165 KASSERT(LIST_EMPTY(&sh->nodes), 1166 ("%s: source node hash not empty", __func__)); 1167 mtx_destroy(&sh->lock); 1168 } 1169 free(V_pf_srchash, M_PFHASH); 1170 1171 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1172 m_freem(pfse->pfse_m); 1173 free(pfse, M_PFTEMP); 1174 } 1175 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1176 1177 uma_zdestroy(V_pf_sources_z); 1178 uma_zdestroy(V_pf_state_z); 1179 uma_zdestroy(V_pf_state_key_z); 1180 } 1181 1182 static int 1183 pf_mtag_uminit(void *mem, int size, int how) 1184 { 1185 struct m_tag *t; 1186 1187 t = (struct m_tag *)mem; 1188 t->m_tag_cookie = MTAG_ABI_COMPAT; 1189 t->m_tag_id = PACKET_TAG_PF; 1190 t->m_tag_len = sizeof(struct pf_mtag); 1191 t->m_tag_free = pf_mtag_free; 1192 1193 return (0); 1194 } 1195 1196 static void 1197 pf_mtag_free(struct m_tag *t) 1198 { 1199 1200 uma_zfree(pf_mtag_z, t); 1201 } 1202 1203 struct pf_mtag * 1204 pf_get_mtag(struct mbuf *m) 1205 { 1206 struct m_tag *mtag; 1207 1208 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1209 return ((struct pf_mtag *)(mtag + 1)); 1210 1211 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1212 if (mtag == NULL) 1213 return (NULL); 1214 bzero(mtag + 1, sizeof(struct pf_mtag)); 1215 m_tag_prepend(m, mtag); 1216 1217 return ((struct pf_mtag *)(mtag + 1)); 1218 } 1219 1220 static int 1221 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1222 struct pf_kstate *s) 1223 { 1224 struct pf_keyhash *khs, *khw, *kh; 1225 struct pf_state_key *sk, *cur; 1226 struct pf_kstate *si, *olds = NULL; 1227 int idx; 1228 1229 NET_EPOCH_ASSERT(); 1230 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1231 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1232 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1233 1234 /* 1235 * We need to lock hash slots of both keys. To avoid deadlock 1236 * we always lock the slot with lower address first. Unlock order 1237 * isn't important. 1238 * 1239 * We also need to lock ID hash slot before dropping key 1240 * locks. On success we return with ID hash slot locked. 1241 */ 1242 1243 if (skw == sks) { 1244 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1245 PF_HASHROW_LOCK(khs); 1246 } else { 1247 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1248 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1249 if (khs == khw) { 1250 PF_HASHROW_LOCK(khs); 1251 } else if (khs < khw) { 1252 PF_HASHROW_LOCK(khs); 1253 PF_HASHROW_LOCK(khw); 1254 } else { 1255 PF_HASHROW_LOCK(khw); 1256 PF_HASHROW_LOCK(khs); 1257 } 1258 } 1259 1260 #define KEYS_UNLOCK() do { \ 1261 if (khs != khw) { \ 1262 PF_HASHROW_UNLOCK(khs); \ 1263 PF_HASHROW_UNLOCK(khw); \ 1264 } else \ 1265 PF_HASHROW_UNLOCK(khs); \ 1266 } while (0) 1267 1268 /* 1269 * First run: start with wire key. 1270 */ 1271 sk = skw; 1272 kh = khw; 1273 idx = PF_SK_WIRE; 1274 1275 MPASS(s->lock == NULL); 1276 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1277 1278 keyattach: 1279 LIST_FOREACH(cur, &kh->keys, entry) 1280 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1281 break; 1282 1283 if (cur != NULL) { 1284 /* Key exists. Check for same kif, if none, add to key. */ 1285 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1286 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1287 1288 PF_HASHROW_LOCK(ih); 1289 if (si->kif == s->kif && 1290 si->direction == s->direction) { 1291 if (sk->proto == IPPROTO_TCP && 1292 si->src.state >= TCPS_FIN_WAIT_2 && 1293 si->dst.state >= TCPS_FIN_WAIT_2) { 1294 /* 1295 * New state matches an old >FIN_WAIT_2 1296 * state. We can't drop key hash locks, 1297 * thus we can't unlink it properly. 1298 * 1299 * As a workaround we drop it into 1300 * TCPS_CLOSED state, schedule purge 1301 * ASAP and push it into the very end 1302 * of the slot TAILQ, so that it won't 1303 * conflict with our new state. 1304 */ 1305 pf_set_protostate(si, PF_PEER_BOTH, 1306 TCPS_CLOSED); 1307 si->timeout = PFTM_PURGE; 1308 olds = si; 1309 } else { 1310 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1311 printf("pf: %s key attach " 1312 "failed on %s: ", 1313 (idx == PF_SK_WIRE) ? 1314 "wire" : "stack", 1315 s->kif->pfik_name); 1316 pf_print_state_parts(s, 1317 (idx == PF_SK_WIRE) ? 1318 sk : NULL, 1319 (idx == PF_SK_STACK) ? 1320 sk : NULL); 1321 printf(", existing: "); 1322 pf_print_state_parts(si, 1323 (idx == PF_SK_WIRE) ? 1324 sk : NULL, 1325 (idx == PF_SK_STACK) ? 1326 sk : NULL); 1327 printf("\n"); 1328 } 1329 PF_HASHROW_UNLOCK(ih); 1330 KEYS_UNLOCK(); 1331 uma_zfree(V_pf_state_key_z, sk); 1332 if (idx == PF_SK_STACK) 1333 pf_detach_state(s); 1334 return (EEXIST); /* collision! */ 1335 } 1336 } 1337 PF_HASHROW_UNLOCK(ih); 1338 } 1339 uma_zfree(V_pf_state_key_z, sk); 1340 s->key[idx] = cur; 1341 } else { 1342 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1343 s->key[idx] = sk; 1344 } 1345 1346 stateattach: 1347 /* List is sorted, if-bound states before floating. */ 1348 if (s->kif == V_pfi_all) 1349 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1350 else 1351 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1352 1353 if (olds) { 1354 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1355 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1356 key_list[idx]); 1357 olds = NULL; 1358 } 1359 1360 /* 1361 * Attach done. See how should we (or should not?) 1362 * attach a second key. 1363 */ 1364 if (sks == skw) { 1365 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1366 idx = PF_SK_STACK; 1367 sks = NULL; 1368 goto stateattach; 1369 } else if (sks != NULL) { 1370 /* 1371 * Continue attaching with stack key. 1372 */ 1373 sk = sks; 1374 kh = khs; 1375 idx = PF_SK_STACK; 1376 sks = NULL; 1377 goto keyattach; 1378 } 1379 1380 PF_STATE_LOCK(s); 1381 KEYS_UNLOCK(); 1382 1383 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1384 ("%s failure", __func__)); 1385 1386 return (0); 1387 #undef KEYS_UNLOCK 1388 } 1389 1390 static void 1391 pf_detach_state(struct pf_kstate *s) 1392 { 1393 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1394 struct pf_keyhash *kh; 1395 1396 NET_EPOCH_ASSERT(); 1397 1398 pf_sctp_multihome_detach_addr(s); 1399 1400 if (sks != NULL) { 1401 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1402 PF_HASHROW_LOCK(kh); 1403 if (s->key[PF_SK_STACK] != NULL) 1404 pf_state_key_detach(s, PF_SK_STACK); 1405 /* 1406 * If both point to same key, then we are done. 1407 */ 1408 if (sks == s->key[PF_SK_WIRE]) { 1409 pf_state_key_detach(s, PF_SK_WIRE); 1410 PF_HASHROW_UNLOCK(kh); 1411 return; 1412 } 1413 PF_HASHROW_UNLOCK(kh); 1414 } 1415 1416 if (s->key[PF_SK_WIRE] != NULL) { 1417 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1418 PF_HASHROW_LOCK(kh); 1419 if (s->key[PF_SK_WIRE] != NULL) 1420 pf_state_key_detach(s, PF_SK_WIRE); 1421 PF_HASHROW_UNLOCK(kh); 1422 } 1423 } 1424 1425 static void 1426 pf_state_key_detach(struct pf_kstate *s, int idx) 1427 { 1428 struct pf_state_key *sk = s->key[idx]; 1429 #ifdef INVARIANTS 1430 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1431 1432 PF_HASHROW_ASSERT(kh); 1433 #endif 1434 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1435 s->key[idx] = NULL; 1436 1437 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1438 LIST_REMOVE(sk, entry); 1439 uma_zfree(V_pf_state_key_z, sk); 1440 } 1441 } 1442 1443 static int 1444 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1445 { 1446 struct pf_state_key *sk = mem; 1447 1448 bzero(sk, sizeof(struct pf_state_key_cmp)); 1449 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1450 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1451 1452 return (0); 1453 } 1454 1455 struct pf_state_key * 1456 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1457 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1458 { 1459 struct pf_state_key *sk; 1460 1461 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1462 if (sk == NULL) 1463 return (NULL); 1464 1465 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1466 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1467 sk->port[pd->sidx] = sport; 1468 sk->port[pd->didx] = dport; 1469 sk->proto = pd->proto; 1470 sk->af = pd->af; 1471 1472 return (sk); 1473 } 1474 1475 struct pf_state_key * 1476 pf_state_key_clone(struct pf_state_key *orig) 1477 { 1478 struct pf_state_key *sk; 1479 1480 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1481 if (sk == NULL) 1482 return (NULL); 1483 1484 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1485 1486 return (sk); 1487 } 1488 1489 int 1490 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1491 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1492 { 1493 struct pf_idhash *ih; 1494 struct pf_kstate *cur; 1495 int error; 1496 1497 NET_EPOCH_ASSERT(); 1498 1499 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1500 ("%s: sks not pristine", __func__)); 1501 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1502 ("%s: skw not pristine", __func__)); 1503 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1504 1505 s->kif = kif; 1506 s->orig_kif = orig_kif; 1507 1508 if (s->id == 0 && s->creatorid == 0) { 1509 s->id = alloc_unr64(&V_pf_stateid); 1510 s->id = htobe64(s->id); 1511 s->creatorid = V_pf_status.hostid; 1512 } 1513 1514 /* Returns with ID locked on success. */ 1515 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1516 return (error); 1517 1518 ih = &V_pf_idhash[PF_IDHASH(s)]; 1519 PF_HASHROW_ASSERT(ih); 1520 LIST_FOREACH(cur, &ih->states, entry) 1521 if (cur->id == s->id && cur->creatorid == s->creatorid) 1522 break; 1523 1524 if (cur != NULL) { 1525 PF_HASHROW_UNLOCK(ih); 1526 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1527 printf("pf: state ID collision: " 1528 "id: %016llx creatorid: %08x\n", 1529 (unsigned long long)be64toh(s->id), 1530 ntohl(s->creatorid)); 1531 } 1532 pf_detach_state(s); 1533 return (EEXIST); 1534 } 1535 LIST_INSERT_HEAD(&ih->states, s, entry); 1536 /* One for keys, one for ID hash. */ 1537 refcount_init(&s->refs, 2); 1538 1539 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1540 if (V_pfsync_insert_state_ptr != NULL) 1541 V_pfsync_insert_state_ptr(s); 1542 1543 /* Returns locked. */ 1544 return (0); 1545 } 1546 1547 /* 1548 * Find state by ID: returns with locked row on success. 1549 */ 1550 struct pf_kstate * 1551 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1552 { 1553 struct pf_idhash *ih; 1554 struct pf_kstate *s; 1555 1556 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1557 1558 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1559 1560 PF_HASHROW_LOCK(ih); 1561 LIST_FOREACH(s, &ih->states, entry) 1562 if (s->id == id && s->creatorid == creatorid) 1563 break; 1564 1565 if (s == NULL) 1566 PF_HASHROW_UNLOCK(ih); 1567 1568 return (s); 1569 } 1570 1571 /* 1572 * Find state by key. 1573 * Returns with ID hash slot locked on success. 1574 */ 1575 static struct pf_kstate * 1576 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1577 { 1578 struct pf_keyhash *kh; 1579 struct pf_state_key *sk; 1580 struct pf_kstate *s; 1581 int idx; 1582 1583 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1584 1585 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1586 1587 PF_HASHROW_LOCK(kh); 1588 LIST_FOREACH(sk, &kh->keys, entry) 1589 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1590 break; 1591 if (sk == NULL) { 1592 PF_HASHROW_UNLOCK(kh); 1593 return (NULL); 1594 } 1595 1596 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1597 1598 /* List is sorted, if-bound states before floating ones. */ 1599 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1600 if (s->kif == V_pfi_all || s->kif == kif) { 1601 PF_STATE_LOCK(s); 1602 PF_HASHROW_UNLOCK(kh); 1603 if (__predict_false(s->timeout >= PFTM_MAX)) { 1604 /* 1605 * State is either being processed by 1606 * pf_unlink_state() in an other thread, or 1607 * is scheduled for immediate expiry. 1608 */ 1609 PF_STATE_UNLOCK(s); 1610 return (NULL); 1611 } 1612 return (s); 1613 } 1614 PF_HASHROW_UNLOCK(kh); 1615 1616 return (NULL); 1617 } 1618 1619 /* 1620 * Returns with ID hash slot locked on success. 1621 */ 1622 struct pf_kstate * 1623 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1624 { 1625 struct pf_keyhash *kh; 1626 struct pf_state_key *sk; 1627 struct pf_kstate *s, *ret = NULL; 1628 int idx, inout = 0; 1629 1630 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1631 1632 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1633 1634 PF_HASHROW_LOCK(kh); 1635 LIST_FOREACH(sk, &kh->keys, entry) 1636 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1637 break; 1638 if (sk == NULL) { 1639 PF_HASHROW_UNLOCK(kh); 1640 return (NULL); 1641 } 1642 switch (dir) { 1643 case PF_IN: 1644 idx = PF_SK_WIRE; 1645 break; 1646 case PF_OUT: 1647 idx = PF_SK_STACK; 1648 break; 1649 case PF_INOUT: 1650 idx = PF_SK_WIRE; 1651 inout = 1; 1652 break; 1653 default: 1654 panic("%s: dir %u", __func__, dir); 1655 } 1656 second_run: 1657 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1658 if (more == NULL) { 1659 PF_STATE_LOCK(s); 1660 PF_HASHROW_UNLOCK(kh); 1661 return (s); 1662 } 1663 1664 if (ret) 1665 (*more)++; 1666 else { 1667 ret = s; 1668 PF_STATE_LOCK(s); 1669 } 1670 } 1671 if (inout == 1) { 1672 inout = 0; 1673 idx = PF_SK_STACK; 1674 goto second_run; 1675 } 1676 PF_HASHROW_UNLOCK(kh); 1677 1678 return (ret); 1679 } 1680 1681 /* 1682 * FIXME 1683 * This routine is inefficient -- locks the state only to unlock immediately on 1684 * return. 1685 * It is racy -- after the state is unlocked nothing stops other threads from 1686 * removing it. 1687 */ 1688 bool 1689 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1690 { 1691 struct pf_kstate *s; 1692 1693 s = pf_find_state_all(key, dir, NULL); 1694 if (s != NULL) { 1695 PF_STATE_UNLOCK(s); 1696 return (true); 1697 } 1698 return (false); 1699 } 1700 1701 /* END state table stuff */ 1702 1703 static void 1704 pf_send(struct pf_send_entry *pfse) 1705 { 1706 1707 PF_SENDQ_LOCK(); 1708 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1709 PF_SENDQ_UNLOCK(); 1710 swi_sched(V_pf_swi_cookie, 0); 1711 } 1712 1713 static bool 1714 pf_isforlocal(struct mbuf *m, int af) 1715 { 1716 switch (af) { 1717 #ifdef INET 1718 case AF_INET: { 1719 struct ip *ip = mtod(m, struct ip *); 1720 1721 return (in_localip(ip->ip_dst)); 1722 } 1723 #endif 1724 #ifdef INET6 1725 case AF_INET6: { 1726 struct ip6_hdr *ip6; 1727 struct in6_ifaddr *ia; 1728 ip6 = mtod(m, struct ip6_hdr *); 1729 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1730 if (ia == NULL) 1731 return (false); 1732 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1733 } 1734 #endif 1735 default: 1736 panic("Unsupported af %d", af); 1737 } 1738 1739 return (false); 1740 } 1741 1742 void 1743 pf_intr(void *v) 1744 { 1745 struct epoch_tracker et; 1746 struct pf_send_head queue; 1747 struct pf_send_entry *pfse, *next; 1748 1749 CURVNET_SET((struct vnet *)v); 1750 1751 PF_SENDQ_LOCK(); 1752 queue = V_pf_sendqueue; 1753 STAILQ_INIT(&V_pf_sendqueue); 1754 PF_SENDQ_UNLOCK(); 1755 1756 NET_EPOCH_ENTER(et); 1757 1758 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1759 switch (pfse->pfse_type) { 1760 #ifdef INET 1761 case PFSE_IP: { 1762 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1763 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1764 pfse->pfse_m->m_pkthdr.csum_flags |= 1765 CSUM_IP_VALID | CSUM_IP_CHECKED; 1766 ip_input(pfse->pfse_m); 1767 } else { 1768 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1769 NULL); 1770 } 1771 break; 1772 } 1773 case PFSE_ICMP: 1774 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1775 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1776 break; 1777 #endif /* INET */ 1778 #ifdef INET6 1779 case PFSE_IP6: 1780 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1781 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1782 ip6_input(pfse->pfse_m); 1783 } else { 1784 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1785 NULL, NULL); 1786 } 1787 break; 1788 case PFSE_ICMP6: 1789 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1790 pfse->icmpopts.code, pfse->icmpopts.mtu); 1791 break; 1792 #endif /* INET6 */ 1793 default: 1794 panic("%s: unknown type", __func__); 1795 } 1796 free(pfse, M_PFTEMP); 1797 } 1798 NET_EPOCH_EXIT(et); 1799 CURVNET_RESTORE(); 1800 } 1801 1802 #define pf_purge_thread_period (hz / 10) 1803 1804 #ifdef PF_WANT_32_TO_64_COUNTER 1805 static void 1806 pf_status_counter_u64_periodic(void) 1807 { 1808 1809 PF_RULES_RASSERT(); 1810 1811 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1812 return; 1813 } 1814 1815 for (int i = 0; i < FCNT_MAX; i++) { 1816 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1817 } 1818 } 1819 1820 static void 1821 pf_kif_counter_u64_periodic(void) 1822 { 1823 struct pfi_kkif *kif; 1824 size_t r, run; 1825 1826 PF_RULES_RASSERT(); 1827 1828 if (__predict_false(V_pf_allkifcount == 0)) { 1829 return; 1830 } 1831 1832 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1833 return; 1834 } 1835 1836 run = V_pf_allkifcount / 10; 1837 if (run < 5) 1838 run = 5; 1839 1840 for (r = 0; r < run; r++) { 1841 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1842 if (kif == NULL) { 1843 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1844 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1845 break; 1846 } 1847 1848 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1849 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1850 1851 for (int i = 0; i < 2; i++) { 1852 for (int j = 0; j < 2; j++) { 1853 for (int k = 0; k < 2; k++) { 1854 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1855 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1856 } 1857 } 1858 } 1859 } 1860 } 1861 1862 static void 1863 pf_rule_counter_u64_periodic(void) 1864 { 1865 struct pf_krule *rule; 1866 size_t r, run; 1867 1868 PF_RULES_RASSERT(); 1869 1870 if (__predict_false(V_pf_allrulecount == 0)) { 1871 return; 1872 } 1873 1874 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1875 return; 1876 } 1877 1878 run = V_pf_allrulecount / 10; 1879 if (run < 5) 1880 run = 5; 1881 1882 for (r = 0; r < run; r++) { 1883 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1884 if (rule == NULL) { 1885 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1886 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1887 break; 1888 } 1889 1890 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1891 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1892 1893 pf_counter_u64_periodic(&rule->evaluations); 1894 for (int i = 0; i < 2; i++) { 1895 pf_counter_u64_periodic(&rule->packets[i]); 1896 pf_counter_u64_periodic(&rule->bytes[i]); 1897 } 1898 } 1899 } 1900 1901 static void 1902 pf_counter_u64_periodic_main(void) 1903 { 1904 PF_RULES_RLOCK_TRACKER; 1905 1906 V_pf_counter_periodic_iter++; 1907 1908 PF_RULES_RLOCK(); 1909 pf_counter_u64_critical_enter(); 1910 pf_status_counter_u64_periodic(); 1911 pf_kif_counter_u64_periodic(); 1912 pf_rule_counter_u64_periodic(); 1913 pf_counter_u64_critical_exit(); 1914 PF_RULES_RUNLOCK(); 1915 } 1916 #else 1917 #define pf_counter_u64_periodic_main() do { } while (0) 1918 #endif 1919 1920 void 1921 pf_purge_thread(void *unused __unused) 1922 { 1923 struct epoch_tracker et; 1924 1925 VNET_ITERATOR_DECL(vnet_iter); 1926 1927 sx_xlock(&pf_end_lock); 1928 while (pf_end_threads == 0) { 1929 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1930 1931 VNET_LIST_RLOCK(); 1932 NET_EPOCH_ENTER(et); 1933 VNET_FOREACH(vnet_iter) { 1934 CURVNET_SET(vnet_iter); 1935 1936 /* Wait until V_pf_default_rule is initialized. */ 1937 if (V_pf_vnet_active == 0) { 1938 CURVNET_RESTORE(); 1939 continue; 1940 } 1941 1942 pf_counter_u64_periodic_main(); 1943 1944 /* 1945 * Process 1/interval fraction of the state 1946 * table every run. 1947 */ 1948 V_pf_purge_idx = 1949 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1950 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1951 1952 /* 1953 * Purge other expired types every 1954 * PFTM_INTERVAL seconds. 1955 */ 1956 if (V_pf_purge_idx == 0) { 1957 /* 1958 * Order is important: 1959 * - states and src nodes reference rules 1960 * - states and rules reference kifs 1961 */ 1962 pf_purge_expired_fragments(); 1963 pf_purge_expired_src_nodes(); 1964 pf_purge_unlinked_rules(); 1965 pfi_kkif_purge(); 1966 } 1967 CURVNET_RESTORE(); 1968 } 1969 NET_EPOCH_EXIT(et); 1970 VNET_LIST_RUNLOCK(); 1971 } 1972 1973 pf_end_threads++; 1974 sx_xunlock(&pf_end_lock); 1975 kproc_exit(0); 1976 } 1977 1978 void 1979 pf_unload_vnet_purge(void) 1980 { 1981 1982 /* 1983 * To cleanse up all kifs and rules we need 1984 * two runs: first one clears reference flags, 1985 * then pf_purge_expired_states() doesn't 1986 * raise them, and then second run frees. 1987 */ 1988 pf_purge_unlinked_rules(); 1989 pfi_kkif_purge(); 1990 1991 /* 1992 * Now purge everything. 1993 */ 1994 pf_purge_expired_states(0, pf_hashmask); 1995 pf_purge_fragments(UINT_MAX); 1996 pf_purge_expired_src_nodes(); 1997 1998 /* 1999 * Now all kifs & rules should be unreferenced, 2000 * thus should be successfully freed. 2001 */ 2002 pf_purge_unlinked_rules(); 2003 pfi_kkif_purge(); 2004 } 2005 2006 u_int32_t 2007 pf_state_expires(const struct pf_kstate *state) 2008 { 2009 u_int32_t timeout; 2010 u_int32_t start; 2011 u_int32_t end; 2012 u_int32_t states; 2013 2014 /* handle all PFTM_* > PFTM_MAX here */ 2015 if (state->timeout == PFTM_PURGE) 2016 return (time_uptime); 2017 KASSERT(state->timeout != PFTM_UNLINKED, 2018 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2019 KASSERT((state->timeout < PFTM_MAX), 2020 ("pf_state_expires: timeout > PFTM_MAX")); 2021 timeout = state->rule.ptr->timeout[state->timeout]; 2022 if (!timeout) 2023 timeout = V_pf_default_rule.timeout[state->timeout]; 2024 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2025 if (start && state->rule.ptr != &V_pf_default_rule) { 2026 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2027 states = counter_u64_fetch(state->rule.ptr->states_cur); 2028 } else { 2029 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2030 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2031 states = V_pf_status.states; 2032 } 2033 if (end && states > start && start < end) { 2034 if (states < end) { 2035 timeout = (u_int64_t)timeout * (end - states) / 2036 (end - start); 2037 return (state->expire + timeout); 2038 } 2039 else 2040 return (time_uptime); 2041 } 2042 return (state->expire + timeout); 2043 } 2044 2045 void 2046 pf_purge_expired_src_nodes(void) 2047 { 2048 struct pf_ksrc_node_list freelist; 2049 struct pf_srchash *sh; 2050 struct pf_ksrc_node *cur, *next; 2051 int i; 2052 2053 LIST_INIT(&freelist); 2054 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 2055 PF_HASHROW_LOCK(sh); 2056 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2057 if (cur->states == 0 && cur->expire <= time_uptime) { 2058 pf_unlink_src_node(cur); 2059 LIST_INSERT_HEAD(&freelist, cur, entry); 2060 } else if (cur->rule.ptr != NULL) 2061 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2062 PF_HASHROW_UNLOCK(sh); 2063 } 2064 2065 pf_free_src_nodes(&freelist); 2066 2067 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2068 } 2069 2070 static void 2071 pf_src_tree_remove_state(struct pf_kstate *s) 2072 { 2073 struct pf_ksrc_node *sn; 2074 uint32_t timeout; 2075 2076 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2077 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2078 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2079 2080 if (s->src_node != NULL) { 2081 sn = s->src_node; 2082 PF_SRC_NODE_LOCK(sn); 2083 if (s->src.tcp_est) 2084 --sn->conn; 2085 if (--sn->states == 0) 2086 sn->expire = time_uptime + timeout; 2087 PF_SRC_NODE_UNLOCK(sn); 2088 } 2089 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2090 sn = s->nat_src_node; 2091 PF_SRC_NODE_LOCK(sn); 2092 if (--sn->states == 0) 2093 sn->expire = time_uptime + timeout; 2094 PF_SRC_NODE_UNLOCK(sn); 2095 } 2096 s->src_node = s->nat_src_node = NULL; 2097 } 2098 2099 /* 2100 * Unlink and potentilly free a state. Function may be 2101 * called with ID hash row locked, but always returns 2102 * unlocked, since it needs to go through key hash locking. 2103 */ 2104 int 2105 pf_unlink_state(struct pf_kstate *s) 2106 { 2107 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2108 2109 NET_EPOCH_ASSERT(); 2110 PF_HASHROW_ASSERT(ih); 2111 2112 if (s->timeout == PFTM_UNLINKED) { 2113 /* 2114 * State is being processed 2115 * by pf_unlink_state() in 2116 * an other thread. 2117 */ 2118 PF_HASHROW_UNLOCK(ih); 2119 return (0); /* XXXGL: undefined actually */ 2120 } 2121 2122 if (s->src.state == PF_TCPS_PROXY_DST) { 2123 /* XXX wire key the right one? */ 2124 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2125 &s->key[PF_SK_WIRE]->addr[1], 2126 &s->key[PF_SK_WIRE]->addr[0], 2127 s->key[PF_SK_WIRE]->port[1], 2128 s->key[PF_SK_WIRE]->port[0], 2129 s->src.seqhi, s->src.seqlo + 1, 2130 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2131 } 2132 2133 LIST_REMOVE(s, entry); 2134 pf_src_tree_remove_state(s); 2135 2136 if (V_pfsync_delete_state_ptr != NULL) 2137 V_pfsync_delete_state_ptr(s); 2138 2139 STATE_DEC_COUNTERS(s); 2140 2141 s->timeout = PFTM_UNLINKED; 2142 2143 /* Ensure we remove it from the list of halfopen states, if needed. */ 2144 if (s->key[PF_SK_STACK] != NULL && 2145 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2146 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2147 2148 PF_HASHROW_UNLOCK(ih); 2149 2150 pf_detach_state(s); 2151 /* pf_state_insert() initialises refs to 2 */ 2152 return (pf_release_staten(s, 2)); 2153 } 2154 2155 struct pf_kstate * 2156 pf_alloc_state(int flags) 2157 { 2158 2159 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2160 } 2161 2162 void 2163 pf_free_state(struct pf_kstate *cur) 2164 { 2165 struct pf_krule_item *ri; 2166 2167 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2168 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2169 cur->timeout)); 2170 2171 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2172 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2173 free(ri, M_PF_RULE_ITEM); 2174 } 2175 2176 pf_normalize_tcp_cleanup(cur); 2177 uma_zfree(V_pf_state_z, cur); 2178 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2179 } 2180 2181 /* 2182 * Called only from pf_purge_thread(), thus serialized. 2183 */ 2184 static u_int 2185 pf_purge_expired_states(u_int i, int maxcheck) 2186 { 2187 struct pf_idhash *ih; 2188 struct pf_kstate *s; 2189 struct pf_krule_item *mrm; 2190 size_t count __unused; 2191 2192 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2193 2194 /* 2195 * Go through hash and unlink states that expire now. 2196 */ 2197 while (maxcheck > 0) { 2198 count = 0; 2199 ih = &V_pf_idhash[i]; 2200 2201 /* only take the lock if we expect to do work */ 2202 if (!LIST_EMPTY(&ih->states)) { 2203 relock: 2204 PF_HASHROW_LOCK(ih); 2205 LIST_FOREACH(s, &ih->states, entry) { 2206 if (pf_state_expires(s) <= time_uptime) { 2207 V_pf_status.states -= 2208 pf_unlink_state(s); 2209 goto relock; 2210 } 2211 s->rule.ptr->rule_ref |= PFRULE_REFS; 2212 if (s->nat_rule.ptr != NULL) 2213 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2214 if (s->anchor.ptr != NULL) 2215 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2216 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2217 SLIST_FOREACH(mrm, &s->match_rules, entry) 2218 mrm->r->rule_ref |= PFRULE_REFS; 2219 if (s->rt_kif) 2220 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2221 count++; 2222 } 2223 PF_HASHROW_UNLOCK(ih); 2224 } 2225 2226 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2227 2228 /* Return when we hit end of hash. */ 2229 if (++i > pf_hashmask) { 2230 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2231 return (0); 2232 } 2233 2234 maxcheck--; 2235 } 2236 2237 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2238 2239 return (i); 2240 } 2241 2242 static void 2243 pf_purge_unlinked_rules(void) 2244 { 2245 struct pf_krulequeue tmpq; 2246 struct pf_krule *r, *r1; 2247 2248 /* 2249 * If we have overloading task pending, then we'd 2250 * better skip purging this time. There is a tiny 2251 * probability that overloading task references 2252 * an already unlinked rule. 2253 */ 2254 PF_OVERLOADQ_LOCK(); 2255 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2256 PF_OVERLOADQ_UNLOCK(); 2257 return; 2258 } 2259 PF_OVERLOADQ_UNLOCK(); 2260 2261 /* 2262 * Do naive mark-and-sweep garbage collecting of old rules. 