1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 #include "opt_bpf.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_pf.h" 45 #include "opt_sctp.h" 46 47 #include <sys/param.h> 48 #include <sys/bus.h> 49 #include <sys/endian.h> 50 #include <sys/gsb_crc32.h> 51 #include <sys/hash.h> 52 #include <sys/interrupt.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/limits.h> 56 #include <sys/mbuf.h> 57 #include <sys/md5.h> 58 #include <sys/random.h> 59 #include <sys/refcount.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/sysctl.h> 63 #include <sys/taskqueue.h> 64 #include <sys/ucred.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/if_private.h> 69 #include <net/if_types.h> 70 #include <net/if_vlan_var.h> 71 #include <net/route.h> 72 #include <net/route/nhop.h> 73 #include <net/vnet.h> 74 75 #include <net/pfil.h> 76 #include <net/pfvar.h> 77 #include <net/if_pflog.h> 78 #include <net/if_pfsync.h> 79 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_var.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_fw.h> 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/tcp.h> 89 #include <netinet/tcp_fsm.h> 90 #include <netinet/tcp_seq.h> 91 #include <netinet/tcp_timer.h> 92 #include <netinet/tcp_var.h> 93 #include <netinet/udp.h> 94 #include <netinet/udp_var.h> 95 96 /* dummynet */ 97 #include <netinet/ip_dummynet.h> 98 #include <netinet/ip_fw.h> 99 #include <netpfil/ipfw/dn_heap.h> 100 #include <netpfil/ipfw/ip_fw_private.h> 101 #include <netpfil/ipfw/ip_dn_private.h> 102 103 #ifdef INET6 104 #include <netinet/ip6.h> 105 #include <netinet/icmp6.h> 106 #include <netinet6/nd6.h> 107 #include <netinet6/ip6_var.h> 108 #include <netinet6/in6_pcb.h> 109 #include <netinet6/in6_fib.h> 110 #include <netinet6/scope6_var.h> 111 #endif /* INET6 */ 112 113 #include <netinet/sctp_header.h> 114 #include <netinet/sctp_crc32.h> 115 116 #include <machine/in_cksum.h> 117 #include <security/mac/mac_framework.h> 118 119 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121 SDT_PROVIDER_DEFINE(pf); 122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 130 "struct pf_krule *", "struct mbuf *", "int"); 131 132 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 133 "struct mbuf *"); 134 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 135 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 136 "int", "struct pf_keth_rule *", "char *"); 137 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 138 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 139 "int", "struct pf_keth_rule *"); 140 141 /* 142 * Global variables 143 */ 144 145 /* state tables */ 146 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 147 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 148 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 149 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 150 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 151 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 152 VNET_DEFINE(struct pf_kstatus, pf_status); 153 154 VNET_DEFINE(u_int32_t, ticket_altqs_active); 155 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 156 VNET_DEFINE(int, altqs_inactive_open); 157 VNET_DEFINE(u_int32_t, ticket_pabuf); 158 159 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 160 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 161 VNET_DEFINE(u_char, pf_tcp_secret[16]); 162 #define V_pf_tcp_secret VNET(pf_tcp_secret) 163 VNET_DEFINE(int, pf_tcp_secret_init); 164 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 165 VNET_DEFINE(int, pf_tcp_iss_off); 166 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 167 VNET_DECLARE(int, pf_vnet_active); 168 #define V_pf_vnet_active VNET(pf_vnet_active) 169 170 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 171 #define V_pf_purge_idx VNET(pf_purge_idx) 172 173 #ifdef PF_WANT_32_TO_64_COUNTER 174 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 175 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 176 177 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 178 VNET_DEFINE(size_t, pf_allrulecount); 179 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 180 #endif 181 182 struct pf_sctp_endpoint; 183 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 184 struct pf_sctp_source { 185 sa_family_t af; 186 struct pf_addr addr; 187 TAILQ_ENTRY(pf_sctp_source) entry; 188 }; 189 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 190 struct pf_sctp_endpoint 191 { 192 uint32_t v_tag; 193 struct pf_sctp_sources sources; 194 RB_ENTRY(pf_sctp_endpoint) entry; 195 }; 196 static int 197 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 198 { 199 return (a->v_tag - b->v_tag); 200 } 201 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 202 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 203 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 204 #define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 205 static struct mtx_padalign pf_sctp_endpoints_mtx; 206 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 207 #define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 208 #define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 209 210 /* 211 * Queue for pf_intr() sends. 212 */ 213 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 214 struct pf_send_entry { 215 STAILQ_ENTRY(pf_send_entry) pfse_next; 216 struct mbuf *pfse_m; 217 enum { 218 PFSE_IP, 219 PFSE_IP6, 220 PFSE_ICMP, 221 PFSE_ICMP6, 222 } pfse_type; 223 struct { 224 int type; 225 int code; 226 int mtu; 227 } icmpopts; 228 }; 229 230 STAILQ_HEAD(pf_send_head, pf_send_entry); 231 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 232 #define V_pf_sendqueue VNET(pf_sendqueue) 233 234 static struct mtx_padalign pf_sendqueue_mtx; 235 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 236 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 237 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 238 239 /* 240 * Queue for pf_overload_task() tasks. 241 */ 242 struct pf_overload_entry { 243 SLIST_ENTRY(pf_overload_entry) next; 244 struct pf_addr addr; 245 sa_family_t af; 246 uint8_t dir; 247 struct pf_krule *rule; 248 }; 249 250 SLIST_HEAD(pf_overload_head, pf_overload_entry); 251 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 252 #define V_pf_overloadqueue VNET(pf_overloadqueue) 253 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 254 #define V_pf_overloadtask VNET(pf_overloadtask) 255 256 static struct mtx_padalign pf_overloadqueue_mtx; 257 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 258 "pf overload/flush queue", MTX_DEF); 259 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 260 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 261 262 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 263 struct mtx_padalign pf_unlnkdrules_mtx; 264 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 265 MTX_DEF); 266 267 struct sx pf_config_lock; 268 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 269 270 struct mtx_padalign pf_table_stats_lock; 271 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 272 MTX_DEF); 273 274 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 275 #define V_pf_sources_z VNET(pf_sources_z) 276 uma_zone_t pf_mtag_z; 277 VNET_DEFINE(uma_zone_t, pf_state_z); 278 VNET_DEFINE(uma_zone_t, pf_state_key_z); 279 280 VNET_DEFINE(struct unrhdr64, pf_stateid); 281 282 static void pf_src_tree_remove_state(struct pf_kstate *); 283 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 284 u_int32_t); 285 static void pf_add_threshold(struct pf_threshold *); 286 static int pf_check_threshold(struct pf_threshold *); 287 288 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 289 u_int16_t *, u_int16_t *, struct pf_addr *, 290 u_int16_t, u_int8_t, sa_family_t); 291 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 292 struct tcphdr *, struct pf_state_peer *); 293 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 294 struct pf_addr *, struct pf_addr *, u_int16_t, 295 u_int16_t *, u_int16_t *, u_int16_t *, 296 u_int16_t *, u_int8_t, sa_family_t); 297 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 298 sa_family_t, struct pf_krule *, int); 299 static void pf_detach_state(struct pf_kstate *); 300 static int pf_state_key_attach(struct pf_state_key *, 301 struct pf_state_key *, struct pf_kstate *); 302 static void pf_state_key_detach(struct pf_kstate *, int); 303 static int pf_state_key_ctor(void *, int, void *, int); 304 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 305 static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 306 struct pf_krule *, struct mbuf **); 307 static int pf_dummynet_route(struct pf_pdesc *, 308 struct pf_kstate *, struct pf_krule *, 309 struct ifnet *, struct sockaddr *, struct mbuf **); 310 static int pf_test_eth_rule(int, struct pfi_kkif *, 311 struct mbuf **); 312 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 313 struct pfi_kkif *, struct mbuf *, int, 314 struct pf_pdesc *, struct pf_krule **, 315 struct pf_kruleset **, struct inpcb *); 316 static int pf_create_state(struct pf_krule *, struct pf_krule *, 317 struct pf_krule *, struct pf_pdesc *, 318 struct pf_ksrc_node *, struct pf_state_key *, 319 struct pf_state_key *, struct mbuf *, int, 320 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 321 struct pf_kstate **, int, u_int16_t, u_int16_t, 322 int, struct pf_krule_slist *); 323 static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 324 struct mbuf *, void *, struct pf_pdesc *, 325 struct pf_krule **, struct pf_kruleset **); 326 static int pf_tcp_track_full(struct pf_kstate **, 327 struct pfi_kkif *, struct mbuf *, int, 328 struct pf_pdesc *, u_short *, int *); 329 static int pf_tcp_track_sloppy(struct pf_kstate **, 330 struct pf_pdesc *, u_short *); 331 static int pf_test_state_tcp(struct pf_kstate **, 332 struct pfi_kkif *, struct mbuf *, int, 333 void *, struct pf_pdesc *, u_short *); 334 static int pf_test_state_udp(struct pf_kstate **, 335 struct pfi_kkif *, struct mbuf *, int, 336 void *, struct pf_pdesc *); 337 static int pf_test_state_icmp(struct pf_kstate **, 338 struct pfi_kkif *, struct mbuf *, int, 339 void *, struct pf_pdesc *, u_short *); 340 static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 341 static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 342 struct pfi_kkif *, struct pf_kstate *, int); 343 static int pf_test_state_sctp(struct pf_kstate **, 344 struct pfi_kkif *, struct mbuf *, int, 345 void *, struct pf_pdesc *, u_short *); 346 static int pf_test_state_other(struct pf_kstate **, 347 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 348 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 349 int, u_int16_t); 350 static int pf_check_proto_cksum(struct mbuf *, int, int, 351 u_int8_t, sa_family_t); 352 static void pf_print_state_parts(struct pf_kstate *, 353 struct pf_state_key *, struct pf_state_key *); 354 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 355 bool, u_int8_t); 356 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 357 struct pf_state_key_cmp *, u_int); 358 static int pf_src_connlimit(struct pf_kstate **); 359 static void pf_overload_task(void *v, int pending); 360 static u_short pf_insert_src_node(struct pf_ksrc_node **, 361 struct pf_krule *, struct pf_addr *, sa_family_t); 362 static u_int pf_purge_expired_states(u_int, int); 363 static void pf_purge_unlinked_rules(void); 364 static int pf_mtag_uminit(void *, int, int); 365 static void pf_mtag_free(struct m_tag *); 366 static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 367 int, struct pf_state_key *); 368 #ifdef INET 369 static void pf_route(struct mbuf **, struct pf_krule *, 370 struct ifnet *, struct pf_kstate *, 371 struct pf_pdesc *, struct inpcb *); 372 #endif /* INET */ 373 #ifdef INET6 374 static void pf_change_a6(struct pf_addr *, u_int16_t *, 375 struct pf_addr *, u_int8_t); 376 static void pf_route6(struct mbuf **, struct pf_krule *, 377 struct ifnet *, struct pf_kstate *, 378 struct pf_pdesc *, struct inpcb *); 379 #endif /* INET6 */ 380 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 381 382 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 383 384 extern int pf_end_threads; 385 extern struct proc *pf_purge_proc; 386 387 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 388 389 #define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 390 do { \ 391 struct pf_state_key *nk; \ 392 if ((pd->dir) == PF_OUT) \ 393 nk = (_s)->key[PF_SK_STACK]; \ 394 else \ 395 nk = (_s)->key[PF_SK_WIRE]; \ 396 pf_packet_rework_nat(_m, _pd, _off, nk); \ 397 } while (0) 398 399 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 400 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 401 402 #define STATE_LOOKUP(i, k, s, pd) \ 403 do { \ 404 (s) = pf_find_state((i), (k), (pd->dir)); \ 405 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 406 if ((s) == NULL) \ 407 return (PF_DROP); \ 408 if (PACKET_LOOPED(pd)) \ 409 return (PF_PASS); \ 410 } while (0) 411 412 #define BOUND_IFACE(r, k) \ 413 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 414 415 #define STATE_INC_COUNTERS(s) \ 416 do { \ 417 struct pf_krule_item *mrm; \ 418 counter_u64_add(s->rule.ptr->states_cur, 1); \ 419 counter_u64_add(s->rule.ptr->states_tot, 1); \ 420 if (s->anchor.ptr != NULL) { \ 421 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 422 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 423 } \ 424 if (s->nat_rule.ptr != NULL) { \ 425 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 426 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 427 } \ 428 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 429 counter_u64_add(mrm->r->states_cur, 1); \ 430 counter_u64_add(mrm->r->states_tot, 1); \ 431 } \ 432 } while (0) 433 434 #define STATE_DEC_COUNTERS(s) \ 435 do { \ 436 struct pf_krule_item *mrm; \ 437 if (s->nat_rule.ptr != NULL) \ 438 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 439 if (s->anchor.ptr != NULL) \ 440 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 441 counter_u64_add(s->rule.ptr->states_cur, -1); \ 442 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 443 counter_u64_add(mrm->r->states_cur, -1); \ 444 } while (0) 445 446 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 447 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 448 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 449 VNET_DEFINE(struct pf_idhash *, pf_idhash); 450 VNET_DEFINE(struct pf_srchash *, pf_srchash); 451 452 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 453 "pf(4)"); 454 455 u_long pf_hashmask; 456 u_long pf_srchashmask; 457 static u_long pf_hashsize; 458 static u_long pf_srchashsize; 459 u_long pf_ioctl_maxcount = 65535; 460 461 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 462 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 463 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 464 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 465 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 466 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 467 468 VNET_DEFINE(void *, pf_swi_cookie); 469 VNET_DEFINE(struct intr_event *, pf_swi_ie); 470 471 VNET_DEFINE(uint32_t, pf_hashseed); 472 #define V_pf_hashseed VNET(pf_hashseed) 473 474 static void 475 pf_sctp_checksum(struct mbuf *m, int off) 476 { 477 uint32_t sum = 0; 478 479 /* Zero out the checksum, to enable recalculation. */ 480 m_copyback(m, off + offsetof(struct sctphdr, checksum), 481 sizeof(sum), (caddr_t)&sum); 482 483 sum = sctp_calculate_cksum(m, off); 484 485 m_copyback(m, off + offsetof(struct sctphdr, checksum), 486 sizeof(sum), (caddr_t)&sum); 487 } 488 489 int 490 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 491 { 492 493 switch (af) { 494 #ifdef INET 495 case AF_INET: 496 if (a->addr32[0] > b->addr32[0]) 497 return (1); 498 if (a->addr32[0] < b->addr32[0]) 499 return (-1); 500 break; 501 #endif /* INET */ 502 #ifdef INET6 503 case AF_INET6: 504 if (a->addr32[3] > b->addr32[3]) 505 return (1); 506 if (a->addr32[3] < b->addr32[3]) 507 return (-1); 508 if (a->addr32[2] > b->addr32[2]) 509 return (1); 510 if (a->addr32[2] < b->addr32[2]) 511 return (-1); 512 if (a->addr32[1] > b->addr32[1]) 513 return (1); 514 if (a->addr32[1] < b->addr32[1]) 515 return (-1); 516 if (a->addr32[0] > b->addr32[0]) 517 return (1); 518 if (a->addr32[0] < b->addr32[0]) 519 return (-1); 520 break; 521 #endif /* INET6 */ 522 default: 523 panic("%s: unknown address family %u", __func__, af); 524 } 525 return (0); 526 } 527 528 static void 529 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 530 struct pf_state_key *nk) 531 { 532 533 switch (pd->proto) { 534 case IPPROTO_TCP: { 535 struct tcphdr *th = &pd->hdr.tcp; 536 537 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 538 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 539 &th->th_sum, &nk->addr[pd->sidx], 540 nk->port[pd->sidx], 0, pd->af); 541 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 542 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 543 &th->th_sum, &nk->addr[pd->didx], 544 nk->port[pd->didx], 0, pd->af); 545 m_copyback(m, off, sizeof(*th), (caddr_t)th); 546 break; 547 } 548 case IPPROTO_UDP: { 549 struct udphdr *uh = &pd->hdr.udp; 550 551 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 552 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 553 &uh->uh_sum, &nk->addr[pd->sidx], 554 nk->port[pd->sidx], 1, pd->af); 555 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 556 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 557 &uh->uh_sum, &nk->addr[pd->didx], 558 nk->port[pd->didx], 1, pd->af); 559 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 560 break; 561 } 562 case IPPROTO_SCTP: { 563 struct sctphdr *sh = &pd->hdr.sctp; 564 uint16_t checksum = 0; 565 566 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 567 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 568 &checksum, &nk->addr[pd->sidx], 569 nk->port[pd->sidx], 1, pd->af); 570 } 571 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 572 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 573 &checksum, &nk->addr[pd->didx], 574 nk->port[pd->didx], 1, pd->af); 575 } 576 577 break; 578 } 579 case IPPROTO_ICMP: { 580 struct icmp *ih = &pd->hdr.icmp; 581 582 if (nk->port[pd->sidx] != ih->icmp_id) { 583 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 584 ih->icmp_cksum, ih->icmp_id, 585 nk->port[pd->sidx], 0); 586 ih->icmp_id = nk->port[pd->sidx]; 587 pd->sport = &ih->icmp_id; 588 589 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 590 } 591 /* FALLTHROUGH */ 592 } 593 default: 594 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 595 switch (pd->af) { 596 case AF_INET: 597 pf_change_a(&pd->src->v4.s_addr, 598 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 599 0); 600 break; 601 case AF_INET6: 602 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 603 break; 604 } 605 } 606 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 607 switch (pd->af) { 608 case AF_INET: 609 pf_change_a(&pd->dst->v4.s_addr, 610 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 611 0); 612 break; 613 case AF_INET6: 614 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 615 break; 616 } 617 } 618 break; 619 } 620 } 621 622 static __inline uint32_t 623 pf_hashkey(struct pf_state_key *sk) 624 { 625 uint32_t h; 626 627 h = murmur3_32_hash32((uint32_t *)sk, 628 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 629 V_pf_hashseed); 630 631 return (h & pf_hashmask); 632 } 633 634 static __inline uint32_t 635 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 636 { 637 uint32_t h; 638 639 switch (af) { 640 case AF_INET: 641 h = murmur3_32_hash32((uint32_t *)&addr->v4, 642 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 643 break; 644 case AF_INET6: 645 h = murmur3_32_hash32((uint32_t *)&addr->v6, 646 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 647 break; 648 default: 649 panic("%s: unknown address family %u", __func__, af); 650 } 651 652 return (h & pf_srchashmask); 653 } 654 655 #ifdef ALTQ 656 static int 657 pf_state_hash(struct pf_kstate *s) 658 { 659 u_int32_t hv = (intptr_t)s / sizeof(*s); 660 661 hv ^= crc32(&s->src, sizeof(s->src)); 662 hv ^= crc32(&s->dst, sizeof(s->dst)); 663 if (hv == 0) 664 hv = 1; 665 return (hv); 666 } 667 #endif 668 669 static __inline void 670 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 671 { 672 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 673 s->dst.state = newstate; 674 if (which == PF_PEER_DST) 675 return; 676 if (s->src.state == newstate) 677 return; 678 if (s->creatorid == V_pf_status.hostid && 679 s->key[PF_SK_STACK] != NULL && 680 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 681 !(TCPS_HAVEESTABLISHED(s->src.state) || 682 s->src.state == TCPS_CLOSED) && 683 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 684 atomic_add_32(&V_pf_status.states_halfopen, -1); 685 686 s->src.state = newstate; 687 } 688 689 #ifdef INET6 690 void 691 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 692 { 693 switch (af) { 694 #ifdef INET 695 case AF_INET: 696 dst->addr32[0] = src->addr32[0]; 697 break; 698 #endif /* INET */ 699 case AF_INET6: 700 dst->addr32[0] = src->addr32[0]; 701 dst->addr32[1] = src->addr32[1]; 702 dst->addr32[2] = src->addr32[2]; 703 dst->addr32[3] = src->addr32[3]; 704 break; 705 } 706 } 707 #endif /* INET6 */ 708 709 static void 710 pf_init_threshold(struct pf_threshold *threshold, 711 u_int32_t limit, u_int32_t seconds) 712 { 713 threshold->limit = limit * PF_THRESHOLD_MULT; 714 threshold->seconds = seconds; 715 threshold->count = 0; 716 threshold->last = time_uptime; 717 } 718 719 static void 720 pf_add_threshold(struct pf_threshold *threshold) 721 { 722 u_int32_t t = time_uptime, diff = t - threshold->last; 723 724 if (diff >= threshold->seconds) 725 threshold->count = 0; 726 else 727 threshold->count -= threshold->count * diff / 728 threshold->seconds; 729 threshold->count += PF_THRESHOLD_MULT; 730 threshold->last = t; 731 } 732 733 static int 734 pf_check_threshold(struct pf_threshold *threshold) 735 { 736 return (threshold->count > threshold->limit); 737 } 738 739 static int 740 pf_src_connlimit(struct pf_kstate **state) 741 { 742 struct pf_overload_entry *pfoe; 743 int bad = 0; 744 745 PF_STATE_LOCK_ASSERT(*state); 746 /* 747 * XXXKS: The src node is accessed unlocked! 748 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 749 */ 750 751 (*state)->src_node->conn++; 752 (*state)->src.tcp_est = 1; 753 pf_add_threshold(&(*state)->src_node->conn_rate); 754 755 if ((*state)->rule.ptr->max_src_conn && 756 (*state)->rule.ptr->max_src_conn < 757 (*state)->src_node->conn) { 758 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 759 bad++; 760 } 761 762 if ((*state)->rule.ptr->max_src_conn_rate.limit && 763 pf_check_threshold(&(*state)->src_node->conn_rate)) { 764 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 765 bad++; 766 } 767 768 if (!bad) 769 return (0); 770 771 /* Kill this state. */ 772 (*state)->timeout = PFTM_PURGE; 773 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 774 775 if ((*state)->rule.ptr->overload_tbl == NULL) 776 return (1); 777 778 /* Schedule overloading and flushing task. */ 779 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 780 if (pfoe == NULL) 781 return (1); /* too bad :( */ 782 783 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 784 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 785 pfoe->rule = (*state)->rule.ptr; 786 pfoe->dir = (*state)->direction; 787 PF_OVERLOADQ_LOCK(); 788 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 789 PF_OVERLOADQ_UNLOCK(); 790 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 791 792 return (1); 793 } 794 795 static void 796 pf_overload_task(void *v, int pending) 797 { 798 struct pf_overload_head queue; 799 struct pfr_addr p; 800 struct pf_overload_entry *pfoe, *pfoe1; 801 uint32_t killed = 0; 802 803 CURVNET_SET((struct vnet *)v); 804 805 PF_OVERLOADQ_LOCK(); 806 queue = V_pf_overloadqueue; 807 SLIST_INIT(&V_pf_overloadqueue); 808 PF_OVERLOADQ_UNLOCK(); 809 810 bzero(&p, sizeof(p)); 811 SLIST_FOREACH(pfoe, &queue, next) { 812 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 813 if (V_pf_status.debug >= PF_DEBUG_MISC) { 814 printf("%s: blocking address ", __func__); 815 pf_print_host(&pfoe->addr, 0, pfoe->af); 816 printf("\n"); 817 } 818 819 p.pfra_af = pfoe->af; 820 switch (pfoe->af) { 821 #ifdef INET 822 case AF_INET: 823 p.pfra_net = 32; 824 p.pfra_ip4addr = pfoe->addr.v4; 825 break; 826 #endif 827 #ifdef INET6 828 case AF_INET6: 829 p.pfra_net = 128; 830 p.pfra_ip6addr = pfoe->addr.v6; 831 break; 832 #endif 833 } 834 835 PF_RULES_WLOCK(); 836 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 837 PF_RULES_WUNLOCK(); 838 } 839 840 /* 841 * Remove those entries, that don't need flushing. 842 */ 843 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 844 if (pfoe->rule->flush == 0) { 845 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 846 free(pfoe, M_PFTEMP); 847 } else 848 counter_u64_add( 849 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 850 851 /* If nothing to flush, return. */ 852 if (SLIST_EMPTY(&queue)) { 853 CURVNET_RESTORE(); 854 return; 855 } 856 857 for (int i = 0; i <= pf_hashmask; i++) { 858 struct pf_idhash *ih = &V_pf_idhash[i]; 859 struct pf_state_key *sk; 860 struct pf_kstate *s; 861 862 PF_HASHROW_LOCK(ih); 863 LIST_FOREACH(s, &ih->states, entry) { 864 sk = s->key[PF_SK_WIRE]; 865 SLIST_FOREACH(pfoe, &queue, next) 866 if (sk->af == pfoe->af && 867 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 868 pfoe->rule == s->rule.ptr) && 869 ((pfoe->dir == PF_OUT && 870 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 871 (pfoe->dir == PF_IN && 872 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 873 s->timeout = PFTM_PURGE; 874 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 875 killed++; 876 } 877 } 878 PF_HASHROW_UNLOCK(ih); 879 } 880 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 881 free(pfoe, M_PFTEMP); 882 if (V_pf_status.debug >= PF_DEBUG_MISC) 883 printf("%s: %u states killed", __func__, killed); 884 885 CURVNET_RESTORE(); 886 } 887 888 /* 889 * Can return locked on failure, so that we can consistently 890 * allocate and insert a new one. 891 */ 892 struct pf_ksrc_node * 893 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 894 struct pf_srchash **sh, bool returnlocked) 895 { 896 struct pf_ksrc_node *n; 897 898 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 899 900 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 901 PF_HASHROW_LOCK(*sh); 902 LIST_FOREACH(n, &(*sh)->nodes, entry) 903 if (n->rule.ptr == rule && n->af == af && 904 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 905 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 906 break; 907 908 if (n != NULL) { 909 n->states++; 910 PF_HASHROW_UNLOCK(*sh); 911 } else if (returnlocked == false) 912 PF_HASHROW_UNLOCK(*sh); 913 914 return (n); 915 } 916 917 static void 918 pf_free_src_node(struct pf_ksrc_node *sn) 919 { 920 921 for (int i = 0; i < 2; i++) { 922 counter_u64_free(sn->bytes[i]); 923 counter_u64_free(sn->packets[i]); 924 } 925 uma_zfree(V_pf_sources_z, sn); 926 } 927 928 static u_short 929 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 930 struct pf_addr *src, sa_family_t af) 931 { 932 u_short reason = 0; 933 struct pf_srchash *sh = NULL; 934 935 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 936 rule->rpool.opts & PF_POOL_STICKYADDR), 937 ("%s for non-tracking rule %p", __func__, rule)); 938 939 if (*sn == NULL) 940 *sn = pf_find_src_node(src, rule, af, &sh, true); 941 942 if (*sn == NULL) { 943 PF_HASHROW_ASSERT(sh); 944 945 if (rule->max_src_nodes && 946 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 947 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 948 PF_HASHROW_UNLOCK(sh); 949 reason = PFRES_SRCLIMIT; 950 goto done; 951 } 952 953 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 954 if ((*sn) == NULL) { 955 PF_HASHROW_UNLOCK(sh); 956 reason = PFRES_MEMORY; 957 goto done; 958 } 959 960 for (int i = 0; i < 2; i++) { 961 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 962 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 963 964 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 965 pf_free_src_node(*sn); 966 PF_HASHROW_UNLOCK(sh); 967 reason = PFRES_MEMORY; 968 goto done; 969 } 970 } 971 972 pf_init_threshold(&(*sn)->conn_rate, 973 rule->max_src_conn_rate.limit, 974 rule->max_src_conn_rate.seconds); 975 976 MPASS((*sn)->lock == NULL); 977 (*sn)->lock = &sh->lock; 978 979 (*sn)->af = af; 980 (*sn)->rule.ptr = rule; 981 PF_ACPY(&(*sn)->addr, src, af); 982 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 983 (*sn)->creation = time_uptime; 984 (*sn)->ruletype = rule->action; 985 (*sn)->states = 1; 986 if ((*sn)->rule.ptr != NULL) 987 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 988 PF_HASHROW_UNLOCK(sh); 989 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 990 } else { 991 if (rule->max_src_states && 992 (*sn)->states >= rule->max_src_states) { 993 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 994 1); 995 reason = PFRES_SRCLIMIT; 996 goto done; 997 } 998 } 999 done: 1000 return (reason); 1001 } 1002 1003 void 1004 pf_unlink_src_node(struct pf_ksrc_node *src) 1005 { 1006 PF_SRC_NODE_LOCK_ASSERT(src); 1007 1008 LIST_REMOVE(src, entry); 1009 if (src->rule.ptr) 1010 counter_u64_add(src->rule.ptr->src_nodes, -1); 1011 } 1012 1013 u_int 1014 pf_free_src_nodes(struct pf_ksrc_node_list *head) 1015 { 1016 struct pf_ksrc_node *sn, *tmp; 1017 u_int count = 0; 1018 1019 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1020 pf_free_src_node(sn); 1021 count++; 1022 } 1023 1024 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1025 1026 return (count); 1027 } 1028 1029 void 1030 pf_mtag_initialize(void) 1031 { 1032 1033 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1034 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1035 UMA_ALIGN_PTR, 0); 1036 } 1037 1038 /* Per-vnet data storage structures initialization. */ 1039 void 1040 pf_initialize(void) 1041 { 1042 struct pf_keyhash *kh; 1043 struct pf_idhash *ih; 1044 struct pf_srchash *sh; 1045 u_int i; 1046 1047 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 1048 pf_hashsize = PF_HASHSIZ; 1049 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 1050 pf_srchashsize = PF_SRCHASHSIZ; 1051 1052 V_pf_hashseed = arc4random(); 1053 1054 /* States and state keys storage. */ 1055 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1056 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1057 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1058 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1059 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1060 1061 V_pf_state_key_z = uma_zcreate("pf state keys", 1062 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1063 UMA_ALIGN_PTR, 0); 1064 1065 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 1066 M_PFHASH, M_NOWAIT | M_ZERO); 1067 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1068 M_PFHASH, M_NOWAIT | M_ZERO); 1069 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1070 printf("pf: Unable to allocate memory for " 1071 "state_hashsize %lu.