pf.c revision 265008
1/*- 2 * Copyright (c) 2001 Daniel Hartmeier 3 * Copyright (c) 2002 - 2008 Henning Brauer 4 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * - Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * - Redistributions in binary form must reproduce the above 14 * copyright notice, this list of conditions and the following 15 * disclaimer in the documentation and/or other materials provided 16 * with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 22 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 25 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 26 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 28 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 * 31 * Effort sponsored in part by the Defense Advanced Research Projects 32 * Agency (DARPA) and Air Force Research Laboratory, Air Force 33 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 34 * 35 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 36 */ 37 38#include <sys/cdefs.h> 39__FBSDID("$FreeBSD: stable/10/sys/netpfil/pf/pf.c 265008 2014-04-27 09:05:34Z mm $"); 40 41#include "opt_inet.h" 42#include "opt_inet6.h" 43#include "opt_bpf.h" 44#include "opt_pf.h" 45 46#include <sys/param.h> 47#include <sys/bus.h> 48#include <sys/endian.h> 49#include <sys/hash.h> 50#include <sys/interrupt.h> 51#include <sys/kernel.h> 52#include <sys/kthread.h> 53#include <sys/limits.h> 54#include <sys/mbuf.h> 55#include <sys/md5.h> 56#include <sys/random.h> 57#include <sys/refcount.h> 58#include <sys/socket.h> 59#include <sys/sysctl.h> 60#include <sys/taskqueue.h> 61#include <sys/ucred.h> 62 63#include <net/if.h> 64#include <net/if_types.h> 65#include <net/route.h> 66#include <net/radix_mpath.h> 67#include <net/vnet.h> 68 69#include <net/pfvar.h> 70#include <net/if_pflog.h> 71#include <net/if_pfsync.h> 72 73#include <netinet/in_pcb.h> 74#include <netinet/in_var.h> 75#include <netinet/ip.h> 76#include <netinet/ip_fw.h> 77#include <netinet/ip_icmp.h> 78#include <netinet/icmp_var.h> 79#include <netinet/ip_var.h> 80#include <netinet/tcp.h> 81#include <netinet/tcp_fsm.h> 82#include <netinet/tcp_seq.h> 83#include <netinet/tcp_timer.h> 84#include <netinet/tcp_var.h> 85#include <netinet/udp.h> 86#include <netinet/udp_var.h> 87 88#include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */ 89 90#ifdef INET6 91#include <netinet/ip6.h> 92#include <netinet/icmp6.h> 93#include <netinet6/nd6.h> 94#include <netinet6/ip6_var.h> 95#include <netinet6/in6_pcb.h> 96#endif /* INET6 */ 97 98#include <machine/in_cksum.h> 99#include <security/mac/mac_framework.h> 100 101#define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 102 103/* 104 * Global variables 105 */ 106 107/* state tables */ 108VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]); 109VNET_DEFINE(struct pf_palist, pf_pabuf); 110VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 111VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 112VNET_DEFINE(struct pf_status, pf_status); 113 114VNET_DEFINE(u_int32_t, ticket_altqs_active); 115VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 116VNET_DEFINE(int, altqs_inactive_open); 117VNET_DEFINE(u_int32_t, ticket_pabuf); 118 119VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 120#define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 121VNET_DEFINE(u_char, pf_tcp_secret[16]); 122#define V_pf_tcp_secret VNET(pf_tcp_secret) 123VNET_DEFINE(int, pf_tcp_secret_init); 124#define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 125VNET_DEFINE(int, pf_tcp_iss_off); 126#define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 127 128/* 129 * Queue for pf_intr() sends. 130 */ 131static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 132struct pf_send_entry { 133 STAILQ_ENTRY(pf_send_entry) pfse_next; 134 struct mbuf *pfse_m; 135 enum { 136 PFSE_IP, 137 PFSE_IP6, 138 PFSE_ICMP, 139 PFSE_ICMP6, 140 } pfse_type; 141 union { 142 struct route ro; 143 struct { 144 int type; 145 int code; 146 int mtu; 147 } icmpopts; 148 } u; 149#define pfse_ro u.ro 150#define pfse_icmp_type u.icmpopts.type 151#define pfse_icmp_code u.icmpopts.code 152#define pfse_icmp_mtu u.icmpopts.mtu 153}; 154 155STAILQ_HEAD(pf_send_head, pf_send_entry); 156static VNET_DEFINE(struct pf_send_head, pf_sendqueue); 157#define V_pf_sendqueue VNET(pf_sendqueue) 158 159static struct mtx pf_sendqueue_mtx; 160#define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 161#define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 162 163/* 164 * Queue for pf_overload_task() tasks. 165 */ 166struct pf_overload_entry { 167 SLIST_ENTRY(pf_overload_entry) next; 168 struct pf_addr addr; 169 sa_family_t af; 170 uint8_t dir; 171 struct pf_rule *rule; 172}; 173 174SLIST_HEAD(pf_overload_head, pf_overload_entry); 175static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue); 176#define V_pf_overloadqueue VNET(pf_overloadqueue) 177static VNET_DEFINE(struct task, pf_overloadtask); 178#define V_pf_overloadtask VNET(pf_overloadtask) 179 180static struct mtx pf_overloadqueue_mtx; 181#define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 182#define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 183 184VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules); 185struct mtx pf_unlnkdrules_mtx; 186 187static VNET_DEFINE(uma_zone_t, pf_sources_z); 188#define V_pf_sources_z VNET(pf_sources_z) 189uma_zone_t pf_mtag_z; 190VNET_DEFINE(uma_zone_t, pf_state_z); 191VNET_DEFINE(uma_zone_t, pf_state_key_z); 192 193VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 194#define PFID_CPUBITS 8 195#define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 196#define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 197#define PFID_MAXID (~PFID_CPUMASK) 198CTASSERT((1 << PFID_CPUBITS) > MAXCPU); 199 200static void pf_src_tree_remove_state(struct pf_state *); 201static void pf_init_threshold(struct pf_threshold *, u_int32_t, 202 u_int32_t); 203static void pf_add_threshold(struct pf_threshold *); 204static int pf_check_threshold(struct pf_threshold *); 205 206static void pf_change_ap(struct pf_addr *, u_int16_t *, 207 u_int16_t *, u_int16_t *, struct pf_addr *, 208 u_int16_t, u_int8_t, sa_family_t); 209static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 210 struct tcphdr *, struct pf_state_peer *); 211static void pf_change_icmp(struct pf_addr *, u_int16_t *, 212 struct pf_addr *, struct pf_addr *, u_int16_t, 213 u_int16_t *, u_int16_t *, u_int16_t *, 214 u_int16_t *, u_int8_t, sa_family_t); 215static void pf_send_tcp(struct mbuf *, 216 const struct pf_rule *, sa_family_t, 217 const struct pf_addr *, const struct pf_addr *, 218 u_int16_t, u_int16_t, u_int32_t, u_int32_t, 219 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, 220 u_int16_t, struct ifnet *); 221static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 222 sa_family_t, struct pf_rule *); 223static void pf_detach_state(struct pf_state *); 224static int pf_state_key_attach(struct pf_state_key *, 225 struct pf_state_key *, struct pf_state *); 226static void pf_state_key_detach(struct pf_state *, int); 227static int pf_state_key_ctor(void *, int, void *, int); 228static u_int32_t pf_tcp_iss(struct pf_pdesc *); 229static int pf_test_rule(struct pf_rule **, struct pf_state **, 230 int, struct pfi_kif *, struct mbuf *, int, 231 struct pf_pdesc *, struct pf_rule **, 232 struct pf_ruleset **, struct inpcb *); 233static int pf_create_state(struct pf_rule *, struct pf_rule *, 234 struct pf_rule *, struct pf_pdesc *, 235 struct pf_src_node *, struct pf_state_key *, 236 struct pf_state_key *, struct mbuf *, int, 237 u_int16_t, u_int16_t, int *, struct pfi_kif *, 238 struct pf_state **, int, u_int16_t, u_int16_t, 239 int); 240static int pf_test_fragment(struct pf_rule **, int, 241 struct pfi_kif *, struct mbuf *, void *, 242 struct pf_pdesc *, struct pf_rule **, 243 struct pf_ruleset **); 244static int pf_tcp_track_full(struct pf_state_peer *, 245 struct pf_state_peer *, struct pf_state **, 246 struct pfi_kif *, struct mbuf *, int, 247 struct pf_pdesc *, u_short *, int *); 248static int pf_tcp_track_sloppy(struct pf_state_peer *, 249 struct pf_state_peer *, struct pf_state **, 250 struct pf_pdesc *, u_short *); 251static int pf_test_state_tcp(struct pf_state **, int, 252 struct pfi_kif *, struct mbuf *, int, 253 void *, struct pf_pdesc *, u_short *); 254static int pf_test_state_udp(struct pf_state **, int, 255 struct pfi_kif *, struct mbuf *, int, 256 void *, struct pf_pdesc *); 257static int pf_test_state_icmp(struct pf_state **, int, 258 struct pfi_kif *, struct mbuf *, int, 259 void *, struct pf_pdesc *, u_short *); 260static int pf_test_state_other(struct pf_state **, int, 261 struct pfi_kif *, struct mbuf *, struct pf_pdesc *); 262static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, 263 sa_family_t); 264static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, 265 sa_family_t); 266static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 267 int, u_int16_t); 268static void pf_set_rt_ifp(struct pf_state *, 269 struct pf_addr *); 270static int pf_check_proto_cksum(struct mbuf *, int, int, 271 u_int8_t, sa_family_t); 272static void pf_print_state_parts(struct pf_state *, 273 struct pf_state_key *, struct pf_state_key *); 274static int pf_addr_wrap_neq(struct pf_addr_wrap *, 275 struct pf_addr_wrap *); 276static struct pf_state *pf_find_state(struct pfi_kif *, 277 struct pf_state_key_cmp *, u_int); 278static int pf_src_connlimit(struct pf_state **); 279static void pf_overload_task(void *v, int pending); 280static int pf_insert_src_node(struct pf_src_node **, 281 struct pf_rule *, struct pf_addr *, sa_family_t); 282static u_int pf_purge_expired_states(u_int, int); 283static void pf_purge_unlinked_rules(void); 284static int pf_mtag_uminit(void *, int, int); 285static void pf_mtag_free(struct m_tag *); 286#ifdef INET 287static void pf_route(struct mbuf **, struct pf_rule *, int, 288 struct ifnet *, struct pf_state *, 289 struct pf_pdesc *); 290#endif /* INET */ 291#ifdef INET6 292static void pf_change_a6(struct pf_addr *, u_int16_t *, 293 struct pf_addr *, u_int8_t); 294static void pf_route6(struct mbuf **, struct pf_rule *, int, 295 struct ifnet *, struct pf_state *, 296 struct pf_pdesc *); 297#endif /* INET6 */ 298 299int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 300 301VNET_DECLARE(int, pf_end_threads); 302 303VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 304 305#define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 306 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 307 308#define STATE_LOOKUP(i, k, d, s, pd) \ 309 do { \ 310 (s) = pf_find_state((i), (k), (d)); \ 311 if ((s) == NULL) \ 312 return (PF_DROP); \ 313 if (PACKET_LOOPED(pd)) \ 314 return (PF_PASS); \ 315 if ((d) == PF_OUT && \ 316 (((s)->rule.ptr->rt == PF_ROUTETO && \ 317 (s)->rule.ptr->direction == PF_OUT) || \ 318 ((s)->rule.ptr->rt == PF_REPLYTO && \ 319 (s)->rule.ptr->direction == PF_IN)) && \ 320 (s)->rt_kif != NULL && \ 321 (s)->rt_kif != (i)) \ 322 return (PF_PASS); \ 323 } while (0) 324 325#define BOUND_IFACE(r, k) \ 326 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 327 328#define STATE_INC_COUNTERS(s) \ 329 do { \ 330 counter_u64_add(s->rule.ptr->states_cur, 1); \ 331 counter_u64_add(s->rule.ptr->states_tot, 1); \ 332 if (s->anchor.ptr != NULL) { \ 333 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 334 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 335 } \ 336 if (s->nat_rule.ptr != NULL) { \ 337 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 338 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 339 } \ 340 } while (0) 341 342#define STATE_DEC_COUNTERS(s) \ 343 do { \ 344 if (s->nat_rule.ptr != NULL) \ 345 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 346 if (s->anchor.ptr != NULL) \ 347 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 348 counter_u64_add(s->rule.ptr->states_cur, -1); \ 349 } while (0) 350 351static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 352VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 353VNET_DEFINE(struct pf_idhash *, pf_idhash); 354VNET_DEFINE(u_long, pf_hashmask); 355VNET_DEFINE(struct pf_srchash *, pf_srchash); 356VNET_DEFINE(u_long, pf_srchashmask); 357 358SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)"); 359 360VNET_DEFINE(u_long, pf_hashsize); 361#define V_pf_hashsize VNET(pf_hashsize) 362SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 363 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); 364 365VNET_DEFINE(u_long, pf_srchashsize); 366#define V_pf_srchashsize VNET(pf_srchashsize) 367SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 368 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); 369 370VNET_DEFINE(void *, pf_swi_cookie); 371 372VNET_DEFINE(uint32_t, pf_hashseed); 373#define V_pf_hashseed VNET(pf_hashseed) 374 375static __inline uint32_t 376pf_hashkey(struct pf_state_key *sk) 377{ 378 uint32_t h; 379 380 h = jenkins_hash32((uint32_t *)sk, 381 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 382 V_pf_hashseed); 383 384 return (h & V_pf_hashmask); 385} 386 387static __inline uint32_t 388pf_hashsrc(struct pf_addr *addr, sa_family_t af) 389{ 390 uint32_t h; 391 392 switch (af) { 393 case AF_INET: 394 h = jenkins_hash32((uint32_t *)&addr->v4, 395 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 396 break; 397 case AF_INET6: 398 h = jenkins_hash32((uint32_t *)&addr->v6, 399 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 400 break; 401 default: 402 panic("%s: unknown address family %u", __func__, af); 403 } 404 405 return (h & V_pf_srchashmask); 406} 407 408#ifdef INET6 409void 410pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 411{ 412 switch (af) { 413#ifdef INET 414 case AF_INET: 415 dst->addr32[0] = src->addr32[0]; 416 break; 417#endif /* INET */ 418 case AF_INET6: 419 dst->addr32[0] = src->addr32[0]; 420 dst->addr32[1] = src->addr32[1]; 421 dst->addr32[2] = src->addr32[2]; 422 dst->addr32[3] = src->addr32[3]; 423 break; 424 } 425} 426#endif /* INET6 */ 427 428static void 429pf_init_threshold(struct pf_threshold *threshold, 430 u_int32_t limit, u_int32_t seconds) 431{ 432 threshold->limit = limit * PF_THRESHOLD_MULT; 433 threshold->seconds = seconds; 434 threshold->count = 0; 435 threshold->last = time_uptime; 436} 437 438static void 439pf_add_threshold(struct pf_threshold *threshold) 440{ 441 u_int32_t t = time_uptime, diff = t - threshold->last; 442 443 if (diff >= threshold->seconds) 444 threshold->count = 0; 445 else 446 threshold->count -= threshold->count * diff / 447 threshold->seconds; 448 threshold->count += PF_THRESHOLD_MULT; 449 threshold->last = t; 450} 451 452static int 453pf_check_threshold(struct pf_threshold *threshold) 454{ 455 return (threshold->count > threshold->limit); 456} 457 458static int 459pf_src_connlimit(struct pf_state **state) 460{ 461 struct pf_overload_entry *pfoe; 462 int bad = 0; 463 464 PF_STATE_LOCK_ASSERT(*state); 465 466 (*state)->src_node->conn++; 467 (*state)->src.tcp_est = 1; 468 pf_add_threshold(&(*state)->src_node->conn_rate); 469 470 if ((*state)->rule.ptr->max_src_conn && 471 (*state)->rule.ptr->max_src_conn < 472 (*state)->src_node->conn) { 473 V_pf_status.lcounters[LCNT_SRCCONN]++; 474 bad++; 475 } 476 477 if ((*state)->rule.ptr->max_src_conn_rate.limit && 478 pf_check_threshold(&(*state)->src_node->conn_rate)) { 479 V_pf_status.lcounters[LCNT_SRCCONNRATE]++; 480 bad++; 481 } 482 483 if (!bad) 484 return (0); 485 486 /* Kill this state. */ 487 (*state)->timeout = PFTM_PURGE; 488 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 489 490 if ((*state)->rule.ptr->overload_tbl == NULL) 491 return (1); 492 493 /* Schedule overloading and flushing task. */ 494 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 495 if (pfoe == NULL) 496 return (1); /* too bad :( */ 497 498 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 499 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 500 pfoe->rule = (*state)->rule.ptr; 501 pfoe->dir = (*state)->direction; 502 PF_OVERLOADQ_LOCK(); 503 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 504 PF_OVERLOADQ_UNLOCK(); 505 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 506 507 return (1); 508} 509 510static void 511pf_overload_task(void *v, int pending) 512{ 513 struct pf_overload_head queue; 514 struct pfr_addr p; 515 struct pf_overload_entry *pfoe, *pfoe1; 516 uint32_t killed = 0; 517 518 CURVNET_SET((struct vnet *)v); 519 520 PF_OVERLOADQ_LOCK(); 521 queue = V_pf_overloadqueue; 522 SLIST_INIT(&V_pf_overloadqueue); 523 PF_OVERLOADQ_UNLOCK(); 524 525 bzero(&p, sizeof(p)); 526 SLIST_FOREACH(pfoe, &queue, next) { 527 V_pf_status.lcounters[LCNT_OVERLOAD_TABLE]++; 528 if (V_pf_status.debug >= PF_DEBUG_MISC) { 529 printf("%s: blocking address ", __func__); 530 pf_print_host(&pfoe->addr, 0, pfoe->af); 531 printf("\n"); 532 } 533 534 p.pfra_af = pfoe->af; 535 switch (pfoe->af) { 536#ifdef INET 537 case AF_INET: 538 p.pfra_net = 32; 539 p.pfra_ip4addr = pfoe->addr.v4; 540 break; 541#endif 542#ifdef INET6 543 case AF_INET6: 544 p.pfra_net = 128; 545 p.pfra_ip6addr = pfoe->addr.v6; 546 break; 547#endif 548 } 549 550 PF_RULES_WLOCK(); 551 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 552 PF_RULES_WUNLOCK(); 553 } 554 555 /* 556 * Remove those entries, that don't need flushing. 557 */ 558 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 559 if (pfoe->rule->flush == 0) { 560 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 561 free(pfoe, M_PFTEMP); 562 } else 563 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++; 564 565 /* If nothing to flush, return. */ 566 if (SLIST_EMPTY(&queue)) { 567 CURVNET_RESTORE(); 568 return; 569 } 570 571 for (int i = 0; i <= V_pf_hashmask; i++) { 572 struct pf_idhash *ih = &V_pf_idhash[i]; 573 struct pf_state_key *sk; 574 struct pf_state *s; 575 576 PF_HASHROW_LOCK(ih); 577 LIST_FOREACH(s, &ih->states, entry) { 578 sk = s->key[PF_SK_WIRE]; 579 SLIST_FOREACH(pfoe, &queue, next) 580 if (sk->af == pfoe->af && 581 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 582 pfoe->rule == s->rule.ptr) && 583 ((pfoe->dir == PF_OUT && 584 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 585 (pfoe->dir == PF_IN && 586 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 587 s->timeout = PFTM_PURGE; 588 s->src.state = s->dst.state = TCPS_CLOSED; 589 killed++; 590 } 591 } 592 PF_HASHROW_UNLOCK(ih); 593 } 594 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 595 free(pfoe, M_PFTEMP); 596 if (V_pf_status.debug >= PF_DEBUG_MISC) 597 printf("%s: %u states killed", __func__, killed); 598 599 CURVNET_RESTORE(); 600} 601 602/* 603 * Can return locked on failure, so that we can consistently 604 * allocate and insert a new one. 605 */ 606struct pf_src_node * 607pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af, 608 int returnlocked) 609{ 610 struct pf_srchash *sh; 611 struct pf_src_node *n; 612 613 V_pf_status.scounters[SCNT_SRC_NODE_SEARCH]++; 614 615 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 616 PF_HASHROW_LOCK(sh); 617 LIST_FOREACH(n, &sh->nodes, entry) 618 if (n->rule.ptr == rule && n->af == af && 619 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 620 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 621 break; 622 if (n != NULL || returnlocked == 0) 623 PF_HASHROW_UNLOCK(sh); 624 625 return (n); 626} 627 628static int 629pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, 630 struct pf_addr *src, sa_family_t af) 631{ 632 633 KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK || 634 rule->rpool.opts & PF_POOL_STICKYADDR), 635 ("%s for non-tracking rule %p", __func__, rule)); 636 637 if (*sn == NULL) 638 *sn = pf_find_src_node(src, rule, af, 1); 639 640 if (*sn == NULL) { 641 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 642 643 PF_HASHROW_ASSERT(sh); 644 645 if (!rule->max_src_nodes || 646 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 647 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 648 else 649 V_pf_status.lcounters[LCNT_SRCNODES]++; 650 if ((*sn) == NULL) { 651 PF_HASHROW_UNLOCK(sh); 652 return (-1); 653 } 654 655 pf_init_threshold(&(*sn)->conn_rate, 656 rule->max_src_conn_rate.limit, 657 rule->max_src_conn_rate.seconds); 658 659 (*sn)->af = af; 660 (*sn)->rule.ptr = rule; 661 PF_ACPY(&(*sn)->addr, src, af); 662 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 663 (*sn)->creation = time_uptime; 664 (*sn)->ruletype = rule->action; 665 if ((*sn)->rule.ptr != NULL) 666 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 667 PF_HASHROW_UNLOCK(sh); 668 V_pf_status.scounters[SCNT_SRC_NODE_INSERT]++; 669 V_pf_status.src_nodes++; 670 } else { 671 if (rule->max_src_states && 672 (*sn)->states >= rule->max_src_states) { 673 V_pf_status.lcounters[LCNT_SRCSTATES]++; 674 return (-1); 675 } 676 } 677 return (0); 678} 679 680void 681pf_unlink_src_node_locked(struct pf_src_node *src) 682{ 683#ifdef INVARIANTS 684 struct pf_srchash *sh; 685 686 sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)]; 687 PF_HASHROW_ASSERT(sh); 688#endif 689 LIST_REMOVE(src, entry); 690 if (src->rule.ptr) 691 counter_u64_add(src->rule.ptr->src_nodes, -1); 692 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++; 693 V_pf_status.src_nodes--; 694} 695 696void 697pf_unlink_src_node(struct pf_src_node *src) 698{ 699 struct pf_srchash *sh; 700 701 sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)]; 702 PF_HASHROW_LOCK(sh); 703 pf_unlink_src_node_locked(src); 704 PF_HASHROW_UNLOCK(sh); 705} 706 707static void 708pf_free_src_node(struct pf_src_node *sn) 709{ 710 711 KASSERT(sn->states == 0, ("%s: %p has refs", __func__, sn)); 712 uma_zfree(V_pf_sources_z, sn); 713} 714 715u_int 716pf_free_src_nodes(struct pf_src_node_list *head) 717{ 718 struct pf_src_node *sn, *tmp; 719 u_int count = 0; 720 721 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 722 pf_free_src_node(sn); 723 count++; 724 } 725 726 return (count); 727} 728 729void 730pf_mtag_initialize() 731{ 732 733 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 734 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 735 UMA_ALIGN_PTR, 0); 736} 737 738/* Per-vnet data storage structures initialization. */ 739void 740pf_initialize() 741{ 742 struct pf_keyhash *kh; 743 struct pf_idhash *ih; 744 struct pf_srchash *sh; 745 u_int i; 746 747 TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize); 748 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) 749 V_pf_hashsize = PF_HASHSIZ; 750 TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize); 751 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) 752 V_pf_srchashsize = PF_HASHSIZ / 4; 753 754 V_pf_hashseed = arc4random(); 755 