2263 * Reference flag is raised by pf_purge_expired_states() 2264 * and pf_purge_expired_src_nodes(). 2265 * 2266 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2267 * use a temporary queue. 2268 */ 2269 TAILQ_INIT(&tmpq); 2270 PF_UNLNKDRULES_LOCK(); 2271 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2272 if (!(r->rule_ref & PFRULE_REFS)) { 2273 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2274 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2275 } else 2276 r->rule_ref &= ~PFRULE_REFS; 2277 } 2278 PF_UNLNKDRULES_UNLOCK(); 2279 2280 if (!TAILQ_EMPTY(&tmpq)) { 2281 PF_CONFIG_LOCK(); 2282 PF_RULES_WLOCK(); 2283 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2284 TAILQ_REMOVE(&tmpq, r, entries); 2285 pf_free_rule(r); 2286 } 2287 PF_RULES_WUNLOCK(); 2288 PF_CONFIG_UNLOCK(); 2289 } 2290 } 2291 2292 void 2293 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2294 { 2295 switch (af) { 2296 #ifdef INET 2297 case AF_INET: { 2298 u_int32_t a = ntohl(addr->addr32[0]); 2299 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2300 (a>>8)&255, a&255); 2301 if (p) { 2302 p = ntohs(p); 2303 printf(":%u", p); 2304 } 2305 break; 2306 } 2307 #endif /* INET */ 2308 #ifdef INET6 2309 case AF_INET6: { 2310 u_int16_t b; 2311 u_int8_t i, curstart, curend, maxstart, maxend; 2312 curstart = curend = maxstart = maxend = 255; 2313 for (i = 0; i < 8; i++) { 2314 if (!addr->addr16[i]) { 2315 if (curstart == 255) 2316 curstart = i; 2317 curend = i; 2318 } else { 2319 if ((curend - curstart) > 2320 (maxend - maxstart)) { 2321 maxstart = curstart; 2322 maxend = curend; 2323 } 2324 curstart = curend = 255; 2325 } 2326 } 2327 if ((curend - curstart) > 2328 (maxend - maxstart)) { 2329 maxstart = curstart; 2330 maxend = curend; 2331 } 2332 for (i = 0; i < 8; i++) { 2333 if (i >= maxstart && i <= maxend) { 2334 if (i == 0) 2335 printf(":"); 2336 if (i == maxend) 2337 printf(":"); 2338 } else { 2339 b = ntohs(addr->addr16[i]); 2340 printf("%x", b); 2341 if (i < 7) 2342 printf(":"); 2343 } 2344 } 2345 if (p) { 2346 p = ntohs(p); 2347 printf("[%u]", p); 2348 } 2349 break; 2350 } 2351 #endif /* INET6 */ 2352 } 2353 } 2354 2355 void 2356 pf_print_state(struct pf_kstate *s) 2357 { 2358 pf_print_state_parts(s, NULL, NULL); 2359 } 2360 2361 static void 2362 pf_print_state_parts(struct pf_kstate *s, 2363 struct pf_state_key *skwp, struct pf_state_key *sksp) 2364 { 2365 struct pf_state_key *skw, *sks; 2366 u_int8_t proto, dir; 2367 2368 /* Do our best to fill these, but they're skipped if NULL */ 2369 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2370 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2371 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2372 dir = s ? s->direction : 0; 2373 2374 switch (proto) { 2375 case IPPROTO_IPV4: 2376 printf("IPv4"); 2377 break; 2378 case IPPROTO_IPV6: 2379 printf("IPv6"); 2380 break; 2381 case IPPROTO_TCP: 2382 printf("TCP"); 2383 break; 2384 case IPPROTO_UDP: 2385 printf("UDP"); 2386 break; 2387 case IPPROTO_ICMP: 2388 printf("ICMP"); 2389 break; 2390 case IPPROTO_ICMPV6: 2391 printf("ICMPv6"); 2392 break; 2393 default: 2394 printf("%u", proto); 2395 break; 2396 } 2397 switch (dir) { 2398 case PF_IN: 2399 printf(" in"); 2400 break; 2401 case PF_OUT: 2402 printf(" out"); 2403 break; 2404 } 2405 if (skw) { 2406 printf(" wire: "); 2407 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2408 printf(" "); 2409 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2410 } 2411 if (sks) { 2412 printf(" stack: "); 2413 if (sks != skw) { 2414 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2415 printf(" "); 2416 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2417 } else 2418 printf("-"); 2419 } 2420 if (s) { 2421 if (proto == IPPROTO_TCP) { 2422 printf(" [lo=%u high=%u win=%u modulator=%u", 2423 s->src.seqlo, s->src.seqhi, 2424 s->src.max_win, s->src.seqdiff); 2425 if (s->src.wscale && s->dst.wscale) 2426 printf(" wscale=%u", 2427 s->src.wscale & PF_WSCALE_MASK); 2428 printf("]"); 2429 printf(" [lo=%u high=%u win=%u modulator=%u", 2430 s->dst.seqlo, s->dst.seqhi, 2431 s->dst.max_win, s->dst.seqdiff); 2432 if (s->src.wscale && s->dst.wscale) 2433 printf(" wscale=%u", 2434 s->dst.wscale & PF_WSCALE_MASK); 2435 printf("]"); 2436 } 2437 printf(" %u:%u", s->src.state, s->dst.state); 2438 } 2439 } 2440 2441 void 2442 pf_print_flags(u_int8_t f) 2443 { 2444 if (f) 2445 printf(" "); 2446 if (f & TH_FIN) 2447 printf("F"); 2448 if (f & TH_SYN) 2449 printf("S"); 2450 if (f & TH_RST) 2451 printf("R"); 2452 if (f & TH_PUSH) 2453 printf("P"); 2454 if (f & TH_ACK) 2455 printf("A"); 2456 if (f & TH_URG) 2457 printf("U"); 2458 if (f & TH_ECE) 2459 printf("E"); 2460 if (f & TH_CWR) 2461 printf("W"); 2462 } 2463 2464 #define PF_SET_SKIP_STEPS(i) \ 2465 do { \ 2466 while (head[i] != cur) { \ 2467 head[i]->skip[i].ptr = cur; \ 2468 head[i] = TAILQ_NEXT(head[i], entries); \ 2469 } \ 2470 } while (0) 2471 2472 void 2473 pf_calc_skip_steps(struct pf_krulequeue *rules) 2474 { 2475 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2476 int i; 2477 2478 cur = TAILQ_FIRST(rules); 2479 prev = cur; 2480 for (i = 0; i < PF_SKIP_COUNT; ++i) 2481 head[i] = cur; 2482 while (cur != NULL) { 2483 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2484 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2485 if (cur->direction != prev->direction) 2486 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2487 if (cur->af != prev->af) 2488 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2489 if (cur->proto != prev->proto) 2490 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2491 if (cur->src.neg != prev->src.neg || 2492 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2493 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2494 if (cur->src.port[0] != prev->src.port[0] || 2495 cur->src.port[1] != prev->src.port[1] || 2496 cur->src.port_op != prev->src.port_op) 2497 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2498 if (cur->dst.neg != prev->dst.neg || 2499 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2500 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2501 if (cur->dst.port[0] != prev->dst.port[0] || 2502 cur->dst.port[1] != prev->dst.port[1] || 2503 cur->dst.port_op != prev->dst.port_op) 2504 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2505 2506 prev = cur; 2507 cur = TAILQ_NEXT(cur, entries); 2508 } 2509 for (i = 0; i < PF_SKIP_COUNT; ++i) 2510 PF_SET_SKIP_STEPS(i); 2511 } 2512 2513 int 2514 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2515 { 2516 if (aw1->type != aw2->type) 2517 return (1); 2518 switch (aw1->type) { 2519 case PF_ADDR_ADDRMASK: 2520 case PF_ADDR_RANGE: 2521 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2522 return (1); 2523 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2524 return (1); 2525 return (0); 2526 case PF_ADDR_DYNIFTL: 2527 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2528 case PF_ADDR_NOROUTE: 2529 case PF_ADDR_URPFFAILED: 2530 return (0); 2531 case PF_ADDR_TABLE: 2532 return (aw1->p.tbl != aw2->p.tbl); 2533 default: 2534 printf("invalid address type: %d\n", aw1->type); 2535 return (1); 2536 } 2537 } 2538 2539 /** 2540 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2541 * header isn't always a full checksum. In some cases (i.e. output) it's a 2542 * pseudo-header checksum, which is a partial checksum over src/dst IP 2543 * addresses, protocol number and length. 2544 * 2545 * That means we have the following cases: 2546 * * Input or forwarding: we don't have TSO, the checksum fields are full 2547 * checksums, we need to update the checksum whenever we change anything. 2548 * * Output (i.e. the checksum is a pseudo-header checksum): 2549 * x The field being updated is src/dst address or affects the length of 2550 * the packet. We need to update the pseudo-header checksum (note that this 2551 * checksum is not ones' complement). 2552 * x Some other field is being modified (e.g. src/dst port numbers): We 2553 * don't have to update anything. 2554 **/ 2555 u_int16_t 2556 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2557 { 2558 u_int32_t x; 2559 2560 x = cksum + old - new; 2561 x = (x + (x >> 16)) & 0xffff; 2562 2563 /* optimise: eliminate a branch when not udp */ 2564 if (udp && cksum == 0x0000) 2565 return cksum; 2566 if (udp && x == 0x0000) 2567 x = 0xffff; 2568 2569 return (u_int16_t)(x); 2570 } 2571 2572 static void 2573 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2574 u_int8_t udp) 2575 { 2576 u_int16_t old = htons(hi ? (*f << 8) : *f); 2577 u_int16_t new = htons(hi ? ( v << 8) : v); 2578 2579 if (*f == v) 2580 return; 2581 2582 *f = v; 2583 2584 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2585 return; 2586 2587 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2588 } 2589 2590 void 2591 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2592 bool hi, u_int8_t udp) 2593 { 2594 u_int8_t *fb = (u_int8_t *)f; 2595 u_int8_t *vb = (u_int8_t *)&v; 2596 2597 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2598 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2599 } 2600 2601 void 2602 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2603 bool hi, u_int8_t udp) 2604 { 2605 u_int8_t *fb = (u_int8_t *)f; 2606 u_int8_t *vb = (u_int8_t *)&v; 2607 2608 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2609 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2610 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2611 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2612 } 2613 2614 u_int16_t 2615 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2616 u_int16_t new, u_int8_t udp) 2617 { 2618 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2619 return (cksum); 2620 2621 return (pf_cksum_fixup(cksum, old, new, udp)); 2622 } 2623 2624 static void 2625 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2626 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2627 sa_family_t af) 2628 { 2629 struct pf_addr ao; 2630 u_int16_t po = *p; 2631 2632 PF_ACPY(&ao, a, af); 2633 PF_ACPY(a, an, af); 2634 2635 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2636 *pc = ~*pc; 2637 2638 *p = pn; 2639 2640 switch (af) { 2641 #ifdef INET 2642 case AF_INET: 2643 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2644 ao.addr16[0], an->addr16[0], 0), 2645 ao.addr16[1], an->addr16[1], 0); 2646 *p = pn; 2647 2648 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2649 ao.addr16[0], an->addr16[0], u), 2650 ao.addr16[1], an->addr16[1], u); 2651 2652 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2653 break; 2654 #endif /* INET */ 2655 #ifdef INET6 2656 case AF_INET6: 2657 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2658 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2659 pf_cksum_fixup(pf_cksum_fixup(*pc, 2660 ao.addr16[0], an->addr16[0], u), 2661 ao.addr16[1], an->addr16[1], u), 2662 ao.addr16[2], an->addr16[2], u), 2663 ao.addr16[3], an->addr16[3], u), 2664 ao.addr16[4], an->addr16[4], u), 2665 ao.addr16[5], an->addr16[5], u), 2666 ao.addr16[6], an->addr16[6], u), 2667 ao.addr16[7], an->addr16[7], u); 2668 2669 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2670 break; 2671 #endif /* INET6 */ 2672 } 2673 2674 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2675 CSUM_DELAY_DATA_IPV6)) { 2676 *pc = ~*pc; 2677 if (! *pc) 2678 *pc = 0xffff; 2679 } 2680 } 2681 2682 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2683 void 2684 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2685 { 2686 u_int32_t ao; 2687 2688 memcpy(&ao, a, sizeof(ao)); 2689 memcpy(a, &an, sizeof(u_int32_t)); 2690 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2691 ao % 65536, an % 65536, u); 2692 } 2693 2694 void 2695 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2696 { 2697 u_int32_t ao; 2698 2699 memcpy(&ao, a, sizeof(ao)); 2700 memcpy(a, &an, sizeof(u_int32_t)); 2701 2702 *c = pf_proto_cksum_fixup(m, 2703 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2704 ao % 65536, an % 65536, udp); 2705 } 2706 2707 #ifdef INET6 2708 static void 2709 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2710 { 2711 struct pf_addr ao; 2712 2713 PF_ACPY(&ao, a, AF_INET6); 2714 PF_ACPY(a, an, AF_INET6); 2715 2716 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2717 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2718 pf_cksum_fixup(pf_cksum_fixup(*c, 2719 ao.addr16[0], an->addr16[0], u), 2720 ao.addr16[1], an->addr16[1], u), 2721 ao.addr16[2], an->addr16[2], u), 2722 ao.addr16[3], an->addr16[3], u), 2723 ao.addr16[4], an->addr16[4], u), 2724 ao.addr16[5], an->addr16[5], u), 2725 ao.addr16[6], an->addr16[6], u), 2726 ao.addr16[7], an->addr16[7], u); 2727 } 2728 #endif /* INET6 */ 2729 2730 static void 2731 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2732 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2733 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2734 { 2735 struct pf_addr oia, ooa; 2736 2737 PF_ACPY(&oia, ia, af); 2738 if (oa) 2739 PF_ACPY(&ooa, oa, af); 2740 2741 /* Change inner protocol port, fix inner protocol checksum. */ 2742 if (ip != NULL) { 2743 u_int16_t oip = *ip; 2744 u_int32_t opc; 2745 2746 if (pc != NULL) 2747 opc = *pc; 2748 *ip = np; 2749 if (pc != NULL) 2750 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2751 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2752 if (pc != NULL) 2753 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2754 } 2755 /* Change inner ip address, fix inner ip and icmp checksums. */ 2756 PF_ACPY(ia, na, af); 2757 switch (af) { 2758 #ifdef INET 2759 case AF_INET: { 2760 u_int32_t oh2c = *h2c; 2761 2762 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2763 oia.addr16[0], ia->addr16[0], 0), 2764 oia.addr16[1], ia->addr16[1], 0); 2765 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2766 oia.addr16[0], ia->addr16[0], 0), 2767 oia.addr16[1], ia->addr16[1], 0); 2768 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2769 break; 2770 } 2771 #endif /* INET */ 2772 #ifdef INET6 2773 case AF_INET6: 2774 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2775 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2776 pf_cksum_fixup(pf_cksum_fixup(*ic, 2777 oia.addr16[0], ia->addr16[0], u), 2778 oia.addr16[1], ia->addr16[1], u), 2779 oia.addr16[2], ia->addr16[2], u), 2780 oia.addr16[3], ia->addr16[3], u), 2781 oia.addr16[4], ia->addr16[4], u), 2782 oia.addr16[5], ia->addr16[5], u), 2783 oia.addr16[6], ia->addr16[6], u), 2784 oia.addr16[7], ia->addr16[7], u); 2785 break; 2786 #endif /* INET6 */ 2787 } 2788 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2789 if (oa) { 2790 PF_ACPY(oa, na, af); 2791 switch (af) { 2792 #ifdef INET 2793 case AF_INET: 2794 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2795 ooa.addr16[0], oa->addr16[0], 0), 2796 ooa.addr16[1], oa->addr16[1], 0); 2797 break; 2798 #endif /* INET */ 2799 #ifdef INET6 2800 case AF_INET6: 2801 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2802 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2803 pf_cksum_fixup(pf_cksum_fixup(*ic, 2804 ooa.addr16[0], oa->addr16[0], u), 2805 ooa.addr16[1], oa->addr16[1], u), 2806 ooa.addr16[2], oa->addr16[2], u), 2807 ooa.addr16[3], oa->addr16[3], u), 2808 ooa.addr16[4], oa->addr16[4], u), 2809 ooa.addr16[5], oa->addr16[5], u), 2810 ooa.addr16[6], oa->addr16[6], u), 2811 ooa.addr16[7], oa->addr16[7], u); 2812 break; 2813 #endif /* INET6 */ 2814 } 2815 } 2816 } 2817 2818 /* 2819 * Need to modulate the sequence numbers in the TCP SACK option 2820 * (credits to Krzysztof Pfaff for report and patch) 2821 */ 2822 static int 2823 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2824 struct tcphdr *th, struct pf_state_peer *dst) 2825 { 2826 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2827 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2828 int copyback = 0, i, olen; 2829 struct sackblk sack; 2830 2831 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2832 if (hlen < TCPOLEN_SACKLEN || 2833 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2834 return 0; 2835 2836 while (hlen >= TCPOLEN_SACKLEN) { 2837 size_t startoff = opt - opts; 2838 olen = opt[1]; 2839 switch (*opt) { 2840 case TCPOPT_EOL: /* FALLTHROUGH */ 2841 case TCPOPT_NOP: 2842 opt++; 2843 hlen--; 2844 break; 2845 case TCPOPT_SACK: 2846 if (olen > hlen) 2847 olen = hlen; 2848 if (olen >= TCPOLEN_SACKLEN) { 2849 for (i = 2; i + TCPOLEN_SACK <= olen; 2850 i += TCPOLEN_SACK) { 2851 memcpy(&sack, &opt[i], sizeof(sack)); 2852 pf_patch_32_unaligned(m, 2853 &th->th_sum, &sack.start, 2854 htonl(ntohl(sack.start) - dst->seqdiff), 2855 PF_ALGNMNT(startoff), 2856 0); 2857 pf_patch_32_unaligned(m, &th->th_sum, 2858 &sack.end, 2859 htonl(ntohl(sack.end) - dst->seqdiff), 2860 PF_ALGNMNT(startoff), 2861 0); 2862 memcpy(&opt[i], &sack, sizeof(sack)); 2863 } 2864 copyback = 1; 2865 } 2866 /* FALLTHROUGH */ 2867 default: 2868 if (olen < 2) 2869 olen = 2; 2870 hlen -= olen; 2871 opt += olen; 2872 } 2873 } 2874 2875 if (copyback) 2876 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2877 return (copyback); 2878 } 2879 2880 struct mbuf * 2881 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2882 const struct pf_addr *saddr, const struct pf_addr *daddr, 2883 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2884 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 2885 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 2886 { 2887 struct mbuf *m; 2888 int len, tlen; 2889 #ifdef INET 2890 struct ip *h = NULL; 2891 #endif /* INET */ 2892 #ifdef INET6 2893 struct ip6_hdr *h6 = NULL; 2894 #endif /* INET6 */ 2895 struct tcphdr *th; 2896 char *opt; 2897 struct pf_mtag *pf_mtag; 2898 2899 len = 0; 2900 th = NULL; 2901 2902 /* maximum segment size tcp option */ 2903 tlen = sizeof(struct tcphdr); 2904 if (mss) 2905 tlen += 4; 2906 2907 switch (af) { 2908 #ifdef INET 2909 case AF_INET: 2910 len = sizeof(struct ip) + tlen; 2911 break; 2912 #endif /* INET */ 2913 #ifdef INET6 2914 case AF_INET6: 2915 len = sizeof(struct ip6_hdr) + tlen; 2916 break; 2917 #endif /* INET6 */ 2918 default: 2919 panic("%s: unsupported af %d", __func__, af); 2920 } 2921 2922 m = m_gethdr(M_NOWAIT, MT_DATA); 2923 if (m == NULL) 2924 return (NULL); 2925 2926 #ifdef MAC 2927 mac_netinet_firewall_send(m); 2928 #endif 2929 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2930 m_freem(m); 2931 return (NULL); 2932 } 2933 if (skip_firewall) 2934 m->m_flags |= M_SKIP_FIREWALL; 2935 pf_mtag->tag = mtag_tag; 2936 pf_mtag->flags = mtag_flags; 2937 2938 if (rtableid >= 0) 2939 M_SETFIB(m, rtableid); 2940 2941 #ifdef ALTQ 2942 if (r != NULL && r->qid) { 2943 pf_mtag->qid = r->qid; 2944 2945 /* add hints for ecn */ 2946 pf_mtag->hdr = mtod(m, struct ip *); 2947 } 2948 #endif /* ALTQ */ 2949 m->m_data += max_linkhdr; 2950 m->m_pkthdr.len = m->m_len = len; 2951 /* The rest of the stack assumes a rcvif, so provide one. 2952 * This is a locally generated packet, so .. close enough. */ 2953 m->m_pkthdr.rcvif = V_loif; 2954 bzero(m->m_data, len); 2955 switch (af) { 2956 #ifdef INET 2957 case AF_INET: 2958 h = mtod(m, struct ip *); 2959 2960 /* IP header fields included in the TCP checksum */ 2961 h->ip_p = IPPROTO_TCP; 2962 h->ip_len = htons(tlen); 2963 h->ip_src.s_addr = saddr->v4.s_addr; 2964 h->ip_dst.s_addr = daddr->v4.s_addr; 2965 2966 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2967 break; 2968 #endif /* INET */ 2969 #ifdef INET6 2970 case AF_INET6: 2971 h6 = mtod(m, struct ip6_hdr *); 2972 2973 /* IP header fields included in the TCP checksum */ 2974 h6->ip6_nxt = IPPROTO_TCP; 2975 h6->ip6_plen = htons(tlen); 2976 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2977 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2978 2979 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2980 break; 2981 #endif /* INET6 */ 2982 } 2983 2984 /* TCP header */ 2985 th->th_sport = sport; 2986 th->th_dport = dport; 2987 th->th_seq = htonl(seq); 2988 th->th_ack = htonl(ack); 2989 th->th_off = tlen >> 2; 2990 th->th_flags = tcp_flags; 2991 th->th_win = htons(win); 2992 2993 if (mss) { 2994 opt = (char *)(th + 1); 2995 opt[0] = TCPOPT_MAXSEG; 2996 opt[1] = 4; 2997 HTONS(mss); 2998 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2999 } 3000 3001 switch (af) { 3002 #ifdef INET 3003 case AF_INET: 3004 /* TCP checksum */ 3005 th->th_sum = in_cksum(m, len); 3006 3007 /* Finish the IP header */ 3008 h->ip_v = 4; 3009 h->ip_hl = sizeof(*h) >> 2; 3010 h->ip_tos = IPTOS_LOWDELAY; 3011 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3012 h->ip_len = htons(len); 3013 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3014 h->ip_sum = 0; 3015 break; 3016 #endif /* INET */ 3017 #ifdef INET6 3018 case AF_INET6: 3019 /* TCP checksum */ 3020 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3021 sizeof(struct ip6_hdr), tlen); 3022 3023 h6->ip6_vfc |= IPV6_VERSION; 3024 h6->ip6_hlim = IPV6_DEFHLIM; 3025 break; 3026 #endif /* INET6 */ 3027 } 3028 3029 return (m); 3030 } 3031 3032 static void 3033 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3034 uint8_t ttl, int rtableid) 3035 { 3036 struct mbuf *m; 3037 #ifdef INET 3038 struct ip *h = NULL; 3039 #endif /* INET */ 3040 #ifdef INET6 3041 struct ip6_hdr *h6 = NULL; 3042 #endif /* INET6 */ 3043 struct sctphdr *hdr; 3044 struct sctp_chunkhdr *chunk; 3045 struct pf_send_entry *pfse; 3046 int off = 0; 3047 3048 MPASS(af == pd->af); 3049 3050 m = m_gethdr(M_NOWAIT, MT_DATA); 3051 if (m == NULL) 3052 return; 3053 3054 m->m_data += max_linkhdr; 3055 m->m_flags |= M_SKIP_FIREWALL; 3056 /* The rest of the stack assumes a rcvif, so provide one. 3057 * This is a locally generated packet, so .. close enough. */ 3058 m->m_pkthdr.rcvif = V_loif; 3059 3060 /* IPv4|6 header */ 3061 switch (af) { 3062 #ifdef INET 3063 case AF_INET: 3064 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3065 3066 h = mtod(m, struct ip *); 3067 3068 /* IP header fields included in the TCP checksum */ 3069 3070 h->ip_p = IPPROTO_SCTP; 3071 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3072 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3073 h->ip_src = pd->dst->v4; 3074 h->ip_dst = pd->src->v4; 3075 3076 off += sizeof(struct ip); 3077 break; 3078 #endif /* INET */ 3079 #ifdef INET6 3080 case AF_INET6: 3081 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3082 3083 h6 = mtod(m, struct ip6_hdr *); 3084 3085 /* IP header fields included in the TCP checksum */ 3086 h6->ip6_vfc |= IPV6_VERSION; 3087 h6->ip6_nxt = IPPROTO_SCTP; 3088 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3089 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3090 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3091 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3092 3093 off += sizeof(struct ip6_hdr); 3094 break; 3095 #endif /* INET6 */ 3096 } 3097 3098 /* SCTP header */ 3099 hdr = mtodo(m, off); 3100 3101 hdr->src_port = pd->hdr.sctp.dest_port; 3102 hdr->dest_port = pd->hdr.sctp.src_port; 3103 hdr->v_tag = pd->sctp_initiate_tag; 3104 hdr->checksum = 0; 3105 3106 /* Abort chunk. */ 3107 off += sizeof(struct sctphdr); 3108 chunk = mtodo(m, off); 3109 3110 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3111 chunk->chunk_length = htons(sizeof(*chunk)); 3112 3113 /* SCTP checksum */ 3114 off += sizeof(*chunk); 3115 m->m_pkthdr.len = m->m_len = off; 3116 3117 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));; 3118 3119 if (rtableid >= 0) 3120 M_SETFIB(m, rtableid); 3121 3122 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3123 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3124 if (pfse == NULL) { 3125 m_freem(m); 3126 return; 3127 } 3128 3129 switch (af) { 3130 #ifdef INET 3131 case AF_INET: 3132 pfse->pfse_type = PFSE_IP; 3133 break; 3134 #endif /* INET */ 3135 #ifdef INET6 3136 case AF_INET6: 3137 pfse->pfse_type = PFSE_IP6; 3138 break; 3139 #endif /* INET6 */ 3140 } 3141 3142 pfse->pfse_m = m; 3143 pf_send(pfse); 3144 } 3145 3146 void 3147 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3148 const struct pf_addr *saddr, const struct pf_addr *daddr, 3149 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3150 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3151 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3152 { 3153 struct pf_send_entry *pfse; 3154 struct mbuf *m; 3155 3156 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3157 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3158 if (m == NULL) 3159 return; 3160 3161 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3162 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3163 if (pfse == NULL) { 3164 m_freem(m); 3165 return; 3166 } 3167 3168 switch (af) { 3169 #ifdef INET 3170 case AF_INET: 3171 pfse->pfse_type = PFSE_IP; 3172 break; 3173 #endif /* INET */ 3174 #ifdef INET6 3175 case AF_INET6: 3176 pfse->pfse_type = PFSE_IP6; 3177 break; 3178 #endif /* INET6 */ 3179 } 3180 3181 pfse->pfse_m = m; 3182 pf_send(pfse); 3183 } 3184 3185 static void 3186 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3187 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3188 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3189 u_short *reason, int rtableid) 3190 { 3191 struct pf_addr * const saddr = pd->src; 3192 struct pf_addr * const daddr = pd->dst; 3193 sa_family_t af = pd->af; 3194 3195 /* undo NAT changes, if they have taken place */ 3196 if (nr != NULL) { 3197 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3198 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3199 if (pd->sport) 3200 *pd->sport = sk->port[pd->sidx]; 3201 if (pd->dport) 3202 *pd->dport = sk->port[pd->didx]; 3203 if (pd->proto_sum) 3204 *pd->proto_sum = bproto_sum; 3205 if (pd->ip_sum) 3206 *pd->ip_sum = bip_sum; 3207 m_copyback(m, off, hdrlen, pd->hdr.any); 3208 } 3209 if (pd->proto == IPPROTO_TCP && 3210 ((r->rule_flag & PFRULE_RETURNRST) || 3211 (r->rule_flag & PFRULE_RETURN)) && 3212 !