\n", pf_hashsize); 1072 1073 free(V_pf_keyhash, M_PFHASH); 1074 free(V_pf_idhash, M_PFHASH); 1075 1076 pf_hashsize = PF_HASHSIZ; 1077 V_pf_keyhash = mallocarray(pf_hashsize, 1078 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1079 V_pf_idhash = mallocarray(pf_hashsize, 1080 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1081 } 1082 1083 pf_hashmask = pf_hashsize - 1; 1084 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1085 i++, kh++, ih++) { 1086 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1087 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1088 } 1089 1090 /* Source nodes. */ 1091 V_pf_sources_z = uma_zcreate("pf source nodes", 1092 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1093 0); 1094 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1095 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1096 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1097 1098 V_pf_srchash = mallocarray(pf_srchashsize, 1099 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1100 if (V_pf_srchash == NULL) { 1101 printf("pf: Unable to allocate memory for " 1102 "source_hashsize %lu.\n", pf_srchashsize); 1103 1104 pf_srchashsize = PF_SRCHASHSIZ; 1105 V_pf_srchash = mallocarray(pf_srchashsize, 1106 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1107 } 1108 1109 pf_srchashmask = pf_srchashsize - 1; 1110 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1111 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1112 1113 /* ALTQ */ 1114 TAILQ_INIT(&V_pf_altqs[0]); 1115 TAILQ_INIT(&V_pf_altqs[1]); 1116 TAILQ_INIT(&V_pf_altqs[2]); 1117 TAILQ_INIT(&V_pf_altqs[3]); 1118 TAILQ_INIT(&V_pf_pabuf); 1119 V_pf_altqs_active = &V_pf_altqs[0]; 1120 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1121 V_pf_altqs_inactive = &V_pf_altqs[2]; 1122 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1123 1124 /* Send & overload+flush queues. */ 1125 STAILQ_INIT(&V_pf_sendqueue); 1126 SLIST_INIT(&V_pf_overloadqueue); 1127 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1128 1129 /* Unlinked, but may be referenced rules. */ 1130 TAILQ_INIT(&V_pf_unlinked_rules); 1131 } 1132 1133 void 1134 pf_mtag_cleanup(void) 1135 { 1136 1137 uma_zdestroy(pf_mtag_z); 1138 } 1139 1140 void 1141 pf_cleanup(void) 1142 { 1143 struct pf_keyhash *kh; 1144 struct pf_idhash *ih; 1145 struct pf_srchash *sh; 1146 struct pf_send_entry *pfse, *next; 1147 u_int i; 1148 1149 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1150 i++, kh++, ih++) { 1151 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1152 __func__)); 1153 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1154 __func__)); 1155 mtx_destroy(&kh->lock); 1156 mtx_destroy(&ih->lock); 1157 } 1158 free(V_pf_keyhash, M_PFHASH); 1159 free(V_pf_idhash, M_PFHASH); 1160 1161 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1162 KASSERT(LIST_EMPTY(&sh->nodes), 1163 ("%s: source node hash not empty", __func__)); 1164 mtx_destroy(&sh->lock); 1165 } 1166 free(V_pf_srchash, M_PFHASH); 1167 1168 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1169 m_freem(pfse->pfse_m); 1170 free(pfse, M_PFTEMP); 1171 } 1172 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1173 1174 uma_zdestroy(V_pf_sources_z); 1175 uma_zdestroy(V_pf_state_z); 1176 uma_zdestroy(V_pf_state_key_z); 1177 } 1178 1179 static int 1180 pf_mtag_uminit(void *mem, int size, int how) 1181 { 1182 struct m_tag *t; 1183 1184 t = (struct m_tag *)mem; 1185 t->m_tag_cookie = MTAG_ABI_COMPAT; 1186 t->m_tag_id = PACKET_TAG_PF; 1187 t->m_tag_len = sizeof(struct pf_mtag); 1188 t->m_tag_free = pf_mtag_free; 1189 1190 return (0); 1191 } 1192 1193 static void 1194 pf_mtag_free(struct m_tag *t) 1195 { 1196 1197 uma_zfree(pf_mtag_z, t); 1198 } 1199 1200 struct pf_mtag * 1201 pf_get_mtag(struct mbuf *m) 1202 { 1203 struct m_tag *mtag; 1204 1205 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1206 return ((struct pf_mtag *)(mtag + 1)); 1207 1208 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1209 if (mtag == NULL) 1210 return (NULL); 1211 bzero(mtag + 1, sizeof(struct pf_mtag)); 1212 m_tag_prepend(m, mtag); 1213 1214 return ((struct pf_mtag *)(mtag + 1)); 1215 } 1216 1217 static int 1218 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1219 struct pf_kstate *s) 1220 { 1221 struct pf_keyhash *khs, *khw, *kh; 1222 struct pf_state_key *sk, *cur; 1223 struct pf_kstate *si, *olds = NULL; 1224 int idx; 1225 1226 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1227 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1228 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1229 1230 /* 1231 * We need to lock hash slots of both keys. To avoid deadlock 1232 * we always lock the slot with lower address first. Unlock order 1233 * isn't important. 1234 * 1235 * We also need to lock ID hash slot before dropping key 1236 * locks. On success we return with ID hash slot locked. 1237 */ 1238 1239 if (skw == sks) { 1240 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1241 PF_HASHROW_LOCK(khs); 1242 } else { 1243 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1244 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1245 if (khs == khw) { 1246 PF_HASHROW_LOCK(khs); 1247 } else if (khs < khw) { 1248 PF_HASHROW_LOCK(khs); 1249 PF_HASHROW_LOCK(khw); 1250 } else { 1251 PF_HASHROW_LOCK(khw); 1252 PF_HASHROW_LOCK(khs); 1253 } 1254 } 1255 1256 #define KEYS_UNLOCK() do { \ 1257 if (khs != khw) { \ 1258 PF_HASHROW_UNLOCK(khs); \ 1259 PF_HASHROW_UNLOCK(khw); \ 1260 } else \ 1261 PF_HASHROW_UNLOCK(khs); \ 1262 } while (0) 1263 1264 /* 1265 * First run: start with wire key. 1266 */ 1267 sk = skw; 1268 kh = khw; 1269 idx = PF_SK_WIRE; 1270 1271 MPASS(s->lock == NULL); 1272 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1273 1274 keyattach: 1275 LIST_FOREACH(cur, &kh->keys, entry) 1276 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1277 break; 1278 1279 if (cur != NULL) { 1280 /* Key exists. Check for same kif, if none, add to key. */ 1281 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1282 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1283 1284 PF_HASHROW_LOCK(ih); 1285 if (si->kif == s->kif && 1286 si->direction == s->direction) { 1287 if (sk->proto == IPPROTO_TCP && 1288 si->src.state >= TCPS_FIN_WAIT_2 && 1289 si->dst.state >= TCPS_FIN_WAIT_2) { 1290 /* 1291 * New state matches an old >FIN_WAIT_2 1292 * state. We can't drop key hash locks, 1293 * thus we can't unlink it properly. 1294 * 1295 * As a workaround we drop it into 1296 * TCPS_CLOSED state, schedule purge 1297 * ASAP and push it into the very end 1298 * of the slot TAILQ, so that it won't 1299 * conflict with our new state. 1300 */ 1301 pf_set_protostate(si, PF_PEER_BOTH, 1302 TCPS_CLOSED); 1303 si->timeout = PFTM_PURGE; 1304 olds = si; 1305 } else { 1306 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1307 printf("pf: %s key attach " 1308 "failed on %s: ", 1309 (idx == PF_SK_WIRE) ? 1310 "wire" : "stack", 1311 s->kif->pfik_name); 1312 pf_print_state_parts(s, 1313 (idx == PF_SK_WIRE) ? 1314 sk : NULL, 1315 (idx == PF_SK_STACK) ? 1316 sk : NULL); 1317 printf(", existing: "); 1318 pf_print_state_parts(si, 1319 (idx == PF_SK_WIRE) ? 1320 sk : NULL, 1321 (idx == PF_SK_STACK) ? 1322 sk : NULL); 1323 printf("\n"); 1324 } 1325 PF_HASHROW_UNLOCK(ih); 1326 KEYS_UNLOCK(); 1327 uma_zfree(V_pf_state_key_z, sk); 1328 if (idx == PF_SK_STACK) 1329 pf_detach_state(s); 1330 return (EEXIST); /* collision! */ 1331 } 1332 } 1333 PF_HASHROW_UNLOCK(ih); 1334 } 1335 uma_zfree(V_pf_state_key_z, sk); 1336 s->key[idx] = cur; 1337 } else { 1338 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1339 s->key[idx] = sk; 1340 } 1341 1342 stateattach: 1343 /* List is sorted, if-bound states before floating. */ 1344 if (s->kif == V_pfi_all) 1345 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1346 else 1347 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1348 1349 if (olds) { 1350 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1351 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1352 key_list[idx]); 1353 olds = NULL; 1354 } 1355 1356 /* 1357 * Attach done. See how should we (or should not?) 1358 * attach a second key. 1359 */ 1360 if (sks == skw) { 1361 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1362 idx = PF_SK_STACK; 1363 sks = NULL; 1364 goto stateattach; 1365 } else if (sks != NULL) { 1366 /* 1367 * Continue attaching with stack key. 1368 */ 1369 sk = sks; 1370 kh = khs; 1371 idx = PF_SK_STACK; 1372 sks = NULL; 1373 goto keyattach; 1374 } 1375 1376 PF_STATE_LOCK(s); 1377 KEYS_UNLOCK(); 1378 1379 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1380 ("%s failure", __func__)); 1381 1382 return (0); 1383 #undef KEYS_UNLOCK 1384 } 1385 1386 static void 1387 pf_detach_state(struct pf_kstate *s) 1388 { 1389 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1390 struct pf_keyhash *kh; 1391 1392 pf_sctp_multihome_detach_addr(s); 1393 1394 if (sks != NULL) { 1395 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1396 PF_HASHROW_LOCK(kh); 1397 if (s->key[PF_SK_STACK] != NULL) 1398 pf_state_key_detach(s, PF_SK_STACK); 1399 /* 1400 * If both point to same key, then we are done. 1401 */ 1402 if (sks == s->key[PF_SK_WIRE]) { 1403 pf_state_key_detach(s, PF_SK_WIRE); 1404 PF_HASHROW_UNLOCK(kh); 1405 return; 1406 } 1407 PF_HASHROW_UNLOCK(kh); 1408 } 1409 1410 if (s->key[PF_SK_WIRE] != NULL) { 1411 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1412 PF_HASHROW_LOCK(kh); 1413 if (s->key[PF_SK_WIRE] != NULL) 1414 pf_state_key_detach(s, PF_SK_WIRE); 1415 PF_HASHROW_UNLOCK(kh); 1416 } 1417 } 1418 1419 static void 1420 pf_state_key_detach(struct pf_kstate *s, int idx) 1421 { 1422 struct pf_state_key *sk = s->key[idx]; 1423 #ifdef INVARIANTS 1424 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1425 1426 PF_HASHROW_ASSERT(kh); 1427 #endif 1428 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1429 s->key[idx] = NULL; 1430 1431 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1432 LIST_REMOVE(sk, entry); 1433 uma_zfree(V_pf_state_key_z, sk); 1434 } 1435 } 1436 1437 static int 1438 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1439 { 1440 struct pf_state_key *sk = mem; 1441 1442 bzero(sk, sizeof(struct pf_state_key_cmp)); 1443 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1444 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1445 1446 return (0); 1447 } 1448 1449 struct pf_state_key * 1450 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1451 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1452 { 1453 struct pf_state_key *sk; 1454 1455 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1456 if (sk == NULL) 1457 return (NULL); 1458 1459 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1460 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1461 sk->port[pd->sidx] = sport; 1462 sk->port[pd->didx] = dport; 1463 sk->proto = pd->proto; 1464 sk->af = pd->af; 1465 1466 return (sk); 1467 } 1468 1469 struct pf_state_key * 1470 pf_state_key_clone(struct pf_state_key *orig) 1471 { 1472 struct pf_state_key *sk; 1473 1474 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1475 if (sk == NULL) 1476 return (NULL); 1477 1478 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1479 1480 return (sk); 1481 } 1482 1483 int 1484 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1485 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1486 { 1487 struct pf_idhash *ih; 1488 struct pf_kstate *cur; 1489 int error; 1490 1491 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1492 ("%s: sks not pristine", __func__)); 1493 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1494 ("%s: skw not pristine", __func__)); 1495 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1496 1497 s->kif = kif; 1498 s->orig_kif = orig_kif; 1499 1500 if (s->id == 0 && s->creatorid == 0) { 1501 s->id = alloc_unr64(&V_pf_stateid); 1502 s->id = htobe64(s->id); 1503 s->creatorid = V_pf_status.hostid; 1504 } 1505 1506 /* Returns with ID locked on success. */ 1507 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1508 return (error); 1509 1510 ih = &V_pf_idhash[PF_IDHASH(s)]; 1511 PF_HASHROW_ASSERT(ih); 1512 LIST_FOREACH(cur, &ih->states, entry) 1513 if (cur->id == s->id && cur->creatorid == s->creatorid) 1514 break; 1515 1516 if (cur != NULL) { 1517 PF_HASHROW_UNLOCK(ih); 1518 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1519 printf("pf: state ID collision: " 1520 "id: %016llx creatorid: %08x\n", 1521 (unsigned long long)be64toh(s->id), 1522 ntohl(s->creatorid)); 1523 } 1524 pf_detach_state(s); 1525 return (EEXIST); 1526 } 1527 LIST_INSERT_HEAD(&ih->states, s, entry); 1528 /* One for keys, one for ID hash. */ 1529 refcount_init(&s->refs, 2); 1530 1531 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1532 if (V_pfsync_insert_state_ptr != NULL) 1533 V_pfsync_insert_state_ptr(s); 1534 1535 /* Returns locked. */ 1536 return (0); 1537 } 1538 1539 /* 1540 * Find state by ID: returns with locked row on success. 1541 */ 1542 struct pf_kstate * 1543 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1544 { 1545 struct pf_idhash *ih; 1546 struct pf_kstate *s; 1547 1548 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1549 1550 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1551 1552 PF_HASHROW_LOCK(ih); 1553 LIST_FOREACH(s, &ih->states, entry) 1554 if (s->id == id && s->creatorid == creatorid) 1555 break; 1556 1557 if (s == NULL) 1558 PF_HASHROW_UNLOCK(ih); 1559 1560 return (s); 1561 } 1562 1563 /* 1564 * Find state by key. 1565 * Returns with ID hash slot locked on success. 1566 */ 1567 static struct pf_kstate * 1568 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1569 { 1570 struct pf_keyhash *kh; 1571 struct pf_state_key *sk; 1572 struct pf_kstate *s; 1573 int idx; 1574 1575 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1576 1577 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1578 1579 PF_HASHROW_LOCK(kh); 1580 LIST_FOREACH(sk, &kh->keys, entry) 1581 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1582 break; 1583 if (sk == NULL) { 1584 PF_HASHROW_UNLOCK(kh); 1585 return (NULL); 1586 } 1587 1588 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1589 1590 /* List is sorted, if-bound states before floating ones. */ 1591 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1592 if (s->kif == V_pfi_all || s->kif == kif) { 1593 PF_STATE_LOCK(s); 1594 PF_HASHROW_UNLOCK(kh); 1595 if (__predict_false(s->timeout >= PFTM_MAX)) { 1596 /* 1597 * State is either being processed by 1598 * pf_unlink_state() in an other thread, or 1599 * is scheduled for immediate expiry. 1600 */ 1601 PF_STATE_UNLOCK(s); 1602 return (NULL); 1603 } 1604 return (s); 1605 } 1606 PF_HASHROW_UNLOCK(kh); 1607 1608 return (NULL); 1609 } 1610 1611 /* 1612 * Returns with ID hash slot locked on success. 1613 */ 1614 struct pf_kstate * 1615 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1616 { 1617 struct pf_keyhash *kh; 1618 struct pf_state_key *sk; 1619 struct pf_kstate *s, *ret = NULL; 1620 int idx, inout = 0; 1621 1622 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1623 1624 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1625 1626 PF_HASHROW_LOCK(kh); 1627 LIST_FOREACH(sk, &kh->keys, entry) 1628 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1629 break; 1630 if (sk == NULL) { 1631 PF_HASHROW_UNLOCK(kh); 1632 return (NULL); 1633 } 1634 switch (dir) { 1635 case PF_IN: 1636 idx = PF_SK_WIRE; 1637 break; 1638 case PF_OUT: 1639 idx = PF_SK_STACK; 1640 break; 1641 case PF_INOUT: 1642 idx = PF_SK_WIRE; 1643 inout = 1; 1644 break; 1645 default: 1646 panic("%s: dir %u", __func__, dir); 1647 } 1648 second_run: 1649 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1650 if (more == NULL) { 1651 PF_STATE_LOCK(s); 1652 PF_HASHROW_UNLOCK(kh); 1653 return (s); 1654 } 1655 1656 if (ret) 1657 (*more)++; 1658 else { 1659 ret = s; 1660 PF_STATE_LOCK(s); 1661 } 1662 } 1663 if (inout == 1) { 1664 inout = 0; 1665 idx = PF_SK_STACK; 1666 goto second_run; 1667 } 1668 PF_HASHROW_UNLOCK(kh); 1669 1670 return (ret); 1671 } 1672 1673 /* 1674 * FIXME 1675 * This routine is inefficient -- locks the state only to unlock immediately on 1676 * return. 1677 * It is racy -- after the state is unlocked nothing stops other threads from 1678 * removing it. 1679 */ 1680 bool 1681 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1682 { 1683 struct pf_kstate *s; 1684 1685 s = pf_find_state_all(key, dir, NULL); 1686 if (s != NULL) { 1687 PF_STATE_UNLOCK(s); 1688 return (true); 1689 } 1690 return (false); 1691 } 1692 1693 /* END state table stuff */ 1694 1695 static void 1696 pf_send(struct pf_send_entry *pfse) 1697 { 1698 1699 PF_SENDQ_LOCK(); 1700 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1701 PF_SENDQ_UNLOCK(); 1702 swi_sched(V_pf_swi_cookie, 0); 1703 } 1704 1705 static bool 1706 pf_isforlocal(struct mbuf *m, int af) 1707 { 1708 switch (af) { 1709 #ifdef INET 1710 case AF_INET: { 1711 struct ip *ip = mtod(m, struct ip *); 1712 1713 return (in_localip(ip->ip_dst)); 1714 } 1715 #endif 1716 #ifdef INET6 1717 case AF_INET6: { 1718 struct ip6_hdr *ip6; 1719 struct in6_ifaddr *ia; 1720 ip6 = mtod(m, struct ip6_hdr *); 1721 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1722 if (ia == NULL) 1723 return (false); 1724 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1725 } 1726 #endif 1727 default: 1728 panic("Unsupported af %d", af); 1729 } 1730 1731 return (false); 1732 } 1733 1734 void 1735 pf_intr(void *v) 1736 { 1737 struct epoch_tracker et; 1738 struct pf_send_head queue; 1739 struct pf_send_entry *pfse, *next; 1740 1741 CURVNET_SET((struct vnet *)v); 1742 1743 PF_SENDQ_LOCK(); 1744 queue = V_pf_sendqueue; 1745 STAILQ_INIT(&V_pf_sendqueue); 1746 PF_SENDQ_UNLOCK(); 1747 1748 NET_EPOCH_ENTER(et); 1749 1750 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1751 switch (pfse->pfse_type) { 1752 #ifdef INET 1753 case PFSE_IP: { 1754 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1755 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1756 pfse->pfse_m->m_pkthdr.csum_flags |= 1757 CSUM_IP_VALID | CSUM_IP_CHECKED; 1758 ip_input(pfse->pfse_m); 1759 } else { 1760 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1761 NULL); 1762 } 1763 break; 1764 } 1765 case PFSE_ICMP: 1766 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1767 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1768 break; 1769 #endif /* INET */ 1770 #ifdef INET6 1771 case PFSE_IP6: 1772 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1773 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1774 ip6_input(pfse->pfse_m); 1775 } else { 1776 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1777 NULL, NULL); 1778 } 1779 break; 1780 case PFSE_ICMP6: 1781 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1782 pfse->icmpopts.code, pfse->icmpopts.mtu); 1783 break; 1784 #endif /* INET6 */ 1785 default: 1786 panic("%s: unknown type", __func__); 1787 } 1788 free(pfse, M_PFTEMP); 1789 } 1790 NET_EPOCH_EXIT(et); 1791 CURVNET_RESTORE(); 1792 } 1793 1794 #define pf_purge_thread_period (hz / 10) 1795 1796 #ifdef PF_WANT_32_TO_64_COUNTER 1797 static void 1798 pf_status_counter_u64_periodic(void) 1799 { 1800 1801 PF_RULES_RASSERT(); 1802 1803 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1804 return; 1805 } 1806 1807 for (int i = 0; i < FCNT_MAX; i++) { 1808 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1809 } 1810 } 1811 1812 static void 1813 pf_kif_counter_u64_periodic(void) 1814 { 1815 struct pfi_kkif *kif; 1816 size_t r, run; 1817 1818 PF_RULES_RASSERT(); 1819 1820 if (__predict_false(V_pf_allkifcount == 0)) { 1821 return; 1822 } 1823 1824 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1825 return; 1826 } 1827 1828 run = V_pf_allkifcount / 10; 1829 if (run < 5) 1830 run = 5; 1831 1832 for (r = 0; r < run; r++) { 1833 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1834 if (kif == NULL) { 1835 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1836 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1837 break; 1838 } 1839 1840 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1841 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1842 1843 for (int i = 0; i < 2; i++) { 1844 for (int j = 0; j < 2; j++) { 1845 for (int k = 0; k < 2; k++) { 1846 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1847 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1848 } 1849 } 1850 } 1851 } 1852 } 1853 1854 static void 1855 pf_rule_counter_u64_periodic(void) 1856 { 1857 struct pf_krule *rule; 1858 size_t r, run; 1859 1860 PF_RULES_RASSERT(); 1861 1862 if (__predict_false(V_pf_allrulecount == 0)) { 1863 return; 1864 } 1865 1866 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1867 return; 1868 } 1869 1870 run = V_pf_allrulecount / 10; 1871 if (run < 5) 1872 run = 5; 1873 1874 for (r = 0; r < run; r++) { 1875 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1876 if (rule == NULL) { 1877 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1878 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1879 break; 1880 } 1881 1882 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1883 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1884 1885 pf_counter_u64_periodic(&rule->evaluations); 1886 for (int i = 0; i < 2; i++) { 1887 pf_counter_u64_periodic(&rule->packets[i]); 1888 pf_counter_u64_periodic(&rule->bytes[i]); 1889 } 1890 } 1891 } 1892 1893 static void 1894 pf_counter_u64_periodic_main(void) 1895 { 1896 PF_RULES_RLOCK_TRACKER; 1897 1898 V_pf_counter_periodic_iter++; 1899 1900 PF_RULES_RLOCK(); 1901 pf_counter_u64_critical_enter(); 1902 pf_status_counter_u64_periodic(); 1903 pf_kif_counter_u64_periodic(); 1904 pf_rule_counter_u64_periodic(); 1905 pf_counter_u64_critical_exit(); 1906 PF_RULES_RUNLOCK(); 1907 } 1908 #else 1909 #define pf_counter_u64_periodic_main() do { } while (0) 1910 #endif 1911 1912 void 1913 pf_purge_thread(void *unused __unused) 1914 { 1915 VNET_ITERATOR_DECL(vnet_iter); 1916 1917 sx_xlock(&pf_end_lock); 1918 while (pf_end_threads == 0) { 1919 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1920 1921 VNET_LIST_RLOCK(); 1922 VNET_FOREACH(vnet_iter) { 1923 CURVNET_SET(vnet_iter); 1924 1925 /* Wait until V_pf_default_rule is initialized. */ 1926 if (V_pf_vnet_active == 0) { 1927 CURVNET_RESTORE(); 1928 continue; 1929 } 1930 1931 pf_counter_u64_periodic_main(); 1932 1933 /* 1934 * Process 1/interval fraction of the state 1935 * table every run. 1936 */ 1937 V_pf_purge_idx = 1938 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1939 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1940 1941 /* 1942 * Purge other expired types every 1943 * PFTM_INTERVAL seconds. 1944 */ 1945 if (V_pf_purge_idx == 0) { 1946 /* 1947 * Order is important: 1948 * - states and src nodes reference rules 1949 * - states and rules reference kifs 1950 */ 1951 pf_purge_expired_fragments(); 1952 pf_purge_expired_src_nodes(); 1953 pf_purge_unlinked_rules(); 1954 pfi_kkif_purge(); 1955 } 1956 CURVNET_RESTORE(); 1957 } 1958 VNET_LIST_RUNLOCK(); 1959 } 1960 1961 pf_end_threads++; 1962 sx_xunlock(&pf_end_lock); 1963 kproc_exit(0); 1964 } 1965 1966 void 1967 pf_unload_vnet_purge(void) 1968 { 1969 1970 /* 1971 * To cleanse up all kifs and rules we need 1972 * two runs: first one clears reference flags, 1973 * then pf_purge_expired_states() doesn't 1974 * raise them, and then second run frees. 1975 */ 1976 pf_purge_unlinked_rules(); 1977 pfi_kkif_purge(); 1978 1979 /* 1980 * Now purge everything. 1981 */ 1982 pf_purge_expired_states(0, pf_hashmask); 1983 pf_purge_fragments(UINT_MAX); 1984 pf_purge_expired_src_nodes(); 1985 1986 /* 1987 * Now all kifs & rules should be unreferenced, 1988 * thus should be successfully freed. 1989 */ 1990 pf_purge_unlinked_rules(); 1991 pfi_kkif_purge(); 1992 } 1993 1994 u_int32_t 1995 pf_state_expires(const struct pf_kstate *state) 1996 { 1997 u_int32_t timeout; 1998 u_int32_t start; 1999 u_int32_t end; 2000 u_int32_t states; 2001 2002 /* handle all PFTM_* > PFTM_MAX here */ 2003 if (state->timeout == PFTM_PURGE) 2004 return (time_uptime); 2005 KASSERT(state->timeout != PFTM_UNLINKED, 2006 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2007 KASSERT((state->timeout < PFTM_MAX), 2008 ("pf_state_expires: timeout > PFTM_MAX")); 2009 timeout = state->rule.ptr->timeout[state->timeout]; 2010 if (!timeout) 2011 timeout = V_pf_default_rule.timeout[state->timeout]; 2012 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2013 if (start && state->rule.ptr != &V_pf_default_rule) { 2014 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2015 states = counter_u64_fetch(state->rule.ptr->states_cur); 2016 } else { 2017 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2018 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2019 states = V_pf_status.states; 2020 } 2021 if (end && states > start && start < end) { 2022 if (states < end) { 2023 timeout = (u_int64_t)timeout * (end - states) / 2024 (end - start); 2025 return (state->expire + timeout); 2026 } 2027 else 2028 return (time_uptime); 2029 } 2030 return (state->expire + timeout); 2031 } 2032 2033 void 2034 pf_purge_expired_src_nodes(void) 2035 { 2036 struct pf_ksrc_node_list freelist; 2037 struct pf_srchash *sh; 2038 struct pf_ksrc_node *cur, *next; 2039 int i; 2040 2041 LIST_INIT(&freelist); 2042 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 2043 PF_HASHROW_LOCK(sh); 2044 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2045 if (cur->states == 0 && cur->expire <= time_uptime) { 2046 pf_unlink_src_node(cur); 2047 LIST_INSERT_HEAD(&freelist, cur, entry); 2048 } else if (cur->rule.ptr != NULL) 2049 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2050 PF_HASHROW_UNLOCK(sh); 2051 } 2052 2053 pf_free_src_nodes(&freelist); 2054 2055 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2056 } 2057 2058 static void 2059 pf_src_tree_remove_state(struct pf_kstate *s) 2060 { 2061 struct pf_ksrc_node *sn; 2062 uint32_t timeout; 2063 2064 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2065 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2066 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2067 2068 if (s->src_node != NULL) { 2069 sn = s->src_node; 2070 PF_SRC_NODE_LOCK(sn); 2071 if (s->src.tcp_est) 2072 --sn->conn; 2073 if (--sn->states == 0) 2074 sn->expire = time_uptime + timeout; 2075 PF_SRC_NODE_UNLOCK(sn); 2076 } 2077 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2078 sn = s->nat_src_node; 2079 PF_SRC_NODE_LOCK(sn); 2080 if (--sn->states == 0) 2081 sn->expire = time_uptime + timeout; 2082 PF_SRC_NODE_UNLOCK(sn); 2083 } 2084 s->src_node = s->nat_src_node = NULL; 2085 } 2086 2087 /* 2088 * Unlink and potentilly free a state. Function may be 2089 * called with ID hash row locked, but always returns 2090 * unlocked, since it needs to go through key hash locking. 2091 */ 2092 int 2093 pf_unlink_state(struct pf_kstate *s) 2094 { 2095 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2096 2097 PF_HASHROW_ASSERT(ih); 2098 2099 if (s->timeout == PFTM_UNLINKED) { 2100 /* 2101 * State is being processed 2102 * by pf_unlink_state() in 2103 * an other thread. 2104 */ 2105 PF_HASHROW_UNLOCK(ih); 2106 return (0); /* XXXGL: undefined actually */ 2107 } 2108 2109 if (s->src.state == PF_TCPS_PROXY_DST) { 2110 /* XXX wire key the right one? */ 2111 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2112 &s->key[PF_SK_WIRE]->addr[1], 2113 &s->key[PF_SK_WIRE]->addr[0], 2114 s->key[PF_SK_WIRE]->port[1], 2115 s->key[PF_SK_WIRE]->port[0], 2116 s->src.seqhi, s->src.seqlo + 1, 2117 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2118 } 2119 2120 LIST_REMOVE(s, entry); 2121 pf_src_tree_remove_state(s); 2122 2123 if (V_pfsync_delete_state_ptr != NULL) 2124 V_pfsync_delete_state_ptr(s); 2125 2126 STATE_DEC_COUNTERS(s); 2127 2128 s->timeout = PFTM_UNLINKED; 2129 2130 /* Ensure we remove it from the list of halfopen states, if needed. */ 2131 if (s->key[PF_SK_STACK] != NULL && 2132 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2133 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2134 2135 PF_HASHROW_UNLOCK(ih); 2136 2137 pf_detach_state(s); 2138 /* pf_state_insert() initialises refs to 2 */ 2139 return (pf_release_staten(s, 2)); 2140 } 2141 2142 struct pf_kstate * 2143 pf_alloc_state(int flags) 2144 { 2145 2146 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2147 } 2148 2149 void 2150 pf_free_state(struct pf_kstate *cur) 2151 { 2152 struct pf_krule_item *ri; 2153 2154 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2155 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2156 cur->timeout)); 2157 2158 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2159 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2160 free(ri, M_PF_RULE_ITEM); 2161 } 2162 2163 pf_normalize_tcp_cleanup(cur); 2164 uma_zfree(V_pf_state_z, cur); 2165 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2166 } 2167 2168 /* 2169 * Called only from pf_purge_thread(), thus serialized. 2170 */ 2171 static u_int 2172 pf_purge_expired_states(u_int i, int maxcheck) 2173 { 2174 struct pf_idhash *ih; 2175 struct pf_kstate *s; 2176 struct pf_krule_item *mrm; 2177 2178 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2179 2180 /* 2181 * Go through hash and unlink states that expire now. 2182 */ 2183 while (maxcheck > 0) { 2184 ih = &V_pf_idhash[i]; 2185 2186 /* only take the lock if we expect to do work */ 2187 if (!LIST_EMPTY(&ih->states)) { 2188 relock: 2189 PF_HASHROW_LOCK(ih); 2190 LIST_FOREACH(s, &ih->states, entry) { 2191 if (pf_state_expires(s) <= time_uptime) { 2192 V_pf_status.states -= 2193 pf_unlink_state(s); 2194 goto relock; 2195 } 2196 s->rule.ptr->rule_ref |= PFRULE_REFS; 2197 if (s->nat_rule.ptr != NULL) 2198 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2199 if (s->anchor.ptr != NULL) 2200 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2201 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2202 SLIST_FOREACH(mrm, &s->match_rules, entry) 2203 mrm->r->rule_ref |= PFRULE_REFS; 2204 if (s->rt_kif) 2205 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2206 } 2207 PF_HASHROW_UNLOCK(ih); 2208 } 2209 2210 /* Return when we hit end of hash. */ 2211 if (++i > pf_hashmask) { 2212 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2213 return (0); 2214 } 2215 2216 maxcheck--; 2217 } 2218 2219 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2220 2221 return (i); 2222 } 2223 2224 static void 2225 pf_purge_unlinked_rules(void) 2226 { 2227 struct pf_krulequeue tmpq; 2228 struct pf_krule *r, *r1; 2229 2230 /* 2231 * If we have overloading task pending, then we'd 2232 * better skip purging this time. There is a tiny 2233 * probability that overloading task references 2234 * an already unlinked rule. 2235 */ 2236 PF_OVERLOADQ_LOCK(); 2237 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2238 PF_OVERLOADQ_UNLOCK(); 2239 return; 2240 } 2241 PF_OVERLOADQ_UNLOCK(); 2242 2243 /* 2244 * Do naive mark-and-sweep garbage collecting of old rules. 2245 * Reference flag is raised by pf_purge_expired_states() 2246 * and pf_purge_expired_src_nodes(). 2247 * 2248 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2249 * use a temporary queue. 2250 */ 2251 TAILQ_INIT(&tmpq); 2252 PF_UNLNKDRULES_LOCK(); 2253 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2254 if (!