756 /* States and state keys storage. */ 757 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), 758 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 759 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 760 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 761 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 762 763 V_pf_state_key_z = uma_zcreate("pf state keys", 764 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 765 UMA_ALIGN_PTR, 0); 766 V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash), 767 M_PFHASH, M_WAITOK | M_ZERO); 768 V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash), 769 M_PFHASH, M_WAITOK | M_ZERO); 770 V_pf_hashmask = V_pf_hashsize - 1; 771 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 772 i++, kh++, ih++) { 773 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 774 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 775 } 776 777 /* Source nodes. */ 778 V_pf_sources_z = uma_zcreate("pf source nodes", 779 sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 780 0); 781 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 782 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 783 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 784 V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash), 785 M_PFHASH, M_WAITOK|M_ZERO); 786 V_pf_srchashmask = V_pf_srchashsize - 1; 787 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) 788 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 789 790 /* ALTQ */ 791 TAILQ_INIT(&V_pf_altqs[0]); 792 TAILQ_INIT(&V_pf_altqs[1]); 793 TAILQ_INIT(&V_pf_pabuf); 794 V_pf_altqs_active = &V_pf_altqs[0]; 795 V_pf_altqs_inactive = &V_pf_altqs[1]; 796 797 798 /* Send & overload+flush queues. */ 799 STAILQ_INIT(&V_pf_sendqueue); 800 SLIST_INIT(&V_pf_overloadqueue); 801 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 802 mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF); 803 mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL, 804 MTX_DEF); 805 806 /* Unlinked, but may be referenced rules. */ 807 TAILQ_INIT(&V_pf_unlinked_rules); 808 mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF); 809} 810 811void 812pf_mtag_cleanup() 813{ 814 815 uma_zdestroy(pf_mtag_z); 816} 817 818void 819pf_cleanup() 820{ 821 struct pf_keyhash *kh; 822 struct pf_idhash *ih; 823 struct pf_srchash *sh; 824 struct pf_send_entry *pfse, *next; 825 u_int i; 826 827 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; 828 i++, kh++, ih++) { 829 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 830 __func__)); 831 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 832 __func__)); 833 mtx_destroy(&kh->lock); 834 mtx_destroy(&ih->lock); 835 } 836 free(V_pf_keyhash, M_PFHASH); 837 free(V_pf_idhash, M_PFHASH); 838 839 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 840 KASSERT(LIST_EMPTY(&sh->nodes), 841 ("%s: source node hash not empty", __func__)); 842 mtx_destroy(&sh->lock); 843 } 844 free(V_pf_srchash, M_PFHASH); 845 846 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 847 m_freem(pfse->pfse_m); 848 free(pfse, M_PFTEMP); 849 } 850 851 mtx_destroy(&pf_sendqueue_mtx); 852 mtx_destroy(&pf_overloadqueue_mtx); 853 mtx_destroy(&pf_unlnkdrules_mtx); 854 855 uma_zdestroy(V_pf_sources_z); 856 uma_zdestroy(V_pf_state_z); 857 uma_zdestroy(V_pf_state_key_z); 858} 859 860static int 861pf_mtag_uminit(void *mem, int size, int how) 862{ 863 struct m_tag *t; 864 865 t = (struct m_tag *)mem; 866 t->m_tag_cookie = MTAG_ABI_COMPAT; 867 t->m_tag_id = PACKET_TAG_PF; 868 t->m_tag_len = sizeof(struct pf_mtag); 869 t->m_tag_free = pf_mtag_free; 870 871 return (0); 872} 873 874static void 875pf_mtag_free(struct m_tag *t) 876{ 877 878 uma_zfree(pf_mtag_z, t); 879} 880 881struct pf_mtag * 882pf_get_mtag(struct mbuf *m) 883{ 884 struct m_tag *mtag; 885 886 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 887 return ((struct pf_mtag *)(mtag + 1)); 888 889 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 890 if (mtag == NULL) 891 return (NULL); 892 bzero(mtag + 1, sizeof(struct pf_mtag)); 893 m_tag_prepend(m, mtag); 894 895 return ((struct pf_mtag *)(mtag + 1)); 896} 897 898static int 899pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 900 struct pf_state *s) 901{ 902 struct pf_keyhash *khs, *khw, *kh; 903 struct pf_state_key *sk, *cur; 904 struct pf_state *si, *olds = NULL; 905 int idx; 906 907 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 908 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 909 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 910 911 /* 912 * We need to lock hash slots of both keys. To avoid deadlock 913 * we always lock the slot with lower address first. Unlock order 914 * isn't important. 915 * 916 * We also need to lock ID hash slot before dropping key 917 * locks. On success we return with ID hash slot locked. 918 */ 919 920 if (skw == sks) { 921 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 922 PF_HASHROW_LOCK(khs); 923 } else { 924 khs = &V_pf_keyhash[pf_hashkey(sks)]; 925 khw = &V_pf_keyhash[pf_hashkey(skw)]; 926 if (khs == khw) { 927 PF_HASHROW_LOCK(khs); 928 } else if (khs < khw) { 929 PF_HASHROW_LOCK(khs); 930 PF_HASHROW_LOCK(khw); 931 } else { 932 PF_HASHROW_LOCK(khw); 933 PF_HASHROW_LOCK(khs); 934 } 935 } 936 937#define KEYS_UNLOCK() do { \ 938 if (khs != khw) { \ 939 PF_HASHROW_UNLOCK(khs); \ 940 PF_HASHROW_UNLOCK(khw); \ 941 } else \ 942 PF_HASHROW_UNLOCK(khs); \ 943} while (0) 944 945 /* 946 * First run: start with wire key. 947 */ 948 sk = skw; 949 kh = khw; 950 idx = PF_SK_WIRE; 951 952keyattach: 953 LIST_FOREACH(cur, &kh->keys, entry) 954 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 955 break; 956 957 if (cur != NULL) { 958 /* Key exists. Check for same kif, if none, add to key. */ 959 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 960 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 961 962 PF_HASHROW_LOCK(ih); 963 if (si->kif == s->kif && 964 si->direction == s->direction) { 965 if (sk->proto == IPPROTO_TCP && 966 si->src.state >= TCPS_FIN_WAIT_2 && 967 si->dst.state >= TCPS_FIN_WAIT_2) { 968 /* 969 * New state matches an old >FIN_WAIT_2 970 * state. We can't drop key hash locks, 971 * thus we can't unlink it properly. 972 * 973 * As a workaround we drop it into 974 * TCPS_CLOSED state, schedule purge 975 * ASAP and push it into the very end 976 * of the slot TAILQ, so that it won't 977 * conflict with our new state. 978 */ 979 si->src.state = si->dst.state = 980 TCPS_CLOSED; 981 si->timeout = PFTM_PURGE; 982 olds = si; 983 } else { 984 if (V_pf_status.debug >= PF_DEBUG_MISC) { 985 printf("pf: %s key attach " 986 "failed on %s: ", 987 (idx == PF_SK_WIRE) ? 988 "wire" : "stack", 989 s->kif->pfik_name); 990 pf_print_state_parts(s, 991 (idx == PF_SK_WIRE) ? 992 sk : NULL, 993 (idx == PF_SK_STACK) ? 994 sk : NULL); 995 printf(", existing: "); 996 pf_print_state_parts(si, 997 (idx == PF_SK_WIRE) ? 998 sk : NULL, 999 (idx == PF_SK_STACK) ? 1000 sk : NULL); 1001 printf("\n"); 1002 } 1003 PF_HASHROW_UNLOCK(ih); 1004 KEYS_UNLOCK(); 1005 uma_zfree(V_pf_state_key_z, sk); 1006 if (idx == PF_SK_STACK) 1007 pf_detach_state(s); 1008 return (EEXIST); /* collision! */ 1009 } 1010 } 1011 PF_HASHROW_UNLOCK(ih); 1012 } 1013 uma_zfree(V_pf_state_key_z, sk); 1014 s->key[idx] = cur; 1015 } else { 1016 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1017 s->key[idx] = sk; 1018 } 1019 1020stateattach: 1021 /* List is sorted, if-bound states before floating. */ 1022 if (s->kif == V_pfi_all) 1023 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1024 else 1025 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1026 1027 if (olds) { 1028 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1029 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1030 key_list[idx]); 1031 olds = NULL; 1032 } 1033 1034 /* 1035 * Attach done. See how should we (or should not?) 1036 * attach a second key. 1037 */ 1038 if (sks == skw) { 1039 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1040 idx = PF_SK_STACK; 1041 sks = NULL; 1042 goto stateattach; 1043 } else if (sks != NULL) { 1044 /* 1045 * Continue attaching with stack key. 1046 */ 1047 sk = sks; 1048 kh = khs; 1049 idx = PF_SK_STACK; 1050 sks = NULL; 1051 goto keyattach; 1052 } 1053 1054 PF_STATE_LOCK(s); 1055 KEYS_UNLOCK(); 1056 1057 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1058 ("%s failure", __func__)); 1059 1060 return (0); 1061#undef KEYS_UNLOCK 1062} 1063 1064static void 1065pf_detach_state(struct pf_state *s) 1066{ 1067 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1068 struct pf_keyhash *kh; 1069 1070 if (sks != NULL) { 1071 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1072 PF_HASHROW_LOCK(kh); 1073 if (s->key[PF_SK_STACK] != NULL) 1074 pf_state_key_detach(s, PF_SK_STACK); 1075 /* 1076 * If both point to same key, then we are done. 1077 */ 1078 if (sks == s->key[PF_SK_WIRE]) { 1079 pf_state_key_detach(s, PF_SK_WIRE); 1080 PF_HASHROW_UNLOCK(kh); 1081 return; 1082 } 1083 PF_HASHROW_UNLOCK(kh); 1084 } 1085 1086 if (s->key[PF_SK_WIRE] != NULL) { 1087 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1088 PF_HASHROW_LOCK(kh); 1089 if (s->key[PF_SK_WIRE] != NULL) 1090 pf_state_key_detach(s, PF_SK_WIRE); 1091 PF_HASHROW_UNLOCK(kh); 1092 } 1093} 1094 1095static void 1096pf_state_key_detach(struct pf_state *s, int idx) 1097{ 1098 struct pf_state_key *sk = s->key[idx]; 1099#ifdef INVARIANTS 1100 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1101 1102 PF_HASHROW_ASSERT(kh); 1103#endif 1104 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1105 s->key[idx] = NULL; 1106 1107 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1108 LIST_REMOVE(sk, entry); 1109 uma_zfree(V_pf_state_key_z, sk); 1110 } 1111} 1112 1113static int 1114pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1115{ 1116 struct pf_state_key *sk = mem; 1117 1118 bzero(sk, sizeof(struct pf_state_key_cmp)); 1119 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1120 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1121 1122 return (0); 1123} 1124 1125struct pf_state_key * 1126pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1127 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1128{ 1129 struct pf_state_key *sk; 1130 1131 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1132 if (sk == NULL) 1133 return (NULL); 1134 1135 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1136 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1137 sk->port[pd->sidx] = sport; 1138 sk->port[pd->didx] = dport; 1139 sk->proto = pd->proto; 1140 sk->af = pd->af; 1141 1142 return (sk); 1143} 1144 1145struct pf_state_key * 1146pf_state_key_clone(struct pf_state_key *orig) 1147{ 1148 struct pf_state_key *sk; 1149 1150 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1151 if (sk == NULL) 1152 return (NULL); 1153 1154 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1155 1156 return (sk); 1157} 1158 1159int 1160pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, 1161 struct pf_state_key *sks, struct pf_state *s) 1162{ 1163 struct pf_idhash *ih; 1164 struct pf_state *cur; 1165 int error; 1166 1167 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1168 ("%s: sks not pristine", __func__)); 1169 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1170 ("%s: skw not pristine", __func__)); 1171 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1172 1173 s->kif = kif; 1174 1175 if (s->id == 0 && s->creatorid == 0) { 1176 /* XXX: should be atomic, but probability of collision low */ 1177 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1178 V_pf_stateid[curcpu] = 1; 1179 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1180 s->id = htobe64(s->id); 1181 s->creatorid = V_pf_status.hostid; 1182 } 1183 1184 /* Returns with ID locked on success. */ 1185 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1186 return (error); 1187 1188 ih = &V_pf_idhash[PF_IDHASH(s)]; 1189 PF_HASHROW_ASSERT(ih); 1190 LIST_FOREACH(cur, &ih->states, entry) 1191 if (cur->id == s->id && cur->creatorid == s->creatorid) 1192 break; 1193 1194 if (cur != NULL) { 1195 PF_HASHROW_UNLOCK(ih); 1196 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1197 printf("pf: state ID collision: " 1198 "id: %016llx creatorid: %08x\n", 1199 (unsigned long long)be64toh(s->id), 1200 ntohl(s->creatorid)); 1201 } 1202 pf_detach_state(s); 1203 return (EEXIST); 1204 } 1205 LIST_INSERT_HEAD(&ih->states, s, entry); 1206 /* One for keys, one for ID hash. */ 1207 refcount_init(&s->refs, 2); 1208 1209 V_pf_status.fcounters[FCNT_STATE_INSERT]++; 1210 if (pfsync_insert_state_ptr != NULL) 1211 pfsync_insert_state_ptr(s); 1212 1213 /* Returns locked. */ 1214 return (0); 1215} 1216 1217/* 1218 * Find state by ID: returns with locked row on success. 1219 */ 1220struct pf_state * 1221pf_find_state_byid(uint64_t id, uint32_t creatorid) 1222{ 1223 struct pf_idhash *ih; 1224 struct pf_state *s; 1225 1226 V_pf_status.fcounters[FCNT_STATE_SEARCH]++; 1227 1228 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; 1229 1230 PF_HASHROW_LOCK(ih); 1231 LIST_FOREACH(s, &ih->states, entry) 1232 if (s->id == id && s->creatorid == creatorid) 1233 break; 1234 1235 if (s == NULL) 1236 PF_HASHROW_UNLOCK(ih); 1237 1238 return (s); 1239} 1240 1241/* 1242 * Find state by key. 1243 * Returns with ID hash slot locked on success. 1244 */ 1245static struct pf_state * 1246pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir) 1247{ 1248 struct pf_keyhash *kh; 1249 struct pf_state_key *sk; 1250 struct pf_state *s; 1251 int idx; 1252 1253 V_pf_status.fcounters[FCNT_STATE_SEARCH]++; 1254 1255 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1256 1257 PF_HASHROW_LOCK(kh); 1258 LIST_FOREACH(sk, &kh->keys, entry) 1259 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1260 break; 1261 if (sk == NULL) { 1262 PF_HASHROW_UNLOCK(kh); 1263 return (NULL); 1264 } 1265 1266 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1267 1268 /* List is sorted, if-bound states before floating ones. */ 1269 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1270 if (s->kif == V_pfi_all || s->kif == kif) { 1271 PF_STATE_LOCK(s); 1272 PF_HASHROW_UNLOCK(kh); 1273 if (s->timeout >= PFTM_MAX) { 1274 /* 1275 * State is either being processed by 1276 * pf_unlink_state() in an other thread, or 1277 * is scheduled for immediate expiry. 1278 */ 1279 PF_STATE_UNLOCK(s); 1280 return (NULL); 1281 } 1282 return (s); 1283 } 1284 PF_HASHROW_UNLOCK(kh); 1285 1286 return (NULL); 1287} 1288 1289struct pf_state * 1290pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1291{ 1292 struct pf_keyhash *kh; 1293 struct pf_state_key *sk; 1294 struct pf_state *s, *ret = NULL; 1295 int idx, inout = 0; 1296 1297 V_pf_status.fcounters[FCNT_STATE_SEARCH]++; 1298 1299 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1300 1301 PF_HASHROW_LOCK(kh); 1302 LIST_FOREACH(sk, &kh->keys, entry) 1303 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1304 break; 1305 if (sk == NULL) { 1306 PF_HASHROW_UNLOCK(kh); 1307 return (NULL); 1308 } 1309 switch (dir) { 1310 case PF_IN: 1311 idx = PF_SK_WIRE; 1312 break; 1313 case PF_OUT: 1314 idx = PF_SK_STACK; 1315 break; 1316 case PF_INOUT: 1317 idx = PF_SK_WIRE; 1318 inout = 1; 1319 break; 1320 default: 1321 panic("%s: dir %u", __func__, dir); 1322 } 1323second_run: 1324 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1325 if (more == NULL) { 1326 PF_HASHROW_UNLOCK(kh); 1327 return (s); 1328 } 1329 1330 if (ret) 1331 (*more)++; 1332 else 1333 ret = s; 1334 } 1335 if (inout == 1) { 1336 inout = 0; 1337 idx = PF_SK_STACK; 1338 goto second_run; 1339 } 1340 PF_HASHROW_UNLOCK(kh); 1341 1342 return (ret); 1343} 1344 1345/* END state table stuff */ 1346 1347static void 1348pf_send(struct pf_send_entry *pfse) 1349{ 1350 1351 PF_SENDQ_LOCK(); 1352 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1353 PF_SENDQ_UNLOCK(); 1354 swi_sched(V_pf_swi_cookie, 0); 1355} 1356 1357void 1358pf_intr(void *v) 1359{ 1360 struct pf_send_head queue; 1361 struct pf_send_entry *pfse, *next; 1362 1363 CURVNET_SET((struct vnet *)v); 1364 1365 PF_SENDQ_LOCK(); 1366 queue = V_pf_sendqueue; 1367 STAILQ_INIT(&V_pf_sendqueue); 1368 PF_SENDQ_UNLOCK(); 1369 1370 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1371 switch (pfse->pfse_type) { 1372#ifdef INET 1373 case PFSE_IP: 1374 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); 1375 break; 1376 case PFSE_ICMP: 1377 icmp_error(pfse->pfse_m, pfse->pfse_icmp_type, 1378 pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu); 1379 break; 1380#endif /* INET */ 1381#ifdef INET6 1382 case PFSE_IP6: 1383 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, 1384 NULL); 1385 break; 1386 case PFSE_ICMP6: 1387 icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type, 1388 pfse->pfse_icmp_code, pfse->pfse_icmp_mtu); 1389 break; 1390#endif /* INET6 */ 1391 default: 1392 panic("%s: unknown type", __func__); 1393 } 1394 free(pfse, M_PFTEMP); 1395 } 1396 CURVNET_RESTORE(); 1397} 1398 1399void 1400pf_purge_thread(void *v) 1401{ 1402 u_int idx = 0; 1403 1404 CURVNET_SET((struct vnet *)v); 1405 1406 for (;;) { 1407 PF_RULES_RLOCK(); 1408 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10); 1409 1410 if (V_pf_end_threads) { 1411 /* 1412 * To cleanse up all kifs and rules we need 1413 * two runs: first one clears reference flags, 1414 * then pf_purge_expired_states() doesn't 1415 * raise them, and then second run frees. 1416 */ 1417 PF_RULES_RUNLOCK(); 1418 pf_purge_unlinked_rules(); 1419 pfi_kif_purge(); 1420 1421 /* 1422 * Now purge everything. 1423 */ 1424 pf_purge_expired_states(0, V_pf_hashmask); 1425 pf_purge_expired_fragments(); 1426 pf_purge_expired_src_nodes(); 1427 1428 /* 1429 * Now all kifs & rules should be unreferenced, 1430 * thus should be successfully freed. 1431 */ 1432 pf_purge_unlinked_rules(); 1433 pfi_kif_purge(); 1434 1435 /* 1436 * Announce success and exit. 1437 */ 1438 PF_RULES_RLOCK(); 1439 V_pf_end_threads++; 1440 PF_RULES_RUNLOCK(); 1441 wakeup(pf_purge_thread); 1442 kproc_exit(0); 1443 } 1444 PF_RULES_RUNLOCK(); 1445 1446 /* Process 1/interval fraction of the state table every run. */ 1447 idx = pf_purge_expired_states(idx, V_pf_hashmask / 1448 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1449 1450 /* Purge other expired types every PFTM_INTERVAL seconds. */ 1451 if (idx == 0) { 1452 /* 1453 * Order is important: 1454 * - states and src nodes reference rules 1455 * - states and rules reference kifs 1456 */ 1457 pf_purge_expired_fragments(); 1458 pf_purge_expired_src_nodes(); 1459 pf_purge_unlinked_rules(); 1460 pfi_kif_purge(); 1461 } 1462 } 1463 /* not reached */ 1464 CURVNET_RESTORE(); 1465} 1466 1467u_int32_t 1468pf_state_expires(const struct pf_state *state) 1469{ 1470 u_int32_t timeout; 1471 u_int32_t start; 1472 u_int32_t end; 1473 u_int32_t states; 1474 1475 /* handle all PFTM_* > PFTM_MAX here */ 1476 if (state->timeout == PFTM_PURGE) 1477 return (time_uptime); 1478 KASSERT(state->timeout != PFTM_UNLINKED, 1479 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1480 KASSERT((state->timeout < PFTM_MAX), 1481 ("pf_state_expires: timeout > PFTM_MAX")); 1482 timeout = state->rule.ptr->timeout[state->timeout]; 1483 if (!timeout) 1484 timeout = V_pf_default_rule.timeout[state->timeout]; 1485 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1486 if (start) { 1487 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1488 states = counter_u64_fetch(state->rule.ptr->states_cur); 1489 } else { 1490 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1491 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1492 states = V_pf_status.states; 1493 } 1494 if (end && states > start && start < end) { 1495 if (states < end) 1496 return (state->expire + timeout * (end - states) / 1497 (end - start)); 1498 else 1499 return (time_uptime); 1500 } 1501 return (state->expire + timeout); 1502} 1503 1504void 1505pf_purge_expired_src_nodes() 1506{ 1507 struct pf_src_node_list freelist; 1508 struct pf_srchash *sh; 1509 struct pf_src_node *cur, *next; 1510 int i; 1511 1512 LIST_INIT(&freelist); 1513 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { 1514 PF_HASHROW_LOCK(sh); 1515 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1516 if (cur->states == 0 && cur->expire <= time_uptime) { 1517 pf_unlink_src_node_locked(cur); 1518 LIST_INSERT_HEAD(&freelist, cur, entry); 1519 } else if (cur->rule.ptr != NULL) 1520 cur->rule.ptr->rule_flag |= PFRULE_REFS; 1521 PF_HASHROW_UNLOCK(sh); 1522 } 1523 1524 pf_free_src_nodes(&freelist); 1525} 1526 1527static void 1528pf_src_tree_remove_state(struct pf_state *s) 1529{ 1530 u_int32_t timeout; 1531 1532 if (s->src_node != NULL) { 1533 if (s->src.tcp_est) 1534 --s->src_node->conn; 1535 if (--s->src_node->states == 0) { 1536 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1537 if (!timeout) 1538 timeout = 1539 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1540 s->src_node->expire = time_uptime + timeout; 1541 } 1542 } 1543 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1544 if (--s->nat_src_node->states == 0) { 1545 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE]; 1546 if (!timeout) 1547 timeout = 1548 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1549 s->nat_src_node->expire = time_uptime + timeout; 1550 } 1551 } 1552 s->src_node = s->nat_src_node = NULL; 1553} 1554 1555/* 1556 * Unlink and potentilly free a state. Function may be 1557 * called with ID hash row locked, but always returns 1558 * unlocked, since it needs to go through key hash locking. 1559 */ 1560int 1561pf_unlink_state(struct pf_state *s, u_int flags) 1562{ 1563 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 1564 1565 if ((flags & PF_ENTER_LOCKED) == 0) 1566 PF_HASHROW_LOCK(ih); 1567 else 1568 PF_HASHROW_ASSERT(ih); 1569 1570 if (s->timeout == PFTM_UNLINKED) { 1571 /* 1572 * State is being processed 1573 * by pf_unlink_state() in 1574 * an other thread. 1575 */ 1576 PF_HASHROW_UNLOCK(ih); 1577 return (0); /* XXXGL: undefined actually */ 1578 } 1579 1580 if (s->src.state == PF_TCPS_PROXY_DST) { 1581 /* XXX wire key the right one? */ 1582 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, 1583 &s->key[PF_SK_WIRE]->addr[1], 1584 &s->key[PF_SK_WIRE]->addr[0], 1585 s->key[PF_SK_WIRE]->port[1], 1586 s->key[PF_SK_WIRE]->port[0], 1587 s->src.seqhi, s->src.seqlo + 1, 1588 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); 1589 } 1590 1591 LIST_REMOVE(s, entry); 1592 pf_src_tree_remove_state(s); 1593 1594 if (pfsync_delete_state_ptr != NULL) 1595 pfsync_delete_state_ptr(s); 1596 1597 STATE_DEC_COUNTERS(s); 1598 1599 s->timeout = PFTM_UNLINKED; 1600 1601 PF_HASHROW_UNLOCK(ih); 1602 1603 pf_detach_state(s); 1604 refcount_release(&s->refs); 1605 1606 return (pf_release_state(s)); 1607} 1608 1609void 1610pf_free_state(struct pf_state *cur) 1611{ 1612 1613 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 1614 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 1615 cur->timeout)); 1616 1617 pf_normalize_tcp_cleanup(cur); 1618 uma_zfree(V_pf_state_z, cur); 1619 V_pf_status.fcounters[FCNT_STATE_REMOVALS]++; 1620} 1621 1622/* 1623 * Called only from pf_purge_thread(), thus serialized. 