(th->th_flags & TH_RST)) { 3213 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3214 int len = 0; 3215 #ifdef INET 3216 struct ip *h4; 3217 #endif 3218 #ifdef INET6 3219 struct ip6_hdr *h6; 3220 #endif 3221 3222 switch (af) { 3223 #ifdef INET 3224 case AF_INET: 3225 h4 = mtod(m, struct ip *); 3226 len = ntohs(h4->ip_len) - off; 3227 break; 3228 #endif 3229 #ifdef INET6 3230 case AF_INET6: 3231 h6 = mtod(m, struct ip6_hdr *); 3232 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3233 break; 3234 #endif 3235 } 3236 3237 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3238 REASON_SET(reason, PFRES_PROTCKSUM); 3239 else { 3240 if (th->th_flags & TH_SYN) 3241 ack++; 3242 if (th->th_flags & TH_FIN) 3243 ack++; 3244 pf_send_tcp(r, af, pd->dst, 3245 pd->src, th->th_dport, th->th_sport, 3246 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3247 r->return_ttl, true, 0, 0, rtableid); 3248 } 3249 } else if (pd->proto == IPPROTO_SCTP && 3250 (r->rule_flag & PFRULE_RETURN)) { 3251 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3252 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3253 r->return_icmp) 3254 pf_send_icmp(m, r->return_icmp >> 8, 3255 r->return_icmp & 255, af, r, rtableid); 3256 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3257 r->return_icmp6) 3258 pf_send_icmp(m, r->return_icmp6 >> 8, 3259 r->return_icmp6 & 255, af, r, rtableid); 3260 } 3261 3262 static int 3263 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3264 { 3265 struct m_tag *mtag; 3266 u_int8_t mpcp; 3267 3268 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3269 if (mtag == NULL) 3270 return (0); 3271 3272 if (prio == PF_PRIO_ZERO) 3273 prio = 0; 3274 3275 mpcp = *(uint8_t *)(mtag + 1); 3276 3277 return (mpcp == prio); 3278 } 3279 3280 static int 3281 pf_icmp_to_bandlim(uint8_t type) 3282 { 3283 switch (type) { 3284 case ICMP_ECHO: 3285 case ICMP_ECHOREPLY: 3286 return (BANDLIM_ICMP_ECHO); 3287 case ICMP_TSTAMP: 3288 case ICMP_TSTAMPREPLY: 3289 return (BANDLIM_ICMP_TSTAMP); 3290 case ICMP_UNREACH: 3291 default: 3292 return (BANDLIM_ICMP_UNREACH); 3293 } 3294 } 3295 3296 static void 3297 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3298 struct pf_krule *r, int rtableid) 3299 { 3300 struct pf_send_entry *pfse; 3301 struct mbuf *m0; 3302 struct pf_mtag *pf_mtag; 3303 3304 /* ICMP packet rate limitation. */ 3305 #ifdef INET6 3306 if (af == AF_INET6) { 3307 if (icmp6_ratelimit(NULL, type, code)) 3308 return; 3309 } 3310 #endif 3311 #ifdef INET 3312 if (af == AF_INET) { 3313 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3314 return; 3315 } 3316 #endif 3317 3318 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3319 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3320 if (pfse == NULL) 3321 return; 3322 3323 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3324 free(pfse, M_PFTEMP); 3325 return; 3326 } 3327 3328 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3329 free(pfse, M_PFTEMP); 3330 return; 3331 } 3332 /* XXX: revisit */ 3333 m0->m_flags |= M_SKIP_FIREWALL; 3334 3335 if (rtableid >= 0) 3336 M_SETFIB(m0, rtableid); 3337 3338 #ifdef ALTQ 3339 if (r->qid) { 3340 pf_mtag->qid = r->qid; 3341 /* add hints for ecn */ 3342 pf_mtag->hdr = mtod(m0, struct ip *); 3343 } 3344 #endif /* ALTQ */ 3345 3346 switch (af) { 3347 #ifdef INET 3348 case AF_INET: 3349 pfse->pfse_type = PFSE_ICMP; 3350 break; 3351 #endif /* INET */ 3352 #ifdef INET6 3353 case AF_INET6: 3354 pfse->pfse_type = PFSE_ICMP6; 3355 break; 3356 #endif /* INET6 */ 3357 } 3358 pfse->pfse_m = m0; 3359 pfse->icmpopts.type = type; 3360 pfse->icmpopts.code = code; 3361 pf_send(pfse); 3362 } 3363 3364 /* 3365 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3366 * If n is 0, they match if they are equal. If n is != 0, they match if they 3367 * are different. 3368 */ 3369 int 3370 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3371 struct pf_addr *b, sa_family_t af) 3372 { 3373 int match = 0; 3374 3375 switch (af) { 3376 #ifdef INET 3377 case AF_INET: 3378 if ((a->addr32[0] & m->addr32[0]) == 3379 (b->addr32[0] & m->addr32[0])) 3380 match++; 3381 break; 3382 #endif /* INET */ 3383 #ifdef INET6 3384 case AF_INET6: 3385 if (((a->addr32[0] & m->addr32[0]) == 3386 (b->addr32[0] & m->addr32[0])) && 3387 ((a->addr32[1] & m->addr32[1]) == 3388 (b->addr32[1] & m->addr32[1])) && 3389 ((a->addr32[2] & m->addr32[2]) == 3390 (b->addr32[2] & m->addr32[2])) && 3391 ((a->addr32[3] & m->addr32[3]) == 3392 (b->addr32[3] & m->addr32[3]))) 3393 match++; 3394 break; 3395 #endif /* INET6 */ 3396 } 3397 if (match) { 3398 if (n) 3399 return (0); 3400 else 3401 return (1); 3402 } else { 3403 if (n) 3404 return (1); 3405 else 3406 return (0); 3407 } 3408 } 3409 3410 /* 3411 * Return 1 if b <= a <= e, otherwise return 0. 3412 */ 3413 int 3414 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3415 struct pf_addr *a, sa_family_t af) 3416 { 3417 switch (af) { 3418 #ifdef INET 3419 case AF_INET: 3420 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3421 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3422 return (0); 3423 break; 3424 #endif /* INET */ 3425 #ifdef INET6 3426 case AF_INET6: { 3427 int i; 3428 3429 /* check a >= b */ 3430 for (i = 0; i < 4; ++i) 3431 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3432 break; 3433 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3434 return (0); 3435 /* check a <= e */ 3436 for (i = 0; i < 4; ++i) 3437 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3438 break; 3439 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3440 return (0); 3441 break; 3442 } 3443 #endif /* INET6 */ 3444 } 3445 return (1); 3446 } 3447 3448 static int 3449 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3450 { 3451 switch (op) { 3452 case PF_OP_IRG: 3453 return ((p > a1) && (p < a2)); 3454 case PF_OP_XRG: 3455 return ((p < a1) || (p > a2)); 3456 case PF_OP_RRG: 3457 return ((p >= a1) && (p <= a2)); 3458 case PF_OP_EQ: 3459 return (p == a1); 3460 case PF_OP_NE: 3461 return (p != a1); 3462 case PF_OP_LT: 3463 return (p < a1); 3464 case PF_OP_LE: 3465 return (p <= a1); 3466 case PF_OP_GT: 3467 return (p > a1); 3468 case PF_OP_GE: 3469 return (p >= a1); 3470 } 3471 return (0); /* never reached */ 3472 } 3473 3474 int 3475 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3476 { 3477 NTOHS(a1); 3478 NTOHS(a2); 3479 NTOHS(p); 3480 return (pf_match(op, a1, a2, p)); 3481 } 3482 3483 static int 3484 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3485 { 3486 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3487 return (0); 3488 return (pf_match(op, a1, a2, u)); 3489 } 3490 3491 static int 3492 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3493 { 3494 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3495 return (0); 3496 return (pf_match(op, a1, a2, g)); 3497 } 3498 3499 int 3500 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3501 { 3502 if (*tag == -1) 3503 *tag = mtag; 3504 3505 return ((!r->match_tag_not && r->match_tag == *tag) || 3506 (r->match_tag_not && r->match_tag != *tag)); 3507 } 3508 3509 int 3510 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3511 { 3512 3513 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3514 3515 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3516 return (ENOMEM); 3517 3518 pd->pf_mtag->tag = tag; 3519 3520 return (0); 3521 } 3522 3523 #define PF_ANCHOR_STACKSIZE 32 3524 struct pf_kanchor_stackframe { 3525 struct pf_kruleset *rs; 3526 struct pf_krule *r; /* XXX: + match bit */ 3527 struct pf_kanchor *child; 3528 }; 3529 3530 /* 3531 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3532 */ 3533 #define PF_ANCHORSTACK_MATCH 0x00000001 3534 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3535 3536 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3537 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3538 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3539 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3540 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3541 } while (0) 3542 3543 void 3544 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3545 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3546 int *match) 3547 { 3548 struct pf_kanchor_stackframe *f; 3549 3550 PF_RULES_RASSERT(); 3551 3552 if (match) 3553 *match = 0; 3554 if (*depth >= PF_ANCHOR_STACKSIZE) { 3555 printf("%s: anchor stack overflow on %s\n", 3556 __func__, (*r)->anchor->name); 3557 *r = TAILQ_NEXT(*r, entries); 3558 return; 3559 } else if (*depth == 0 && a != NULL) 3560 *a = *r; 3561 f = stack + (*depth)++; 3562 f->rs = *rs; 3563 f->r = *r; 3564 if ((*r)->anchor_wildcard) { 3565 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3566 3567 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3568 *r = NULL; 3569 return; 3570 } 3571 *rs = &f->child->ruleset; 3572 } else { 3573 f->child = NULL; 3574 *rs = &(*r)->anchor->ruleset; 3575 } 3576 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3577 } 3578 3579 int 3580 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3581 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3582 int *match) 3583 { 3584 struct pf_kanchor_stackframe *f; 3585 struct pf_krule *fr; 3586 int quick = 0; 3587 3588 PF_RULES_RASSERT(); 3589 3590 do { 3591 if (*depth <= 0) 3592 break; 3593 f = stack + *depth - 1; 3594 fr = PF_ANCHOR_RULE(f); 3595 if (f->child != NULL) { 3596 /* 3597 * This block traverses through 3598 * a wildcard anchor. 3599 */ 3600 if (match != NULL && *match) { 3601 /* 3602 * If any of "*" matched, then 3603 * "foo/ *" matched, mark frame 3604 * appropriately. 3605 */ 3606 PF_ANCHOR_SET_MATCH(f); 3607 *match = 0; 3608 } 3609 f->child = RB_NEXT(pf_kanchor_node, 3610 &fr->anchor->children, f->child); 3611 if (f->child != NULL) { 3612 *rs = &f->child->ruleset; 3613 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3614 if (*r == NULL) 3615 continue; 3616 else 3617 break; 3618 } 3619 } 3620 (*depth)--; 3621 if (*depth == 0 && a != NULL) 3622 *a = NULL; 3623 *rs = f->rs; 3624 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3625 quick = fr->quick; 3626 *r = TAILQ_NEXT(fr, entries); 3627 } while (*r == NULL); 3628 3629 return (quick); 3630 } 3631 3632 struct pf_keth_anchor_stackframe { 3633 struct pf_keth_ruleset *rs; 3634 struct pf_keth_rule *r; /* XXX: + match bit */ 3635 struct pf_keth_anchor *child; 3636 }; 3637 3638 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3639 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3640 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3641 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3642 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3643 } while (0) 3644 3645 void 3646 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3647 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3648 struct pf_keth_rule **a, int *match) 3649 { 3650 struct pf_keth_anchor_stackframe *f; 3651 3652 NET_EPOCH_ASSERT(); 3653 3654 if (match) 3655 *match = 0; 3656 if (*depth >= PF_ANCHOR_STACKSIZE) { 3657 printf("%s: anchor stack overflow on %s\n", 3658 __func__, (*r)->anchor->name); 3659 *r = TAILQ_NEXT(*r, entries); 3660 return; 3661 } else if (*depth == 0 && a != NULL) 3662 *a = *r; 3663 f = stack + (*depth)++; 3664 f->rs = *rs; 3665 f->r = *r; 3666 if ((*r)->anchor_wildcard) { 3667 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3668 3669 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3670 *r = NULL; 3671 return; 3672 } 3673 *rs = &f->child->ruleset; 3674 } else { 3675 f->child = NULL; 3676 *rs = &(*r)->anchor->ruleset; 3677 } 3678 *r = TAILQ_FIRST((*rs)->active.rules); 3679 } 3680 3681 int 3682 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3683 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3684 struct pf_keth_rule **a, int *match) 3685 { 3686 struct pf_keth_anchor_stackframe *f; 3687 struct pf_keth_rule *fr; 3688 int quick = 0; 3689 3690 NET_EPOCH_ASSERT(); 3691 3692 do { 3693 if (*depth <= 0) 3694 break; 3695 f = stack + *depth - 1; 3696 fr = PF_ETH_ANCHOR_RULE(f); 3697 if (f->child != NULL) { 3698 /* 3699 * This block traverses through 3700 * a wildcard anchor. 3701 */ 3702 if (match != NULL && *match) { 3703 /* 3704 * If any of "*" matched, then 3705 * "foo/ *" matched, mark frame 3706 * appropriately. 3707 */ 3708 PF_ETH_ANCHOR_SET_MATCH(f); 3709 *match = 0; 3710 } 3711 f->child = RB_NEXT(pf_keth_anchor_node, 3712 &fr->anchor->children, f->child); 3713 if (f->child != NULL) { 3714 *rs = &f->child->ruleset; 3715 *r = TAILQ_FIRST((*rs)->active.rules); 3716 if (*r == NULL) 3717 continue; 3718 else 3719 break; 3720 } 3721 } 3722 (*depth)--; 3723 if (*depth == 0 && a != NULL) 3724 *a = NULL; 3725 *rs = f->rs; 3726 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3727 quick = fr->quick; 3728 *r = TAILQ_NEXT(fr, entries); 3729 } while (*r == NULL); 3730 3731 return (quick); 3732 } 3733 3734 #ifdef INET6 3735 void 3736 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3737 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3738 { 3739 switch (af) { 3740 #ifdef INET 3741 case AF_INET: 3742 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3743 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3744 break; 3745 #endif /* INET */ 3746 case AF_INET6: 3747 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3748 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3749 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3750 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3751 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3752 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3753 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3754 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3755 break; 3756 } 3757 } 3758 3759 void 3760 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3761 { 3762 switch (af) { 3763 #ifdef INET 3764 case AF_INET: 3765 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3766 break; 3767 #endif /* INET */ 3768 case AF_INET6: 3769 if (addr->addr32[3] == 0xffffffff) { 3770 addr->addr32[3] = 0; 3771 if (addr->addr32[2] == 0xffffffff) { 3772 addr->addr32[2] = 0; 3773 if (addr->addr32[1] == 0xffffffff) { 3774 addr->addr32[1] = 0; 3775 addr->addr32[0] = 3776 htonl(ntohl(addr->addr32[0]) + 1); 3777 } else 3778 addr->addr32[1] = 3779 htonl(ntohl(addr->addr32[1]) + 1); 3780 } else 3781 addr->addr32[2] = 3782 htonl(ntohl(addr->addr32[2]) + 1); 3783 } else 3784 addr->addr32[3] = 3785 htonl(ntohl(addr->addr32[3]) + 1); 3786 break; 3787 } 3788 } 3789 #endif /* INET6 */ 3790 3791 void 3792 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3793 { 3794 /* 3795 * Modern rules use the same flags in rules as they do in states. 3796 */ 3797 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3798 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3799 3800 /* 3801 * Old-style scrub rules have different flags which need to be translated. 3802 */ 3803 if (r->rule_flag & PFRULE_RANDOMID) 3804 a->flags |= PFSTATE_RANDOMID; 3805 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 3806 a->flags |= PFSTATE_SETTOS; 3807 a->set_tos = r->set_tos; 3808 } 3809 3810 if (r->qid) 3811 a->qid = r->qid; 3812 if (r->pqid) 3813 a->pqid = r->pqid; 3814 if (r->rtableid >= 0) 3815 a->rtableid = r->rtableid; 3816 a->log |= r->log; 3817 if (r->min_ttl) 3818 a->min_ttl = r->min_ttl; 3819 if (r->max_mss) 3820 a->max_mss = r->max_mss; 3821 if (r->dnpipe) 3822 a->dnpipe = r->dnpipe; 3823 if (r->dnrpipe) 3824 a->dnrpipe = r->dnrpipe; 3825 if (r->dnpipe || r->dnrpipe) { 3826 if (r->free_flags & PFRULE_DN_IS_PIPE) 3827 a->flags |= PFSTATE_DN_IS_PIPE; 3828 else 3829 a->flags &= ~PFSTATE_DN_IS_PIPE; 3830 } 3831 if (r->scrub_flags & PFSTATE_SETPRIO) { 3832 a->set_prio[0] = r->set_prio[0]; 3833 a->set_prio[1] = r->set_prio[1]; 3834 } 3835 } 3836 3837 int 3838 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 3839 { 3840 struct pf_addr *saddr, *daddr; 3841 u_int16_t sport, dport; 3842 struct inpcbinfo *pi; 3843 struct inpcb *inp; 3844 3845 pd->lookup.uid = UID_MAX; 3846 pd->lookup.gid = GID_MAX; 3847 3848 switch (pd->proto) { 3849 case IPPROTO_TCP: 3850 sport = pd->hdr.tcp.th_sport; 3851 dport = pd->hdr.tcp.th_dport; 3852 pi = &V_tcbinfo; 3853 break; 3854 case IPPROTO_UDP: 3855 sport = pd->hdr.udp.uh_sport; 3856 dport = pd->hdr.udp.uh_dport; 3857 pi = &V_udbinfo; 3858 break; 3859 default: 3860 return (-1); 3861 } 3862 if (pd->dir == PF_IN) { 3863 saddr = pd->src; 3864 daddr = pd->dst; 3865 } else { 3866 u_int16_t p; 3867 3868 p = sport; 3869 sport = dport; 3870 dport = p; 3871 saddr = pd->dst; 3872 daddr = pd->src; 3873 } 3874 switch (pd->af) { 3875 #ifdef INET 3876 case AF_INET: 3877 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3878 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3879 if (inp == NULL) { 3880 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3881 daddr->v4, dport, INPLOOKUP_WILDCARD | 3882 INPLOOKUP_RLOCKPCB, NULL, m); 3883 if (inp == NULL) 3884 return (-1); 3885 } 3886 break; 3887 #endif /* INET */ 3888 #ifdef INET6 3889 case AF_INET6: 3890 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3891 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3892 if (inp == NULL) { 3893 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3894 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3895 INPLOOKUP_RLOCKPCB, NULL, m); 3896 if (inp == NULL) 3897 return (-1); 3898 } 3899 break; 3900 #endif /* INET6 */ 3901 3902 default: 3903 return (-1); 3904 } 3905 INP_RLOCK_ASSERT(inp); 3906 pd->lookup.uid = inp->inp_cred->cr_uid; 3907 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3908 INP_RUNLOCK(inp); 3909 3910 return (1); 3911 } 3912 3913 u_int8_t 3914 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3915 { 3916 int hlen; 3917 u_int8_t hdr[60]; 3918 u_int8_t *opt, optlen; 3919 u_int8_t wscale = 0; 3920 3921 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3922 if (hlen <= sizeof(struct tcphdr)) 3923 return (0); 3924 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3925 return (0); 3926 opt = hdr + sizeof(struct tcphdr); 3927 hlen -= sizeof(struct tcphdr); 3928 while (hlen >= 3) { 3929 switch (*opt) { 3930 case TCPOPT_EOL: 3931 case TCPOPT_NOP: 3932 ++opt; 3933 --hlen; 3934 break; 3935 case TCPOPT_WINDOW: 3936 wscale = opt[2]; 3937 if (wscale > TCP_MAX_WINSHIFT) 3938 wscale = TCP_MAX_WINSHIFT; 3939 wscale |= PF_WSCALE_FLAG; 3940 /* FALLTHROUGH */ 3941 default: 3942 optlen = opt[1]; 3943 if (optlen < 2) 3944 optlen = 2; 3945 hlen -= optlen; 3946 opt += optlen; 3947 break; 3948 } 3949 } 3950 return (wscale); 3951 } 3952 3953 u_int16_t 3954 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3955 { 3956 int hlen; 3957 u_int8_t hdr[60]; 3958 u_int8_t *opt, optlen; 3959 u_int16_t mss = V_tcp_mssdflt; 3960 3961 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3962 if (hlen <= sizeof(struct tcphdr)) 3963 return (0); 3964 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3965 return (0); 3966 opt = hdr + sizeof(struct tcphdr); 3967 hlen -= sizeof(struct tcphdr); 3968 while (hlen >= TCPOLEN_MAXSEG) { 3969 switch (*opt) { 3970 case TCPOPT_EOL: 3971 case TCPOPT_NOP: 3972 ++opt; 3973 --hlen; 3974 break; 3975 case TCPOPT_MAXSEG: 3976 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3977 NTOHS(mss); 3978 /* FALLTHROUGH */ 3979 default: 3980 optlen = opt[1]; 3981 if (optlen < 2) 3982 optlen = 2; 3983 hlen -= optlen; 3984 opt += optlen; 3985 break; 3986 } 3987 } 3988 return (mss); 3989 } 3990 3991 static u_int16_t 3992 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3993 { 3994 struct nhop_object *nh; 3995 #ifdef INET6 3996 struct in6_addr dst6; 3997 uint32_t scopeid; 3998 #endif /* INET6 */ 3999 int hlen = 0; 4000 uint16_t mss = 0; 4001 4002 NET_EPOCH_ASSERT(); 4003 4004 switch (af) { 4005 #ifdef INET 4006 case AF_INET: 4007 hlen = sizeof(struct ip); 4008 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4009 if (nh != NULL) 4010 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4011 break; 4012 #endif /* INET */ 4013 #ifdef INET6 4014 case AF_INET6: 4015 hlen = sizeof(struct ip6_hdr); 4016 in6_splitscope(&addr->v6, &dst6, &scopeid); 4017 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4018 if (nh != NULL) 4019 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4020 break; 4021 #endif /* INET6 */ 4022 } 4023 4024 mss = max(V_tcp_mssdflt, mss); 4025 mss = min(mss, offer); 4026 mss = max(mss, 64); /* sanity - at least max opt space */ 4027 return (mss); 4028 } 4029 4030 static u_int32_t 4031 pf_tcp_iss(struct pf_pdesc *pd) 4032 { 4033 MD5_CTX ctx; 4034 u_int32_t digest[4]; 4035 4036 if (V_pf_tcp_secret_init == 0) { 4037 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4038 MD5Init(&V_pf_tcp_secret_ctx); 4039 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4040 sizeof(V_pf_tcp_secret)); 4041 V_pf_tcp_secret_init = 1; 4042 } 4043 4044 ctx = V_pf_tcp_secret_ctx; 4045 4046 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4047 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4048 if (pd->af == AF_INET6) { 4049 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4050 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4051 } else { 4052 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4053 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4054 } 4055 MD5Final((u_char *)digest, &ctx); 4056 V_pf_tcp_iss_off += 4096; 4057 #define ISN_RANDOM_INCREMENT (4096 - 1) 4058 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4059 V_pf_tcp_iss_off); 4060 #undef ISN_RANDOM_INCREMENT 4061 } 4062 4063 static bool 4064 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4065 { 4066 bool match = true; 4067 4068 /* Always matches if not set */ 4069 if (! r->isset) 4070 return (!r->neg); 4071 4072 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4073 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4074 match = false; 4075 break; 4076 } 4077 } 4078 4079 return (match ^ r->neg); 4080 } 4081 4082 static int 4083 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4084 { 4085 if (*tag == -1) 4086 *tag = mtag; 4087 4088 return ((!r->match_tag_not && r->match_tag == *tag) || 4089 (r->match_tag_not && r->match_tag != *tag)); 4090 } 4091 4092 static void 4093 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4094 { 4095 /* If we don't have the interface drop the packet. */ 4096 if (ifp == NULL) { 4097 m_freem(m); 4098 return; 4099 } 4100 4101 switch (ifp->if_type) { 4102 case IFT_ETHER: 4103 case IFT_XETHER: 4104 case IFT_L2VLAN: 4105 case IFT_BRIDGE: 4106 case IFT_IEEE8023ADLAG: 4107 break; 4108 default: 4109 m_freem(m); 4110 return; 4111 } 4112 4113 ifp->if_transmit(ifp, m); 4114 } 4115 4116 static int 4117 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4118 { 4119 #ifdef INET 4120 struct ip ip; 4121 #endif 4122 #ifdef INET6 4123 struct ip6_hdr ip6; 4124 #endif 4125 struct mbuf *m = *m0; 4126 struct ether_header *e; 4127 struct pf_keth_rule *r, *rm, *a = NULL; 4128 struct pf_keth_ruleset *ruleset = NULL; 4129 struct pf_mtag *mtag; 4130 struct pf_keth_ruleq *rules; 4131 struct pf_addr *src = NULL, *dst = NULL; 4132 struct pfi_kkif *bridge_to; 4133 sa_family_t af = 0; 4134 uint16_t proto; 4135 int asd = 0, match = 0; 4136 int tag = -1; 4137 uint8_t action; 4138 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4139 4140 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4141 NET_EPOCH_ASSERT(); 4142 4143 PF_RULES_RLOCK_TRACKER; 4144 4145 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4146 4147 mtag = pf_find_mtag(m); 4148 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4149 /* Dummynet re-injects packets after they've 4150 * completed their delay. We've already 4151 * processed them, so pass unconditionally. */ 4152 4153 /* But only once. We may see the packet multiple times (e.g. 4154 * PFIL_IN/PFIL_OUT). */ 4155 pf_dummynet_flag_remove(m, mtag); 4156 4157 return (PF_PASS); 4158 } 4159 4160 ruleset = V_pf_keth; 4161 rules = ck_pr_load_ptr(&ruleset->active.rules); 4162 r = TAILQ_FIRST(rules); 4163 rm = NULL; 4164 4165 e = mtod(m, struct ether_header *); 4166 proto = ntohs(e->ether_type); 4167 4168 switch (proto) { 4169 #ifdef INET 4170 case ETHERTYPE_IP: { 4171 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4172 sizeof(ip))) 4173 return (PF_DROP); 4174 4175 af = AF_INET; 4176 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4177 (caddr_t)&ip); 4178 src = (struct pf_addr *)&ip.ip_src; 4179 dst = (struct pf_addr *)&ip.ip_dst; 4180 break; 4181 } 4182 #endif /* INET */ 4183 #ifdef INET6 4184 case ETHERTYPE_IPV6: { 4185 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4186 sizeof(ip6))) 4187 return (PF_DROP); 4188 4189 af = AF_INET6; 4190 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4191 (caddr_t)&ip6); 4192 src = (struct pf_addr *)&ip6.ip6_src; 4193 dst = (struct pf_addr *)&ip6.ip6_dst; 4194 break; 4195 } 4196 #endif /* INET6 */ 4197 } 4198 4199 PF_RULES_RLOCK(); 4200 4201 while (r != NULL) { 4202 counter_u64_add(r->evaluations, 1); 4203 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4204 4205 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4206 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4207 "kif"); 4208 r = r->skip[PFE_SKIP_IFP].ptr; 4209 } 4210 else if (r->direction && r->direction != dir) { 4211 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4212 "dir"); 4213 r = r->skip[PFE_SKIP_DIR].ptr; 4214 } 4215 else if (r->proto && r->proto != proto) { 4216 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4217 "proto"); 4218 r = r->skip[PFE_SKIP_PROTO].ptr; 4219 } 4220 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4221 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4222 "src"); 4223 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4224 } 4225 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4226 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4227 "dst"); 4228 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4229 } 4230 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4231 r->ipsrc.neg, kif, M_GETFIB(m))) { 4232 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4233 "ip_src"); 4234 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4235 } 4236 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4237 r->ipdst.neg, kif, M_GETFIB(m))) { 4238 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4239 "ip_dst"); 4240 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4241 } 4242 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4243 mtag ? mtag->tag : 0)) { 4244 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4245 "match_tag"); 4246 r = TAILQ_NEXT(r, entries); 4247 } 4248 else { 4249 if (r->tag) 4250 tag = r->tag; 4251 if (r->anchor == NULL) { 4252 /* Rule matches */ 4253 rm = r; 4254 4255 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4256 4257 if (r->quick) 4258 break; 4259 4260 r = TAILQ_NEXT(r, entries); 4261 } else { 4262 pf_step_into_keth_anchor(anchor_stack, &asd, 4263 &ruleset, &r, &a, &match); 4264 } 4265 } 4266 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4267 &ruleset, &r, &a, &match)) 4268 break; 4269 } 4270 4271 r = rm; 4272 4273 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4274 4275 /* Default to pass. */ 4276 if (r == NULL) { 4277 PF_RULES_RUNLOCK(); 4278 return (PF_PASS); 4279 } 4280 4281 /* Execute action. */ 4282 counter_u64_add(r->packets[dir == PF_OUT], 1); 4283 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4284 pf_update_timestamp(r); 4285 4286 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4287 if (r->action == PF_DROP) { 4288 PF_RULES_RUNLOCK(); 4289 return (PF_DROP); 4290 } 4291 4292 if (tag > 0) { 4293 if (mtag == NULL) 4294 mtag = pf_get_mtag(m); 4295 if (mtag == NULL) { 4296 PF_RULES_RUNLOCK(); 4297 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4298 return (PF_DROP); 4299 } 4300 mtag->tag = tag; 4301 } 4302 4303 if (r->qid != 0) { 4304 if (mtag == NULL) 4305 mtag = pf_get_mtag(m); 4306 if (mtag == NULL) { 4307 PF_RULES_RUNLOCK(); 4308 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4309 return (PF_DROP); 4310 } 4311 mtag->qid = r->qid; 4312 } 4313 4314 action = r->action; 4315 bridge_to = r->bridge_to; 4316 4317 /* Dummynet */ 4318 if (r->dnpipe) { 4319 struct ip_fw_args dnflow; 4320 4321 /* Drop packet if dummynet is not loaded. */ 4322 if (ip_dn_io_ptr == NULL) { 4323 PF_RULES_RUNLOCK(); 4324 m_freem(m); 4325 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4326 return (PF_DROP); 4327 } 4328 if (mtag == NULL) 4329 mtag = pf_get_mtag(m); 4330 if (mtag == NULL) { 4331 PF_RULES_RUNLOCK(); 4332 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4333 return (PF_DROP); 4334 } 4335 4336 bzero(&dnflow, sizeof(dnflow)); 4337 4338 /* We