(r->rule_ref & PFRULE_REFS)) { 2255 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2256 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2257 } else 2258 r->rule_ref &= ~PFRULE_REFS; 2259 } 2260 PF_UNLNKDRULES_UNLOCK(); 2261 2262 if (!TAILQ_EMPTY(&tmpq)) { 2263 PF_CONFIG_LOCK(); 2264 PF_RULES_WLOCK(); 2265 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2266 TAILQ_REMOVE(&tmpq, r, entries); 2267 pf_free_rule(r); 2268 } 2269 PF_RULES_WUNLOCK(); 2270 PF_CONFIG_UNLOCK(); 2271 } 2272 } 2273 2274 void 2275 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2276 { 2277 switch (af) { 2278 #ifdef INET 2279 case AF_INET: { 2280 u_int32_t a = ntohl(addr->addr32[0]); 2281 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2282 (a>>8)&255, a&255); 2283 if (p) { 2284 p = ntohs(p); 2285 printf(":%u", p); 2286 } 2287 break; 2288 } 2289 #endif /* INET */ 2290 #ifdef INET6 2291 case AF_INET6: { 2292 u_int16_t b; 2293 u_int8_t i, curstart, curend, maxstart, maxend; 2294 curstart = curend = maxstart = maxend = 255; 2295 for (i = 0; i < 8; i++) { 2296 if (!addr->addr16[i]) { 2297 if (curstart == 255) 2298 curstart = i; 2299 curend = i; 2300 } else { 2301 if ((curend - curstart) > 2302 (maxend - maxstart)) { 2303 maxstart = curstart; 2304 maxend = curend; 2305 } 2306 curstart = curend = 255; 2307 } 2308 } 2309 if ((curend - curstart) > 2310 (maxend - maxstart)) { 2311 maxstart = curstart; 2312 maxend = curend; 2313 } 2314 for (i = 0; i < 8; i++) { 2315 if (i >= maxstart && i <= maxend) { 2316 if (i == 0) 2317 printf(":"); 2318 if (i == maxend) 2319 printf(":"); 2320 } else { 2321 b = ntohs(addr->addr16[i]); 2322 printf("%x", b); 2323 if (i < 7) 2324 printf(":"); 2325 } 2326 } 2327 if (p) { 2328 p = ntohs(p); 2329 printf("[%u]", p); 2330 } 2331 break; 2332 } 2333 #endif /* INET6 */ 2334 } 2335 } 2336 2337 void 2338 pf_print_state(struct pf_kstate *s) 2339 { 2340 pf_print_state_parts(s, NULL, NULL); 2341 } 2342 2343 static void 2344 pf_print_state_parts(struct pf_kstate *s, 2345 struct pf_state_key *skwp, struct pf_state_key *sksp) 2346 { 2347 struct pf_state_key *skw, *sks; 2348 u_int8_t proto, dir; 2349 2350 /* Do our best to fill these, but they're skipped if NULL */ 2351 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2352 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2353 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2354 dir = s ? s->direction : 0; 2355 2356 switch (proto) { 2357 case IPPROTO_IPV4: 2358 printf("IPv4"); 2359 break; 2360 case IPPROTO_IPV6: 2361 printf("IPv6"); 2362 break; 2363 case IPPROTO_TCP: 2364 printf("TCP"); 2365 break; 2366 case IPPROTO_UDP: 2367 printf("UDP"); 2368 break; 2369 case IPPROTO_ICMP: 2370 printf("ICMP"); 2371 break; 2372 case IPPROTO_ICMPV6: 2373 printf("ICMPv6"); 2374 break; 2375 default: 2376 printf("%u", proto); 2377 break; 2378 } 2379 switch (dir) { 2380 case PF_IN: 2381 printf(" in"); 2382 break; 2383 case PF_OUT: 2384 printf(" out"); 2385 break; 2386 } 2387 if (skw) { 2388 printf(" wire: "); 2389 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2390 printf(" "); 2391 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2392 } 2393 if (sks) { 2394 printf(" stack: "); 2395 if (sks != skw) { 2396 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2397 printf(" "); 2398 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2399 } else 2400 printf("-"); 2401 } 2402 if (s) { 2403 if (proto == IPPROTO_TCP) { 2404 printf(" [lo=%u high=%u win=%u modulator=%u", 2405 s->src.seqlo, s->src.seqhi, 2406 s->src.max_win, s->src.seqdiff); 2407 if (s->src.wscale && s->dst.wscale) 2408 printf(" wscale=%u", 2409 s->src.wscale & PF_WSCALE_MASK); 2410 printf("]"); 2411 printf(" [lo=%u high=%u win=%u modulator=%u", 2412 s->dst.seqlo, s->dst.seqhi, 2413 s->dst.max_win, s->dst.seqdiff); 2414 if (s->src.wscale && s->dst.wscale) 2415 printf(" wscale=%u", 2416 s->dst.wscale & PF_WSCALE_MASK); 2417 printf("]"); 2418 } 2419 printf(" %u:%u", s->src.state, s->dst.state); 2420 } 2421 } 2422 2423 void 2424 pf_print_flags(u_int8_t f) 2425 { 2426 if (f) 2427 printf(" "); 2428 if (f & TH_FIN) 2429 printf("F"); 2430 if (f & TH_SYN) 2431 printf("S"); 2432 if (f & TH_RST) 2433 printf("R"); 2434 if (f & TH_PUSH) 2435 printf("P"); 2436 if (f & TH_ACK) 2437 printf("A"); 2438 if (f & TH_URG) 2439 printf("U"); 2440 if (f & TH_ECE) 2441 printf("E"); 2442 if (f & TH_CWR) 2443 printf("W"); 2444 } 2445 2446 #define PF_SET_SKIP_STEPS(i) \ 2447 do { \ 2448 while (head[i] != cur) { \ 2449 head[i]->skip[i].ptr = cur; \ 2450 head[i] = TAILQ_NEXT(head[i], entries); \ 2451 } \ 2452 } while (0) 2453 2454 void 2455 pf_calc_skip_steps(struct pf_krulequeue *rules) 2456 { 2457 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2458 int i; 2459 2460 cur = TAILQ_FIRST(rules); 2461 prev = cur; 2462 for (i = 0; i < PF_SKIP_COUNT; ++i) 2463 head[i] = cur; 2464 while (cur != NULL) { 2465 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2466 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2467 if (cur->direction != prev->direction) 2468 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2469 if (cur->af != prev->af) 2470 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2471 if (cur->proto != prev->proto) 2472 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2473 if (cur->src.neg != prev->src.neg || 2474 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2475 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2476 if (cur->src.port[0] != prev->src.port[0] || 2477 cur->src.port[1] != prev->src.port[1] || 2478 cur->src.port_op != prev->src.port_op) 2479 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2480 if (cur->dst.neg != prev->dst.neg || 2481 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2482 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2483 if (cur->dst.port[0] != prev->dst.port[0] || 2484 cur->dst.port[1] != prev->dst.port[1] || 2485 cur->dst.port_op != prev->dst.port_op) 2486 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2487 2488 prev = cur; 2489 cur = TAILQ_NEXT(cur, entries); 2490 } 2491 for (i = 0; i < PF_SKIP_COUNT; ++i) 2492 PF_SET_SKIP_STEPS(i); 2493 } 2494 2495 int 2496 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2497 { 2498 if (aw1->type != aw2->type) 2499 return (1); 2500 switch (aw1->type) { 2501 case PF_ADDR_ADDRMASK: 2502 case PF_ADDR_RANGE: 2503 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2504 return (1); 2505 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2506 return (1); 2507 return (0); 2508 case PF_ADDR_DYNIFTL: 2509 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2510 case PF_ADDR_NOROUTE: 2511 case PF_ADDR_URPFFAILED: 2512 return (0); 2513 case PF_ADDR_TABLE: 2514 return (aw1->p.tbl != aw2->p.tbl); 2515 default: 2516 printf("invalid address type: %d\n", aw1->type); 2517 return (1); 2518 } 2519 } 2520 2521 /** 2522 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2523 * header isn't always a full checksum. In some cases (i.e. output) it's a 2524 * pseudo-header checksum, which is a partial checksum over src/dst IP 2525 * addresses, protocol number and length. 2526 * 2527 * That means we have the following cases: 2528 * * Input or forwarding: we don't have TSO, the checksum fields are full 2529 * checksums, we need to update the checksum whenever we change anything. 2530 * * Output (i.e. the checksum is a pseudo-header checksum): 2531 * x The field being updated is src/dst address or affects the length of 2532 * the packet. We need to update the pseudo-header checksum (note that this 2533 * checksum is not ones' complement). 2534 * x Some other field is being modified (e.g. src/dst port numbers): We 2535 * don't have to update anything. 2536 **/ 2537 u_int16_t 2538 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2539 { 2540 u_int32_t x; 2541 2542 x = cksum + old - new; 2543 x = (x + (x >> 16)) & 0xffff; 2544 2545 /* optimise: eliminate a branch when not udp */ 2546 if (udp && cksum == 0x0000) 2547 return cksum; 2548 if (udp && x == 0x0000) 2549 x = 0xffff; 2550 2551 return (u_int16_t)(x); 2552 } 2553 2554 static void 2555 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2556 u_int8_t udp) 2557 { 2558 u_int16_t old = htons(hi ? (*f << 8) : *f); 2559 u_int16_t new = htons(hi ? ( v << 8) : v); 2560 2561 if (*f == v) 2562 return; 2563 2564 *f = v; 2565 2566 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2567 return; 2568 2569 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2570 } 2571 2572 void 2573 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2574 bool hi, u_int8_t udp) 2575 { 2576 u_int8_t *fb = (u_int8_t *)f; 2577 u_int8_t *vb = (u_int8_t *)&v; 2578 2579 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2580 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2581 } 2582 2583 void 2584 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2585 bool hi, u_int8_t udp) 2586 { 2587 u_int8_t *fb = (u_int8_t *)f; 2588 u_int8_t *vb = (u_int8_t *)&v; 2589 2590 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2591 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2592 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2593 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2594 } 2595 2596 u_int16_t 2597 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2598 u_int16_t new, u_int8_t udp) 2599 { 2600 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2601 return (cksum); 2602 2603 return (pf_cksum_fixup(cksum, old, new, udp)); 2604 } 2605 2606 static void 2607 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2608 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2609 sa_family_t af) 2610 { 2611 struct pf_addr ao; 2612 u_int16_t po = *p; 2613 2614 PF_ACPY(&ao, a, af); 2615 PF_ACPY(a, an, af); 2616 2617 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2618 *pc = ~*pc; 2619 2620 *p = pn; 2621 2622 switch (af) { 2623 #ifdef INET 2624 case AF_INET: 2625 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2626 ao.addr16[0], an->addr16[0], 0), 2627 ao.addr16[1], an->addr16[1], 0); 2628 *p = pn; 2629 2630 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2631 ao.addr16[0], an->addr16[0], u), 2632 ao.addr16[1], an->addr16[1], u); 2633 2634 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2635 break; 2636 #endif /* INET */ 2637 #ifdef INET6 2638 case AF_INET6: 2639 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2640 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2641 pf_cksum_fixup(pf_cksum_fixup(*pc, 2642 ao.addr16[0], an->addr16[0], u), 2643 ao.addr16[1], an->addr16[1], u), 2644 ao.addr16[2], an->addr16[2], u), 2645 ao.addr16[3], an->addr16[3], u), 2646 ao.addr16[4], an->addr16[4], u), 2647 ao.addr16[5], an->addr16[5], u), 2648 ao.addr16[6], an->addr16[6], u), 2649 ao.addr16[7], an->addr16[7], u); 2650 2651 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2652 break; 2653 #endif /* INET6 */ 2654 } 2655 2656 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2657 CSUM_DELAY_DATA_IPV6)) { 2658 *pc = ~*pc; 2659 if (! *pc) 2660 *pc = 0xffff; 2661 } 2662 } 2663 2664 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2665 void 2666 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2667 { 2668 u_int32_t ao; 2669 2670 memcpy(&ao, a, sizeof(ao)); 2671 memcpy(a, &an, sizeof(u_int32_t)); 2672 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2673 ao % 65536, an % 65536, u); 2674 } 2675 2676 void 2677 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2678 { 2679 u_int32_t ao; 2680 2681 memcpy(&ao, a, sizeof(ao)); 2682 memcpy(a, &an, sizeof(u_int32_t)); 2683 2684 *c = pf_proto_cksum_fixup(m, 2685 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2686 ao % 65536, an % 65536, udp); 2687 } 2688 2689 #ifdef INET6 2690 static void 2691 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2692 { 2693 struct pf_addr ao; 2694 2695 PF_ACPY(&ao, a, AF_INET6); 2696 PF_ACPY(a, an, AF_INET6); 2697 2698 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2699 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2700 pf_cksum_fixup(pf_cksum_fixup(*c, 2701 ao.addr16[0], an->addr16[0], u), 2702 ao.addr16[1], an->addr16[1], u), 2703 ao.addr16[2], an->addr16[2], u), 2704 ao.addr16[3], an->addr16[3], u), 2705 ao.addr16[4], an->addr16[4], u), 2706 ao.addr16[5], an->addr16[5], u), 2707 ao.addr16[6], an->addr16[6], u), 2708 ao.addr16[7], an->addr16[7], u); 2709 } 2710 #endif /* INET6 */ 2711 2712 static void 2713 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2714 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2715 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2716 { 2717 struct pf_addr oia, ooa; 2718 2719 PF_ACPY(&oia, ia, af); 2720 if (oa) 2721 PF_ACPY(&ooa, oa, af); 2722 2723 /* Change inner protocol port, fix inner protocol checksum. */ 2724 if (ip != NULL) { 2725 u_int16_t oip = *ip; 2726 u_int32_t opc; 2727 2728 if (pc != NULL) 2729 opc = *pc; 2730 *ip = np; 2731 if (pc != NULL) 2732 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2733 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2734 if (pc != NULL) 2735 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2736 } 2737 /* Change inner ip address, fix inner ip and icmp checksums. */ 2738 PF_ACPY(ia, na, af); 2739 switch (af) { 2740 #ifdef INET 2741 case AF_INET: { 2742 u_int32_t oh2c = *h2c; 2743 2744 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2745 oia.addr16[0], ia->addr16[0], 0), 2746 oia.addr16[1], ia->addr16[1], 0); 2747 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2748 oia.addr16[0], ia->addr16[0], 0), 2749 oia.addr16[1], ia->addr16[1], 0); 2750 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2751 break; 2752 } 2753 #endif /* INET */ 2754 #ifdef INET6 2755 case AF_INET6: 2756 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2757 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2758 pf_cksum_fixup(pf_cksum_fixup(*ic, 2759 oia.addr16[0], ia->addr16[0], u), 2760 oia.addr16[1], ia->addr16[1], u), 2761 oia.addr16[2], ia->addr16[2], u), 2762 oia.addr16[3], ia->addr16[3], u), 2763 oia.addr16[4], ia->addr16[4], u), 2764 oia.addr16[5], ia->addr16[5], u), 2765 oia.addr16[6], ia->addr16[6], u), 2766 oia.addr16[7], ia->addr16[7], u); 2767 break; 2768 #endif /* INET6 */ 2769 } 2770 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2771 if (oa) { 2772 PF_ACPY(oa, na, af); 2773 switch (af) { 2774 #ifdef INET 2775 case AF_INET: 2776 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2777 ooa.addr16[0], oa->addr16[0], 0), 2778 ooa.addr16[1], oa->addr16[1], 0); 2779 break; 2780 #endif /* INET */ 2781 #ifdef INET6 2782 case AF_INET6: 2783 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2784 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2785 pf_cksum_fixup(pf_cksum_fixup(*ic, 2786 ooa.addr16[0], oa->addr16[0], u), 2787 ooa.addr16[1], oa->addr16[1], u), 2788 ooa.addr16[2], oa->addr16[2], u), 2789 ooa.addr16[3], oa->addr16[3], u), 2790 ooa.addr16[4], oa->addr16[4], u), 2791 ooa.addr16[5], oa->addr16[5], u), 2792 ooa.addr16[6], oa->addr16[6], u), 2793 ooa.addr16[7], oa->addr16[7], u); 2794 break; 2795 #endif /* INET6 */ 2796 } 2797 } 2798 } 2799 2800 /* 2801 * Need to modulate the sequence numbers in the TCP SACK option 2802 * (credits to Krzysztof Pfaff for report and patch) 2803 */ 2804 static int 2805 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2806 struct tcphdr *th, struct pf_state_peer *dst) 2807 { 2808 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2809 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2810 int copyback = 0, i, olen; 2811 struct sackblk sack; 2812 2813 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2814 if (hlen < TCPOLEN_SACKLEN || 2815 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2816 return 0; 2817 2818 while (hlen >= TCPOLEN_SACKLEN) { 2819 size_t startoff = opt - opts; 2820 olen = opt[1]; 2821 switch (*opt) { 2822 case TCPOPT_EOL: /* FALLTHROUGH */ 2823 case TCPOPT_NOP: 2824 opt++; 2825 hlen--; 2826 break; 2827 case TCPOPT_SACK: 2828 if (olen > hlen) 2829 olen = hlen; 2830 if (olen >= TCPOLEN_SACKLEN) { 2831 for (i = 2; i + TCPOLEN_SACK <= olen; 2832 i += TCPOLEN_SACK) { 2833 memcpy(&sack, &opt[i], sizeof(sack)); 2834 pf_patch_32_unaligned(m, 2835 &th->th_sum, &sack.start, 2836 htonl(ntohl(sack.start) - dst->seqdiff), 2837 PF_ALGNMNT(startoff), 2838 0); 2839 pf_patch_32_unaligned(m, &th->th_sum, 2840 &sack.end, 2841 htonl(ntohl(sack.end) - dst->seqdiff), 2842 PF_ALGNMNT(startoff), 2843 0); 2844 memcpy(&opt[i], &sack, sizeof(sack)); 2845 } 2846 copyback = 1; 2847 } 2848 /* FALLTHROUGH */ 2849 default: 2850 if (olen < 2) 2851 olen = 2; 2852 hlen -= olen; 2853 opt += olen; 2854 } 2855 } 2856 2857 if (copyback) 2858 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2859 return (copyback); 2860 } 2861 2862 struct mbuf * 2863 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2864 const struct pf_addr *saddr, const struct pf_addr *daddr, 2865 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2866 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 2867 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 2868 { 2869 struct mbuf *m; 2870 int len, tlen; 2871 #ifdef INET 2872 struct ip *h = NULL; 2873 #endif /* INET */ 2874 #ifdef INET6 2875 struct ip6_hdr *h6 = NULL; 2876 #endif /* INET6 */ 2877 struct tcphdr *th; 2878 char *opt; 2879 struct pf_mtag *pf_mtag; 2880 2881 len = 0; 2882 th = NULL; 2883 2884 /* maximum segment size tcp option */ 2885 tlen = sizeof(struct tcphdr); 2886 if (mss) 2887 tlen += 4; 2888 2889 switch (af) { 2890 #ifdef INET 2891 case AF_INET: 2892 len = sizeof(struct ip) + tlen; 2893 break; 2894 #endif /* INET */ 2895 #ifdef INET6 2896 case AF_INET6: 2897 len = sizeof(struct ip6_hdr) + tlen; 2898 break; 2899 #endif /* INET6 */ 2900 default: 2901 panic("%s: unsupported af %d", __func__, af); 2902 } 2903 2904 m = m_gethdr(M_NOWAIT, MT_DATA); 2905 if (m == NULL) 2906 return (NULL); 2907 2908 #ifdef MAC 2909 mac_netinet_firewall_send(m); 2910 #endif 2911 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2912 m_freem(m); 2913 return (NULL); 2914 } 2915 if (skip_firewall) 2916 m->m_flags |= M_SKIP_FIREWALL; 2917 pf_mtag->tag = mtag_tag; 2918 pf_mtag->flags = mtag_flags; 2919 2920 if (rtableid >= 0) 2921 M_SETFIB(m, rtableid); 2922 2923 #ifdef ALTQ 2924 if (r != NULL && r->qid) { 2925 pf_mtag->qid = r->qid; 2926 2927 /* add hints for ecn */ 2928 pf_mtag->hdr = mtod(m, struct ip *); 2929 } 2930 #endif /* ALTQ */ 2931 m->m_data += max_linkhdr; 2932 m->m_pkthdr.len = m->m_len = len; 2933 /* The rest of the stack assumes a rcvif, so provide one. 2934 * This is a locally generated packet, so .. close enough. */ 2935 m->m_pkthdr.rcvif = V_loif; 2936 bzero(m->m_data, len); 2937 switch (af) { 2938 #ifdef INET 2939 case AF_INET: 2940 h = mtod(m, struct ip *); 2941 2942 /* IP header fields included in the TCP checksum */ 2943 h->ip_p = IPPROTO_TCP; 2944 h->ip_len = htons(tlen); 2945 h->ip_src.s_addr = saddr->v4.s_addr; 2946 h->ip_dst.s_addr = daddr->v4.s_addr; 2947 2948 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2949 break; 2950 #endif /* INET */ 2951 #ifdef INET6 2952 case AF_INET6: 2953 h6 = mtod(m, struct ip6_hdr *); 2954 2955 /* IP header fields included in the TCP checksum */ 2956 h6->ip6_nxt = IPPROTO_TCP; 2957 h6->ip6_plen = htons(tlen); 2958 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2959 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2960 2961 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2962 break; 2963 #endif /* INET6 */ 2964 } 2965 2966 /* TCP header */ 2967 th->th_sport = sport; 2968 th->th_dport = dport; 2969 th->th_seq = htonl(seq); 2970 th->th_ack = htonl(ack); 2971 th->th_off = tlen >> 2; 2972 th->th_flags = tcp_flags; 2973 th->th_win = htons(win); 2974 2975 if (mss) { 2976 opt = (char *)(th + 1); 2977 opt[0] = TCPOPT_MAXSEG; 2978 opt[1] = 4; 2979 HTONS(mss); 2980 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2981 } 2982 2983 switch (af) { 2984 #ifdef INET 2985 case AF_INET: 2986 /* TCP checksum */ 2987 th->th_sum = in_cksum(m, len); 2988 2989 /* Finish the IP header */ 2990 h->ip_v = 4; 2991 h->ip_hl = sizeof(*h) >> 2; 2992 h->ip_tos = IPTOS_LOWDELAY; 2993 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2994 h->ip_len = htons(len); 2995 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2996 h->ip_sum = 0; 2997 break; 2998 #endif /* INET */ 2999 #ifdef INET6 3000 case AF_INET6: 3001 /* TCP checksum */ 3002 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3003 sizeof(struct ip6_hdr), tlen); 3004 3005 h6->ip6_vfc |= IPV6_VERSION; 3006 h6->ip6_hlim = IPV6_DEFHLIM; 3007 break; 3008 #endif /* INET6 */ 3009 } 3010 3011 return (m); 3012 } 3013 3014 static void 3015 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3016 uint8_t ttl, int rtableid) 3017 { 3018 struct mbuf *m; 3019 #ifdef INET 3020 struct ip *h = NULL; 3021 #endif /* INET */ 3022 #ifdef INET6 3023 struct ip6_hdr *h6 = NULL; 3024 #endif /* INET6 */ 3025 struct sctphdr *hdr; 3026 struct sctp_chunkhdr *chunk; 3027 struct pf_send_entry *pfse; 3028 int off = 0; 3029 3030 MPASS(af == pd->af); 3031 3032 m = m_gethdr(M_NOWAIT, MT_DATA); 3033 if (m == NULL) 3034 return; 3035 3036 m->m_data += max_linkhdr; 3037 m->m_flags |= M_SKIP_FIREWALL; 3038 /* The rest of the stack assumes a rcvif, so provide one. 3039 * This is a locally generated packet, so .. close enough. */ 3040 m->m_pkthdr.rcvif = V_loif; 3041 3042 /* IPv4|6 header */ 3043 switch (af) { 3044 #ifdef INET 3045 case AF_INET: 3046 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3047 3048 h = mtod(m, struct ip *); 3049 3050 /* IP header fields included in the TCP checksum */ 3051 3052 h->ip_p = IPPROTO_SCTP; 3053 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3054 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3055 h->ip_src = pd->dst->v4; 3056 h->ip_dst = pd->src->v4; 3057 3058 off += sizeof(struct ip); 3059 break; 3060 #endif /* INET */ 3061 #ifdef INET6 3062 case AF_INET6: 3063 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3064 3065 h6 = mtod(m, struct ip6_hdr *); 3066 3067 /* IP header fields included in the TCP checksum */ 3068 h6->ip6_vfc |= IPV6_VERSION; 3069 h6->ip6_nxt = IPPROTO_SCTP; 3070 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3071 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3072 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3073 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3074 3075 off += sizeof(struct ip6_hdr); 3076 break; 3077 #endif /* INET6 */ 3078 } 3079 3080 /* SCTP header */ 3081 hdr = mtodo(m, off); 3082 3083 hdr->src_port = pd->hdr.sctp.dest_port; 3084 hdr->dest_port = pd->hdr.sctp.src_port; 3085 hdr->v_tag = pd->sctp_initiate_tag; 3086 hdr->checksum = 0; 3087 3088 /* Abort chunk. */ 3089 off += sizeof(struct sctphdr); 3090 chunk = mtodo(m, off); 3091 3092 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3093 chunk->chunk_length = htons(sizeof(*chunk)); 3094 3095 /* SCTP checksum */ 3096 off += sizeof(*chunk); 3097 m->m_pkthdr.len = m->m_len = off; 3098 3099 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));; 3100 3101 if (rtableid >= 0) 3102 M_SETFIB(m, rtableid); 3103 3104 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3105 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3106 if (pfse == NULL) { 3107 m_freem(m); 3108 return; 3109 } 3110 3111 switch (af) { 3112 #ifdef INET 3113 case AF_INET: 3114 pfse->pfse_type = PFSE_IP; 3115 break; 3116 #endif /* INET */ 3117 #ifdef INET6 3118 case AF_INET6: 3119 pfse->pfse_type = PFSE_IP6; 3120 break; 3121 #endif /* INET6 */ 3122 } 3123 3124 pfse->pfse_m = m; 3125 pf_send(pfse); 3126 } 3127 3128 void 3129 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3130 const struct pf_addr *saddr, const struct pf_addr *daddr, 3131 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3132 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3133 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3134 { 3135 struct pf_send_entry *pfse; 3136 struct mbuf *m; 3137 3138 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3139 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3140 if (m == NULL) 3141 return; 3142 3143 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3144 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3145 if (pfse == NULL) { 3146 m_freem(m); 3147 return; 3148 } 3149 3150 switch (af) { 3151 #ifdef INET 3152 case AF_INET: 3153 pfse->pfse_type = PFSE_IP; 3154 break; 3155 #endif /* INET */ 3156 #ifdef INET6 3157 case AF_INET6: 3158 pfse->pfse_type = PFSE_IP6; 3159 break; 3160 #endif /* INET6 */ 3161 } 3162 3163 pfse->pfse_m = m; 3164 pf_send(pfse); 3165 } 3166 3167 static void 3168 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3169 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3170 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3171 u_short *reason, int rtableid) 3172 { 3173 struct pf_addr * const saddr = pd->src; 3174 struct pf_addr * const daddr = pd->dst; 3175 sa_family_t af = pd->af; 3176 3177 /* undo NAT changes, if they have taken place */ 3178 if (nr != NULL) { 3179 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3180 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3181 if (pd->sport) 3182 *pd->sport = sk->port[pd->sidx]; 3183 if (pd->dport) 3184 *pd->dport = sk->port[pd->didx]; 3185 if (pd->proto_sum) 3186 *pd->proto_sum = bproto_sum; 3187 if (pd->ip_sum) 3188 *pd->ip_sum = bip_sum; 3189 m_copyback(m, off, hdrlen, pd->hdr.any); 3190 } 3191 if (pd->proto == IPPROTO_TCP && 3192 ((r->rule_flag & PFRULE_RETURNRST) || 3193 (r->rule_flag & PFRULE_RETURN)) && 3194 !(th->th_flags & TH_RST)) { 3195 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3196 int len = 0; 3197 #ifdef INET 3198 struct ip *h4; 3199 #endif 3200 #ifdef INET6 3201 struct ip6_hdr *h6; 3202 #endif 3203 3204 switch (af) { 3205 #ifdef INET 3206 case AF_INET: 3207 h4 = mtod(m, struct ip *); 3208 len = ntohs(h4->ip_len) - off; 3209 break; 3210 #endif 3211 #ifdef INET6 3212 case AF_INET6: 3213 h6 = mtod(m, struct ip6_hdr *); 3214 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3215 break; 3216 #endif 3217 } 3218 3219 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3220 REASON_SET(reason, PFRES_PROTCKSUM); 3221 else { 3222 if (th->th_flags & TH_SYN) 3223 ack++; 3224 if (th->th_flags & TH_FIN) 3225 ack++; 3226 pf_send_tcp(r, af, pd->dst, 3227 pd->src, th->th_dport, th->th_sport, 3228 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3229 r->return_ttl, true, 0, 0, rtableid); 3230 } 3231 } else if (pd->proto == IPPROTO_SCTP && 3232 (r->rule_flag & PFRULE_RETURN)) { 3233 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3234 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3235 r->return_icmp) 3236 pf_send_icmp(m, r->return_icmp >> 8, 3237 r->return_icmp & 255, af, r, rtableid); 3238 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3239 r->return_icmp6) 3240 pf_send_icmp(m, r->return_icmp6 >> 8, 3241 r->return_icmp6 & 255, af, r, rtableid); 3242 } 3243 3244 static int 3245 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3246 { 3247 struct m_tag *mtag; 3248 u_int8_t mpcp; 3249 3250 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3251 if (mtag == NULL) 3252 return (0); 3253 3254 if (prio == PF_PRIO_ZERO) 3255 prio = 0; 3256 3257 mpcp = *(uint8_t *)(mtag + 1); 3258 3259 return (mpcp == prio); 3260 } 3261 3262 static int 3263 pf_icmp_to_bandlim(uint8_t type) 3264 { 3265 switch (type) { 3266 case ICMP_ECHO: 3267 case ICMP_ECHOREPLY: 3268 return (BANDLIM_ICMP_ECHO); 3269 case ICMP_TSTAMP: 3270 case ICMP_TSTAMPREPLY: 3271 return (BANDLIM_ICMP_TSTAMP); 3272 case ICMP_UNREACH: 3273 default: 3274 return (BANDLIM_ICMP_UNREACH); 3275 } 3276 } 3277 3278 static void 3279 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3280 struct pf_krule *r, int rtableid) 3281 { 3282 struct pf_send_entry *pfse; 3283 struct mbuf *m0; 3284 struct pf_mtag *pf_mtag; 3285 3286 /* ICMP packet rate limitation. */ 3287 #ifdef INET6 3288 if (af == AF_INET6) { 3289 if (icmp6_ratelimit(NULL, type, code)) 3290 return; 3291 } 3292 #endif 3293 #ifdef INET 3294 if (af == AF_INET) { 3295 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3296 return; 3297 } 3298 #endif 3299 3300 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3301 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3302 if (pfse == NULL) 3303 return; 3304 3305 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3306 free(pfse, M_PFTEMP); 3307 return; 3308 } 3309 3310 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3311 free(pfse, M_PFTEMP); 3312 return; 3313 } 3314 /* XXX: revisit */ 3315 m0->m_flags |= M_SKIP_FIREWALL; 3316 3317 if (rtableid >= 0) 3318 M_SETFIB(m0, rtableid); 3319 3320 #ifdef ALTQ 3321 if (r->qid) { 3322 pf_mtag->qid = r->qid; 3323 /* add hints for ecn */ 3324 pf_mtag->hdr = mtod(m0, struct ip *); 3325 } 3326 #endif /* ALTQ */ 3327 3328 switch (af) { 3329 #ifdef INET 3330 case AF_INET: 3331 pfse->pfse_type = PFSE_ICMP; 3332 break; 3333 #endif /* INET */ 3334 #ifdef INET6 3335 case AF_INET6: 3336 pfse->pfse_type = PFSE_ICMP6; 3337 break; 3338 #endif /* INET6 */ 3339 } 3340 pfse->pfse_m = m0; 3341 pfse->icmpopts.type = type; 3342 pfse->icmpopts.code = code; 3343 pf_send(pfse); 3344 } 3345 3346 /* 3347 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3348 * If n is 0, they match if they are equal. If n is != 0, they match if they 3349 * are different. 3350 */ 3351 int 3352 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3353 struct pf_addr *b, sa_family_t af) 3354 { 3355 int match = 0; 3356 3357 switch (af) { 3358 #ifdef INET 3359 case AF_INET: 3360 if ((a->addr32[0] & m->addr32[0]) == 3361 (b->addr32[0] & m->addr32[0])) 3362 match++; 3363 break; 3364 #endif /* INET */ 3365 #ifdef INET6 3366 case AF_INET6: 3367 if (((a->addr32[0] & m->addr32[0]) == 3368 (b->addr32[0] & m->addr32[0])) && 3369 ((a->addr32[1] & m->addr32[1]) == 3370 (b->addr32[1] & m->addr32[1])) && 3371 ((a->addr32[2] & m->addr32[2]) == 3372 (b->addr32[2] & m->addr32[2])) && 3373 ((a->addr32[3] & m->addr32[3]) == 3374 (b->addr32[3] & m->addr32[3]))) 3375 match++; 3376 break; 3377 #endif /* INET6 */ 3378 } 3379 if (match) { 3380 if (n) 3381 return (0); 3382 else 3383 return (1); 3384 } else { 3385 if (n) 3386 return (1); 3387 else 3388 return (0); 3389 } 3390 } 3391 3392 /* 3393 * Return 1 if b <= a <= e, otherwise return 0. 3394 */ 3395 int 3396 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3397 struct pf_addr *a, sa_family_t af) 3398 { 3399 switch (af) { 3400 #ifdef INET 3401 case AF_INET: 3402 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3403 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3404 return (0); 3405 break; 3406 #endif /* INET */ 3407 #ifdef INET6 3408 case AF_INET6: { 3409 int i; 3410 3411 /* check a >= b */ 3412 for (i = 0; i < 4; ++i) 3413 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3414 break; 3415 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3416 return (0); 3417 /* check a <= e */ 3418 for (i = 0; i < 4; ++i) 3419 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3420 break; 3421 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3422 return (0); 3423 break; 3424 } 3425 #endif /* INET6 */ 3426 } 3427 return (1); 3428 } 3429 3430 static int 3431 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3432 { 3433 switch (op) { 3434 case PF_OP_IRG: 3435 return ((p > a1) && (p < a2)); 3436 case PF_OP_XRG: 3437 return ((p < a1) || (p > a2)); 3438 case PF_OP_RRG: 3439 return ((p >= a1) && (p <= a2)); 3440 case PF_OP_EQ: 3441 return (p == a1); 3442 case PF_OP_NE: 3443 return (p != a1); 3444 case PF_OP_LT: 3445 return (p < a1); 3446 case PF_OP_LE: 3447 return (p <= a1); 3448 case PF_OP_GT: 3449 return (p > a1); 3450 case PF_OP_GE: 3451 return (p >= a1); 3452 } 3453 return (0); /* never reached */ 3454 } 3455 3456 int 3457 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3458 { 3459 NTOHS(a1); 3460 NTOHS(a2); 3461 NTOHS(p); 3462 return (pf_match(op, a1, a2, p)); 3463 } 3464 3465 static int 3466 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3467 { 3468 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3469 return (0); 3470 return (pf_match(op, a1, a2, u)); 3471 } 3472 3473 static int 3474 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3475 { 3476 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3477 return (0); 3478 return (pf_match(op, a1, a2, g)); 3479 } 3480 3481 int 3482 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3483 { 3484 if (*tag == -1) 3485 *tag = mtag; 3486 3487 return ((!r->match_tag_not && r->match_tag == *tag) || 3488 (r->match_tag_not && r->match_tag != *tag)); 3489 } 3490 3491 int 3492 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3493 { 3494 3495 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3496 3497 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3498 return (ENOMEM); 3499 3500 pd->pf_mtag->tag = tag; 3501 3502 return (0); 3503 } 3504 3505 #define PF_ANCHOR_STACKSIZE 32 3506 struct pf_kanchor_stackframe { 3507 struct pf_kruleset *rs; 3508 struct pf_krule *r; /* XXX: + match bit */ 3509 struct pf_kanchor *child; 3510 }; 3511 3512 /* 3513 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3514 */ 3515 #define PF_ANCHORSTACK_MATCH 0x00000001 3516 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3517 3518 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3519 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3520 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3521 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3522 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3523 } while (0) 3524 3525 void 3526 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3527 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3528 int *match) 3529 { 3530 struct pf_kanchor_stackframe *f; 3531 3532 PF_RULES_RASSERT(); 3533 3534 if (match) 3535 *match = 0; 3536 if (*depth >= PF_ANCHOR_STACKSIZE) { 3537 printf("%s: anchor stack overflow on %s\n", 3538 __func__, (*r)->anchor->name); 3539 *r = TAILQ_NEXT(*r, entries); 3540 return; 3541 } else if (*depth == 0 && a != NULL) 3542 *a = *r; 3543 f = stack + (*depth)++; 3544 f->rs = *rs; 3545 f->r = *r; 3546 if ((*r)->anchor_wildcard) { 3547 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3548 3549 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3550 *r = NULL; 3551 return; 3552 } 3553 *rs = &f->child->ruleset; 3554 } else { 3555 f->child = NULL; 3556 *rs = &(*r)->anchor->ruleset; 3557 } 3558 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3559 } 3560 3561 int 3562 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3563 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3564 int *match) 3565 { 3566 struct pf_kanchor_stackframe *f; 3567 struct pf_krule *fr; 3568 int quick = 0; 3569 3570 PF_RULES_RASSERT(); 3571 3572 do { 3573 if (*depth <= 0) 3574 break; 3575 f = stack + *depth - 1; 3576 fr = PF_ANCHOR_RULE(f); 3577 if (f->child != NULL) { 3578 /* 3579 * This block traverses through 3580 * a wildcard anchor. 