1624 */ 1625static u_int 1626pf_purge_expired_states(u_int i, int maxcheck) 1627{ 1628 struct pf_idhash *ih; 1629 struct pf_state *s; 1630 1631 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1632 1633 /* 1634 * Go through hash and unlink states that expire now. 1635 */ 1636 while (maxcheck > 0) { 1637 1638 ih = &V_pf_idhash[i]; 1639relock: 1640 PF_HASHROW_LOCK(ih); 1641 LIST_FOREACH(s, &ih->states, entry) { 1642 if (pf_state_expires(s) <= time_uptime) { 1643 V_pf_status.states -= 1644 pf_unlink_state(s, PF_ENTER_LOCKED); 1645 goto relock; 1646 } 1647 s->rule.ptr->rule_flag |= PFRULE_REFS; 1648 if (s->nat_rule.ptr != NULL) 1649 s->nat_rule.ptr->rule_flag |= PFRULE_REFS; 1650 if (s->anchor.ptr != NULL) 1651 s->anchor.ptr->rule_flag |= PFRULE_REFS; 1652 s->kif->pfik_flags |= PFI_IFLAG_REFS; 1653 if (s->rt_kif) 1654 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 1655 } 1656 PF_HASHROW_UNLOCK(ih); 1657 1658 /* Return when we hit end of hash. */ 1659 if (++i > V_pf_hashmask) { 1660 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1661 return (0); 1662 } 1663 1664 maxcheck--; 1665 } 1666 1667 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1668 1669 return (i); 1670} 1671 1672static void 1673pf_purge_unlinked_rules() 1674{ 1675 struct pf_rulequeue tmpq; 1676 struct pf_rule *r, *r1; 1677 1678 /* 1679 * If we have overloading task pending, then we'd 1680 * better skip purging this time. There is a tiny 1681 * probability that overloading task references 1682 * an already unlinked rule. 1683 */ 1684 PF_OVERLOADQ_LOCK(); 1685 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 1686 PF_OVERLOADQ_UNLOCK(); 1687 return; 1688 } 1689 PF_OVERLOADQ_UNLOCK(); 1690 1691 /* 1692 * Do naive mark-and-sweep garbage collecting of old rules. 1693 * Reference flag is raised by pf_purge_expired_states() 1694 * and pf_purge_expired_src_nodes(). 1695 * 1696 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 1697 * use a temporary queue. 1698 */ 1699 TAILQ_INIT(&tmpq); 1700 PF_UNLNKDRULES_LOCK(); 1701 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 1702 if (!(r->rule_flag & PFRULE_REFS)) { 1703 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 1704 TAILQ_INSERT_TAIL(&tmpq, r, entries); 1705 } else 1706 r->rule_flag &= ~PFRULE_REFS; 1707 } 1708 PF_UNLNKDRULES_UNLOCK(); 1709 1710 if (!TAILQ_EMPTY(&tmpq)) { 1711 PF_RULES_WLOCK(); 1712 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 1713 TAILQ_REMOVE(&tmpq, r, entries); 1714 pf_free_rule(r); 1715 } 1716 PF_RULES_WUNLOCK(); 1717 } 1718} 1719 1720void 1721pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 1722{ 1723 switch (af) { 1724#ifdef INET 1725 case AF_INET: { 1726 u_int32_t a = ntohl(addr->addr32[0]); 1727 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 1728 (a>>8)&255, a&255); 1729 if (p) { 1730 p = ntohs(p); 1731 printf(":%u", p); 1732 } 1733 break; 1734 } 1735#endif /* INET */ 1736#ifdef INET6 1737 case AF_INET6: { 1738 u_int16_t b; 1739 u_int8_t i, curstart, curend, maxstart, maxend; 1740 curstart = curend = maxstart = maxend = 255; 1741 for (i = 0; i < 8; i++) { 1742 if (!addr->addr16[i]) { 1743 if (curstart == 255) 1744 curstart = i; 1745 curend = i; 1746 } else { 1747 if ((curend - curstart) > 1748 (maxend - maxstart)) { 1749 maxstart = curstart; 1750 maxend = curend; 1751 } 1752 curstart = curend = 255; 1753 } 1754 } 1755 if ((curend - curstart) > 1756 (maxend - maxstart)) { 1757 maxstart = curstart; 1758 maxend = curend; 1759 } 1760 for (i = 0; i < 8; i++) { 1761 if (i >= maxstart && i <= maxend) { 1762 if (i == 0) 1763 printf(":"); 1764 if (i == maxend) 1765 printf(":"); 1766 } else { 1767 b = ntohs(addr->addr16[i]); 1768 printf("%x", b); 1769 if (i < 7) 1770 printf(":"); 1771 } 1772 } 1773 if (p) { 1774 p = ntohs(p); 1775 printf("[%u]", p); 1776 } 1777 break; 1778 } 1779#endif /* INET6 */ 1780 } 1781} 1782 1783void 1784pf_print_state(struct pf_state *s) 1785{ 1786 pf_print_state_parts(s, NULL, NULL); 1787} 1788 1789static void 1790pf_print_state_parts(struct pf_state *s, 1791 struct pf_state_key *skwp, struct pf_state_key *sksp) 1792{ 1793 struct pf_state_key *skw, *sks; 1794 u_int8_t proto, dir; 1795 1796 /* Do our best to fill these, but they're skipped if NULL */ 1797 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 1798 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 1799 proto = skw ? skw->proto : (sks ? sks->proto : 0); 1800 dir = s ? s->direction : 0; 1801 1802 switch (proto) { 1803 case IPPROTO_IPV4: 1804 printf("IPv4"); 1805 break; 1806 case IPPROTO_IPV6: 1807 printf("IPv6"); 1808 break; 1809 case IPPROTO_TCP: 1810 printf("TCP"); 1811 break; 1812 case IPPROTO_UDP: 1813 printf("UDP"); 1814 break; 1815 case IPPROTO_ICMP: 1816 printf("ICMP"); 1817 break; 1818 case IPPROTO_ICMPV6: 1819 printf("ICMPv6"); 1820 break; 1821 default: 1822 printf("%u", skw->proto); 1823 break; 1824 } 1825 switch (dir) { 1826 case PF_IN: 1827 printf(" in"); 1828 break; 1829 case PF_OUT: 1830 printf(" out"); 1831 break; 1832 } 1833 if (skw) { 1834 printf(" wire: "); 1835 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 1836 printf(" "); 1837 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 1838 } 1839 if (sks) { 1840 printf(" stack: "); 1841 if (sks != skw) { 1842 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 1843 printf(" "); 1844 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 1845 } else 1846 printf("-"); 1847 } 1848 if (s) { 1849 if (proto == IPPROTO_TCP) { 1850 printf(" [lo=%u high=%u win=%u modulator=%u", 1851 s->src.seqlo, s->src.seqhi, 1852 s->src.max_win, s->src.seqdiff); 1853 if (s->src.wscale && s->dst.wscale) 1854 printf(" wscale=%u", 1855 s->src.wscale & PF_WSCALE_MASK); 1856 printf("]"); 1857 printf(" [lo=%u high=%u win=%u modulator=%u", 1858 s->dst.seqlo, s->dst.seqhi, 1859 s->dst.max_win, s->dst.seqdiff); 1860 if (s->src.wscale && s->dst.wscale) 1861 printf(" wscale=%u", 1862 s->dst.wscale & PF_WSCALE_MASK); 1863 printf("]"); 1864 } 1865 printf(" %u:%u", s->src.state, s->dst.state); 1866 } 1867} 1868 1869void 1870pf_print_flags(u_int8_t f) 1871{ 1872 if (f) 1873 printf(" "); 1874 if (f & TH_FIN) 1875 printf("F"); 1876 if (f & TH_SYN) 1877 printf("S"); 1878 if (f & TH_RST) 1879 printf("R"); 1880 if (f & TH_PUSH) 1881 printf("P"); 1882 if (f & TH_ACK) 1883 printf("A"); 1884 if (f & TH_URG) 1885 printf("U"); 1886 if (f & TH_ECE) 1887 printf("E"); 1888 if (f & TH_CWR) 1889 printf("W"); 1890} 1891 1892#define PF_SET_SKIP_STEPS(i) \ 1893 do { \ 1894 while (head[i] != cur) { \ 1895 head[i]->skip[i].ptr = cur; \ 1896 head[i] = TAILQ_NEXT(head[i], entries); \ 1897 } \ 1898 } while (0) 1899 1900void 1901pf_calc_skip_steps(struct pf_rulequeue *rules) 1902{ 1903 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; 1904 int i; 1905 1906 cur = TAILQ_FIRST(rules); 1907 prev = cur; 1908 for (i = 0; i < PF_SKIP_COUNT; ++i) 1909 head[i] = cur; 1910 while (cur != NULL) { 1911 1912 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 1913 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 1914 if (cur->direction != prev->direction) 1915 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 1916 if (cur->af != prev->af) 1917 PF_SET_SKIP_STEPS(PF_SKIP_AF); 1918 if (cur->proto != prev->proto) 1919 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 1920 if (cur->src.neg != prev->src.neg || 1921 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 1922 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 1923 if (cur->src.port[0] != prev->src.port[0] || 1924 cur->src.port[1] != prev->src.port[1] || 1925 cur->src.port_op != prev->src.port_op) 1926 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 1927 if (cur->dst.neg != prev->dst.neg || 1928 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 1929 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 1930 if (cur->dst.port[0] != prev->dst.port[0] || 1931 cur->dst.port[1] != prev->dst.port[1] || 1932 cur->dst.port_op != prev->dst.port_op) 1933 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 1934 1935 prev = cur; 1936 cur = TAILQ_NEXT(cur, entries); 1937 } 1938 for (i = 0; i < PF_SKIP_COUNT; ++i) 1939 PF_SET_SKIP_STEPS(i); 1940} 1941 1942static int 1943pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 1944{ 1945 if (aw1->type != aw2->type) 1946 return (1); 1947 switch (aw1->type) { 1948 case PF_ADDR_ADDRMASK: 1949 case PF_ADDR_RANGE: 1950 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0)) 1951 return (1); 1952 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0)) 1953 return (1); 1954 return (0); 1955 case PF_ADDR_DYNIFTL: 1956 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 1957 case PF_ADDR_NOROUTE: 1958 case PF_ADDR_URPFFAILED: 1959 return (0); 1960 case PF_ADDR_TABLE: 1961 return (aw1->p.tbl != aw2->p.tbl); 1962 default: 1963 printf("invalid address type: %d\n", aw1->type); 1964 return (1); 1965 } 1966} 1967 1968u_int16_t 1969pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 1970{ 1971 u_int32_t l; 1972 1973 if (udp && !cksum) 1974 return (0x0000); 1975 l = cksum + old - new; 1976 l = (l >> 16) + (l & 65535); 1977 l = l & 65535; 1978 if (udp && !l) 1979 return (0xFFFF); 1980 return (l); 1981} 1982 1983static void 1984pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, 1985 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) 1986{ 1987 struct pf_addr ao; 1988 u_int16_t po = *p; 1989 1990 PF_ACPY(&ao, a, af); 1991 PF_ACPY(a, an, af); 1992 1993 *p = pn; 1994 1995 switch (af) { 1996#ifdef INET 1997 case AF_INET: 1998 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 1999 ao.addr16[0], an->addr16[0], 0), 2000 ao.addr16[1], an->addr16[1], 0); 2001 *p = pn; 2002 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 2003 ao.addr16[0], an->addr16[0], u), 2004 ao.addr16[1], an->addr16[1], u), 2005 po, pn, u); 2006 break; 2007#endif /* INET */ 2008#ifdef INET6 2009 case AF_INET6: 2010 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2011 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2012 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc, 2013 ao.addr16[0], an->addr16[0], u), 2014 ao.addr16[1], an->addr16[1], u), 2015 ao.addr16[2], an->addr16[2], u), 2016 ao.addr16[3], an->addr16[3], u), 2017 ao.addr16[4], an->addr16[4], u), 2018 ao.addr16[5], an->addr16[5], u), 2019 ao.addr16[6], an->addr16[6], u), 2020 ao.addr16[7], an->addr16[7], u), 2021 po, pn, u); 2022 break; 2023#endif /* INET6 */ 2024 } 2025} 2026 2027 2028/* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2029void 2030pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2031{ 2032 u_int32_t ao; 2033 2034 memcpy(&ao, a, sizeof(ao)); 2035 memcpy(a, &an, sizeof(u_int32_t)); 2036 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2037 ao % 65536, an % 65536, u); 2038} 2039 2040#ifdef INET6 2041static void 2042pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2043{ 2044 struct pf_addr ao; 2045 2046 PF_ACPY(&ao, a, AF_INET6); 2047 PF_ACPY(a, an, AF_INET6); 2048 2049 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2050 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2051 pf_cksum_fixup(pf_cksum_fixup(*c, 2052 ao.addr16[0], an->addr16[0], u), 2053 ao.addr16[1], an->addr16[1], u), 2054 ao.addr16[2], an->addr16[2], u), 2055 ao.addr16[3], an->addr16[3], u), 2056 ao.addr16[4], an->addr16[4], u), 2057 ao.addr16[5], an->addr16[5], u), 2058 ao.addr16[6], an->addr16[6], u), 2059 ao.addr16[7], an->addr16[7], u); 2060} 2061#endif /* INET6 */ 2062 2063static void 2064pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2065 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2066 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2067{ 2068 struct pf_addr oia, ooa; 2069 2070 PF_ACPY(&oia, ia, af); 2071 if (oa) 2072 PF_ACPY(&ooa, oa, af); 2073 2074 /* Change inner protocol port, fix inner protocol checksum. */ 2075 if (ip != NULL) { 2076 u_int16_t oip = *ip; 2077 u_int32_t opc; 2078 2079 if (pc != NULL) 2080 opc = *pc; 2081 *ip = np; 2082 if (pc != NULL) 2083 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2084 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2085 if (pc != NULL) 2086 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2087 } 2088 /* Change inner ip address, fix inner ip and icmp checksums. */ 2089 PF_ACPY(ia, na, af); 2090 switch (af) { 2091#ifdef INET 2092 case AF_INET: { 2093 u_int32_t oh2c = *h2c; 2094 2095 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2096 oia.addr16[0], ia->addr16[0], 0), 2097 oia.addr16[1], ia->addr16[1], 0); 2098 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2099 oia.addr16[0], ia->addr16[0], 0), 2100 oia.addr16[1], ia->addr16[1], 0); 2101 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2102 break; 2103 } 2104#endif /* INET */ 2105#ifdef INET6 2106 case AF_INET6: 2107 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2108 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2109 pf_cksum_fixup(pf_cksum_fixup(*ic, 2110 oia.addr16[0], ia->addr16[0], u), 2111 oia.addr16[1], ia->addr16[1], u), 2112 oia.addr16[2], ia->addr16[2], u), 2113 oia.addr16[3], ia->addr16[3], u), 2114 oia.addr16[4], ia->addr16[4], u), 2115 oia.addr16[5], ia->addr16[5], u), 2116 oia.addr16[6], ia->addr16[6], u), 2117 oia.addr16[7], ia->addr16[7], u); 2118 break; 2119#endif /* INET6 */ 2120 } 2121 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2122 if (oa) { 2123 PF_ACPY(oa, na, af); 2124 switch (af) { 2125#ifdef INET 2126 case AF_INET: 2127 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2128 ooa.addr16[0], oa->addr16[0], 0), 2129 ooa.addr16[1], oa->addr16[1], 0); 2130 break; 2131#endif /* INET */ 2132#ifdef INET6 2133 case AF_INET6: 2134 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2135 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2136 pf_cksum_fixup(pf_cksum_fixup(*ic, 2137 ooa.addr16[0], oa->addr16[0], u), 2138 ooa.addr16[1], oa->addr16[1], u), 2139 ooa.addr16[2], oa->addr16[2], u), 2140 ooa.addr16[3], oa->addr16[3], u), 2141 ooa.addr16[4], oa->addr16[4], u), 2142 ooa.addr16[5], oa->addr16[5], u), 2143 ooa.addr16[6], oa->addr16[6], u), 2144 ooa.addr16[7], oa->addr16[7], u); 2145 break; 2146#endif /* INET6 */ 2147 } 2148 } 2149} 2150 2151 2152/* 2153 * Need to modulate the sequence numbers in the TCP SACK option 2154 * (credits to Krzysztof Pfaff for report and patch) 2155 */ 2156static int 2157pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2158 struct tcphdr *th, struct pf_state_peer *dst) 2159{ 2160 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2161 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2162 int copyback = 0, i, olen; 2163 struct sackblk sack; 2164 2165#define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2166 if (hlen < TCPOLEN_SACKLEN || 2167 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2168 return 0; 2169 2170 while (hlen >= TCPOLEN_SACKLEN) { 2171 olen = opt[1]; 2172 switch (*opt) { 2173 case TCPOPT_EOL: /* FALLTHROUGH */ 2174 case TCPOPT_NOP: 2175 opt++; 2176 hlen--; 2177 break; 2178 case TCPOPT_SACK: 2179 if (olen > hlen) 2180 olen = hlen; 2181 if (olen >= TCPOLEN_SACKLEN) { 2182 for (i = 2; i + TCPOLEN_SACK <= olen; 2183 i += TCPOLEN_SACK) { 2184 memcpy(&sack, &opt[i], sizeof(sack)); 2185 pf_change_a(&sack.start, &th->th_sum, 2186 htonl(ntohl(sack.start) - 2187 dst->seqdiff), 0); 2188 pf_change_a(&sack.end, &th->th_sum, 2189 htonl(ntohl(sack.end) - 2190 dst->seqdiff), 0); 2191 memcpy(&opt[i], &sack, sizeof(sack)); 2192 } 2193 copyback = 1; 2194 } 2195 /* FALLTHROUGH */ 2196 default: 2197 if (olen < 2) 2198 olen = 2; 2199 hlen -= olen; 2200 opt += olen; 2201 } 2202 } 2203 2204 if (copyback) 2205 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2206 return (copyback); 2207} 2208 2209static void 2210pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af, 2211 const struct pf_addr *saddr, const struct pf_addr *daddr, 2212 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2213 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2214 u_int16_t rtag, struct ifnet *ifp) 2215{ 2216 struct pf_send_entry *pfse; 2217 struct mbuf *m; 2218 int len, tlen; 2219#ifdef INET 2220 struct ip *h = NULL; 2221#endif /* INET */ 2222#ifdef INET6 2223 struct ip6_hdr *h6 = NULL; 2224#endif /* INET6 */ 2225 struct tcphdr *th; 2226 char *opt; 2227 struct pf_mtag *pf_mtag; 2228 2229 len = 0; 2230 th = NULL; 2231 2232 /* maximum segment size tcp option */ 2233 tlen = sizeof(struct tcphdr); 2234 if (mss) 2235 tlen += 4; 2236 2237 switch (af) { 2238#ifdef INET 2239 case AF_INET: 2240 len = sizeof(struct ip) + tlen; 2241 break; 2242#endif /* INET */ 2243#ifdef INET6 2244 case AF_INET6: 2245 len = sizeof(struct ip6_hdr) + tlen; 2246 break; 2247#endif /* INET6 */ 2248 default: 2249 panic("%s: unsupported af %d", __func__, af); 2250 } 2251 2252 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2253 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2254 if (pfse == NULL) 2255 return; 2256 m = m_gethdr(M_NOWAIT, MT_DATA); 2257 if (m == NULL) { 2258 free(pfse, M_PFTEMP); 2259 return; 2260 } 2261#ifdef MAC 2262 mac_netinet_firewall_send(m); 2263#endif 2264 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2265 free(pfse, M_PFTEMP); 2266 m_freem(m); 2267 return; 2268 } 2269 if (tag) 2270 m->m_flags |= M_SKIP_FIREWALL; 2271 pf_mtag->tag = rtag; 2272 2273 if (r != NULL && r->rtableid >= 0) 2274 M_SETFIB(m, r->rtableid); 2275 2276#ifdef ALTQ 2277 if (r != NULL && r->qid) { 2278 pf_mtag->qid = r->qid; 2279 2280 /* add hints for ecn */ 2281 pf_mtag->hdr = mtod(m, struct ip *); 2282 } 2283#endif /* ALTQ */ 2284 m->m_data += max_linkhdr; 2285 m->m_pkthdr.len = m->m_len = len; 2286 m->m_pkthdr.rcvif = NULL; 2287 bzero(m->m_data, len); 2288 switch (af) { 2289#ifdef INET 2290 case AF_INET: 2291 h = mtod(m, struct ip *); 2292 2293 /* IP header fields included in the TCP checksum */ 2294 h->ip_p = IPPROTO_TCP; 2295 h->ip_len = htons(tlen); 2296 h->ip_src.s_addr = saddr->v4.s_addr; 2297 h->ip_dst.s_addr = daddr->v4.s_addr; 2298 2299 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2300 break; 2301#endif /* INET */ 2302#ifdef INET6 2303 case AF_INET6: 2304 h6 = mtod(m, struct ip6_hdr *); 2305 2306 /* IP header fields included in the TCP checksum */ 2307 h6->ip6_nxt = IPPROTO_TCP; 2308 h6->ip6_plen = htons(tlen); 2309 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2310 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2311 2312 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2313 break; 2314#endif /* INET6 */ 2315 } 2316 2317 /* TCP header */ 2318 th->th_sport = sport; 2319 th->th_dport = dport; 2320 th->th_seq = htonl(seq); 2321 th->th_ack = htonl(ack); 2322 th->th_off = tlen >> 2; 2323 th->th_flags = flags; 2324 th->th_win = htons(win); 2325 2326 if (mss) { 2327 opt = (char *)(th + 1); 2328 opt[0] = TCPOPT_MAXSEG; 2329 opt[1] = 4; 2330 HTONS(mss); 2331 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2332 } 2333 2334 switch (af) { 2335#ifdef INET 2336 case AF_INET: 2337 /* TCP checksum */ 2338 th->th_sum = in_cksum(m, len); 2339 2340 /* Finish the IP header */ 2341 h->ip_v = 4; 2342 h->ip_hl = sizeof(*h) >> 2; 2343 h->ip_tos = IPTOS_LOWDELAY; 2344 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2345 h->ip_len = htons(len); 2346 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2347 h->ip_sum = 0; 2348 2349 pfse->pfse_type = PFSE_IP; 2350 break; 2351#endif /* INET */ 2352#ifdef INET6 2353 case AF_INET6: 2354 /* TCP checksum */ 2355 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2356 sizeof(struct ip6_hdr), tlen); 2357 2358 h6->ip6_vfc |= IPV6_VERSION; 2359 h6->ip6_hlim = IPV6_DEFHLIM; 2360 2361 pfse->pfse_type = PFSE_IP6; 2362 break; 2363#endif /* INET6 */ 2364 } 2365 pfse->pfse_m = m; 2366 pf_send(pfse); 2367} 2368 2369static void 2370pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2371 struct pf_rule *r) 2372{ 2373 struct pf_send_entry *pfse; 2374 struct mbuf *m0; 2375 struct pf_mtag *pf_mtag; 2376 2377 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2378 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2379 if (pfse == NULL) 2380 return; 2381 2382 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 2383 free(pfse, M_PFTEMP); 2384 return; 2385 } 2386 2387 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 2388 free(pfse, M_PFTEMP); 2389 return; 2390 } 2391 /* XXX: revisit */ 2392 m0->m_flags |= M_SKIP_FIREWALL; 2393 2394 if (r->rtableid >= 0) 2395 M_SETFIB(m0, r->rtableid); 2396 2397#ifdef ALTQ 2398 if (r->qid) { 2399 pf_mtag->qid = r->qid; 2400 /* add hints for ecn */ 2401 pf_mtag->hdr = mtod(m0, struct ip *); 2402 } 2403#endif /* ALTQ */ 2404 2405 switch (af) { 2406#ifdef INET 2407 case AF_INET: 2408 pfse->pfse_type = PFSE_ICMP; 2409 break; 2410#endif /* INET */ 2411#ifdef INET6 2412 case AF_INET6: 2413 pfse->pfse_type = PFSE_ICMP6; 2414 break; 2415#endif /* INET6 */ 2416 } 2417 pfse->pfse_m = m0; 2418 pfse->pfse_icmp_type = type; 2419 pfse->pfse_icmp_code = code; 2420 pf_send(pfse); 2421} 2422 2423/* 2424 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2425 * If n is 0, they match if they are equal. If n is != 0, they match if they 2426 * are different. 2427 */ 2428int 2429pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2430 struct pf_addr *b, sa_family_t af) 2431{ 2432 int match = 0; 2433 2434 switch (af) { 2435#ifdef INET 2436 case AF_INET: 2437 if ((a->addr32[0] & m->addr32[0]) == 2438 (b->addr32[0] & m->addr32[0])) 2439 match++; 2440 break; 2441#endif /* INET */ 2442#ifdef INET6 2443 case AF_INET6: 2444 if (((a->addr32[0] & m->addr32[0]) == 2445 (b->addr32[0] & m->addr32[0])) && 2446 ((a->addr32[1] & m->addr32[1]) == 2447 (b->addr32[1] & m->addr32[1])) && 2448 ((a->addr32[2] & m->addr32[2]) == 2449 (b->addr32[2] & m->addr32[2])) && 2450 ((a->addr32[3] & m->addr32[3]) == 2451 (b->addr32[3] & m->addr32[3]))) 2452 match++; 2453 break; 2454#endif /* INET6 */ 2455 } 2456 if (match) { 2457 if (n) 2458 return (0); 2459 else 2460 return (1); 2461 } else { 2462 if (n) 2463 return (1); 2464 else 2465 return (0); 2466 } 2467} 2468 2469/* 2470 * Return 1 if b <= a <= e, otherwise return 0. 