don't have port numbers here, so we set 0. That means 4339 * that we'll be somewhat limited in distinguishing flows (i.e. 4340 * only based on IP addresses, not based on port numbers), but 4341 * it's better than nothing. */ 4342 dnflow.f_id.dst_port = 0; 4343 dnflow.f_id.src_port = 0; 4344 dnflow.f_id.proto = 0; 4345 4346 dnflow.rule.info = r->dnpipe; 4347 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4348 if (r->dnflags & PFRULE_DN_IS_PIPE) 4349 dnflow.rule.info |= IPFW_IS_PIPE; 4350 4351 dnflow.f_id.extra = dnflow.rule.info; 4352 4353 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4354 dnflow.flags |= IPFW_ARGS_ETHER; 4355 dnflow.ifp = kif->pfik_ifp; 4356 4357 switch (af) { 4358 case AF_INET: 4359 dnflow.f_id.addr_type = 4; 4360 dnflow.f_id.src_ip = src->v4.s_addr; 4361 dnflow.f_id.dst_ip = dst->v4.s_addr; 4362 break; 4363 case AF_INET6: 4364 dnflow.flags |= IPFW_ARGS_IP6; 4365 dnflow.f_id.addr_type = 6; 4366 dnflow.f_id.src_ip6 = src->v6; 4367 dnflow.f_id.dst_ip6 = dst->v6; 4368 break; 4369 } 4370 4371 PF_RULES_RUNLOCK(); 4372 4373 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4374 ip_dn_io_ptr(m0, &dnflow); 4375 if (*m0 != NULL) 4376 pf_dummynet_flag_remove(m, mtag); 4377 } else { 4378 PF_RULES_RUNLOCK(); 4379 } 4380 4381 if (action == PF_PASS && bridge_to) { 4382 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4383 *m0 = NULL; /* We've eaten the packet. */ 4384 } 4385 4386 return (action); 4387 } 4388 4389 static int 4390 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4391 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4392 struct pf_kruleset **rsm, struct inpcb *inp) 4393 { 4394 struct pf_krule *nr = NULL; 4395 struct pf_addr * const saddr = pd->src; 4396 struct pf_addr * const daddr = pd->dst; 4397 sa_family_t af = pd->af; 4398 struct pf_krule *r, *a = NULL; 4399 struct pf_kruleset *ruleset = NULL; 4400 struct pf_krule_slist match_rules; 4401 struct pf_krule_item *ri; 4402 struct pf_ksrc_node *nsn = NULL; 4403 struct tcphdr *th = &pd->hdr.tcp; 4404 struct pf_state_key *sk = NULL, *nk = NULL; 4405 u_short reason; 4406 int rewrite = 0, hdrlen = 0; 4407 int tag = -1; 4408 int asd = 0; 4409 int match = 0; 4410 int state_icmp = 0; 4411 u_int16_t sport = 0, dport = 0; 4412 u_int16_t bproto_sum = 0, bip_sum = 0; 4413 u_int8_t icmptype = 0, icmpcode = 0; 4414 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4415 4416 PF_RULES_RASSERT(); 4417 4418 if (inp != NULL) { 4419 INP_LOCK_ASSERT(inp); 4420 pd->lookup.uid = inp->inp_cred->cr_uid; 4421 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4422 pd->lookup.done = 1; 4423 } 4424 4425 switch (pd->proto) { 4426 case IPPROTO_TCP: 4427 sport = th->th_sport; 4428 dport = th->th_dport; 4429 hdrlen = sizeof(*th); 4430 break; 4431 case IPPROTO_UDP: 4432 sport = pd->hdr.udp.uh_sport; 4433 dport = pd->hdr.udp.uh_dport; 4434 hdrlen = sizeof(pd->hdr.udp); 4435 break; 4436 case IPPROTO_SCTP: 4437 sport = pd->hdr.sctp.src_port; 4438 dport = pd->hdr.sctp.dest_port; 4439 hdrlen = sizeof(pd->hdr.sctp); 4440 break; 4441 #ifdef INET 4442 case IPPROTO_ICMP: 4443 if (pd->af != AF_INET) 4444 break; 4445 sport = dport = pd->hdr.icmp.icmp_id; 4446 hdrlen = sizeof(pd->hdr.icmp); 4447 icmptype = pd->hdr.icmp.icmp_type; 4448 icmpcode = pd->hdr.icmp.icmp_code; 4449 4450 if (icmptype == ICMP_UNREACH || 4451 icmptype == ICMP_SOURCEQUENCH || 4452 icmptype == ICMP_REDIRECT || 4453 icmptype == ICMP_TIMXCEED || 4454 icmptype == ICMP_PARAMPROB) 4455 state_icmp++; 4456 break; 4457 #endif /* INET */ 4458 #ifdef INET6 4459 case IPPROTO_ICMPV6: 4460 if (af != AF_INET6) 4461 break; 4462 sport = dport = pd->hdr.icmp6.icmp6_id; 4463 hdrlen = sizeof(pd->hdr.icmp6); 4464 icmptype = pd->hdr.icmp6.icmp6_type; 4465 icmpcode = pd->hdr.icmp6.icmp6_code; 4466 4467 if (icmptype == ICMP6_DST_UNREACH || 4468 icmptype == ICMP6_PACKET_TOO_BIG || 4469 icmptype == ICMP6_TIME_EXCEEDED || 4470 icmptype == ICMP6_PARAM_PROB) 4471 state_icmp++; 4472 break; 4473 #endif /* INET6 */ 4474 default: 4475 sport = dport = hdrlen = 0; 4476 break; 4477 } 4478 4479 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4480 4481 /* check packet for BINAT/NAT/RDR */ 4482 if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4483 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4484 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4485 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4486 4487 if (nr->log) { 4488 PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a, 4489 ruleset, pd, 1); 4490 } 4491 4492 if (pd->ip_sum) 4493 bip_sum = *pd->ip_sum; 4494 4495 switch (pd->proto) { 4496 case IPPROTO_TCP: 4497 bproto_sum = th->th_sum; 4498 pd->proto_sum = &th->th_sum; 4499 4500 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4501 nk->port[pd->sidx] != sport) { 4502 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4503 &th->th_sum, &nk->addr[pd->sidx], 4504 nk->port[pd->sidx], 0, af); 4505 pd->sport = &th->th_sport; 4506 sport = th->th_sport; 4507 } 4508 4509 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4510 nk->port[pd->didx] != dport) { 4511 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4512 &th->th_sum, &nk->addr[pd->didx], 4513 nk->port[pd->didx], 0, af); 4514 dport = th->th_dport; 4515 pd->dport = &th->th_dport; 4516 } 4517 rewrite++; 4518 break; 4519 case IPPROTO_UDP: 4520 bproto_sum = pd->hdr.udp.uh_sum; 4521 pd->proto_sum = &pd->hdr.udp.uh_sum; 4522 4523 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4524 nk->port[pd->sidx] != sport) { 4525 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4526 pd->ip_sum, &pd->hdr.udp.uh_sum, 4527 &nk->addr[pd->sidx], 4528 nk->port[pd->sidx], 1, af); 4529 sport = pd->hdr.udp.uh_sport; 4530 pd->sport = &pd->hdr.udp.uh_sport; 4531 } 4532 4533 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4534 nk->port[pd->didx] != dport) { 4535 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4536 pd->ip_sum, &pd->hdr.udp.uh_sum, 4537 &nk->addr[pd->didx], 4538 nk->port[pd->didx], 1, af); 4539 dport = pd->hdr.udp.uh_dport; 4540 pd->dport = &pd->hdr.udp.uh_dport; 4541 } 4542 rewrite++; 4543 break; 4544 case IPPROTO_SCTP: { 4545 uint16_t checksum = 0; 4546 4547 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4548 nk->port[pd->sidx] != sport) { 4549 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 4550 pd->ip_sum, &checksum, 4551 &nk->addr[pd->sidx], 4552 nk->port[pd->sidx], 1, af); 4553 } 4554 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4555 nk->port[pd->didx] != dport) { 4556 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 4557 pd->ip_sum, &checksum, 4558 &nk->addr[pd->didx], 4559 nk->port[pd->didx], 1, af); 4560 } 4561 break; 4562 } 4563 #ifdef INET 4564 case IPPROTO_ICMP: 4565 nk->port[0] = nk->port[1]; 4566 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4567 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4568 nk->addr[pd->sidx].v4.s_addr, 0); 4569 4570 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4571 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4572 nk->addr[pd->didx].v4.s_addr, 0); 4573 4574 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4575 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4576 pd->hdr.icmp.icmp_cksum, sport, 4577 nk->port[1], 0); 4578 pd->hdr.icmp.icmp_id = nk->port[1]; 4579 pd->sport = &pd->hdr.icmp.icmp_id; 4580 } 4581 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4582 break; 4583 #endif /* INET */ 4584 #ifdef INET6 4585 case IPPROTO_ICMPV6: 4586 nk->port[0] = nk->port[1]; 4587 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4588 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4589 &nk->addr[pd->sidx], 0); 4590 4591 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4592 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4593 &nk->addr[pd->didx], 0); 4594 rewrite++; 4595 break; 4596 #endif /* INET */ 4597 default: 4598 switch (af) { 4599 #ifdef INET 4600 case AF_INET: 4601 if (PF_ANEQ(saddr, 4602 &nk->addr[pd->sidx], AF_INET)) 4603 pf_change_a(&saddr->v4.s_addr, 4604 pd->ip_sum, 4605 nk->addr[pd->sidx].v4.s_addr, 0); 4606 4607 if (PF_ANEQ(daddr, 4608 &nk->addr[pd->didx], AF_INET)) 4609 pf_change_a(&daddr->v4.s_addr, 4610 pd->ip_sum, 4611 nk->addr[pd->didx].v4.s_addr, 0); 4612 break; 4613 #endif /* INET */ 4614 #ifdef INET6 4615 case AF_INET6: 4616 if (PF_ANEQ(saddr, 4617 &nk->addr[pd->sidx], AF_INET6)) 4618 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4619 4620 if (PF_ANEQ(daddr, 4621 &nk->addr[pd->didx], AF_INET6)) 4622 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4623 break; 4624 #endif /* INET */ 4625 } 4626 break; 4627 } 4628 if (nr->natpass) 4629 r = NULL; 4630 pd->nat_rule = nr; 4631 } 4632 4633 SLIST_INIT(&match_rules); 4634 while (r != NULL) { 4635 pf_counter_u64_add(&r->evaluations, 1); 4636 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4637 r = r->skip[PF_SKIP_IFP].ptr; 4638 else if (r->direction && r->direction != pd->dir) 4639 r = r->skip[PF_SKIP_DIR].ptr; 4640 else if (r->af && r->af != af) 4641 r = r->skip[PF_SKIP_AF].ptr; 4642 else if (r->proto && r->proto != pd->proto) 4643 r = r->skip[PF_SKIP_PROTO].ptr; 4644 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4645 r->src.neg, kif, M_GETFIB(m))) 4646 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4647 /* tcp/udp only. port_op always 0 in other cases */ 4648 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4649 r->src.port[0], r->src.port[1], sport)) 4650 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4651 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4652 r->dst.neg, NULL, M_GETFIB(m))) 4653 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4654 /* tcp/udp only. port_op always 0 in other cases */ 4655 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4656 r->dst.port[0], r->dst.port[1], dport)) 4657 r = r->skip[PF_SKIP_DST_PORT].ptr; 4658 /* icmp only. type always 0 in other cases */ 4659 else if (r->type && r->type != icmptype + 1) 4660 r = TAILQ_NEXT(r, entries); 4661 /* icmp only. type always 0 in other cases */ 4662 else if (r->code && r->code != icmpcode + 1) 4663 r = TAILQ_NEXT(r, entries); 4664 else if (r->tos && !(r->tos == pd->tos)) 4665 r = TAILQ_NEXT(r, entries); 4666 else if (r->rule_flag & PFRULE_FRAGMENT) 4667 r = TAILQ_NEXT(r, entries); 4668 else if (pd->proto == IPPROTO_TCP && 4669 (r->flagset & th->th_flags) != r->flags) 4670 r = TAILQ_NEXT(r, entries); 4671 /* tcp/udp only. uid.op always 0 in other cases */ 4672 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4673 pf_socket_lookup(pd, m), 1)) && 4674 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4675 pd->lookup.uid)) 4676 r = TAILQ_NEXT(r, entries); 4677 /* tcp/udp only. gid.op always 0 in other cases */ 4678 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4679 pf_socket_lookup(pd, m), 1)) && 4680 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4681 pd->lookup.gid)) 4682 r = TAILQ_NEXT(r, entries); 4683 else if (r->prio && 4684 !pf_match_ieee8021q_pcp(r->prio, m)) 4685 r = TAILQ_NEXT(r, entries); 4686 else if (r->prob && 4687 r->prob <= arc4random()) 4688 r = TAILQ_NEXT(r, entries); 4689 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4690 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4691 r = TAILQ_NEXT(r, entries); 4692 else if (r->os_fingerprint != PF_OSFP_ANY && 4693 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4694 pf_osfp_fingerprint(pd, m, off, th), 4695 r->os_fingerprint))) 4696 r = TAILQ_NEXT(r, entries); 4697 else { 4698 if (r->tag) 4699 tag = r->tag; 4700 if (r->anchor == NULL) { 4701 if (r->action == PF_MATCH) { 4702 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4703 if (ri == NULL) { 4704 REASON_SET(&reason, PFRES_MEMORY); 4705 goto cleanup; 4706 } 4707 ri->r = r; 4708 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4709 pf_counter_u64_critical_enter(); 4710 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 4711 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 4712 pf_counter_u64_critical_exit(); 4713 pf_rule_to_actions(r, &pd->act); 4714 if (r->log) 4715 PFLOG_PACKET(kif, m, af, 4716 r->action, PFRES_MATCH, r, 4717 a, ruleset, pd, 1); 4718 } else { 4719 match = 1; 4720 *rm = r; 4721 *am = a; 4722 *rsm = ruleset; 4723 } 4724 if ((*rm)->quick) 4725 break; 4726 r = TAILQ_NEXT(r, entries); 4727 } else 4728 pf_step_into_anchor(anchor_stack, &asd, 4729 &ruleset, PF_RULESET_FILTER, &r, &a, 4730 &match); 4731 } 4732 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4733 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4734 break; 4735 } 4736 r = *rm; 4737 a = *am; 4738 ruleset = *rsm; 4739 4740 REASON_SET(&reason, PFRES_MATCH); 4741 4742 /* apply actions for last matching pass/block rule */ 4743 pf_rule_to_actions(r, &pd->act); 4744 4745 if (r->log) { 4746 if (rewrite) 4747 m_copyback(m, off, hdrlen, pd->hdr.any); 4748 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 4749 } 4750 4751 if ((r->action == PF_DROP) && 4752 ((r->rule_flag & PFRULE_RETURNRST) || 4753 (r->rule_flag & PFRULE_RETURNICMP) || 4754 (r->rule_flag & PFRULE_RETURN))) { 4755 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4756 bip_sum, hdrlen, &reason, r->rtableid); 4757 } 4758 4759 if (r->action == PF_DROP) 4760 goto cleanup; 4761 4762 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4763 REASON_SET(&reason, PFRES_MEMORY); 4764 goto cleanup; 4765 } 4766 if (pd->act.rtableid >= 0) 4767 M_SETFIB(m, pd->act.rtableid); 4768 4769 if (!state_icmp && (r->keep_state || nr != NULL || 4770 (pd->flags & PFDESC_TCP_NORM))) { 4771 int action; 4772 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4773 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4774 hdrlen, &match_rules); 4775 if (action != PF_PASS) { 4776 if (action == PF_DROP && 4777 (r->rule_flag & PFRULE_RETURN)) 4778 pf_return(r, nr, pd, sk, off, m, th, kif, 4779 bproto_sum, bip_sum, hdrlen, &reason, 4780 pd->act.rtableid); 4781 return (action); 4782 } 4783 } else { 4784 while ((ri = SLIST_FIRST(&match_rules))) { 4785 SLIST_REMOVE_HEAD(&match_rules, entry); 4786 free(ri, M_PF_RULE_ITEM); 4787 } 4788 4789 uma_zfree(V_pf_state_key_z, sk); 4790 uma_zfree(V_pf_state_key_z, nk); 4791 } 4792 4793 /* copy back packet headers if we performed NAT operations */ 4794 if (rewrite) 4795 m_copyback(m, off, hdrlen, pd->hdr.any); 4796 4797 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4798 pd->dir == PF_OUT && 4799 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4800 /* 4801 * We want the state created, but we dont 4802 * want to send this in case a partner 4803 * firewall has to know about it to allow 4804 * replies through it. 4805 */ 4806 return (PF_DEFER); 4807 4808 return (PF_PASS); 4809 4810 cleanup: 4811 while ((ri = SLIST_FIRST(&match_rules))) { 4812 SLIST_REMOVE_HEAD(&match_rules, entry); 4813 free(ri, M_PF_RULE_ITEM); 4814 } 4815 4816 uma_zfree(V_pf_state_key_z, sk); 4817 uma_zfree(V_pf_state_key_z, nk); 4818 return (PF_DROP); 4819 } 4820 4821 static int 4822 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4823 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4824 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4825 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4826 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 4827 struct pf_krule_slist *match_rules) 4828 { 4829 struct pf_kstate *s = NULL; 4830 struct pf_ksrc_node *sn = NULL; 4831 struct tcphdr *th = &pd->hdr.tcp; 4832 u_int16_t mss = V_tcp_mssdflt; 4833 u_short reason, sn_reason; 4834 struct pf_krule_item *ri; 4835 4836 /* check maximums */ 4837 if (r->max_states && 4838 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4839 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4840 REASON_SET(&reason, PFRES_MAXSTATES); 4841 goto csfailed; 4842 } 4843 /* src node for filter rule */ 4844 if ((r->rule_flag & PFRULE_SRCTRACK || 4845 r->rpool.opts & PF_POOL_STICKYADDR) && 4846 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 4847 REASON_SET(&reason, sn_reason); 4848 goto csfailed; 4849 } 4850 /* src node for translation rule */ 4851 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4852 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 4853 pd->af)) != 0 ) { 4854 REASON_SET(&reason, sn_reason); 4855 goto csfailed; 4856 } 4857 s = pf_alloc_state(M_NOWAIT); 4858 if (s == NULL) { 4859 REASON_SET(&reason, PFRES_MEMORY); 4860 goto csfailed; 4861 } 4862 s->rule.ptr = r; 4863 s->nat_rule.ptr = nr; 4864 s->anchor.ptr = a; 4865 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 4866 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 4867 4868 STATE_INC_COUNTERS(s); 4869 if (r->allow_opts) 4870 s->state_flags |= PFSTATE_ALLOWOPTS; 4871 if (r->rule_flag & PFRULE_STATESLOPPY) 4872 s->state_flags |= PFSTATE_SLOPPY; 4873 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 4874 s->state_flags |= PFSTATE_SCRUB_TCP; 4875 4876 s->act.log = pd->act.log & PF_LOG_ALL; 4877 s->sync_state = PFSYNC_S_NONE; 4878 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 4879 4880 if (nr != NULL) 4881 s->act.log |= nr->log & PF_LOG_ALL; 4882 switch (pd->proto) { 4883 case IPPROTO_TCP: 4884 s->src.seqlo = ntohl(th->th_seq); 4885 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4886 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4887 r->keep_state == PF_STATE_MODULATE) { 4888 /* Generate sequence number modulator */ 4889 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4890 0) 4891 s->src.seqdiff = 1; 4892 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4893 htonl(s->src.seqlo + s->src.seqdiff), 0); 4894 *rewrite = 1; 4895 } else 4896 s->src.seqdiff = 0; 4897 if (th->th_flags & TH_SYN) { 4898 s->src.seqhi++; 4899 s->src.wscale = pf_get_wscale(m, off, 4900 th->th_off, pd->af); 4901 } 4902 s->src.max_win = MAX(ntohs(th->th_win), 1); 4903 if (s->src.wscale & PF_WSCALE_MASK) { 4904 /* Remove scale factor from initial window */ 4905 int win = s->src.max_win; 4906 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4907 s->src.max_win = (win - 1) >> 4908 (s->src.wscale & PF_WSCALE_MASK); 4909 } 4910 if (th->th_flags & TH_FIN) 4911 s->src.seqhi++; 4912 s->dst.seqhi = 1; 4913 s->dst.max_win = 1; 4914 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4915 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4916 s->timeout = PFTM_TCP_FIRST_PACKET; 4917 atomic_add_32(&V_pf_status.states_halfopen, 1); 4918 break; 4919 case IPPROTO_UDP: 4920 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4921 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4922 s->timeout = PFTM_UDP_FIRST_PACKET; 4923 break; 4924 case IPPROTO_SCTP: 4925 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 4926 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 4927 s->timeout = PFTM_SCTP_FIRST_PACKET; 4928 break; 4929 case IPPROTO_ICMP: 4930 #ifdef INET6 4931 case IPPROTO_ICMPV6: 4932 #endif 4933 s->timeout = PFTM_ICMP_FIRST_PACKET; 4934 break; 4935 default: 4936 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4937 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4938 s->timeout = PFTM_OTHER_FIRST_PACKET; 4939 } 4940 4941 if (r->rt) { 4942 /* pf_map_addr increases the reason counters */ 4943 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 4944 &s->rt_kif, NULL, &sn)) != 0) 4945 goto csfailed; 4946 s->rt = r->rt; 4947 } 4948 4949 s->creation = time_uptime; 4950 s->expire = time_uptime; 4951 4952 if (sn != NULL) 4953 s->src_node = sn; 4954 if (nsn != NULL) { 4955 /* XXX We only modify one side for now. */ 4956 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4957 s->nat_src_node = nsn; 4958 } 4959 if (pd->proto == IPPROTO_TCP) { 4960 if (s->state_flags & PFSTATE_SCRUB_TCP && 4961 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 4962 REASON_SET(&reason, PFRES_MEMORY); 4963 goto drop; 4964 } 4965 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 4966 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4967 &s->src, &s->dst, rewrite)) { 4968 /* This really shouldn't happen!!! */ 4969 DPFPRINTF(PF_DEBUG_URGENT, 4970 ("pf_normalize_tcp_stateful failed on first " 4971 "pkt\n")); 4972 goto drop; 4973 } 4974 } else if (pd->proto == IPPROTO_SCTP) { 4975 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 4976 goto drop; 4977 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 4978 goto drop; 4979 } 4980 s->direction = pd->dir; 4981 4982 /* 4983 * sk/nk could already been setup by pf_get_translation(). 4984 */ 4985 if (nr == NULL) { 4986 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4987 __func__, nr, sk, nk)); 4988 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4989 if (sk == NULL) 4990 goto csfailed; 4991 nk = sk; 4992 } else 4993 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4994 __func__, nr, sk, nk)); 4995 4996 /* Swap sk/nk for PF_OUT. */ 4997 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4998 (pd->dir == PF_IN) ? sk : nk, 4999 (pd->dir == PF_IN) ? nk : sk, s)) { 5000 REASON_SET(&reason, PFRES_STATEINS); 5001 goto drop; 5002 } else 5003 *sm = s; 5004 5005 if (tag > 0) 5006 s->tag = tag; 5007 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5008 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5009 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5010 /* undo NAT changes, if they have taken place */ 5011 if (nr != NULL) { 5012 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5013 if (pd->dir == PF_OUT) 5014 skt = s->key[PF_SK_STACK]; 5015 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5016 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5017 if (pd->sport) 5018 *pd->sport = skt->port[pd->sidx]; 5019 if (pd->dport) 5020 *pd->dport = skt->port[pd->didx]; 5021 if (pd->proto_sum) 5022 *pd->proto_sum = bproto_sum; 5023 if (pd->ip_sum) 5024 *pd->ip_sum = bip_sum; 5025 m_copyback(m, off, hdrlen, pd->hdr.any); 5026 } 5027 s->src.seqhi = htonl(arc4random()); 5028 /* Find mss option */ 5029 int rtid = M_GETFIB(m); 5030 mss = pf_get_mss(m, off, th->th_off, pd->af); 5031 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5032 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5033 s->src.mss = mss; 5034 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5035 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5036 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5037 pd->act.rtableid); 5038 REASON_SET(&reason, PFRES_SYNPROXY); 5039 return (PF_SYNPROXY_DROP); 5040 } 5041 5042 return (PF_PASS); 5043 5044 csfailed: 5045 while ((ri = SLIST_FIRST(match_rules))) { 5046 SLIST_REMOVE_HEAD(match_rules, entry); 5047 free(ri, M_PF_RULE_ITEM); 5048 } 5049 5050 uma_zfree(V_pf_state_key_z, sk); 5051 uma_zfree(V_pf_state_key_z, nk); 5052 5053 if (sn != NULL) { 5054 PF_SRC_NODE_LOCK(sn); 5055 if (--sn->states == 0 && sn->expire == 0) { 5056 pf_unlink_src_node(sn); 5057 uma_zfree(V_pf_sources_z, sn); 5058 counter_u64_add( 5059 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5060 } 5061 PF_SRC_NODE_UNLOCK(sn); 5062 } 5063 5064 if (nsn != sn && nsn != NULL) { 5065 PF_SRC_NODE_LOCK(nsn); 5066 if (--nsn->states == 0 && nsn->expire == 0) { 5067 pf_unlink_src_node(nsn); 5068 uma_zfree(V_pf_sources_z, nsn); 5069 counter_u64_add( 5070 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5071 } 5072 PF_SRC_NODE_UNLOCK(nsn); 5073 } 5074 5075 drop: 5076 if (s != NULL) { 5077 pf_src_tree_remove_state(s); 5078 s->timeout = PFTM_UNLINKED; 5079 STATE_DEC_COUNTERS(s); 5080 pf_free_state(s); 5081 } 5082 5083 return (PF_DROP); 5084 } 5085 5086 static int 5087 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5088 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5089 struct pf_kruleset **rsm) 5090 { 5091 struct pf_krule *r, *a = NULL; 5092 struct pf_kruleset *ruleset = NULL; 5093 struct pf_krule_slist match_rules; 5094 struct pf_krule_item *ri; 5095 sa_family_t af = pd->af; 5096 u_short reason; 5097 int tag = -1; 5098 int asd = 0; 5099 int match = 0; 5100 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5101 5102 PF_RULES_RASSERT(); 5103 5104 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5105 SLIST_INIT(&match_rules); 5106 while (r != NULL) { 5107 pf_counter_u64_add(&r->evaluations, 1); 5108 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5109 r = r->skip[PF_SKIP_IFP].ptr; 5110 else if (r->direction && r->direction != pd->dir) 5111 r = r->skip[PF_SKIP_DIR].ptr; 5112 else if (r->af && r->af != af) 5113 r = r->skip[PF_SKIP_AF].ptr; 5114 else if (r->proto && r->proto != pd->proto) 5115 r = r->skip[PF_SKIP_PROTO].ptr; 5116 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5117 r->src.neg, kif, M_GETFIB(m))) 5118 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5119 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5120 r->dst.neg, NULL, M_GETFIB(m))) 5121 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5122 else if (r->tos && !