3581 */ 3582 if (match != NULL && *match) { 3583 /* 3584 * If any of "*" matched, then 3585 * "foo/ *" matched, mark frame 3586 * appropriately. 3587 */ 3588 PF_ANCHOR_SET_MATCH(f); 3589 *match = 0; 3590 } 3591 f->child = RB_NEXT(pf_kanchor_node, 3592 &fr->anchor->children, f->child); 3593 if (f->child != NULL) { 3594 *rs = &f->child->ruleset; 3595 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3596 if (*r == NULL) 3597 continue; 3598 else 3599 break; 3600 } 3601 } 3602 (*depth)--; 3603 if (*depth == 0 && a != NULL) 3604 *a = NULL; 3605 *rs = f->rs; 3606 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3607 quick = fr->quick; 3608 *r = TAILQ_NEXT(fr, entries); 3609 } while (*r == NULL); 3610 3611 return (quick); 3612 } 3613 3614 struct pf_keth_anchor_stackframe { 3615 struct pf_keth_ruleset *rs; 3616 struct pf_keth_rule *r; /* XXX: + match bit */ 3617 struct pf_keth_anchor *child; 3618 }; 3619 3620 #define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3621 #define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3622 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3623 #define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3624 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3625 } while (0) 3626 3627 void 3628 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3629 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3630 struct pf_keth_rule **a, int *match) 3631 { 3632 struct pf_keth_anchor_stackframe *f; 3633 3634 NET_EPOCH_ASSERT(); 3635 3636 if (match) 3637 *match = 0; 3638 if (*depth >= PF_ANCHOR_STACKSIZE) { 3639 printf("%s: anchor stack overflow on %s\n", 3640 __func__, (*r)->anchor->name); 3641 *r = TAILQ_NEXT(*r, entries); 3642 return; 3643 } else if (*depth == 0 && a != NULL) 3644 *a = *r; 3645 f = stack + (*depth)++; 3646 f->rs = *rs; 3647 f->r = *r; 3648 if ((*r)->anchor_wildcard) { 3649 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3650 3651 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3652 *r = NULL; 3653 return; 3654 } 3655 *rs = &f->child->ruleset; 3656 } else { 3657 f->child = NULL; 3658 *rs = &(*r)->anchor->ruleset; 3659 } 3660 *r = TAILQ_FIRST((*rs)->active.rules); 3661 } 3662 3663 int 3664 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3665 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3666 struct pf_keth_rule **a, int *match) 3667 { 3668 struct pf_keth_anchor_stackframe *f; 3669 struct pf_keth_rule *fr; 3670 int quick = 0; 3671 3672 NET_EPOCH_ASSERT(); 3673 3674 do { 3675 if (*depth <= 0) 3676 break; 3677 f = stack + *depth - 1; 3678 fr = PF_ETH_ANCHOR_RULE(f); 3679 if (f->child != NULL) { 3680 /* 3681 * This block traverses through 3682 * a wildcard anchor. 3683 */ 3684 if (match != NULL && *match) { 3685 /* 3686 * If any of "*" matched, then 3687 * "foo/ *" matched, mark frame 3688 * appropriately. 3689 */ 3690 PF_ETH_ANCHOR_SET_MATCH(f); 3691 *match = 0; 3692 } 3693 f->child = RB_NEXT(pf_keth_anchor_node, 3694 &fr->anchor->children, f->child); 3695 if (f->child != NULL) { 3696 *rs = &f->child->ruleset; 3697 *r = TAILQ_FIRST((*rs)->active.rules); 3698 if (*r == NULL) 3699 continue; 3700 else 3701 break; 3702 } 3703 } 3704 (*depth)--; 3705 if (*depth == 0 && a != NULL) 3706 *a = NULL; 3707 *rs = f->rs; 3708 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3709 quick = fr->quick; 3710 *r = TAILQ_NEXT(fr, entries); 3711 } while (*r == NULL); 3712 3713 return (quick); 3714 } 3715 3716 #ifdef INET6 3717 void 3718 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3719 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3720 { 3721 switch (af) { 3722 #ifdef INET 3723 case AF_INET: 3724 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3725 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3726 break; 3727 #endif /* INET */ 3728 case AF_INET6: 3729 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3730 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3731 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3732 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3733 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3734 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3735 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3736 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3737 break; 3738 } 3739 } 3740 3741 void 3742 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3743 { 3744 switch (af) { 3745 #ifdef INET 3746 case AF_INET: 3747 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3748 break; 3749 #endif /* INET */ 3750 case AF_INET6: 3751 if (addr->addr32[3] == 0xffffffff) { 3752 addr->addr32[3] = 0; 3753 if (addr->addr32[2] == 0xffffffff) { 3754 addr->addr32[2] = 0; 3755 if (addr->addr32[1] == 0xffffffff) { 3756 addr->addr32[1] = 0; 3757 addr->addr32[0] = 3758 htonl(ntohl(addr->addr32[0]) + 1); 3759 } else 3760 addr->addr32[1] = 3761 htonl(ntohl(addr->addr32[1]) + 1); 3762 } else 3763 addr->addr32[2] = 3764 htonl(ntohl(addr->addr32[2]) + 1); 3765 } else 3766 addr->addr32[3] = 3767 htonl(ntohl(addr->addr32[3]) + 1); 3768 break; 3769 } 3770 } 3771 #endif /* INET6 */ 3772 3773 void 3774 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3775 { 3776 /* 3777 * Modern rules use the same flags in rules as they do in states. 3778 */ 3779 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3780 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3781 3782 /* 3783 * Old-style scrub rules have different flags which need to be translated. 3784 */ 3785 if (r->rule_flag & PFRULE_RANDOMID) 3786 a->flags |= PFSTATE_RANDOMID; 3787 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 3788 a->flags |= PFSTATE_SETTOS; 3789 a->set_tos = r->set_tos; 3790 } 3791 3792 if (r->qid) 3793 a->qid = r->qid; 3794 if (r->pqid) 3795 a->pqid = r->pqid; 3796 if (r->rtableid >= 0) 3797 a->rtableid = r->rtableid; 3798 a->log |= r->log; 3799 if (r->min_ttl) 3800 a->min_ttl = r->min_ttl; 3801 if (r->max_mss) 3802 a->max_mss = r->max_mss; 3803 if (r->dnpipe) 3804 a->dnpipe = r->dnpipe; 3805 if (r->dnrpipe) 3806 a->dnrpipe = r->dnrpipe; 3807 if (r->dnpipe || r->dnrpipe) { 3808 if (r->free_flags & PFRULE_DN_IS_PIPE) 3809 a->flags |= PFSTATE_DN_IS_PIPE; 3810 else 3811 a->flags &= ~PFSTATE_DN_IS_PIPE; 3812 } 3813 if (r->scrub_flags & PFSTATE_SETPRIO) { 3814 a->set_prio[0] = r->set_prio[0]; 3815 a->set_prio[1] = r->set_prio[1]; 3816 } 3817 } 3818 3819 int 3820 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 3821 { 3822 struct pf_addr *saddr, *daddr; 3823 u_int16_t sport, dport; 3824 struct inpcbinfo *pi; 3825 struct inpcb *inp; 3826 3827 pd->lookup.uid = UID_MAX; 3828 pd->lookup.gid = GID_MAX; 3829 3830 switch (pd->proto) { 3831 case IPPROTO_TCP: 3832 sport = pd->hdr.tcp.th_sport; 3833 dport = pd->hdr.tcp.th_dport; 3834 pi = &V_tcbinfo; 3835 break; 3836 case IPPROTO_UDP: 3837 sport = pd->hdr.udp.uh_sport; 3838 dport = pd->hdr.udp.uh_dport; 3839 pi = &V_udbinfo; 3840 break; 3841 default: 3842 return (-1); 3843 } 3844 if (pd->dir == PF_IN) { 3845 saddr = pd->src; 3846 daddr = pd->dst; 3847 } else { 3848 u_int16_t p; 3849 3850 p = sport; 3851 sport = dport; 3852 dport = p; 3853 saddr = pd->dst; 3854 daddr = pd->src; 3855 } 3856 switch (pd->af) { 3857 #ifdef INET 3858 case AF_INET: 3859 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3860 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3861 if (inp == NULL) { 3862 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3863 daddr->v4, dport, INPLOOKUP_WILDCARD | 3864 INPLOOKUP_RLOCKPCB, NULL, m); 3865 if (inp == NULL) 3866 return (-1); 3867 } 3868 break; 3869 #endif /* INET */ 3870 #ifdef INET6 3871 case AF_INET6: 3872 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3873 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3874 if (inp == NULL) { 3875 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3876 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3877 INPLOOKUP_RLOCKPCB, NULL, m); 3878 if (inp == NULL) 3879 return (-1); 3880 } 3881 break; 3882 #endif /* INET6 */ 3883 3884 default: 3885 return (-1); 3886 } 3887 INP_RLOCK_ASSERT(inp); 3888 pd->lookup.uid = inp->inp_cred->cr_uid; 3889 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3890 INP_RUNLOCK(inp); 3891 3892 return (1); 3893 } 3894 3895 u_int8_t 3896 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3897 { 3898 int hlen; 3899 u_int8_t hdr[60]; 3900 u_int8_t *opt, optlen; 3901 u_int8_t wscale = 0; 3902 3903 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3904 if (hlen <= sizeof(struct tcphdr)) 3905 return (0); 3906 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3907 return (0); 3908 opt = hdr + sizeof(struct tcphdr); 3909 hlen -= sizeof(struct tcphdr); 3910 while (hlen >= 3) { 3911 switch (*opt) { 3912 case TCPOPT_EOL: 3913 case TCPOPT_NOP: 3914 ++opt; 3915 --hlen; 3916 break; 3917 case TCPOPT_WINDOW: 3918 wscale = opt[2]; 3919 if (wscale > TCP_MAX_WINSHIFT) 3920 wscale = TCP_MAX_WINSHIFT; 3921 wscale |= PF_WSCALE_FLAG; 3922 /* FALLTHROUGH */ 3923 default: 3924 optlen = opt[1]; 3925 if (optlen < 2) 3926 optlen = 2; 3927 hlen -= optlen; 3928 opt += optlen; 3929 break; 3930 } 3931 } 3932 return (wscale); 3933 } 3934 3935 u_int16_t 3936 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3937 { 3938 int hlen; 3939 u_int8_t hdr[60]; 3940 u_int8_t *opt, optlen; 3941 u_int16_t mss = V_tcp_mssdflt; 3942 3943 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3944 if (hlen <= sizeof(struct tcphdr)) 3945 return (0); 3946 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3947 return (0); 3948 opt = hdr + sizeof(struct tcphdr); 3949 hlen -= sizeof(struct tcphdr); 3950 while (hlen >= TCPOLEN_MAXSEG) { 3951 switch (*opt) { 3952 case TCPOPT_EOL: 3953 case TCPOPT_NOP: 3954 ++opt; 3955 --hlen; 3956 break; 3957 case TCPOPT_MAXSEG: 3958 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3959 NTOHS(mss); 3960 /* FALLTHROUGH */ 3961 default: 3962 optlen = opt[1]; 3963 if (optlen < 2) 3964 optlen = 2; 3965 hlen -= optlen; 3966 opt += optlen; 3967 break; 3968 } 3969 } 3970 return (mss); 3971 } 3972 3973 static u_int16_t 3974 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3975 { 3976 struct nhop_object *nh; 3977 #ifdef INET6 3978 struct in6_addr dst6; 3979 uint32_t scopeid; 3980 #endif /* INET6 */ 3981 int hlen = 0; 3982 uint16_t mss = 0; 3983 3984 NET_EPOCH_ASSERT(); 3985 3986 switch (af) { 3987 #ifdef INET 3988 case AF_INET: 3989 hlen = sizeof(struct ip); 3990 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3991 if (nh != NULL) 3992 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3993 break; 3994 #endif /* INET */ 3995 #ifdef INET6 3996 case AF_INET6: 3997 hlen = sizeof(struct ip6_hdr); 3998 in6_splitscope(&addr->v6, &dst6, &scopeid); 3999 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4000 if (nh != NULL) 4001 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4002 break; 4003 #endif /* INET6 */ 4004 } 4005 4006 mss = max(V_tcp_mssdflt, mss); 4007 mss = min(mss, offer); 4008 mss = max(mss, 64); /* sanity - at least max opt space */ 4009 return (mss); 4010 } 4011 4012 static u_int32_t 4013 pf_tcp_iss(struct pf_pdesc *pd) 4014 { 4015 MD5_CTX ctx; 4016 u_int32_t digest[4]; 4017 4018 if (V_pf_tcp_secret_init == 0) { 4019 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4020 MD5Init(&V_pf_tcp_secret_ctx); 4021 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4022 sizeof(V_pf_tcp_secret)); 4023 V_pf_tcp_secret_init = 1; 4024 } 4025 4026 ctx = V_pf_tcp_secret_ctx; 4027 4028 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4029 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4030 if (pd->af == AF_INET6) { 4031 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4032 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4033 } else { 4034 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4035 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4036 } 4037 MD5Final((u_char *)digest, &ctx); 4038 V_pf_tcp_iss_off += 4096; 4039 #define ISN_RANDOM_INCREMENT (4096 - 1) 4040 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4041 V_pf_tcp_iss_off); 4042 #undef ISN_RANDOM_INCREMENT 4043 } 4044 4045 static bool 4046 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4047 { 4048 bool match = true; 4049 4050 /* Always matches if not set */ 4051 if (! r->isset) 4052 return (!r->neg); 4053 4054 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4055 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4056 match = false; 4057 break; 4058 } 4059 } 4060 4061 return (match ^ r->neg); 4062 } 4063 4064 static int 4065 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4066 { 4067 if (*tag == -1) 4068 *tag = mtag; 4069 4070 return ((!r->match_tag_not && r->match_tag == *tag) || 4071 (r->match_tag_not && r->match_tag != *tag)); 4072 } 4073 4074 static void 4075 pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4076 { 4077 /* If we don't have the interface drop the packet. */ 4078 if (ifp == NULL) { 4079 m_freem(m); 4080 return; 4081 } 4082 4083 switch (ifp->if_type) { 4084 case IFT_ETHER: 4085 case IFT_XETHER: 4086 case IFT_L2VLAN: 4087 case IFT_BRIDGE: 4088 case IFT_IEEE8023ADLAG: 4089 break; 4090 default: 4091 m_freem(m); 4092 return; 4093 } 4094 4095 ifp->if_transmit(ifp, m); 4096 } 4097 4098 static int 4099 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4100 { 4101 #ifdef INET 4102 struct ip ip; 4103 #endif 4104 #ifdef INET6 4105 struct ip6_hdr ip6; 4106 #endif 4107 struct mbuf *m = *m0; 4108 struct ether_header *e; 4109 struct pf_keth_rule *r, *rm, *a = NULL; 4110 struct pf_keth_ruleset *ruleset = NULL; 4111 struct pf_mtag *mtag; 4112 struct pf_keth_ruleq *rules; 4113 struct pf_addr *src = NULL, *dst = NULL; 4114 struct pfi_kkif *bridge_to; 4115 sa_family_t af = 0; 4116 uint16_t proto; 4117 int asd = 0, match = 0; 4118 int tag = -1; 4119 uint8_t action; 4120 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4121 4122 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4123 NET_EPOCH_ASSERT(); 4124 4125 PF_RULES_RLOCK_TRACKER; 4126 4127 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4128 4129 mtag = pf_find_mtag(m); 4130 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4131 /* Dummynet re-injects packets after they've 4132 * completed their delay. We've already 4133 * processed them, so pass unconditionally. */ 4134 4135 /* But only once. We may see the packet multiple times (e.g. 4136 * PFIL_IN/PFIL_OUT). */ 4137 mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 4138 4139 return (PF_PASS); 4140 } 4141 4142 ruleset = V_pf_keth; 4143 rules = ck_pr_load_ptr(&ruleset->active.rules); 4144 r = TAILQ_FIRST(rules); 4145 rm = NULL; 4146 4147 e = mtod(m, struct ether_header *); 4148 proto = ntohs(e->ether_type); 4149 4150 switch (proto) { 4151 #ifdef INET 4152 case ETHERTYPE_IP: { 4153 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4154 sizeof(ip))) 4155 return (PF_DROP); 4156 4157 af = AF_INET; 4158 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4159 (caddr_t)&ip); 4160 src = (struct pf_addr *)&ip.ip_src; 4161 dst = (struct pf_addr *)&ip.ip_dst; 4162 break; 4163 } 4164 #endif /* INET */ 4165 #ifdef INET6 4166 case ETHERTYPE_IPV6: { 4167 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4168 sizeof(ip6))) 4169 return (PF_DROP); 4170 4171 af = AF_INET6; 4172 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4173 (caddr_t)&ip6); 4174 src = (struct pf_addr *)&ip6.ip6_src; 4175 dst = (struct pf_addr *)&ip6.ip6_dst; 4176 break; 4177 } 4178 #endif /* INET6 */ 4179 } 4180 4181 PF_RULES_RLOCK(); 4182 4183 while (r != NULL) { 4184 counter_u64_add(r->evaluations, 1); 4185 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4186 4187 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4188 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4189 "kif"); 4190 r = r->skip[PFE_SKIP_IFP].ptr; 4191 } 4192 else if (r->direction && r->direction != dir) { 4193 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4194 "dir"); 4195 r = r->skip[PFE_SKIP_DIR].ptr; 4196 } 4197 else if (r->proto && r->proto != proto) { 4198 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4199 "proto"); 4200 r = r->skip[PFE_SKIP_PROTO].ptr; 4201 } 4202 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4203 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4204 "src"); 4205 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4206 } 4207 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4208 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4209 "dst"); 4210 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4211 } 4212 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4213 r->ipsrc.neg, kif, M_GETFIB(m))) { 4214 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4215 "ip_src"); 4216 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4217 } 4218 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4219 r->ipdst.neg, kif, M_GETFIB(m))) { 4220 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4221 "ip_dst"); 4222 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4223 } 4224 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4225 mtag ? mtag->tag : 0)) { 4226 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4227 "match_tag"); 4228 r = TAILQ_NEXT(r, entries); 4229 } 4230 else { 4231 if (r->tag) 4232 tag = r->tag; 4233 if (r->anchor == NULL) { 4234 /* Rule matches */ 4235 rm = r; 4236 4237 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4238 4239 if (r->quick) 4240 break; 4241 4242 r = TAILQ_NEXT(r, entries); 4243 } else { 4244 pf_step_into_keth_anchor(anchor_stack, &asd, 4245 &ruleset, &r, &a, &match); 4246 } 4247 } 4248 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4249 &ruleset, &r, &a, &match)) 4250 break; 4251 } 4252 4253 r = rm; 4254 4255 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4256 4257 /* Default to pass. */ 4258 if (r == NULL) { 4259 PF_RULES_RUNLOCK(); 4260 return (PF_PASS); 4261 } 4262 4263 /* Execute action. */ 4264 counter_u64_add(r->packets[dir == PF_OUT], 1); 4265 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4266 pf_update_timestamp(r); 4267 4268 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4269 if (r->action == PF_DROP) { 4270 PF_RULES_RUNLOCK(); 4271 return (PF_DROP); 4272 } 4273 4274 if (tag > 0) { 4275 if (mtag == NULL) 4276 mtag = pf_get_mtag(m); 4277 if (mtag == NULL) { 4278 PF_RULES_RUNLOCK(); 4279 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4280 return (PF_DROP); 4281 } 4282 mtag->tag = tag; 4283 } 4284 4285 if (r->qid != 0) { 4286 if (mtag == NULL) 4287 mtag = pf_get_mtag(m); 4288 if (mtag == NULL) { 4289 PF_RULES_RUNLOCK(); 4290 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4291 return (PF_DROP); 4292 } 4293 mtag->qid = r->qid; 4294 } 4295 4296 action = r->action; 4297 bridge_to = r->bridge_to; 4298 4299 /* Dummynet */ 4300 if (r->dnpipe) { 4301 struct ip_fw_args dnflow; 4302 4303 /* Drop packet if dummynet is not loaded. */ 4304 if (ip_dn_io_ptr == NULL) { 4305 PF_RULES_RUNLOCK(); 4306 m_freem(m); 4307 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4308 return (PF_DROP); 4309 } 4310 if (mtag == NULL) 4311 mtag = pf_get_mtag(m); 4312 if (mtag == NULL) { 4313 PF_RULES_RUNLOCK(); 4314 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4315 return (PF_DROP); 4316 } 4317 4318 bzero(&dnflow, sizeof(dnflow)); 4319 4320 /* We don't have port numbers here, so we set 0. That means 4321 * that we'll be somewhat limited in distinguishing flows (i.e. 4322 * only based on IP addresses, not based on port numbers), but 4323 * it's better than nothing. */ 4324 dnflow.f_id.dst_port = 0; 4325 dnflow.f_id.src_port = 0; 4326 dnflow.f_id.proto = 0; 4327 4328 dnflow.rule.info = r->dnpipe; 4329 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4330 if (r->dnflags & PFRULE_DN_IS_PIPE) 4331 dnflow.rule.info |= IPFW_IS_PIPE; 4332 4333 dnflow.f_id.extra = dnflow.rule.info; 4334 4335 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4336 dnflow.flags |= IPFW_ARGS_ETHER; 4337 dnflow.ifp = kif->pfik_ifp; 4338 4339 switch (af) { 4340 case AF_INET: 4341 dnflow.f_id.addr_type = 4; 4342 dnflow.f_id.src_ip = src->v4.s_addr; 4343 dnflow.f_id.dst_ip = dst->v4.s_addr; 4344 break; 4345 case AF_INET6: 4346 dnflow.flags |= IPFW_ARGS_IP6; 4347 dnflow.f_id.addr_type = 6; 4348 dnflow.f_id.src_ip6 = src->v6; 4349 dnflow.f_id.dst_ip6 = dst->v6; 4350 break; 4351 } 4352 4353 PF_RULES_RUNLOCK(); 4354 4355 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4356 ip_dn_io_ptr(m0, &dnflow); 4357 if (*m0 != NULL) 4358 mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 4359 } else { 4360 PF_RULES_RUNLOCK(); 4361 } 4362 4363 if (action == PF_PASS && bridge_to) { 4364 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4365 *m0 = NULL; /* We've eaten the packet. */ 4366 } 4367 4368 return (action); 4369 } 4370 4371 static int 4372 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4373 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4374 struct pf_kruleset **rsm, struct inpcb *inp) 4375 { 4376 struct pf_krule *nr = NULL; 4377 struct pf_addr * const saddr = pd->src; 4378 struct pf_addr * const daddr = pd->dst; 4379 sa_family_t af = pd->af; 4380 struct pf_krule *r, *a = NULL; 4381 struct pf_kruleset *ruleset = NULL; 4382 struct pf_krule_slist match_rules; 4383 struct pf_krule_item *ri; 4384 struct pf_ksrc_node *nsn = NULL; 4385 struct tcphdr *th = &pd->hdr.tcp; 4386 struct pf_state_key *sk = NULL, *nk = NULL; 4387 u_short reason; 4388 int rewrite = 0, hdrlen = 0; 4389 int tag = -1; 4390 int asd = 0; 4391 int match = 0; 4392 int state_icmp = 0; 4393 u_int16_t sport = 0, dport = 0; 4394 u_int16_t bproto_sum = 0, bip_sum = 0; 4395 u_int8_t icmptype = 0, icmpcode = 0; 4396 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4397 4398 PF_RULES_RASSERT(); 4399 4400 if (inp != NULL) { 4401 INP_LOCK_ASSERT(inp); 4402 pd->lookup.uid = inp->inp_cred->cr_uid; 4403 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4404 pd->lookup.done = 1; 4405 } 4406 4407 switch (pd->proto) { 4408 case IPPROTO_TCP: 4409 sport = th->th_sport; 4410 dport = th->th_dport; 4411 hdrlen = sizeof(*th); 4412 break; 4413 case IPPROTO_UDP: 4414 sport = pd->hdr.udp.uh_sport; 4415 dport = pd->hdr.udp.uh_dport; 4416 hdrlen = sizeof(pd->hdr.udp); 4417 break; 4418 case IPPROTO_SCTP: 4419 sport = pd->hdr.sctp.src_port; 4420 dport = pd->hdr.sctp.dest_port; 4421 hdrlen = sizeof(pd->hdr.sctp); 4422 break; 4423 #ifdef INET 4424 case IPPROTO_ICMP: 4425 if (pd->af != AF_INET) 4426 break; 4427 sport = dport = pd->hdr.icmp.icmp_id; 4428 hdrlen = sizeof(pd->hdr.icmp); 4429 icmptype = pd->hdr.icmp.icmp_type; 4430 icmpcode = pd->hdr.icmp.icmp_code; 4431 4432 if (icmptype == ICMP_UNREACH || 4433 icmptype == ICMP_SOURCEQUENCH || 4434 icmptype == ICMP_REDIRECT || 4435 icmptype == ICMP_TIMXCEED || 4436 icmptype == ICMP_PARAMPROB) 4437 state_icmp++; 4438 break; 4439 #endif /* INET */ 4440 #ifdef INET6 4441 case IPPROTO_ICMPV6: 4442 if (af != AF_INET6) 4443 break; 4444 sport = dport = pd->hdr.icmp6.icmp6_id; 4445 hdrlen = sizeof(pd->hdr.icmp6); 4446 icmptype = pd->hdr.icmp6.icmp6_type; 4447 icmpcode = pd->hdr.icmp6.icmp6_code; 4448 4449 if (icmptype == ICMP6_DST_UNREACH || 4450 icmptype == ICMP6_PACKET_TOO_BIG || 4451 icmptype == ICMP6_TIME_EXCEEDED || 4452 icmptype == ICMP6_PARAM_PROB) 4453 state_icmp++; 4454 break; 4455 #endif /* INET6 */ 4456 default: 4457 sport = dport = hdrlen = 0; 4458 break; 4459 } 4460 4461 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4462 4463 /* check packet for BINAT/NAT/RDR */ 4464 if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4465 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4466 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4467 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4468 4469 if (nr->log) { 4470 PFLOG_PACKET(kif, m, af, PFRES_MATCH, nr, a, 4471 ruleset, pd, 1); 4472 } 4473 4474 if (pd->ip_sum) 4475 bip_sum = *pd->ip_sum; 4476 4477 switch (pd->proto) { 4478 case IPPROTO_TCP: 4479 bproto_sum = th->th_sum; 4480 pd->proto_sum = &th->th_sum; 4481 4482 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4483 nk->port[pd->sidx] != sport) { 4484 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4485 &th->th_sum, &nk->addr[pd->sidx], 4486 nk->port[pd->sidx], 0, af); 4487 pd->sport = &th->th_sport; 4488 sport = th->th_sport; 4489 } 4490 4491 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4492 nk->port[pd->didx] != dport) { 4493 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4494 &th->th_sum, &nk->addr[pd->didx], 4495 nk->port[pd->didx], 0, af); 4496 dport = th->th_dport; 4497 pd->dport = &th->th_dport; 4498 } 4499 rewrite++; 4500 break; 4501 case IPPROTO_UDP: 4502 bproto_sum = pd->hdr.udp.uh_sum; 4503 pd->proto_sum = &pd->hdr.udp.uh_sum; 4504 4505 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4506 nk->port[pd->sidx] != sport) { 4507 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4508 pd->ip_sum, &pd->hdr.udp.uh_sum, 4509 &nk->addr[pd->sidx], 4510 nk->port[pd->sidx], 1, af); 4511 sport = pd->hdr.udp.uh_sport; 4512 pd->sport = &pd->hdr.udp.uh_sport; 4513 } 4514 4515 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4516 nk->port[pd->didx] != dport) { 4517 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4518 pd->ip_sum, &pd->hdr.udp.uh_sum, 4519 &nk->addr[pd->didx], 4520 nk->port[pd->didx], 1, af); 4521 dport = pd->hdr.udp.uh_dport; 4522 pd->dport = &pd->hdr.udp.uh_dport; 4523 } 4524 rewrite++; 4525 break; 4526 case IPPROTO_SCTP: { 4527 uint16_t checksum = 0; 4528 4529 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4530 nk->port[pd->sidx] != sport) { 4531 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 4532 pd->ip_sum, &checksum, 4533 &nk->addr[pd->sidx], 4534 nk->port[pd->sidx], 1, af); 4535 } 4536 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4537 nk->port[pd->didx] != dport) { 4538 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 4539 pd->ip_sum, &checksum, 4540 &nk->addr[pd->didx], 4541 nk->port[pd->didx], 1, af); 4542 } 4543 break; 4544 } 4545 #ifdef INET 4546 case IPPROTO_ICMP: 4547 nk->port[0] = nk->port[1]; 4548 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4549 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4550 nk->addr[pd->sidx].v4.s_addr, 0); 4551 4552 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4553 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4554 nk->addr[pd->didx].v4.s_addr, 0); 4555 4556 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4557 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4558 pd->hdr.icmp.icmp_cksum, sport, 4559 nk->port[1], 0); 4560 pd->hdr.icmp.icmp_id = nk->port[1]; 4561 pd->sport = &pd->hdr.icmp.icmp_id; 4562 } 4563 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4564 break; 4565 #endif /* INET */ 4566 #ifdef INET6 4567 case IPPROTO_ICMPV6: 4568 nk->port[0] = nk->port[1]; 4569 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4570 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4571 &nk->addr[pd->sidx], 0); 4572 4573 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4574 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4575 &nk->addr[pd->didx], 0); 4576 rewrite++; 4577 break; 4578 #endif /* INET */ 4579 default: 4580 switch (af) { 4581 #ifdef INET 4582 case AF_INET: 4583 if (PF_ANEQ(saddr, 4584 &nk->addr[pd->sidx], AF_INET)) 4585 pf_change_a(&saddr->v4.s_addr, 4586 pd->ip_sum, 4587 nk->addr[pd->sidx].v4.s_addr, 0); 4588 4589 if (PF_ANEQ(daddr, 4590 &nk->addr[pd->didx], AF_INET)) 4591 pf_change_a(&daddr->v4.s_addr, 4592 pd->ip_sum, 4593 nk->addr[pd->didx].v4.s_addr, 0); 4594 break; 4595 #endif /* INET */ 4596 #ifdef INET6 4597 case AF_INET6: 4598 if (PF_ANEQ(saddr, 4599 &nk->addr[pd->sidx], AF_INET6)) 4600 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4601 4602 if (PF_ANEQ(daddr, 4603 &nk->addr[pd->didx], AF_INET6)) 4604 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4605 break; 4606 #endif /* INET */ 4607 } 4608 break; 4609 } 4610 if (nr->natpass) 4611 r = NULL; 4612 pd->nat_rule = nr; 4613 } 4614 4615 SLIST_INIT(&match_rules); 4616 while (r != NULL) { 4617 pf_counter_u64_add(&r->evaluations, 1); 4618 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4619 r = r->skip[PF_SKIP_IFP].ptr; 4620 else if (r->direction && r->direction != pd->dir) 4621 r = r->skip[PF_SKIP_DIR].ptr; 4622 else if (r->af && r->af != af) 4623 r = r->skip[PF_SKIP_AF].ptr; 4624 else if (r->proto && r->proto != pd->proto) 4625 r = r->skip[PF_SKIP_PROTO].ptr; 4626 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4627 r->src.neg, kif, M_GETFIB(m))) 4628 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4629 /* tcp/udp only. port_op always 0 in other cases */ 4630 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4631 r->src.port[0], r->src.port[1], sport)) 4632 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4633 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4634 r->dst.neg, NULL, M_GETFIB(m))) 4635 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4636 /* tcp/udp only. port_op always 0 in other cases */ 4637 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4638 r->dst.port[0], r->dst.port[1], dport)) 4639 r = r->skip[PF_SKIP_DST_PORT].ptr; 4640 /* icmp only. type always 0 in other cases */ 4641 else if (r->type && r->type != icmptype + 1) 4642 r = TAILQ_NEXT(r, entries); 4643 /* icmp only. type always 0 in other cases */ 4644 else if (r->code && r->code != icmpcode + 1) 4645 r = TAILQ_NEXT(r, entries); 4646 else if (r->tos && !