2471 */ 2472int 2473pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 2474 struct pf_addr *a, sa_family_t af) 2475{ 2476 switch (af) { 2477#ifdef INET 2478 case AF_INET: 2479 if ((a->addr32[0] < b->addr32[0]) || 2480 (a->addr32[0] > e->addr32[0])) 2481 return (0); 2482 break; 2483#endif /* INET */ 2484#ifdef INET6 2485 case AF_INET6: { 2486 int i; 2487 2488 /* check a >= b */ 2489 for (i = 0; i < 4; ++i) 2490 if (a->addr32[i] > b->addr32[i]) 2491 break; 2492 else if (a->addr32[i] < b->addr32[i]) 2493 return (0); 2494 /* check a <= e */ 2495 for (i = 0; i < 4; ++i) 2496 if (a->addr32[i] < e->addr32[i]) 2497 break; 2498 else if (a->addr32[i] > e->addr32[i]) 2499 return (0); 2500 break; 2501 } 2502#endif /* INET6 */ 2503 } 2504 return (1); 2505} 2506 2507static int 2508pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 2509{ 2510 switch (op) { 2511 case PF_OP_IRG: 2512 return ((p > a1) && (p < a2)); 2513 case PF_OP_XRG: 2514 return ((p < a1) || (p > a2)); 2515 case PF_OP_RRG: 2516 return ((p >= a1) && (p <= a2)); 2517 case PF_OP_EQ: 2518 return (p == a1); 2519 case PF_OP_NE: 2520 return (p != a1); 2521 case PF_OP_LT: 2522 return (p < a1); 2523 case PF_OP_LE: 2524 return (p <= a1); 2525 case PF_OP_GT: 2526 return (p > a1); 2527 case PF_OP_GE: 2528 return (p >= a1); 2529 } 2530 return (0); /* never reached */ 2531} 2532 2533int 2534pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 2535{ 2536 NTOHS(a1); 2537 NTOHS(a2); 2538 NTOHS(p); 2539 return (pf_match(op, a1, a2, p)); 2540} 2541 2542static int 2543pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 2544{ 2545 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2546 return (0); 2547 return (pf_match(op, a1, a2, u)); 2548} 2549 2550static int 2551pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 2552{ 2553 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 2554 return (0); 2555 return (pf_match(op, a1, a2, g)); 2556} 2557 2558int 2559pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag) 2560{ 2561 if (*tag == -1) 2562 *tag = mtag; 2563 2564 return ((!r->match_tag_not && r->match_tag == *tag) || 2565 (r->match_tag_not && r->match_tag != *tag)); 2566} 2567 2568int 2569pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 2570{ 2571 2572 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 2573 2574 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 2575 return (ENOMEM); 2576 2577 pd->pf_mtag->tag = tag; 2578 2579 return (0); 2580} 2581 2582#define PF_ANCHOR_STACKSIZE 32 2583struct pf_anchor_stackframe { 2584 struct pf_ruleset *rs; 2585 struct pf_rule *r; /* XXX: + match bit */ 2586 struct pf_anchor *child; 2587}; 2588 2589/* 2590 * XXX: We rely on malloc(9) returning pointer aligned addresses. 2591 */ 2592#define PF_ANCHORSTACK_MATCH 0x00000001 2593#define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 2594 2595#define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 2596#define PF_ANCHOR_RULE(f) (struct pf_rule *) \ 2597 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 2598#define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 2599 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 2600} while (0) 2601 2602void 2603pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth, 2604 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2605 int *match) 2606{ 2607 struct pf_anchor_stackframe *f; 2608 2609 PF_RULES_RASSERT(); 2610 2611 if (match) 2612 *match = 0; 2613 if (*depth >= PF_ANCHOR_STACKSIZE) { 2614 printf("%s: anchor stack overflow on %s\n", 2615 __func__, (*r)->anchor->name); 2616 *r = TAILQ_NEXT(*r, entries); 2617 return; 2618 } else if (*depth == 0 && a != NULL) 2619 *a = *r; 2620 f = stack + (*depth)++; 2621 f->rs = *rs; 2622 f->r = *r; 2623 if ((*r)->anchor_wildcard) { 2624 struct pf_anchor_node *parent = &(*r)->anchor->children; 2625 2626 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) { 2627 *r = NULL; 2628 return; 2629 } 2630 *rs = &f->child->ruleset; 2631 } else { 2632 f->child = NULL; 2633 *rs = &(*r)->anchor->ruleset; 2634 } 2635 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2636} 2637 2638int 2639pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth, 2640 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, 2641 int *match) 2642{ 2643 struct pf_anchor_stackframe *f; 2644 struct pf_rule *fr; 2645 int quick = 0; 2646 2647 PF_RULES_RASSERT(); 2648 2649 do { 2650 if (*depth <= 0) 2651 break; 2652 f = stack + *depth - 1; 2653 fr = PF_ANCHOR_RULE(f); 2654 if (f->child != NULL) { 2655 struct pf_anchor_node *parent; 2656 2657 /* 2658 * This block traverses through 2659 * a wildcard anchor. 2660 */ 2661 parent = &fr->anchor->children; 2662 if (match != NULL && *match) { 2663 /* 2664 * If any of "*" matched, then 2665 * "foo/ *" matched, mark frame 2666 * appropriately. 2667 */ 2668 PF_ANCHOR_SET_MATCH(f); 2669 *match = 0; 2670 } 2671 f->child = RB_NEXT(pf_anchor_node, parent, f->child); 2672 if (f->child != NULL) { 2673 *rs = &f->child->ruleset; 2674 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 2675 if (*r == NULL) 2676 continue; 2677 else 2678 break; 2679 } 2680 } 2681 (*depth)--; 2682 if (*depth == 0 && a != NULL) 2683 *a = NULL; 2684 *rs = f->rs; 2685 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 2686 quick = fr->quick; 2687 *r = TAILQ_NEXT(fr, entries); 2688 } while (*r == NULL); 2689 2690 return (quick); 2691} 2692 2693#ifdef INET6 2694void 2695pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 2696 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 2697{ 2698 switch (af) { 2699#ifdef INET 2700 case AF_INET: 2701 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2702 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2703 break; 2704#endif /* INET */ 2705 case AF_INET6: 2706 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 2707 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 2708 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 2709 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 2710 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 2711 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 2712 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 2713 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 2714 break; 2715 } 2716} 2717 2718void 2719pf_addr_inc(struct pf_addr *addr, sa_family_t af) 2720{ 2721 switch (af) { 2722#ifdef INET 2723 case AF_INET: 2724 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 2725 break; 2726#endif /* INET */ 2727 case AF_INET6: 2728 if (addr->addr32[3] == 0xffffffff) { 2729 addr->addr32[3] = 0; 2730 if (addr->addr32[2] == 0xffffffff) { 2731 addr->addr32[2] = 0; 2732 if (addr->addr32[1] == 0xffffffff) { 2733 addr->addr32[1] = 0; 2734 addr->addr32[0] = 2735 htonl(ntohl(addr->addr32[0]) + 1); 2736 } else 2737 addr->addr32[1] = 2738 htonl(ntohl(addr->addr32[1]) + 1); 2739 } else 2740 addr->addr32[2] = 2741 htonl(ntohl(addr->addr32[2]) + 1); 2742 } else 2743 addr->addr32[3] = 2744 htonl(ntohl(addr->addr32[3]) + 1); 2745 break; 2746 } 2747} 2748#endif /* INET6 */ 2749 2750int 2751pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 2752{ 2753 struct pf_addr *saddr, *daddr; 2754 u_int16_t sport, dport; 2755 struct inpcbinfo *pi; 2756 struct inpcb *inp; 2757 2758 pd->lookup.uid = UID_MAX; 2759 pd->lookup.gid = GID_MAX; 2760 2761 switch (pd->proto) { 2762 case IPPROTO_TCP: 2763 if (pd->hdr.tcp == NULL) 2764 return (-1); 2765 sport = pd->hdr.tcp->th_sport; 2766 dport = pd->hdr.tcp->th_dport; 2767 pi = &V_tcbinfo; 2768 break; 2769 case IPPROTO_UDP: 2770 if (pd->hdr.udp == NULL) 2771 return (-1); 2772 sport = pd->hdr.udp->uh_sport; 2773 dport = pd->hdr.udp->uh_dport; 2774 pi = &V_udbinfo; 2775 break; 2776 default: 2777 return (-1); 2778 } 2779 if (direction == PF_IN) { 2780 saddr = pd->src; 2781 daddr = pd->dst; 2782 } else { 2783 u_int16_t p; 2784 2785 p = sport; 2786 sport = dport; 2787 dport = p; 2788 saddr = pd->dst; 2789 daddr = pd->src; 2790 } 2791 switch (pd->af) { 2792#ifdef INET 2793 case AF_INET: 2794 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 2795 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2796 if (inp == NULL) { 2797 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 2798 daddr->v4, dport, INPLOOKUP_WILDCARD | 2799 INPLOOKUP_RLOCKPCB, NULL, m); 2800 if (inp == NULL) 2801 return (-1); 2802 } 2803 break; 2804#endif /* INET */ 2805#ifdef INET6 2806 case AF_INET6: 2807 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 2808 dport, INPLOOKUP_RLOCKPCB, NULL, m); 2809 if (inp == NULL) { 2810 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 2811 &daddr->v6, dport, INPLOOKUP_WILDCARD | 2812 INPLOOKUP_RLOCKPCB, NULL, m); 2813 if (inp == NULL) 2814 return (-1); 2815 } 2816 break; 2817#endif /* INET6 */ 2818 2819 default: 2820 return (-1); 2821 } 2822 INP_RLOCK_ASSERT(inp); 2823 pd->lookup.uid = inp->inp_cred->cr_uid; 2824 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 2825 INP_RUNLOCK(inp); 2826 2827 return (1); 2828} 2829 2830static u_int8_t 2831pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2832{ 2833 int hlen; 2834 u_int8_t hdr[60]; 2835 u_int8_t *opt, optlen; 2836 u_int8_t wscale = 0; 2837 2838 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2839 if (hlen <= sizeof(struct tcphdr)) 2840 return (0); 2841 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2842 return (0); 2843 opt = hdr + sizeof(struct tcphdr); 2844 hlen -= sizeof(struct tcphdr); 2845 while (hlen >= 3) { 2846 switch (*opt) { 2847 case TCPOPT_EOL: 2848 case TCPOPT_NOP: 2849 ++opt; 2850 --hlen; 2851 break; 2852 case TCPOPT_WINDOW: 2853 wscale = opt[2]; 2854 if (wscale > TCP_MAX_WINSHIFT) 2855 wscale = TCP_MAX_WINSHIFT; 2856 wscale |= PF_WSCALE_FLAG; 2857 /* FALLTHROUGH */ 2858 default: 2859 optlen = opt[1]; 2860 if (optlen < 2) 2861 optlen = 2; 2862 hlen -= optlen; 2863 opt += optlen; 2864 break; 2865 } 2866 } 2867 return (wscale); 2868} 2869 2870static u_int16_t 2871pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 2872{ 2873 int hlen; 2874 u_int8_t hdr[60]; 2875 u_int8_t *opt, optlen; 2876 u_int16_t mss = V_tcp_mssdflt; 2877 2878 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 2879 if (hlen <= sizeof(struct tcphdr)) 2880 return (0); 2881 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 2882 return (0); 2883 opt = hdr + sizeof(struct tcphdr); 2884 hlen -= sizeof(struct tcphdr); 2885 while (hlen >= TCPOLEN_MAXSEG) { 2886 switch (*opt) { 2887 case TCPOPT_EOL: 2888 case TCPOPT_NOP: 2889 ++opt; 2890 --hlen; 2891 break; 2892 case TCPOPT_MAXSEG: 2893 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 2894 NTOHS(mss); 2895 /* FALLTHROUGH */ 2896 default: 2897 optlen = opt[1]; 2898 if (optlen < 2) 2899 optlen = 2; 2900 hlen -= optlen; 2901 opt += optlen; 2902 break; 2903 } 2904 } 2905 return (mss); 2906} 2907 2908static u_int16_t 2909pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 2910{ 2911#ifdef INET 2912 struct sockaddr_in *dst; 2913 struct route ro; 2914#endif /* INET */ 2915#ifdef INET6 2916 struct sockaddr_in6 *dst6; 2917 struct route_in6 ro6; 2918#endif /* INET6 */ 2919 struct rtentry *rt = NULL; 2920 int hlen = 0; 2921 u_int16_t mss = V_tcp_mssdflt; 2922 2923 switch (af) { 2924#ifdef INET 2925 case AF_INET: 2926 hlen = sizeof(struct ip); 2927 bzero(&ro, sizeof(ro)); 2928 dst = (struct sockaddr_in *)&ro.ro_dst; 2929 dst->sin_family = AF_INET; 2930 dst->sin_len = sizeof(*dst); 2931 dst->sin_addr = addr->v4; 2932 in_rtalloc_ign(&ro, 0, rtableid); 2933 rt = ro.ro_rt; 2934 break; 2935#endif /* INET */ 2936#ifdef INET6 2937 case AF_INET6: 2938 hlen = sizeof(struct ip6_hdr); 2939 bzero(&ro6, sizeof(ro6)); 2940 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst; 2941 dst6->sin6_family = AF_INET6; 2942 dst6->sin6_len = sizeof(*dst6); 2943 dst6->sin6_addr = addr->v6; 2944 in6_rtalloc_ign(&ro6, 0, rtableid); 2945 rt = ro6.ro_rt; 2946 break; 2947#endif /* INET6 */ 2948 } 2949 2950 if (rt && rt->rt_ifp) { 2951 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr); 2952 mss = max(V_tcp_mssdflt, mss); 2953 RTFREE(rt); 2954 } 2955 mss = min(mss, offer); 2956 mss = max(mss, 64); /* sanity - at least max opt space */ 2957 return (mss); 2958} 2959 2960static void 2961pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr) 2962{ 2963 struct pf_rule *r = s->rule.ptr; 2964 struct pf_src_node *sn = NULL; 2965 2966 s->rt_kif = NULL; 2967 if (!r->rt || r->rt == PF_FASTROUTE) 2968 return; 2969 switch (s->key[PF_SK_WIRE]->af) { 2970#ifdef INET 2971 case AF_INET: 2972 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, &sn); 2973 s->rt_kif = r->rpool.cur->kif; 2974 break; 2975#endif /* INET */ 2976#ifdef INET6 2977 case AF_INET6: 2978 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, &sn); 2979 s->rt_kif = r->rpool.cur->kif; 2980 break; 2981#endif /* INET6 */ 2982 } 2983} 2984 2985static u_int32_t 2986pf_tcp_iss(struct pf_pdesc *pd) 2987{ 2988 MD5_CTX ctx; 2989 u_int32_t digest[4]; 2990 2991 if (V_pf_tcp_secret_init == 0) { 2992 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 2993 MD5Init(&V_pf_tcp_secret_ctx); 2994 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 2995 sizeof(V_pf_tcp_secret)); 2996 V_pf_tcp_secret_init = 1; 2997 } 2998 2999 ctx = V_pf_tcp_secret_ctx; 3000 3001 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); 3002 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); 3003 if (pd->af == AF_INET6) { 3004 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3005 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3006 } else { 3007 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3008 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3009 } 3010 MD5Final((u_char *)digest, &ctx); 3011 V_pf_tcp_iss_off += 4096; 3012#define ISN_RANDOM_INCREMENT (4096 - 1) 3013 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3014 V_pf_tcp_iss_off); 3015#undef ISN_RANDOM_INCREMENT 3016} 3017 3018static int 3019pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, 3020 struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 3021 struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp) 3022{ 3023 struct pf_rule *nr = NULL; 3024 struct pf_addr * const saddr = pd->src; 3025 struct pf_addr * const daddr = pd->dst; 3026 sa_family_t af = pd->af; 3027 struct pf_rule *r, *a = NULL; 3028 struct pf_ruleset *ruleset = NULL; 3029 struct pf_src_node *nsn = NULL; 3030 struct tcphdr *th = pd->hdr.tcp; 3031 struct pf_state_key *sk = NULL, *nk = NULL; 3032 u_short reason; 3033 int rewrite = 0, hdrlen = 0; 3034 int tag = -1, rtableid = -1; 3035 int asd = 0; 3036 int match = 0; 3037 int state_icmp = 0; 3038 u_int16_t sport = 0, dport = 0; 3039 u_int16_t bproto_sum = 0, bip_sum = 0; 3040 u_int8_t icmptype = 0, icmpcode = 0; 3041 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3042 3043 PF_RULES_RASSERT(); 3044 3045 if (inp != NULL) { 3046 INP_LOCK_ASSERT(inp); 3047 pd->lookup.uid = inp->inp_cred->cr_uid; 3048 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3049 pd->lookup.done = 1; 3050 } 3051 3052 switch (pd->proto) { 3053 case IPPROTO_TCP: 3054 sport = th->th_sport; 3055 dport = th->th_dport; 3056 hdrlen = sizeof(*th); 3057 break; 3058 case IPPROTO_UDP: 3059 sport = pd->hdr.udp->uh_sport; 3060 dport = pd->hdr.udp->uh_dport; 3061 hdrlen = sizeof(*pd->hdr.udp); 3062 break; 3063#ifdef INET 3064 case IPPROTO_ICMP: 3065 if (pd->af != AF_INET) 3066 break; 3067 sport = dport = pd->hdr.icmp->icmp_id; 3068 hdrlen = sizeof(*pd->hdr.icmp); 3069 icmptype = pd->hdr.icmp->icmp_type; 3070 icmpcode = pd->hdr.icmp->icmp_code; 3071 3072 if (icmptype == ICMP_UNREACH || 3073 icmptype == ICMP_SOURCEQUENCH || 3074 icmptype == ICMP_REDIRECT || 3075 icmptype == ICMP_TIMXCEED || 3076 icmptype == ICMP_PARAMPROB) 3077 state_icmp++; 3078 break; 3079#endif /* INET */ 3080#ifdef INET6 3081 case IPPROTO_ICMPV6: 3082 if (af != AF_INET6) 3083 break; 3084 sport = dport = pd->hdr.icmp6->icmp6_id; 3085 hdrlen = sizeof(*pd->hdr.icmp6); 3086 icmptype = pd->hdr.icmp6->icmp6_type; 3087 icmpcode = pd->hdr.icmp6->icmp6_code; 3088 3089 if (icmptype == ICMP6_DST_UNREACH || 3090 icmptype == ICMP6_PACKET_TOO_BIG || 3091 icmptype == ICMP6_TIME_EXCEEDED || 3092 icmptype == ICMP6_PARAM_PROB) 3093 state_icmp++; 3094 break; 3095#endif /* INET6 */ 3096 default: 3097 sport = dport = hdrlen = 0; 3098 break; 3099 } 3100 3101 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3102 3103 /* check packet for BINAT/NAT/RDR */ 3104 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 3105 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 3106 KASSERT(sk != NULL, ("%s: null sk", __func__)); 3107 KASSERT(nk != NULL, ("%s: null nk", __func__)); 3108 3109 if (pd->ip_sum) 3110 bip_sum = *pd->ip_sum; 3111 3112 switch (pd->proto) { 3113 case IPPROTO_TCP: 3114 bproto_sum = th->th_sum; 3115 pd->proto_sum = &th->th_sum; 3116 3117 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3118 nk->port[pd->sidx] != sport) { 3119 pf_change_ap(saddr, &th->th_sport, pd->ip_sum, 3120 &th->th_sum, &nk->addr[pd->sidx], 3121 nk->port[pd->sidx], 0, af); 3122 pd->sport = &th->th_sport; 3123 sport = th->th_sport; 3124 } 3125 3126 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3127 nk->port[pd->didx] != dport) { 3128 pf_change_ap(daddr, &th->th_dport, pd->ip_sum, 3129 &th->th_sum, &nk->addr[pd->didx], 3130 nk->port[pd->didx], 0, af); 3131 dport = th->th_dport; 3132 pd->dport = &th->th_dport; 3133 } 3134 rewrite++; 3135 break; 3136 case IPPROTO_UDP: 3137 bproto_sum = pd->hdr.udp->uh_sum; 3138 pd->proto_sum = &pd->hdr.udp->uh_sum; 3139 3140 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3141 nk->port[pd->sidx] != sport) { 3142 pf_change_ap(saddr, &pd->hdr.udp->uh_sport, 3143 pd->ip_sum, &pd->hdr.udp->uh_sum, 3144 &nk->addr[pd->sidx], 3145 nk->port[pd->sidx], 1, af); 3146 sport = pd->hdr.udp->uh_sport; 3147 pd->sport = &pd->hdr.udp->uh_sport; 3148 } 3149 3150 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3151 nk->port[pd->didx] != dport) { 3152 pf_change_ap(daddr, &pd->hdr.udp->uh_dport, 3153 pd->ip_sum, &pd->hdr.udp->uh_sum, 3154 &nk->addr[pd->didx], 3155 nk->port[pd->didx], 1, af); 3156 dport = pd->hdr.udp->uh_dport; 3157 pd->dport = &pd->hdr.udp->uh_dport; 3158 } 3159 rewrite++; 3160 break; 3161#ifdef INET 3162 case IPPROTO_ICMP: 3163 nk->port[0] = nk->port[1]; 3164 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3165 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3166 nk->addr[pd->sidx].v4.s_addr, 0); 3167 3168 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3169 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3170 nk->addr[pd->didx].v4.s_addr, 0); 3171 3172 if (nk->port[1] != pd->hdr.icmp->icmp_id) { 3173 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( 3174 pd->hdr.icmp->icmp_cksum, sport, 3175 nk->port[1], 0); 3176 pd->hdr.icmp->icmp_id = nk->port[1]; 3177 pd->sport = &pd->hdr.icmp->icmp_id; 3178 } 3179 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 3180 break; 3181#endif /* INET */ 3182#ifdef INET6 3183 case IPPROTO_ICMPV6: 3184 nk->port[0] = nk->port[1]; 3185 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3186 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, 3187 &nk->addr[pd->sidx], 0); 3188 3189 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3190 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, 3191 &nk->addr[pd->didx], 0); 3192 rewrite++; 3193 break; 3194#endif /* INET */ 3195 default: 3196 switch (af) { 3197#ifdef INET 3198 case AF_INET: 3199 if (PF_ANEQ(saddr, 3200 &nk->addr[pd->sidx], AF_INET)) 3201 pf_change_a(&saddr->v4.s_addr, 3202 pd->ip_sum, 3203 nk->addr[pd->sidx].v4.s_addr, 0); 3204 3205 if (PF_ANEQ(daddr, 3206 &nk->addr[pd->didx], AF_INET)) 3207 pf_change_a(&daddr->v4.s_addr, 3208 pd->ip_sum, 3209 nk->addr[pd->didx].v4.s_addr, 0); 3210 break; 3211#endif /* INET */ 3212#ifdef INET6 3213 case AF_INET6: 3214 if (PF_ANEQ(saddr, 3215 &nk->addr[pd->sidx], AF_INET6)) 3216 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3217 3218 if (PF_ANEQ(daddr, 3219 &nk->addr[pd->didx], AF_INET6)) 3220 PF_ACPY(saddr, &nk->addr[pd->didx], af); 3221 break; 3222#endif /* INET */ 3223 } 3224 break; 3225 } 3226 if (nr->natpass) 3227 r = NULL; 3228 pd->nat_rule = nr; 3229 } 3230 3231 while (r != NULL) { 3232 r->evaluations++; 3233 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3234 r = r->skip[PF_SKIP_IFP].ptr; 3235 else if (r->direction && r->direction != direction) 3236 r = r->skip[PF_SKIP_DIR].ptr; 3237 else if (r->af && r->af != af) 3238 r = r->skip[PF_SKIP_AF].ptr; 3239 else if (r->proto && r->proto != pd->proto) 3240 r = r->skip[PF_SKIP_PROTO].ptr; 3241 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3242 r->src.neg, kif, M_GETFIB(m))) 3243 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3244 /* tcp/udp only. port_op always 0 in other cases */ 3245 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3246 r->src.port[0], r->src.port[1], sport)) 3247 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3248 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3249 r->dst.neg, NULL, M_GETFIB(m))) 3250 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3251 /* tcp/udp only. port_op always 0 in other cases */ 3252 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3253 r->dst.port[0], r->dst.port[1], dport)) 3254 r = r->skip[PF_SKIP_DST_PORT].ptr; 3255 /* icmp only. type always 0 in other cases */ 3256 else if (r->type && r->type != icmptype + 1) 3257 r = TAILQ_NEXT(r, entries); 3258 /* icmp only. type always 0 in other cases */ 3259 else if (r->code && r->code != icmpcode + 1) 3260 r = TAILQ_NEXT(r, entries); 3261 else if (r->tos && !