(r->tos == pd->tos)) 5123 r = TAILQ_NEXT(r, entries); 5124 else if (r->os_fingerprint != PF_OSFP_ANY) 5125 r = TAILQ_NEXT(r, entries); 5126 else if (pd->proto == IPPROTO_UDP && 5127 (r->src.port_op || r->dst.port_op)) 5128 r = TAILQ_NEXT(r, entries); 5129 else if (pd->proto == IPPROTO_TCP && 5130 (r->src.port_op || r->dst.port_op || r->flagset)) 5131 r = TAILQ_NEXT(r, entries); 5132 else if ((pd->proto == IPPROTO_ICMP || 5133 pd->proto == IPPROTO_ICMPV6) && 5134 (r->type || r->code)) 5135 r = TAILQ_NEXT(r, entries); 5136 else if (r->prio && 5137 !pf_match_ieee8021q_pcp(r->prio, m)) 5138 r = TAILQ_NEXT(r, entries); 5139 else if (r->prob && r->prob <= 5140 (arc4random() % (UINT_MAX - 1) + 1)) 5141 r = TAILQ_NEXT(r, entries); 5142 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5143 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5144 r = TAILQ_NEXT(r, entries); 5145 else { 5146 if (r->anchor == NULL) { 5147 if (r->action == PF_MATCH) { 5148 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5149 if (ri == NULL) { 5150 REASON_SET(&reason, PFRES_MEMORY); 5151 goto cleanup; 5152 } 5153 ri->r = r; 5154 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5155 pf_counter_u64_critical_enter(); 5156 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5157 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5158 pf_counter_u64_critical_exit(); 5159 pf_rule_to_actions(r, &pd->act); 5160 if (r->log) 5161 PFLOG_PACKET(kif, m, af, 5162 r->action, PFRES_MATCH, r, 5163 a, ruleset, pd, 1); 5164 } else { 5165 match = 1; 5166 *rm = r; 5167 *am = a; 5168 *rsm = ruleset; 5169 } 5170 if ((*rm)->quick) 5171 break; 5172 r = TAILQ_NEXT(r, entries); 5173 } else 5174 pf_step_into_anchor(anchor_stack, &asd, 5175 &ruleset, PF_RULESET_FILTER, &r, &a, 5176 &match); 5177 } 5178 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5179 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5180 break; 5181 } 5182 r = *rm; 5183 a = *am; 5184 ruleset = *rsm; 5185 5186 REASON_SET(&reason, PFRES_MATCH); 5187 5188 /* apply actions for last matching pass/block rule */ 5189 pf_rule_to_actions(r, &pd->act); 5190 5191 if (r->log) 5192 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 5193 5194 if (r->action != PF_PASS) 5195 return (PF_DROP); 5196 5197 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5198 REASON_SET(&reason, PFRES_MEMORY); 5199 goto cleanup; 5200 } 5201 5202 return (PF_PASS); 5203 5204 cleanup: 5205 while ((ri = SLIST_FIRST(&match_rules))) { 5206 SLIST_REMOVE_HEAD(&match_rules, entry); 5207 free(ri, M_PF_RULE_ITEM); 5208 } 5209 5210 return (PF_DROP); 5211 } 5212 5213 static int 5214 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5215 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5216 int *copyback) 5217 { 5218 struct tcphdr *th = &pd->hdr.tcp; 5219 struct pf_state_peer *src, *dst; 5220 u_int16_t win = ntohs(th->th_win); 5221 u_int32_t ack, end, seq, orig_seq; 5222 u_int8_t sws, dws, psrc, pdst; 5223 int ackskew; 5224 5225 if (pd->dir == (*state)->direction) { 5226 src = &(*state)->src; 5227 dst = &(*state)->dst; 5228 psrc = PF_PEER_SRC; 5229 pdst = PF_PEER_DST; 5230 } else { 5231 src = &(*state)->dst; 5232 dst = &(*state)->src; 5233 psrc = PF_PEER_DST; 5234 pdst = PF_PEER_SRC; 5235 } 5236 5237 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5238 sws = src->wscale & PF_WSCALE_MASK; 5239 dws = dst->wscale & PF_WSCALE_MASK; 5240 } else 5241 sws = dws = 0; 5242 5243 /* 5244 * Sequence tracking algorithm from Guido van Rooij's paper: 5245 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5246 * tcp_filtering.ps 5247 */ 5248 5249 orig_seq = seq = ntohl(th->th_seq); 5250 if (src->seqlo == 0) { 5251 /* First packet from this end. Set its state */ 5252 5253 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5254 src->scrub == NULL) { 5255 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5256 REASON_SET(reason, PFRES_MEMORY); 5257 return (PF_DROP); 5258 } 5259 } 5260 5261 /* Deferred generation of sequence number modulator */ 5262 if (dst->seqdiff && !src->seqdiff) { 5263 /* use random iss for the TCP server */ 5264 while ((src->seqdiff = arc4random() - seq) == 0) 5265 ; 5266 ack = ntohl(th->th_ack) - dst->seqdiff; 5267 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5268 src->seqdiff), 0); 5269 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5270 *copyback = 1; 5271 } else { 5272 ack = ntohl(th->th_ack); 5273 } 5274 5275 end = seq + pd->p_len; 5276 if (th->th_flags & TH_SYN) { 5277 end++; 5278 if (dst->wscale & PF_WSCALE_FLAG) { 5279 src->wscale = pf_get_wscale(m, off, th->th_off, 5280 pd->af); 5281 if (src->wscale & PF_WSCALE_FLAG) { 5282 /* Remove scale factor from initial 5283 * window */ 5284 sws = src->wscale & PF_WSCALE_MASK; 5285 win = ((u_int32_t)win + (1 << sws) - 1) 5286 >> sws; 5287 dws = dst->wscale & PF_WSCALE_MASK; 5288 } else { 5289 /* fixup other window */ 5290 dst->max_win <<= dst->wscale & 5291 PF_WSCALE_MASK; 5292 /* in case of a retrans SYN|ACK */ 5293 dst->wscale = 0; 5294 } 5295 } 5296 } 5297 if (th->th_flags & TH_FIN) 5298 end++; 5299 5300 src->seqlo = seq; 5301 if (src->state < TCPS_SYN_SENT) 5302 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5303 5304 /* 5305 * May need to slide the window (seqhi may have been set by 5306 * the crappy stack check or if we picked up the connection 5307 * after establishment) 5308 */ 5309 if (src->seqhi == 1 || 5310 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5311 src->seqhi = end + MAX(1, dst->max_win << dws); 5312 if (win > src->max_win) 5313 src->max_win = win; 5314 5315 } else { 5316 ack = ntohl(th->th_ack) - dst->seqdiff; 5317 if (src->seqdiff) { 5318 /* Modulate sequence numbers */ 5319 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5320 src->seqdiff), 0); 5321 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5322 *copyback = 1; 5323 } 5324 end = seq + pd->p_len; 5325 if (th->th_flags & TH_SYN) 5326 end++; 5327 if (th->th_flags & TH_FIN) 5328 end++; 5329 } 5330 5331 if ((th->th_flags & TH_ACK) == 0) { 5332 /* Let it pass through the ack skew check */ 5333 ack = dst->seqlo; 5334 } else if ((ack == 0 && 5335 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5336 /* broken tcp stacks do not set ack */ 5337 (dst->state < TCPS_SYN_SENT)) { 5338 /* 5339 * Many stacks (ours included) will set the ACK number in an 5340 * FIN|ACK if the SYN times out -- no sequence to ACK. 5341 */ 5342 ack = dst->seqlo; 5343 } 5344 5345 if (seq == end) { 5346 /* Ease sequencing restrictions on no data packets */ 5347 seq = src->seqlo; 5348 end = seq; 5349 } 5350 5351 ackskew = dst->seqlo - ack; 5352 5353 /* 5354 * Need to demodulate the sequence numbers in any TCP SACK options 5355 * (Selective ACK). We could optionally validate the SACK values 5356 * against the current ACK window, either forwards or backwards, but 5357 * I'm not confident that SACK has been implemented properly 5358 * everywhere. It wouldn't surprise me if several stacks accidentally 5359 * SACK too far backwards of previously ACKed data. There really aren't 5360 * any security implications of bad SACKing unless the target stack 5361 * doesn't validate the option length correctly. Someone trying to 5362 * spoof into a TCP connection won't bother blindly sending SACK 5363 * options anyway. 5364 */ 5365 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5366 if (pf_modulate_sack(m, off, pd, th, dst)) 5367 *copyback = 1; 5368 } 5369 5370 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5371 if (SEQ_GEQ(src->seqhi, end) && 5372 /* Last octet inside other's window space */ 5373 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5374 /* Retrans: not more than one window back */ 5375 (ackskew >= -MAXACKWINDOW) && 5376 /* Acking not more than one reassembled fragment backwards */ 5377 (ackskew <= (MAXACKWINDOW << sws)) && 5378 /* Acking not more than one window forward */ 5379 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5380 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5381 /* Require an exact/+1 sequence match on resets when possible */ 5382 5383 if (dst->scrub || src->scrub) { 5384 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5385 *state, src, dst, copyback)) 5386 return (PF_DROP); 5387 } 5388 5389 /* update max window */ 5390 if (src->max_win < win) 5391 src->max_win = win; 5392 /* synchronize sequencing */ 5393 if (SEQ_GT(end, src->seqlo)) 5394 src->seqlo = end; 5395 /* slide the window of what the other end can send */ 5396 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5397 dst->seqhi = ack + MAX((win << sws), 1); 5398 5399 /* update states */ 5400 if (th->th_flags & TH_SYN) 5401 if (src->state < TCPS_SYN_SENT) 5402 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5403 if (th->th_flags & TH_FIN) 5404 if (src->state < TCPS_CLOSING) 5405 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5406 if (th->th_flags & TH_ACK) { 5407 if (dst->state == TCPS_SYN_SENT) { 5408 pf_set_protostate(*state, pdst, 5409 TCPS_ESTABLISHED); 5410 if (src->state == TCPS_ESTABLISHED && 5411 (*state)->src_node != NULL && 5412 pf_src_connlimit(state)) { 5413 REASON_SET(reason, PFRES_SRCLIMIT); 5414 return (PF_DROP); 5415 } 5416 } else if (dst->state == TCPS_CLOSING) 5417 pf_set_protostate(*state, pdst, 5418 TCPS_FIN_WAIT_2); 5419 } 5420 if (th->th_flags & TH_RST) 5421 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5422 5423 /* update expire time */ 5424 (*state)->expire = time_uptime; 5425 if (src->state >= TCPS_FIN_WAIT_2 && 5426 dst->state >= TCPS_FIN_WAIT_2) 5427 (*state)->timeout = PFTM_TCP_CLOSED; 5428 else if (src->state >= TCPS_CLOSING && 5429 dst->state >= TCPS_CLOSING) 5430 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5431 else if (src->state < TCPS_ESTABLISHED || 5432 dst->state < TCPS_ESTABLISHED) 5433 (*state)->timeout = PFTM_TCP_OPENING; 5434 else if (src->state >= TCPS_CLOSING || 5435 dst->state >= TCPS_CLOSING) 5436 (*state)->timeout = PFTM_TCP_CLOSING; 5437 else 5438 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5439 5440 /* Fall through to PASS packet */ 5441 5442 } else if ((dst->state < TCPS_SYN_SENT || 5443 dst->state >= TCPS_FIN_WAIT_2 || 5444 src->state >= TCPS_FIN_WAIT_2) && 5445 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5446 /* Within a window forward of the originating packet */ 5447 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5448 /* Within a window backward of the originating packet */ 5449 5450 /* 5451 * This currently handles three situations: 5452 * 1) Stupid stacks will shotgun SYNs before their peer 5453 * replies. 5454 * 2) When PF catches an already established stream (the 5455 * firewall rebooted, the state table was flushed, routes 5456 * changed...) 5457 * 3) Packets get funky immediately after the connection 5458 * closes (this should catch Solaris spurious ACK|FINs 5459 * that web servers like to spew after a close) 5460 * 5461 * This must be a little more careful than the above code 5462 * since packet floods will also be caught here. We don't 5463 * update the TTL here to mitigate the damage of a packet 5464 * flood and so the same code can handle awkward establishment 5465 * and a loosened connection close. 5466 * In the establishment case, a correct peer response will 5467 * validate the connection, go through the normal state code 5468 * and keep updating the state TTL. 5469 */ 5470 5471 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5472 printf("pf: loose state match: "); 5473 pf_print_state(*state); 5474 pf_print_flags(th->th_flags); 5475 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5476 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5477 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5478 (unsigned long long)(*state)->packets[1], 5479 pd->dir == PF_IN ? "in" : "out", 5480 pd->dir == (*state)->direction ? "fwd" : "rev"); 5481 } 5482 5483 if (dst->scrub || src->scrub) { 5484 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5485 *state, src, dst, copyback)) 5486 return (PF_DROP); 5487 } 5488 5489 /* update max window */ 5490 if (src->max_win < win) 5491 src->max_win = win; 5492 /* synchronize sequencing */ 5493 if (SEQ_GT(end, src->seqlo)) 5494 src->seqlo = end; 5495 /* slide the window of what the other end can send */ 5496 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5497 dst->seqhi = ack + MAX((win << sws), 1); 5498 5499 /* 5500 * Cannot set dst->seqhi here since this could be a shotgunned 5501 * SYN and not an already established connection. 5502 */ 5503 5504 if (th->th_flags & TH_FIN) 5505 if (src->state < TCPS_CLOSING) 5506 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5507 if (th->th_flags & TH_RST) 5508 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5509 5510 /* Fall through to PASS packet */ 5511 5512 } else { 5513 if ((*state)->dst.state == TCPS_SYN_SENT && 5514 (*state)->src.state == TCPS_SYN_SENT) { 5515 /* Send RST for state mismatches during handshake */ 5516 if (!(th->th_flags & TH_RST)) 5517 pf_send_tcp((*state)->rule.ptr, pd->af, 5518 pd->dst, pd->src, th->th_dport, 5519 th->th_sport, ntohl(th->th_ack), 0, 5520 TH_RST, 0, 0, 5521 (*state)->rule.ptr->return_ttl, true, 0, 0, 5522 (*state)->act.rtableid); 5523 src->seqlo = 0; 5524 src->seqhi = 1; 5525 src->max_win = 1; 5526 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5527 printf("pf: BAD state: "); 5528 pf_print_state(*state); 5529 pf_print_flags(th->th_flags); 5530 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5531 "pkts=%llu:%llu dir=%s,%s\n", 5532 seq, orig_seq, ack, pd->p_len, ackskew, 5533 (unsigned long long)(*state)->packets[0], 5534 (unsigned long long)(*state)->packets[1], 5535 pd->dir == PF_IN ? "in" : "out", 5536 pd->dir == (*state)->direction ? "fwd" : "rev"); 5537 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5538 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5539 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5540 ' ': '2', 5541 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5542 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5543 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5544 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5545 } 5546 REASON_SET(reason, PFRES_BADSTATE); 5547 return (PF_DROP); 5548 } 5549 5550 return (PF_PASS); 5551 } 5552 5553 static int 5554 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5555 { 5556 struct tcphdr *th = &pd->hdr.tcp; 5557 struct pf_state_peer *src, *dst; 5558 u_int8_t psrc, pdst; 5559 5560 if (pd->dir == (*state)->direction) { 5561 src = &(*state)->src; 5562 dst = &(*state)->dst; 5563 psrc = PF_PEER_SRC; 5564 pdst = PF_PEER_DST; 5565 } else { 5566 src = &(*state)->dst; 5567 dst = &(*state)->src; 5568 psrc = PF_PEER_DST; 5569 pdst = PF_PEER_SRC; 5570 } 5571 5572 if (th->th_flags & TH_SYN) 5573 if (src->state < TCPS_SYN_SENT) 5574 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5575 if (th->th_flags & TH_FIN) 5576 if (src->state < TCPS_CLOSING) 5577 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5578 if (th->th_flags & TH_ACK) { 5579 if (dst->state == TCPS_SYN_SENT) { 5580 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5581 if (src->state == TCPS_ESTABLISHED && 5582 (*state)->src_node != NULL && 5583 pf_src_connlimit(state)) { 5584 REASON_SET(reason, PFRES_SRCLIMIT); 5585 return (PF_DROP); 5586 } 5587 } else if (dst->state == TCPS_CLOSING) { 5588 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5589 } else if (src->state == TCPS_SYN_SENT && 5590 dst->state < TCPS_SYN_SENT) { 5591 /* 5592 * Handle a special sloppy case where we only see one 5593 * half of the connection. If there is a ACK after 5594 * the initial SYN without ever seeing a packet from 5595 * the destination, set the connection to established. 5596 */ 5597 pf_set_protostate(*state, PF_PEER_BOTH, 5598 TCPS_ESTABLISHED); 5599 dst->state = src->state = TCPS_ESTABLISHED; 5600 if ((*state)->src_node != NULL && 5601 pf_src_connlimit(state)) { 5602 REASON_SET(reason, PFRES_SRCLIMIT); 5603 return (PF_DROP); 5604 } 5605 } else if (src->state == TCPS_CLOSING && 5606 dst->state == TCPS_ESTABLISHED && 5607 dst->seqlo == 0) { 5608 /* 5609 * Handle the closing of half connections where we 5610 * don't see the full bidirectional FIN/ACK+ACK 5611 * handshake. 5612 */ 5613 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5614 } 5615 } 5616 if (th->th_flags & TH_RST) 5617 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5618 5619 /* update expire time */ 5620 (*state)->expire = time_uptime; 5621 if (src->state >= TCPS_FIN_WAIT_2 && 5622 dst->state >= TCPS_FIN_WAIT_2) 5623 (*state)->timeout = PFTM_TCP_CLOSED; 5624 else if (src->state >= TCPS_CLOSING && 5625 dst->state >= TCPS_CLOSING) 5626 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5627 else if (src->state < TCPS_ESTABLISHED || 5628 dst->state < TCPS_ESTABLISHED) 5629 (*state)->timeout = PFTM_TCP_OPENING; 5630 else if (src->state >= TCPS_CLOSING || 5631 dst->state >= TCPS_CLOSING) 5632 (*state)->timeout = PFTM_TCP_CLOSING; 5633 else 5634 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5635 5636 return (PF_PASS); 5637 } 5638 5639 static int 5640 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5641 { 5642 struct pf_state_key *sk = (*state)->key[pd->didx]; 5643 struct tcphdr *th = &pd->hdr.tcp; 5644 5645 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5646 if (pd->dir != (*state)->direction) { 5647 REASON_SET(reason, PFRES_SYNPROXY); 5648 return (PF_SYNPROXY_DROP); 5649 } 5650 if (th->th_flags & TH_SYN) { 5651 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5652 REASON_SET(reason, PFRES_SYNPROXY); 5653 return (PF_DROP); 5654 } 5655 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5656 pd->src, th->th_dport, th->th_sport, 5657 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5658 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 5659 (*state)->act.rtableid); 5660 REASON_SET(reason, PFRES_SYNPROXY); 5661 return (PF_SYNPROXY_DROP); 5662 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5663 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5664 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5665 REASON_SET(reason, PFRES_SYNPROXY); 5666 return (PF_DROP); 5667 } else if ((*state)->src_node != NULL && 5668 pf_src_connlimit(state)) { 5669 REASON_SET(reason, PFRES_SRCLIMIT); 5670 return (PF_DROP); 5671 } else 5672 pf_set_protostate(*state, PF_PEER_SRC, 5673 PF_TCPS_PROXY_DST); 5674 } 5675 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5676 if (pd->dir == (*state)->direction) { 5677 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5678 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5679 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5680 REASON_SET(reason, PFRES_SYNPROXY); 5681 return (PF_DROP); 5682 } 5683 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5684 if ((*state)->dst.seqhi == 1) 5685 (*state)->dst.seqhi = htonl(arc4random()); 5686 pf_send_tcp((*state)->rule.ptr, pd->af, 5687 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5688 sk->port[pd->sidx], sk->port[pd->didx], 5689 (*state)->dst.seqhi, 0, TH_SYN, 0, 5690 (*state)->src.mss, 0, false, (*state)->tag, 0, 5691 (*state)->act.rtableid); 5692 REASON_SET(reason, PFRES_SYNPROXY); 5693 return (PF_SYNPROXY_DROP); 5694 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5695 (TH_SYN|TH_ACK)) || 5696 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5697 REASON_SET(reason, PFRES_SYNPROXY); 5698 return (PF_DROP); 5699 } else { 5700 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5701 (*state)->dst.seqlo = ntohl(th->th_seq); 5702 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5703 pd->src, th->th_dport, th->th_sport, 5704 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5705 TH_ACK, (*state)->src.max_win, 0, 0, false, 5706 (*state)->tag, 0, (*state)->act.rtableid); 5707 pf_send_tcp((*state)->rule.ptr, pd->af, 5708 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5709 sk->port[pd->sidx], sk->port[pd->didx], 5710 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5711 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 5712 (*state)->act.rtableid); 5713 (*state)->src.seqdiff = (*state)->dst.seqhi - 5714 (*state)->src.seqlo; 5715 (*state)->dst.seqdiff = (*state)->src.seqhi - 5716 (*state)->dst.seqlo; 5717 (*state)->src.seqhi = (*state)->src.seqlo + 5718 (*state)->dst.max_win; 5719 (*state)->dst.seqhi = (*state)->dst.seqlo + 5720 (*state)->src.max_win; 5721 (*state)->src.wscale = (*state)->dst.wscale = 0; 5722 pf_set_protostate(*state, PF_PEER_BOTH, 5723 TCPS_ESTABLISHED); 5724 REASON_SET(reason, PFRES_SYNPROXY); 5725 return (PF_SYNPROXY_DROP); 5726 } 5727 } 5728 5729 return (PF_PASS); 5730 } 5731 5732 static int 5733 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 5734 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5735 u_short *reason) 5736 { 5737 struct pf_state_key_cmp key; 5738 struct tcphdr *th = &pd->hdr.tcp; 5739 int copyback = 0; 5740 int action; 5741 struct pf_state_peer *src, *dst; 5742 5743 bzero(&key, sizeof(key)); 5744 key.af = pd->af; 5745 key.proto = IPPROTO_TCP; 5746 if (pd->dir == PF_IN) { /* wire side, straight */ 5747 PF_ACPY(&key.addr[0], pd->src, key.af); 5748 PF_ACPY(&key.addr[1], pd->dst, key.af); 5749 key.port[0] = th->th_sport; 5750 key.port[1] = th->th_dport; 5751 } else { /* stack side, reverse */ 5752 PF_ACPY(&key.addr[1], pd->src, key.af); 5753 PF_ACPY(&key.addr[0], pd->dst, key.af); 5754 key.port[1] = th->th_sport; 5755 key.port[0] = th->th_dport; 5756 } 5757 5758 STATE_LOOKUP(kif, &key, *state, pd); 5759 5760 if (pd->dir == (*state)->direction) { 5761 src = &(*state)->src; 5762 dst = &(*state)->dst; 5763 } else { 5764 src = &(*state)->dst; 5765 dst = &(*state)->src; 5766 } 5767 5768 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5769 return (action); 5770 5771 if (dst->state >= TCPS_FIN_WAIT_2 && 5772 src->state >= TCPS_FIN_WAIT_2 && 5773 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 5774 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 5775 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 5776 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5777 printf("pf: state reuse "); 5778 pf_print_state(*state); 5779 pf_print_flags(th->th_flags); 5780 printf("\n"); 5781 } 5782 /* XXX make sure it's the same direction ?? */ 5783 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5784 pf_unlink_state(*state); 5785 *state = NULL; 5786 return (PF_DROP); 5787 } 5788 5789 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5790 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5791 return (PF_DROP); 5792 } else { 5793 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5794 ©back) == PF_DROP) 5795 return (PF_DROP); 5796 } 5797 5798 /* translate source/destination address, if necessary */ 5799 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5800 struct pf_state_key *nk = (*state)->key[pd->didx]; 5801 5802 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5803 nk->port[pd->sidx] != th->th_sport) 5804 pf_change_ap(m, pd->src, &th->th_sport, 5805 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5806 nk->port[pd->sidx], 0, pd->af); 5807 5808 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5809 nk->port[pd->didx] != th->th_dport) 5810 pf_change_ap(m, pd->dst, &th->th_dport, 5811 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5812 nk->port[pd->didx], 0, pd->af); 5813 copyback = 1; 5814 } 5815 5816 /* Copyback sequence modulation or stateful scrub changes if needed */ 5817 if (copyback) 5818 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5819 5820 return (PF_PASS); 5821 } 5822 5823 static int 5824 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 5825 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5826 { 5827 struct pf_state_peer *src, *dst; 5828 struct pf_state_key_cmp key; 5829 struct udphdr *uh = &pd->hdr.udp; 5830 uint8_t psrc, pdst; 5831 5832 bzero(&key, sizeof(key)); 5833 key.af = pd->af; 5834 key.proto = IPPROTO_UDP; 5835 if (pd->dir == PF_IN) { /* wire side, straight */ 5836 PF_ACPY(&key.addr[0], pd->src, key.af); 5837 PF_ACPY(&key.addr[1], pd->dst, key.af); 5838 key.port[0] = uh->uh_sport; 5839 key.port[1] = uh->uh_dport; 5840 } else { /* stack side, reverse */ 5841 PF_ACPY(&key.addr[1], pd->src, key.af); 5842 PF_ACPY(&key.addr[0], pd->dst, key.af); 5843 key.port[1] = uh->uh_sport; 5844 key.port[0] = uh->uh_dport; 5845 } 5846 5847 STATE_LOOKUP(kif, &key, *state, pd); 5848 5849 if (pd->dir == (*state)->direction) { 5850 src = &(*state)->src; 5851 dst = &(*state)->dst; 5852 psrc = PF_PEER_SRC; 5853 pdst = PF_PEER_DST; 5854 } else { 5855 src = &(*state)->dst; 5856 dst = &(*state)->src; 5857 psrc = PF_PEER_DST; 5858 pdst = PF_PEER_SRC; 5859 } 5860 5861 /* update states */ 5862 if (src->state < PFUDPS_SINGLE) 5863 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5864 if (dst->state == PFUDPS_SINGLE) 5865 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5866 5867 /* update expire time */ 5868 (*state)->expire = time_uptime; 5869 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5870 (*state)->timeout = PFTM_UDP_MULTIPLE; 5871 else 5872 (*state)->timeout = PFTM_UDP_SINGLE; 5873 5874 /* translate source/destination address, if necessary */ 5875 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5876 struct pf_state_key *nk = (*state)->key[pd->didx]; 5877 5878 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5879 nk->port[pd->sidx] != uh->uh_sport) 5880 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5881 &uh->uh_sum, &nk->addr[pd->sidx], 5882 nk->port[pd->sidx], 1, pd->af); 5883 5884 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5885 nk->port[pd->didx] != uh->uh_dport) 5886 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5887 &uh->uh_sum, &nk->addr[pd->didx], 5888 nk->port[pd->didx], 1, pd->af); 5889 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5890 } 5891 5892 return (PF_PASS); 5893 } 5894 5895 static int 5896 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 5897 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5898 { 5899 struct pf_state_key_cmp key; 5900 struct pf_state_peer *src, *dst; 5901 struct sctphdr *sh = &pd->hdr.sctp; 5902 u_int8_t psrc; //, pdst; 5903 5904 bzero(&key, sizeof(key)); 5905 key.af = pd->af; 5906 key.proto = IPPROTO_SCTP; 5907 if (pd->dir == PF_IN) { /* wire side, straight */ 5908 PF_ACPY(&key.addr[0], pd->src, key.af); 5909 PF_ACPY(&key.addr[1], pd->dst, key.af); 5910 key.port[0] = sh->src_port; 5911 key.port[1] = sh->dest_port; 5912 } else { /* stack side, reverse */ 5913 PF_ACPY(&key.addr[1], pd->src, key.af); 5914 PF_ACPY(&key.addr[0], pd->dst, key.af); 5915 key.port[1] = sh->src_port; 5916 key.port[0] = sh->dest_port; 5917 } 5918 5919 STATE_LOOKUP(kif, &key, *state, pd); 5920 5921 if (pd->dir == (*state)->direction) { 5922 src = &(*state)->src; 5923 dst = &(*state)->dst; 5924 psrc = PF_PEER_SRC; 5925 } else { 5926 src = &(*state)->dst; 5927 dst = &(*state)->src; 5928 psrc = PF_PEER_DST; 5929 } 5930 5931 /* Track state. */ 5932 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 5933 if (src->state < SCTP_COOKIE_WAIT) { 5934 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 5935 (*state)->timeout = PFTM_SCTP_OPENING; 5936 } 5937 } 5938 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 5939 MPASS(dst->scrub != NULL); 5940 if (dst->scrub->pfss_v_tag == 0) 5941 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 5942 } 5943 5944 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 5945 if (src->state < SCTP_ESTABLISHED) { 5946 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 5947 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 5948 } 5949 } 5950 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 5951 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5952 if (src->state < SCTP_SHUTDOWN_PENDING) { 5953 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 5954 (*state)->timeout = PFTM_SCTP_CLOSING; 5955 } 5956 } 5957 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5958 pf_set_protostate(*state, psrc, SCTP_CLOSED); 5959 (*state)->timeout = PFTM_SCTP_CLOSED; 5960 } 5961 5962 if (src->scrub != NULL) { 5963 if (src->scrub->pfss_v_tag == 0) { 5964 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 5965 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 5966 return (PF_DROP); 5967 } 5968 5969 (*state)->expire = time_uptime; 5970 5971 /* translate source/destination address, if necessary */ 5972 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5973 uint16_t checksum = 0; 5974 struct pf_state_key *nk = (*state)->key[pd->didx]; 5975 5976 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5977 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 5978 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 5979 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 5980 nk->port[pd->sidx], 1, pd->af); 5981 } 5982 5983 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5984 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 5985 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 5986 pd->ip_sum, &checksum, &nk->addr[pd->didx], 5987 nk->port[pd->didx], 1, pd->af); 5988 } 5989 } 5990 5991 return (PF_PASS); 5992 } 5993 5994 static void 5995 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 5996 { 5997 struct pf_sctp_endpoint key; 5998 struct pf_sctp_endpoint *ep; 5999 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6000 struct pf_sctp_source *i, *tmp; 6001 6002 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6003 return; 6004 6005 PF_SCTP_ENDPOINTS_LOCK(); 6006 6007 key.v_tag = s->dst.scrub->pfss_v_tag; 6008 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6009 if (ep != NULL) { 6010 /* XXX Actually remove! */ 6011 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6012 if (pf_addr_cmp(&i->addr, 6013 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6014 s->key[PF_SK_WIRE]->af) == 0) { 6015 TAILQ_REMOVE(&ep->sources, i, entry); 6016 free(i, M_PFTEMP); 6017 break; 6018 } 6019 } 6020 6021 if (TAILQ_EMPTY(&ep->sources)) { 6022 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6023 free(ep, M_PFTEMP); 6024 } 6025 } 6026 6027 /* Other direction. */ 6028 key.v_tag = s->src.scrub->pfss_v_tag; 6029 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6030 if (ep != NULL) { 6031 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6032 if (pf_addr_cmp(&i->addr, 6033 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6034 s->key[PF_SK_WIRE]->af) == 0) { 6035 TAILQ_REMOVE(&ep->sources, i, entry); 6036 free(i, M_PFTEMP); 6037 break; 6038 } 6039 } 6040 6041 if (TAILQ_EMPTY(&ep->sources)) { 6042 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6043 free(ep, M_PFTEMP); 6044 } 6045 } 6046 6047 PF_SCTP_ENDPOINTS_UNLOCK(); 6048 } 6049 6050 static void 6051 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6052 { 6053 struct pf_sctp_endpoint key = { 6054 .v_tag = v_tag, 6055 }; 6056 struct pf_sctp_source *i; 6057 struct pf_sctp_endpoint *ep; 6058 6059 PF_SCTP_ENDPOINTS_LOCK(); 6060 6061 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6062 if (ep == NULL) { 6063 ep = malloc(sizeof(struct pf_sctp_endpoint), 6064 M_PFTEMP, M_NOWAIT); 6065 if (ep == NULL) { 6066 PF_SCTP_ENDPOINTS_UNLOCK(); 6067 return; 6068 } 6069 6070 ep->v_tag = v_tag; 6071 TAILQ_INIT(&ep->sources); 6072 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6073 } 6074 6075 /* Avoid inserting duplicates. */ 6076 TAILQ_FOREACH(i, &ep->sources, entry) { 6077 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6078 PF_SCTP_ENDPOINTS_UNLOCK(); 6079 return; 6080 } 6081 } 6082 6083 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6084 if (i == NULL) { 6085 PF_SCTP_ENDPOINTS_UNLOCK(); 6086 return; 6087 } 6088 6089 i->af = pd->af; 6090 memcpy(&i->addr, a, sizeof(*a)); 6091 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6092 6093 PF_SCTP_ENDPOINTS_UNLOCK(); 6094 } 6095 6096 static void 6097 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6098 struct pf_kstate *s, int action) 6099 { 6100 struct pf_sctp_multihome_job *j, *tmp; 6101 struct pf_sctp_source *i; 6102 int ret __unused;; 6103 struct pf_kstate *sm = NULL; 6104 struct pf_krule *ra = NULL; 6105 struct pf_krule *r = &V_pf_default_rule; 6106 struct pf_kruleset *rs = NULL; 6107 bool do_extra = true; 6108 6109 PF_RULES_RLOCK_TRACKER; 6110 6111 again: 6112 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6113 if (s == NULL || action != PF_PASS) 6114 goto free; 6115 6116 /* Confirm we don't recurse here. */ 6117 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6118 6119 switch (j->op) { 6120 case SCTP_ADD_IP_ADDRESS: { 6121 uint32_t v_tag = pd->sctp_initiate_tag; 6122 6123 if (v_tag == 0) { 6124 if (s->direction == pd->dir) 6125 v_tag = s->src.scrub->pfss_v_tag; 6126 else 6127 v_tag = s->dst.scrub->pfss_v_tag; 6128 } 6129 6130 /* 6131 * Avoid duplicating states. We'll already have 6132 * created a state based on the source address of 6133 * the packet, but SCTP endpoints may also list this 6134 * address again in the INIT(_ACK) parameters. 