1/*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40#include <sys/cdefs.h> 41#include "opt_bpf.h" 42#include "opt_inet.h" 43#include "opt_inet6.h" 44#include "opt_pf.h" 45#include "opt_sctp.h" 46 47#include <sys/param.h> 48#include <sys/bus.h> 49#include <sys/endian.h> 50#include <sys/gsb_crc32.h> 51#include <sys/hash.h> 52#include <sys/interrupt.h> 53#include <sys/kernel.h> 54#include <sys/kthread.h> 55#include <sys/limits.h> 56#include <sys/mbuf.h> 57#include <sys/md5.h> 58#include <sys/random.h> 59#include <sys/refcount.h> 60#include <sys/sdt.h> 61#include <sys/socket.h> 62#include <sys/sysctl.h> 63#include <sys/taskqueue.h> 64#include <sys/ucred.h> 65 66#include <net/if.h> 67#include <net/if_var.h> 68#include <net/if_private.h> 69#include <net/if_types.h> 70#include <net/if_vlan_var.h> 71#include <net/route.h> 72#include <net/route/nhop.h> 73#include <net/vnet.h> 74 75#include <net/pfil.h> 76#include <net/pfvar.h> 77#include <net/if_pflog.h> 78#include <net/if_pfsync.h> 79 80#include <netinet/in_pcb.h> 81#include <netinet/in_var.h> 82#include <netinet/in_fib.h> 83#include <netinet/ip.h> 84#include <netinet/ip_fw.h> 85#include <netinet/ip_icmp.h> 86#include <netinet/icmp_var.h> 87#include <netinet/ip_var.h> 88#include <netinet/tcp.h> 89#include <netinet/tcp_fsm.h> 90#include <netinet/tcp_seq.h> 91#include <netinet/tcp_timer.h> 92#include <netinet/tcp_var.h> 93#include <netinet/udp.h> 94#include <netinet/udp_var.h> 95 96/* dummynet */ 97#include <netinet/ip_dummynet.h> 98#include <netinet/ip_fw.h> 99#include <netpfil/ipfw/dn_heap.h> 100#include <netpfil/ipfw/ip_fw_private.h> 101#include <netpfil/ipfw/ip_dn_private.h> 102 103#ifdef INET6 104#include <netinet/ip6.h> 105#include <netinet/icmp6.h> 106#include <netinet6/nd6.h> 107#include <netinet6/ip6_var.h> 108#include <netinet6/in6_pcb.h> 109#include <netinet6/in6_fib.h> 110#include <netinet6/scope6_var.h> 111#endif /* INET6 */ 112 113#include <netinet/sctp_header.h> 114#include <netinet/sctp_crc32.h> 115 116#include <machine/in_cksum.h> 117#include <security/mac/mac_framework.h> 118 119#define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 120 121SDT_PROVIDER_DEFINE(pf); 122SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 123 "struct pf_kstate *"); 124SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 127 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 128 "struct pf_kstate *"); 129SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *", 130 "struct pfi_kkif *"); 131SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *", 132 "struct pf_krule *", "struct mbuf *", "int"); 133SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t", 134 "struct pf_sctp_source *"); 135SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t", 136 "struct pf_kstate *", "struct pf_sctp_source *"); 137 138SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *", 139 "struct mbuf *"); 140SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *"); 141SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch, 142 "int", "struct pf_keth_rule *", "char *"); 143SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *"); 144SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match, 145 "int", "struct pf_keth_rule *"); 146SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t"); 147 148/* 149 * Global variables 150 */ 151 152/* state tables */ 153VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 154VNET_DEFINE(struct pf_kpalist, pf_pabuf); 155VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 156VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 157VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 158VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 159VNET_DEFINE(struct pf_kstatus, pf_status); 160 161VNET_DEFINE(u_int32_t, ticket_altqs_active); 162VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 163VNET_DEFINE(int, altqs_inactive_open); 164VNET_DEFINE(u_int32_t, ticket_pabuf); 165 166VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 167#define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 168VNET_DEFINE(u_char, pf_tcp_secret[16]); 169#define V_pf_tcp_secret VNET(pf_tcp_secret) 170VNET_DEFINE(int, pf_tcp_secret_init); 171#define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 172VNET_DEFINE(int, pf_tcp_iss_off); 173#define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 