(r->tos == pd->tos)) 3262 r = TAILQ_NEXT(r, entries); 3263 else if (r->rule_flag & PFRULE_FRAGMENT) 3264 r = TAILQ_NEXT(r, entries); 3265 else if (pd->proto == IPPROTO_TCP && 3266 (r->flagset & th->th_flags) != r->flags) 3267 r = TAILQ_NEXT(r, entries); 3268 /* tcp/udp only. uid.op always 0 in other cases */ 3269 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3270 pf_socket_lookup(direction, pd, m), 1)) && 3271 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3272 pd->lookup.uid)) 3273 r = TAILQ_NEXT(r, entries); 3274 /* tcp/udp only. gid.op always 0 in other cases */ 3275 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3276 pf_socket_lookup(direction, pd, m), 1)) && 3277 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3278 pd->lookup.gid)) 3279 r = TAILQ_NEXT(r, entries); 3280 else if (r->prob && 3281 r->prob <= arc4random()) 3282 r = TAILQ_NEXT(r, entries); 3283 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3284 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3285 r = TAILQ_NEXT(r, entries); 3286 else if (r->os_fingerprint != PF_OSFP_ANY && 3287 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3288 pf_osfp_fingerprint(pd, m, off, th), 3289 r->os_fingerprint))) 3290 r = TAILQ_NEXT(r, entries); 3291 else { 3292 if (r->tag) 3293 tag = r->tag; 3294 if (r->rtableid >= 0) 3295 rtableid = r->rtableid; 3296 if (r->anchor == NULL) { 3297 match = 1; 3298 *rm = r; 3299 *am = a; 3300 *rsm = ruleset; 3301 if ((*rm)->quick) 3302 break; 3303 r = TAILQ_NEXT(r, entries); 3304 } else 3305 pf_step_into_anchor(anchor_stack, &asd, 3306 &ruleset, PF_RULESET_FILTER, &r, &a, 3307 &match); 3308 } 3309 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3310 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3311 break; 3312 } 3313 r = *rm; 3314 a = *am; 3315 ruleset = *rsm; 3316 3317 REASON_SET(&reason, PFRES_MATCH); 3318 3319 if (r->log || (nr != NULL && nr->log)) { 3320 if (rewrite) 3321 m_copyback(m, off, hdrlen, pd->hdr.any); 3322 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, 3323 ruleset, pd, 1); 3324 } 3325 3326 if ((r->action == PF_DROP) && 3327 ((r->rule_flag & PFRULE_RETURNRST) || 3328 (r->rule_flag & PFRULE_RETURNICMP) || 3329 (r->rule_flag & PFRULE_RETURN))) { 3330 /* undo NAT changes, if they have taken place */ 3331 if (nr != NULL) { 3332 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3333 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3334 if (pd->sport) 3335 *pd->sport = sk->port[pd->sidx]; 3336 if (pd->dport) 3337 *pd->dport = sk->port[pd->didx]; 3338 if (pd->proto_sum) 3339 *pd->proto_sum = bproto_sum; 3340 if (pd->ip_sum) 3341 *pd->ip_sum = bip_sum; 3342 m_copyback(m, off, hdrlen, pd->hdr.any); 3343 } 3344 if (pd->proto == IPPROTO_TCP && 3345 ((r->rule_flag & PFRULE_RETURNRST) || 3346 (r->rule_flag & PFRULE_RETURN)) && 3347 !(th->th_flags & TH_RST)) { 3348 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3349 int len = 0; 3350#ifdef INET 3351 struct ip *h4; 3352#endif 3353#ifdef INET6 3354 struct ip6_hdr *h6; 3355#endif 3356 3357 switch (af) { 3358#ifdef INET 3359 case AF_INET: 3360 h4 = mtod(m, struct ip *); 3361 len = ntohs(h4->ip_len) - off; 3362 break; 3363#endif 3364#ifdef INET6 3365 case AF_INET6: 3366 h6 = mtod(m, struct ip6_hdr *); 3367 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3368 break; 3369#endif 3370 } 3371 3372 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3373 REASON_SET(&reason, PFRES_PROTCKSUM); 3374 else { 3375 if (th->th_flags & TH_SYN) 3376 ack++; 3377 if (th->th_flags & TH_FIN) 3378 ack++; 3379 pf_send_tcp(m, r, af, pd->dst, 3380 pd->src, th->th_dport, th->th_sport, 3381 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3382 r->return_ttl, 1, 0, kif->pfik_ifp); 3383 } 3384 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3385 r->return_icmp) 3386 pf_send_icmp(m, r->return_icmp >> 8, 3387 r->return_icmp & 255, af, r); 3388 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3389 r->return_icmp6) 3390 pf_send_icmp(m, r->return_icmp6 >> 8, 3391 r->return_icmp6 & 255, af, r); 3392 } 3393 3394 if (r->action == PF_DROP) 3395 goto cleanup; 3396 3397 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3398 REASON_SET(&reason, PFRES_MEMORY); 3399 goto cleanup; 3400 } 3401 if (rtableid >= 0) 3402 M_SETFIB(m, rtableid); 3403 3404 if (!state_icmp && (r->keep_state || nr != NULL || 3405 (pd->flags & PFDESC_TCP_NORM))) { 3406 int action; 3407 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 3408 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 3409 hdrlen); 3410 if (action != PF_PASS) 3411 return (action); 3412 } else { 3413 if (sk != NULL) 3414 uma_zfree(V_pf_state_key_z, sk); 3415 if (nk != NULL) 3416 uma_zfree(V_pf_state_key_z, nk); 3417 } 3418 3419 /* copy back packet headers if we performed NAT operations */ 3420 if (rewrite) 3421 m_copyback(m, off, hdrlen, pd->hdr.any); 3422 3423 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 3424 direction == PF_OUT && 3425 pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m)) 3426 /* 3427 * We want the state created, but we dont 3428 * want to send this in case a partner 3429 * firewall has to know about it to allow 3430 * replies through it. 3431 */ 3432 return (PF_DEFER); 3433 3434 return (PF_PASS); 3435 3436cleanup: 3437 if (sk != NULL) 3438 uma_zfree(V_pf_state_key_z, sk); 3439 if (nk != NULL) 3440 uma_zfree(V_pf_state_key_z, nk); 3441 return (PF_DROP); 3442} 3443 3444static int 3445pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, 3446 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk, 3447 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 3448 u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm, 3449 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 3450{ 3451 struct pf_state *s = NULL; 3452 struct pf_src_node *sn = NULL; 3453 struct tcphdr *th = pd->hdr.tcp; 3454 u_int16_t mss = V_tcp_mssdflt; 3455 u_short reason; 3456 3457 /* check maximums */ 3458 if (r->max_states && 3459 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 3460 V_pf_status.lcounters[LCNT_STATES]++; 3461 REASON_SET(&reason, PFRES_MAXSTATES); 3462 return (PF_DROP); 3463 } 3464 /* src node for filter rule */ 3465 if ((r->rule_flag & PFRULE_SRCTRACK || 3466 r->rpool.opts & PF_POOL_STICKYADDR) && 3467 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 3468 REASON_SET(&reason, PFRES_SRCLIMIT); 3469 goto csfailed; 3470 } 3471 /* src node for translation rule */ 3472 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 3473 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 3474 REASON_SET(&reason, PFRES_SRCLIMIT); 3475 goto csfailed; 3476 } 3477 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); 3478 if (s == NULL) { 3479 REASON_SET(&reason, PFRES_MEMORY); 3480 goto csfailed; 3481 } 3482 s->rule.ptr = r; 3483 s->nat_rule.ptr = nr; 3484 s->anchor.ptr = a; 3485 STATE_INC_COUNTERS(s); 3486 if (r->allow_opts) 3487 s->state_flags |= PFSTATE_ALLOWOPTS; 3488 if (r->rule_flag & PFRULE_STATESLOPPY) 3489 s->state_flags |= PFSTATE_SLOPPY; 3490 s->log = r->log & PF_LOG_ALL; 3491 s->sync_state = PFSYNC_S_NONE; 3492 if (nr != NULL) 3493 s->log |= nr->log & PF_LOG_ALL; 3494 switch (pd->proto) { 3495 case IPPROTO_TCP: 3496 s->src.seqlo = ntohl(th->th_seq); 3497 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 3498 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 3499 r->keep_state == PF_STATE_MODULATE) { 3500 /* Generate sequence number modulator */ 3501 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 3502 0) 3503 s->src.seqdiff = 1; 3504 pf_change_a(&th->th_seq, &th->th_sum, 3505 htonl(s->src.seqlo + s->src.seqdiff), 0); 3506 *rewrite = 1; 3507 } else 3508 s->src.seqdiff = 0; 3509 if (th->th_flags & TH_SYN) { 3510 s->src.seqhi++; 3511 s->src.wscale = pf_get_wscale(m, off, 3512 th->th_off, pd->af); 3513 } 3514 s->src.max_win = MAX(ntohs(th->th_win), 1); 3515 if (s->src.wscale & PF_WSCALE_MASK) { 3516 /* Remove scale factor from initial window */ 3517 int win = s->src.max_win; 3518 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 3519 s->src.max_win = (win - 1) >> 3520 (s->src.wscale & PF_WSCALE_MASK); 3521 } 3522 if (th->th_flags & TH_FIN) 3523 s->src.seqhi++; 3524 s->dst.seqhi = 1; 3525 s->dst.max_win = 1; 3526 s->src.state = TCPS_SYN_SENT; 3527 s->dst.state = TCPS_CLOSED; 3528 s->timeout = PFTM_TCP_FIRST_PACKET; 3529 break; 3530 case IPPROTO_UDP: 3531 s->src.state = PFUDPS_SINGLE; 3532 s->dst.state = PFUDPS_NO_TRAFFIC; 3533 s->timeout = PFTM_UDP_FIRST_PACKET; 3534 break; 3535 case IPPROTO_ICMP: 3536#ifdef INET6 3537 case IPPROTO_ICMPV6: 3538#endif 3539 s->timeout = PFTM_ICMP_FIRST_PACKET; 3540 break; 3541 default: 3542 s->src.state = PFOTHERS_SINGLE; 3543 s->dst.state = PFOTHERS_NO_TRAFFIC; 3544 s->timeout = PFTM_OTHER_FIRST_PACKET; 3545 } 3546 3547 s->creation = time_uptime; 3548 s->expire = time_uptime; 3549 3550 if (sn != NULL) { 3551 s->src_node = sn; 3552 s->src_node->states++; 3553 } 3554 if (nsn != NULL) { 3555 /* XXX We only modify one side for now. */ 3556 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 3557 s->nat_src_node = nsn; 3558 s->nat_src_node->states++; 3559 } 3560 if (pd->proto == IPPROTO_TCP) { 3561 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 3562 off, pd, th, &s->src, &s->dst)) { 3563 REASON_SET(&reason, PFRES_MEMORY); 3564 pf_src_tree_remove_state(s); 3565 STATE_DEC_COUNTERS(s); 3566 uma_zfree(V_pf_state_z, s); 3567 return (PF_DROP); 3568 } 3569 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 3570 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 3571 &s->src, &s->dst, rewrite)) { 3572 /* This really shouldn't happen!!! */ 3573 DPFPRINTF(PF_DEBUG_URGENT, 3574 ("pf_normalize_tcp_stateful failed on first pkt")); 3575 pf_normalize_tcp_cleanup(s); 3576 pf_src_tree_remove_state(s); 3577 STATE_DEC_COUNTERS(s); 3578 uma_zfree(V_pf_state_z, s); 3579 return (PF_DROP); 3580 } 3581 } 3582 s->direction = pd->dir; 3583 3584 /* 3585 * sk/nk could already been setup by pf_get_translation(). 3586 */ 3587 if (nr == NULL) { 3588 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 3589 __func__, nr, sk, nk)); 3590 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 3591 if (sk == NULL) 3592 goto csfailed; 3593 nk = sk; 3594 } else 3595 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 3596 __func__, nr, sk, nk)); 3597 3598 /* Swap sk/nk for PF_OUT. */ 3599 if (pf_state_insert(BOUND_IFACE(r, kif), 3600 (pd->dir == PF_IN) ? sk : nk, 3601 (pd->dir == PF_IN) ? nk : sk, s)) { 3602 if (pd->proto == IPPROTO_TCP) 3603 pf_normalize_tcp_cleanup(s); 3604 REASON_SET(&reason, PFRES_STATEINS); 3605 pf_src_tree_remove_state(s); 3606 STATE_DEC_COUNTERS(s); 3607 uma_zfree(V_pf_state_z, s); 3608 return (PF_DROP); 3609 } else 3610 *sm = s; 3611 3612 pf_set_rt_ifp(s, pd->src); /* needs s->state_key set */ 3613 if (tag > 0) 3614 s->tag = tag; 3615 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 3616 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 3617 s->src.state = PF_TCPS_PROXY_SRC; 3618 /* undo NAT changes, if they have taken place */ 3619 if (nr != NULL) { 3620 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 3621 if (pd->dir == PF_OUT) 3622 skt = s->key[PF_SK_STACK]; 3623 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 3624 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 3625 if (pd->sport) 3626 *pd->sport = skt->port[pd->sidx]; 3627 if (pd->dport) 3628 *pd->dport = skt->port[pd->didx]; 3629 if (pd->proto_sum) 3630 *pd->proto_sum = bproto_sum; 3631 if (pd->ip_sum) 3632 *pd->ip_sum = bip_sum; 3633 m_copyback(m, off, hdrlen, pd->hdr.any); 3634 } 3635 s->src.seqhi = htonl(arc4random()); 3636 /* Find mss option */ 3637 int rtid = M_GETFIB(m); 3638 mss = pf_get_mss(m, off, th->th_off, pd->af); 3639 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 3640 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 3641 s->src.mss = mss; 3642 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, 3643 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 3644 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); 3645 REASON_SET(&reason, PFRES_SYNPROXY); 3646 return (PF_SYNPROXY_DROP); 3647 } 3648 3649 return (PF_PASS); 3650 3651csfailed: 3652 if (sk != NULL) 3653 uma_zfree(V_pf_state_key_z, sk); 3654 if (nk != NULL) 3655 uma_zfree(V_pf_state_key_z, nk); 3656 3657 if (sn != NULL && sn->states == 0 && sn->expire == 0) { 3658 pf_unlink_src_node(sn); 3659 pf_free_src_node(sn); 3660 } 3661 3662 if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) { 3663 pf_unlink_src_node(nsn); 3664 pf_free_src_node(nsn); 3665 } 3666 3667 return (PF_DROP); 3668} 3669 3670static int 3671pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, 3672 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, 3673 struct pf_ruleset **rsm) 3674{ 3675 struct pf_rule *r, *a = NULL; 3676 struct pf_ruleset *ruleset = NULL; 3677 sa_family_t af = pd->af; 3678 u_short reason; 3679 int tag = -1; 3680 int asd = 0; 3681 int match = 0; 3682 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3683 3684 PF_RULES_RASSERT(); 3685 3686 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3687 while (r != NULL) { 3688 r->evaluations++; 3689 if (pfi_kif_match(r->kif, kif) == r->ifnot) 3690 r = r->skip[PF_SKIP_IFP].ptr; 3691 else if (r->direction && r->direction != direction) 3692 r = r->skip[PF_SKIP_DIR].ptr; 3693 else if (r->af && r->af != af) 3694 r = r->skip[PF_SKIP_AF].ptr; 3695 else if (r->proto && r->proto != pd->proto) 3696 r = r->skip[PF_SKIP_PROTO].ptr; 3697 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 3698 r->src.neg, kif, M_GETFIB(m))) 3699 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3700 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 3701 r->dst.neg, NULL, M_GETFIB(m))) 3702 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3703 else if (r->tos && !(r->tos == pd->tos)) 3704 r = TAILQ_NEXT(r, entries); 3705 else if (r->os_fingerprint != PF_OSFP_ANY) 3706 r = TAILQ_NEXT(r, entries); 3707 else if (pd->proto == IPPROTO_UDP && 3708 (r->src.port_op || r->dst.port_op)) 3709 r = TAILQ_NEXT(r, entries); 3710 else if (pd->proto == IPPROTO_TCP && 3711 (r->src.port_op || r->dst.port_op || r->flagset)) 3712 r = TAILQ_NEXT(r, entries); 3713 else if ((pd->proto == IPPROTO_ICMP || 3714 pd->proto == IPPROTO_ICMPV6) && 3715 (r->type || r->code)) 3716 r = TAILQ_NEXT(r, entries); 3717 else if (r->prob && r->prob <= 3718 (arc4random() % (UINT_MAX - 1) + 1)) 3719 r = TAILQ_NEXT(r, entries); 3720 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3721 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3722 r = TAILQ_NEXT(r, entries); 3723 else { 3724 if (r->anchor == NULL) { 3725 match = 1; 3726 *rm = r; 3727 *am = a; 3728 *rsm = ruleset; 3729 if ((*rm)->quick) 3730 break; 3731 r = TAILQ_NEXT(r, entries); 3732 } else 3733 pf_step_into_anchor(anchor_stack, &asd, 3734 &ruleset, PF_RULESET_FILTER, &r, &a, 3735 &match); 3736 } 3737 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3738 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3739 break; 3740 } 3741 r = *rm; 3742 a = *am; 3743 ruleset = *rsm; 3744 3745 REASON_SET(&reason, PFRES_MATCH); 3746 3747 if (r->log) 3748 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 3749 1); 3750 3751 if (r->action != PF_PASS) 3752 return (PF_DROP); 3753 3754 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3755 REASON_SET(&reason, PFRES_MEMORY); 3756 return (PF_DROP); 3757 } 3758 3759 return (PF_PASS); 3760} 3761 3762static int 3763pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, 3764 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, 3765 struct pf_pdesc *pd, u_short *reason, int *copyback) 3766{ 3767 struct tcphdr *th = pd->hdr.tcp; 3768 u_int16_t win = ntohs(th->th_win); 3769 u_int32_t ack, end, seq, orig_seq; 3770 u_int8_t sws, dws; 3771 int ackskew; 3772 3773 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 3774 sws = src->wscale & PF_WSCALE_MASK; 3775 dws = dst->wscale & PF_WSCALE_MASK; 3776 } else 3777 sws = dws = 0; 3778 3779 /* 3780 * Sequence tracking algorithm from Guido van Rooij's paper: 3781 * http://www.madison-gurkha.com/publications/tcp_filtering/ 3782 * tcp_filtering.ps 3783 */ 3784 3785 orig_seq = seq = ntohl(th->th_seq); 3786 if (src->seqlo == 0) { 3787 /* First packet from this end. Set its state */ 3788 3789 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 3790 src->scrub == NULL) { 3791 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 3792 REASON_SET(reason, PFRES_MEMORY); 3793 return (PF_DROP); 3794 } 3795 } 3796 3797 /* Deferred generation of sequence number modulator */ 3798 if (dst->seqdiff && !src->seqdiff) { 3799 /* use random iss for the TCP server */ 3800 while ((src->seqdiff = arc4random() - seq) == 0) 3801 ; 3802 ack = ntohl(th->th_ack) - dst->seqdiff; 3803 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 3804 src->seqdiff), 0); 3805 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 3806 *copyback = 1; 3807 } else { 3808 ack = ntohl(th->th_ack); 3809 } 3810 3811 end = seq + pd->p_len; 3812 if (th->th_flags & TH_SYN) { 3813 end++; 3814 if (dst->wscale & PF_WSCALE_FLAG) { 3815 src->wscale = pf_get_wscale(m, off, th->th_off, 3816 pd->af); 3817 if (src->wscale & PF_WSCALE_FLAG) { 3818 /* Remove scale factor from initial 3819 * window */ 3820 sws = src->wscale & PF_WSCALE_MASK; 3821 win = ((u_int32_t)win + (1 << sws) - 1) 3822 >> sws; 3823 dws = dst->wscale & PF_WSCALE_MASK; 3824 } else { 3825 /* fixup other window */ 3826 dst->max_win <<= dst->wscale & 3827 PF_WSCALE_MASK; 3828 /* in case of a retrans SYN|ACK */ 3829 dst->wscale = 0; 3830 } 3831 } 3832 } 3833 if (th->th_flags & TH_FIN) 3834 end++; 3835 3836 src->seqlo = seq; 3837 if (src->state < TCPS_SYN_SENT) 3838 src->state = TCPS_SYN_SENT; 3839 3840 /* 3841 * May need to slide the window (seqhi may have been set by 3842 * the crappy stack check or if we picked up the connection 3843 * after establishment) 3844 */ 3845 if (src->seqhi == 1 || 3846 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 3847 src->seqhi = end + MAX(1, dst->max_win << dws); 3848 if (win > src->max_win) 3849 src->max_win = win; 3850 3851 } else { 3852 ack = ntohl(th->th_ack) - dst->seqdiff; 3853 if (src->seqdiff) { 3854 /* Modulate sequence numbers */ 3855 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq + 3856 src->seqdiff), 0); 3857 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0); 3858 *copyback = 1; 3859 } 3860 end = seq + pd->p_len; 3861 if (th->th_flags & TH_SYN) 3862 end++; 3863 if (th->th_flags & TH_FIN) 3864 end++; 3865 } 3866 3867 if ((th->th_flags & TH_ACK) == 0) { 3868 /* Let it pass through the ack skew check */ 3869 ack = dst->seqlo; 3870 } else if ((ack == 0 && 3871 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 3872 /* broken tcp stacks do not set ack */ 3873 (dst->state < TCPS_SYN_SENT)) { 3874 /* 3875 * Many stacks (ours included) will set the ACK number in an 3876 * FIN|ACK if the SYN times out -- no sequence to ACK. 3877 */ 3878 ack = dst->seqlo; 3879 } 3880 3881 if (seq == end) { 3882 /* Ease sequencing restrictions on no data packets */ 3883 seq = src->seqlo; 3884 end = seq; 3885 } 3886 3887 ackskew = dst->seqlo - ack; 3888 3889 3890 /* 3891 * Need to demodulate the sequence numbers in any TCP SACK options 3892 * (Selective ACK). We could optionally validate the SACK values 3893 * against the current ACK window, either forwards or backwards, but 3894 * I'm not confident that SACK has been implemented properly 3895 * everywhere. It wouldn't surprise me if several stacks accidently 3896 * SACK too far backwards of previously ACKed data. There really aren't 3897 * any security implications of bad SACKing unless the target stack 3898 * doesn't validate the option length correctly. Someone trying to 3899 * spoof into a TCP connection won't bother blindly sending SACK 3900 * options anyway. 3901 */ 3902 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 3903 if (pf_modulate_sack(m, off, pd, th, dst)) 3904 *copyback = 1; 3905 } 3906 3907 3908#define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 3909 if (SEQ_GEQ(src->seqhi, end) && 3910 /* Last octet inside other's window space */ 3911 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 3912 /* Retrans: not more than one window back */ 3913 (ackskew >= -MAXACKWINDOW) && 3914 /* Acking not more than one reassembled fragment backwards */ 3915 (ackskew <= (MAXACKWINDOW << sws)) && 3916 /* Acking not more than one window forward */ 3917 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 3918 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 3919 (pd->flags & PFDESC_IP_REAS) == 0)) { 3920 /* Require an exact/+1 sequence match on resets when possible */ 3921 3922 if (dst->scrub || src->scrub) { 3923 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 3924 *state, src, dst, copyback)) 3925 return (PF_DROP); 3926 } 3927 3928 /* update max window */ 3929 if (src->max_win < win) 3930 src->max_win = win; 3931 /* synchronize sequencing */ 3932 if (SEQ_GT(end, src->seqlo)) 3933 src->seqlo = end; 3934 /* slide the window of what the other end can send */ 3935 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 3936 dst->seqhi = ack + MAX((win << sws), 1); 3937 3938 3939 /* update states */ 3940 if (th->th_flags & TH_SYN) 3941 if (src->state < TCPS_SYN_SENT) 3942 src->state = TCPS_SYN_SENT; 3943 if (th->th_flags & TH_FIN) 3944 if (src->state < TCPS_CLOSING) 3945 src->state = TCPS_CLOSING; 3946 if (th->th_flags & TH_ACK) { 3947 if (dst->state == TCPS_SYN_SENT) { 3948 dst->state = TCPS_ESTABLISHED; 3949 if (src->state == TCPS_ESTABLISHED && 3950 (*state)->src_node != NULL && 3951 pf_src_connlimit(state)) { 3952 REASON_SET(reason, PFRES_SRCLIMIT); 3953 return (PF_DROP); 3954 } 3955 } else if (dst->state == TCPS_CLOSING) 3956 dst->state = TCPS_FIN_WAIT_2; 3957 } 3958 if (th->th_flags & TH_RST) 3959 src->state = dst->state = TCPS_TIME_WAIT; 3960 3961 /* update expire time */ 3962 (*state)->expire = time_uptime; 3963 if (src->state >= TCPS_FIN_WAIT_2 && 3964 dst->state >= TCPS_FIN_WAIT_2) 3965 (*state)->timeout = PFTM_TCP_CLOSED; 3966 else if (src->state >= TCPS_CLOSING && 3967 dst->state >= TCPS_CLOSING) 3968 (*state)->timeout = PFTM_TCP_FIN_WAIT; 3969 else if (src->state < TCPS_ESTABLISHED || 3970 dst->state < TCPS_ESTABLISHED) 3971 (*state)->timeout = PFTM_TCP_OPENING; 3972 else if (src->state >= TCPS_CLOSING || 3973 dst->state >= TCPS_CLOSING) 3974 (*state)->timeout = PFTM_TCP_CLOSING; 3975 else 3976 (*state)->timeout = PFTM_TCP_ESTABLISHED; 3977 3978 /* Fall through to PASS packet */ 3979 3980 } else if ((dst->state < TCPS_SYN_SENT || 3981 dst->state >= TCPS_FIN_WAIT_2 || 3982 src->state >= TCPS_FIN_WAIT_2) && 3983 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 3984 /* Within a window forward of the originating packet */ 3985 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 3986 /* Within a window backward of the originating packet */ 3987 3988 /* 3989 * This currently handles three situations: 3990 * 1) Stupid stacks will shotgun SYNs before their peer 3991 * replies. 3992 * 2) When PF catches an already established stream (the 3993 * firewall rebooted, the state table was flushed, routes 3994 * changed...) 3995 * 3) Packets get funky immediately after the connection 3996 * closes (this should catch Solaris spurious ACK|FINs 3997 * that web servers like to spew after a close) 3998 * 3999 * This must be a little more careful than the above code 4000 * since packet floods will also be caught here. We don't 4001 * update the TTL here to mitigate the damage of a packet 4002 * flood and so the same code can handle awkward establishment 4003 * and a loosened connection close. 