(r->tos == pd->tos)) 4647 r = TAILQ_NEXT(r, entries); 4648 else if (r->rule_flag & PFRULE_FRAGMENT) 4649 r = TAILQ_NEXT(r, entries); 4650 else if (pd->proto == IPPROTO_TCP && 4651 (r->flagset & th->th_flags) != r->flags) 4652 r = TAILQ_NEXT(r, entries); 4653 /* tcp/udp only. uid.op always 0 in other cases */ 4654 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4655 pf_socket_lookup(pd, m), 1)) && 4656 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4657 pd->lookup.uid)) 4658 r = TAILQ_NEXT(r, entries); 4659 /* tcp/udp only. gid.op always 0 in other cases */ 4660 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4661 pf_socket_lookup(pd, m), 1)) && 4662 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4663 pd->lookup.gid)) 4664 r = TAILQ_NEXT(r, entries); 4665 else if (r->prio && 4666 !pf_match_ieee8021q_pcp(r->prio, m)) 4667 r = TAILQ_NEXT(r, entries); 4668 else if (r->prob && 4669 r->prob <= arc4random()) 4670 r = TAILQ_NEXT(r, entries); 4671 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4672 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4673 r = TAILQ_NEXT(r, entries); 4674 else if (r->os_fingerprint != PF_OSFP_ANY && 4675 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4676 pf_osfp_fingerprint(pd, m, off, th), 4677 r->os_fingerprint))) 4678 r = TAILQ_NEXT(r, entries); 4679 else { 4680 if (r->tag) 4681 tag = r->tag; 4682 if (r->anchor == NULL) { 4683 if (r->action == PF_MATCH) { 4684 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4685 if (ri == NULL) { 4686 REASON_SET(&reason, PFRES_MEMORY); 4687 goto cleanup; 4688 } 4689 ri->r = r; 4690 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4691 pf_counter_u64_critical_enter(); 4692 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 4693 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 4694 pf_counter_u64_critical_exit(); 4695 pf_rule_to_actions(r, &pd->act); 4696 if (r->log) 4697 PFLOG_PACKET(kif, m, af, 4698 PFRES_MATCH, r, 4699 a, ruleset, pd, 1); 4700 } else { 4701 match = 1; 4702 *rm = r; 4703 *am = a; 4704 *rsm = ruleset; 4705 } 4706 if ((*rm)->quick) 4707 break; 4708 r = TAILQ_NEXT(r, entries); 4709 } else 4710 pf_step_into_anchor(anchor_stack, &asd, 4711 &ruleset, PF_RULESET_FILTER, &r, &a, 4712 &match); 4713 } 4714 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4715 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4716 break; 4717 } 4718 r = *rm; 4719 a = *am; 4720 ruleset = *rsm; 4721 4722 REASON_SET(&reason, PFRES_MATCH); 4723 4724 /* apply actions for last matching pass/block rule */ 4725 pf_rule_to_actions(r, &pd->act); 4726 4727 if (r->log) { 4728 if (rewrite) 4729 m_copyback(m, off, hdrlen, pd->hdr.any); 4730 PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1); 4731 } 4732 4733 if ((r->action == PF_DROP) && 4734 ((r->rule_flag & PFRULE_RETURNRST) || 4735 (r->rule_flag & PFRULE_RETURNICMP) || 4736 (r->rule_flag & PFRULE_RETURN))) { 4737 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4738 bip_sum, hdrlen, &reason, r->rtableid); 4739 } 4740 4741 if (r->action == PF_DROP) 4742 goto cleanup; 4743 4744 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4745 REASON_SET(&reason, PFRES_MEMORY); 4746 goto cleanup; 4747 } 4748 if (pd->act.rtableid >= 0) 4749 M_SETFIB(m, pd->act.rtableid); 4750 4751 if (!state_icmp && (r->keep_state || nr != NULL || 4752 (pd->flags & PFDESC_TCP_NORM))) { 4753 int action; 4754 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4755 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4756 hdrlen, &match_rules); 4757 if (action != PF_PASS) { 4758 if (action == PF_DROP && 4759 (r->rule_flag & PFRULE_RETURN)) 4760 pf_return(r, nr, pd, sk, off, m, th, kif, 4761 bproto_sum, bip_sum, hdrlen, &reason, 4762 pd->act.rtableid); 4763 return (action); 4764 } 4765 } else { 4766 while ((ri = SLIST_FIRST(&match_rules))) { 4767 SLIST_REMOVE_HEAD(&match_rules, entry); 4768 free(ri, M_PF_RULE_ITEM); 4769 } 4770 4771 uma_zfree(V_pf_state_key_z, sk); 4772 uma_zfree(V_pf_state_key_z, nk); 4773 } 4774 4775 /* copy back packet headers if we performed NAT operations */ 4776 if (rewrite) 4777 m_copyback(m, off, hdrlen, pd->hdr.any); 4778 4779 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4780 pd->dir == PF_OUT && 4781 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4782 /* 4783 * We want the state created, but we dont 4784 * want to send this in case a partner 4785 * firewall has to know about it to allow 4786 * replies through it. 4787 */ 4788 return (PF_DEFER); 4789 4790 return (PF_PASS); 4791 4792 cleanup: 4793 while ((ri = SLIST_FIRST(&match_rules))) { 4794 SLIST_REMOVE_HEAD(&match_rules, entry); 4795 free(ri, M_PF_RULE_ITEM); 4796 } 4797 4798 uma_zfree(V_pf_state_key_z, sk); 4799 uma_zfree(V_pf_state_key_z, nk); 4800 return (PF_DROP); 4801 } 4802 4803 static int 4804 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4805 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4806 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4807 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4808 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 4809 struct pf_krule_slist *match_rules) 4810 { 4811 struct pf_kstate *s = NULL; 4812 struct pf_ksrc_node *sn = NULL; 4813 struct tcphdr *th = &pd->hdr.tcp; 4814 u_int16_t mss = V_tcp_mssdflt; 4815 u_short reason, sn_reason; 4816 struct pf_krule_item *ri; 4817 4818 /* check maximums */ 4819 if (r->max_states && 4820 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4821 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4822 REASON_SET(&reason, PFRES_MAXSTATES); 4823 goto csfailed; 4824 } 4825 /* src node for filter rule */ 4826 if ((r->rule_flag & PFRULE_SRCTRACK || 4827 r->rpool.opts & PF_POOL_STICKYADDR) && 4828 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 4829 REASON_SET(&reason, sn_reason); 4830 goto csfailed; 4831 } 4832 /* src node for translation rule */ 4833 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4834 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 4835 pd->af)) != 0 ) { 4836 REASON_SET(&reason, sn_reason); 4837 goto csfailed; 4838 } 4839 s = pf_alloc_state(M_NOWAIT); 4840 if (s == NULL) { 4841 REASON_SET(&reason, PFRES_MEMORY); 4842 goto csfailed; 4843 } 4844 s->rule.ptr = r; 4845 s->nat_rule.ptr = nr; 4846 s->anchor.ptr = a; 4847 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 4848 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 4849 4850 STATE_INC_COUNTERS(s); 4851 if (r->allow_opts) 4852 s->state_flags |= PFSTATE_ALLOWOPTS; 4853 if (r->rule_flag & PFRULE_STATESLOPPY) 4854 s->state_flags |= PFSTATE_SLOPPY; 4855 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 4856 s->state_flags |= PFSTATE_SCRUB_TCP; 4857 4858 s->act.log = pd->act.log & PF_LOG_ALL; 4859 s->sync_state = PFSYNC_S_NONE; 4860 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 4861 4862 if (nr != NULL) 4863 s->act.log |= nr->log & PF_LOG_ALL; 4864 switch (pd->proto) { 4865 case IPPROTO_TCP: 4866 s->src.seqlo = ntohl(th->th_seq); 4867 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4868 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4869 r->keep_state == PF_STATE_MODULATE) { 4870 /* Generate sequence number modulator */ 4871 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4872 0) 4873 s->src.seqdiff = 1; 4874 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4875 htonl(s->src.seqlo + s->src.seqdiff), 0); 4876 *rewrite = 1; 4877 } else 4878 s->src.seqdiff = 0; 4879 if (th->th_flags & TH_SYN) { 4880 s->src.seqhi++; 4881 s->src.wscale = pf_get_wscale(m, off, 4882 th->th_off, pd->af); 4883 } 4884 s->src.max_win = MAX(ntohs(th->th_win), 1); 4885 if (s->src.wscale & PF_WSCALE_MASK) { 4886 /* Remove scale factor from initial window */ 4887 int win = s->src.max_win; 4888 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4889 s->src.max_win = (win - 1) >> 4890 (s->src.wscale & PF_WSCALE_MASK); 4891 } 4892 if (th->th_flags & TH_FIN) 4893 s->src.seqhi++; 4894 s->dst.seqhi = 1; 4895 s->dst.max_win = 1; 4896 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4897 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4898 s->timeout = PFTM_TCP_FIRST_PACKET; 4899 atomic_add_32(&V_pf_status.states_halfopen, 1); 4900 break; 4901 case IPPROTO_UDP: 4902 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4903 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4904 s->timeout = PFTM_UDP_FIRST_PACKET; 4905 break; 4906 case IPPROTO_SCTP: 4907 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 4908 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 4909 s->timeout = PFTM_SCTP_FIRST_PACKET; 4910 break; 4911 case IPPROTO_ICMP: 4912 #ifdef INET6 4913 case IPPROTO_ICMPV6: 4914 #endif 4915 s->timeout = PFTM_ICMP_FIRST_PACKET; 4916 break; 4917 default: 4918 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4919 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4920 s->timeout = PFTM_OTHER_FIRST_PACKET; 4921 } 4922 4923 if (r->rt) { 4924 /* pf_map_addr increases the reason counters */ 4925 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 4926 &s->rt_kif, NULL, &sn)) != 0) 4927 goto csfailed; 4928 s->rt = r->rt; 4929 } 4930 4931 s->creation = time_uptime; 4932 s->expire = time_uptime; 4933 4934 if (sn != NULL) 4935 s->src_node = sn; 4936 if (nsn != NULL) { 4937 /* XXX We only modify one side for now. */ 4938 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4939 s->nat_src_node = nsn; 4940 } 4941 if (pd->proto == IPPROTO_TCP) { 4942 if (s->state_flags & PFSTATE_SCRUB_TCP && 4943 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 4944 REASON_SET(&reason, PFRES_MEMORY); 4945 goto drop; 4946 } 4947 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 4948 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4949 &s->src, &s->dst, rewrite)) { 4950 /* This really shouldn't happen!!! */ 4951 DPFPRINTF(PF_DEBUG_URGENT, 4952 ("pf_normalize_tcp_stateful failed on first " 4953 "pkt\n")); 4954 goto drop; 4955 } 4956 } else if (pd->proto == IPPROTO_SCTP) { 4957 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 4958 goto drop; 4959 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 4960 goto drop; 4961 } 4962 s->direction = pd->dir; 4963 4964 /* 4965 * sk/nk could already been setup by pf_get_translation(). 4966 */ 4967 if (nr == NULL) { 4968 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4969 __func__, nr, sk, nk)); 4970 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4971 if (sk == NULL) 4972 goto csfailed; 4973 nk = sk; 4974 } else 4975 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4976 __func__, nr, sk, nk)); 4977 4978 /* Swap sk/nk for PF_OUT. */ 4979 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4980 (pd->dir == PF_IN) ? sk : nk, 4981 (pd->dir == PF_IN) ? nk : sk, s)) { 4982 REASON_SET(&reason, PFRES_STATEINS); 4983 goto drop; 4984 } else 4985 *sm = s; 4986 4987 if (tag > 0) 4988 s->tag = tag; 4989 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4990 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4991 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4992 /* undo NAT changes, if they have taken place */ 4993 if (nr != NULL) { 4994 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4995 if (pd->dir == PF_OUT) 4996 skt = s->key[PF_SK_STACK]; 4997 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4998 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4999 if (pd->sport) 5000 *pd->sport = skt->port[pd->sidx]; 5001 if (pd->dport) 5002 *pd->dport = skt->port[pd->didx]; 5003 if (pd->proto_sum) 5004 *pd->proto_sum = bproto_sum; 5005 if (pd->ip_sum) 5006 *pd->ip_sum = bip_sum; 5007 m_copyback(m, off, hdrlen, pd->hdr.any); 5008 } 5009 s->src.seqhi = htonl(arc4random()); 5010 /* Find mss option */ 5011 int rtid = M_GETFIB(m); 5012 mss = pf_get_mss(m, off, th->th_off, pd->af); 5013 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5014 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5015 s->src.mss = mss; 5016 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5017 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5018 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5019 pd->act.rtableid); 5020 REASON_SET(&reason, PFRES_SYNPROXY); 5021 return (PF_SYNPROXY_DROP); 5022 } 5023 5024 return (PF_PASS); 5025 5026 csfailed: 5027 while ((ri = SLIST_FIRST(match_rules))) { 5028 SLIST_REMOVE_HEAD(match_rules, entry); 5029 free(ri, M_PF_RULE_ITEM); 5030 } 5031 5032 uma_zfree(V_pf_state_key_z, sk); 5033 uma_zfree(V_pf_state_key_z, nk); 5034 5035 if (sn != NULL) { 5036 PF_SRC_NODE_LOCK(sn); 5037 if (--sn->states == 0 && sn->expire == 0) { 5038 pf_unlink_src_node(sn); 5039 uma_zfree(V_pf_sources_z, sn); 5040 counter_u64_add( 5041 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5042 } 5043 PF_SRC_NODE_UNLOCK(sn); 5044 } 5045 5046 if (nsn != sn && nsn != NULL) { 5047 PF_SRC_NODE_LOCK(nsn); 5048 if (--nsn->states == 0 && nsn->expire == 0) { 5049 pf_unlink_src_node(nsn); 5050 uma_zfree(V_pf_sources_z, nsn); 5051 counter_u64_add( 5052 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5053 } 5054 PF_SRC_NODE_UNLOCK(nsn); 5055 } 5056 5057 drop: 5058 if (s != NULL) { 5059 pf_src_tree_remove_state(s); 5060 s->timeout = PFTM_UNLINKED; 5061 STATE_DEC_COUNTERS(s); 5062 pf_free_state(s); 5063 } 5064 5065 return (PF_DROP); 5066 } 5067 5068 static int 5069 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5070 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5071 struct pf_kruleset **rsm) 5072 { 5073 struct pf_krule *r, *a = NULL; 5074 struct pf_kruleset *ruleset = NULL; 5075 struct pf_krule_slist match_rules; 5076 struct pf_krule_item *ri; 5077 sa_family_t af = pd->af; 5078 u_short reason; 5079 int tag = -1; 5080 int asd = 0; 5081 int match = 0; 5082 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5083 5084 PF_RULES_RASSERT(); 5085 5086 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5087 SLIST_INIT(&match_rules); 5088 while (r != NULL) { 5089 pf_counter_u64_add(&r->evaluations, 1); 5090 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5091 r = r->skip[PF_SKIP_IFP].ptr; 5092 else if (r->direction && r->direction != pd->dir) 5093 r = r->skip[PF_SKIP_DIR].ptr; 5094 else if (r->af && r->af != af) 5095 r = r->skip[PF_SKIP_AF].ptr; 5096 else if (r->proto && r->proto != pd->proto) 5097 r = r->skip[PF_SKIP_PROTO].ptr; 5098 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5099 r->src.neg, kif, M_GETFIB(m))) 5100 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5101 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5102 r->dst.neg, NULL, M_GETFIB(m))) 5103 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5104 else if (r->tos && !(r->tos == pd->tos)) 5105 r = TAILQ_NEXT(r, entries); 5106 else if (r->os_fingerprint != PF_OSFP_ANY) 5107 r = TAILQ_NEXT(r, entries); 5108 else if (pd->proto == IPPROTO_UDP && 5109 (r->src.port_op || r->dst.port_op)) 5110 r = TAILQ_NEXT(r, entries); 5111 else if (pd->proto == IPPROTO_TCP && 5112 (r->src.port_op || r->dst.port_op || r->flagset)) 5113 r = TAILQ_NEXT(r, entries); 5114 else if ((pd->proto == IPPROTO_ICMP || 5115 pd->proto == IPPROTO_ICMPV6) && 5116 (r->type || r->code)) 5117 r = TAILQ_NEXT(r, entries); 5118 else if (r->prio && 5119 !pf_match_ieee8021q_pcp(r->prio, m)) 5120 r = TAILQ_NEXT(r, entries); 5121 else if (r->prob && r->prob <= 5122 (arc4random() % (UINT_MAX - 1) + 1)) 5123 r = TAILQ_NEXT(r, entries); 5124 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5125 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5126 r = TAILQ_NEXT(r, entries); 5127 else { 5128 if (r->anchor == NULL) { 5129 if (r->action == PF_MATCH) { 5130 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5131 if (ri == NULL) { 5132 REASON_SET(&reason, PFRES_MEMORY); 5133 goto cleanup; 5134 } 5135 ri->r = r; 5136 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5137 pf_counter_u64_critical_enter(); 5138 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5139 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5140 pf_counter_u64_critical_exit(); 5141 pf_rule_to_actions(r, &pd->act); 5142 if (r->log) 5143 PFLOG_PACKET(kif, m, af, 5144 PFRES_MATCH, r, 5145 a, ruleset, pd, 1); 5146 } else { 5147 match = 1; 5148 *rm = r; 5149 *am = a; 5150 *rsm = ruleset; 5151 } 5152 if ((*rm)->quick) 5153 break; 5154 r = TAILQ_NEXT(r, entries); 5155 } else 5156 pf_step_into_anchor(anchor_stack, &asd, 5157 &ruleset, PF_RULESET_FILTER, &r, &a, 5158 &match); 5159 } 5160 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5161 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5162 break; 5163 } 5164 r = *rm; 5165 a = *am; 5166 ruleset = *rsm; 5167 5168 REASON_SET(&reason, PFRES_MATCH); 5169 5170 /* apply actions for last matching pass/block rule */ 5171 pf_rule_to_actions(r, &pd->act); 5172 5173 if (r->log) 5174 PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1); 5175 5176 if (r->action != PF_PASS) 5177 return (PF_DROP); 5178 5179 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5180 REASON_SET(&reason, PFRES_MEMORY); 5181 goto cleanup; 5182 } 5183 5184 return (PF_PASS); 5185 5186 cleanup: 5187 while ((ri = SLIST_FIRST(&match_rules))) { 5188 SLIST_REMOVE_HEAD(&match_rules, entry); 5189 free(ri, M_PF_RULE_ITEM); 5190 } 5191 5192 return (PF_DROP); 5193 } 5194 5195 static int 5196 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5197 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5198 int *copyback) 5199 { 5200 struct tcphdr *th = &pd->hdr.tcp; 5201 struct pf_state_peer *src, *dst; 5202 u_int16_t win = ntohs(th->th_win); 5203 u_int32_t ack, end, seq, orig_seq; 5204 u_int8_t sws, dws, psrc, pdst; 5205 int ackskew; 5206 5207 if (pd->dir == (*state)->direction) { 5208 src = &(*state)->src; 5209 dst = &(*state)->dst; 5210 psrc = PF_PEER_SRC; 5211 pdst = PF_PEER_DST; 5212 } else { 5213 src = &(*state)->dst; 5214 dst = &(*state)->src; 5215 psrc = PF_PEER_DST; 5216 pdst = PF_PEER_SRC; 5217 } 5218 5219 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5220 sws = src->wscale & PF_WSCALE_MASK; 5221 dws = dst->wscale & PF_WSCALE_MASK; 5222 } else 5223 sws = dws = 0; 5224 5225 /* 5226 * Sequence tracking algorithm from Guido van Rooij's paper: 5227 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5228 * tcp_filtering.ps 5229 */ 5230 5231 orig_seq = seq = ntohl(th->th_seq); 5232 if (src->seqlo == 0) { 5233 /* First packet from this end. Set its state */ 5234 5235 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5236 src->scrub == NULL) { 5237 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5238 REASON_SET(reason, PFRES_MEMORY); 5239 return (PF_DROP); 5240 } 5241 } 5242 5243 /* Deferred generation of sequence number modulator */ 5244 if (dst->seqdiff && !src->seqdiff) { 5245 /* use random iss for the TCP server */ 5246 while ((src->seqdiff = arc4random() - seq) == 0) 5247 ; 5248 ack = ntohl(th->th_ack) - dst->seqdiff; 5249 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5250 src->seqdiff), 0); 5251 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5252 *copyback = 1; 5253 } else { 5254 ack = ntohl(th->th_ack); 5255 } 5256 5257 end = seq + pd->p_len; 5258 if (th->th_flags & TH_SYN) { 5259 end++; 5260 if (dst->wscale & PF_WSCALE_FLAG) { 5261 src->wscale = pf_get_wscale(m, off, th->th_off, 5262 pd->af); 5263 if (src->wscale & PF_WSCALE_FLAG) { 5264 /* Remove scale factor from initial 5265 * window */ 5266 sws = src->wscale & PF_WSCALE_MASK; 5267 win = ((u_int32_t)win + (1 << sws) - 1) 5268 >> sws; 5269 dws = dst->wscale & PF_WSCALE_MASK; 5270 } else { 5271 /* fixup other window */ 5272 dst->max_win <<= dst->wscale & 5273 PF_WSCALE_MASK; 5274 /* in case of a retrans SYN|ACK */ 5275 dst->wscale = 0; 5276 } 5277 } 5278 } 5279 if (th->th_flags & TH_FIN) 5280 end++; 5281 5282 src->seqlo = seq; 5283 if (src->state < TCPS_SYN_SENT) 5284 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5285 5286 /* 5287 * May need to slide the window (seqhi may have been set by 5288 * the crappy stack check or if we picked up the connection 5289 * after establishment) 5290 */ 5291 if (src->seqhi == 1 || 5292 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5293 src->seqhi = end + MAX(1, dst->max_win << dws); 5294 if (win > src->max_win) 5295 src->max_win = win; 5296 5297 } else { 5298 ack = ntohl(th->th_ack) - dst->seqdiff; 5299 if (src->seqdiff) { 5300 /* Modulate sequence numbers */ 5301 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5302 src->seqdiff), 0); 5303 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5304 *copyback = 1; 5305 } 5306 end = seq + pd->p_len; 5307 if (th->th_flags & TH_SYN) 5308 end++; 5309 if (th->th_flags & TH_FIN) 5310 end++; 5311 } 5312 5313 if ((th->th_flags & TH_ACK) == 0) { 5314 /* Let it pass through the ack skew check */ 5315 ack = dst->seqlo; 5316 } else if ((ack == 0 && 5317 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5318 /* broken tcp stacks do not set ack */ 5319 (dst->state < TCPS_SYN_SENT)) { 5320 /* 5321 * Many stacks (ours included) will set the ACK number in an 5322 * FIN|ACK if the SYN times out -- no sequence to ACK. 5323 */ 5324 ack = dst->seqlo; 5325 } 5326 5327 if (seq == end) { 5328 /* Ease sequencing restrictions on no data packets */ 5329 seq = src->seqlo; 5330 end = seq; 5331 } 5332 5333 ackskew = dst->seqlo - ack; 5334 5335 /* 5336 * Need to demodulate the sequence numbers in any TCP SACK options 5337 * (Selective ACK). We could optionally validate the SACK values 5338 * against the current ACK window, either forwards or backwards, but 5339 * I'm not confident that SACK has been implemented properly 5340 * everywhere. It wouldn't surprise me if several stacks accidentally 5341 * SACK too far backwards of previously ACKed data. There really aren't 5342 * any security implications of bad SACKing unless the target stack 5343 * doesn't validate the option length correctly. Someone trying to 5344 * spoof into a TCP connection won't bother blindly sending SACK 5345 * options anyway. 5346 */ 5347 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5348 if (pf_modulate_sack(m, off, pd, th, dst)) 5349 *copyback = 1; 5350 } 5351 5352 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5353 if (SEQ_GEQ(src->seqhi, end) && 5354 /* Last octet inside other's window space */ 5355 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5356 /* Retrans: not more than one window back */ 5357 (ackskew >= -MAXACKWINDOW) && 5358 /* Acking not more than one reassembled fragment backwards */ 5359 (ackskew <= (MAXACKWINDOW << sws)) && 5360 /* Acking not more than one window forward */ 5361 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5362 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 5363 (pd->flags & PFDESC_IP_REAS) == 0)) { 5364 /* Require an exact/+1 sequence match on resets when possible */ 5365 5366 if (dst->scrub || src->scrub) { 5367 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5368 *state, src, dst, copyback)) 5369 return (PF_DROP); 5370 } 5371 5372 /* update max window */ 5373 if (src->max_win < win) 5374 src->max_win = win; 5375 /* synchronize sequencing */ 5376 if (SEQ_GT(end, src->seqlo)) 5377 src->seqlo = end; 5378 /* slide the window of what the other end can send */ 5379 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5380 dst->seqhi = ack + MAX((win << sws), 1); 5381 5382 /* update states */ 5383 if (th->th_flags & TH_SYN) 5384 if (src->state < TCPS_SYN_SENT) 5385 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5386 if (th->th_flags & TH_FIN) 5387 if (src->state < TCPS_CLOSING) 5388 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5389 if (th->th_flags & TH_ACK) { 5390 if (dst->state == TCPS_SYN_SENT) { 5391 pf_set_protostate(*state, pdst, 5392 TCPS_ESTABLISHED); 5393 if (src->state == TCPS_ESTABLISHED && 5394 (*state)->src_node != NULL && 5395 pf_src_connlimit(state)) { 5396 REASON_SET(reason, PFRES_SRCLIMIT); 5397 return (PF_DROP); 5398 } 5399 } else if (dst->state == TCPS_CLOSING) 5400 pf_set_protostate(*state, pdst, 5401 TCPS_FIN_WAIT_2); 5402 } 5403 if (th->th_flags & TH_RST) 5404 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5405 5406 /* update expire time */ 5407 (*state)->expire = time_uptime; 5408 if (src->state >= TCPS_FIN_WAIT_2 && 5409 dst->state >= TCPS_FIN_WAIT_2) 5410 (*state)->timeout = PFTM_TCP_CLOSED; 5411 else if (src->state >= TCPS_CLOSING && 5412 dst->state >= TCPS_CLOSING) 5413 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5414 else if (src->state < TCPS_ESTABLISHED || 5415 dst->state < TCPS_ESTABLISHED) 5416 (*state)->timeout = PFTM_TCP_OPENING; 5417 else if (src->state >= TCPS_CLOSING || 5418 dst->state >= TCPS_CLOSING) 5419 (*state)->timeout = PFTM_TCP_CLOSING; 5420 else 5421 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5422 5423 /* Fall through to PASS packet */ 5424 5425 } else if ((dst->state < TCPS_SYN_SENT || 5426 dst->state >= TCPS_FIN_WAIT_2 || 5427 src->state >= TCPS_FIN_WAIT_2) && 5428 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5429 /* Within a window forward of the originating packet */ 5430 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5431 /* Within a window backward of the originating packet */ 5432 5433 /* 5434 * This currently handles three situations: 5435 * 1) Stupid stacks will shotgun SYNs before their peer 5436 * replies. 5437 * 2) When PF catches an already established stream (the 5438 * firewall rebooted, the state table was flushed, routes 5439 * changed...) 5440 * 3) Packets get funky immediately after the connection 5441 * closes (this should catch Solaris spurious ACK|FINs 5442 * that web servers like to spew after a close) 5443 * 5444 * This must be a little more careful than the above code 5445 * since packet floods will also be caught here. We don't 5446 * update the TTL here to mitigate the damage of a packet 5447 * flood and so the same code can handle awkward establishment 5448 * and a loosened connection close. 5449 * In the establishment case, a correct peer response will 5450 * validate the connection, go through the normal state code 5451 * and keep updating the state TTL. 5452 */ 5453 5454 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5455 printf("pf: loose state match: "); 5456 pf_print_state(*state); 5457 pf_print_flags(th->th_flags); 5458 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5459 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5460 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5461 (unsigned long long)(*state)->packets[1], 5462 pd->dir == PF_IN ? "in" : "out", 5463 pd->dir == (*state)->direction ? "fwd" : "rev"); 5464 } 5465 5466 if (dst->scrub || src->scrub) { 5467 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5468 *state, src, dst, copyback)) 5469 return (PF_DROP); 5470 } 5471 5472 /* update max window */ 5473 if (src->max_win < win) 5474 src->max_win = win; 5475 /* synchronize sequencing */ 5476 if (SEQ_GT(end, src->seqlo)) 5477 src->seqlo = end; 5478 /* slide the window of what the other end can send */ 5479 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5480 dst->seqhi = ack + MAX((win << sws), 1); 5481 5482 /* 5483 * Cannot set dst->seqhi here since this could be a shotgunned 5484 * SYN and not an already established connection. 5485 */ 5486 5487 if (th->th_flags & TH_FIN) 5488 if (src->state < TCPS_CLOSING) 5489 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5490 if (th->th_flags & TH_RST) 5491 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5492 5493 /* Fall through to PASS packet */ 5494 5495 } else { 5496 if ((*state)->dst.state == TCPS_SYN_SENT && 5497 (*state)->src.state == TCPS_SYN_SENT) { 5498 /* Send RST for state mismatches during handshake */ 5499 if (!(th->th_flags & TH_RST)) 5500 pf_send_tcp((*state)->rule.ptr, pd->af, 5501 pd->dst, pd->src, th->th_dport, 5502 th->th_sport, ntohl(th->th_ack), 0, 5503 TH_RST, 0, 0, 5504 (*state)->rule.ptr->return_ttl, true, 0, 0, 5505 (*state)->act.rtableid); 5506 src->seqlo = 0; 5507 src->seqhi = 1; 5508 src->max_win = 1; 5509 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5510 printf("pf: BAD state: "); 5511 pf_print_state(*state); 5512 pf_print_flags(th->th_flags); 5513 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5514 "pkts=%llu:%llu dir=%s,%s\n", 5515 seq, orig_seq, ack, pd->p_len, ackskew, 5516 (unsigned long long)(*state)->packets[0], 5517 (unsigned long long)(*state)->packets[1], 5518 pd->dir == PF_IN ? "in" : "out", 5519 pd->dir == (*state)->direction ? "fwd" : "rev"); 5520 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5521 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5522 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5523 ' ': '2', 5524 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5525 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5526 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5527 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5528 } 5529 REASON_SET(reason, PFRES_BADSTATE); 5530 return (PF_DROP); 5531 } 5532 5533 return (PF_PASS); 5534 } 5535 5536 static int 5537 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5538 { 5539 struct tcphdr *th = &pd->hdr.tcp; 5540 struct pf_state_peer *src, *dst; 5541 u_int8_t psrc, pdst; 5542 5543 if (pd->dir == (*state)->direction) { 5544 src = &(*state)->src; 5545 dst = &(*state)->dst; 5546 psrc = PF_PEER_SRC; 5547 pdst = PF_PEER_DST; 5548 } else { 5549 src = &(*state)->dst; 5550 dst = &(*state)->src; 5551 psrc = PF_PEER_DST; 5552 pdst = PF_PEER_SRC; 5553 } 5554 5555 if (th->th_flags & TH_SYN) 5556 if (src->state < TCPS_SYN_SENT) 5557 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5558 if (th->th_flags & TH_FIN) 5559 if (src->state < TCPS_CLOSING) 5560 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5561 if (th->th_flags & TH_ACK) { 5562 if (dst->state == TCPS_SYN_SENT) { 5563 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5564 if (src->state == TCPS_ESTABLISHED && 5565 (*state)->src_node != NULL && 5566 pf_src_connlimit(state)) { 5567 REASON_SET(reason, PFRES_SRCLIMIT); 5568 return (PF_DROP); 5569 } 5570 } else if (dst->state == TCPS_CLOSING) { 5571 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5572 } else if (src->state == TCPS_SYN_SENT && 5573 dst->state < TCPS_SYN_SENT) { 5574 /* 5575 * Handle a special sloppy case where we only see one 5576 * half of the connection. If there is a ACK after 5577 * the initial SYN without ever seeing a packet from 5578 * the destination, set the connection to established. 5579 */ 5580 pf_set_protostate(*state, PF_PEER_BOTH, 5581 TCPS_ESTABLISHED); 5582 dst->state = src->state = TCPS_ESTABLISHED; 5583 if ((*state)->src_node != NULL && 5584 pf_src_connlimit(state)) { 5585 REASON_SET(reason, PFRES_SRCLIMIT); 5586 return (PF_DROP); 5587 } 5588 } else if (src->state == TCPS_CLOSING && 5589 dst->state == TCPS_ESTABLISHED && 5590 dst->seqlo == 0) { 5591 /* 5592 * Handle the closing of half connections where we 5593 * don't see the full bidirectional FIN/ACK+ACK 5594 * handshake. 