174VNET_DECLARE(int, pf_vnet_active); 175#define V_pf_vnet_active VNET(pf_vnet_active) 176 177VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 178#define V_pf_purge_idx VNET(pf_purge_idx) 179 180#ifdef PF_WANT_32_TO_64_COUNTER 181VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 182#define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 183 184VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 185VNET_DEFINE(size_t, pf_allrulecount); 186VNET_DEFINE(struct pf_krule *, pf_rulemarker); 187#endif 188 189struct pf_sctp_endpoint; 190RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint); 191struct pf_sctp_source { 192 sa_family_t af; 193 struct pf_addr addr; 194 TAILQ_ENTRY(pf_sctp_source) entry; 195}; 196TAILQ_HEAD(pf_sctp_sources, pf_sctp_source); 197struct pf_sctp_endpoint 198{ 199 uint32_t v_tag; 200 struct pf_sctp_sources sources; 201 RB_ENTRY(pf_sctp_endpoint) entry; 202}; 203static int 204pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b) 205{ 206 return (a->v_tag - b->v_tag); 207} 208RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 209RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare); 210VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints); 211#define V_pf_sctp_endpoints VNET(pf_sctp_endpoints) 212static struct mtx_padalign pf_sctp_endpoints_mtx; 213MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF); 214#define PF_SCTP_ENDPOINTS_LOCK() mtx_lock(&pf_sctp_endpoints_mtx) 215#define PF_SCTP_ENDPOINTS_UNLOCK() mtx_unlock(&pf_sctp_endpoints_mtx) 216 217/* 218 * Queue for pf_intr() sends. 219 */ 220static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 221struct pf_send_entry { 222 STAILQ_ENTRY(pf_send_entry) pfse_next; 223 struct mbuf *pfse_m; 224 enum { 225 PFSE_IP, 226 PFSE_IP6, 227 PFSE_ICMP, 228 PFSE_ICMP6, 229 } pfse_type; 230 struct { 231 int type; 232 int code; 233 int mtu; 234 } icmpopts; 235}; 236 237STAILQ_HEAD(pf_send_head, pf_send_entry); 238VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 239#define V_pf_sendqueue VNET(pf_sendqueue) 240 241static struct mtx_padalign pf_sendqueue_mtx; 242MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 243#define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 244#define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 245 246/* 247 * Queue for pf_overload_task() tasks. 248 */ 249struct pf_overload_entry { 250 SLIST_ENTRY(pf_overload_entry) next; 251 struct pf_addr addr; 252 sa_family_t af; 253 uint8_t dir; 254 struct pf_krule *rule; 255}; 256 257SLIST_HEAD(pf_overload_head, pf_overload_entry); 258VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 259#define V_pf_overloadqueue VNET(pf_overloadqueue) 260VNET_DEFINE_STATIC(struct task, pf_overloadtask); 261#define V_pf_overloadtask VNET(pf_overloadtask) 262 263static struct mtx_padalign pf_overloadqueue_mtx; 264MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 265 "pf overload/flush queue", MTX_DEF); 266#define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 267#define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 268 269VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 270struct mtx_padalign pf_unlnkdrules_mtx; 271MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 272 MTX_DEF); 273 274struct sx pf_config_lock; 275SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config"); 276 277struct mtx_padalign pf_table_stats_lock; 278MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 279 MTX_DEF); 280 281VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 282#define V_pf_sources_z VNET(pf_sources_z) 283uma_zone_t pf_mtag_z; 284VNET_DEFINE(uma_zone_t, pf_state_z); 285VNET_DEFINE(uma_zone_t, pf_state_key_z); 286 287VNET_DEFINE(struct unrhdr64, pf_stateid); 288 289static void pf_src_tree_remove_state(struct pf_kstate *); 290static void pf_init_threshold(struct pf_threshold *, u_int32_t, 291 u_int32_t); 292static void pf_add_threshold(struct pf_threshold *); 293static int pf_check_threshold(struct pf_threshold *); 294 295static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 296 u_int16_t *, u_int16_t *, struct pf_addr *, 297 u_int16_t, u_int8_t, sa_family_t); 298static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 299 struct tcphdr *, struct pf_state_peer *); 300static void pf_change_icmp(struct pf_addr *, u_int16_t *, 301 struct pf_addr *, struct pf_addr *, u_int16_t, 302 u_int16_t *, u_int16_t *, u_int16_t *, 303 u_int16_t *, u_int8_t, sa_family_t); 304static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 305 sa_family_t, struct pf_krule *, int); 306static void pf_detach_state(struct pf_kstate *); 307static int pf_state_key_attach(struct pf_state_key *, 308 struct pf_state_key *, struct pf_kstate *); 309static void pf_state_key_detach(struct pf_kstate *, int); 310static int pf_state_key_ctor(void *, int, void *, int); 311static u_int32_t pf_tcp_iss(struct pf_pdesc *); 312static __inline void pf_dummynet_flag_remove(struct mbuf *m, 313 struct pf_mtag *pf_mtag); 314static int pf_dummynet(struct pf_pdesc *, struct pf_kstate *, 315 struct pf_krule *, struct mbuf **); 316static int pf_dummynet_route(struct pf_pdesc *, 317 struct pf_kstate *, struct pf_krule *, 318 struct ifnet *, struct sockaddr *, struct mbuf **); 319static int pf_test_eth_rule(int, struct pfi_kkif *, 320 struct mbuf **); 321static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 322 struct pfi_kkif *, struct mbuf *, int, 323 struct pf_pdesc *, struct pf_krule **, 324 struct pf_kruleset **, struct inpcb *); 325static int pf_create_state(struct pf_krule *, struct pf_krule *, 326 struct pf_krule *, struct pf_pdesc *, 327 struct pf_ksrc_node *, struct pf_state_key *, 328 struct pf_state_key *, struct mbuf *, int, 329 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 330 struct pf_kstate **, int, u_int16_t, u_int16_t, 331 int, struct pf_krule_slist *); 332static int pf_test_fragment(struct pf_krule **, struct pfi_kkif *, 333 struct mbuf *, void *, struct pf_pdesc *, 334 struct pf_krule **, struct pf_kruleset **); 335static int pf_tcp_track_full(struct pf_kstate **, 336 struct pfi_kkif *, struct mbuf *, int, 337 struct pf_pdesc *, u_short *, int *); 338static int pf_tcp_track_sloppy(struct pf_kstate **, 339 struct pf_pdesc *, u_short *); 340static int pf_test_state_tcp(struct pf_kstate **, 341 struct pfi_kkif *, struct mbuf *, int, 342 void *, struct pf_pdesc *, u_short *); 343static int pf_test_state_udp(struct pf_kstate **, 344 struct pfi_kkif *, struct mbuf *, int, 345 void *, struct pf_pdesc *); 346static int pf_test_state_icmp(struct pf_kstate **, 347 struct pfi_kkif *, struct mbuf *, int, 348 void *, struct pf_pdesc *, u_short *); 349static void pf_sctp_multihome_detach_addr(const struct pf_kstate *); 350static void pf_sctp_multihome_delayed(struct pf_pdesc *, int, 351 struct pfi_kkif *, struct pf_kstate *, int); 352static int pf_test_state_sctp(struct pf_kstate **, 353 struct pfi_kkif *, struct mbuf *, int, 354 void *, struct pf_pdesc *, u_short *); 355static int pf_test_state_other(struct pf_kstate **, 356 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 357static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 358 int, u_int16_t); 359static int pf_check_proto_cksum(struct mbuf *, int, int, 360 u_int8_t, sa_family_t); 361static void pf_print_state_parts(struct pf_kstate *, 362 struct pf_state_key *, struct pf_state_key *); 363static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 364 bool, u_int8_t); 365static struct pf_kstate *pf_find_state(struct pfi_kkif *, 366 struct pf_state_key_cmp *, u_int); 367static int pf_src_connlimit(struct pf_kstate **); 368static void pf_overload_task(void *v, int pending); 369static u_short pf_insert_src_node(struct pf_ksrc_node **, 370 struct pf_krule *, struct pf_addr *, sa_family_t); 371static u_int pf_purge_expired_states(u_int, int); 372static void pf_purge_unlinked_rules(void); 373static int pf_mtag_uminit(void *, int, int); 374static void pf_mtag_free(struct m_tag *); 375static void pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *, 376 int, struct pf_state_key *); 377#ifdef INET 378static void pf_route(struct mbuf **, struct pf_krule *, 379 struct ifnet *, struct pf_kstate *, 380 struct pf_pdesc *, struct inpcb *); 381#endif /* INET */ 382#ifdef INET6 383static void pf_change_a6(struct pf_addr *, u_int16_t *, 384 struct pf_addr *, u_int8_t); 385static void pf_route6(struct mbuf **, struct pf_krule *, 386 struct ifnet *, struct pf_kstate *, 387 struct pf_pdesc *, struct inpcb *); 388#endif /* INET6 */ 389static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 390 391int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 392 393extern int pf_end_threads; 394extern struct proc *pf_purge_proc; 395 396VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 397 398#define PACKET_UNDO_NAT(_m, _pd, _off, _s) \ 399 do { \ 400 struct pf_state_key *nk; \ 401 if ((pd->dir) == PF_OUT) \ 402 nk = (_s)->key[PF_SK_STACK]; \ 403 else \ 404 nk = (_s)->key[PF_SK_WIRE]; \ 405 pf_packet_rework_nat(_m, _pd, _off, nk); \ 406 } while (0) 407 408#define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 409 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED) 410 411#define STATE_LOOKUP(i, k, s, pd) \ 412 do { \ 413 (s) = pf_find_state((i), (k), (pd->dir)); \ 414 SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s)); \ 415 if ((s) == NULL) \ 416 return (PF_DROP); \ 417 if (PACKET_LOOPED(pd)) \ 418 return (PF_PASS); \ 419 } while (0) 420 421static struct pfi_kkif * 422BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k) 423{ 424 SDT_PROBE2(pf, ip, , bound_iface, st, k); 425 426 /* Floating unless otherwise specified. */ 427 if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND)) 428 return (V_pfi_all); 429 430 /* 431 * Initially set to all, because we don't know what interface we'll be 432 * sending this out when we create the state. 433 */ 434 if (st->rule.ptr->rt == PF_REPLYTO) 435 return (V_pfi_all); 436 437 /* Don't overrule the interface for states created on incoming packets. */ 438 if (st->direction == PF_IN) 439 return (k); 440 441 /* No route-to, so don't overrule. */ 442 if (st->rt != PF_ROUTETO) 443 return (k); 444 445 /* Bind to the route-to interface. */ 446 return (st->rt_kif); 447} 448 449#define STATE_INC_COUNTERS(s) \ 450 do { \ 451 struct pf_krule_item *mrm; \ 452 counter_u64_add(s->rule.ptr->states_cur, 1); \ 453 counter_u64_add(s->rule.ptr->states_tot, 1); \ 454 if (s->anchor.ptr != NULL) { \ 455 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 456 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 457 } \ 458 if (s->nat_rule.ptr != NULL) { \ 459 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 460 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 461 } \ 462 SLIST_FOREACH(mrm, &s->match_rules, entry) { \ 463 counter_u64_add(mrm->r->states_cur, 1); \ 464 counter_u64_add(mrm->r->states_tot, 1); \ 465 } \ 466 } while (0) 467 468#define STATE_DEC_COUNTERS(s) \ 469 do { \ 470 struct pf_krule_item *mrm; \ 471 if (s->nat_rule.ptr != NULL) \ 472 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 473 if (s->anchor.ptr != NULL) \ 474 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 475 counter_u64_add(s->rule.ptr->states_cur, -1); \ 476 SLIST_FOREACH(mrm, &s->match_rules, entry) \ 477 counter_u64_add(mrm->r->states_cur, -1); \ 478 } while (0) 479 480MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 481MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items"); 482VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 483VNET_DEFINE(struct pf_idhash *, pf_idhash); 484VNET_DEFINE(struct pf_srchash *, pf_srchash); 485 486SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 487 "pf(4)"); 488 489u_long pf_hashmask; 490u_long pf_srchashmask; 491static u_long pf_hashsize; 492static u_long pf_srchashsize; 493u_long pf_ioctl_maxcount = 65535; 494 495SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 496 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 497SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 498 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 499SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 500 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 501 502VNET_DEFINE(void *, pf_swi_cookie); 503VNET_DEFINE(struct intr_event *, pf_swi_ie); 504 505VNET_DEFINE(uint32_t, pf_hashseed); 506#define V_pf_hashseed VNET(pf_hashseed) 507 508static void 509pf_sctp_checksum(struct mbuf *m, int off) 510{ 511 uint32_t sum = 0; 512 513 /* Zero out the checksum, to enable recalculation. */ 514 m_copyback(m, off + offsetof(struct sctphdr, checksum), 515 sizeof(sum), (caddr_t)&sum); 516 517 sum = sctp_calculate_cksum(m, off); 518 519 m_copyback(m, off + offsetof(struct sctphdr, checksum), 520 sizeof(sum), (caddr_t)&sum); 521} 522 523int 524pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 525{ 526 527 switch (af) { 528#ifdef INET 529 case AF_INET: 530 if (a->addr32[0] > b->addr32[0]) 531 return (1); 532 if (a->addr32[0] < b->addr32[0]) 533 return (-1); 534 break; 535#endif /* INET */ 536#ifdef INET6 537 case AF_INET6: 538 if (a->addr32[3] > b->addr32[3]) 539 return (1); 540 if (a->addr32[3] < b->addr32[3]) 541 return (-1); 542 if (a->addr32[2] > b->addr32[2]) 543 return (1); 544 if (a->addr32[2] < b->addr32[2]) 545 return (-1); 546 if (a->addr32[1] > b->addr32[1]) 547 return (1); 548 if (a->addr32[1] < b->addr32[1]) 549 return (-1); 550 if (a->addr32[0] > b->addr32[0]) 551 return (1); 552 if (a->addr32[0] < b->addr32[0]) 553 return (-1); 554 break; 555#endif /* INET6 */ 556 default: 557 panic("%s: unknown address family %u", __func__, af); 558 } 559 return (0); 560} 561 562static void 563pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off, 564 struct pf_state_key *nk) 565{ 566 567 switch (pd->proto) { 568 case IPPROTO_TCP: { 569 struct tcphdr *th = &pd->hdr.tcp; 570 571 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 572 pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, 573 &th->th_sum, &nk->addr[pd->sidx], 574 nk->port[pd->sidx], 0, pd->af); 575 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 576 pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, 577 &th->th_sum, &nk->addr[pd->didx], 578 nk->port[pd->didx], 0, pd->af); 579 m_copyback(m, off, sizeof(*th), (caddr_t)th); 580 break; 581 } 582 case IPPROTO_UDP: { 583 struct udphdr *uh = &pd->hdr.udp; 584 585 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) 586 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 587 &uh->uh_sum, &nk->addr[pd->sidx], 588 nk->port[pd->sidx], 1, pd->af); 589 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) 590 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 591 &uh->uh_sum, &nk->addr[pd->didx], 592 nk->port[pd->didx], 1, pd->af); 593 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 594 break; 595 } 596 case IPPROTO_SCTP: { 597 struct sctphdr *sh = &pd->hdr.sctp; 598 uint16_t checksum = 0; 599 600 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 601 pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum, 602 &checksum, &nk->addr[pd->sidx], 603 nk->port[pd->sidx], 1, pd->af); 604 } 605 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 606 pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum, 607 &checksum, &nk->addr[pd->didx], 608 nk->port[pd->didx], 1, pd->af); 609 } 610 611 break; 612 } 613 case IPPROTO_ICMP: { 614 struct icmp *ih = &pd->hdr.icmp; 615 616 if (nk->port[pd->sidx] != ih->icmp_id) { 617 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 618 ih->icmp_cksum, ih->icmp_id, 619 nk->port[pd->sidx], 0); 620 ih->icmp_id = nk->port[pd->sidx]; 621 pd->sport = &ih->icmp_id; 622 623 m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih); 624 } 625 /* FALLTHROUGH */ 626 } 627 default: 628 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) { 629 switch (pd->af) { 630 case AF_INET: 631 pf_change_a(&pd->src->v4.s_addr, 632 pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 633 0); 634 break; 635 case AF_INET6: 636 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 637 break; 638 } 639 } 640 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) { 641 switch (pd->af) { 642 case AF_INET: 643 pf_change_a(&pd->dst->v4.s_addr, 644 pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 645 0); 646 break; 647 case AF_INET6: 648 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 649 break; 650 } 651 } 652 break; 653 } 654} 655 656static __inline uint32_t 657pf_hashkey(struct pf_state_key *sk) 658{ 659 uint32_t h; 660 661 h = murmur3_32_hash32((uint32_t *)sk, 662 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 663 V_pf_hashseed); 664 665 return (h & pf_hashmask); 666} 667 668static __inline uint32_t 669pf_hashsrc(struct pf_addr *addr, sa_family_t af) 670{ 671 uint32_t h; 672 673 switch (af) { 674 case AF_INET: 675 h = murmur3_32_hash32((uint32_t *)&addr->v4, 676 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 677 break; 678 case AF_INET6: 679 h = murmur3_32_hash32((uint32_t *)&addr->v6, 680 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 681 break; 682 default: 683 panic("%s: unknown address family %u", __func__, af); 684 } 685 686 return (h & pf_srchashmask); 687} 688 689#ifdef ALTQ 690static int 691pf_state_hash(struct pf_kstate *s) 692{ 693 u_int32_t hv = (intptr_t)s / sizeof(*s); 694 695 hv ^= crc32(&s->src, sizeof(s->src)); 696 hv ^= crc32(&s->dst, sizeof(s->dst)); 697 if (hv == 0) 698 hv = 1; 699 return (hv); 700} 701#endif 702 703static __inline void 704pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 705{ 706 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 707 s->dst.state = newstate; 708 if (which == PF_PEER_DST) 709 return; 710 if (s->src.state == newstate) 711 return; 712 if (s->creatorid == V_pf_status.hostid && 713 s->key[PF_SK_STACK] != NULL && 714 s->key[PF_SK_STACK]->proto == IPPROTO_TCP && 715 !