6135 */ 6136 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6137 break; 6138 } 6139 6140 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6141 PF_RULES_RLOCK(); 6142 sm = NULL; 6143 /* 6144 * New connections need to be floating, because 6145 * we cannot know what interfaces it will use. 6146 * That's why we pass V_pfi_all rather than kif. 6147 */ 6148 ret = pf_test_rule(&r, &sm, V_pfi_all, 6149 j->m, off, &j->pd, &ra, &rs, NULL); 6150 PF_RULES_RUNLOCK(); 6151 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6152 if (ret != PF_DROP && sm != NULL) { 6153 /* Inherit v_tag values. */ 6154 if (sm->direction == s->direction) { 6155 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6156 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6157 } else { 6158 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6159 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6160 } 6161 PF_STATE_UNLOCK(sm); 6162 } else { 6163 /* If we try duplicate inserts? */ 6164 break; 6165 } 6166 6167 /* Only add the addres if we've actually allowed the state. */ 6168 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6169 6170 if (! do_extra) { 6171 break; 6172 } 6173 /* 6174 * We need to do this for each of our source addresses. 6175 * Find those based on the verification tag. 6176 */ 6177 struct pf_sctp_endpoint key = { 6178 .v_tag = pd->hdr.sctp.v_tag, 6179 }; 6180 struct pf_sctp_endpoint *ep; 6181 6182 PF_SCTP_ENDPOINTS_LOCK(); 6183 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6184 if (ep == NULL) { 6185 PF_SCTP_ENDPOINTS_UNLOCK(); 6186 break; 6187 } 6188 MPASS(ep != NULL); 6189 6190 TAILQ_FOREACH(i, &ep->sources, entry) { 6191 struct pf_sctp_multihome_job *nj; 6192 6193 /* SCTP can intermingle IPv4 and IPv6. */ 6194 if (i->af != pd->af) 6195 continue; 6196 6197 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6198 if (! nj) { 6199 continue; 6200 } 6201 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6202 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6203 nj->pd.src = &nj->src; 6204 // New destination address! 6205 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6206 nj->pd.dst = &nj->dst; 6207 nj->m = j->m; 6208 nj->op = j->op; 6209 6210 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6211 } 6212 PF_SCTP_ENDPOINTS_UNLOCK(); 6213 6214 break; 6215 } 6216 case SCTP_DEL_IP_ADDRESS: { 6217 struct pf_state_key_cmp key; 6218 uint8_t psrc; 6219 6220 bzero(&key, sizeof(key)); 6221 key.af = j->pd.af; 6222 key.proto = IPPROTO_SCTP; 6223 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6224 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6225 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6226 key.port[0] = j->pd.hdr.sctp.src_port; 6227 key.port[1] = j->pd.hdr.sctp.dest_port; 6228 } else { /* stack side, reverse */ 6229 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6230 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6231 key.port[1] = j->pd.hdr.sctp.src_port; 6232 key.port[0] = j->pd.hdr.sctp.dest_port; 6233 } 6234 6235 sm = pf_find_state(kif, &key, j->pd.dir); 6236 if (sm != NULL) { 6237 PF_STATE_LOCK_ASSERT(sm); 6238 if (j->pd.dir == sm->direction) { 6239 psrc = PF_PEER_SRC; 6240 } else { 6241 psrc = PF_PEER_DST; 6242 } 6243 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6244 sm->timeout = PFTM_SCTP_CLOSING; 6245 PF_STATE_UNLOCK(sm); 6246 } 6247 break; 6248 default: 6249 panic("Unknown op %#x", j->op); 6250 } 6251 } 6252 6253 free: 6254 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6255 free(j, M_PFTEMP); 6256 } 6257 6258 /* We may have inserted extra work while processing the list. */ 6259 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6260 do_extra = false; 6261 goto again; 6262 } 6263 } 6264 6265 static int 6266 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6267 struct pfi_kkif *kif, int op) 6268 { 6269 int off = 0; 6270 struct pf_sctp_multihome_job *job; 6271 6272 while (off < len) { 6273 struct sctp_paramhdr h; 6274 6275 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6276 pd->af)) 6277 return (PF_DROP); 6278 6279 /* Parameters are at least 4 bytes. */ 6280 if (ntohs(h.param_length) < 4) 6281 return (PF_DROP); 6282 6283 switch (ntohs(h.param_type)) { 6284 case SCTP_IPV4_ADDRESS: { 6285 struct in_addr t; 6286 6287 if (ntohs(h.param_length) != 6288 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6289 return (PF_DROP); 6290 6291 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6292 NULL, NULL, pd->af)) 6293 return (PF_DROP); 6294 6295 if (in_nullhost(t)) 6296 t.s_addr = pd->src->v4.s_addr; 6297 6298 /* 6299 * We hold the state lock (idhash) here, which means 6300 * that we can't acquire the keyhash, or we'll get a 6301 * LOR (and potentially double-lock things too). We also 6302 * can't release the state lock here, so instead we'll 6303 * enqueue this for async handling. 6304 * There's a relatively small race here, in that a 6305 * packet using the new addresses could arrive already, 6306 * but that's just though luck for it. 6307 */ 6308 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6309 if (! job) 6310 return (PF_DROP); 6311 6312 memcpy(&job->pd, pd, sizeof(*pd)); 6313 6314 // New source address! 6315 memcpy(&job->src, &t, sizeof(t)); 6316 job->pd.src = &job->src; 6317 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6318 job->pd.dst = &job->dst; 6319 job->m = m; 6320 job->op = op; 6321 6322 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6323 break; 6324 } 6325 #ifdef INET6 6326 case SCTP_IPV6_ADDRESS: { 6327 struct in6_addr t; 6328 6329 if (ntohs(h.param_length) != 6330 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6331 return (PF_DROP); 6332 6333 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6334 NULL, NULL, pd->af)) 6335 return (PF_DROP); 6336 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6337 break; 6338 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6339 memcpy(&t, &pd->src->v6, sizeof(t)); 6340 6341 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6342 if (! job) 6343 return (PF_DROP); 6344 6345 memcpy(&job->pd, pd, sizeof(*pd)); 6346 memcpy(&job->src, &t, sizeof(t)); 6347 job->pd.src = &job->src; 6348 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6349 job->pd.dst = &job->dst; 6350 job->m = m; 6351 job->op = op; 6352 6353 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6354 break; 6355 } 6356 #endif 6357 case SCTP_ADD_IP_ADDRESS: { 6358 int ret; 6359 struct sctp_asconf_paramhdr ah; 6360 6361 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6362 NULL, NULL, pd->af)) 6363 return (PF_DROP); 6364 6365 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6366 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6367 SCTP_ADD_IP_ADDRESS); 6368 if (ret != PF_PASS) 6369 return (ret); 6370 break; 6371 } 6372 case SCTP_DEL_IP_ADDRESS: { 6373 int ret; 6374 struct sctp_asconf_paramhdr ah; 6375 6376 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6377 NULL, NULL, pd->af)) 6378 return (PF_DROP); 6379 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6380 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6381 SCTP_DEL_IP_ADDRESS); 6382 if (ret != PF_PASS) 6383 return (ret); 6384 break; 6385 } 6386 default: 6387 break; 6388 } 6389 6390 off += roundup(ntohs(h.param_length), 4); 6391 } 6392 6393 return (PF_PASS); 6394 } 6395 int 6396 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6397 struct pfi_kkif *kif) 6398 { 6399 start += sizeof(struct sctp_init_chunk); 6400 len -= sizeof(struct sctp_init_chunk); 6401 6402 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6403 } 6404 6405 int 6406 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6407 struct pf_pdesc *pd, struct pfi_kkif *kif) 6408 { 6409 start += sizeof(struct sctp_asconf_chunk); 6410 len -= sizeof(struct sctp_asconf_chunk); 6411 6412 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6413 } 6414 6415 static int 6416 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6417 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6418 { 6419 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6420 u_int16_t icmpid = 0, *icmpsum; 6421 u_int8_t icmptype, icmpcode; 6422 int state_icmp = 0; 6423 struct pf_state_key_cmp key; 6424 6425 bzero(&key, sizeof(key)); 6426 switch (pd->proto) { 6427 #ifdef INET 6428 case IPPROTO_ICMP: 6429 icmptype = pd->hdr.icmp.icmp_type; 6430 icmpcode = pd->hdr.icmp.icmp_code; 6431 icmpid = pd->hdr.icmp.icmp_id; 6432 icmpsum = &pd->hdr.icmp.icmp_cksum; 6433 6434 if (icmptype == ICMP_UNREACH || 6435 icmptype == ICMP_SOURCEQUENCH || 6436 icmptype == ICMP_REDIRECT || 6437 icmptype == ICMP_TIMXCEED || 6438 icmptype == ICMP_PARAMPROB) 6439 state_icmp++; 6440 break; 6441 #endif /* INET */ 6442 #ifdef INET6 6443 case IPPROTO_ICMPV6: 6444 icmptype = pd->hdr.icmp6.icmp6_type; 6445 icmpcode = pd->hdr.icmp6.icmp6_code; 6446 icmpid = pd->hdr.icmp6.icmp6_id; 6447 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6448 6449 if (icmptype == ICMP6_DST_UNREACH || 6450 icmptype == ICMP6_PACKET_TOO_BIG || 6451 icmptype == ICMP6_TIME_EXCEEDED || 6452 icmptype == ICMP6_PARAM_PROB) 6453 state_icmp++; 6454 break; 6455 #endif /* INET6 */ 6456 } 6457 6458 if (!state_icmp) { 6459 /* 6460 * ICMP query/reply message not related to a TCP/UDP packet. 6461 * Search for an ICMP state. 6462 */ 6463 key.af = pd->af; 6464 key.proto = pd->proto; 6465 key.port[0] = key.port[1] = icmpid; 6466 if (pd->dir == PF_IN) { /* wire side, straight */ 6467 PF_ACPY(&key.addr[0], pd->src, key.af); 6468 PF_ACPY(&key.addr[1], pd->dst, key.af); 6469 } else { /* stack side, reverse */ 6470 PF_ACPY(&key.addr[1], pd->src, key.af); 6471 PF_ACPY(&key.addr[0], pd->dst, key.af); 6472 } 6473 6474 STATE_LOOKUP(kif, &key, *state, pd); 6475 6476 (*state)->expire = time_uptime; 6477 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6478 6479 /* translate source/destination address, if necessary */ 6480 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6481 struct pf_state_key *nk = (*state)->key[pd->didx]; 6482 6483 switch (pd->af) { 6484 #ifdef INET 6485 case AF_INET: 6486 if (PF_ANEQ(pd->src, 6487 &nk->addr[pd->sidx], AF_INET)) 6488 pf_change_a(&saddr->v4.s_addr, 6489 pd->ip_sum, 6490 nk->addr[pd->sidx].v4.s_addr, 0); 6491 6492 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6493 AF_INET)) 6494 pf_change_a(&daddr->v4.s_addr, 6495 pd->ip_sum, 6496 nk->addr[pd->didx].v4.s_addr, 0); 6497 6498 if (nk->port[0] != 6499 pd->hdr.icmp.icmp_id) { 6500 pd->hdr.icmp.icmp_cksum = 6501 pf_cksum_fixup( 6502 pd->hdr.icmp.icmp_cksum, icmpid, 6503 nk->port[pd->sidx], 0); 6504 pd->hdr.icmp.icmp_id = 6505 nk->port[pd->sidx]; 6506 } 6507 6508 m_copyback(m, off, ICMP_MINLEN, 6509 (caddr_t )&pd->hdr.icmp); 6510 break; 6511 #endif /* INET */ 6512 #ifdef INET6 6513 case AF_INET6: 6514 if (PF_ANEQ(pd->src, 6515 &nk->addr[pd->sidx], AF_INET6)) 6516 pf_change_a6(saddr, 6517 &pd->hdr.icmp6.icmp6_cksum, 6518 &nk->addr[pd->sidx], 0); 6519 6520 if (PF_ANEQ(pd->dst, 6521 &nk->addr[pd->didx], AF_INET6)) 6522 pf_change_a6(daddr, 6523 &pd->hdr.icmp6.icmp6_cksum, 6524 &nk->addr[pd->didx], 0); 6525 6526 m_copyback(m, off, sizeof(struct icmp6_hdr), 6527 (caddr_t )&pd->hdr.icmp6); 6528 break; 6529 #endif /* INET6 */ 6530 } 6531 } 6532 return (PF_PASS); 6533 6534 } else { 6535 /* 6536 * ICMP error message in response to a TCP/UDP packet. 6537 * Extract the inner TCP/UDP header and search for that state. 6538 */ 6539 6540 struct pf_pdesc pd2; 6541 bzero(&pd2, sizeof pd2); 6542 #ifdef INET 6543 struct ip h2; 6544 #endif /* INET */ 6545 #ifdef INET6 6546 struct ip6_hdr h2_6; 6547 int terminal = 0; 6548 #endif /* INET6 */ 6549 int ipoff2 = 0; 6550 int off2 = 0; 6551 6552 pd2.af = pd->af; 6553 /* Payload packet is from the opposite direction. */ 6554 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 6555 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 6556 switch (pd->af) { 6557 #ifdef INET 6558 case AF_INET: 6559 /* offset of h2 in mbuf chain */ 6560 ipoff2 = off + ICMP_MINLEN; 6561 6562 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 6563 NULL, reason, pd2.af)) { 6564 DPFPRINTF(PF_DEBUG_MISC, 6565 ("pf: ICMP error message too short " 6566 "(ip)\n")); 6567 return (PF_DROP); 6568 } 6569 /* 6570 * ICMP error messages don't refer to non-first 6571 * fragments 6572 */ 6573 if (h2.ip_off & htons(IP_OFFMASK)) { 6574 REASON_SET(reason, PFRES_FRAG); 6575 return (PF_DROP); 6576 } 6577 6578 /* offset of protocol header that follows h2 */ 6579 off2 = ipoff2 + (h2.ip_hl << 2); 6580 6581 pd2.proto = h2.ip_p; 6582 pd2.src = (struct pf_addr *)&h2.ip_src; 6583 pd2.dst = (struct pf_addr *)&h2.ip_dst; 6584 pd2.ip_sum = &h2.ip_sum; 6585 break; 6586 #endif /* INET */ 6587 #ifdef INET6 6588 case AF_INET6: 6589 ipoff2 = off + sizeof(struct icmp6_hdr); 6590 6591 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 6592 NULL, reason, pd2.af)) { 6593 DPFPRINTF(PF_DEBUG_MISC, 6594 ("pf: ICMP error message too short " 6595 "(ip6)\n")); 6596 return (PF_DROP); 6597 } 6598 pd2.proto = h2_6.ip6_nxt; 6599 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 6600 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 6601 pd2.ip_sum = NULL; 6602 off2 = ipoff2 + sizeof(h2_6); 6603 do { 6604 switch (pd2.proto) { 6605 case IPPROTO_FRAGMENT: 6606 /* 6607 * ICMPv6 error messages for 6608 * non-first fragments 6609 */ 6610 REASON_SET(reason, PFRES_FRAG); 6611 return (PF_DROP); 6612 case IPPROTO_AH: 6613 case IPPROTO_HOPOPTS: 6614 case IPPROTO_ROUTING: 6615 case IPPROTO_DSTOPTS: { 6616 /* get next header and header length */ 6617 struct ip6_ext opt6; 6618 6619 if (!pf_pull_hdr(m, off2, &opt6, 6620 sizeof(opt6), NULL, reason, 6621 pd2.af)) { 6622 DPFPRINTF(PF_DEBUG_MISC, 6623 ("pf: ICMPv6 short opt\n")); 6624 return (PF_DROP); 6625 } 6626 if (pd2.proto == IPPROTO_AH) 6627 off2 += (opt6.ip6e_len + 2) * 4; 6628 else 6629 off2 += (opt6.ip6e_len + 1) * 8; 6630 pd2.proto = opt6.ip6e_nxt; 6631 /* goto the next header */ 6632 break; 6633 } 6634 default: 6635 terminal++; 6636 break; 6637 } 6638 } while (!terminal); 6639 break; 6640 #endif /* INET6 */ 6641 } 6642 6643 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 6644 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6645 printf("pf: BAD ICMP %d:%d outer dst: ", 6646 icmptype, icmpcode); 6647 pf_print_host(pd->src, 0, pd->af); 6648 printf(" -> "); 6649 pf_print_host(pd->dst, 0, pd->af); 6650 printf(" inner src: "); 6651 pf_print_host(pd2.src, 0, pd2.af); 6652 printf(" -> "); 6653 pf_print_host(pd2.dst, 0, pd2.af); 6654 printf("\n"); 6655 } 6656 REASON_SET(reason, PFRES_BADSTATE); 6657 return (PF_DROP); 6658 } 6659 6660 switch (pd2.proto) { 6661 case IPPROTO_TCP: { 6662 struct tcphdr th; 6663 u_int32_t seq; 6664 struct pf_state_peer *src, *dst; 6665 u_int8_t dws; 6666 int copyback = 0; 6667 6668 /* 6669 * Only the first 8 bytes of the TCP header can be 6670 * expected. Don't access any TCP header fields after 6671 * th_seq, an ackskew test is not possible. 6672 */ 6673 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 6674 pd2.af)) { 6675 DPFPRINTF(PF_DEBUG_MISC, 6676 ("pf: ICMP error message too short " 6677 "(tcp)\n")); 6678 return (PF_DROP); 6679 } 6680 6681 key.af = pd2.af; 6682 key.proto = IPPROTO_TCP; 6683 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6684 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6685 key.port[pd2.sidx] = th.th_sport; 6686 key.port[pd2.didx] = th.th_dport; 6687 6688 STATE_LOOKUP(kif, &key, *state, pd); 6689 6690 if (pd->dir == (*state)->direction) { 6691 src = &(*state)->dst; 6692 dst = &(*state)->src; 6693 } else { 6694 src = &(*state)->src; 6695 dst = &(*state)->dst; 6696 } 6697 6698 if (src->wscale && dst->wscale) 6699 dws = dst->wscale & PF_WSCALE_MASK; 6700 else 6701 dws = 0; 6702 6703 /* Demodulate sequence number */ 6704 seq = ntohl(th.th_seq) - src->seqdiff; 6705 if (src->seqdiff) { 6706 pf_change_a(&th.th_seq, icmpsum, 6707 htonl(seq), 0); 6708 copyback = 1; 6709 } 6710 6711 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 6712 (!SEQ_GEQ(src->seqhi, seq) || 6713 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 6714 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6715 printf("pf: BAD ICMP %d:%d ", 6716 icmptype, icmpcode); 6717 pf_print_host(pd->src, 0, pd->af); 6718 printf(" -> "); 6719 pf_print_host(pd->dst, 0, pd->af); 6720 printf(" state: "); 6721 pf_print_state(*state); 6722 printf(" seq=%u\n", seq); 6723 } 6724 REASON_SET(reason, PFRES_BADSTATE); 6725 return (PF_DROP); 6726 } else { 6727 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6728 printf("pf: OK ICMP %d:%d ", 6729 icmptype, icmpcode); 6730 pf_print_host(pd->src, 0, pd->af); 6731 printf(" -> "); 6732 pf_print_host(pd->dst, 0, pd->af); 6733 printf(" state: "); 6734 pf_print_state(*state); 6735 printf(" seq=%u\n", seq); 6736 } 6737 } 6738 6739 /* translate source/destination address, if necessary */ 6740 if ((*state)->key[PF_SK_WIRE] != 6741 (*state)->key[PF_SK_STACK]) { 6742 struct pf_state_key *nk = 6743 (*state)->key[pd->didx]; 6744 6745 if (PF_ANEQ(pd2.src, 6746 &nk->addr[pd2.sidx], pd2.af) || 6747 nk->port[pd2.sidx] != th.th_sport) 6748 pf_change_icmp(pd2.src, &th.th_sport, 6749 daddr, &nk->addr[pd2.sidx], 6750 nk->port[pd2.sidx], NULL, 6751 pd2.ip_sum, icmpsum, 6752 pd->ip_sum, 0, pd2.af); 6753 6754 if (PF_ANEQ(pd2.dst, 6755 &nk->addr[pd2.didx], pd2.af) || 6756 nk->port[pd2.didx] != th.th_dport) 6757 pf_change_icmp(pd2.dst, &th.th_dport, 6758 saddr, &nk->addr[pd2.didx], 6759 nk->port[pd2.didx], NULL, 6760 pd2.ip_sum, icmpsum, 6761 pd->ip_sum, 0, pd2.af); 6762 copyback = 1; 6763 } 6764 6765 if (copyback) { 6766 switch (pd2.af) { 6767 #ifdef INET 6768 case AF_INET: 6769 m_copyback(m, off, ICMP_MINLEN, 6770 (caddr_t )&pd->hdr.icmp); 6771 m_copyback(m, ipoff2, sizeof(h2), 6772 (caddr_t )&h2); 6773 break; 6774 #endif /* INET */ 6775 #ifdef INET6 6776 case AF_INET6: 6777 m_copyback(m, off, 6778 sizeof(struct icmp6_hdr), 6779 (caddr_t )&pd->hdr.icmp6); 6780 m_copyback(m, ipoff2, sizeof(h2_6), 6781 (caddr_t )&h2_6); 6782 break; 6783 #endif /* INET6 */ 6784 } 6785 m_copyback(m, off2, 8, (caddr_t)&th); 6786 } 6787 6788 return (PF_PASS); 6789 break; 6790 } 6791 case IPPROTO_UDP: { 6792 struct udphdr uh; 6793 6794 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 6795 NULL, reason, pd2.af)) { 6796 DPFPRINTF(PF_DEBUG_MISC, 6797 ("pf: ICMP error message too short " 6798 "(udp)\n")); 6799 return (PF_DROP); 6800 } 6801 6802 key.af = pd2.af; 6803 key.proto = IPPROTO_UDP; 6804 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6805 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6806 key.port[pd2.sidx] = uh.uh_sport; 6807 key.port[pd2.didx] = uh.uh_dport; 6808 6809 STATE_LOOKUP(kif, &key, *state, pd); 6810 6811 /* translate source/destination address, if necessary */ 6812 if ((*state)->key[PF_SK_WIRE] != 6813 (*state)->key[PF_SK_STACK]) { 6814 struct pf_state_key *nk = 6815 (*state)->key[pd->didx]; 6816 6817 if (PF_ANEQ(pd2.src, 6818 &nk->addr[pd2.sidx], pd2.af) || 6819 nk->port[pd2.sidx] != uh.uh_sport) 6820 pf_change_icmp(pd2.src, &uh.uh_sport, 6821 daddr, &nk->addr[pd2.sidx], 6822 nk->port[pd2.sidx], &uh.uh_sum, 6823 pd2.ip_sum, icmpsum, 6824 pd->ip_sum, 1, pd2.af); 6825 6826 if (PF_ANEQ(pd2.dst, 6827 &nk->addr[pd2.didx], pd2.af) || 6828 nk->port[pd2.didx] != uh.uh_dport) 6829 pf_change_icmp(pd2.dst, &uh.uh_dport, 6830 saddr, &nk->addr[pd2.didx], 6831 nk->port[pd2.didx], &uh.uh_sum, 6832 pd2.ip_sum, icmpsum, 6833 pd->ip_sum, 1, pd2.af); 6834 6835 switch (pd2.af) { 6836 #ifdef INET 6837 case AF_INET: 6838 m_copyback(m, off, ICMP_MINLEN, 6839 (caddr_t )&pd->hdr.icmp); 6840 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6841 break; 6842 #endif /* INET */ 6843 #ifdef INET6 6844 case AF_INET6: 6845 m_copyback(m, off, 6846 sizeof(struct icmp6_hdr), 6847 (caddr_t )&pd->hdr.icmp6); 6848 m_copyback(m, ipoff2, sizeof(h2_6), 6849 (caddr_t )&h2_6); 6850 break; 6851 #endif /* INET6 */ 6852 } 6853 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 6854 } 6855 return (PF_PASS); 6856 break; 6857 } 6858 #ifdef INET 6859 case IPPROTO_ICMP: { 6860 struct icmp iih; 6861 6862 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6863 NULL, reason, pd2.af)) { 6864 DPFPRINTF(PF_DEBUG_MISC, 6865 ("pf: ICMP error message too short i" 6866 "(icmp)\n")); 6867 return (PF_DROP); 6868 } 6869 6870 key.af = pd2.af; 6871 key.proto = IPPROTO_ICMP; 6872 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6873 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6874 key.port[0] = key.port[1] = iih.icmp_id; 6875 6876 STATE_LOOKUP(kif, &key, *state, pd); 6877 6878 /* translate source/destination address, if necessary */ 6879 if ((*state)->key[PF_SK_WIRE] != 6880 (*state)->key[PF_SK_STACK]) { 6881 struct pf_state_key *nk = 6882 (*state)->key[pd->didx]; 6883 6884 if (PF_ANEQ(pd2.src, 6885 &nk->addr[pd2.sidx], pd2.af) || 6886 nk->port[pd2.sidx] != iih.icmp_id) 6887 pf_change_icmp(pd2.src, &iih.icmp_id, 6888 daddr, &nk->addr[pd2.sidx], 6889 nk->port[pd2.sidx], NULL, 6890 pd2.ip_sum, icmpsum, 6891 pd->ip_sum, 0, AF_INET); 6892 6893 if (PF_ANEQ(pd2.dst, 6894 &nk->addr[pd2.didx], pd2.af) || 6895 nk->port[pd2.didx] != iih.icmp_id) 6896 pf_change_icmp(pd2.dst, &iih.icmp_id, 6897 saddr, &nk->addr[pd2.didx], 6898 nk->port[pd2.didx], NULL, 6899 pd2.ip_sum, icmpsum, 6900 pd->ip_sum, 0, AF_INET); 6901 6902 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6903 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6904 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6905 } 6906 return (PF_PASS); 6907 break; 6908 } 6909 #endif /* INET */ 6910 #ifdef INET6 6911 case IPPROTO_ICMPV6: { 6912 struct icmp6_hdr iih; 6913 6914 if (!pf_pull_hdr(m, off2, &iih, 6915 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6916 DPFPRINTF(PF_DEBUG_MISC, 6917 ("pf: ICMP error message too short " 6918 "(icmp6)\n")); 6919 return (PF_DROP); 6920 } 6921 6922 key.af = pd2.af; 6923 key.proto = IPPROTO_ICMPV6; 6924 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6925 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6926 key.port[0] = key.port[1] = iih.icmp6_id; 6927 6928 STATE_LOOKUP(kif, &key, *state, pd); 6929 6930 /* translate source/destination address, if necessary */ 6931 if ((*state)->key[PF_SK_WIRE] != 6932 (*state)->key[PF_SK_STACK]) { 6933 struct pf_state_key *nk = 6934 (*state)->key[pd->didx]; 6935 6936 if (PF_ANEQ(pd2.src, 6937 &nk->addr[pd2.sidx], pd2.af) || 6938 nk->port[pd2.sidx] != iih.icmp6_id) 6939 pf_change_icmp(pd2.src, &iih.icmp6_id, 6940 daddr, &nk->addr[pd2.sidx], 6941 nk->port[pd2.sidx], NULL, 6942 pd2.ip_sum, icmpsum, 6943 pd->ip_sum, 0, AF_INET6); 6944 6945 if (PF_ANEQ(pd2.dst, 6946 &nk->addr[pd2.didx], pd2.af) || 6947 nk->port[pd2.didx] != iih.icmp6_id) 6948 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6949 saddr, &nk->addr[pd2.didx], 6950 nk->port[pd2.didx], NULL, 6951 pd2.ip_sum, icmpsum, 6952 pd->ip_sum, 0, AF_INET6); 6953 6954 m_copyback(m, off, sizeof(struct icmp6_hdr), 6955 (caddr_t)&pd->hdr.icmp6); 6956 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6957 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6958 (caddr_t)&iih); 6959 } 6960 return (PF_PASS); 6961 break; 6962 } 6963 #endif /* INET6 */ 6964 default: { 6965 key.af = pd2.af; 6966 key.proto = pd2.proto; 6967 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6968 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6969 key.port[0] = key.port[1] = 0; 6970 6971 STATE_LOOKUP(kif, &key, *state, pd); 6972 6973 /* translate source/destination address, if necessary */ 6974 if ((*state)->key[PF_SK_WIRE] != 6975 (*state)->key[PF_SK_STACK]) { 6976 struct pf_state_key *nk = 6977 (*state)->key[pd->didx]; 6978 6979 if (PF_ANEQ(pd2.src, 6980 &nk->addr[pd2.sidx], pd2.af)) 6981 pf_change_icmp(pd2.src, NULL, daddr, 6982 &nk->addr[pd2.sidx], 0, NULL, 6983 pd2.ip_sum, icmpsum, 6984 pd->ip_sum, 0, pd2.af); 6985 6986 if (PF_ANEQ(pd2.dst, 6987 &nk->addr[pd2.didx], pd2.af)) 6988 pf_change_icmp(pd2.dst, NULL, saddr, 6989 &nk->addr[pd2.didx], 0, NULL, 6990 pd2.ip_sum, icmpsum, 6991 pd->ip_sum, 0, pd2.af); 6992 6993 switch (pd2.af) { 6994 #ifdef INET 6995 case AF_INET: 6996 m_copyback(m, off, ICMP_MINLEN, 6997 (caddr_t)&pd->hdr.icmp); 6998 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6999 break; 7000 #endif /* INET */ 7001 #ifdef INET6 7002 case AF_INET6: 7003 m_copyback(m, off, 7004 sizeof(struct icmp6_hdr), 7005 (caddr_t )&pd->hdr.icmp6); 7006 m_copyback(m, ipoff2, sizeof(h2_6), 7007 (caddr_t )&h2_6); 7008 break; 7009 #endif /* INET6 */ 7010 } 7011 } 7012 return (PF_PASS); 7013 break; 7014 } 7015 } 7016 } 7017 } 7018 7019 static int 7020 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7021 struct mbuf *m, struct pf_pdesc *pd) 7022 { 7023 struct pf_state_peer *src, *dst; 7024 struct pf_state_key_cmp key; 7025 uint8_t psrc, pdst; 7026 7027 bzero(&key, sizeof(key)); 7028 key.af = pd->af; 7029 key.proto = pd->proto; 7030 if (pd->dir == PF_IN) { 7031 PF_ACPY(&key.addr[0], pd->src, key.af); 7032 PF_ACPY(&key.addr[1], pd->dst, key.af); 7033 key.port[0] = key.port[1] = 0; 7034 } else { 7035 PF_ACPY(&key.addr[1], pd->src, key.af); 7036 PF_ACPY(&key.addr[0], pd->dst, key.af); 7037 key.port[1] = key.port[0] = 0; 7038 } 7039 7040 STATE_LOOKUP(kif, &key, *state, pd); 7041 7042 if (pd->dir == (*state)->direction) { 7043 src = &(*state)->src; 7044 dst = &(*state)->dst; 7045 psrc = PF_PEER_SRC; 7046 pdst = PF_PEER_DST; 7047 } else { 7048 src = &(*state)->dst; 7049 dst = &(*state)->src; 7050 psrc = PF_PEER_DST; 7051 pdst = PF_PEER_SRC; 7052 } 7053 7054 /* update states */ 7055 if (src->state < PFOTHERS_SINGLE) 7056 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7057 if (dst->state == PFOTHERS_SINGLE) 7058 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7059 7060 /* update expire time */ 7061 (*state)->expire = time_uptime; 7062 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7063 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7064 else 7065 (*state)->timeout = PFTM_OTHER_SINGLE; 7066 7067 /* translate source/destination address, if necessary */ 7068 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7069 struct pf_state_key *nk = (*state)->key[pd->didx]; 7070 7071 KASSERT(nk, ("%s: nk is null", __func__)); 7072 KASSERT(pd, ("%s: pd is null", __func__)); 7073 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7074 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7075 switch (pd->af) { 7076 #ifdef INET 7077 case AF_INET: 7078 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7079 pf_change_a(&pd->src->v4.s_addr, 7080 pd->ip_sum, 7081 nk->addr[pd->sidx].v4.s_addr, 7082 0); 7083 7084 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7085 pf_change_a(&pd->dst->v4.s_addr, 7086 pd->ip_sum, 7087 nk->addr[pd->didx].v4.s_addr, 7088 0); 7089 7090 break; 7091 #endif /* INET */ 7092 #ifdef INET6 7093 case AF_INET6: 7094 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7095 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7096 7097 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7098 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7099 #endif /* INET6 */ 7100 } 7101 } 7102 return (PF_PASS); 7103 } 7104 7105 /* 7106 * ipoff and off are measured from the start of the mbuf chain. 