4004 * In the establishment case, a correct peer response will 4005 * validate the connection, go through the normal state code 4006 * and keep updating the state TTL. 4007 */ 4008 4009 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4010 printf("pf: loose state match: "); 4011 pf_print_state(*state); 4012 pf_print_flags(th->th_flags); 4013 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4014 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4015 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 4016 (unsigned long long)(*state)->packets[1], 4017 pd->dir == PF_IN ? "in" : "out", 4018 pd->dir == (*state)->direction ? "fwd" : "rev"); 4019 } 4020 4021 if (dst->scrub || src->scrub) { 4022 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4023 *state, src, dst, copyback)) 4024 return (PF_DROP); 4025 } 4026 4027 /* update max window */ 4028 if (src->max_win < win) 4029 src->max_win = win; 4030 /* synchronize sequencing */ 4031 if (SEQ_GT(end, src->seqlo)) 4032 src->seqlo = end; 4033 /* slide the window of what the other end can send */ 4034 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4035 dst->seqhi = ack + MAX((win << sws), 1); 4036 4037 /* 4038 * Cannot set dst->seqhi here since this could be a shotgunned 4039 * SYN and not an already established connection. 4040 */ 4041 4042 if (th->th_flags & TH_FIN) 4043 if (src->state < TCPS_CLOSING) 4044 src->state = TCPS_CLOSING; 4045 if (th->th_flags & TH_RST) 4046 src->state = dst->state = TCPS_TIME_WAIT; 4047 4048 /* Fall through to PASS packet */ 4049 4050 } else { 4051 if ((*state)->dst.state == TCPS_SYN_SENT && 4052 (*state)->src.state == TCPS_SYN_SENT) { 4053 /* Send RST for state mismatches during handshake */ 4054 if (!(th->th_flags & TH_RST)) 4055 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4056 pd->dst, pd->src, th->th_dport, 4057 th->th_sport, ntohl(th->th_ack), 0, 4058 TH_RST, 0, 0, 4059 (*state)->rule.ptr->return_ttl, 1, 0, 4060 kif->pfik_ifp); 4061 src->seqlo = 0; 4062 src->seqhi = 1; 4063 src->max_win = 1; 4064 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 4065 printf("pf: BAD state: "); 4066 pf_print_state(*state); 4067 pf_print_flags(th->th_flags); 4068 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4069 "pkts=%llu:%llu dir=%s,%s\n", 4070 seq, orig_seq, ack, pd->p_len, ackskew, 4071 (unsigned long long)(*state)->packets[0], 4072 (unsigned long long)(*state)->packets[1], 4073 pd->dir == PF_IN ? "in" : "out", 4074 pd->dir == (*state)->direction ? "fwd" : "rev"); 4075 printf("pf: State failure on: %c %c %c %c | %c %c\n", 4076 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4077 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4078 ' ': '2', 4079 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4080 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4081 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4082 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4083 } 4084 REASON_SET(reason, PFRES_BADSTATE); 4085 return (PF_DROP); 4086 } 4087 4088 return (PF_PASS); 4089} 4090 4091static int 4092pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, 4093 struct pf_state **state, struct pf_pdesc *pd, u_short *reason) 4094{ 4095 struct tcphdr *th = pd->hdr.tcp; 4096 4097 if (th->th_flags & TH_SYN) 4098 if (src->state < TCPS_SYN_SENT) 4099 src->state = TCPS_SYN_SENT; 4100 if (th->th_flags & TH_FIN) 4101 if (src->state < TCPS_CLOSING) 4102 src->state = TCPS_CLOSING; 4103 if (th->th_flags & TH_ACK) { 4104 if (dst->state == TCPS_SYN_SENT) { 4105 dst->state = TCPS_ESTABLISHED; 4106 if (src->state == TCPS_ESTABLISHED && 4107 (*state)->src_node != NULL && 4108 pf_src_connlimit(state)) { 4109 REASON_SET(reason, PFRES_SRCLIMIT); 4110 return (PF_DROP); 4111 } 4112 } else if (dst->state == TCPS_CLOSING) { 4113 dst->state = TCPS_FIN_WAIT_2; 4114 } else if (src->state == TCPS_SYN_SENT && 4115 dst->state < TCPS_SYN_SENT) { 4116 /* 4117 * Handle a special sloppy case where we only see one 4118 * half of the connection. If there is a ACK after 4119 * the initial SYN without ever seeing a packet from 4120 * the destination, set the connection to established. 4121 */ 4122 dst->state = src->state = TCPS_ESTABLISHED; 4123 if ((*state)->src_node != NULL && 4124 pf_src_connlimit(state)) { 4125 REASON_SET(reason, PFRES_SRCLIMIT); 4126 return (PF_DROP); 4127 } 4128 } else if (src->state == TCPS_CLOSING && 4129 dst->state == TCPS_ESTABLISHED && 4130 dst->seqlo == 0) { 4131 /* 4132 * Handle the closing of half connections where we 4133 * don't see the full bidirectional FIN/ACK+ACK 4134 * handshake. 4135 */ 4136 dst->state = TCPS_CLOSING; 4137 } 4138 } 4139 if (th->th_flags & TH_RST) 4140 src->state = dst->state = TCPS_TIME_WAIT; 4141 4142 /* update expire time */ 4143 (*state)->expire = time_uptime; 4144 if (src->state >= TCPS_FIN_WAIT_2 && 4145 dst->state >= TCPS_FIN_WAIT_2) 4146 (*state)->timeout = PFTM_TCP_CLOSED; 4147 else if (src->state >= TCPS_CLOSING && 4148 dst->state >= TCPS_CLOSING) 4149 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4150 else if (src->state < TCPS_ESTABLISHED || 4151 dst->state < TCPS_ESTABLISHED) 4152 (*state)->timeout = PFTM_TCP_OPENING; 4153 else if (src->state >= TCPS_CLOSING || 4154 dst->state >= TCPS_CLOSING) 4155 (*state)->timeout = PFTM_TCP_CLOSING; 4156 else 4157 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4158 4159 return (PF_PASS); 4160} 4161 4162static int 4163pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, 4164 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4165 u_short *reason) 4166{ 4167 struct pf_state_key_cmp key; 4168 struct tcphdr *th = pd->hdr.tcp; 4169 int copyback = 0; 4170 struct pf_state_peer *src, *dst; 4171 struct pf_state_key *sk; 4172 4173 bzero(&key, sizeof(key)); 4174 key.af = pd->af; 4175 key.proto = IPPROTO_TCP; 4176 if (direction == PF_IN) { /* wire side, straight */ 4177 PF_ACPY(&key.addr[0], pd->src, key.af); 4178 PF_ACPY(&key.addr[1], pd->dst, key.af); 4179 key.port[0] = th->th_sport; 4180 key.port[1] = th->th_dport; 4181 } else { /* stack side, reverse */ 4182 PF_ACPY(&key.addr[1], pd->src, key.af); 4183 PF_ACPY(&key.addr[0], pd->dst, key.af); 4184 key.port[1] = th->th_sport; 4185 key.port[0] = th->th_dport; 4186 } 4187 4188 STATE_LOOKUP(kif, &key, direction, *state, pd); 4189 4190 if (direction == (*state)->direction) { 4191 src = &(*state)->src; 4192 dst = &(*state)->dst; 4193 } else { 4194 src = &(*state)->dst; 4195 dst = &(*state)->src; 4196 } 4197 4198 sk = (*state)->key[pd->didx]; 4199 4200 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4201 if (direction != (*state)->direction) { 4202 REASON_SET(reason, PFRES_SYNPROXY); 4203 return (PF_SYNPROXY_DROP); 4204 } 4205 if (th->th_flags & TH_SYN) { 4206 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4207 REASON_SET(reason, PFRES_SYNPROXY); 4208 return (PF_DROP); 4209 } 4210 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4211 pd->src, th->th_dport, th->th_sport, 4212 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4213 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); 4214 REASON_SET(reason, PFRES_SYNPROXY); 4215 return (PF_SYNPROXY_DROP); 4216 } else if (!(th->th_flags & TH_ACK) || 4217 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4218 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4219 REASON_SET(reason, PFRES_SYNPROXY); 4220 return (PF_DROP); 4221 } else if ((*state)->src_node != NULL && 4222 pf_src_connlimit(state)) { 4223 REASON_SET(reason, PFRES_SRCLIMIT); 4224 return (PF_DROP); 4225 } else 4226 (*state)->src.state = PF_TCPS_PROXY_DST; 4227 } 4228 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4229 if (direction == (*state)->direction) { 4230 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4231 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4232 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4233 REASON_SET(reason, PFRES_SYNPROXY); 4234 return (PF_DROP); 4235 } 4236 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4237 if ((*state)->dst.seqhi == 1) 4238 (*state)->dst.seqhi = htonl(arc4random()); 4239 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4240 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4241 sk->port[pd->sidx], sk->port[pd->didx], 4242 (*state)->dst.seqhi, 0, TH_SYN, 0, 4243 (*state)->src.mss, 0, 0, (*state)->tag, NULL); 4244 REASON_SET(reason, PFRES_SYNPROXY); 4245 return (PF_SYNPROXY_DROP); 4246 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4247 (TH_SYN|TH_ACK)) || 4248 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4249 REASON_SET(reason, PFRES_SYNPROXY); 4250 return (PF_DROP); 4251 } else { 4252 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4253 (*state)->dst.seqlo = ntohl(th->th_seq); 4254 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, 4255 pd->src, th->th_dport, th->th_sport, 4256 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4257 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4258 (*state)->tag, NULL); 4259 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, 4260 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4261 sk->port[pd->sidx], sk->port[pd->didx], 4262 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4263 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); 4264 (*state)->src.seqdiff = (*state)->dst.seqhi - 4265 (*state)->src.seqlo; 4266 (*state)->dst.seqdiff = (*state)->src.seqhi - 4267 (*state)->dst.seqlo; 4268 (*state)->src.seqhi = (*state)->src.seqlo + 4269 (*state)->dst.max_win; 4270 (*state)->dst.seqhi = (*state)->dst.seqlo + 4271 (*state)->src.max_win; 4272 (*state)->src.wscale = (*state)->dst.wscale = 0; 4273 (*state)->src.state = (*state)->dst.state = 4274 TCPS_ESTABLISHED; 4275 REASON_SET(reason, PFRES_SYNPROXY); 4276 return (PF_SYNPROXY_DROP); 4277 } 4278 } 4279 4280 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4281 dst->state >= TCPS_FIN_WAIT_2 && 4282 src->state >= TCPS_FIN_WAIT_2) { 4283 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4284 printf("pf: state reuse "); 4285 pf_print_state(*state); 4286 pf_print_flags(th->th_flags); 4287 printf("\n"); 4288 } 4289 /* XXX make sure it's the same direction ?? */ 4290 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; 4291 pf_unlink_state(*state, PF_ENTER_LOCKED); 4292 *state = NULL; 4293 return (PF_DROP); 4294 } 4295 4296 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4297 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) 4298 return (PF_DROP); 4299 } else { 4300 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, 4301 ©back) == PF_DROP) 4302 return (PF_DROP); 4303 } 4304 4305 /* translate source/destination address, if necessary */ 4306 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4307 struct pf_state_key *nk = (*state)->key[pd->didx]; 4308 4309 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4310 nk->port[pd->sidx] != th->th_sport) 4311 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum, 4312 &th->th_sum, &nk->addr[pd->sidx], 4313 nk->port[pd->sidx], 0, pd->af); 4314 4315 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4316 nk->port[pd->didx] != th->th_dport) 4317 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum, 4318 &th->th_sum, &nk->addr[pd->didx], 4319 nk->port[pd->didx], 0, pd->af); 4320 copyback = 1; 4321 } 4322 4323 /* Copyback sequence modulation or stateful scrub changes if needed */ 4324 if (copyback) 4325 m_copyback(m, off, sizeof(*th), (caddr_t)th); 4326 4327 return (PF_PASS); 4328} 4329 4330static int 4331pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, 4332 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 4333{ 4334 struct pf_state_peer *src, *dst; 4335 struct pf_state_key_cmp key; 4336 struct udphdr *uh = pd->hdr.udp; 4337 4338 bzero(&key, sizeof(key)); 4339 key.af = pd->af; 4340 key.proto = IPPROTO_UDP; 4341 if (direction == PF_IN) { /* wire side, straight */ 4342 PF_ACPY(&key.addr[0], pd->src, key.af); 4343 PF_ACPY(&key.addr[1], pd->dst, key.af); 4344 key.port[0] = uh->uh_sport; 4345 key.port[1] = uh->uh_dport; 4346 } else { /* stack side, reverse */ 4347 PF_ACPY(&key.addr[1], pd->src, key.af); 4348 PF_ACPY(&key.addr[0], pd->dst, key.af); 4349 key.port[1] = uh->uh_sport; 4350 key.port[0] = uh->uh_dport; 4351 } 4352 4353 STATE_LOOKUP(kif, &key, direction, *state, pd); 4354 4355 if (direction == (*state)->direction) { 4356 src = &(*state)->src; 4357 dst = &(*state)->dst; 4358 } else { 4359 src = &(*state)->dst; 4360 dst = &(*state)->src; 4361 } 4362 4363 /* update states */ 4364 if (src->state < PFUDPS_SINGLE) 4365 src->state = PFUDPS_SINGLE; 4366 if (dst->state == PFUDPS_SINGLE) 4367 dst->state = PFUDPS_MULTIPLE; 4368 4369 /* update expire time */ 4370 (*state)->expire = time_uptime; 4371 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 4372 (*state)->timeout = PFTM_UDP_MULTIPLE; 4373 else 4374 (*state)->timeout = PFTM_UDP_SINGLE; 4375 4376 /* translate source/destination address, if necessary */ 4377 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4378 struct pf_state_key *nk = (*state)->key[pd->didx]; 4379 4380 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4381 nk->port[pd->sidx] != uh->uh_sport) 4382 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum, 4383 &uh->uh_sum, &nk->addr[pd->sidx], 4384 nk->port[pd->sidx], 1, pd->af); 4385 4386 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4387 nk->port[pd->didx] != uh->uh_dport) 4388 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum, 4389 &uh->uh_sum, &nk->addr[pd->didx], 4390 nk->port[pd->didx], 1, pd->af); 4391 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 4392 } 4393 4394 return (PF_PASS); 4395} 4396 4397static int 4398pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, 4399 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 4400{ 4401 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 4402 u_int16_t icmpid = 0, *icmpsum; 4403 u_int8_t icmptype; 4404 int state_icmp = 0; 4405 struct pf_state_key_cmp key; 4406 4407 bzero(&key, sizeof(key)); 4408 switch (pd->proto) { 4409#ifdef INET 4410 case IPPROTO_ICMP: 4411 icmptype = pd->hdr.icmp->icmp_type; 4412 icmpid = pd->hdr.icmp->icmp_id; 4413 icmpsum = &pd->hdr.icmp->icmp_cksum; 4414 4415 if (icmptype == ICMP_UNREACH || 4416 icmptype == ICMP_SOURCEQUENCH || 4417 icmptype == ICMP_REDIRECT || 4418 icmptype == ICMP_TIMXCEED || 4419 icmptype == ICMP_PARAMPROB) 4420 state_icmp++; 4421 break; 4422#endif /* INET */ 4423#ifdef INET6 4424 case IPPROTO_ICMPV6: 4425 icmptype = pd->hdr.icmp6->icmp6_type; 4426 icmpid = pd->hdr.icmp6->icmp6_id; 4427 icmpsum = &pd->hdr.icmp6->icmp6_cksum; 4428 4429 if (icmptype == ICMP6_DST_UNREACH || 4430 icmptype == ICMP6_PACKET_TOO_BIG || 4431 icmptype == ICMP6_TIME_EXCEEDED || 4432 icmptype == ICMP6_PARAM_PROB) 4433 state_icmp++; 4434 break; 4435#endif /* INET6 */ 4436 } 4437 4438 if (!state_icmp) { 4439 4440 /* 4441 * ICMP query/reply message not related to a TCP/UDP packet. 4442 * Search for an ICMP state. 4443 */ 4444 key.af = pd->af; 4445 key.proto = pd->proto; 4446 key.port[0] = key.port[1] = icmpid; 4447 if (direction == PF_IN) { /* wire side, straight */ 4448 PF_ACPY(&key.addr[0], pd->src, key.af); 4449 PF_ACPY(&key.addr[1], pd->dst, key.af); 4450 } else { /* stack side, reverse */ 4451 PF_ACPY(&key.addr[1], pd->src, key.af); 4452 PF_ACPY(&key.addr[0], pd->dst, key.af); 4453 } 4454 4455 STATE_LOOKUP(kif, &key, direction, *state, pd); 4456 4457 (*state)->expire = time_uptime; 4458 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 4459 4460 /* translate source/destination address, if necessary */ 4461 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4462 struct pf_state_key *nk = (*state)->key[pd->didx]; 4463 4464 switch (pd->af) { 4465#ifdef INET 4466 case AF_INET: 4467 if (PF_ANEQ(pd->src, 4468 &nk->addr[pd->sidx], AF_INET)) 4469 pf_change_a(&saddr->v4.s_addr, 4470 pd->ip_sum, 4471 nk->addr[pd->sidx].v4.s_addr, 0); 4472 4473 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 4474 AF_INET)) 4475 pf_change_a(&daddr->v4.s_addr, 4476 pd->ip_sum, 4477 nk->addr[pd->didx].v4.s_addr, 0); 4478 4479 if (nk->port[0] != 4480 pd->hdr.icmp->icmp_id) { 4481 pd->hdr.icmp->icmp_cksum = 4482 pf_cksum_fixup( 4483 pd->hdr.icmp->icmp_cksum, icmpid, 4484 nk->port[pd->sidx], 0); 4485 pd->hdr.icmp->icmp_id = 4486 nk->port[pd->sidx]; 4487 } 4488 4489 m_copyback(m, off, ICMP_MINLEN, 4490 (caddr_t )pd->hdr.icmp); 4491 break; 4492#endif /* INET */ 4493#ifdef INET6 4494 case AF_INET6: 4495 if (PF_ANEQ(pd->src, 4496 &nk->addr[pd->sidx], AF_INET6)) 4497 pf_change_a6(saddr, 4498 &pd->hdr.icmp6->icmp6_cksum, 4499 &nk->addr[pd->sidx], 0); 4500 4501 if (PF_ANEQ(pd->dst, 4502 &nk->addr[pd->didx], AF_INET6)) 4503 pf_change_a6(daddr, 4504 &pd->hdr.icmp6->icmp6_cksum, 4505 &nk->addr[pd->didx], 0); 4506 4507 m_copyback(m, off, sizeof(struct icmp6_hdr), 4508 (caddr_t )pd->hdr.icmp6); 4509 break; 4510#endif /* INET6 */ 4511 } 4512 } 4513 return (PF_PASS); 4514 4515 } else { 4516 /* 4517 * ICMP error message in response to a TCP/UDP packet. 4518 * Extract the inner TCP/UDP header and search for that state. 4519 */ 4520 4521 struct pf_pdesc pd2; 4522 bzero(&pd2, sizeof pd2); 4523#ifdef INET 4524 struct ip h2; 4525#endif /* INET */ 4526#ifdef INET6 4527 struct ip6_hdr h2_6; 4528 int terminal = 0; 4529#endif /* INET6 */ 4530 int ipoff2 = 0; 4531 int off2 = 0; 4532 4533 pd2.af = pd->af; 4534 /* Payload packet is from the opposite direction. */ 4535 pd2.sidx = (direction == PF_IN) ? 1 : 0; 4536 pd2.didx = (direction == PF_IN) ? 0 : 1; 4537 switch (pd->af) { 4538#ifdef INET 4539 case AF_INET: 4540 /* offset of h2 in mbuf chain */ 4541 ipoff2 = off + ICMP_MINLEN; 4542 4543 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 4544 NULL, reason, pd2.af)) { 4545 DPFPRINTF(PF_DEBUG_MISC, 4546 ("pf: ICMP error message too short " 4547 "(ip)\n")); 4548 return (PF_DROP); 4549 } 4550 /* 4551 * ICMP error messages don't refer to non-first 4552 * fragments 4553 */ 4554 if (h2.ip_off & htons(IP_OFFMASK)) { 4555 REASON_SET(reason, PFRES_FRAG); 4556 return (PF_DROP); 4557 } 4558 4559 /* offset of protocol header that follows h2 */ 4560 off2 = ipoff2 + (h2.ip_hl << 2); 4561 4562 pd2.proto = h2.ip_p; 4563 pd2.src = (struct pf_addr *)&h2.ip_src; 4564 pd2.dst = (struct pf_addr *)&h2.ip_dst; 4565 pd2.ip_sum = &h2.ip_sum; 4566 break; 4567#endif /* INET */ 4568#ifdef INET6 4569 case AF_INET6: 4570 ipoff2 = off + sizeof(struct icmp6_hdr); 4571 4572 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 4573 NULL, reason, pd2.af)) { 4574 DPFPRINTF(PF_DEBUG_MISC, 4575 ("pf: ICMP error message too short " 4576 "(ip6)\n")); 4577 return (PF_DROP); 4578 } 4579 pd2.proto = h2_6.ip6_nxt; 4580 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 4581 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 4582 pd2.ip_sum = NULL; 4583 off2 = ipoff2 + sizeof(h2_6); 4584 do { 4585 switch (pd2.proto) { 4586 case IPPROTO_FRAGMENT: 4587 /* 4588 * ICMPv6 error messages for 4589 * non-first fragments 4590 */ 4591 REASON_SET(reason, PFRES_FRAG); 4592 return (PF_DROP); 4593 case IPPROTO_AH: 4594 case IPPROTO_HOPOPTS: 4595 case IPPROTO_ROUTING: 4596 case IPPROTO_DSTOPTS: { 4597 /* get next header and header length */ 4598 struct ip6_ext opt6; 4599 4600 if (!pf_pull_hdr(m, off2, &opt6, 4601 sizeof(opt6), NULL, reason, 4602 pd2.af)) { 4603 DPFPRINTF(PF_DEBUG_MISC, 4604 ("pf: ICMPv6 short opt\n")); 4605 return (PF_DROP); 4606 } 4607 if (pd2.proto == IPPROTO_AH) 4608 off2 += (opt6.ip6e_len + 2) * 4; 4609 else 4610 off2 += (opt6.ip6e_len + 1) * 8; 4611 pd2.proto = opt6.ip6e_nxt; 4612 /* goto the next header */ 4613 break; 4614 } 4615 default: 4616 terminal++; 4617 break; 4618 } 4619 } while (!terminal); 4620 break; 4621#endif /* INET6 */ 4622 } 4623 4624 switch (pd2.proto) { 4625 case IPPROTO_TCP: { 4626 struct tcphdr th; 4627 u_int32_t seq; 4628 struct pf_state_peer *src, *dst; 4629 u_int8_t dws; 4630 int copyback = 0; 4631 4632 /* 4633 * Only the first 8 bytes of the TCP header can be 4634 * expected. Don't access any TCP header fields after 4635 * th_seq, an ackskew test is not possible. 