(TCPS_HAVEESTABLISHED(s->src.state) || 716 s->src.state == TCPS_CLOSED) && 717 (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED)) 718 atomic_add_32(&V_pf_status.states_halfopen, -1); 719 720 s->src.state = newstate; 721} 722 723#ifdef INET6 724void 725pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 726{ 727 switch (af) { 728#ifdef INET 729 case AF_INET: 730 memcpy(&dst->v4, &src->v4, sizeof(dst->v4)); 731 break; 732#endif /* INET */ 733 case AF_INET6: 734 memcpy(&dst->v6, &src->v6, sizeof(dst->v6)); 735 break; 736 } 737} 738#endif /* INET6 */ 739 740static void 741pf_init_threshold(struct pf_threshold *threshold, 742 u_int32_t limit, u_int32_t seconds) 743{ 744 threshold->limit = limit * PF_THRESHOLD_MULT; 745 threshold->seconds = seconds; 746 threshold->count = 0; 747 threshold->last = time_uptime; 748} 749 750static void 751pf_add_threshold(struct pf_threshold *threshold) 752{ 753 u_int32_t t = time_uptime, diff = t - threshold->last; 754 755 if (diff >= threshold->seconds) 756 threshold->count = 0; 757 else 758 threshold->count -= threshold->count * diff / 759 threshold->seconds; 760 threshold->count += PF_THRESHOLD_MULT; 761 threshold->last = t; 762} 763 764static int 765pf_check_threshold(struct pf_threshold *threshold) 766{ 767 return (threshold->count > threshold->limit); 768} 769 770static int 771pf_src_connlimit(struct pf_kstate **state) 772{ 773 struct pf_overload_entry *pfoe; 774 int bad = 0; 775 776 PF_STATE_LOCK_ASSERT(*state); 777 /* 778 * XXXKS: The src node is accessed unlocked! 779 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node); 780 */ 781 782 (*state)->src_node->conn++; 783 (*state)->src.tcp_est = 1; 784 pf_add_threshold(&(*state)->src_node->conn_rate); 785 786 if ((*state)->rule.ptr->max_src_conn && 787 (*state)->rule.ptr->max_src_conn < 788 (*state)->src_node->conn) { 789 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 790 bad++; 791 } 792 793 if ((*state)->rule.ptr->max_src_conn_rate.limit && 794 pf_check_threshold(&(*state)->src_node->conn_rate)) { 795 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 796 bad++; 797 } 798 799 if (!bad) 800 return (0); 801 802 /* Kill this state. */ 803 (*state)->timeout = PFTM_PURGE; 804 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 805 806 if ((*state)->rule.ptr->overload_tbl == NULL) 807 return (1); 808 809 /* Schedule overloading and flushing task. */ 810 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 811 if (pfoe == NULL) 812 return (1); /* too bad :( */ 813 814 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 815 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 816 pfoe->rule = (*state)->rule.ptr; 817 pfoe->dir = (*state)->direction; 818 PF_OVERLOADQ_LOCK(); 819 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 820 PF_OVERLOADQ_UNLOCK(); 821 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 822 823 return (1); 824} 825 826static void 827pf_overload_task(void *v, int pending) 828{ 829 struct pf_overload_head queue; 830 struct pfr_addr p; 831 struct pf_overload_entry *pfoe, *pfoe1; 832 uint32_t killed = 0; 833 834 CURVNET_SET((struct vnet *)v); 835 836 PF_OVERLOADQ_LOCK(); 837 queue = V_pf_overloadqueue; 838 SLIST_INIT(&V_pf_overloadqueue); 839 PF_OVERLOADQ_UNLOCK(); 840 841 bzero(&p, sizeof(p)); 842 SLIST_FOREACH(pfoe, &queue, next) { 843 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 844 if (V_pf_status.debug >= PF_DEBUG_MISC) { 845 printf("%s: blocking address ", __func__); 846 pf_print_host(&pfoe->addr, 0, pfoe->af); 847 printf("\n"); 848 } 849 850 p.pfra_af = pfoe->af; 851 switch (pfoe->af) { 852#ifdef INET 853 case AF_INET: 854 p.pfra_net = 32; 855 p.pfra_ip4addr = pfoe->addr.v4; 856 break; 857#endif 858#ifdef INET6 859 case AF_INET6: 860 p.pfra_net = 128; 861 p.pfra_ip6addr = pfoe->addr.v6; 862 break; 863#endif 864 } 865 866 PF_RULES_WLOCK(); 867 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 868 PF_RULES_WUNLOCK(); 869 } 870 871 /* 872 * Remove those entries, that don't need flushing. 