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