4636 */ 4637 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 4638 pd2.af)) { 4639 DPFPRINTF(PF_DEBUG_MISC, 4640 ("pf: ICMP error message too short " 4641 "(tcp)\n")); 4642 return (PF_DROP); 4643 } 4644 4645 key.af = pd2.af; 4646 key.proto = IPPROTO_TCP; 4647 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4648 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4649 key.port[pd2.sidx] = th.th_sport; 4650 key.port[pd2.didx] = th.th_dport; 4651 4652 STATE_LOOKUP(kif, &key, direction, *state, pd); 4653 4654 if (direction == (*state)->direction) { 4655 src = &(*state)->dst; 4656 dst = &(*state)->src; 4657 } else { 4658 src = &(*state)->src; 4659 dst = &(*state)->dst; 4660 } 4661 4662 if (src->wscale && dst->wscale) 4663 dws = dst->wscale & PF_WSCALE_MASK; 4664 else 4665 dws = 0; 4666 4667 /* Demodulate sequence number */ 4668 seq = ntohl(th.th_seq) - src->seqdiff; 4669 if (src->seqdiff) { 4670 pf_change_a(&th.th_seq, icmpsum, 4671 htonl(seq), 0); 4672 copyback = 1; 4673 } 4674 4675 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 4676 (!SEQ_GEQ(src->seqhi, seq) || 4677 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 4678 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4679 printf("pf: BAD ICMP %d:%d ", 4680 icmptype, pd->hdr.icmp->icmp_code); 4681 pf_print_host(pd->src, 0, pd->af); 4682 printf(" -> "); 4683 pf_print_host(pd->dst, 0, pd->af); 4684 printf(" state: "); 4685 pf_print_state(*state); 4686 printf(" seq=%u\n", seq); 4687 } 4688 REASON_SET(reason, PFRES_BADSTATE); 4689 return (PF_DROP); 4690 } else { 4691 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4692 printf("pf: OK ICMP %d:%d ", 4693 icmptype, pd->hdr.icmp->icmp_code); 4694 pf_print_host(pd->src, 0, pd->af); 4695 printf(" -> "); 4696 pf_print_host(pd->dst, 0, pd->af); 4697 printf(" state: "); 4698 pf_print_state(*state); 4699 printf(" seq=%u\n", seq); 4700 } 4701 } 4702 4703 /* translate source/destination address, if necessary */ 4704 if ((*state)->key[PF_SK_WIRE] != 4705 (*state)->key[PF_SK_STACK]) { 4706 struct pf_state_key *nk = 4707 (*state)->key[pd->didx]; 4708 4709 if (PF_ANEQ(pd2.src, 4710 &nk->addr[pd2.sidx], pd2.af) || 4711 nk->port[pd2.sidx] != th.th_sport) 4712 pf_change_icmp(pd2.src, &th.th_sport, 4713 daddr, &nk->addr[pd2.sidx], 4714 nk->port[pd2.sidx], NULL, 4715 pd2.ip_sum, icmpsum, 4716 pd->ip_sum, 0, pd2.af); 4717 4718 if (PF_ANEQ(pd2.dst, 4719 &nk->addr[pd2.didx], pd2.af) || 4720 nk->port[pd2.didx] != th.th_dport) 4721 pf_change_icmp(pd2.dst, &th.th_dport, 4722 NULL, /* XXX Inbound NAT? */ 4723 &nk->addr[pd2.didx], 4724 nk->port[pd2.didx], NULL, 4725 pd2.ip_sum, icmpsum, 4726 pd->ip_sum, 0, pd2.af); 4727 copyback = 1; 4728 } 4729 4730 if (copyback) { 4731 switch (pd2.af) { 4732#ifdef INET 4733 case AF_INET: 4734 m_copyback(m, off, ICMP_MINLEN, 4735 (caddr_t )pd->hdr.icmp); 4736 m_copyback(m, ipoff2, sizeof(h2), 4737 (caddr_t )&h2); 4738 break; 4739#endif /* INET */ 4740#ifdef INET6 4741 case AF_INET6: 4742 m_copyback(m, off, 4743 sizeof(struct icmp6_hdr), 4744 (caddr_t )pd->hdr.icmp6); 4745 m_copyback(m, ipoff2, sizeof(h2_6), 4746 (caddr_t )&h2_6); 4747 break; 4748#endif /* INET6 */ 4749 } 4750 m_copyback(m, off2, 8, (caddr_t)&th); 4751 } 4752 4753 return (PF_PASS); 4754 break; 4755 } 4756 case IPPROTO_UDP: { 4757 struct udphdr uh; 4758 4759 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 4760 NULL, reason, pd2.af)) { 4761 DPFPRINTF(PF_DEBUG_MISC, 4762 ("pf: ICMP error message too short " 4763 "(udp)\n")); 4764 return (PF_DROP); 4765 } 4766 4767 key.af = pd2.af; 4768 key.proto = IPPROTO_UDP; 4769 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4770 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4771 key.port[pd2.sidx] = uh.uh_sport; 4772 key.port[pd2.didx] = uh.uh_dport; 4773 4774 STATE_LOOKUP(kif, &key, direction, *state, pd); 4775 4776 /* translate source/destination address, if necessary */ 4777 if ((*state)->key[PF_SK_WIRE] != 4778 (*state)->key[PF_SK_STACK]) { 4779 struct pf_state_key *nk = 4780 (*state)->key[pd->didx]; 4781 4782 if (PF_ANEQ(pd2.src, 4783 &nk->addr[pd2.sidx], pd2.af) || 4784 nk->port[pd2.sidx] != uh.uh_sport) 4785 pf_change_icmp(pd2.src, &uh.uh_sport, 4786 daddr, &nk->addr[pd2.sidx], 4787 nk->port[pd2.sidx], &uh.uh_sum, 4788 pd2.ip_sum, icmpsum, 4789 pd->ip_sum, 1, pd2.af); 4790 4791 if (PF_ANEQ(pd2.dst, 4792 &nk->addr[pd2.didx], pd2.af) || 4793 nk->port[pd2.didx] != uh.uh_dport) 4794 pf_change_icmp(pd2.dst, &uh.uh_dport, 4795 NULL, /* XXX Inbound NAT? */ 4796 &nk->addr[pd2.didx], 4797 nk->port[pd2.didx], &uh.uh_sum, 4798 pd2.ip_sum, icmpsum, 4799 pd->ip_sum, 1, pd2.af); 4800 4801 switch (pd2.af) { 4802#ifdef INET 4803 case AF_INET: 4804 m_copyback(m, off, ICMP_MINLEN, 4805 (caddr_t )pd->hdr.icmp); 4806 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4807 break; 4808#endif /* INET */ 4809#ifdef INET6 4810 case AF_INET6: 4811 m_copyback(m, off, 4812 sizeof(struct icmp6_hdr), 4813 (caddr_t )pd->hdr.icmp6); 4814 m_copyback(m, ipoff2, sizeof(h2_6), 4815 (caddr_t )&h2_6); 4816 break; 4817#endif /* INET6 */ 4818 } 4819 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 4820 } 4821 return (PF_PASS); 4822 break; 4823 } 4824#ifdef INET 4825 case IPPROTO_ICMP: { 4826 struct icmp iih; 4827 4828 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 4829 NULL, reason, pd2.af)) { 4830 DPFPRINTF(PF_DEBUG_MISC, 4831 ("pf: ICMP error message too short i" 4832 "(icmp)\n")); 4833 return (PF_DROP); 4834 } 4835 4836 key.af = pd2.af; 4837 key.proto = IPPROTO_ICMP; 4838 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4839 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4840 key.port[0] = key.port[1] = iih.icmp_id; 4841 4842 STATE_LOOKUP(kif, &key, direction, *state, pd); 4843 4844 /* translate source/destination address, if necessary */ 4845 if ((*state)->key[PF_SK_WIRE] != 4846 (*state)->key[PF_SK_STACK]) { 4847 struct pf_state_key *nk = 4848 (*state)->key[pd->didx]; 4849 4850 if (PF_ANEQ(pd2.src, 4851 &nk->addr[pd2.sidx], pd2.af) || 4852 nk->port[pd2.sidx] != iih.icmp_id) 4853 pf_change_icmp(pd2.src, &iih.icmp_id, 4854 daddr, &nk->addr[pd2.sidx], 4855 nk->port[pd2.sidx], NULL, 4856 pd2.ip_sum, icmpsum, 4857 pd->ip_sum, 0, AF_INET); 4858 4859 if (PF_ANEQ(pd2.dst, 4860 &nk->addr[pd2.didx], pd2.af) || 4861 nk->port[pd2.didx] != iih.icmp_id) 4862 pf_change_icmp(pd2.dst, &iih.icmp_id, 4863 NULL, /* XXX Inbound NAT? */ 4864 &nk->addr[pd2.didx], 4865 nk->port[pd2.didx], NULL, 4866 pd2.ip_sum, icmpsum, 4867 pd->ip_sum, 0, AF_INET); 4868 4869 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); 4870 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4871 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 4872 } 4873 return (PF_PASS); 4874 break; 4875 } 4876#endif /* INET */ 4877#ifdef INET6 4878 case IPPROTO_ICMPV6: { 4879 struct icmp6_hdr iih; 4880 4881 if (!pf_pull_hdr(m, off2, &iih, 4882 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 4883 DPFPRINTF(PF_DEBUG_MISC, 4884 ("pf: ICMP error message too short " 4885 "(icmp6)\n")); 4886 return (PF_DROP); 4887 } 4888 4889 key.af = pd2.af; 4890 key.proto = IPPROTO_ICMPV6; 4891 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4892 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4893 key.port[0] = key.port[1] = iih.icmp6_id; 4894 4895 STATE_LOOKUP(kif, &key, direction, *state, pd); 4896 4897 /* translate source/destination address, if necessary */ 4898 if ((*state)->key[PF_SK_WIRE] != 4899 (*state)->key[PF_SK_STACK]) { 4900 struct pf_state_key *nk = 4901 (*state)->key[pd->didx]; 4902 4903 if (PF_ANEQ(pd2.src, 4904 &nk->addr[pd2.sidx], pd2.af) || 4905 nk->port[pd2.sidx] != iih.icmp6_id) 4906 pf_change_icmp(pd2.src, &iih.icmp6_id, 4907 daddr, &nk->addr[pd2.sidx], 4908 nk->port[pd2.sidx], NULL, 4909 pd2.ip_sum, icmpsum, 4910 pd->ip_sum, 0, AF_INET6); 4911 4912 if (PF_ANEQ(pd2.dst, 4913 &nk->addr[pd2.didx], pd2.af) || 4914 nk->port[pd2.didx] != iih.icmp6_id) 4915 pf_change_icmp(pd2.dst, &iih.icmp6_id, 4916 NULL, /* XXX Inbound NAT? */ 4917 &nk->addr[pd2.didx], 4918 nk->port[pd2.didx], NULL, 4919 pd2.ip_sum, icmpsum, 4920 pd->ip_sum, 0, AF_INET6); 4921 4922 m_copyback(m, off, sizeof(struct icmp6_hdr), 4923 (caddr_t)pd->hdr.icmp6); 4924 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 4925 m_copyback(m, off2, sizeof(struct icmp6_hdr), 4926 (caddr_t)&iih); 4927 } 4928 return (PF_PASS); 4929 break; 4930 } 4931#endif /* INET6 */ 4932 default: { 4933 key.af = pd2.af; 4934 key.proto = pd2.proto; 4935 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 4936 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 4937 key.port[0] = key.port[1] = 0; 4938 4939 STATE_LOOKUP(kif, &key, direction, *state, pd); 4940 4941 /* translate source/destination address, if necessary */ 4942 if ((*state)->key[PF_SK_WIRE] != 4943 (*state)->key[PF_SK_STACK]) { 4944 struct pf_state_key *nk = 4945 (*state)->key[pd->didx]; 4946 4947 if (PF_ANEQ(pd2.src, 4948 &nk->addr[pd2.sidx], pd2.af)) 4949 pf_change_icmp(pd2.src, NULL, daddr, 4950 &nk->addr[pd2.sidx], 0, NULL, 4951 pd2.ip_sum, icmpsum, 4952 pd->ip_sum, 0, pd2.af); 4953 4954 if (PF_ANEQ(pd2.dst, 4955 &nk->addr[pd2.didx], pd2.af)) 4956 pf_change_icmp(pd2.src, NULL, 4957 NULL, /* XXX Inbound NAT? */ 4958 &nk->addr[pd2.didx], 0, NULL, 4959 pd2.ip_sum, icmpsum, 4960 pd->ip_sum, 0, pd2.af); 4961 4962 switch (pd2.af) { 4963#ifdef INET 4964 case AF_INET: 4965 m_copyback(m, off, ICMP_MINLEN, 4966 (caddr_t)pd->hdr.icmp); 4967 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 4968 break; 4969#endif /* INET */ 4970#ifdef INET6 4971 case AF_INET6: 4972 m_copyback(m, off, 4973 sizeof(struct icmp6_hdr), 4974 (caddr_t )pd->hdr.icmp6); 4975 m_copyback(m, ipoff2, sizeof(h2_6), 4976 (caddr_t )&h2_6); 4977 break; 4978#endif /* INET6 */ 4979 } 4980 } 4981 return (PF_PASS); 4982 break; 4983 } 4984 } 4985 } 4986} 4987 4988static int 4989pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, 4990 struct mbuf *m, struct pf_pdesc *pd) 4991{ 4992 struct pf_state_peer *src, *dst; 4993 struct pf_state_key_cmp key; 4994 4995 bzero(&key, sizeof(key)); 4996 key.af = pd->af; 4997 key.proto = pd->proto; 4998 if (direction == PF_IN) { 4999 PF_ACPY(&key.addr[0], pd->src, key.af); 5000 PF_ACPY(&key.addr[1], pd->dst, key.af); 5001 key.port[0] = key.port[1] = 0; 5002 } else { 5003 PF_ACPY(&key.addr[1], pd->src, key.af); 5004 PF_ACPY(&key.addr[0], pd->dst, key.af); 5005 key.port[1] = key.port[0] = 0; 5006 } 5007 5008 STATE_LOOKUP(kif, &key, direction, *state, pd); 5009 5010 if (direction == (*state)->direction) { 5011 src = &(*state)->src; 5012 dst = &(*state)->dst; 5013 } else { 5014 src = &(*state)->dst; 5015 dst = &(*state)->src; 5016 } 5017 5018 /* update states */ 5019 if (src->state < PFOTHERS_SINGLE) 5020 src->state = PFOTHERS_SINGLE; 5021 if (dst->state == PFOTHERS_SINGLE) 5022 dst->state = PFOTHERS_MULTIPLE; 5023 5024 /* update expire time */ 5025 (*state)->expire = time_uptime; 5026 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5027 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5028 else 5029 (*state)->timeout = PFTM_OTHER_SINGLE; 5030 5031 /* translate source/destination address, if necessary */ 5032 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5033 struct pf_state_key *nk = (*state)->key[pd->didx]; 5034 5035 KASSERT(nk, ("%s: nk is null", __func__)); 5036 KASSERT(pd, ("%s: pd is null", __func__)); 5037 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 5038 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 5039 switch (pd->af) { 5040#ifdef INET 5041 case AF_INET: 5042 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5043 pf_change_a(&pd->src->v4.s_addr, 5044 pd->ip_sum, 5045 nk->addr[pd->sidx].v4.s_addr, 5046 0); 5047 5048 5049 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5050 pf_change_a(&pd->dst->v4.s_addr, 5051 pd->ip_sum, 5052 nk->addr[pd->didx].v4.s_addr, 5053 0); 5054 5055 break; 5056#endif /* INET */ 5057#ifdef INET6 5058 case AF_INET6: 5059 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5060 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5061 5062 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5063 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5064#endif /* INET6 */ 5065 } 5066 } 5067 return (PF_PASS); 5068} 5069 5070/* 5071 * ipoff and off are measured from the start of the mbuf chain. 5072 * h must be at "ipoff" on the mbuf chain. 5073 */ 5074void * 5075pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5076 u_short *actionp, u_short *reasonp, sa_family_t af) 5077{ 5078 switch (af) { 5079#ifdef INET 5080 case AF_INET: { 5081 struct ip *h = mtod(m, struct ip *); 5082 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 5083 5084 if (fragoff) { 5085 if (fragoff >= len) 5086 ACTION_SET(actionp, PF_PASS); 5087 else { 5088 ACTION_SET(actionp, PF_DROP); 5089 REASON_SET(reasonp, PFRES_FRAG); 5090 } 5091 return (NULL); 5092 } 5093 if (m->m_pkthdr.len < off + len || 5094 ntohs(h->ip_len) < off + len) { 5095 ACTION_SET(actionp, PF_DROP); 5096 REASON_SET(reasonp, PFRES_SHORT); 5097 return (NULL); 5098 } 5099 break; 5100 } 5101#endif /* INET */ 5102#ifdef INET6 5103 case AF_INET6: { 5104 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5105 5106 if (m->m_pkthdr.len < off + len || 5107 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5108 (unsigned)(off + len)) { 5109 ACTION_SET(actionp, PF_DROP); 5110 REASON_SET(reasonp, PFRES_SHORT); 5111 return (NULL); 5112 } 5113 break; 5114 } 5115#endif /* INET6 */ 5116 } 5117 m_copydata(m, off, len, p); 5118 return (p); 5119} 5120 5121int 5122pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, 5123 int rtableid) 5124{ 5125#ifdef RADIX_MPATH 5126 struct radix_node_head *rnh; 5127#endif 5128 struct sockaddr_in *dst; 5129 int ret = 1; 5130 int check_mpath; 5131#ifdef INET6 5132 struct sockaddr_in6 *dst6; 5133 struct route_in6 ro; 5134#else 5135 struct route ro; 5136#endif 5137 struct radix_node *rn; 5138 struct rtentry *rt; 5139 struct ifnet *ifp; 5140 5141 check_mpath = 0; 5142#ifdef RADIX_MPATH 5143 /* XXX: stick to table 0 for now */ 5144 rnh = rt_tables_get_rnh(0, af); 5145 if (rnh != NULL && rn_mpath_capable(rnh)) 5146 check_mpath = 1; 5147#endif 5148 bzero(&ro, sizeof(ro)); 5149 switch (af) { 5150 case AF_INET: 5151 dst = satosin(&ro.ro_dst); 5152 dst->sin_family = AF_INET; 5153 dst->sin_len = sizeof(*dst); 5154 dst->sin_addr = addr->v4; 5155 break; 5156#ifdef INET6 5157 case AF_INET6: 5158 /* 5159 * Skip check for addresses with embedded interface scope, 5160 * as they would always match anyway. 5161 */ 5162 if (IN6_IS_SCOPE_EMBED(&addr->v6)) 5163 goto out; 5164 dst6 = (struct sockaddr_in6 *)&ro.ro_dst; 5165 dst6->sin6_family = AF_INET6; 5166 dst6->sin6_len = sizeof(*dst6); 5167 dst6->sin6_addr = addr->v6; 5168 break; 5169#endif /* INET6 */ 5170 default: 5171 return (0); 5172 } 5173 5174 /* Skip checks for ipsec interfaces */ 5175 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5176 goto out; 5177 5178 switch (af) { 5179#ifdef INET6 5180 case AF_INET6: 5181 in6_rtalloc_ign(&ro, 0, rtableid); 5182 break; 5183#endif 5184#ifdef INET 5185 case AF_INET: 5186 in_rtalloc_ign((struct route *)&ro, 0, rtableid); 5187 break; 5188#endif 5189 default: 5190 rtalloc_ign((struct route *)&ro, 0); /* No/default FIB. */ 5191 break; 5192 } 5193 5194 if (ro.ro_rt != NULL) { 5195 /* No interface given, this is a no-route check */ 5196 if (kif == NULL) 5197 goto out; 5198 5199 if (kif->pfik_ifp == NULL) { 5200 ret = 0; 5201 goto out; 5202 } 5203 5204 /* Perform uRPF check if passed input interface */ 5205 ret = 0; 5206 rn = (struct radix_node *)ro.ro_rt; 5207 do { 5208 rt = (struct rtentry *)rn; 5209 ifp = rt->rt_ifp; 5210 5211 if (kif->pfik_ifp == ifp) 5212 ret = 1; 5213#ifdef RADIX_MPATH 5214 rn = rn_mpath_next(rn); 5215#endif 5216 } while (check_mpath == 1 && rn != NULL && ret == 0); 5217 } else 5218 ret = 0; 5219out: 5220 if (ro.ro_rt != NULL) 5221 RTFREE(ro.ro_rt); 5222 return (ret); 5223} 5224 5225#ifdef INET 5226static void 5227pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5228 struct pf_state *s, struct pf_pdesc *pd) 5229{ 5230 struct mbuf *m0, *m1; 5231 struct sockaddr_in dst; 5232 struct ip *ip; 5233 struct ifnet *ifp = NULL; 5234 struct pf_addr naddr; 5235 struct pf_src_node *sn = NULL; 5236 int error = 0; 5237 uint16_t ip_len, ip_off; 5238 5239 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5240 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5241 __func__)); 5242 5243 if ((pd->pf_mtag == NULL && 5244 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5245 pd->pf_mtag->routed++ > 3) { 5246 m0 = *m; 5247 *m = NULL; 5248 goto bad_locked; 5249 } 5250 5251 if (r->rt == PF_DUPTO) { 5252 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5253 if (s) 5254 PF_STATE_UNLOCK(s); 5255 return; 5256 } 5257 } else { 5258 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5259 if (s) 5260 PF_STATE_UNLOCK(s); 5261 return; 5262 } 5263 m0 = *m; 5264 } 5265 5266 ip = mtod(m0, struct ip *); 5267 5268 bzero(&dst, sizeof(dst)); 5269 dst.sin_family = AF_INET; 5270 dst.sin_len = sizeof(dst); 5271 dst.sin_addr = ip->ip_dst; 5272 5273 if (r->rt == PF_FASTROUTE) { 5274 struct rtentry *rt; 5275 5276 if (s) 5277 PF_STATE_UNLOCK(s); 5278 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0)); 5279 if (rt == NULL) { 5280 KMOD_IPSTAT_INC(ips_noroute); 5281 error = EHOSTUNREACH; 5282 goto bad; 5283 } 5284 5285 ifp = rt->rt_ifp; 5286 counter_u64_add(rt->rt_pksent, 1); 5287 5288 if (rt->rt_flags & RTF_GATEWAY) 5289 bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst)); 5290 RTFREE_LOCKED(rt); 5291 } else { 5292 if (TAILQ_EMPTY(&r->rpool.list)) { 5293 DPFPRINTF(PF_DEBUG_URGENT, 5294 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5295 goto bad_locked; 5296 } 5297 if (s == NULL) { 5298 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 5299 &naddr, NULL, &sn); 5300 if (!PF_AZERO(&naddr, AF_INET)) 5301 dst.sin_addr.s_addr = naddr.v4.s_addr; 5302 ifp = r->rpool.cur->kif ? 5303 r->rpool.cur->kif->pfik_ifp : NULL; 5304 } else { 5305 if (!PF_AZERO(&s->rt_addr, AF_INET)) 5306 dst.sin_addr.s_addr = 5307 s->rt_addr.v4.s_addr; 5308 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5309 PF_STATE_UNLOCK(s); 5310 } 5311 } 5312 if (ifp == NULL) 5313 goto bad; 5314 5315 if (oifp != ifp) { 5316 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS) 5317 goto bad; 5318 else if (m0 == NULL) 5319 goto done; 5320 if (m0->m_len < sizeof(struct ip)) { 5321 DPFPRINTF(PF_DEBUG_URGENT, 5322 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 5323 goto bad; 5324 } 5325 ip = mtod(m0, struct ip *); 5326 } 5327 5328 if (ifp->if_flags & IFF_LOOPBACK) 5329 m0->m_flags |= M_SKIP_FIREWALL; 5330 5331 ip_len = ntohs(ip->ip_len); 5332 ip_off = ntohs(ip->ip_off); 5333 5334 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 5335 m0->m_pkthdr.csum_flags |= CSUM_IP; 5336 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 5337 in_delayed_cksum(m0); 5338 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 5339 } 5340#ifdef SCTP 5341 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 5342 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 5343 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 5344 } 5345#endif 5346 5347 /* 5348 * If small enough for interface, or the interface will take 5349 * care of the fragmentation for us, we can just send directly. 5350 */ 5351 if (ip_len <= ifp->if_mtu || 5352 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 5353 ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 5354 ip->ip_sum = 0; 5355 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 5356 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 5357 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 5358 } 5359 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 5360 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5361 goto done; 5362 } 5363 5364 /* Balk when DF bit is set or the interface didn't support TSO. */ 5365 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 5366 error = EMSGSIZE; 5367 KMOD_IPSTAT_INC(ips_cantfrag); 5368 if (r->rt != PF_DUPTO) { 5369 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 5370 ifp->if_mtu); 5371 goto done; 5372 } else 5373 goto bad; 5374 } 5375 5376 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 5377 if (error) 5378 goto bad; 5379 5380 for (; m0; m0 = m1) { 5381 m1 = m0->m_nextpkt; 5382 m0->m_nextpkt = NULL; 5383 if (error == 0) { 5384 m_clrprotoflags(m0); 5385 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5386 } else 5387 m_freem(m0); 5388 } 5389 5390 if (error == 0) 5391 KMOD_IPSTAT_INC(ips_fragmented); 5392 5393done: 5394 if (r->rt != PF_DUPTO) 5395 *m = NULL; 5396 return; 5397 5398bad_locked: 5399 if (s) 5400 PF_STATE_UNLOCK(s); 5401bad: 5402 m_freem(m0); 5403 goto done; 5404} 5405#endif /* INET */ 5406 5407#ifdef INET6 5408static void 5409pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, 5410 struct pf_state *s, struct pf_pdesc *pd) 5411{ 5412 struct mbuf *m0; 5413 struct sockaddr_in6 dst; 5414 struct ip6_hdr *ip6; 5415 struct ifnet *ifp = NULL; 5416 struct pf_addr naddr; 5417 struct pf_src_node *sn = NULL; 5418 5419 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5420 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5421 __func__)); 5422 5423 if ((pd->pf_mtag == NULL && 5424 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5425 pd->pf_mtag->routed++ > 3) { 5426 m0 = *m; 5427 *m = NULL; 5428 goto bad_locked; 5429 } 5430 5431 if (r->rt == PF_DUPTO) { 5432 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { 5433 if (s) 5434 PF_STATE_UNLOCK(s); 5435 return; 5436 } 5437 } else { 5438 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5439 if (s) 5440 PF_STATE_UNLOCK(s); 5441 return; 5442 } 5443 m0 = *m; 5444 } 5445 5446 ip6 = mtod(m0, struct ip6_hdr *); 5447 5448 bzero(&dst, sizeof(dst)); 5449 dst.sin6_family = AF_INET6; 5450 dst.sin6_len = sizeof(dst); 5451 dst.sin6_addr = ip6->ip6_dst; 5452 5453 /* Cheat. XXX why only in the v6 case??? */ 5454 if (r->rt == PF_FASTROUTE) { 5455 if (s) 5456 PF_STATE_UNLOCK(s); 5457 m0->m_flags |= M_SKIP_FIREWALL; 5458 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); 5459 return; 5460 } 5461 5462 if (TAILQ_EMPTY(&r->rpool.list)) { 5463 DPFPRINTF(PF_DEBUG_URGENT, 5464 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5465 goto bad_locked; 5466 } 5467 if (s == NULL) { 5468 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 5469 &naddr, NULL, &sn); 5470 if (!PF_AZERO(&naddr, AF_INET6)) 5471 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5472 &naddr, AF_INET6); 5473 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 5474 } else { 5475 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 5476 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 5477 &s->rt_addr, AF_INET6); 5478 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5479 } 5480 5481 if (s) 5482 PF_STATE_UNLOCK(s); 5483 5484 if (ifp == NULL) 5485 goto bad; 5486 5487 if (oifp != ifp) { 5488 if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS) 5489 goto bad; 5490 else if (m0 == NULL) 5491 goto done; 5492 if (m0->m_len < sizeof(struct ip6_hdr)) { 5493 DPFPRINTF(PF_DEBUG_URGENT, 5494 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 5495 __func__)); 5496 goto bad; 5497 } 5498 ip6 = mtod(m0, struct ip6_hdr *); 5499 } 5500 5501 if (ifp->if_flags & IFF_LOOPBACK) 5502 m0->m_flags |= M_SKIP_FIREWALL; 5503 5504 /* 5505 * If the packet is too large for the outgoing interface, 5506 * send back an icmp6 error. 5507 */ 5508 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 5509 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 5510 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) 5511 nd6_output(ifp, ifp, m0, &dst, NULL); 5512 else { 5513 in6_ifstat_inc(ifp, ifs6_in_toobig); 5514 if (r->rt != PF_DUPTO) 5515 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 5516 else 5517 goto bad; 5518 } 5519 5520done: 5521 if (r->rt != PF_DUPTO) 5522 *m = NULL; 5523 return; 5524 5525bad_locked: 5526 if (s) 5527 PF_STATE_UNLOCK(s); 5528bad: 5529 m_freem(m0); 5530 goto done; 5531} 5532#endif /* INET6 */ 5533 5534/* 5535 * FreeBSD supports cksum offloads for the following drivers. 5536 * em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4), 5537 * ti(4), txp(4), xl(4) 5538 * 5539 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 5540 * network driver performed cksum including pseudo header, need to verify 5541 * csum_data 5542 * CSUM_DATA_VALID : 5543 * network driver performed cksum, needs to additional pseudo header 5544 * cksum computation with partial csum_data(i.e. lack of H/W support for 5545 * pseudo header, for instance hme(4), sk(4) and possibly gem(4)) 5546 * 5547 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 5548 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 5549 * TCP/UDP layer. 5550 * Also, set csum_data to 0xffff to force cksum validation. 