873 */ 874 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 875 if (pfoe->rule->flush == 0) { 876 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 877 free(pfoe, M_PFTEMP); 878 } else 879 counter_u64_add( 880 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 881 882 /* If nothing to flush, return. */ 883 if (SLIST_EMPTY(&queue)) { 884 CURVNET_RESTORE(); 885 return; 886 } 887 888 for (int i = 0; i <= pf_hashmask; i++) { 889 struct pf_idhash *ih = &V_pf_idhash[i]; 890 struct pf_state_key *sk; 891 struct pf_kstate *s; 892 893 PF_HASHROW_LOCK(ih); 894 LIST_FOREACH(s, &ih->states, entry) { 895 sk = s->key[PF_SK_WIRE]; 896 SLIST_FOREACH(pfoe, &queue, next) 897 if (sk->af == pfoe->af && 898 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 899 pfoe->rule == s->rule.ptr) && 900 ((pfoe->dir == PF_OUT && 901 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 902 (pfoe->dir == PF_IN && 903 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 904 s->timeout = PFTM_PURGE; 905 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 906 killed++; 907 } 908 } 909 PF_HASHROW_UNLOCK(ih); 910 } 911 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 912 free(pfoe, M_PFTEMP); 913 if (V_pf_status.debug >= PF_DEBUG_MISC) 914 printf("%s: %u states killed", __func__, killed); 915 916 CURVNET_RESTORE(); 917} 918 919/* 920 * Can return locked on failure, so that we can consistently 921 * allocate and insert a new one. 922 */ 923struct pf_ksrc_node * 924pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 925 struct pf_srchash **sh, bool returnlocked) 926{ 927 struct pf_ksrc_node *n; 928 929 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 930 931 *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 932 PF_HASHROW_LOCK(*sh); 933 LIST_FOREACH(n, &(*sh)->nodes, entry) 934 if (n->rule.ptr == rule && n->af == af && 935 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 936 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 937 break; 938 939 if (n != NULL) { 940 n->states++; 941 PF_HASHROW_UNLOCK(*sh); 942 } else if (returnlocked == false) 943 PF_HASHROW_UNLOCK(*sh); 944 945 return (n); 946} 947 948static void 949pf_free_src_node(struct pf_ksrc_node *sn) 950{ 951 952 for (int i = 0; i < 2; i++) { 953 counter_u64_free(sn->bytes[i]); 954 counter_u64_free(sn->packets[i]); 955 } 956 uma_zfree(V_pf_sources_z, sn); 957} 958 959static u_short 960pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 961 struct pf_addr *src, sa_family_t af) 962{ 963 u_short reason = 0; 964 struct pf_srchash *sh = NULL; 965 966 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 967 rule->rpool.opts & PF_POOL_STICKYADDR), 968 ("%s for non-tracking rule %p", __func__, rule)); 969 970 if (*sn == NULL) 971 *sn = pf_find_src_node(src, rule, af, &sh, true); 972 973 if (*sn == NULL) { 974 PF_HASHROW_ASSERT(sh); 975 976 if (rule->max_src_nodes && 977 counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) { 978 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); 979 PF_HASHROW_UNLOCK(sh); 980 reason = PFRES_SRCLIMIT; 981 goto done; 982 } 983 984 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 985 if ((*sn) == NULL) { 986 PF_HASHROW_UNLOCK(sh); 987 reason = PFRES_MEMORY; 988 goto done; 989 } 990 991 for (int i = 0; i < 2; i++) { 992 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 993 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 994 995 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 996 pf_free_src_node(*sn); 997 PF_HASHROW_UNLOCK(sh); 998 reason = PFRES_MEMORY; 999 goto done; 1000 } 1001 } 1002 1003 pf_init_threshold(&(*sn)->conn_rate, 1004 rule->max_src_conn_rate.limit, 1005 rule->max_src_conn_rate.seconds); 1006 1007 MPASS((*sn)->lock == NULL); 1008 (*sn)->lock = &sh->lock; 1009 1010 (*sn)->af = af; 1011 (*sn)->rule.ptr = rule; 1012 PF_ACPY(&(*sn)->addr, src, af); 1013 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 1014 (*sn)->creation = time_uptime; 1015 (*sn)->ruletype = rule->action; 1016 (*sn)->states = 1; 1017 if ((*sn)->rule.ptr != NULL) 1018 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 1019 PF_HASHROW_UNLOCK(sh); 1020 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 1021 } else { 1022 if (rule->max_src_states && 1023 (*sn)->states >= rule->max_src_states) { 1024 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1025 