5595 */ 5596 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5597 } 5598 } 5599 if (th->th_flags & TH_RST) 5600 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5601 5602 /* update expire time */ 5603 (*state)->expire = time_uptime; 5604 if (src->state >= TCPS_FIN_WAIT_2 && 5605 dst->state >= TCPS_FIN_WAIT_2) 5606 (*state)->timeout = PFTM_TCP_CLOSED; 5607 else if (src->state >= TCPS_CLOSING && 5608 dst->state >= TCPS_CLOSING) 5609 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5610 else if (src->state < TCPS_ESTABLISHED || 5611 dst->state < TCPS_ESTABLISHED) 5612 (*state)->timeout = PFTM_TCP_OPENING; 5613 else if (src->state >= TCPS_CLOSING || 5614 dst->state >= TCPS_CLOSING) 5615 (*state)->timeout = PFTM_TCP_CLOSING; 5616 else 5617 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5618 5619 return (PF_PASS); 5620 } 5621 5622 static int 5623 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5624 { 5625 struct pf_state_key *sk = (*state)->key[pd->didx]; 5626 struct tcphdr *th = &pd->hdr.tcp; 5627 5628 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5629 if (pd->dir != (*state)->direction) { 5630 REASON_SET(reason, PFRES_SYNPROXY); 5631 return (PF_SYNPROXY_DROP); 5632 } 5633 if (th->th_flags & TH_SYN) { 5634 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5635 REASON_SET(reason, PFRES_SYNPROXY); 5636 return (PF_DROP); 5637 } 5638 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5639 pd->src, th->th_dport, th->th_sport, 5640 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5641 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 5642 (*state)->act.rtableid); 5643 REASON_SET(reason, PFRES_SYNPROXY); 5644 return (PF_SYNPROXY_DROP); 5645 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5646 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5647 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5648 REASON_SET(reason, PFRES_SYNPROXY); 5649 return (PF_DROP); 5650 } else if ((*state)->src_node != NULL && 5651 pf_src_connlimit(state)) { 5652 REASON_SET(reason, PFRES_SRCLIMIT); 5653 return (PF_DROP); 5654 } else 5655 pf_set_protostate(*state, PF_PEER_SRC, 5656 PF_TCPS_PROXY_DST); 5657 } 5658 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5659 if (pd->dir == (*state)->direction) { 5660 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5661 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5662 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5663 REASON_SET(reason, PFRES_SYNPROXY); 5664 return (PF_DROP); 5665 } 5666 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5667 if ((*state)->dst.seqhi == 1) 5668 (*state)->dst.seqhi = htonl(arc4random()); 5669 pf_send_tcp((*state)->rule.ptr, pd->af, 5670 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5671 sk->port[pd->sidx], sk->port[pd->didx], 5672 (*state)->dst.seqhi, 0, TH_SYN, 0, 5673 (*state)->src.mss, 0, false, (*state)->tag, 0, 5674 (*state)->act.rtableid); 5675 REASON_SET(reason, PFRES_SYNPROXY); 5676 return (PF_SYNPROXY_DROP); 5677 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5678 (TH_SYN|TH_ACK)) || 5679 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5680 REASON_SET(reason, PFRES_SYNPROXY); 5681 return (PF_DROP); 5682 } else { 5683 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5684 (*state)->dst.seqlo = ntohl(th->th_seq); 5685 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5686 pd->src, th->th_dport, th->th_sport, 5687 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5688 TH_ACK, (*state)->src.max_win, 0, 0, false, 5689 (*state)->tag, 0, (*state)->act.rtableid); 5690 pf_send_tcp((*state)->rule.ptr, pd->af, 5691 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5692 sk->port[pd->sidx], sk->port[pd->didx], 5693 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5694 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 5695 (*state)->act.rtableid); 5696 (*state)->src.seqdiff = (*state)->dst.seqhi - 5697 (*state)->src.seqlo; 5698 (*state)->dst.seqdiff = (*state)->src.seqhi - 5699 (*state)->dst.seqlo; 5700 (*state)->src.seqhi = (*state)->src.seqlo + 5701 (*state)->dst.max_win; 5702 (*state)->dst.seqhi = (*state)->dst.seqlo + 5703 (*state)->src.max_win; 5704 (*state)->src.wscale = (*state)->dst.wscale = 0; 5705 pf_set_protostate(*state, PF_PEER_BOTH, 5706 TCPS_ESTABLISHED); 5707 REASON_SET(reason, PFRES_SYNPROXY); 5708 return (PF_SYNPROXY_DROP); 5709 } 5710 } 5711 5712 return (PF_PASS); 5713 } 5714 5715 static int 5716 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 5717 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5718 u_short *reason) 5719 { 5720 struct pf_state_key_cmp key; 5721 struct tcphdr *th = &pd->hdr.tcp; 5722 int copyback = 0; 5723 int action; 5724 struct pf_state_peer *src, *dst; 5725 5726 bzero(&key, sizeof(key)); 5727 key.af = pd->af; 5728 key.proto = IPPROTO_TCP; 5729 if (pd->dir == PF_IN) { /* wire side, straight */ 5730 PF_ACPY(&key.addr[0], pd->src, key.af); 5731 PF_ACPY(&key.addr[1], pd->dst, key.af); 5732 key.port[0] = th->th_sport; 5733 key.port[1] = th->th_dport; 5734 } else { /* stack side, reverse */ 5735 PF_ACPY(&key.addr[1], pd->src, key.af); 5736 PF_ACPY(&key.addr[0], pd->dst, key.af); 5737 key.port[1] = th->th_sport; 5738 key.port[0] = th->th_dport; 5739 } 5740 5741 STATE_LOOKUP(kif, &key, *state, pd); 5742 5743 if (pd->dir == (*state)->direction) { 5744 src = &(*state)->src; 5745 dst = &(*state)->dst; 5746 } else { 5747 src = &(*state)->dst; 5748 dst = &(*state)->src; 5749 } 5750 5751 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5752 return (action); 5753 5754 if (dst->state >= TCPS_FIN_WAIT_2 && 5755 src->state >= TCPS_FIN_WAIT_2 && 5756 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 5757 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 5758 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 5759 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5760 printf("pf: state reuse "); 5761 pf_print_state(*state); 5762 pf_print_flags(th->th_flags); 5763 printf("\n"); 5764 } 5765 /* XXX make sure it's the same direction ?? */ 5766 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5767 pf_unlink_state(*state); 5768 *state = NULL; 5769 return (PF_DROP); 5770 } 5771 5772 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5773 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5774 return (PF_DROP); 5775 } else { 5776 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5777 ©back) == PF_DROP) 5778 return (PF_DROP); 5779 } 5780 5781 /* translate source/destination address, if necessary */ 5782 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5783 struct pf_state_key *nk = (*state)->key[pd->didx]; 5784 5785 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5786 nk->port[pd->sidx] != th->th_sport) 5787 pf_change_ap(m, pd->src, &th->th_sport, 5788 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5789 nk->port[pd->sidx], 0, pd->af); 5790 5791 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5792 nk->port[pd->didx] != th->th_dport) 5793 pf_change_ap(m, pd->dst, &th->th_dport, 5794 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5795 nk->port[pd->didx], 0, pd->af); 5796 copyback = 1; 5797 } 5798 5799 /* Copyback sequence modulation or stateful scrub changes if needed */ 5800 if (copyback) 5801 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5802 5803 return (PF_PASS); 5804 } 5805 5806 static int 5807 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 5808 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5809 { 5810 struct pf_state_peer *src, *dst; 5811 struct pf_state_key_cmp key; 5812 struct udphdr *uh = &pd->hdr.udp; 5813 uint8_t psrc, pdst; 5814 5815 bzero(&key, sizeof(key)); 5816 key.af = pd->af; 5817 key.proto = IPPROTO_UDP; 5818 if (pd->dir == PF_IN) { /* wire side, straight */ 5819 PF_ACPY(&key.addr[0], pd->src, key.af); 5820 PF_ACPY(&key.addr[1], pd->dst, key.af); 5821 key.port[0] = uh->uh_sport; 5822 key.port[1] = uh->uh_dport; 5823 } else { /* stack side, reverse */ 5824 PF_ACPY(&key.addr[1], pd->src, key.af); 5825 PF_ACPY(&key.addr[0], pd->dst, key.af); 5826 key.port[1] = uh->uh_sport; 5827 key.port[0] = uh->uh_dport; 5828 } 5829 5830 STATE_LOOKUP(kif, &key, *state, pd); 5831 5832 if (pd->dir == (*state)->direction) { 5833 src = &(*state)->src; 5834 dst = &(*state)->dst; 5835 psrc = PF_PEER_SRC; 5836 pdst = PF_PEER_DST; 5837 } else { 5838 src = &(*state)->dst; 5839 dst = &(*state)->src; 5840 psrc = PF_PEER_DST; 5841 pdst = PF_PEER_SRC; 5842 } 5843 5844 /* update states */ 5845 if (src->state < PFUDPS_SINGLE) 5846 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5847 if (dst->state == PFUDPS_SINGLE) 5848 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5849 5850 /* update expire time */ 5851 (*state)->expire = time_uptime; 5852 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5853 (*state)->timeout = PFTM_UDP_MULTIPLE; 5854 else 5855 (*state)->timeout = PFTM_UDP_SINGLE; 5856 5857 /* translate source/destination address, if necessary */ 5858 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5859 struct pf_state_key *nk = (*state)->key[pd->didx]; 5860 5861 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5862 nk->port[pd->sidx] != uh->uh_sport) 5863 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5864 &uh->uh_sum, &nk->addr[pd->sidx], 5865 nk->port[pd->sidx], 1, pd->af); 5866 5867 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5868 nk->port[pd->didx] != uh->uh_dport) 5869 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5870 &uh->uh_sum, &nk->addr[pd->didx], 5871 nk->port[pd->didx], 1, pd->af); 5872 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5873 } 5874 5875 return (PF_PASS); 5876 } 5877 5878 static int 5879 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 5880 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5881 { 5882 struct pf_state_key_cmp key; 5883 struct pf_state_peer *src, *dst; 5884 struct sctphdr *sh = &pd->hdr.sctp; 5885 u_int8_t psrc; //, pdst; 5886 5887 bzero(&key, sizeof(key)); 5888 key.af = pd->af; 5889 key.proto = IPPROTO_SCTP; 5890 if (pd->dir == PF_IN) { /* wire side, straight */ 5891 PF_ACPY(&key.addr[0], pd->src, key.af); 5892 PF_ACPY(&key.addr[1], pd->dst, key.af); 5893 key.port[0] = sh->src_port; 5894 key.port[1] = sh->dest_port; 5895 } else { /* stack side, reverse */ 5896 PF_ACPY(&key.addr[1], pd->src, key.af); 5897 PF_ACPY(&key.addr[0], pd->dst, key.af); 5898 key.port[1] = sh->src_port; 5899 key.port[0] = sh->dest_port; 5900 } 5901 5902 STATE_LOOKUP(kif, &key, *state, pd); 5903 5904 if (pd->dir == (*state)->direction) { 5905 src = &(*state)->src; 5906 dst = &(*state)->dst; 5907 psrc = PF_PEER_SRC; 5908 } else { 5909 src = &(*state)->dst; 5910 dst = &(*state)->src; 5911 psrc = PF_PEER_DST; 5912 } 5913 5914 /* Track state. */ 5915 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 5916 if (src->state < SCTP_COOKIE_WAIT) { 5917 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 5918 (*state)->timeout = PFTM_SCTP_OPENING; 5919 } 5920 } 5921 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 5922 MPASS(dst->scrub != NULL); 5923 if (dst->scrub->pfss_v_tag == 0) 5924 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 5925 } 5926 5927 if (pd->sctp_flags & PFDESC_SCTP_COOKIE) { 5928 if (src->state < SCTP_ESTABLISHED) { 5929 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 5930 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 5931 } 5932 } 5933 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 5934 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5935 if (src->state < SCTP_SHUTDOWN_PENDING) { 5936 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 5937 (*state)->timeout = PFTM_SCTP_CLOSING; 5938 } 5939 } 5940 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5941 pf_set_protostate(*state, psrc, SCTP_CLOSED); 5942 (*state)->timeout = PFTM_SCTP_CLOSED; 5943 } 5944 5945 if (src->scrub != NULL) { 5946 if (src->scrub->pfss_v_tag == 0) { 5947 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 5948 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 5949 return (PF_DROP); 5950 } 5951 5952 (*state)->expire = time_uptime; 5953 5954 /* translate source/destination address, if necessary */ 5955 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5956 uint16_t checksum = 0; 5957 struct pf_state_key *nk = (*state)->key[pd->didx]; 5958 5959 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5960 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 5961 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 5962 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 5963 nk->port[pd->sidx], 1, pd->af); 5964 } 5965 5966 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5967 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 5968 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 5969 pd->ip_sum, &checksum, &nk->addr[pd->didx], 5970 nk->port[pd->didx], 1, pd->af); 5971 } 5972 } 5973 5974 return (PF_PASS); 5975 } 5976 5977 static void 5978 pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 5979 { 5980 struct pf_sctp_endpoint key; 5981 struct pf_sctp_endpoint *ep; 5982 struct pf_state_key *sks = s->key[PF_SK_STACK]; 5983 struct pf_sctp_source *i, *tmp; 5984 5985 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 5986 return; 5987 5988 PF_SCTP_ENDPOINTS_LOCK(); 5989 5990 key.v_tag = s->dst.scrub->pfss_v_tag; 5991 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 5992 if (ep != NULL) { 5993 /* XXX Actually remove! */ 5994 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 5995 if (pf_addr_cmp(&i->addr, 5996 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 5997 s->key[PF_SK_WIRE]->af) == 0) { 5998 TAILQ_REMOVE(&ep->sources, i, entry); 5999 free(i, M_PFTEMP); 6000 break; 6001 } 6002 } 6003 6004 if (TAILQ_EMPTY(&ep->sources)) { 6005 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6006 free(ep, M_PFTEMP); 6007 } 6008 } 6009 6010 /* Other direction. */ 6011 key.v_tag = s->src.scrub->pfss_v_tag; 6012 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6013 if (ep != NULL) { 6014 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6015 if (pf_addr_cmp(&i->addr, 6016 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6017 s->key[PF_SK_WIRE]->af) == 0) { 6018 TAILQ_REMOVE(&ep->sources, i, entry); 6019 free(i, M_PFTEMP); 6020 break; 6021 } 6022 } 6023 6024 if (TAILQ_EMPTY(&ep->sources)) { 6025 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6026 free(ep, M_PFTEMP); 6027 } 6028 } 6029 6030 PF_SCTP_ENDPOINTS_UNLOCK(); 6031 } 6032 6033 static void 6034 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6035 { 6036 struct pf_sctp_endpoint key = { 6037 .v_tag = v_tag, 6038 }; 6039 struct pf_sctp_source *i; 6040 struct pf_sctp_endpoint *ep; 6041 6042 PF_SCTP_ENDPOINTS_LOCK(); 6043 6044 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6045 if (ep == NULL) { 6046 ep = malloc(sizeof(struct pf_sctp_endpoint), 6047 M_PFTEMP, M_NOWAIT); 6048 if (ep == NULL) { 6049 PF_SCTP_ENDPOINTS_UNLOCK(); 6050 return; 6051 } 6052 6053 ep->v_tag = v_tag; 6054 TAILQ_INIT(&ep->sources); 6055 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6056 } 6057 6058 /* Avoid inserting duplicates. */ 6059 TAILQ_FOREACH(i, &ep->sources, entry) { 6060 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6061 PF_SCTP_ENDPOINTS_UNLOCK(); 6062 return; 6063 } 6064 } 6065 6066 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6067 if (i == NULL) { 6068 PF_SCTP_ENDPOINTS_UNLOCK(); 6069 return; 6070 } 6071 6072 i->af = pd->af; 6073 memcpy(&i->addr, a, sizeof(*a)); 6074 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6075 6076 PF_SCTP_ENDPOINTS_UNLOCK(); 6077 } 6078 6079 static void 6080 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6081 struct pf_kstate *s, int action) 6082 { 6083 struct pf_sctp_multihome_job *j, *tmp; 6084 struct pf_sctp_source *i; 6085 int ret __unused;; 6086 struct pf_kstate *sm = NULL; 6087 struct pf_krule *ra = NULL; 6088 struct pf_krule *r = &V_pf_default_rule; 6089 struct pf_kruleset *rs = NULL; 6090 bool do_extra = true; 6091 6092 PF_RULES_RLOCK_TRACKER; 6093 6094 again: 6095 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6096 if (s == NULL || action != PF_PASS) 6097 goto free; 6098 6099 /* Confirm we don't recurse here. */ 6100 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6101 6102 switch (j->op) { 6103 case SCTP_ADD_IP_ADDRESS: { 6104 uint32_t v_tag = pd->sctp_initiate_tag; 6105 6106 if (v_tag == 0) { 6107 if (s->direction == pd->dir) 6108 v_tag = s->src.scrub->pfss_v_tag; 6109 else 6110 v_tag = s->dst.scrub->pfss_v_tag; 6111 } 6112 6113 /* 6114 * Avoid duplicating states. We'll already have 6115 * created a state based on the source address of 6116 * the packet, but SCTP endpoints may also list this 6117 * address again in the INIT(_ACK) parameters. 6118 */ 6119 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6120 break; 6121 } 6122 6123 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6124 PF_RULES_RLOCK(); 6125 sm = NULL; 6126 /* XXX: May generated unwanted abort if we try to insert a duplicate state. */ 6127 ret = pf_test_rule(&r, &sm, kif, 6128 j->m, off, &j->pd, &ra, &rs, NULL); 6129 PF_RULES_RUNLOCK(); 6130 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6131 if (ret != PF_DROP && sm != NULL) { 6132 /* Inherit v_tag values. */ 6133 if (sm->direction == s->direction) { 6134 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6135 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6136 } else { 6137 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6138 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6139 } 6140 PF_STATE_UNLOCK(sm); 6141 } else { 6142 /* If we try duplicate inserts? */ 6143 break; 6144 } 6145 6146 /* Only add the addres if we've actually allowed the state. */ 6147 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6148 6149 if (! do_extra) { 6150 break; 6151 } 6152 /* 6153 * We need to do this for each of our source addresses. 6154 * Find those based on the verification tag. 6155 */ 6156 struct pf_sctp_endpoint key = { 6157 .v_tag = pd->hdr.sctp.v_tag, 6158 }; 6159 struct pf_sctp_endpoint *ep; 6160 6161 PF_SCTP_ENDPOINTS_LOCK(); 6162 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6163 if (ep == NULL) { 6164 PF_SCTP_ENDPOINTS_UNLOCK(); 6165 break; 6166 } 6167 MPASS(ep != NULL); 6168 6169 TAILQ_FOREACH(i, &ep->sources, entry) { 6170 struct pf_sctp_multihome_job *nj; 6171 6172 /* SCTP can intermingle IPv4 and IPv6. */ 6173 if (i->af != pd->af) 6174 continue; 6175 6176 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6177 if (! nj) { 6178 continue; 6179 } 6180 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6181 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6182 nj->pd.src = &nj->src; 6183 // New destination address! 6184 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6185 nj->pd.dst = &nj->dst; 6186 nj->m = j->m; 6187 nj->op = j->op; 6188 6189 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6190 } 6191 PF_SCTP_ENDPOINTS_UNLOCK(); 6192 6193 break; 6194 } 6195 case SCTP_DEL_IP_ADDRESS: { 6196 struct pf_state_key_cmp key; 6197 uint8_t psrc; 6198 6199 bzero(&key, sizeof(key)); 6200 key.af = j->pd.af; 6201 key.proto = IPPROTO_SCTP; 6202 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6203 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6204 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6205 key.port[0] = j->pd.hdr.sctp.src_port; 6206 key.port[1] = j->pd.hdr.sctp.dest_port; 6207 } else { /* stack side, reverse */ 6208 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6209 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6210 key.port[1] = j->pd.hdr.sctp.src_port; 6211 key.port[0] = j->pd.hdr.sctp.dest_port; 6212 } 6213 6214 sm = pf_find_state(kif, &key, j->pd.dir); 6215 if (sm != NULL) { 6216 PF_STATE_LOCK_ASSERT(sm); 6217 if (j->pd.dir == sm->direction) { 6218 psrc = PF_PEER_SRC; 6219 } else { 6220 psrc = PF_PEER_DST; 6221 } 6222 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6223 sm->timeout = PFTM_SCTP_CLOSING; 6224 PF_STATE_UNLOCK(sm); 6225 } 6226 break; 6227 default: 6228 panic("Unknown op %#x", j->op); 6229 } 6230 } 6231 6232 free: 6233 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6234 free(j, M_PFTEMP); 6235 } 6236 6237 /* We may have inserted extra work while processing the list. */ 6238 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6239 do_extra = false; 6240 goto again; 6241 } 6242 } 6243 6244 static int 6245 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6246 struct pfi_kkif *kif, int op) 6247 { 6248 int off = 0; 6249 struct pf_sctp_multihome_job *job; 6250 6251 while (off < len) { 6252 struct sctp_paramhdr h; 6253 6254 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6255 pd->af)) 6256 return (PF_DROP); 6257 6258 /* Parameters are at least 4 bytes. */ 6259 if (ntohs(h.param_length) < 4) 6260 return (PF_DROP); 6261 6262 switch (ntohs(h.param_type)) { 6263 case SCTP_IPV4_ADDRESS: { 6264 struct in_addr t; 6265 6266 if (ntohs(h.param_length) != 6267 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6268 return (PF_DROP); 6269 6270 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6271 NULL, NULL, pd->af)) 6272 return (PF_DROP); 6273 6274 if (in_nullhost(t)) 6275 t.s_addr = pd->src->v4.s_addr; 6276 6277 /* 6278 * We hold the state lock (idhash) here, which means 6279 * that we can't acquire the keyhash, or we'll get a 6280 * LOR (and potentially double-lock things too). We also 6281 * can't release the state lock here, so instead we'll 6282 * enqueue this for async handling. 6283 * There's a relatively small race here, in that a 6284 * packet using the new addresses could arrive already, 6285 * but that's just though luck for it. 6286 */ 6287 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6288 if (! job) 6289 return (PF_DROP); 6290 6291 memcpy(&job->pd, pd, sizeof(*pd)); 6292 6293 // New source address! 6294 memcpy(&job->src, &t, sizeof(t)); 6295 job->pd.src = &job->src; 6296 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6297 job->pd.dst = &job->dst; 6298 job->m = m; 6299 job->op = op; 6300 6301 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6302 break; 6303 } 6304 #ifdef INET6 6305 case SCTP_IPV6_ADDRESS: { 6306 struct in6_addr t; 6307 6308 if (ntohs(h.param_length) != 6309 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6310 return (PF_DROP); 6311 6312 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6313 NULL, NULL, pd->af)) 6314 return (PF_DROP); 6315 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6316 break; 6317 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6318 memcpy(&t, &pd->src->v6, sizeof(t)); 6319 6320 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6321 if (! job) 6322 return (PF_DROP); 6323 6324 memcpy(&job->pd, pd, sizeof(*pd)); 6325 memcpy(&job->src, &t, sizeof(t)); 6326 job->pd.src = &job->src; 6327 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6328 job->pd.dst = &job->dst; 6329 job->m = m; 6330 job->op = op; 6331 6332 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6333 break; 6334 } 6335 #endif 6336 case SCTP_ADD_IP_ADDRESS: { 6337 int ret; 6338 struct sctp_asconf_paramhdr ah; 6339 6340 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6341 NULL, NULL, pd->af)) 6342 return (PF_DROP); 6343 6344 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6345 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6346 SCTP_ADD_IP_ADDRESS); 6347 if (ret != PF_PASS) 6348 return (ret); 6349 break; 6350 } 6351 case SCTP_DEL_IP_ADDRESS: { 6352 int ret; 6353 struct sctp_asconf_paramhdr ah; 6354 6355 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6356 NULL, NULL, pd->af)) 6357 return (PF_DROP); 6358 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6359 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6360 SCTP_DEL_IP_ADDRESS); 6361 if (ret != PF_PASS) 6362 return (ret); 6363 break; 6364 } 6365 default: 6366 break; 6367 } 6368 6369 off += roundup(ntohs(h.param_length), 4); 6370 } 6371 6372 return (PF_PASS); 6373 } 6374 int 6375 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6376 struct pfi_kkif *kif) 6377 { 6378 start += sizeof(struct sctp_init_chunk); 6379 len -= sizeof(struct sctp_init_chunk); 6380 6381 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6382 } 6383 6384 int 6385 pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6386 struct pf_pdesc *pd, struct pfi_kkif *kif) 6387 { 6388 start += sizeof(struct sctp_asconf_chunk); 6389 len -= sizeof(struct sctp_asconf_chunk); 6390 6391 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6392 } 6393 6394 static int 6395 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6396 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6397 { 6398 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6399 u_int16_t icmpid = 0, *icmpsum; 6400 u_int8_t icmptype, icmpcode; 6401 int state_icmp = 0; 6402 struct pf_state_key_cmp key; 6403 6404 bzero(&key, sizeof(key)); 6405 switch (pd->proto) { 6406 #ifdef INET 6407 case IPPROTO_ICMP: 6408 icmptype = pd->hdr.icmp.icmp_type; 6409 icmpcode = pd->hdr.icmp.icmp_code; 6410 icmpid = pd->hdr.icmp.icmp_id; 6411 icmpsum = &pd->hdr.icmp.icmp_cksum; 6412 6413 if (icmptype == ICMP_UNREACH || 6414 icmptype == ICMP_SOURCEQUENCH || 6415 icmptype == ICMP_REDIRECT || 6416 icmptype == ICMP_TIMXCEED || 6417 icmptype == ICMP_PARAMPROB) 6418 state_icmp++; 6419 break; 6420 #endif /* INET */ 6421 #ifdef INET6 6422 case IPPROTO_ICMPV6: 6423 icmptype = pd->hdr.icmp6.icmp6_type; 6424 icmpcode = pd->hdr.icmp6.icmp6_code; 6425 icmpid = pd->hdr.icmp6.icmp6_id; 6426 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6427 6428 if (icmptype == ICMP6_DST_UNREACH || 6429 icmptype == ICMP6_PACKET_TOO_BIG || 6430 icmptype == ICMP6_TIME_EXCEEDED || 6431 icmptype == ICMP6_PARAM_PROB) 6432 state_icmp++; 6433 break; 6434 #endif /* INET6 */ 6435 } 6436 6437 if (!state_icmp) { 6438 /* 6439 * ICMP query/reply message not related to a TCP/UDP packet. 6440 * Search for an ICMP state. 6441 */ 6442 key.af = pd->af; 6443 key.proto = pd->proto; 6444 key.port[0] = key.port[1] = icmpid; 6445 if (pd->dir == PF_IN) { /* wire side, straight */ 6446 PF_ACPY(&key.addr[0], pd->src, key.af); 6447 PF_ACPY(&key.addr[1], pd->dst, key.af); 6448 } else { /* stack side, reverse */ 6449 PF_ACPY(&key.addr[1], pd->src, key.af); 6450 PF_ACPY(&key.addr[0], pd->dst, key.af); 6451 } 6452 6453 STATE_LOOKUP(kif, &key, *state, pd); 6454 6455 (*state)->expire = time_uptime; 6456 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6457 6458 /* translate source/destination address, if necessary */ 6459 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6460 struct pf_state_key *nk = (*state)->key[pd->didx]; 6461 6462 switch (pd->af) { 6463 #ifdef INET 6464 case AF_INET: 6465 if (PF_ANEQ(pd->src, 6466 &nk->addr[pd->sidx], AF_INET)) 6467 pf_change_a(&saddr->v4.s_addr, 6468 pd->ip_sum, 6469 nk->addr[pd->sidx].v4.s_addr, 0); 6470 6471 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6472 AF_INET)) 6473 pf_change_a(&daddr->v4.s_addr, 6474 pd->ip_sum, 6475 nk->addr[pd->didx].v4.s_addr, 0); 6476 6477 if (nk->port[0] != 6478 pd->hdr.icmp.icmp_id) { 6479 pd->hdr.icmp.icmp_cksum = 6480 pf_cksum_fixup( 6481 pd->hdr.icmp.icmp_cksum, icmpid, 6482 nk->port[pd->sidx], 0); 6483 pd->hdr.icmp.icmp_id = 6484 nk->port[pd->sidx]; 6485 } 6486 6487 m_copyback(m, off, ICMP_MINLEN, 6488 (caddr_t )&pd->hdr.icmp); 6489 break; 6490 #endif /* INET */ 6491 #ifdef INET6 6492 case AF_INET6: 6493 if (PF_ANEQ(pd->src, 6494 &nk->addr[pd->sidx], AF_INET6)) 6495 pf_change_a6(saddr, 6496 &pd->hdr.icmp6.icmp6_cksum, 6497 &nk->addr[pd->sidx], 0); 6498 6499 if (PF_ANEQ(pd->dst, 6500 &nk->addr[pd->didx], AF_INET6)) 6501 pf_change_a6(daddr, 6502 &pd->hdr.icmp6.icmp6_cksum, 6503 &nk->addr[pd->didx], 0); 6504 6505 m_copyback(m, off, sizeof(struct icmp6_hdr), 6506 (caddr_t )&pd->hdr.icmp6); 6507 break; 6508 #endif /* INET6 */ 6509 } 6510 } 6511 return (PF_PASS); 6512 6513 } else { 6514 /* 6515 * ICMP error message in response to a TCP/UDP packet. 6516 * Extract the inner TCP/UDP header and search for that state. 6517 */ 6518 6519 struct pf_pdesc pd2; 6520 bzero(&pd2, sizeof pd2); 6521 #ifdef INET 6522 struct ip h2; 6523 #endif /* INET */ 6524 #ifdef INET6 6525 struct ip6_hdr h2_6; 6526 int terminal = 0; 6527 #endif /* INET6 */ 6528 int ipoff2 = 0; 6529 int off2 = 0; 6530 6531 pd2.af = pd->af; 6532 /* Payload packet is from the opposite direction. */ 6533 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 6534 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 6535 switch (pd->af) { 6536 #ifdef INET 6537 case AF_INET: 6538 /* offset of h2 in mbuf chain */ 6539 ipoff2 = off + ICMP_MINLEN; 6540 6541 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 6542 NULL, reason, pd2.af)) { 6543 DPFPRINTF(PF_DEBUG_MISC, 6544 ("pf: ICMP error message too short " 6545 "(ip)\n")); 6546 return (PF_DROP); 6547 } 6548 /* 6549 * ICMP error messages don't refer to non-first 6550 * fragments 6551 */ 6552 if (h2.ip_off & htons(IP_OFFMASK)) { 6553 REASON_SET(reason, PFRES_FRAG); 6554 return (PF_DROP); 6555 } 6556 6557 /* offset of protocol header that follows h2 */ 6558 off2 = ipoff2 + (h2.ip_hl << 2); 6559 6560 pd2.proto = h2.ip_p; 6561 pd2.src = (struct pf_addr *)&h2.ip_src; 6562 pd2.dst = (struct pf_addr *)&h2.ip_dst; 6563 pd2.ip_sum = &h2.ip_sum; 6564 break; 6565 #endif /* INET */ 6566 #ifdef INET6 6567 case AF_INET6: 6568 ipoff2 = off + sizeof(struct icmp6_hdr); 6569 6570 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 6571 NULL, reason, pd2.af)) { 6572 DPFPRINTF(PF_DEBUG_MISC, 6573 ("pf: ICMP error message too short " 6574 "(ip6)\n")); 6575 return (PF_DROP); 6576 } 6577 pd2.proto = h2_6.ip6_nxt; 6578 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 6579 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 6580 pd2.ip_sum = NULL; 6581 off2 = ipoff2 + sizeof(h2_6); 6582 do { 6583 switch (pd2.proto) { 6584 case IPPROTO_FRAGMENT: 6585 /* 6586 * ICMPv6 error messages for 6587 * non-first fragments 6588 */ 6589 REASON_SET(reason, PFRES_FRAG); 6590 return (PF_DROP); 6591 case IPPROTO_AH: 6592 case IPPROTO_HOPOPTS: 6593 case IPPROTO_ROUTING: 6594 case IPPROTO_DSTOPTS: { 6595 /* get next header and header length */ 6596 struct ip6_ext opt6; 6597 6598 if (!pf_pull_hdr(m, off2, &opt6, 6599 sizeof(opt6), NULL, reason, 6600 pd2.af)) { 6601 DPFPRINTF(PF_DEBUG_MISC, 6602 ("pf: ICMPv6 short opt\n")); 6603 return (PF_DROP); 6604 } 6605 if (pd2.proto == IPPROTO_AH) 6606 off2 += (opt6.ip6e_len + 2) * 4; 6607 else 6608 off2 += (opt6.ip6e_len + 1) * 8; 6609 pd2.proto = opt6.ip6e_nxt; 6610 /* goto the next header */ 6611 break; 6612 } 6613 default: 6614 terminal++; 6615 break; 6616 } 6617 } while (!terminal); 6618 break; 6619 #endif /* INET6 */ 6620 } 6621 6622 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 6623 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6624 printf("pf: BAD ICMP %d:%d outer dst: ", 6625 icmptype, icmpcode); 6626 pf_print_host(pd->src, 0, pd->af); 6627 printf(" -> "); 6628 pf_print_host(pd->dst, 0, pd->af); 6629 printf(" inner src: "); 6630 pf_print_host(pd2.src, 0, pd2.af); 6631 printf(" -> "); 6632 pf_print_host(pd2.dst, 0, pd2.af); 6633 printf("\n"); 6634 } 6635 REASON_SET(reason, PFRES_BADSTATE); 6636 return (PF_DROP); 6637 } 6638 6639 switch (pd2.proto) { 6640 case IPPROTO_TCP: { 6641 struct tcphdr th; 6642 u_int32_t seq; 6643 struct pf_state_peer *src, *dst; 6644 u_int8_t dws; 6645 int copyback = 0; 6646 6647 /* 6648 * Only the first 8 bytes of the TCP header can be 6649 * expected. Don't access any TCP header fields after 6650 * th_seq, an ackskew test is not possible. 