7107 * h must be at "ipoff" on the mbuf chain. 7108 */ 7109 void * 7110 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 7111 u_short *actionp, u_short *reasonp, sa_family_t af) 7112 { 7113 switch (af) { 7114 #ifdef INET 7115 case AF_INET: { 7116 struct ip *h = mtod(m, struct ip *); 7117 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7118 7119 if (fragoff) { 7120 if (fragoff >= len) 7121 ACTION_SET(actionp, PF_PASS); 7122 else { 7123 ACTION_SET(actionp, PF_DROP); 7124 REASON_SET(reasonp, PFRES_FRAG); 7125 } 7126 return (NULL); 7127 } 7128 if (m->m_pkthdr.len < off + len || 7129 ntohs(h->ip_len) < off + len) { 7130 ACTION_SET(actionp, PF_DROP); 7131 REASON_SET(reasonp, PFRES_SHORT); 7132 return (NULL); 7133 } 7134 break; 7135 } 7136 #endif /* INET */ 7137 #ifdef INET6 7138 case AF_INET6: { 7139 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7140 7141 if (m->m_pkthdr.len < off + len || 7142 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7143 (unsigned)(off + len)) { 7144 ACTION_SET(actionp, PF_DROP); 7145 REASON_SET(reasonp, PFRES_SHORT); 7146 return (NULL); 7147 } 7148 break; 7149 } 7150 #endif /* INET6 */ 7151 } 7152 m_copydata(m, off, len, p); 7153 return (p); 7154 } 7155 7156 int 7157 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7158 int rtableid) 7159 { 7160 struct ifnet *ifp; 7161 7162 /* 7163 * Skip check for addresses with embedded interface scope, 7164 * as they would always match anyway. 7165 */ 7166 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7167 return (1); 7168 7169 if (af != AF_INET && af != AF_INET6) 7170 return (0); 7171 7172 if (kif == V_pfi_all) 7173 return (1); 7174 7175 /* Skip checks for ipsec interfaces */ 7176 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7177 return (1); 7178 7179 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7180 7181 switch (af) { 7182 #ifdef INET6 7183 case AF_INET6: 7184 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7185 ifp)); 7186 #endif 7187 #ifdef INET 7188 case AF_INET: 7189 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7190 ifp)); 7191 #endif 7192 } 7193 7194 return (0); 7195 } 7196 7197 #ifdef INET 7198 static void 7199 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7200 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7201 { 7202 struct mbuf *m0, *m1, *md; 7203 struct sockaddr_in dst; 7204 struct ip *ip; 7205 struct pfi_kkif *nkif = NULL; 7206 struct ifnet *ifp = NULL; 7207 struct pf_addr naddr; 7208 struct pf_ksrc_node *sn = NULL; 7209 int error = 0; 7210 uint16_t ip_len, ip_off; 7211 int r_rt, r_dir; 7212 7213 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7214 7215 if (s) { 7216 r_rt = s->rt; 7217 r_dir = s->direction; 7218 } else { 7219 r_rt = r->rt; 7220 r_dir = r->direction; 7221 } 7222 7223 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7224 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7225 __func__)); 7226 7227 if ((pd->pf_mtag == NULL && 7228 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7229 pd->pf_mtag->routed++ > 3) { 7230 m0 = *m; 7231 *m = NULL; 7232 goto bad_locked; 7233 } 7234 7235 if (r_rt == PF_DUPTO) { 7236 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7237 if (s == NULL) { 7238 ifp = r->rpool.cur->kif ? 7239 r->rpool.cur->kif->pfik_ifp : NULL; 7240 } else { 7241 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7242 /* If pfsync'd */ 7243 if (ifp == NULL && r->rpool.cur != NULL) 7244 ifp = r->rpool.cur->kif ? 7245 r->rpool.cur->kif->pfik_ifp : NULL; 7246 PF_STATE_UNLOCK(s); 7247 } 7248 if (ifp == oifp) { 7249 /* When the 2nd interface is not skipped */ 7250 return; 7251 } else { 7252 m0 = *m; 7253 *m = NULL; 7254 goto bad; 7255 } 7256 } else { 7257 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7258 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7259 if (s) 7260 PF_STATE_UNLOCK(s); 7261 return; 7262 } 7263 } 7264 } else { 7265 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7266 pf_dummynet(pd, s, r, m); 7267 if (s) 7268 PF_STATE_UNLOCK(s); 7269 return; 7270 } 7271 m0 = *m; 7272 } 7273 7274 ip = mtod(m0, struct ip *); 7275 7276 bzero(&dst, sizeof(dst)); 7277 dst.sin_family = AF_INET; 7278 dst.sin_len = sizeof(dst); 7279 dst.sin_addr = ip->ip_dst; 7280 7281 bzero(&naddr, sizeof(naddr)); 7282 7283 if (s == NULL) { 7284 if (TAILQ_EMPTY(&r->rpool.list)) { 7285 DPFPRINTF(PF_DEBUG_URGENT, 7286 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7287 goto bad_locked; 7288 } 7289 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7290 &naddr, &nkif, NULL, &sn); 7291 if (!PF_AZERO(&naddr, AF_INET)) 7292 dst.sin_addr.s_addr = naddr.v4.s_addr; 7293 ifp = nkif ? nkif->pfik_ifp : NULL; 7294 } else { 7295 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7296 dst.sin_addr.s_addr = 7297 s->rt_addr.v4.s_addr; 7298 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7299 /* If pfsync'd */ 7300 if (ifp == NULL && r->rpool.cur != NULL) { 7301 ifp = r->rpool.cur->kif ? 7302 r->rpool.cur->kif->pfik_ifp : NULL; 7303 } 7304 PF_STATE_UNLOCK(s); 7305 } 7306 7307 if (ifp == NULL) 7308 goto bad; 7309 7310 if (pd->dir == PF_IN) { 7311 if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7312 goto bad; 7313 else if (m0 == NULL) 7314 goto done; 7315 if (m0->m_len < sizeof(struct ip)) { 7316 DPFPRINTF(PF_DEBUG_URGENT, 7317 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7318 goto bad; 7319 } 7320 ip = mtod(m0, struct ip *); 7321 } 7322 7323 if (ifp->if_flags & IFF_LOOPBACK) 7324 m0->m_flags |= M_SKIP_FIREWALL; 7325 7326 ip_len = ntohs(ip->ip_len); 7327 ip_off = ntohs(ip->ip_off); 7328 7329 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7330 m0->m_pkthdr.csum_flags |= CSUM_IP; 7331 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7332 in_delayed_cksum(m0); 7333 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7334 } 7335 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7336 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7337 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7338 } 7339 7340 /* 7341 * If small enough for interface, or the interface will take 7342 * care of the fragmentation for us, we can just send directly. 7343 */ 7344 if (ip_len <= ifp->if_mtu || 7345 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7346 ip->ip_sum = 0; 7347 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7348 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7349 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7350 } 7351 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7352 7353 md = m0; 7354 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7355 if (md != NULL) 7356 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7357 goto done; 7358 } 7359 7360 /* Balk when DF bit is set or the interface didn't support TSO. */ 7361 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7362 error = EMSGSIZE; 7363 KMOD_IPSTAT_INC(ips_cantfrag); 7364 if (r_rt != PF_DUPTO) { 7365 if (s && pd->nat_rule != NULL) 7366 PACKET_UNDO_NAT(m0, pd, 7367 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7368 s); 7369 7370 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7371 ifp->if_mtu); 7372 goto done; 7373 } else 7374 goto bad; 7375 } 7376 7377 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7378 if (error) 7379 goto bad; 7380 7381 for (; m0; m0 = m1) { 7382 m1 = m0->m_nextpkt; 7383 m0->m_nextpkt = NULL; 7384 if (error == 0) { 7385 m_clrprotoflags(m0); 7386 md = m0; 7387 error = pf_dummynet_route(pd, s, r, ifp, 7388 sintosa(&dst), &md); 7389 if (md != NULL) 7390 error = (*ifp->if_output)(ifp, md, 7391 sintosa(&dst), NULL); 7392 } else 7393 m_freem(m0); 7394 } 7395 7396 if (error == 0) 7397 KMOD_IPSTAT_INC(ips_fragmented); 7398 7399 done: 7400 if (r_rt != PF_DUPTO) 7401 *m = NULL; 7402 return; 7403 7404 bad_locked: 7405 if (s) 7406 PF_STATE_UNLOCK(s); 7407 bad: 7408 m_freem(m0); 7409 goto done; 7410 } 7411 #endif /* INET */ 7412 7413 #ifdef INET6 7414 static void 7415 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7416 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7417 { 7418 struct mbuf *m0, *md; 7419 struct sockaddr_in6 dst; 7420 struct ip6_hdr *ip6; 7421 struct pfi_kkif *nkif = NULL; 7422 struct ifnet *ifp = NULL; 7423 struct pf_addr naddr; 7424 struct pf_ksrc_node *sn = NULL; 7425 int r_rt, r_dir; 7426 7427 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7428 7429 if (s) { 7430 r_rt = s->rt; 7431 r_dir = s->direction; 7432 } else { 7433 r_rt = r->rt; 7434 r_dir = r->direction; 7435 } 7436 7437 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7438 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7439 __func__)); 7440 7441 if ((pd->pf_mtag == NULL && 7442 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7443 pd->pf_mtag->routed++ > 3) { 7444 m0 = *m; 7445 *m = NULL; 7446 goto bad_locked; 7447 } 7448 7449 if (r_rt == PF_DUPTO) { 7450 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7451 if (s == NULL) { 7452 ifp = r->rpool.cur->kif ? 7453 r->rpool.cur->kif->pfik_ifp : NULL; 7454 } else { 7455 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7456 /* If pfsync'd */ 7457 if (ifp == NULL && r->rpool.cur != NULL) 7458 ifp = r->rpool.cur->kif ? 7459 r->rpool.cur->kif->pfik_ifp : NULL; 7460 PF_STATE_UNLOCK(s); 7461 } 7462 if (ifp == oifp) { 7463 /* When the 2nd interface is not skipped */ 7464 return; 7465 } else { 7466 m0 = *m; 7467 *m = NULL; 7468 goto bad; 7469 } 7470 } else { 7471 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7472 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7473 if (s) 7474 PF_STATE_UNLOCK(s); 7475 return; 7476 } 7477 } 7478 } else { 7479 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7480 pf_dummynet(pd, s, r, m); 7481 if (s) 7482 PF_STATE_UNLOCK(s); 7483 return; 7484 } 7485 m0 = *m; 7486 } 7487 7488 ip6 = mtod(m0, struct ip6_hdr *); 7489 7490 bzero(&dst, sizeof(dst)); 7491 dst.sin6_family = AF_INET6; 7492 dst.sin6_len = sizeof(dst); 7493 dst.sin6_addr = ip6->ip6_dst; 7494 7495 bzero(&naddr, sizeof(naddr)); 7496 7497 if (s == NULL) { 7498 if (TAILQ_EMPTY(&r->rpool.list)) { 7499 DPFPRINTF(PF_DEBUG_URGENT, 7500 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7501 goto bad_locked; 7502 } 7503 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7504 &naddr, &nkif, NULL, &sn); 7505 if (!PF_AZERO(&naddr, AF_INET6)) 7506 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7507 &naddr, AF_INET6); 7508 ifp = nkif ? nkif->pfik_ifp : NULL; 7509 } else { 7510 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7511 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7512 &s->rt_addr, AF_INET6); 7513 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7514 /* If pfsync'd */ 7515 if (ifp == NULL && r->rpool.cur != NULL) 7516 ifp = r->rpool.cur->kif ? 7517 r->rpool.cur->kif->pfik_ifp : NULL; 7518 } 7519 7520 if (s) 7521 PF_STATE_UNLOCK(s); 7522 7523 if (ifp == NULL) 7524 goto bad; 7525 7526 if (pd->dir == PF_IN) { 7527 if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7528 goto bad; 7529 else if (m0 == NULL) 7530 goto done; 7531 if (m0->m_len < sizeof(struct ip6_hdr)) { 7532 DPFPRINTF(PF_DEBUG_URGENT, 7533 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 7534 __func__)); 7535 goto bad; 7536 } 7537 ip6 = mtod(m0, struct ip6_hdr *); 7538 } 7539 7540 if (ifp->if_flags & IFF_LOOPBACK) 7541 m0->m_flags |= M_SKIP_FIREWALL; 7542 7543 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 7544 ~ifp->if_hwassist) { 7545 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 7546 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 7547 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 7548 } 7549 7550 /* 7551 * If the packet is too large for the outgoing interface, 7552 * send back an icmp6 error. 7553 */ 7554 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 7555 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 7556 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 7557 md = m0; 7558 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7559 if (md != NULL) 7560 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 7561 } 7562 else { 7563 in6_ifstat_inc(ifp, ifs6_in_toobig); 7564 if (r_rt != PF_DUPTO) { 7565 if (s && pd->nat_rule != NULL) 7566 PACKET_UNDO_NAT(m0, pd, 7567 ((caddr_t)ip6 - m0->m_data) + 7568 sizeof(struct ip6_hdr), s); 7569 7570 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 7571 } else 7572 goto bad; 7573 } 7574 7575 done: 7576 if (r_rt != PF_DUPTO) 7577 *m = NULL; 7578 return; 7579 7580 bad_locked: 7581 if (s) 7582 PF_STATE_UNLOCK(s); 7583 bad: 7584 m_freem(m0); 7585 goto done; 7586 } 7587 #endif /* INET6 */ 7588 7589 /* 7590 * FreeBSD supports cksum offloads for the following drivers. 7591 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 7592 * 7593 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 7594 * network driver performed cksum including pseudo header, need to verify 7595 * csum_data 7596 * CSUM_DATA_VALID : 7597 * network driver performed cksum, needs to additional pseudo header 7598 * cksum computation with partial csum_data(i.e. lack of H/W support for 7599 * pseudo header, for instance sk(4) and possibly gem(4)) 7600 * 7601 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 7602 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 7603 * TCP/UDP layer. 7604 * Also, set csum_data to 0xffff to force cksum validation. 7605 */ 7606 static int 7607 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 7608 { 7609 u_int16_t sum = 0; 7610 int hw_assist = 0; 7611 struct ip *ip; 7612 7613 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 7614 return (1); 7615 if (m->m_pkthdr.len < off + len) 7616 return (1); 7617 7618 switch (p) { 7619 case IPPROTO_TCP: 7620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7622 sum = m->m_pkthdr.csum_data; 7623 } else { 7624 ip = mtod(m, struct ip *); 7625 sum = in_pseudo(ip->ip_src.s_addr, 7626 ip->ip_dst.s_addr, htonl((u_short)len + 7627 m->m_pkthdr.csum_data + IPPROTO_TCP)); 7628 } 7629 sum ^= 0xffff; 7630 ++hw_assist; 7631 } 7632 break; 7633 case IPPROTO_UDP: 7634 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7635 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7636 sum = m->m_pkthdr.csum_data; 7637 } else { 7638 ip = mtod(m, struct ip *); 7639 sum = in_pseudo(ip->ip_src.s_addr, 7640 ip->ip_dst.s_addr, htonl((u_short)len + 7641 m->m_pkthdr.csum_data + IPPROTO_UDP)); 7642 } 7643 sum ^= 0xffff; 7644 ++hw_assist; 7645 } 7646 break; 7647 case IPPROTO_ICMP: 7648 #ifdef INET6 7649 case IPPROTO_ICMPV6: 7650 #endif /* INET6 */ 7651 break; 7652 default: 7653 return (1); 7654 } 7655 7656 if (!hw_assist) { 7657 switch (af) { 7658 case AF_INET: 7659 if (p == IPPROTO_ICMP) { 7660 if (m->m_len < off) 7661 return (1); 7662 m->m_data += off; 7663 m->m_len -= off; 7664 sum = in_cksum(m, len); 7665 m->m_data -= off; 7666 m->m_len += off; 7667 } else { 7668 if (m->m_len < sizeof(struct ip)) 7669 return (1); 7670 sum = in4_cksum(m, p, off, len); 7671 } 7672 break; 7673 #ifdef INET6 7674 case AF_INET6: 7675 if (m->m_len < sizeof(struct ip6_hdr)) 7676 return (1); 7677 sum = in6_cksum(m, p, off, len); 7678 break; 7679 #endif /* INET6 */ 7680 default: 7681 return (1); 7682 } 7683 } 7684 if (sum) { 7685 switch (p) { 7686 case IPPROTO_TCP: 7687 { 7688 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 7689 break; 7690 } 7691 case IPPROTO_UDP: 7692 { 7693 KMOD_UDPSTAT_INC(udps_badsum); 7694 break; 7695 } 7696 #ifdef INET 7697 case IPPROTO_ICMP: 7698 { 7699 KMOD_ICMPSTAT_INC(icps_checksum); 7700 break; 7701 } 7702 #endif 7703 #ifdef INET6 7704 case IPPROTO_ICMPV6: 7705 { 7706 KMOD_ICMP6STAT_INC(icp6s_checksum); 7707 break; 7708 } 7709 #endif /* INET6 */ 7710 } 7711 return (1); 7712 } else { 7713 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 7714 m->m_pkthdr.csum_flags |= 7715 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 7716 m->m_pkthdr.csum_data = 0xffff; 7717 } 7718 } 7719 return (0); 7720 } 7721 7722 static bool 7723 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 7724 const struct pf_kstate *s, struct ip_fw_args *dnflow) 7725 { 7726 int dndir = r->direction; 7727 7728 if (s && dndir == PF_INOUT) { 7729 dndir = s->direction; 7730 } else if (dndir == PF_INOUT) { 7731 /* Assume primary direction. Happens when we've set dnpipe in 7732 * the ethernet level code. */ 7733 dndir = pd->dir; 7734 } 7735 7736 memset(dnflow, 0, sizeof(*dnflow)); 7737 7738 if (pd->dport != NULL) 7739 dnflow->f_id.dst_port = ntohs(*pd->dport); 7740 if (pd->sport != NULL) 7741 dnflow->f_id.src_port = ntohs(*pd->sport); 7742 7743 if (pd->dir == PF_IN) 7744 dnflow->flags |= IPFW_ARGS_IN; 7745 else 7746 dnflow->flags |= IPFW_ARGS_OUT; 7747 7748 if (pd->dir != dndir && pd->act.dnrpipe) { 7749 dnflow->rule.info = pd->act.dnrpipe; 7750 } 7751 else if (pd->dir == dndir && pd->act.dnpipe) { 7752 dnflow->rule.info = pd->act.dnpipe; 7753 } 7754 else { 7755 return (false); 7756 } 7757 7758 dnflow->rule.info |= IPFW_IS_DUMMYNET; 7759 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 7760 dnflow->rule.info |= IPFW_IS_PIPE; 7761 7762 dnflow->f_id.proto = pd->proto; 7763 dnflow->f_id.extra = dnflow->rule.info; 7764 switch (pd->af) { 7765 case AF_INET: 7766 dnflow->f_id.addr_type = 4; 7767 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 7768 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 7769 break; 7770 case AF_INET6: 7771 dnflow->flags |= IPFW_ARGS_IP6; 7772 dnflow->f_id.addr_type = 6; 7773 dnflow->f_id.src_ip6 = pd->src->v6; 7774 dnflow->f_id.dst_ip6 = pd->dst->v6; 7775 break; 7776 default: 7777 panic("Invalid AF"); 7778 break; 7779 } 7780 7781 return (true); 7782 } 7783 7784 int 7785 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7786 struct inpcb *inp) 7787 { 7788 struct pfi_kkif *kif; 7789 struct mbuf *m = *m0; 7790 7791 M_ASSERTPKTHDR(m); 7792 MPASS(ifp->if_vnet == curvnet); 7793 NET_EPOCH_ASSERT(); 7794 7795 if (!V_pf_status.running) 7796 return (PF_PASS); 7797 7798 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7799 7800 if (kif == NULL) { 7801 DPFPRINTF(PF_DEBUG_URGENT, 7802 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 7803 return (PF_DROP); 7804 } 7805 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7806 return (PF_PASS); 7807 7808 if (m->m_flags & M_SKIP_FIREWALL) 7809 return (PF_PASS); 7810 7811 /* Stateless! */ 7812 return (pf_test_eth_rule(dir, kif, m0)); 7813 } 7814 7815 static __inline void 7816 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 7817 { 7818 struct m_tag *mtag; 7819 7820 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7821 7822 /* dummynet adds this tag, but pf does not need it, 7823 * and keeping it creates unexpected behavior, 7824 * e.g. in case of divert(4) usage right after dummynet. */ 7825 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 7826 if (mtag != NULL) 7827 m_tag_delete(m, mtag); 7828 } 7829 7830 static int 7831 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 7832 struct pf_krule *r, struct mbuf **m0) 7833 { 7834 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 7835 } 7836 7837 static int 7838 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 7839 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 7840 struct mbuf **m0) 7841 { 7842 NET_EPOCH_ASSERT(); 7843 7844 if (pd->act.dnpipe || pd->act.dnrpipe) { 7845 struct ip_fw_args dnflow; 7846 if (ip_dn_io_ptr == NULL) { 7847 m_freem(*m0); 7848 *m0 = NULL; 7849 return (ENOMEM); 7850 } 7851 7852 if (pd->pf_mtag == NULL && 7853 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 7854 m_freem(*m0); 7855 *m0 = NULL; 7856 return (ENOMEM); 7857 } 7858 7859 if (ifp != NULL) { 7860 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 7861 7862 pd->pf_mtag->if_index = ifp->if_index; 7863 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 7864 7865 MPASS(sa != NULL); 7866 7867 if (pd->af == AF_INET) 7868 memcpy(&pd->pf_mtag->dst, sa, 7869 sizeof(struct sockaddr_in)); 7870 else 7871 memcpy(&pd->pf_mtag->dst, sa, 7872 sizeof(struct sockaddr_in6)); 7873 } 7874 7875 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 7876 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 7877 ip_dn_io_ptr(m0, &dnflow); 7878 if (*m0 != NULL) { 7879 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7880 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 7881 } 7882 } 7883 } 7884 7885 return (0); 7886 } 7887 7888 #ifdef INET 7889 int 7890 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7891 struct inpcb *inp, struct pf_rule_actions *default_actions) 7892 { 7893 struct pfi_kkif *kif; 7894 u_short action, reason = 0; 7895 struct mbuf *m = *m0; 7896 struct ip *h = NULL; 7897 struct m_tag *mtag; 7898 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7899 struct pf_kstate *s = NULL; 7900 struct pf_kruleset *ruleset = NULL; 7901 struct pf_pdesc pd; 7902 int off, dirndx, use_2nd_queue = 0; 7903 uint16_t tag; 7904 uint8_t rt; 7905 7906 PF_RULES_RLOCK_TRACKER; 7907 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7908 M_ASSERTPKTHDR(m); 7909 7910 if (!V_pf_status.running) 7911 return (PF_PASS); 7912 7913 PF_RULES_RLOCK(); 7914 7915 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7916 7917 if (__predict_false(kif == NULL)) { 7918 DPFPRINTF(PF_DEBUG_URGENT, 7919 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7920 PF_RULES_RUNLOCK(); 7921 return (PF_DROP); 7922 } 7923 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7924 PF_RULES_RUNLOCK(); 7925 return (PF_PASS); 7926 } 7927 7928 if (m->m_flags & M_SKIP_FIREWALL) { 7929 PF_RULES_RUNLOCK(); 7930 return (PF_PASS); 7931 } 7932 7933 memset(&pd, 0, sizeof(pd)); 7934 TAILQ_INIT(&pd.sctp_multihome_jobs); 7935 if (default_actions != NULL) 7936 memcpy(&pd.act, default_actions, sizeof(pd.act)); 7937 pd.pf_mtag = pf_find_mtag(m); 7938 7939 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 7940 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7941 7942 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7943 pd.pf_mtag->if_idxgen); 7944 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7945 PF_RULES_RUNLOCK(); 7946 m_freem(*m0); 7947 *m0 = NULL; 7948 return (PF_PASS); 7949 } 7950 PF_RULES_RUNLOCK(); 7951 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 7952 *m0 = NULL; 7953 return (PF_PASS); 7954 } 7955 7956 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7957 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7958 pd.act.flags = pd.pf_mtag->dnflags; 7959 } 7960 7961 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7962 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 7963 /* Dummynet re-injects packets after they've 7964 * completed their delay. We've already 7965 * processed them, so pass unconditionally. */ 7966 7967 /* But only once. We may see the packet multiple times (e.g. 7968 * PFIL_IN/PFIL_OUT). */ 7969 pf_dummynet_flag_remove(m, pd.pf_mtag); 7970 PF_RULES_RUNLOCK(); 7971 7972 return (PF_PASS); 7973 } 7974 7975 pd.sport = pd.dport = NULL; 7976 pd.proto_sum = NULL; 7977 pd.dir = dir; 7978 pd.sidx = (dir == PF_IN) ? 0 : 1; 7979 pd.didx = (dir == PF_IN) ? 