5551 */ 5552static int 5553pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 5554{ 5555 u_int16_t sum = 0; 5556 int hw_assist = 0; 5557 struct ip *ip; 5558 5559 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 5560 return (1); 5561 if (m->m_pkthdr.len < off + len) 5562 return (1); 5563 5564 switch (p) { 5565 case IPPROTO_TCP: 5566 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5567 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5568 sum = m->m_pkthdr.csum_data; 5569 } else { 5570 ip = mtod(m, struct ip *); 5571 sum = in_pseudo(ip->ip_src.s_addr, 5572 ip->ip_dst.s_addr, htonl((u_short)len + 5573 m->m_pkthdr.csum_data + IPPROTO_TCP)); 5574 } 5575 sum ^= 0xffff; 5576 ++hw_assist; 5577 } 5578 break; 5579 case IPPROTO_UDP: 5580 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 5581 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 5582 sum = m->m_pkthdr.csum_data; 5583 } else { 5584 ip = mtod(m, struct ip *); 5585 sum = in_pseudo(ip->ip_src.s_addr, 5586 ip->ip_dst.s_addr, htonl((u_short)len + 5587 m->m_pkthdr.csum_data + IPPROTO_UDP)); 5588 } 5589 sum ^= 0xffff; 5590 ++hw_assist; 5591 } 5592 break; 5593 case IPPROTO_ICMP: 5594#ifdef INET6 5595 case IPPROTO_ICMPV6: 5596#endif /* INET6 */ 5597 break; 5598 default: 5599 return (1); 5600 } 5601 5602 if (!hw_assist) { 5603 switch (af) { 5604 case AF_INET: 5605 if (p == IPPROTO_ICMP) { 5606 if (m->m_len < off) 5607 return (1); 5608 m->m_data += off; 5609 m->m_len -= off; 5610 sum = in_cksum(m, len); 5611 m->m_data -= off; 5612 m->m_len += off; 5613 } else { 5614 if (m->m_len < sizeof(struct ip)) 5615 return (1); 5616 sum = in4_cksum(m, p, off, len); 5617 } 5618 break; 5619#ifdef INET6 5620 case AF_INET6: 5621 if (m->m_len < sizeof(struct ip6_hdr)) 5622 return (1); 5623 sum = in6_cksum(m, p, off, len); 5624 break; 5625#endif /* INET6 */ 5626 default: 5627 return (1); 5628 } 5629 } 5630 if (sum) { 5631 switch (p) { 5632 case IPPROTO_TCP: 5633 { 5634 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 5635 break; 5636 } 5637 case IPPROTO_UDP: 5638 { 5639 KMOD_UDPSTAT_INC(udps_badsum); 5640 break; 5641 } 5642#ifdef INET 5643 case IPPROTO_ICMP: 5644 { 5645 KMOD_ICMPSTAT_INC(icps_checksum); 5646 break; 5647 } 5648#endif 5649#ifdef INET6 5650 case IPPROTO_ICMPV6: 5651 { 5652 KMOD_ICMP6STAT_INC(icp6s_checksum); 5653 break; 5654 } 5655#endif /* INET6 */ 5656 } 5657 return (1); 5658 } else { 5659 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 5660 m->m_pkthdr.csum_flags |= 5661 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 5662 m->m_pkthdr.csum_data = 0xffff; 5663 } 5664 } 5665 return (0); 5666} 5667 5668 5669#ifdef INET 5670int 5671pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 5672{ 5673 struct pfi_kif *kif; 5674 u_short action, reason = 0, log = 0; 5675 struct mbuf *m = *m0; 5676 struct ip *h = NULL; 5677 struct m_tag *ipfwtag; 5678 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 5679 struct pf_state *s = NULL; 5680 struct pf_ruleset *ruleset = NULL; 5681 struct pf_pdesc pd; 5682 int off, dirndx, pqid = 0; 5683 5684 M_ASSERTPKTHDR(m); 5685 5686 if (!V_pf_status.running) 5687 return (PF_PASS); 5688 5689 memset(&pd, 0, sizeof(pd)); 5690 5691 kif = (struct pfi_kif *)ifp->if_pf_kif; 5692 5693 if (kif == NULL) { 5694 DPFPRINTF(PF_DEBUG_URGENT, 5695 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 5696 return (PF_DROP); 5697 } 5698 if (kif->pfik_flags & PFI_IFLAG_SKIP) 5699 return (PF_PASS); 5700 5701 if (m->m_flags & M_SKIP_FIREWALL) 5702 return (PF_PASS); 5703 5704 pd.pf_mtag = pf_find_mtag(m); 5705 5706 PF_RULES_RLOCK(); 5707 5708 if (ip_divert_ptr != NULL && 5709 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 5710 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 5711 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { 5712 if (pd.pf_mtag == NULL && 5713 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5714 action = PF_DROP; 5715 goto done; 5716 } 5717 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 5718 m_tag_delete(m, ipfwtag); 5719 } 5720 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 5721 m->m_flags |= M_FASTFWD_OURS; 5722 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 5723 } 5724 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 5725 /* We do IP header normalization and packet reassembly here */ 5726 action = PF_DROP; 5727 goto done; 5728 } 5729 m = *m0; /* pf_normalize messes with m0 */ 5730 h = mtod(m, struct ip *); 5731 5732 off = h->ip_hl << 2; 5733 if (off < (int)sizeof(struct ip)) { 5734 action = PF_DROP; 5735 REASON_SET(&reason, PFRES_SHORT); 5736 log = 1; 5737 goto done; 5738 } 5739 5740 pd.src = (struct pf_addr *)&h->ip_src; 5741 pd.dst = (struct pf_addr *)&h->ip_dst; 5742 pd.sport = pd.dport = NULL; 5743 pd.ip_sum = &h->ip_sum; 5744 pd.proto_sum = NULL; 5745 pd.proto = h->ip_p; 5746 pd.dir = dir; 5747 pd.sidx = (dir == PF_IN) ? 0 : 1; 5748 pd.didx = (dir == PF_IN) ? 1 : 0; 5749 pd.af = AF_INET; 5750 pd.tos = h->ip_tos; 5751 pd.tot_len = ntohs(h->ip_len); 5752 5753 /* handle fragments that didn't get reassembled by normalization */ 5754 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 5755 action = pf_test_fragment(&r, dir, kif, m, h, 5756 &pd, &a, &ruleset); 5757 goto done; 5758 } 5759 5760 switch (h->ip_p) { 5761 5762 case IPPROTO_TCP: { 5763 struct tcphdr th; 5764 5765 pd.hdr.tcp = &th; 5766 if (!pf_pull_hdr(m, off, &th, sizeof(th), 5767 &action, &reason, AF_INET)) { 5768 log = action != PF_PASS; 5769 goto done; 5770 } 5771 pd.p_len = pd.tot_len - off - (th.th_off << 2); 5772 if ((th.th_flags & TH_ACK) && pd.p_len == 0) 5773 pqid = 1; 5774 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 5775 if (action == PF_DROP) 5776 goto done; 5777 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 5778 &reason); 5779 if (action == PF_PASS) { 5780 if (pfsync_update_state_ptr != NULL) 5781 pfsync_update_state_ptr(s); 5782 r = s->rule.ptr; 5783 a = s->anchor.ptr; 5784 log = s->log; 5785 } else if (s == NULL) 5786 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5787 &a, &ruleset, inp); 5788 break; 5789 } 5790 5791 case IPPROTO_UDP: { 5792 struct udphdr uh; 5793 5794 pd.hdr.udp = &uh; 5795 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 5796 &action, &reason, AF_INET)) { 5797 log = action != PF_PASS; 5798 goto done; 5799 } 5800 if (uh.uh_dport == 0 || 5801 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 5802 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 5803 action = PF_DROP; 5804 REASON_SET(&reason, PFRES_SHORT); 5805 goto done; 5806 } 5807 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 5808 if (action == PF_PASS) { 5809 if (pfsync_update_state_ptr != NULL) 5810 pfsync_update_state_ptr(s); 5811 r = s->rule.ptr; 5812 a = s->anchor.ptr; 5813 log = s->log; 5814 } else if (s == NULL) 5815 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5816 &a, &ruleset, inp); 5817 break; 5818 } 5819 5820 case IPPROTO_ICMP: { 5821 struct icmp ih; 5822 5823 pd.hdr.icmp = &ih; 5824 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, 5825 &action, &reason, AF_INET)) { 5826 log = action != PF_PASS; 5827 goto done; 5828 } 5829 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 5830 &reason); 5831 if (action == PF_PASS) { 5832 if (pfsync_update_state_ptr != NULL) 5833 pfsync_update_state_ptr(s); 5834 r = s->rule.ptr; 5835 a = s->anchor.ptr; 5836 log = s->log; 5837 } else if (s == NULL) 5838 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5839 &a, &ruleset, inp); 5840 break; 5841 } 5842 5843#ifdef INET6 5844 case IPPROTO_ICMPV6: { 5845 action = PF_DROP; 5846 DPFPRINTF(PF_DEBUG_MISC, 5847 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 5848 goto done; 5849 } 5850#endif 5851 5852 default: 5853 action = pf_test_state_other(&s, dir, kif, m, &pd); 5854 if (action == PF_PASS) { 5855 if (pfsync_update_state_ptr != NULL) 5856 pfsync_update_state_ptr(s); 5857 r = s->rule.ptr; 5858 a = s->anchor.ptr; 5859 log = s->log; 5860 } else if (s == NULL) 5861 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 5862 &a, &ruleset, inp); 5863 break; 5864 } 5865 5866done: 5867 PF_RULES_RUNLOCK(); 5868 if (action == PF_PASS && h->ip_hl > 5 && 5869 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 5870 action = PF_DROP; 5871 REASON_SET(&reason, PFRES_IPOPTIONS); 5872 log = 1; 5873 DPFPRINTF(PF_DEBUG_MISC, 5874 ("pf: dropping packet with ip options\n")); 5875 } 5876 5877 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 5878 action = PF_DROP; 5879 REASON_SET(&reason, PFRES_MEMORY); 5880 } 5881 if (r->rtableid >= 0) 5882 M_SETFIB(m, r->rtableid); 5883 5884#ifdef ALTQ 5885 if (action == PF_PASS && r->qid) { 5886 if (pd.pf_mtag == NULL && 5887 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5888 action = PF_DROP; 5889 REASON_SET(&reason, PFRES_MEMORY); 5890 } 5891 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 5892 pd.pf_mtag->qid = r->pqid; 5893 else 5894 pd.pf_mtag->qid = r->qid; 5895 /* add hints for ecn */ 5896 pd.pf_mtag->hdr = h; 5897 5898 } 5899#endif /* ALTQ */ 5900 5901 /* 5902 * connections redirected to loopback should not match sockets 5903 * bound specifically to loopback due to security implications, 5904 * see tcp_input() and in_pcblookup_listen(). 5905 */ 5906 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 5907 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 5908 (s->nat_rule.ptr->action == PF_RDR || 5909 s->nat_rule.ptr->action == PF_BINAT) && 5910 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) 5911 m->m_flags |= M_SKIP_FIREWALL; 5912 5913 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && 5914 !PACKET_LOOPED(&pd)) { 5915 5916 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 5917 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 5918 if (ipfwtag != NULL) { 5919 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 5920 ntohs(r->divert.port); 5921 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 5922 5923 if (s) 5924 PF_STATE_UNLOCK(s); 5925 5926 m_tag_prepend(m, ipfwtag); 5927 if (m->m_flags & M_FASTFWD_OURS) { 5928 if (pd.pf_mtag == NULL && 5929 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 5930 action = PF_DROP; 5931 REASON_SET(&reason, PFRES_MEMORY); 5932 log = 1; 5933 DPFPRINTF(PF_DEBUG_MISC, 5934 ("pf: failed to allocate tag\n")); 5935 } 5936 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT; 5937 m->m_flags &= ~M_FASTFWD_OURS; 5938 } 5939 ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT); 5940 *m0 = NULL; 5941 5942 return (action); 5943 } else { 5944 /* XXX: ipfw has the same behaviour! */ 5945 action = PF_DROP; 5946 REASON_SET(&reason, PFRES_MEMORY); 5947 log = 1; 5948 DPFPRINTF(PF_DEBUG_MISC, 5949 ("pf: failed to allocate divert tag\n")); 5950 } 5951 } 5952 5953 if (log) { 5954 struct pf_rule *lr; 5955 5956 if (s != NULL && s->nat_rule.ptr != NULL && 5957 s->nat_rule.ptr->log & PF_LOG_ALL) 5958 lr = s->nat_rule.ptr; 5959 else 5960 lr = r; 5961 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 5962 (s == NULL)); 5963 } 5964 5965 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 5966 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; 5967 5968 if (action == PF_PASS || r->action == PF_DROP) { 5969 dirndx = (dir == PF_OUT); 5970 r->packets[dirndx]++; 5971 r->bytes[dirndx] += pd.tot_len; 5972 if (a != NULL) { 5973 a->packets[dirndx]++; 5974 a->bytes[dirndx] += pd.tot_len; 5975 } 5976 if (s != NULL) { 5977 if (s->nat_rule.ptr != NULL) { 5978 s->nat_rule.ptr->packets[dirndx]++; 5979 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 5980 } 5981 if (s->src_node != NULL) { 5982 s->src_node->packets[dirndx]++; 5983 s->src_node->bytes[dirndx] += pd.tot_len; 5984 } 5985 if (s->nat_src_node != NULL) { 5986 s->nat_src_node->packets[dirndx]++; 5987 s->nat_src_node->bytes[dirndx] += pd.tot_len; 5988 } 5989 dirndx = (dir == s->direction) ? 0 : 1; 5990 s->packets[dirndx]++; 5991 s->bytes[dirndx] += pd.tot_len; 5992 } 5993 tr = r; 5994 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 5995 if (nr != NULL && r == &V_pf_default_rule) 5996 tr = nr; 5997 if (tr->src.addr.type == PF_ADDR_TABLE) 5998 pfr_update_stats(tr->src.addr.p.tbl, 5999 (s == NULL) ? pd.src : 6000 &s->key[(s->direction == PF_IN)]-> 6001 addr[(s->direction == PF_OUT)], 6002 pd.af, pd.tot_len, dir == PF_OUT, 6003 r->action == PF_PASS, tr->src.neg); 6004 if (tr->dst.addr.type == PF_ADDR_TABLE) 6005 pfr_update_stats(tr->dst.addr.p.tbl, 6006 (s == NULL) ? pd.dst : 6007 &s->key[(s->direction == PF_IN)]-> 6008 addr[(s->direction == PF_IN)], 6009 pd.af, pd.tot_len, dir == PF_OUT, 6010 r->action == PF_PASS, tr->dst.neg); 6011 } 6012 6013 switch (action) { 6014 case PF_SYNPROXY_DROP: 6015 m_freem(*m0); 6016 case PF_DEFER: 6017 *m0 = NULL; 6018 action = PF_PASS; 6019 break; 6020 default: 6021 /* pf_route() returns unlocked. */ 6022 if (r->rt) { 6023 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd); 6024 return (action); 6025 } 6026 break; 6027 } 6028 if (s) 6029 PF_STATE_UNLOCK(s); 6030 6031 return (action); 6032} 6033#endif /* INET */ 6034 6035#ifdef INET6 6036int 6037pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6038{ 6039 struct pfi_kif *kif; 6040 u_short action, reason = 0, log = 0; 6041 struct mbuf *m = *m0, *n = NULL; 6042 struct ip6_hdr *h = NULL; 6043 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6044 struct pf_state *s = NULL; 6045 struct pf_ruleset *ruleset = NULL; 6046 struct pf_pdesc pd; 6047 int off, terminal = 0, dirndx, rh_cnt = 0; 6048 6049 M_ASSERTPKTHDR(m); 6050 6051 if (!V_pf_status.running) 6052 return (PF_PASS); 6053 6054 memset(&pd, 0, sizeof(pd)); 6055 pd.pf_mtag = pf_find_mtag(m); 6056 6057 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) 6058 return (PF_PASS); 6059 6060 kif = (struct pfi_kif *)ifp->if_pf_kif; 6061 if (kif == NULL) { 6062 DPFPRINTF(PF_DEBUG_URGENT, 6063 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6064 return (PF_DROP); 6065 } 6066 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6067 return (PF_PASS); 6068 6069 PF_RULES_RLOCK(); 6070 6071 /* We do IP header normalization and packet reassembly here */ 6072 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6073 action = PF_DROP; 6074 goto done; 6075 } 6076 m = *m0; /* pf_normalize messes with m0 */ 6077 h = mtod(m, struct ip6_hdr *); 6078 6079#if 1 6080 /* 6081 * we do not support jumbogram yet. if we keep going, zero ip6_plen 6082 * will do something bad, so drop the packet for now. 6083 */ 6084 if (htons(h->ip6_plen) == 0) { 6085 action = PF_DROP; 6086 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6087 goto done; 6088 } 6089#endif 6090 6091 pd.src = (struct pf_addr *)&h->ip6_src; 6092 pd.dst = (struct pf_addr *)&h->ip6_dst; 6093 pd.sport = pd.dport = NULL; 6094 pd.ip_sum = NULL; 6095 pd.proto_sum = NULL; 6096 pd.dir = dir; 6097 pd.sidx = (dir == PF_IN) ? 0 : 1; 6098 pd.didx = (dir == PF_IN) ? 1 : 0; 6099 pd.af = AF_INET6; 6100 pd.tos = 0; 6101 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6102 6103 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6104 pd.proto = h->ip6_nxt; 6105 do { 6106 switch (pd.proto) { 6107 case IPPROTO_FRAGMENT: 6108 action = pf_test_fragment(&r, dir, kif, m, h, 6109 &pd, &a, &ruleset); 6110 if (action == PF_DROP) 6111 REASON_SET(&reason, PFRES_FRAG); 6112 goto done; 6113 case IPPROTO_ROUTING: { 6114 struct ip6_rthdr rthdr; 6115 6116 if (rh_cnt++) { 6117 DPFPRINTF(PF_DEBUG_MISC, 6118 ("pf: IPv6 more than one rthdr\n")); 6119 action = PF_DROP; 6120 REASON_SET(&reason, PFRES_IPOPTIONS); 6121 log = 1; 6122 goto done; 6123 } 6124 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6125 &reason, pd.af)) { 6126 DPFPRINTF(PF_DEBUG_MISC, 6127 ("pf: IPv6 short rthdr\n")); 6128 action = PF_DROP; 6129 REASON_SET(&reason, PFRES_SHORT); 6130 log = 1; 6131 goto done; 6132 } 6133 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6134 DPFPRINTF(PF_DEBUG_MISC, 6135 ("pf: IPv6 rthdr0\n")); 6136 action = PF_DROP; 6137 REASON_SET(&reason, PFRES_IPOPTIONS); 6138 log = 1; 6139 goto done; 6140 } 6141 /* FALLTHROUGH */ 6142 } 6143 case IPPROTO_AH: 6144 case IPPROTO_HOPOPTS: 6145 case IPPROTO_DSTOPTS: { 6146 /* get next header and header length */ 6147 struct ip6_ext opt6; 6148 6149 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6150 NULL, &reason, pd.af)) { 6151 DPFPRINTF(PF_DEBUG_MISC, 6152 ("pf: IPv6 short opt\n")); 6153 action = PF_DROP; 6154 log = 1; 6155 goto done; 6156 } 6157 if (pd.proto == IPPROTO_AH) 6158 off += (opt6.ip6e_len + 2) * 4; 6159 else 6160 off += (opt6.ip6e_len + 1) * 8; 6161 pd.proto = opt6.ip6e_nxt; 6162 /* goto the next header */ 6163 break; 6164 } 6165 default: 6166 terminal++; 6167 break; 6168 } 6169 } while (!terminal); 6170 6171 /* if there's no routing header, use unmodified mbuf for checksumming */ 6172 if (!n) 6173 n = m; 6174 6175 switch (pd.proto) { 6176 6177 case IPPROTO_TCP: { 6178 struct tcphdr th; 6179 6180 pd.hdr.tcp = &th; 6181 if (!pf_pull_hdr(m, off, &th, sizeof(th), 6182 &action, &reason, AF_INET6)) { 6183 log = action != PF_PASS; 6184 goto done; 6185 } 6186 pd.p_len = pd.tot_len - off - (th.th_off << 2); 6187 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6188 if (action == PF_DROP) 6189 goto done; 6190 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6191 &reason); 6192 if (action == PF_PASS) { 6193 if (pfsync_update_state_ptr != NULL) 6194 pfsync_update_state_ptr(s); 6195 r = s->rule.ptr; 6196 a = s->anchor.ptr; 6197 log = s->log; 6198 } else if (s == NULL) 6199 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6200 &a, &ruleset, inp); 6201 break; 6202 } 6203 6204 case IPPROTO_UDP: { 6205 struct udphdr uh; 6206 6207 pd.hdr.udp = &uh; 6208 if (!pf_pull_hdr(m, off, &uh, sizeof(uh), 6209 &action, &reason, AF_INET6)) { 6210 log = action != PF_PASS; 6211 goto done; 6212 } 6213 if (uh.uh_dport == 0 || 6214 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || 6215 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { 6216 action = PF_DROP; 6217 REASON_SET(&reason, PFRES_SHORT); 6218 goto done; 6219 } 6220 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6221 if (action == PF_PASS) { 6222 if (pfsync_update_state_ptr != NULL) 6223 pfsync_update_state_ptr(s); 6224 r = s->rule.ptr; 6225 a = s->anchor.ptr; 6226 log = s->log; 6227 } else if (s == NULL) 6228 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6229 &a, &ruleset, inp); 6230 break; 6231 } 6232 6233 case IPPROTO_ICMP: { 6234 action = PF_DROP; 6235 DPFPRINTF(PF_DEBUG_MISC, 6236 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 6237 goto done; 6238 } 6239 6240 case IPPROTO_ICMPV6: { 6241 struct icmp6_hdr ih; 6242 6243 pd.hdr.icmp6 = &ih; 6244 if (!pf_pull_hdr(m, off, &ih, sizeof(ih), 6245 &action, &reason, AF_INET6)) { 6246 log = action != PF_PASS; 6247 goto done; 6248 } 6249 action = pf_test_state_icmp(&s, dir, kif, 6250 m, off, h, &pd, &reason); 6251 if (action == PF_PASS) { 6252 if (pfsync_update_state_ptr != NULL) 6253 pfsync_update_state_ptr(s); 6254 r = s->rule.ptr; 6255 a = s->anchor.ptr; 6256 log = s->log; 6257 } else if (s == NULL) 6258 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6259 &a, &ruleset, inp); 6260 break; 6261 } 6262 6263 default: 6264 action = pf_test_state_other(&s, dir, kif, m, &pd); 6265 if (action == PF_PASS) { 6266 if (pfsync_update_state_ptr != NULL) 6267 pfsync_update_state_ptr(s); 6268 r = s->rule.ptr; 6269 a = s->anchor.ptr; 6270 log = s->log; 6271 } else if (s == NULL) 6272 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6273 &a, &ruleset, inp); 6274 break; 6275 } 6276 6277done: 6278 PF_RULES_RUNLOCK(); 6279 if (n != m) { 6280 m_freem(n); 6281 n = NULL; 6282 } 6283 6284 /* handle dangerous IPv6 extension headers. */ 6285 if (action == PF_PASS && rh_cnt && 6286 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6287 action = PF_DROP; 6288 REASON_SET(&reason, PFRES_IPOPTIONS); 6289 log = 1; 6290 DPFPRINTF(PF_DEBUG_MISC, 6291 ("pf: dropping packet with dangerous v6 headers\n")); 6292 } 6293 6294 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6295 action = PF_DROP; 6296 REASON_SET(&reason, PFRES_MEMORY); 6297 } 6298 if (r->rtableid >= 0) 6299 M_SETFIB(m, r->rtableid); 6300 6301#ifdef ALTQ 6302 if (action == PF_PASS && r->qid) { 6303 if (pd.pf_mtag == NULL && 6304 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6305 action = PF_DROP; 6306 REASON_SET(&reason, PFRES_MEMORY); 6307 } 6308 if (pd.tos & IPTOS_LOWDELAY) 6309 pd.pf_mtag->qid = r->pqid; 6310 else 6311 pd.pf_mtag->qid = r->qid; 6312 /* add hints for ecn */ 6313 pd.pf_mtag->hdr = h; 6314 } 6315#endif /* ALTQ */ 6316 6317 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6318 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6319 (s->nat_rule.ptr->action == PF_RDR || 6320 s->nat_rule.ptr->action == PF_BINAT) && 6321 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 6322 m->m_flags |= M_SKIP_FIREWALL; 6323 6324 /* XXX: Anybody working on it?! */ 6325 if (r->divert.port) 6326 printf("pf: divert(9) is not supported for IPv6\n"); 6327 6328 if (log) { 6329 struct pf_rule *lr; 6330 6331 if (s != NULL && s->nat_rule.ptr != NULL && 6332 s->nat_rule.ptr->log & PF_LOG_ALL) 6333 lr = s->nat_rule.ptr; 6334 else 6335 lr = r; 6336 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 6337 &pd, (s == NULL)); 6338 } 6339 6340 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; 6341 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; 6342 6343 if (action == PF_PASS || r->action == PF_DROP) { 6344 dirndx = (dir == PF_OUT); 6345 r->packets[dirndx]++; 6346 r->bytes[dirndx] += pd.tot_len; 6347 if (a != NULL) { 6348 a->packets[dirndx]++; 6349 a->bytes[dirndx] += pd.tot_len; 6350 } 6351 if (s != NULL) { 6352 if (s->nat_rule.ptr != NULL) { 6353 s->nat_rule.ptr->packets[dirndx]++; 6354 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; 6355 } 6356 if (s->src_node != NULL) { 6357 s->src_node->packets[dirndx]++; 6358 s->src_node->bytes[dirndx] += pd.tot_len; 6359 } 6360 if (s->nat_src_node != NULL) { 6361 s->nat_src_node->packets[dirndx]++; 6362 s->nat_src_node->bytes[dirndx] += pd.tot_len; 6363 } 6364 dirndx = (dir == s->direction) ? 0 : 1; 6365 s->packets[dirndx]++; 6366 s->bytes[dirndx] += pd.tot_len; 6367 } 6368 tr = r; 6369 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6370 if (nr != NULL && r == &V_pf_default_rule) 6371 tr = nr; 6372 if (tr->src.addr.type == PF_ADDR_TABLE) 6373 pfr_update_stats(tr->src.addr.p.tbl, 6374 (s == NULL) ? pd.src : 6375 &s->key[(s->direction == PF_IN)]->addr[0], 6376 pd.af, pd.tot_len, dir == PF_OUT, 6377 r->action == PF_PASS, tr->src.neg); 6378 if (tr->dst.addr.type == PF_ADDR_TABLE) 6379 pfr_update_stats(tr->dst.addr.p.tbl, 6380 (s == NULL) ? pd.dst : 6381 &s->key[(s->direction == PF_IN)]->addr[1], 6382 pd.af, pd.tot_len, dir == PF_OUT, 6383 r->action == PF_PASS, tr->dst.neg); 6384 } 6385 6386 switch (action) { 6387 case PF_SYNPROXY_DROP: 6388 m_freem(*m0); 6389 case PF_DEFER: 6390 *m0 = NULL; 6391 action = PF_PASS; 6392 break; 6393 default: 6394 /* pf_route6() returns unlocked. */ 6395 if (r->rt) { 6396 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd); 6397 return (action); 6398 } 6399 break; 6400 } 6401 6402 if (s) 6403 PF_STATE_UNLOCK(s); 6404 6405 return (action); 6406} 6407#endif /* INET6 */ 6408