6651 */ 6652 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 6653 pd2.af)) { 6654 DPFPRINTF(PF_DEBUG_MISC, 6655 ("pf: ICMP error message too short " 6656 "(tcp)\n")); 6657 return (PF_DROP); 6658 } 6659 6660 key.af = pd2.af; 6661 key.proto = IPPROTO_TCP; 6662 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6663 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6664 key.port[pd2.sidx] = th.th_sport; 6665 key.port[pd2.didx] = th.th_dport; 6666 6667 STATE_LOOKUP(kif, &key, *state, pd); 6668 6669 if (pd->dir == (*state)->direction) { 6670 src = &(*state)->dst; 6671 dst = &(*state)->src; 6672 } else { 6673 src = &(*state)->src; 6674 dst = &(*state)->dst; 6675 } 6676 6677 if (src->wscale && dst->wscale) 6678 dws = dst->wscale & PF_WSCALE_MASK; 6679 else 6680 dws = 0; 6681 6682 /* Demodulate sequence number */ 6683 seq = ntohl(th.th_seq) - src->seqdiff; 6684 if (src->seqdiff) { 6685 pf_change_a(&th.th_seq, icmpsum, 6686 htonl(seq), 0); 6687 copyback = 1; 6688 } 6689 6690 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 6691 (!SEQ_GEQ(src->seqhi, seq) || 6692 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 6693 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6694 printf("pf: BAD ICMP %d:%d ", 6695 icmptype, icmpcode); 6696 pf_print_host(pd->src, 0, pd->af); 6697 printf(" -> "); 6698 pf_print_host(pd->dst, 0, pd->af); 6699 printf(" state: "); 6700 pf_print_state(*state); 6701 printf(" seq=%u\n", seq); 6702 } 6703 REASON_SET(reason, PFRES_BADSTATE); 6704 return (PF_DROP); 6705 } else { 6706 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6707 printf("pf: OK ICMP %d:%d ", 6708 icmptype, icmpcode); 6709 pf_print_host(pd->src, 0, pd->af); 6710 printf(" -> "); 6711 pf_print_host(pd->dst, 0, pd->af); 6712 printf(" state: "); 6713 pf_print_state(*state); 6714 printf(" seq=%u\n", seq); 6715 } 6716 } 6717 6718 /* translate source/destination address, if necessary */ 6719 if ((*state)->key[PF_SK_WIRE] != 6720 (*state)->key[PF_SK_STACK]) { 6721 struct pf_state_key *nk = 6722 (*state)->key[pd->didx]; 6723 6724 if (PF_ANEQ(pd2.src, 6725 &nk->addr[pd2.sidx], pd2.af) || 6726 nk->port[pd2.sidx] != th.th_sport) 6727 pf_change_icmp(pd2.src, &th.th_sport, 6728 daddr, &nk->addr[pd2.sidx], 6729 nk->port[pd2.sidx], NULL, 6730 pd2.ip_sum, icmpsum, 6731 pd->ip_sum, 0, pd2.af); 6732 6733 if (PF_ANEQ(pd2.dst, 6734 &nk->addr[pd2.didx], pd2.af) || 6735 nk->port[pd2.didx] != th.th_dport) 6736 pf_change_icmp(pd2.dst, &th.th_dport, 6737 saddr, &nk->addr[pd2.didx], 6738 nk->port[pd2.didx], NULL, 6739 pd2.ip_sum, icmpsum, 6740 pd->ip_sum, 0, pd2.af); 6741 copyback = 1; 6742 } 6743 6744 if (copyback) { 6745 switch (pd2.af) { 6746 #ifdef INET 6747 case AF_INET: 6748 m_copyback(m, off, ICMP_MINLEN, 6749 (caddr_t )&pd->hdr.icmp); 6750 m_copyback(m, ipoff2, sizeof(h2), 6751 (caddr_t )&h2); 6752 break; 6753 #endif /* INET */ 6754 #ifdef INET6 6755 case AF_INET6: 6756 m_copyback(m, off, 6757 sizeof(struct icmp6_hdr), 6758 (caddr_t )&pd->hdr.icmp6); 6759 m_copyback(m, ipoff2, sizeof(h2_6), 6760 (caddr_t )&h2_6); 6761 break; 6762 #endif /* INET6 */ 6763 } 6764 m_copyback(m, off2, 8, (caddr_t)&th); 6765 } 6766 6767 return (PF_PASS); 6768 break; 6769 } 6770 case IPPROTO_UDP: { 6771 struct udphdr uh; 6772 6773 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 6774 NULL, reason, pd2.af)) { 6775 DPFPRINTF(PF_DEBUG_MISC, 6776 ("pf: ICMP error message too short " 6777 "(udp)\n")); 6778 return (PF_DROP); 6779 } 6780 6781 key.af = pd2.af; 6782 key.proto = IPPROTO_UDP; 6783 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6784 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6785 key.port[pd2.sidx] = uh.uh_sport; 6786 key.port[pd2.didx] = uh.uh_dport; 6787 6788 STATE_LOOKUP(kif, &key, *state, pd); 6789 6790 /* translate source/destination address, if necessary */ 6791 if ((*state)->key[PF_SK_WIRE] != 6792 (*state)->key[PF_SK_STACK]) { 6793 struct pf_state_key *nk = 6794 (*state)->key[pd->didx]; 6795 6796 if (PF_ANEQ(pd2.src, 6797 &nk->addr[pd2.sidx], pd2.af) || 6798 nk->port[pd2.sidx] != uh.uh_sport) 6799 pf_change_icmp(pd2.src, &uh.uh_sport, 6800 daddr, &nk->addr[pd2.sidx], 6801 nk->port[pd2.sidx], &uh.uh_sum, 6802 pd2.ip_sum, icmpsum, 6803 pd->ip_sum, 1, pd2.af); 6804 6805 if (PF_ANEQ(pd2.dst, 6806 &nk->addr[pd2.didx], pd2.af) || 6807 nk->port[pd2.didx] != uh.uh_dport) 6808 pf_change_icmp(pd2.dst, &uh.uh_dport, 6809 saddr, &nk->addr[pd2.didx], 6810 nk->port[pd2.didx], &uh.uh_sum, 6811 pd2.ip_sum, icmpsum, 6812 pd->ip_sum, 1, pd2.af); 6813 6814 switch (pd2.af) { 6815 #ifdef INET 6816 case AF_INET: 6817 m_copyback(m, off, ICMP_MINLEN, 6818 (caddr_t )&pd->hdr.icmp); 6819 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6820 break; 6821 #endif /* INET */ 6822 #ifdef INET6 6823 case AF_INET6: 6824 m_copyback(m, off, 6825 sizeof(struct icmp6_hdr), 6826 (caddr_t )&pd->hdr.icmp6); 6827 m_copyback(m, ipoff2, sizeof(h2_6), 6828 (caddr_t )&h2_6); 6829 break; 6830 #endif /* INET6 */ 6831 } 6832 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 6833 } 6834 return (PF_PASS); 6835 break; 6836 } 6837 #ifdef INET 6838 case IPPROTO_ICMP: { 6839 struct icmp iih; 6840 6841 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6842 NULL, reason, pd2.af)) { 6843 DPFPRINTF(PF_DEBUG_MISC, 6844 ("pf: ICMP error message too short i" 6845 "(icmp)\n")); 6846 return (PF_DROP); 6847 } 6848 6849 key.af = pd2.af; 6850 key.proto = IPPROTO_ICMP; 6851 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6852 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6853 key.port[0] = key.port[1] = iih.icmp_id; 6854 6855 STATE_LOOKUP(kif, &key, *state, pd); 6856 6857 /* translate source/destination address, if necessary */ 6858 if ((*state)->key[PF_SK_WIRE] != 6859 (*state)->key[PF_SK_STACK]) { 6860 struct pf_state_key *nk = 6861 (*state)->key[pd->didx]; 6862 6863 if (PF_ANEQ(pd2.src, 6864 &nk->addr[pd2.sidx], pd2.af) || 6865 nk->port[pd2.sidx] != iih.icmp_id) 6866 pf_change_icmp(pd2.src, &iih.icmp_id, 6867 daddr, &nk->addr[pd2.sidx], 6868 nk->port[pd2.sidx], NULL, 6869 pd2.ip_sum, icmpsum, 6870 pd->ip_sum, 0, AF_INET); 6871 6872 if (PF_ANEQ(pd2.dst, 6873 &nk->addr[pd2.didx], pd2.af) || 6874 nk->port[pd2.didx] != iih.icmp_id) 6875 pf_change_icmp(pd2.dst, &iih.icmp_id, 6876 saddr, &nk->addr[pd2.didx], 6877 nk->port[pd2.didx], NULL, 6878 pd2.ip_sum, icmpsum, 6879 pd->ip_sum, 0, AF_INET); 6880 6881 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6882 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6883 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6884 } 6885 return (PF_PASS); 6886 break; 6887 } 6888 #endif /* INET */ 6889 #ifdef INET6 6890 case IPPROTO_ICMPV6: { 6891 struct icmp6_hdr iih; 6892 6893 if (!pf_pull_hdr(m, off2, &iih, 6894 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6895 DPFPRINTF(PF_DEBUG_MISC, 6896 ("pf: ICMP error message too short " 6897 "(icmp6)\n")); 6898 return (PF_DROP); 6899 } 6900 6901 key.af = pd2.af; 6902 key.proto = IPPROTO_ICMPV6; 6903 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6904 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6905 key.port[0] = key.port[1] = iih.icmp6_id; 6906 6907 STATE_LOOKUP(kif, &key, *state, pd); 6908 6909 /* translate source/destination address, if necessary */ 6910 if ((*state)->key[PF_SK_WIRE] != 6911 (*state)->key[PF_SK_STACK]) { 6912 struct pf_state_key *nk = 6913 (*state)->key[pd->didx]; 6914 6915 if (PF_ANEQ(pd2.src, 6916 &nk->addr[pd2.sidx], pd2.af) || 6917 nk->port[pd2.sidx] != iih.icmp6_id) 6918 pf_change_icmp(pd2.src, &iih.icmp6_id, 6919 daddr, &nk->addr[pd2.sidx], 6920 nk->port[pd2.sidx], NULL, 6921 pd2.ip_sum, icmpsum, 6922 pd->ip_sum, 0, AF_INET6); 6923 6924 if (PF_ANEQ(pd2.dst, 6925 &nk->addr[pd2.didx], pd2.af) || 6926 nk->port[pd2.didx] != iih.icmp6_id) 6927 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6928 saddr, &nk->addr[pd2.didx], 6929 nk->port[pd2.didx], NULL, 6930 pd2.ip_sum, icmpsum, 6931 pd->ip_sum, 0, AF_INET6); 6932 6933 m_copyback(m, off, sizeof(struct icmp6_hdr), 6934 (caddr_t)&pd->hdr.icmp6); 6935 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6936 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6937 (caddr_t)&iih); 6938 } 6939 return (PF_PASS); 6940 break; 6941 } 6942 #endif /* INET6 */ 6943 default: { 6944 key.af = pd2.af; 6945 key.proto = pd2.proto; 6946 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6947 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6948 key.port[0] = key.port[1] = 0; 6949 6950 STATE_LOOKUP(kif, &key, *state, pd); 6951 6952 /* translate source/destination address, if necessary */ 6953 if ((*state)->key[PF_SK_WIRE] != 6954 (*state)->key[PF_SK_STACK]) { 6955 struct pf_state_key *nk = 6956 (*state)->key[pd->didx]; 6957 6958 if (PF_ANEQ(pd2.src, 6959 &nk->addr[pd2.sidx], pd2.af)) 6960 pf_change_icmp(pd2.src, NULL, daddr, 6961 &nk->addr[pd2.sidx], 0, NULL, 6962 pd2.ip_sum, icmpsum, 6963 pd->ip_sum, 0, pd2.af); 6964 6965 if (PF_ANEQ(pd2.dst, 6966 &nk->addr[pd2.didx], pd2.af)) 6967 pf_change_icmp(pd2.dst, NULL, saddr, 6968 &nk->addr[pd2.didx], 0, NULL, 6969 pd2.ip_sum, icmpsum, 6970 pd->ip_sum, 0, pd2.af); 6971 6972 switch (pd2.af) { 6973 #ifdef INET 6974 case AF_INET: 6975 m_copyback(m, off, ICMP_MINLEN, 6976 (caddr_t)&pd->hdr.icmp); 6977 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6978 break; 6979 #endif /* INET */ 6980 #ifdef INET6 6981 case AF_INET6: 6982 m_copyback(m, off, 6983 sizeof(struct icmp6_hdr), 6984 (caddr_t )&pd->hdr.icmp6); 6985 m_copyback(m, ipoff2, sizeof(h2_6), 6986 (caddr_t )&h2_6); 6987 break; 6988 #endif /* INET6 */ 6989 } 6990 } 6991 return (PF_PASS); 6992 break; 6993 } 6994 } 6995 } 6996 } 6997 6998 static int 6999 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7000 struct mbuf *m, struct pf_pdesc *pd) 7001 { 7002 struct pf_state_peer *src, *dst; 7003 struct pf_state_key_cmp key; 7004 uint8_t psrc, pdst; 7005 7006 bzero(&key, sizeof(key)); 7007 key.af = pd->af; 7008 key.proto = pd->proto; 7009 if (pd->dir == PF_IN) { 7010 PF_ACPY(&key.addr[0], pd->src, key.af); 7011 PF_ACPY(&key.addr[1], pd->dst, key.af); 7012 key.port[0] = key.port[1] = 0; 7013 } else { 7014 PF_ACPY(&key.addr[1], pd->src, key.af); 7015 PF_ACPY(&key.addr[0], pd->dst, key.af); 7016 key.port[1] = key.port[0] = 0; 7017 } 7018 7019 STATE_LOOKUP(kif, &key, *state, pd); 7020 7021 if (pd->dir == (*state)->direction) { 7022 src = &(*state)->src; 7023 dst = &(*state)->dst; 7024 psrc = PF_PEER_SRC; 7025 pdst = PF_PEER_DST; 7026 } else { 7027 src = &(*state)->dst; 7028 dst = &(*state)->src; 7029 psrc = PF_PEER_DST; 7030 pdst = PF_PEER_SRC; 7031 } 7032 7033 /* update states */ 7034 if (src->state < PFOTHERS_SINGLE) 7035 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7036 if (dst->state == PFOTHERS_SINGLE) 7037 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7038 7039 /* update expire time */ 7040 (*state)->expire = time_uptime; 7041 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7042 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7043 else 7044 (*state)->timeout = PFTM_OTHER_SINGLE; 7045 7046 /* translate source/destination address, if necessary */ 7047 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7048 struct pf_state_key *nk = (*state)->key[pd->didx]; 7049 7050 KASSERT(nk, ("%s: nk is null", __func__)); 7051 KASSERT(pd, ("%s: pd is null", __func__)); 7052 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7053 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7054 switch (pd->af) { 7055 #ifdef INET 7056 case AF_INET: 7057 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7058 pf_change_a(&pd->src->v4.s_addr, 7059 pd->ip_sum, 7060 nk->addr[pd->sidx].v4.s_addr, 7061 0); 7062 7063 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7064 pf_change_a(&pd->dst->v4.s_addr, 7065 pd->ip_sum, 7066 nk->addr[pd->didx].v4.s_addr, 7067 0); 7068 7069 break; 7070 #endif /* INET */ 7071 #ifdef INET6 7072 case AF_INET6: 7073 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7074 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7075 7076 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7077 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7078 #endif /* INET6 */ 7079 } 7080 } 7081 return (PF_PASS); 7082 } 7083 7084 /* 7085 * ipoff and off are measured from the start of the mbuf chain. 7086 * h must be at "ipoff" on the mbuf chain. 7087 */ 7088 void * 7089 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 7090 u_short *actionp, u_short *reasonp, sa_family_t af) 7091 { 7092 switch (af) { 7093 #ifdef INET 7094 case AF_INET: { 7095 struct ip *h = mtod(m, struct ip *); 7096 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7097 7098 if (fragoff) { 7099 if (fragoff >= len) 7100 ACTION_SET(actionp, PF_PASS); 7101 else { 7102 ACTION_SET(actionp, PF_DROP); 7103 REASON_SET(reasonp, PFRES_FRAG); 7104 } 7105 return (NULL); 7106 } 7107 if (m->m_pkthdr.len < off + len || 7108 ntohs(h->ip_len) < off + len) { 7109 ACTION_SET(actionp, PF_DROP); 7110 REASON_SET(reasonp, PFRES_SHORT); 7111 return (NULL); 7112 } 7113 break; 7114 } 7115 #endif /* INET */ 7116 #ifdef INET6 7117 case AF_INET6: { 7118 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7119 7120 if (m->m_pkthdr.len < off + len || 7121 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7122 (unsigned)(off + len)) { 7123 ACTION_SET(actionp, PF_DROP); 7124 REASON_SET(reasonp, PFRES_SHORT); 7125 return (NULL); 7126 } 7127 break; 7128 } 7129 #endif /* INET6 */ 7130 } 7131 m_copydata(m, off, len, p); 7132 return (p); 7133 } 7134 7135 int 7136 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7137 int rtableid) 7138 { 7139 struct ifnet *ifp; 7140 7141 /* 7142 * Skip check for addresses with embedded interface scope, 7143 * as they would always match anyway. 7144 */ 7145 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7146 return (1); 7147 7148 if (af != AF_INET && af != AF_INET6) 7149 return (0); 7150 7151 /* Skip checks for ipsec interfaces */ 7152 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7153 return (1); 7154 7155 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7156 7157 switch (af) { 7158 #ifdef INET6 7159 case AF_INET6: 7160 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7161 ifp)); 7162 #endif 7163 #ifdef INET 7164 case AF_INET: 7165 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7166 ifp)); 7167 #endif 7168 } 7169 7170 return (0); 7171 } 7172 7173 #ifdef INET 7174 static void 7175 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7176 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7177 { 7178 struct mbuf *m0, *m1, *md; 7179 struct sockaddr_in dst; 7180 struct ip *ip; 7181 struct pfi_kkif *nkif = NULL; 7182 struct ifnet *ifp = NULL; 7183 struct pf_addr naddr; 7184 struct pf_ksrc_node *sn = NULL; 7185 int error = 0; 7186 uint16_t ip_len, ip_off; 7187 int r_rt, r_dir; 7188 7189 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7190 7191 if (s) { 7192 r_rt = s->rt; 7193 r_dir = s->direction; 7194 } else { 7195 r_rt = r->rt; 7196 r_dir = r->direction; 7197 } 7198 7199 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7200 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7201 __func__)); 7202 7203 if ((pd->pf_mtag == NULL && 7204 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7205 pd->pf_mtag->routed++ > 3) { 7206 m0 = *m; 7207 *m = NULL; 7208 goto bad_locked; 7209 } 7210 7211 if (r_rt == PF_DUPTO) { 7212 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7213 if (s == NULL) { 7214 ifp = r->rpool.cur->kif ? 7215 r->rpool.cur->kif->pfik_ifp : NULL; 7216 } else { 7217 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7218 /* If pfsync'd */ 7219 if (ifp == NULL && r->rpool.cur != NULL) 7220 ifp = r->rpool.cur->kif ? 7221 r->rpool.cur->kif->pfik_ifp : NULL; 7222 PF_STATE_UNLOCK(s); 7223 } 7224 if (ifp == oifp) { 7225 /* When the 2nd interface is not skipped */ 7226 return; 7227 } else { 7228 m0 = *m; 7229 *m = NULL; 7230 goto bad; 7231 } 7232 } else { 7233 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7234 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7235 if (s) 7236 PF_STATE_UNLOCK(s); 7237 return; 7238 } 7239 } 7240 } else { 7241 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7242 pf_dummynet(pd, s, r, m); 7243 if (s) 7244 PF_STATE_UNLOCK(s); 7245 return; 7246 } 7247 m0 = *m; 7248 } 7249 7250 ip = mtod(m0, struct ip *); 7251 7252 bzero(&dst, sizeof(dst)); 7253 dst.sin_family = AF_INET; 7254 dst.sin_len = sizeof(dst); 7255 dst.sin_addr = ip->ip_dst; 7256 7257 bzero(&naddr, sizeof(naddr)); 7258 7259 if (s == NULL) { 7260 if (TAILQ_EMPTY(&r->rpool.list)) { 7261 DPFPRINTF(PF_DEBUG_URGENT, 7262 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7263 goto bad_locked; 7264 } 7265 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7266 &naddr, &nkif, NULL, &sn); 7267 if (!PF_AZERO(&naddr, AF_INET)) 7268 dst.sin_addr.s_addr = naddr.v4.s_addr; 7269 ifp = nkif ? nkif->pfik_ifp : NULL; 7270 } else { 7271 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7272 dst.sin_addr.s_addr = 7273 s->rt_addr.v4.s_addr; 7274 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7275 /* If pfsync'd */ 7276 if (ifp == NULL && r->rpool.cur != NULL) { 7277 ifp = r->rpool.cur->kif ? 7278 r->rpool.cur->kif->pfik_ifp : NULL; 7279 } 7280 PF_STATE_UNLOCK(s); 7281 } 7282 7283 if (ifp == NULL) 7284 goto bad; 7285 7286 if (pd->dir == PF_IN) { 7287 if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7288 goto bad; 7289 else if (m0 == NULL) 7290 goto done; 7291 if (m0->m_len < sizeof(struct ip)) { 7292 DPFPRINTF(PF_DEBUG_URGENT, 7293 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7294 goto bad; 7295 } 7296 ip = mtod(m0, struct ip *); 7297 } 7298 7299 if (ifp->if_flags & IFF_LOOPBACK) 7300 m0->m_flags |= M_SKIP_FIREWALL; 7301 7302 ip_len = ntohs(ip->ip_len); 7303 ip_off = ntohs(ip->ip_off); 7304 7305 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7306 m0->m_pkthdr.csum_flags |= CSUM_IP; 7307 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7308 in_delayed_cksum(m0); 7309 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7310 } 7311 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7312 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7313 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7314 } 7315 7316 /* 7317 * If small enough for interface, or the interface will take 7318 * care of the fragmentation for us, we can just send directly. 7319 */ 7320 if (ip_len <= ifp->if_mtu || 7321 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7322 ip->ip_sum = 0; 7323 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7324 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7325 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7326 } 7327 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7328 7329 md = m0; 7330 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7331 if (md != NULL) 7332 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7333 goto done; 7334 } 7335 7336 /* Balk when DF bit is set or the interface didn't support TSO. */ 7337 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7338 error = EMSGSIZE; 7339 KMOD_IPSTAT_INC(ips_cantfrag); 7340 if (r_rt != PF_DUPTO) { 7341 if (s && pd->nat_rule != NULL) 7342 PACKET_UNDO_NAT(m0, pd, 7343 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7344 s); 7345 7346 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7347 ifp->if_mtu); 7348 goto done; 7349 } else 7350 goto bad; 7351 } 7352 7353 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7354 if (error) 7355 goto bad; 7356 7357 for (; m0; m0 = m1) { 7358 m1 = m0->m_nextpkt; 7359 m0->m_nextpkt = NULL; 7360 if (error == 0) { 7361 m_clrprotoflags(m0); 7362 md = m0; 7363 error = pf_dummynet_route(pd, s, r, ifp, 7364 sintosa(&dst), &md); 7365 if (md != NULL) 7366 error = (*ifp->if_output)(ifp, md, 7367 sintosa(&dst), NULL); 7368 } else 7369 m_freem(m0); 7370 } 7371 7372 if (error == 0) 7373 KMOD_IPSTAT_INC(ips_fragmented); 7374 7375 done: 7376 if (r_rt != PF_DUPTO) 7377 *m = NULL; 7378 return; 7379 7380 bad_locked: 7381 if (s) 7382 PF_STATE_UNLOCK(s); 7383 bad: 7384 m_freem(m0); 7385 goto done; 7386 } 7387 #endif /* INET */ 7388 7389 #ifdef INET6 7390 static void 7391 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7392 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7393 { 7394 struct mbuf *m0, *md; 7395 struct sockaddr_in6 dst; 7396 struct ip6_hdr *ip6; 7397 struct pfi_kkif *nkif = NULL; 7398 struct ifnet *ifp = NULL; 7399 struct pf_addr naddr; 7400 struct pf_ksrc_node *sn = NULL; 7401 int r_rt, r_dir; 7402 7403 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7404 7405 if (s) { 7406 r_rt = s->rt; 7407 r_dir = s->direction; 7408 } else { 7409 r_rt = r->rt; 7410 r_dir = r->direction; 7411 } 7412 7413 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7414 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7415 __func__)); 7416 7417 if ((pd->pf_mtag == NULL && 7418 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7419 pd->pf_mtag->routed++ > 3) { 7420 m0 = *m; 7421 *m = NULL; 7422 goto bad_locked; 7423 } 7424 7425 if (r_rt == PF_DUPTO) { 7426 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7427 if (s == NULL) { 7428 ifp = r->rpool.cur->kif ? 7429 r->rpool.cur->kif->pfik_ifp : NULL; 7430 } else { 7431 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7432 /* If pfsync'd */ 7433 if (ifp == NULL && r->rpool.cur != NULL) 7434 ifp = r->rpool.cur->kif ? 7435 r->rpool.cur->kif->pfik_ifp : NULL; 7436 PF_STATE_UNLOCK(s); 7437 } 7438 if (ifp == oifp) { 7439 /* When the 2nd interface is not skipped */ 7440 return; 7441 } else { 7442 m0 = *m; 7443 *m = NULL; 7444 goto bad; 7445 } 7446 } else { 7447 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7448 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7449 if (s) 7450 PF_STATE_UNLOCK(s); 7451 return; 7452 } 7453 } 7454 } else { 7455 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7456 pf_dummynet(pd, s, r, m); 7457 if (s) 7458 PF_STATE_UNLOCK(s); 7459 return; 7460 } 7461 m0 = *m; 7462 } 7463 7464 ip6 = mtod(m0, struct ip6_hdr *); 7465 7466 bzero(&dst, sizeof(dst)); 7467 dst.sin6_family = AF_INET6; 7468 dst.sin6_len = sizeof(dst); 7469 dst.sin6_addr = ip6->ip6_dst; 7470 7471 bzero(&naddr, sizeof(naddr)); 7472 7473 if (s == NULL) { 7474 if (TAILQ_EMPTY(&r->rpool.list)) { 7475 DPFPRINTF(PF_DEBUG_URGENT, 7476 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7477 goto bad_locked; 7478 } 7479 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7480 &naddr, &nkif, NULL, &sn); 7481 if (!PF_AZERO(&naddr, AF_INET6)) 7482 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7483 &naddr, AF_INET6); 7484 ifp = nkif ? nkif->pfik_ifp : NULL; 7485 } else { 7486 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7487 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7488 &s->rt_addr, AF_INET6); 7489 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7490 /* If pfsync'd */ 7491 if (ifp == NULL && r->rpool.cur != NULL) 7492 ifp = r->rpool.cur->kif ? 7493 r->rpool.cur->kif->pfik_ifp : NULL; 7494 } 7495 7496 if (s) 7497 PF_STATE_UNLOCK(s); 7498 7499 if (ifp == NULL) 7500 goto bad; 7501 7502 if (pd->dir == PF_IN) { 7503 if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS) 7504 goto bad; 7505 else if (m0 == NULL) 7506 goto done; 7507 if (m0->m_len < sizeof(struct ip6_hdr)) { 7508 DPFPRINTF(PF_DEBUG_URGENT, 7509 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 7510 __func__)); 7511 goto bad; 7512 } 7513 ip6 = mtod(m0, struct ip6_hdr *); 7514 } 7515 7516 if (ifp->if_flags & IFF_LOOPBACK) 7517 m0->m_flags |= M_SKIP_FIREWALL; 7518 7519 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 7520 ~ifp->if_hwassist) { 7521 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 7522 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 7523 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 7524 } 7525 7526 /* 7527 * If the packet is too large for the outgoing interface, 7528 * send back an icmp6 error. 7529 */ 7530 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 7531 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 7532 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 7533 md = m0; 7534 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7535 if (md != NULL) 7536 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 7537 } 7538 else { 7539 in6_ifstat_inc(ifp, ifs6_in_toobig); 7540 if (r_rt != PF_DUPTO) { 7541 if (s && pd->nat_rule != NULL) 7542 PACKET_UNDO_NAT(m0, pd, 7543 ((caddr_t)ip6 - m0->m_data) + 7544 sizeof(struct ip6_hdr), s); 7545 7546 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 7547 } else 7548 goto bad; 7549 } 7550 7551 done: 7552 if (r_rt != PF_DUPTO) 7553 *m = NULL; 7554 return; 7555 7556 bad_locked: 7557 if (s) 7558 PF_STATE_UNLOCK(s); 7559 bad: 7560 m_freem(m0); 7561 goto done; 7562 } 7563 #endif /* INET6 */ 7564 7565 /* 7566 * FreeBSD supports cksum offloads for the following drivers. 7567 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 7568 * 7569 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 7570 * network driver performed cksum including pseudo header, need to verify 7571 * csum_data 7572 * CSUM_DATA_VALID : 7573 * network driver performed cksum, needs to additional pseudo header 7574 * cksum computation with partial csum_data(i.e. lack of H/W support for 7575 * pseudo header, for instance sk(4) and possibly gem(4)) 7576 * 7577 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 7578 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 7579 * TCP/UDP layer. 7580 * Also, set csum_data to 0xffff to force cksum validation. 7581 */ 7582 static int 7583 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 7584 { 7585 u_int16_t sum = 0; 7586 int hw_assist = 0; 7587 struct ip *ip; 7588 7589 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 7590 return (1); 7591 if (m->m_pkthdr.len < off + len) 7592 return (1); 7593 7594 switch (p) { 7595 case IPPROTO_TCP: 7596 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7597 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7598 sum = m->m_pkthdr.csum_data; 7599 } else { 7600 ip = mtod(m, struct ip *); 7601 sum = in_pseudo(ip->ip_src.s_addr, 7602 ip->ip_dst.s_addr, htonl((u_short)len + 7603 m->m_pkthdr.csum_data + IPPROTO_TCP)); 7604 } 7605 sum ^= 0xffff; 7606 ++hw_assist; 7607 } 7608 break; 7609 case IPPROTO_UDP: 7610 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7611 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7612 sum = m->m_pkthdr.csum_data; 7613 } else { 7614 ip = mtod(m, struct ip *); 7615 sum = in_pseudo(ip->ip_src.s_addr, 7616 ip->ip_dst.s_addr, htonl((u_short)len + 7617 m->m_pkthdr.csum_data + IPPROTO_UDP)); 7618 } 7619 sum ^= 0xffff; 7620 ++hw_assist; 7621 } 7622 break; 7623 case IPPROTO_ICMP: 7624 #ifdef INET6 7625 case IPPROTO_ICMPV6: 7626 #endif /* INET6 */ 7627 break; 7628 default: 7629 return (1); 7630 } 7631 7632 if (!hw_assist) { 7633 switch (af) { 7634 case AF_INET: 7635 if (p == IPPROTO_ICMP) { 7636 if (m->m_len < off) 7637 return (1); 7638 m->m_data += off; 7639 m->m_len -= off; 7640 sum = in_cksum(m, len); 7641 m->m_data -= off; 7642 m->m_len += off; 7643 } else { 7644 if (m->m_len < sizeof(struct ip)) 7645 return (1); 7646 sum = in4_cksum(m, p, off, len); 7647 } 7648 break; 7649 #ifdef INET6 7650 case AF_INET6: 7651 if (m->m_len < sizeof(struct ip6_hdr)) 7652 return (1); 7653 sum = in6_cksum(m, p, off, len); 7654 break; 7655 #endif /* INET6 */ 7656 default: 7657 return (1); 7658 } 7659 } 7660 if (sum) { 7661 switch (p) { 7662 case IPPROTO_TCP: 7663 { 7664 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 7665 break; 7666 } 7667 case IPPROTO_UDP: 7668 { 7669 KMOD_UDPSTAT_INC(udps_badsum); 7670 break; 7671 } 7672 #ifdef INET 7673 case IPPROTO_ICMP: 7674 { 7675 KMOD_ICMPSTAT_INC(icps_checksum); 7676 break; 7677 } 7678 #endif 7679 #ifdef INET6 7680 case IPPROTO_ICMPV6: 7681 { 7682 KMOD_ICMP6STAT_INC(icp6s_checksum); 7683 break; 7684 } 7685 #endif /* INET6 */ 7686 } 7687 return (1); 7688 } else { 7689 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 7690 m->m_pkthdr.csum_flags |= 7691 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 7692 m->m_pkthdr.csum_data = 0xffff; 7693 } 7694 } 7695 return (0); 7696 } 7697 7698 static bool 7699 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 7700 const struct pf_kstate *s, struct ip_fw_args *dnflow) 7701 { 7702 int dndir = r->direction; 7703 7704 if (s && dndir == PF_INOUT) { 7705 dndir = s->direction; 7706 } else if (dndir == PF_INOUT) { 7707 /* Assume primary direction. Happens when we've set dnpipe in 7708 * the ethernet level code. */ 7709 dndir = pd->dir; 7710 } 7711 7712 memset(dnflow, 0, sizeof(*dnflow)); 7713 7714 if (pd->dport != NULL) 7715 dnflow->f_id.dst_port = ntohs(*pd->dport); 7716 if (pd->sport != NULL) 7717 dnflow->f_id.src_port = ntohs(*pd->sport); 7718 7719 if (pd->dir == PF_IN) 7720 dnflow->flags |= IPFW_ARGS_IN; 7721 else 7722 dnflow->flags |= IPFW_ARGS_OUT; 7723 7724 if (pd->dir != dndir && pd->act.dnrpipe) { 7725 dnflow->rule.info = pd->act.dnrpipe; 7726 } 7727 else if (pd->dir == dndir && pd->act.dnpipe) { 7728 dnflow->rule.info = pd->act.dnpipe; 7729 } 7730 else { 7731 return (false); 7732 } 7733 7734 dnflow->rule.info |= IPFW_IS_DUMMYNET; 7735 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 7736 dnflow->rule.info |= IPFW_IS_PIPE; 7737 7738 dnflow->f_id.proto = pd->proto; 7739 dnflow->f_id.extra = dnflow->rule.info; 7740 switch (pd->af) { 7741 case AF_INET: 7742 dnflow->f_id.addr_type = 4; 7743 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 7744 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 7745 break; 7746 case AF_INET6: 7747 dnflow->flags |= IPFW_ARGS_IP6; 7748 dnflow->f_id.addr_type = 6; 7749 dnflow->f_id.src_ip6 = pd->src->v6; 7750 dnflow->f_id.dst_ip6 = pd->dst->v6; 7751 break; 7752 default: 7753 panic("Invalid AF"); 7754 break; 7755 } 7756 7757 return (true); 7758 } 7759 7760 int 7761 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7762 struct inpcb *inp) 7763 { 7764 struct pfi_kkif *kif; 7765 struct mbuf *m = *m0; 7766 7767 M_ASSERTPKTHDR(m); 7768 MPASS(ifp->if_vnet == curvnet); 7769 NET_EPOCH_ASSERT(); 7770 7771 if (!V_pf_status.running) 7772 return (PF_PASS); 7773 7774 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7775 7776 if (kif == NULL) { 7777 DPFPRINTF(PF_DEBUG_URGENT, 7778 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 7779 return (PF_DROP); 7780 } 7781 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7782 return (PF_PASS); 7783 7784 if (m->m_flags & M_SKIP_FIREWALL) 7785 return (PF_PASS); 7786 7787 /* Stateless! */ 7788 return (pf_test_eth_rule(dir, kif, m0)); 7789 } 7790 7791 static int 7792 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 7793 struct pf_krule *r, struct mbuf **m0) 7794 { 7795 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 7796 } 7797 7798 static int 7799 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 7800 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 7801 struct mbuf **m0) 7802 { 7803 NET_EPOCH_ASSERT(); 7804 7805 if (pd->act.dnpipe || pd->act.dnrpipe) { 7806 struct ip_fw_args dnflow; 7807 if (ip_dn_io_ptr == NULL) { 7808 m_freem(*m0); 7809 *m0 = NULL; 7810 return (ENOMEM); 7811 } 7812 7813 if (pd->pf_mtag == NULL && 7814 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 7815 m_freem(*m0); 7816 *m0 = NULL; 7817 return (ENOMEM); 7818 } 7819 7820 if (ifp != NULL) { 7821 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 7822 7823 pd->pf_mtag->if_index = ifp->if_index; 7824 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 7825 7826 MPASS(sa != NULL); 7827 7828 if (pd->af == AF_INET) 7829 memcpy(&pd->pf_mtag->dst, sa, 7830 sizeof(struct sockaddr_in)); 7831 else 7832 memcpy(&pd->pf_mtag->dst, sa, 7833 sizeof(struct sockaddr_in6)); 7834 } 7835 7836 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 7837 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 7838 ip_dn_io_ptr(m0, &dnflow); 7839 if (*m0 != NULL) { 7840 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7841 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7842 } 7843 } 7844 } 7845 7846 return (0); 7847 } 7848 7849 #ifdef INET 7850 int 7851 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7852 struct inpcb *inp, struct pf_rule_actions *default_actions) 7853 { 7854 struct pfi_kkif *kif; 7855 u_short action, reason = 0; 7856 struct mbuf *m = *m0; 7857 struct ip *h = NULL; 7858 struct m_tag *mtag; 7859 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7860 struct pf_kstate *s = NULL; 7861 struct pf_kruleset *ruleset = NULL; 7862 struct pf_pdesc pd; 7863 int off, dirndx, use_2nd_queue = 0; 7864 uint16_t tag; 7865 uint8_t rt; 7866 7867 PF_RULES_RLOCK_TRACKER; 7868 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 7869 M_ASSERTPKTHDR(m); 7870 7871 if (!V_pf_status.running) 7872 return (PF_PASS); 7873 7874 PF_RULES_RLOCK(); 7875 7876 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7877 7878 if (__predict_false(kif == NULL)) { 7879 DPFPRINTF(PF_DEBUG_URGENT, 7880 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 7881 PF_RULES_RUNLOCK(); 7882 return (PF_DROP); 7883 } 7884 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 7885 PF_RULES_RUNLOCK(); 7886 return (PF_PASS); 7887 } 7888 7889 if (m->m_flags & M_SKIP_FIREWALL) { 7890 PF_RULES_RUNLOCK(); 7891 return (PF_PASS); 7892 } 7893 7894 memset(&pd, 0, sizeof(pd)); 7895 TAILQ_INIT(&pd.sctp_multihome_jobs); 7896 if (default_actions != NULL) 7897 memcpy(&pd.act, default_actions, sizeof(pd.act)); 7898 pd.pf_mtag = pf_find_mtag(m); 7899 7900 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 7901 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7902 7903 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 7904 pd.pf_mtag->if_idxgen); 7905 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 7906 PF_RULES_RUNLOCK(); 7907 m_freem(*m0); 7908 *m0 = NULL; 7909 return (PF_PASS); 7910 } 7911 PF_RULES_RUNLOCK(); 7912 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 7913 *m0 = NULL; 7914 return (PF_PASS); 7915 } 7916 7917 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 7918 pd.act.dnpipe = pd.pf_mtag->dnpipe; 7919 pd.act.flags = pd.pf_mtag->dnflags; 7920 } 7921 7922 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 7923 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 7924 /* Dummynet re-injects packets after they've 7925 * completed their delay. We've already 7926 * processed them, so pass unconditionally. */ 7927 7928 /* But only once. We may see the packet multiple times (e.g. 7929 * PFIL_IN/PFIL_OUT). */ 7930 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7931 PF_RULES_RUNLOCK(); 7932 7933 return (PF_PASS); 7934 } 7935 7936 pd.sport = pd.dport = NULL; 7937 pd.proto_sum = NULL; 7938 pd.dir = dir; 7939 pd.sidx = (dir == PF_IN) ? 