1 : 0; 7980 pd.af = AF_INET; 7981 pd.act.rtableid = -1; 7982 7983 h = mtod(m, struct ip *); 7984 off = h->ip_hl << 2; 7985 7986 if (__predict_false(ip_divert_ptr != NULL) && 7987 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 7988 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 7989 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 7990 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 7991 if (pd.pf_mtag == NULL && 7992 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7993 action = PF_DROP; 7994 goto done; 7995 } 7996 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 7997 } 7998 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 7999 m->m_flags |= M_FASTFWD_OURS; 8000 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8001 } 8002 m_tag_delete(m, mtag); 8003 8004 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8005 if (mtag != NULL) 8006 m_tag_delete(m, mtag); 8007 } else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 8008 /* We do IP header normalization and packet reassembly here */ 8009 action = PF_DROP; 8010 goto done; 8011 } 8012 m = *m0; /* pf_normalize messes with m0 */ 8013 h = mtod(m, struct ip *); 8014 8015 off = h->ip_hl << 2; 8016 if (off < (int)sizeof(struct ip)) { 8017 action = PF_DROP; 8018 REASON_SET(&reason, PFRES_SHORT); 8019 pd.act.log = PF_LOG_FORCE; 8020 goto done; 8021 } 8022 8023 pd.src = (struct pf_addr *)&h->ip_src; 8024 pd.dst = (struct pf_addr *)&h->ip_dst; 8025 pd.ip_sum = &h->ip_sum; 8026 pd.proto = h->ip_p; 8027 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 8028 pd.tot_len = ntohs(h->ip_len); 8029 8030 /* handle fragments that didn't get reassembled by normalization */ 8031 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8032 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 8033 goto done; 8034 } 8035 8036 switch (h->ip_p) { 8037 case IPPROTO_TCP: { 8038 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8039 &action, &reason, AF_INET)) { 8040 if (action != PF_PASS) 8041 pd.act.log = PF_LOG_FORCE; 8042 goto done; 8043 } 8044 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8045 8046 pd.sport = &pd.hdr.tcp.th_sport; 8047 pd.dport = &pd.hdr.tcp.th_dport; 8048 8049 /* Respond to SYN with a syncookie. */ 8050 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8051 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8052 pf_syncookie_send(m, off, &pd); 8053 action = PF_DROP; 8054 break; 8055 } 8056 8057 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 8058 use_2nd_queue = 1; 8059 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8060 if (action == PF_DROP) 8061 goto done; 8062 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8063 if (action == PF_PASS) { 8064 if (V_pfsync_update_state_ptr != NULL) 8065 V_pfsync_update_state_ptr(s); 8066 r = s->rule.ptr; 8067 a = s->anchor.ptr; 8068 } else if (s == NULL) { 8069 /* Validate remote SYN|ACK, re-create original SYN if 8070 * valid. */ 8071 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8072 TH_ACK && pf_syncookie_validate(&pd) && 8073 pd.dir == PF_IN) { 8074 struct mbuf *msyn; 8075 8076 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 8077 &pd); 8078 if (msyn == NULL) { 8079 action = PF_DROP; 8080 break; 8081 } 8082 8083 action = pf_test(dir, pflags, ifp, &msyn, inp, 8084 &pd.act); 8085 m_freem(msyn); 8086 if (action != PF_PASS) 8087 break; 8088 8089 action = pf_test_state_tcp(&s, kif, m, off, h, 8090 &pd, &reason); 8091 if (action != PF_PASS || s == NULL) { 8092 action = PF_DROP; 8093 break; 8094 } 8095 8096 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8097 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8098 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8099 action = pf_synproxy(&pd, &s, &reason); 8100 break; 8101 } else { 8102 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8103 &a, &ruleset, inp); 8104 } 8105 } 8106 break; 8107 } 8108 8109 case IPPROTO_UDP: { 8110 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8111 &action, &reason, AF_INET)) { 8112 if (action != PF_PASS) 8113 pd.act.log = PF_LOG_FORCE; 8114 goto done; 8115 } 8116 pd.sport = &pd.hdr.udp.uh_sport; 8117 pd.dport = &pd.hdr.udp.uh_dport; 8118 if (pd.hdr.udp.uh_dport == 0 || 8119 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8120 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8121 action = PF_DROP; 8122 REASON_SET(&reason, PFRES_SHORT); 8123 goto done; 8124 } 8125 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8126 if (action == PF_PASS) { 8127 if (V_pfsync_update_state_ptr != NULL) 8128 V_pfsync_update_state_ptr(s); 8129 r = s->rule.ptr; 8130 a = s->anchor.ptr; 8131 } else if (s == NULL) 8132 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8133 &a, &ruleset, inp); 8134 break; 8135 } 8136 8137 case IPPROTO_SCTP: { 8138 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8139 &action, &reason, AF_INET)) { 8140 if (action != PF_PASS) 8141 pd.act.log |= PF_LOG_FORCE; 8142 goto done; 8143 } 8144 pd.p_len = pd.tot_len - off; 8145 8146 pd.sport = &pd.hdr.sctp.src_port; 8147 pd.dport = &pd.hdr.sctp.dest_port; 8148 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8149 action = PF_DROP; 8150 REASON_SET(&reason, PFRES_SHORT); 8151 goto done; 8152 } 8153 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8154 if (action == PF_DROP) 8155 goto done; 8156 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8157 &reason); 8158 if (action == PF_PASS) { 8159 if (V_pfsync_update_state_ptr != NULL) 8160 V_pfsync_update_state_ptr(s); 8161 r = s->rule.ptr; 8162 a = s->anchor.ptr; 8163 } else { 8164 action = pf_test_rule(&r, &s, kif, m, off, 8165 &pd, &a, &ruleset, inp); 8166 } 8167 break; 8168 } 8169 8170 case IPPROTO_ICMP: { 8171 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 8172 &action, &reason, AF_INET)) { 8173 if (action != PF_PASS) 8174 pd.act.log = PF_LOG_FORCE; 8175 goto done; 8176 } 8177 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8178 if (action == PF_PASS) { 8179 if (V_pfsync_update_state_ptr != NULL) 8180 V_pfsync_update_state_ptr(s); 8181 r = s->rule.ptr; 8182 a = s->anchor.ptr; 8183 } else if (s == NULL) 8184 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8185 &a, &ruleset, inp); 8186 break; 8187 } 8188 8189 #ifdef INET6 8190 case IPPROTO_ICMPV6: { 8191 action = PF_DROP; 8192 DPFPRINTF(PF_DEBUG_MISC, 8193 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 8194 goto done; 8195 } 8196 #endif 8197 8198 default: 8199 action = pf_test_state_other(&s, kif, m, &pd); 8200 if (action == PF_PASS) { 8201 if (V_pfsync_update_state_ptr != NULL) 8202 V_pfsync_update_state_ptr(s); 8203 r = s->rule.ptr; 8204 a = s->anchor.ptr; 8205 } else if (s == NULL) 8206 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8207 &a, &ruleset, inp); 8208 break; 8209 } 8210 8211 done: 8212 PF_RULES_RUNLOCK(); 8213 if (action == PF_PASS && h->ip_hl > 5 && 8214 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8215 action = PF_DROP; 8216 REASON_SET(&reason, PFRES_IPOPTIONS); 8217 pd.act.log = PF_LOG_FORCE; 8218 DPFPRINTF(PF_DEBUG_MISC, 8219 ("pf: dropping packet with ip options\n")); 8220 } 8221 8222 if (s) { 8223 uint8_t log = pd.act.log; 8224 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8225 pd.act.log |= log; 8226 tag = s->tag; 8227 rt = s->rt; 8228 } else { 8229 tag = r->tag; 8230 rt = r->rt; 8231 } 8232 8233 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8234 action = PF_DROP; 8235 REASON_SET(&reason, PFRES_MEMORY); 8236 } 8237 8238 pf_scrub_ip(&m, &pd); 8239 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8240 pf_normalize_mss(m, off, &pd); 8241 8242 if (pd.act.rtableid >= 0) 8243 M_SETFIB(m, pd.act.rtableid); 8244 8245 if (pd.act.flags & PFSTATE_SETPRIO) { 8246 if (pd.tos & IPTOS_LOWDELAY) 8247 use_2nd_queue = 1; 8248 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8249 action = PF_DROP; 8250 REASON_SET(&reason, PFRES_MEMORY); 8251 pd.act.log = PF_LOG_FORCE; 8252 DPFPRINTF(PF_DEBUG_MISC, 8253 ("pf: failed to allocate 802.1q mtag\n")); 8254 } 8255 } 8256 8257 #ifdef ALTQ 8258 if (action == PF_PASS && pd.act.qid) { 8259 if (pd.pf_mtag == NULL && 8260 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8261 action = PF_DROP; 8262 REASON_SET(&reason, PFRES_MEMORY); 8263 } else { 8264 if (s != NULL) 8265 pd.pf_mtag->qid_hash = pf_state_hash(s); 8266 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 8267 pd.pf_mtag->qid = pd.act.pqid; 8268 else 8269 pd.pf_mtag->qid = pd.act.qid; 8270 /* Add hints for ecn. */ 8271 pd.pf_mtag->hdr = h; 8272 } 8273 } 8274 #endif /* ALTQ */ 8275 8276 /* 8277 * connections redirected to loopback should not match sockets 8278 * bound specifically to loopback due to security implications, 8279 * see tcp_input() and in_pcblookup_listen(). 8280 */ 8281 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8282 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8283 (s->nat_rule.ptr->action == PF_RDR || 8284 s->nat_rule.ptr->action == PF_BINAT) && 8285 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 8286 m->m_flags |= M_SKIP_FIREWALL; 8287 8288 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 8289 r->divert.port && !PACKET_LOOPED(&pd)) { 8290 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 8291 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 8292 if (mtag != NULL) { 8293 ((struct pf_divert_mtag *)(mtag+1))->port = 8294 ntohs(r->divert.port); 8295 ((struct pf_divert_mtag *)(mtag+1))->idir = 8296 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 8297 PF_DIVERT_MTAG_DIR_OUT; 8298 8299 if (s) 8300 PF_STATE_UNLOCK(s); 8301 8302 m_tag_prepend(m, mtag); 8303 if (m->m_flags & M_FASTFWD_OURS) { 8304 if (pd.pf_mtag == NULL && 8305 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8306 action = PF_DROP; 8307 REASON_SET(&reason, PFRES_MEMORY); 8308 pd.act.log = PF_LOG_FORCE; 8309 DPFPRINTF(PF_DEBUG_MISC, 8310 ("pf: failed to allocate tag\n")); 8311 } else { 8312 pd.pf_mtag->flags |= 8313 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8314 m->m_flags &= ~M_FASTFWD_OURS; 8315 } 8316 } 8317 ip_divert_ptr(*m0, dir == PF_IN); 8318 *m0 = NULL; 8319 8320 return (action); 8321 } else { 8322 /* XXX: ipfw has the same behaviour! */ 8323 action = PF_DROP; 8324 REASON_SET(&reason, PFRES_MEMORY); 8325 pd.act.log = PF_LOG_FORCE; 8326 DPFPRINTF(PF_DEBUG_MISC, 8327 ("pf: failed to allocate divert tag\n")); 8328 } 8329 } 8330 /* this flag will need revising if the pkt is forwarded */ 8331 if (pd.pf_mtag) 8332 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 8333 8334 if (pd.act.log) { 8335 struct pf_krule *lr; 8336 struct pf_krule_item *ri; 8337 8338 if (s != NULL && s->nat_rule.ptr != NULL && 8339 s->nat_rule.ptr->log & PF_LOG_ALL) 8340 lr = s->nat_rule.ptr; 8341 else 8342 lr = r; 8343 8344 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8345 PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a, 8346 ruleset, &pd, (s == NULL)); 8347 if (s) { 8348 SLIST_FOREACH(ri, &s->match_rules, entry) 8349 if (ri->r->log & PF_LOG_ALL) 8350 PFLOG_PACKET(kif, m, AF_INET, action, 8351 reason, ri->r, a, ruleset, &pd, 0); 8352 } 8353 } 8354 8355 pf_counter_u64_critical_enter(); 8356 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 8357 pd.tot_len); 8358 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 8359 1); 8360 8361 if (action == PF_PASS || r->action == PF_DROP) { 8362 dirndx = (dir == PF_OUT); 8363 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8364 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8365 pf_update_timestamp(r); 8366 8367 if (a != NULL) { 8368 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8369 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8370 } 8371 if (s != NULL) { 8372 struct pf_krule_item *ri; 8373 8374 if (s->nat_rule.ptr != NULL) { 8375 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8376 1); 8377 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8378 pd.tot_len); 8379 } 8380 if (s->src_node != NULL) { 8381 counter_u64_add(s->src_node->packets[dirndx], 8382 1); 8383 counter_u64_add(s->src_node->bytes[dirndx], 8384 pd.tot_len); 8385 } 8386 if (s->nat_src_node != NULL) { 8387 counter_u64_add(s->nat_src_node->packets[dirndx], 8388 1); 8389 counter_u64_add(s->nat_src_node->bytes[dirndx], 8390 pd.tot_len); 8391 } 8392 dirndx = (dir == s->direction) ? 0 : 1; 8393 s->packets[dirndx]++; 8394 s->bytes[dirndx] += pd.tot_len; 8395 SLIST_FOREACH(ri, &s->match_rules, entry) { 8396 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8397 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 8398 } 8399 } 8400 tr = r; 8401 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8402 if (nr != NULL && r == &V_pf_default_rule) 8403 tr = nr; 8404 if (tr->src.addr.type == PF_ADDR_TABLE) 8405 pfr_update_stats(tr->src.addr.p.tbl, 8406 (s == NULL) ? pd.src : 8407 &s->key[(s->direction == PF_IN)]-> 8408 addr[(s->direction == PF_OUT)], 8409 pd.af, pd.tot_len, dir == PF_OUT, 8410 r->action == PF_PASS, tr->src.neg); 8411 if (tr->dst.addr.type == PF_ADDR_TABLE) 8412 pfr_update_stats(tr->dst.addr.p.tbl, 8413 (s == NULL) ? pd.dst : 8414 &s->key[(s->direction == PF_IN)]-> 8415 addr[(s->direction == PF_IN)], 8416 pd.af, pd.tot_len, dir == PF_OUT, 8417 r->action == PF_PASS, tr->dst.neg); 8418 } 8419 pf_counter_u64_critical_exit(); 8420 8421 switch (action) { 8422 case PF_SYNPROXY_DROP: 8423 m_freem(*m0); 8424 case PF_DEFER: 8425 *m0 = NULL; 8426 action = PF_PASS; 8427 break; 8428 case PF_DROP: 8429 m_freem(*m0); 8430 *m0 = NULL; 8431 break; 8432 default: 8433 /* pf_route() returns unlocked. */ 8434 if (rt) { 8435 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 8436 goto out; 8437 } 8438 if (pf_dummynet(&pd, s, r, m0) != 0) { 8439 action = PF_DROP; 8440 REASON_SET(&reason, PFRES_MEMORY); 8441 } 8442 break; 8443 } 8444 8445 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 8446 8447 if (s && action != PF_DROP) { 8448 if (!s->if_index_in && dir == PF_IN) 8449 s->if_index_in = ifp->if_index; 8450 else if (!s->if_index_out && dir == PF_OUT) 8451 s->if_index_out = ifp->if_index; 8452 } 8453 8454 if (s) 8455 PF_STATE_UNLOCK(s); 8456 8457 out: 8458 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 8459 8460 return (action); 8461 } 8462 #endif /* INET */ 8463 8464 #ifdef INET6 8465 int 8466 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 8467 struct pf_rule_actions *default_actions) 8468 { 8469 struct pfi_kkif *kif; 8470 u_short action, reason = 0; 8471 struct mbuf *m = *m0, *n = NULL; 8472 struct m_tag *mtag; 8473 struct ip6_hdr *h = NULL; 8474 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8475 struct pf_kstate *s = NULL; 8476 struct pf_kruleset *ruleset = NULL; 8477 struct pf_pdesc pd; 8478 int off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0; 8479 uint16_t tag; 8480 uint8_t rt; 8481 8482 PF_RULES_RLOCK_TRACKER; 8483 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8484 M_ASSERTPKTHDR(m); 8485 8486 if (!V_pf_status.running) 8487 return (PF_PASS); 8488 8489 PF_RULES_RLOCK(); 8490 8491 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8492 if (__predict_false(kif == NULL)) { 8493 DPFPRINTF(PF_DEBUG_URGENT, 8494 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 8495 PF_RULES_RUNLOCK(); 8496 return (PF_DROP); 8497 } 8498 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8499 PF_RULES_RUNLOCK(); 8500 return (PF_PASS); 8501 } 8502 8503 if (m->m_flags & M_SKIP_FIREWALL) { 8504 PF_RULES_RUNLOCK(); 8505 return (PF_PASS); 8506 } 8507 8508 memset(&pd, 0, sizeof(pd)); 8509 TAILQ_INIT(&pd.sctp_multihome_jobs); 8510 if (default_actions != NULL) 8511 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8512 pd.pf_mtag = pf_find_mtag(m); 8513 8514 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8515 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8516 8517 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8518 pd.pf_mtag->if_idxgen); 8519 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8520 PF_RULES_RUNLOCK(); 8521 m_freem(*m0); 8522 *m0 = NULL; 8523 return (PF_PASS); 8524 } 8525 PF_RULES_RUNLOCK(); 8526 nd6_output_ifp(ifp, ifp, m, 8527 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 8528 *m0 = NULL; 8529 return (PF_PASS); 8530 } 8531 8532 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8533 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8534 pd.act.flags = pd.pf_mtag->dnflags; 8535 } 8536 8537 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8538 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8539 pf_dummynet_flag_remove(m, pd.pf_mtag); 8540 /* Dummynet re-injects packets after they've 8541 * completed their delay. We've already 8542 * processed them, so pass unconditionally. */ 8543 PF_RULES_RUNLOCK(); 8544 return (PF_PASS); 8545 } 8546 8547 pd.sport = pd.dport = NULL; 8548 pd.ip_sum = NULL; 8549 pd.proto_sum = NULL; 8550 pd.dir = dir; 8551 pd.sidx = (dir == PF_IN) ? 0 : 1; 8552 pd.didx = (dir == PF_IN) ? 1 : 0; 8553 pd.af = AF_INET6; 8554 pd.act.rtableid = -1; 8555 8556 h = mtod(m, struct ip6_hdr *); 8557 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8558 8559 /* We do IP header normalization and packet reassembly here */ 8560 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 8561 action = PF_DROP; 8562 goto done; 8563 } 8564 m = *m0; /* pf_normalize messes with m0 */ 8565 h = mtod(m, struct ip6_hdr *); 8566 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8567 8568 /* 8569 * we do not support jumbogram. if we keep going, zero ip6_plen 8570 * will do something bad, so drop the packet for now. 8571 */ 8572 if (htons(h->ip6_plen) == 0) { 8573 action = PF_DROP; 8574 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 8575 goto done; 8576 } 8577 8578 pd.src = (struct pf_addr *)&h->ip6_src; 8579 pd.dst = (struct pf_addr *)&h->ip6_dst; 8580 pd.tos = IPV6_DSCP(h); 8581 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8582 8583 pd.proto = h->ip6_nxt; 8584 do { 8585 switch (pd.proto) { 8586 case IPPROTO_FRAGMENT: 8587 action = pf_test_fragment(&r, kif, m, h, &pd, &a, 8588 &ruleset); 8589 if (action == PF_DROP) 8590 REASON_SET(&reason, PFRES_FRAG); 8591 goto done; 8592 case IPPROTO_ROUTING: { 8593 struct ip6_rthdr rthdr; 8594 8595 if (rh_cnt++) { 8596 DPFPRINTF(PF_DEBUG_MISC, 8597 ("pf: IPv6 more than one rthdr\n")); 8598 action = PF_DROP; 8599 REASON_SET(&reason, PFRES_IPOPTIONS); 8600 pd.act.log = PF_LOG_FORCE; 8601 goto done; 8602 } 8603 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 8604 &reason, pd.af)) { 8605 DPFPRINTF(PF_DEBUG_MISC, 8606 ("pf: IPv6 short rthdr\n")); 8607 action = PF_DROP; 8608 REASON_SET(&reason, PFRES_SHORT); 8609 pd.act.log = PF_LOG_FORCE; 8610 goto done; 8611 } 8612 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8613 DPFPRINTF(PF_DEBUG_MISC, 8614 ("pf: IPv6 rthdr0\n")); 8615 action = PF_DROP; 8616 REASON_SET(&reason, PFRES_IPOPTIONS); 8617 pd.act.log = PF_LOG_FORCE; 8618 goto done; 8619 } 8620 /* FALLTHROUGH */ 8621 } 8622 case IPPROTO_AH: 8623 case IPPROTO_HOPOPTS: 8624 case IPPROTO_DSTOPTS: { 8625 /* get next header and header length */ 8626 struct ip6_ext opt6; 8627 8628 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 8629 NULL, &reason, pd.af)) { 8630 DPFPRINTF(PF_DEBUG_MISC, 8631 ("pf: IPv6 short opt\n")); 8632 action = PF_DROP; 8633 pd.act.log = PF_LOG_FORCE; 8634 goto done; 8635 } 8636 if (pd.proto == IPPROTO_AH) 8637 off += (opt6.ip6e_len + 2) * 4; 8638 else 8639 off += (opt6.ip6e_len + 1) * 8; 8640 pd.proto = opt6.ip6e_nxt; 8641 /* goto the next header */ 8642 break; 8643 } 8644 default: 8645 terminal++; 8646 break; 8647 } 8648 } while (!terminal); 8649 8650 /* if there's no routing header, use unmodified mbuf for checksumming */ 8651 if (!n) 8652 n = m; 8653 8654 switch (pd.proto) { 8655 case IPPROTO_TCP: { 8656 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8657 &action, &reason, AF_INET6)) { 8658 if (action != PF_PASS) 8659 pd.act.log |= PF_LOG_FORCE; 8660 goto done; 8661 } 8662 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8663 pd.sport = &pd.hdr.tcp.th_sport; 8664 pd.dport = &pd.hdr.tcp.th_dport; 8665 8666 /* Respond to SYN with a syncookie. */ 8667 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8668 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8669 pf_syncookie_send(m, off, &pd); 8670 action = PF_DROP; 8671 break; 8672 } 8673 8674 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8675 if (action == PF_DROP) 8676 goto done; 8677 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8678 if (action == PF_PASS) { 8679 if (V_pfsync_update_state_ptr != NULL) 8680 V_pfsync_update_state_ptr(s); 8681 r = s->rule.ptr; 8682 a = s->anchor.ptr; 8683 } else if (s == NULL) { 8684 /* Validate remote SYN|ACK, re-create original SYN if 8685 * valid. */ 8686 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8687 TH_ACK && pf_syncookie_validate(&pd) && 8688 pd.dir == PF_IN) { 8689 struct mbuf *msyn; 8690 8691 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 8692 off, &pd); 8693 if (msyn == NULL) { 8694 action = PF_DROP; 8695 break; 8696 } 8697 8698 action = pf_test6(dir, pflags, ifp, &msyn, inp, 8699 &pd.act); 8700 m_freem(msyn); 8701 if (action != PF_PASS) 8702 break; 8703 8704 action = pf_test_state_tcp(&s, kif, m, off, h, 8705 &pd, &reason); 8706 if (action != PF_PASS || s == NULL) { 8707 action = PF_DROP; 8708 break; 8709 } 8710 8711 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8712 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8713 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8714 8715 action = pf_synproxy(&pd, &s, &reason); 8716 break; 8717 } else { 8718 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8719 &a, &ruleset, inp); 8720 } 8721 } 8722 break; 8723 } 8724 8725 case IPPROTO_UDP: { 8726 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8727 &action, &reason, AF_INET6)) { 8728 if (action != PF_PASS) 8729 pd.act.log |= PF_LOG_FORCE; 8730 goto done; 8731 } 8732 pd.sport = &pd.hdr.udp.uh_sport; 8733 pd.dport = &pd.hdr.udp.uh_dport; 8734 if (pd.hdr.udp.uh_dport == 0 || 8735 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8736 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8737 action = PF_DROP; 8738 REASON_SET(&reason, PFRES_SHORT); 8739 goto done; 8740 } 8741 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8742 if (action == PF_PASS) { 8743 if (V_pfsync_update_state_ptr != NULL) 8744 V_pfsync_update_state_ptr(s); 8745 r = s->rule.ptr; 8746 a = s->anchor.ptr; 8747 } else if (s == NULL) 8748 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8749 &a, &ruleset, inp); 8750 break; 8751 } 8752 8753 case IPPROTO_SCTP: { 8754 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8755 &action, &reason, AF_INET6)) { 8756 if (action != PF_PASS) 8757 pd.act.log |= PF_LOG_FORCE; 8758 goto done; 8759 } 8760 pd.sport = &pd.hdr.sctp.src_port; 8761 pd.dport = &pd.hdr.sctp.dest_port; 8762 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8763 action = PF_DROP; 8764 REASON_SET(&reason, PFRES_SHORT); 8765 goto done; 8766 } 8767 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8768 if (action == PF_DROP) 8769 goto done; 8770 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8771 &reason); 8772 if (action == PF_PASS) { 8773 if (V_pfsync_update_state_ptr != NULL) 8774 V_pfsync_update_state_ptr(s); 8775 r = s->rule.ptr; 8776 a = s->anchor.ptr; 8777 } else { 8778 action = pf_test_rule(&r, &s, kif, m, off, 8779 &pd, &a, &ruleset, inp); 8780 } 8781 break; 8782 } 8783 8784 case IPPROTO_ICMP: { 8785 action = PF_DROP; 8786 DPFPRINTF(PF_DEBUG_MISC, 8787 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 8788 goto done; 8789 } 8790 8791 case IPPROTO_ICMPV6: { 8792 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 8793 &action, &reason, AF_INET6)) { 8794 if (action != PF_PASS) 8795 pd.act.log |= PF_LOG_FORCE; 8796 goto done; 8797 } 8798 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8799 if (action == PF_PASS) { 8800 if (V_pfsync_update_state_ptr != NULL) 8801 V_pfsync_update_state_ptr(s); 8802 r = s->rule.ptr; 8803 a = s->anchor.ptr; 8804 } else if (s == NULL) 8805 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8806 &a, &ruleset, inp); 8807 break; 8808 } 8809 8810 default: 8811 action = pf_test_state_other(&s, kif, m, &pd); 8812 if (action == PF_PASS) { 8813 if (V_pfsync_update_state_ptr != NULL) 8814 V_pfsync_update_state_ptr(s); 8815 r = s->rule.ptr; 8816 a = s->anchor.ptr; 8817 } else if (s == NULL) 8818 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8819 &a, &ruleset, inp); 8820 break; 8821 } 8822 8823 done: 8824 PF_RULES_RUNLOCK(); 8825 if (n != m) { 8826 m_freem(n); 8827 n = NULL; 8828 } 8829 8830 /* handle dangerous IPv6 extension headers. */ 8831 if (action == PF_PASS && rh_cnt && 8832 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8833 action = PF_DROP; 8834 REASON_SET(&reason, PFRES_IPOPTIONS); 8835 pd.act.log = r->log; 8836 DPFPRINTF(PF_DEBUG_MISC, 8837 ("pf: dropping packet with dangerous v6 headers\n")); 8838 } 8839 8840 if (s) { 8841 uint8_t log = pd.act.log; 8842 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8843 pd.act.log |= log; 8844 tag = s->tag; 8845 rt = s->rt; 8846 } else { 8847 tag = r->tag; 8848 rt = r->rt; 8849 } 8850 8851 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8852 action = PF_DROP; 8853 REASON_SET(&reason, PFRES_MEMORY); 8854 } 8855 8856 pf_scrub_ip6(&m, &pd); 8857 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8858 pf_normalize_mss(m, off, &pd); 8859 8860 if (pd.act.rtableid >= 0) 8861 M_SETFIB(m, pd.act.rtableid); 8862 8863 if (pd.act.flags & PFSTATE_SETPRIO) { 8864 if (pd.tos & IPTOS_LOWDELAY) 8865 use_2nd_queue = 1; 8866 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8867 action = PF_DROP; 8868 REASON_SET(&reason, PFRES_MEMORY); 8869 pd.act.log = PF_LOG_FORCE; 8870 DPFPRINTF(PF_DEBUG_MISC, 8871 ("pf: failed to allocate 802.1q mtag\n")); 8872 } 8873 } 8874 8875 #ifdef ALTQ 8876 if (action == PF_PASS && pd.act.qid) { 8877 if (pd.pf_mtag == NULL && 8878 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8879 action = PF_DROP; 8880 REASON_SET(&reason, PFRES_MEMORY); 8881 } else { 8882 if (s != NULL) 8883 pd.pf_mtag->qid_hash = pf_state_hash(s); 8884 if (pd.tos & IPTOS_LOWDELAY) 8885 pd.pf_mtag->qid = pd.act.pqid; 8886 else 8887 pd.pf_mtag->qid = pd.act.qid; 8888 /* Add hints for ecn. */ 8889 pd.pf_mtag->hdr = h; 8890 } 8891 } 8892 #endif /* ALTQ */ 8893 8894 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8895 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8896 (s->nat_rule.ptr->action == PF_RDR || 8897 s->nat_rule.ptr->action == PF_BINAT) && 8898 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 8899 m->m_flags |= M_SKIP_FIREWALL; 8900 8901 /* XXX: Anybody working on it?! */ 8902 if (r->divert.port) 8903 printf("pf: divert(9) is not supported for IPv6\n"); 8904 8905 if (pd.act.log) { 8906 struct pf_krule *lr; 8907 struct pf_krule_item *ri; 8908 8909 if (s != NULL && s->nat_rule.ptr != NULL && 8910 s->nat_rule.ptr->log & PF_LOG_ALL) 8911 lr = s->nat_rule.ptr; 8912 else 8913 lr = r; 8914 8915 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8916 PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset, 8917 &pd, (s == NULL)); 8918 if (s) { 8919 SLIST_FOREACH(ri, &s->match_rules, entry) 8920 if (ri->r->log & PF_LOG_ALL) 8921 PFLOG_PACKET(kif, m, AF_INET6, action, reason, 8922 ri->r, a, ruleset, &pd, 0); 8923 } 8924 } 8925 8926 pf_counter_u64_critical_enter(); 8927 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 8928 pd.tot_len); 8929 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 8930 1); 8931 8932 if (action == PF_PASS || r->action == PF_DROP) { 8933 dirndx = (dir == PF_OUT); 8934 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8935 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8936 if (a != NULL) { 8937 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8938 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8939 } 8940 if (s != NULL) { 8941 if (s->nat_rule.ptr != NULL) { 8942 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8943 1); 8944 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8945 pd.tot_len); 8946 } 8947 if (s->src_node != NULL) { 8948 counter_u64_add(s->src_node->packets[dirndx], 8949 1); 8950 counter_u64_add(s->src_node->bytes[dirndx], 8951 pd.tot_len); 8952 } 8953 if (s->nat_src_node != NULL) { 8954 counter_u64_add(s->nat_src_node->packets[dirndx], 8955 1); 8956 counter_u64_add(s->nat_src_node->bytes[dirndx], 8957 pd.tot_len); 8958 } 8959 dirndx = (dir == s->direction) ? 0 : 1; 8960 s->packets[dirndx]++; 8961 s->bytes[dirndx] += pd.tot_len; 8962 } 8963 tr = r; 8964 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8965 if (nr != NULL && r == &V_pf_default_rule) 8966 tr = nr; 8967 if (tr->src.addr.type == PF_ADDR_TABLE) 8968 pfr_update_stats(tr->src.addr.p.tbl, 8969 (s == NULL) ? pd.src : 8970 &s->key[(s->direction == PF_IN)]->addr[0], 8971 pd.af, pd.tot_len, dir == PF_OUT, 8972 r->action == PF_PASS, tr->src.neg); 8973 if (tr->dst.addr.type == PF_ADDR_TABLE) 8974 pfr_update_stats(tr->dst.addr.p.tbl, 8975 (s == NULL) ? pd.dst : 8976 &s->key[(s->direction == PF_IN)]->addr[1], 8977 pd.af, pd.tot_len, dir == PF_OUT, 8978 r->action == PF_PASS, tr->dst.neg); 8979 } 8980 pf_counter_u64_critical_exit(); 8981 8982 switch (action) { 8983 case PF_SYNPROXY_DROP: 8984 m_freem(*m0); 8985 case PF_DEFER: 8986 *m0 = NULL; 8987 action = PF_PASS; 8988 break; 8989 case PF_DROP: 8990 m_freem(*m0); 8991 *m0 = NULL; 8992 break; 8993 default: 8994 /* pf_route6() returns unlocked. */ 8995 if (rt) { 8996 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 8997 goto out; 8998 } 8999 if (pf_dummynet(&pd, s, r, m0) != 0) { 9000 action = PF_DROP; 9001 REASON_SET(&reason, PFRES_MEMORY); 9002 } 9003 break; 9004 } 9005 9006 if (s && action != PF_DROP) { 9007 if (!s->if_index_in && dir == PF_IN) 9008 s->if_index_in = ifp->if_index; 9009 else if (!s->if_index_out && dir == PF_OUT) 9010 s->if_index_out = ifp->if_index; 9011 } 9012 9013 if (s) 9014 PF_STATE_UNLOCK(s); 9015 9016 /* If reassembled packet passed, create new fragments. */ 9017 if (action == PF_PASS && *m0 && dir == PF_OUT && 9018 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9019 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9020 9021 out: 9022 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 9023 9024 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9025 9026 return (action); 9027 } 9028 #endif /* INET6 */ 9029