0 : 1; 7940 pd.didx = (dir == PF_IN) ? 1 : 0; 7941 pd.af = AF_INET; 7942 pd.act.rtableid = -1; 7943 7944 h = mtod(m, struct ip *); 7945 off = h->ip_hl << 2; 7946 7947 if (__predict_false(ip_divert_ptr != NULL) && 7948 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 7949 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 7950 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 7951 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 7952 if (pd.pf_mtag == NULL && 7953 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7954 action = PF_DROP; 7955 goto done; 7956 } 7957 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 7958 } 7959 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 7960 m->m_flags |= M_FASTFWD_OURS; 7961 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 7962 } 7963 m_tag_delete(m, mtag); 7964 7965 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 7966 if (mtag != NULL) 7967 m_tag_delete(m, mtag); 7968 } else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 7969 /* We do IP header normalization and packet reassembly here */ 7970 action = PF_DROP; 7971 goto done; 7972 } 7973 m = *m0; /* pf_normalize messes with m0 */ 7974 h = mtod(m, struct ip *); 7975 7976 off = h->ip_hl << 2; 7977 if (off < (int)sizeof(struct ip)) { 7978 action = PF_DROP; 7979 REASON_SET(&reason, PFRES_SHORT); 7980 pd.act.log = PF_LOG_FORCE; 7981 goto done; 7982 } 7983 7984 pd.src = (struct pf_addr *)&h->ip_src; 7985 pd.dst = (struct pf_addr *)&h->ip_dst; 7986 pd.ip_sum = &h->ip_sum; 7987 pd.proto = h->ip_p; 7988 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 7989 pd.tot_len = ntohs(h->ip_len); 7990 7991 /* handle fragments that didn't get reassembled by normalization */ 7992 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 7993 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 7994 goto done; 7995 } 7996 7997 switch (h->ip_p) { 7998 case IPPROTO_TCP: { 7999 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8000 &action, &reason, AF_INET)) { 8001 if (action != PF_PASS) 8002 pd.act.log = PF_LOG_FORCE; 8003 goto done; 8004 } 8005 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8006 8007 pd.sport = &pd.hdr.tcp.th_sport; 8008 pd.dport = &pd.hdr.tcp.th_dport; 8009 8010 /* Respond to SYN with a syncookie. */ 8011 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8012 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8013 pf_syncookie_send(m, off, &pd); 8014 action = PF_DROP; 8015 break; 8016 } 8017 8018 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 8019 use_2nd_queue = 1; 8020 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8021 if (action == PF_DROP) 8022 goto done; 8023 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8024 if (action == PF_PASS) { 8025 if (V_pfsync_update_state_ptr != NULL) 8026 V_pfsync_update_state_ptr(s); 8027 r = s->rule.ptr; 8028 a = s->anchor.ptr; 8029 } else if (s == NULL) { 8030 /* Validate remote SYN|ACK, re-create original SYN if 8031 * valid. */ 8032 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8033 TH_ACK && pf_syncookie_validate(&pd) && 8034 pd.dir == PF_IN) { 8035 struct mbuf *msyn; 8036 8037 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 8038 &pd); 8039 if (msyn == NULL) { 8040 action = PF_DROP; 8041 break; 8042 } 8043 8044 action = pf_test(dir, pflags, ifp, &msyn, inp, 8045 &pd.act); 8046 m_freem(msyn); 8047 if (action != PF_PASS) 8048 break; 8049 8050 action = pf_test_state_tcp(&s, kif, m, off, h, 8051 &pd, &reason); 8052 if (action != PF_PASS || s == NULL) { 8053 action = PF_DROP; 8054 break; 8055 } 8056 8057 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8058 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8059 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8060 action = pf_synproxy(&pd, &s, &reason); 8061 break; 8062 } else { 8063 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8064 &a, &ruleset, inp); 8065 } 8066 } 8067 break; 8068 } 8069 8070 case IPPROTO_UDP: { 8071 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8072 &action, &reason, AF_INET)) { 8073 if (action != PF_PASS) 8074 pd.act.log = PF_LOG_FORCE; 8075 goto done; 8076 } 8077 pd.sport = &pd.hdr.udp.uh_sport; 8078 pd.dport = &pd.hdr.udp.uh_dport; 8079 if (pd.hdr.udp.uh_dport == 0 || 8080 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8081 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8082 action = PF_DROP; 8083 REASON_SET(&reason, PFRES_SHORT); 8084 goto done; 8085 } 8086 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8087 if (action == PF_PASS) { 8088 if (V_pfsync_update_state_ptr != NULL) 8089 V_pfsync_update_state_ptr(s); 8090 r = s->rule.ptr; 8091 a = s->anchor.ptr; 8092 } else if (s == NULL) 8093 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8094 &a, &ruleset, inp); 8095 break; 8096 } 8097 8098 case IPPROTO_SCTP: { 8099 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8100 &action, &reason, AF_INET)) { 8101 if (action != PF_PASS) 8102 pd.act.log |= PF_LOG_FORCE; 8103 goto done; 8104 } 8105 pd.p_len = pd.tot_len - off; 8106 8107 pd.sport = &pd.hdr.sctp.src_port; 8108 pd.dport = &pd.hdr.sctp.dest_port; 8109 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8110 action = PF_DROP; 8111 REASON_SET(&reason, PFRES_SHORT); 8112 goto done; 8113 } 8114 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8115 if (action == PF_DROP) 8116 goto done; 8117 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8118 &reason); 8119 if (action == PF_PASS) { 8120 if (V_pfsync_update_state_ptr != NULL) 8121 V_pfsync_update_state_ptr(s); 8122 r = s->rule.ptr; 8123 a = s->anchor.ptr; 8124 } else { 8125 action = pf_test_rule(&r, &s, kif, m, off, 8126 &pd, &a, &ruleset, inp); 8127 } 8128 break; 8129 } 8130 8131 case IPPROTO_ICMP: { 8132 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 8133 &action, &reason, AF_INET)) { 8134 if (action != PF_PASS) 8135 pd.act.log = PF_LOG_FORCE; 8136 goto done; 8137 } 8138 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8139 if (action == PF_PASS) { 8140 if (V_pfsync_update_state_ptr != NULL) 8141 V_pfsync_update_state_ptr(s); 8142 r = s->rule.ptr; 8143 a = s->anchor.ptr; 8144 } else if (s == NULL) 8145 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8146 &a, &ruleset, inp); 8147 break; 8148 } 8149 8150 #ifdef INET6 8151 case IPPROTO_ICMPV6: { 8152 action = PF_DROP; 8153 DPFPRINTF(PF_DEBUG_MISC, 8154 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 8155 goto done; 8156 } 8157 #endif 8158 8159 default: 8160 action = pf_test_state_other(&s, kif, m, &pd); 8161 if (action == PF_PASS) { 8162 if (V_pfsync_update_state_ptr != NULL) 8163 V_pfsync_update_state_ptr(s); 8164 r = s->rule.ptr; 8165 a = s->anchor.ptr; 8166 } else if (s == NULL) 8167 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8168 &a, &ruleset, inp); 8169 break; 8170 } 8171 8172 done: 8173 PF_RULES_RUNLOCK(); 8174 if (action == PF_PASS && h->ip_hl > 5 && 8175 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8176 action = PF_DROP; 8177 REASON_SET(&reason, PFRES_IPOPTIONS); 8178 pd.act.log = PF_LOG_FORCE; 8179 DPFPRINTF(PF_DEBUG_MISC, 8180 ("pf: dropping packet with ip options\n")); 8181 } 8182 8183 if (s) { 8184 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8185 tag = s->tag; 8186 rt = s->rt; 8187 } else { 8188 tag = r->tag; 8189 rt = r->rt; 8190 } 8191 8192 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8193 action = PF_DROP; 8194 REASON_SET(&reason, PFRES_MEMORY); 8195 } 8196 8197 pf_scrub_ip(&m, &pd); 8198 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8199 pf_normalize_mss(m, off, &pd); 8200 8201 if (pd.act.rtableid >= 0) 8202 M_SETFIB(m, pd.act.rtableid); 8203 8204 if (pd.act.flags & PFSTATE_SETPRIO) { 8205 if (pd.tos & IPTOS_LOWDELAY) 8206 use_2nd_queue = 1; 8207 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8208 action = PF_DROP; 8209 REASON_SET(&reason, PFRES_MEMORY); 8210 pd.act.log = PF_LOG_FORCE; 8211 DPFPRINTF(PF_DEBUG_MISC, 8212 ("pf: failed to allocate 802.1q mtag\n")); 8213 } 8214 } 8215 8216 #ifdef ALTQ 8217 if (action == PF_PASS && pd.act.qid) { 8218 if (pd.pf_mtag == NULL && 8219 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8220 action = PF_DROP; 8221 REASON_SET(&reason, PFRES_MEMORY); 8222 } else { 8223 if (s != NULL) 8224 pd.pf_mtag->qid_hash = pf_state_hash(s); 8225 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 8226 pd.pf_mtag->qid = pd.act.pqid; 8227 else 8228 pd.pf_mtag->qid = pd.act.qid; 8229 /* Add hints for ecn. */ 8230 pd.pf_mtag->hdr = h; 8231 } 8232 } 8233 #endif /* ALTQ */ 8234 8235 /* 8236 * connections redirected to loopback should not match sockets 8237 * bound specifically to loopback due to security implications, 8238 * see tcp_input() and in_pcblookup_listen(). 8239 */ 8240 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8241 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8242 (s->nat_rule.ptr->action == PF_RDR || 8243 s->nat_rule.ptr->action == PF_BINAT) && 8244 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 8245 m->m_flags |= M_SKIP_FIREWALL; 8246 8247 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 8248 r->divert.port && !PACKET_LOOPED(&pd)) { 8249 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 8250 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 8251 if (mtag != NULL) { 8252 ((struct pf_divert_mtag *)(mtag+1))->port = 8253 ntohs(r->divert.port); 8254 ((struct pf_divert_mtag *)(mtag+1))->idir = 8255 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 8256 PF_DIVERT_MTAG_DIR_OUT; 8257 8258 if (s) 8259 PF_STATE_UNLOCK(s); 8260 8261 m_tag_prepend(m, mtag); 8262 if (m->m_flags & M_FASTFWD_OURS) { 8263 if (pd.pf_mtag == NULL && 8264 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8265 action = PF_DROP; 8266 REASON_SET(&reason, PFRES_MEMORY); 8267 pd.act.log = PF_LOG_FORCE; 8268 DPFPRINTF(PF_DEBUG_MISC, 8269 ("pf: failed to allocate tag\n")); 8270 } else { 8271 pd.pf_mtag->flags |= 8272 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8273 m->m_flags &= ~M_FASTFWD_OURS; 8274 } 8275 } 8276 ip_divert_ptr(*m0, dir == PF_IN); 8277 *m0 = NULL; 8278 8279 return (action); 8280 } else { 8281 /* XXX: ipfw has the same behaviour! */ 8282 action = PF_DROP; 8283 REASON_SET(&reason, PFRES_MEMORY); 8284 pd.act.log = PF_LOG_FORCE; 8285 DPFPRINTF(PF_DEBUG_MISC, 8286 ("pf: failed to allocate divert tag\n")); 8287 } 8288 } 8289 /* this flag will need revising if the pkt is forwarded */ 8290 if (pd.pf_mtag) 8291 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 8292 8293 if (pd.act.log) { 8294 struct pf_krule *lr; 8295 struct pf_krule_item *ri; 8296 8297 if (s != NULL && s->nat_rule.ptr != NULL && 8298 s->nat_rule.ptr->log & PF_LOG_ALL) 8299 lr = s->nat_rule.ptr; 8300 else 8301 lr = r; 8302 8303 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8304 PFLOG_PACKET(kif, m, AF_INET, reason, lr, a, ruleset, 8305 &pd, (s == NULL)); 8306 if (s) { 8307 SLIST_FOREACH(ri, &s->match_rules, entry) 8308 if (ri->r->log & PF_LOG_ALL) 8309 PFLOG_PACKET(kif, m, AF_INET, reason, 8310 ri->r, a, ruleset, &pd, 0); 8311 } 8312 } 8313 8314 pf_counter_u64_critical_enter(); 8315 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 8316 pd.tot_len); 8317 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 8318 1); 8319 8320 if (action == PF_PASS || r->action == PF_DROP) { 8321 dirndx = (dir == PF_OUT); 8322 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8323 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8324 pf_update_timestamp(r); 8325 8326 if (a != NULL) { 8327 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8328 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8329 } 8330 if (s != NULL) { 8331 struct pf_krule_item *ri; 8332 8333 if (s->nat_rule.ptr != NULL) { 8334 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8335 1); 8336 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8337 pd.tot_len); 8338 } 8339 if (s->src_node != NULL) { 8340 counter_u64_add(s->src_node->packets[dirndx], 8341 1); 8342 counter_u64_add(s->src_node->bytes[dirndx], 8343 pd.tot_len); 8344 } 8345 if (s->nat_src_node != NULL) { 8346 counter_u64_add(s->nat_src_node->packets[dirndx], 8347 1); 8348 counter_u64_add(s->nat_src_node->bytes[dirndx], 8349 pd.tot_len); 8350 } 8351 dirndx = (dir == s->direction) ? 0 : 1; 8352 s->packets[dirndx]++; 8353 s->bytes[dirndx] += pd.tot_len; 8354 SLIST_FOREACH(ri, &s->match_rules, entry) { 8355 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8356 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 8357 } 8358 } 8359 tr = r; 8360 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8361 if (nr != NULL && r == &V_pf_default_rule) 8362 tr = nr; 8363 if (tr->src.addr.type == PF_ADDR_TABLE) 8364 pfr_update_stats(tr->src.addr.p.tbl, 8365 (s == NULL) ? pd.src : 8366 &s->key[(s->direction == PF_IN)]-> 8367 addr[(s->direction == PF_OUT)], 8368 pd.af, pd.tot_len, dir == PF_OUT, 8369 r->action == PF_PASS, tr->src.neg); 8370 if (tr->dst.addr.type == PF_ADDR_TABLE) 8371 pfr_update_stats(tr->dst.addr.p.tbl, 8372 (s == NULL) ? pd.dst : 8373 &s->key[(s->direction == PF_IN)]-> 8374 addr[(s->direction == PF_IN)], 8375 pd.af, pd.tot_len, dir == PF_OUT, 8376 r->action == PF_PASS, tr->dst.neg); 8377 } 8378 pf_counter_u64_critical_exit(); 8379 8380 switch (action) { 8381 case PF_SYNPROXY_DROP: 8382 m_freem(*m0); 8383 case PF_DEFER: 8384 *m0 = NULL; 8385 action = PF_PASS; 8386 break; 8387 case PF_DROP: 8388 m_freem(*m0); 8389 *m0 = NULL; 8390 break; 8391 default: 8392 /* pf_route() returns unlocked. */ 8393 if (rt) { 8394 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 8395 goto out; 8396 } 8397 if (pf_dummynet(&pd, s, r, m0) != 0) { 8398 action = PF_DROP; 8399 REASON_SET(&reason, PFRES_MEMORY); 8400 } 8401 break; 8402 } 8403 8404 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 8405 8406 if (s) 8407 PF_STATE_UNLOCK(s); 8408 8409 out: 8410 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 8411 8412 return (action); 8413 } 8414 #endif /* INET */ 8415 8416 #ifdef INET6 8417 int 8418 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 8419 struct pf_rule_actions *default_actions) 8420 { 8421 struct pfi_kkif *kif; 8422 u_short action, reason = 0; 8423 struct mbuf *m = *m0, *n = NULL; 8424 struct m_tag *mtag; 8425 struct ip6_hdr *h = NULL; 8426 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8427 struct pf_kstate *s = NULL; 8428 struct pf_kruleset *ruleset = NULL; 8429 struct pf_pdesc pd; 8430 int off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0; 8431 uint16_t tag; 8432 uint8_t rt; 8433 8434 PF_RULES_RLOCK_TRACKER; 8435 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8436 M_ASSERTPKTHDR(m); 8437 8438 if (!V_pf_status.running) 8439 return (PF_PASS); 8440 8441 PF_RULES_RLOCK(); 8442 8443 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8444 if (__predict_false(kif == NULL)) { 8445 DPFPRINTF(PF_DEBUG_URGENT, 8446 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 8447 PF_RULES_RUNLOCK(); 8448 return (PF_DROP); 8449 } 8450 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8451 PF_RULES_RUNLOCK(); 8452 return (PF_PASS); 8453 } 8454 8455 if (m->m_flags & M_SKIP_FIREWALL) { 8456 PF_RULES_RUNLOCK(); 8457 return (PF_PASS); 8458 } 8459 8460 memset(&pd, 0, sizeof(pd)); 8461 TAILQ_INIT(&pd.sctp_multihome_jobs); 8462 if (default_actions != NULL) 8463 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8464 pd.pf_mtag = pf_find_mtag(m); 8465 8466 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8467 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8468 8469 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8470 pd.pf_mtag->if_idxgen); 8471 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8472 PF_RULES_RUNLOCK(); 8473 m_freem(*m0); 8474 *m0 = NULL; 8475 return (PF_PASS); 8476 } 8477 PF_RULES_RUNLOCK(); 8478 nd6_output_ifp(ifp, ifp, m, 8479 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 8480 *m0 = NULL; 8481 return (PF_PASS); 8482 } 8483 8484 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8485 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8486 pd.act.flags = pd.pf_mtag->dnflags; 8487 } 8488 8489 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8490 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8491 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 8492 /* Dummynet re-injects packets after they've 8493 * completed their delay. We've already 8494 * processed them, so pass unconditionally. */ 8495 PF_RULES_RUNLOCK(); 8496 return (PF_PASS); 8497 } 8498 8499 pd.sport = pd.dport = NULL; 8500 pd.ip_sum = NULL; 8501 pd.proto_sum = NULL; 8502 pd.dir = dir; 8503 pd.sidx = (dir == PF_IN) ? 0 : 1; 8504 pd.didx = (dir == PF_IN) ? 1 : 0; 8505 pd.af = AF_INET6; 8506 pd.act.rtableid = -1; 8507 8508 h = mtod(m, struct ip6_hdr *); 8509 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8510 8511 /* We do IP header normalization and packet reassembly here */ 8512 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 8513 action = PF_DROP; 8514 goto done; 8515 } 8516 m = *m0; /* pf_normalize messes with m0 */ 8517 h = mtod(m, struct ip6_hdr *); 8518 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8519 8520 /* 8521 * we do not support jumbogram. if we keep going, zero ip6_plen 8522 * will do something bad, so drop the packet for now. 8523 */ 8524 if (htons(h->ip6_plen) == 0) { 8525 action = PF_DROP; 8526 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 8527 goto done; 8528 } 8529 8530 pd.src = (struct pf_addr *)&h->ip6_src; 8531 pd.dst = (struct pf_addr *)&h->ip6_dst; 8532 pd.tos = IPV6_DSCP(h); 8533 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8534 8535 pd.proto = h->ip6_nxt; 8536 do { 8537 switch (pd.proto) { 8538 case IPPROTO_FRAGMENT: 8539 action = pf_test_fragment(&r, kif, m, h, &pd, &a, 8540 &ruleset); 8541 if (action == PF_DROP) 8542 REASON_SET(&reason, PFRES_FRAG); 8543 goto done; 8544 case IPPROTO_ROUTING: { 8545 struct ip6_rthdr rthdr; 8546 8547 if (rh_cnt++) { 8548 DPFPRINTF(PF_DEBUG_MISC, 8549 ("pf: IPv6 more than one rthdr\n")); 8550 action = PF_DROP; 8551 REASON_SET(&reason, PFRES_IPOPTIONS); 8552 pd.act.log = PF_LOG_FORCE; 8553 goto done; 8554 } 8555 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 8556 &reason, pd.af)) { 8557 DPFPRINTF(PF_DEBUG_MISC, 8558 ("pf: IPv6 short rthdr\n")); 8559 action = PF_DROP; 8560 REASON_SET(&reason, PFRES_SHORT); 8561 pd.act.log = PF_LOG_FORCE; 8562 goto done; 8563 } 8564 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8565 DPFPRINTF(PF_DEBUG_MISC, 8566 ("pf: IPv6 rthdr0\n")); 8567 action = PF_DROP; 8568 REASON_SET(&reason, PFRES_IPOPTIONS); 8569 pd.act.log = PF_LOG_FORCE; 8570 goto done; 8571 } 8572 /* FALLTHROUGH */ 8573 } 8574 case IPPROTO_AH: 8575 case IPPROTO_HOPOPTS: 8576 case IPPROTO_DSTOPTS: { 8577 /* get next header and header length */ 8578 struct ip6_ext opt6; 8579 8580 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 8581 NULL, &reason, pd.af)) { 8582 DPFPRINTF(PF_DEBUG_MISC, 8583 ("pf: IPv6 short opt\n")); 8584 action = PF_DROP; 8585 pd.act.log = PF_LOG_FORCE; 8586 goto done; 8587 } 8588 if (pd.proto == IPPROTO_AH) 8589 off += (opt6.ip6e_len + 2) * 4; 8590 else 8591 off += (opt6.ip6e_len + 1) * 8; 8592 pd.proto = opt6.ip6e_nxt; 8593 /* goto the next header */ 8594 break; 8595 } 8596 default: 8597 terminal++; 8598 break; 8599 } 8600 } while (!terminal); 8601 8602 /* if there's no routing header, use unmodified mbuf for checksumming */ 8603 if (!n) 8604 n = m; 8605 8606 switch (pd.proto) { 8607 case IPPROTO_TCP: { 8608 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8609 &action, &reason, AF_INET6)) { 8610 if (action != PF_PASS) 8611 pd.act.log |= PF_LOG_FORCE; 8612 goto done; 8613 } 8614 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8615 pd.sport = &pd.hdr.tcp.th_sport; 8616 pd.dport = &pd.hdr.tcp.th_dport; 8617 8618 /* Respond to SYN with a syncookie. */ 8619 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8620 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8621 pf_syncookie_send(m, off, &pd); 8622 action = PF_DROP; 8623 break; 8624 } 8625 8626 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8627 if (action == PF_DROP) 8628 goto done; 8629 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8630 if (action == PF_PASS) { 8631 if (V_pfsync_update_state_ptr != NULL) 8632 V_pfsync_update_state_ptr(s); 8633 r = s->rule.ptr; 8634 a = s->anchor.ptr; 8635 } else if (s == NULL) { 8636 /* Validate remote SYN|ACK, re-create original SYN if 8637 * valid. */ 8638 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8639 TH_ACK && pf_syncookie_validate(&pd) && 8640 pd.dir == PF_IN) { 8641 struct mbuf *msyn; 8642 8643 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 8644 off, &pd); 8645 if (msyn == NULL) { 8646 action = PF_DROP; 8647 break; 8648 } 8649 8650 action = pf_test6(dir, pflags, ifp, &msyn, inp, 8651 &pd.act); 8652 m_freem(msyn); 8653 if (action != PF_PASS) 8654 break; 8655 8656 action = pf_test_state_tcp(&s, kif, m, off, h, 8657 &pd, &reason); 8658 if (action != PF_PASS || s == NULL) { 8659 action = PF_DROP; 8660 break; 8661 } 8662 8663 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8664 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8665 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8666 8667 action = pf_synproxy(&pd, &s, &reason); 8668 break; 8669 } else { 8670 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8671 &a, &ruleset, inp); 8672 } 8673 } 8674 break; 8675 } 8676 8677 case IPPROTO_UDP: { 8678 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8679 &action, &reason, AF_INET6)) { 8680 if (action != PF_PASS) 8681 pd.act.log |= PF_LOG_FORCE; 8682 goto done; 8683 } 8684 pd.sport = &pd.hdr.udp.uh_sport; 8685 pd.dport = &pd.hdr.udp.uh_dport; 8686 if (pd.hdr.udp.uh_dport == 0 || 8687 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8688 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8689 action = PF_DROP; 8690 REASON_SET(&reason, PFRES_SHORT); 8691 goto done; 8692 } 8693 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8694 if (action == PF_PASS) { 8695 if (V_pfsync_update_state_ptr != NULL) 8696 V_pfsync_update_state_ptr(s); 8697 r = s->rule.ptr; 8698 a = s->anchor.ptr; 8699 } else if (s == NULL) 8700 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8701 &a, &ruleset, inp); 8702 break; 8703 } 8704 8705 case IPPROTO_SCTP: { 8706 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8707 &action, &reason, AF_INET6)) { 8708 if (action != PF_PASS) 8709 pd.act.log |= PF_LOG_FORCE; 8710 goto done; 8711 } 8712 pd.sport = &pd.hdr.sctp.src_port; 8713 pd.dport = &pd.hdr.sctp.dest_port; 8714 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8715 action = PF_DROP; 8716 REASON_SET(&reason, PFRES_SHORT); 8717 goto done; 8718 } 8719 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8720 if (action == PF_DROP) 8721 goto done; 8722 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8723 &reason); 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 { 8730 action = pf_test_rule(&r, &s, kif, m, off, 8731 &pd, &a, &ruleset, inp); 8732 } 8733 break; 8734 } 8735 8736 case IPPROTO_ICMP: { 8737 action = PF_DROP; 8738 DPFPRINTF(PF_DEBUG_MISC, 8739 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 8740 goto done; 8741 } 8742 8743 case IPPROTO_ICMPV6: { 8744 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 8745 &action, &reason, AF_INET6)) { 8746 if (action != PF_PASS) 8747 pd.act.log |= PF_LOG_FORCE; 8748 goto done; 8749 } 8750 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8751 if (action == PF_PASS) { 8752 if (V_pfsync_update_state_ptr != NULL) 8753 V_pfsync_update_state_ptr(s); 8754 r = s->rule.ptr; 8755 a = s->anchor.ptr; 8756 } else if (s == NULL) 8757 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8758 &a, &ruleset, inp); 8759 break; 8760 } 8761 8762 default: 8763 action = pf_test_state_other(&s, kif, m, &pd); 8764 if (action == PF_PASS) { 8765 if (V_pfsync_update_state_ptr != NULL) 8766 V_pfsync_update_state_ptr(s); 8767 r = s->rule.ptr; 8768 a = s->anchor.ptr; 8769 } else if (s == NULL) 8770 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8771 &a, &ruleset, inp); 8772 break; 8773 } 8774 8775 done: 8776 PF_RULES_RUNLOCK(); 8777 if (n != m) { 8778 m_freem(n); 8779 n = NULL; 8780 } 8781 8782 /* handle dangerous IPv6 extension headers. */ 8783 if (action == PF_PASS && rh_cnt && 8784 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8785 action = PF_DROP; 8786 REASON_SET(&reason, PFRES_IPOPTIONS); 8787 pd.act.log = r->log; 8788 DPFPRINTF(PF_DEBUG_MISC, 8789 ("pf: dropping packet with dangerous v6 headers\n")); 8790 } 8791 8792 if (s) { 8793 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8794 tag = s->tag; 8795 rt = s->rt; 8796 } else { 8797 tag = r->tag; 8798 rt = r->rt; 8799 } 8800 8801 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8802 action = PF_DROP; 8803 REASON_SET(&reason, PFRES_MEMORY); 8804 } 8805 8806 pf_scrub_ip6(&m, &pd); 8807 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8808 pf_normalize_mss(m, off, &pd); 8809 8810 if (pd.act.rtableid >= 0) 8811 M_SETFIB(m, pd.act.rtableid); 8812 8813 if (pd.act.flags & PFSTATE_SETPRIO) { 8814 if (pd.tos & IPTOS_LOWDELAY) 8815 use_2nd_queue = 1; 8816 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8817 action = PF_DROP; 8818 REASON_SET(&reason, PFRES_MEMORY); 8819 pd.act.log = PF_LOG_FORCE; 8820 DPFPRINTF(PF_DEBUG_MISC, 8821 ("pf: failed to allocate 802.1q mtag\n")); 8822 } 8823 } 8824 8825 #ifdef ALTQ 8826 if (action == PF_PASS && pd.act.qid) { 8827 if (pd.pf_mtag == NULL && 8828 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8829 action = PF_DROP; 8830 REASON_SET(&reason, PFRES_MEMORY); 8831 } else { 8832 if (s != NULL) 8833 pd.pf_mtag->qid_hash = pf_state_hash(s); 8834 if (pd.tos & IPTOS_LOWDELAY) 8835 pd.pf_mtag->qid = pd.act.pqid; 8836 else 8837 pd.pf_mtag->qid = pd.act.qid; 8838 /* Add hints for ecn. */ 8839 pd.pf_mtag->hdr = h; 8840 } 8841 } 8842 #endif /* ALTQ */ 8843 8844 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8845 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8846 (s->nat_rule.ptr->action == PF_RDR || 8847 s->nat_rule.ptr->action == PF_BINAT) && 8848 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 8849 m->m_flags |= M_SKIP_FIREWALL; 8850 8851 /* XXX: Anybody working on it?! */ 8852 if (r->divert.port) 8853 printf("pf: divert(9) is not supported for IPv6\n"); 8854 8855 if (pd.act.log) { 8856 struct pf_krule *lr; 8857 struct pf_krule_item *ri; 8858 8859 if (s != NULL && s->nat_rule.ptr != NULL && 8860 s->nat_rule.ptr->log & PF_LOG_ALL) 8861 lr = s->nat_rule.ptr; 8862 else 8863 lr = r; 8864 8865 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8866 PFLOG_PACKET(kif, m, AF_INET6, reason, lr, a, ruleset, 8867 &pd, (s == NULL)); 8868 if (s) { 8869 SLIST_FOREACH(ri, &s->match_rules, entry) 8870 if (ri->r->log & PF_LOG_ALL) 8871 PFLOG_PACKET(kif, m, AF_INET6, reason, 8872 ri->r, a, ruleset, &pd, 0); 8873 } 8874 } 8875 8876 pf_counter_u64_critical_enter(); 8877 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 8878 pd.tot_len); 8879 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 8880 1); 8881 8882 if (action == PF_PASS || r->action == PF_DROP) { 8883 dirndx = (dir == PF_OUT); 8884 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8885 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8886 if (a != NULL) { 8887 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8888 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8889 } 8890 if (s != NULL) { 8891 if (s->nat_rule.ptr != NULL) { 8892 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8893 1); 8894 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8895 pd.tot_len); 8896 } 8897 if (s->src_node != NULL) { 8898 counter_u64_add(s->src_node->packets[dirndx], 8899 1); 8900 counter_u64_add(s->src_node->bytes[dirndx], 8901 pd.tot_len); 8902 } 8903 if (s->nat_src_node != NULL) { 8904 counter_u64_add(s->nat_src_node->packets[dirndx], 8905 1); 8906 counter_u64_add(s->nat_src_node->bytes[dirndx], 8907 pd.tot_len); 8908 } 8909 dirndx = (dir == s->direction) ? 0 : 1; 8910 s->packets[dirndx]++; 8911 s->bytes[dirndx] += pd.tot_len; 8912 } 8913 tr = r; 8914 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8915 if (nr != NULL && r == &V_pf_default_rule) 8916 tr = nr; 8917 if (tr->src.addr.type == PF_ADDR_TABLE) 8918 pfr_update_stats(tr->src.addr.p.tbl, 8919 (s == NULL) ? pd.src : 8920 &s->key[(s->direction == PF_IN)]->addr[0], 8921 pd.af, pd.tot_len, dir == PF_OUT, 8922 r->action == PF_PASS, tr->src.neg); 8923 if (tr->dst.addr.type == PF_ADDR_TABLE) 8924 pfr_update_stats(tr->dst.addr.p.tbl, 8925 (s == NULL) ? pd.dst : 8926 &s->key[(s->direction == PF_IN)]->addr[1], 8927 pd.af, pd.tot_len, dir == PF_OUT, 8928 r->action == PF_PASS, tr->dst.neg); 8929 } 8930 pf_counter_u64_critical_exit(); 8931 8932 switch (action) { 8933 case PF_SYNPROXY_DROP: 8934 m_freem(*m0); 8935 case PF_DEFER: 8936 *m0 = NULL; 8937 action = PF_PASS; 8938 break; 8939 case PF_DROP: 8940 m_freem(*m0); 8941 *m0 = NULL; 8942 break; 8943 default: 8944 /* pf_route6() returns unlocked. */ 8945 if (rt) { 8946 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 8947 goto out; 8948 } 8949 if (pf_dummynet(&pd, s, r, m0) != 0) { 8950 action = PF_DROP; 8951 REASON_SET(&reason, PFRES_MEMORY); 8952 } 8953 break; 8954 } 8955 8956 if (s) 8957 PF_STATE_UNLOCK(s); 8958 8959 /* If reassembled packet passed, create new fragments. */ 8960 if (action == PF_PASS && *m0 && dir == PF_OUT && 8961 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 8962 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 8963 8964 out: 8965 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 8966 8967 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 8968 8969 return (action); 8970 } 8971 #endif /* INET6 */ 8972