1); 1026 reason = PFRES_SRCLIMIT; 1027 goto done; 1028 } 1029 } 1030done: 1031 return (reason); 1032} 1033 1034void 1035pf_unlink_src_node(struct pf_ksrc_node *src) 1036{ 1037 PF_SRC_NODE_LOCK_ASSERT(src); 1038 1039 LIST_REMOVE(src, entry); 1040 if (src->rule.ptr) 1041 counter_u64_add(src->rule.ptr->src_nodes, -1); 1042} 1043 1044u_int 1045pf_free_src_nodes(struct pf_ksrc_node_list *head) 1046{ 1047 struct pf_ksrc_node *sn, *tmp; 1048 u_int count = 0; 1049 1050 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 1051 pf_free_src_node(sn); 1052 count++; 1053 } 1054 1055 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 1056 1057 return (count); 1058} 1059 1060void 1061pf_mtag_initialize(void) 1062{ 1063 1064 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 1065 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 1066 UMA_ALIGN_PTR, 0); 1067} 1068 1069/* Per-vnet data storage structures initialization. */ 1070void 1071pf_initialize(void) 1072{ 1073 struct pf_keyhash *kh; 1074 struct pf_idhash *ih; 1075 struct pf_srchash *sh; 1076 u_int i; 1077 1078 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 1079 pf_hashsize = PF_HASHSIZ; 1080 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 1081 pf_srchashsize = PF_SRCHASHSIZ; 1082 1083 V_pf_hashseed = arc4random(); 1084 1085 /* States and state keys storage. */ 1086 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 1087 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1088 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 1089 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 1090 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 1091 1092 V_pf_state_key_z = uma_zcreate("pf state keys", 1093 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 1094 UMA_ALIGN_PTR, 0); 1095 1096 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 1097 M_PFHASH, M_NOWAIT | M_ZERO); 1098 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 1099 M_PFHASH, M_NOWAIT | M_ZERO); 1100 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 1101 printf("pf: Unable to allocate memory for " 1102 "state_hashsize %lu.\n", pf_hashsize); 1103 1104 free(V_pf_keyhash, M_PFHASH); 1105 free(V_pf_idhash, M_PFHASH); 1106 1107 pf_hashsize = PF_HASHSIZ; 1108 V_pf_keyhash = mallocarray(pf_hashsize, 1109 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 1110 V_pf_idhash = mallocarray(pf_hashsize, 1111 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 1112 } 1113 1114 pf_hashmask = pf_hashsize - 1; 1115 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1116 i++, kh++, ih++) { 1117 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 1118 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 1119 } 1120 1121 /* Source nodes. */ 1122 V_pf_sources_z = uma_zcreate("pf source nodes", 1123 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1124 0); 1125 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 1126 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 1127 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 1128 1129 V_pf_srchash = mallocarray(pf_srchashsize, 1130 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 1131 if (V_pf_srchash == NULL) { 1132 printf("pf: Unable to allocate memory for " 1133 "source_hashsize %lu.\n", pf_srchashsize); 1134 1135 pf_srchashsize = PF_SRCHASHSIZ; 1136 V_pf_srchash = mallocarray(pf_srchashsize, 1137 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 1138 } 1139 1140 pf_srchashmask = pf_srchashsize - 1; 1141 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 1142 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 1143 1144 /* ALTQ */ 1145 TAILQ_INIT(&V_pf_altqs[0]); 1146 TAILQ_INIT(&V_pf_altqs[1]); 1147 TAILQ_INIT(&V_pf_altqs[2]); 1148 TAILQ_INIT(&V_pf_altqs[3]); 1149 TAILQ_INIT(&V_pf_pabuf); 1150 V_pf_altqs_active = &V_pf_altqs[0]; 1151 V_pf_altq_ifs_active = &V_pf_altqs[1]; 1152 V_pf_altqs_inactive = &V_pf_altqs[2]; 1153 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 1154 1155 /* Send & overload+flush queues. */ 1156 STAILQ_INIT(&V_pf_sendqueue); 1157 SLIST_INIT(&V_pf_overloadqueue); 1158 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 1159 1160 /* Unlinked, but may be referenced rules. */ 1161 TAILQ_INIT(&V_pf_unlinked_rules); 1162} 1163 1164void 1165pf_mtag_cleanup(void) 1166{ 1167 1168 uma_zdestroy(pf_mtag_z); 1169} 1170 1171void 1172pf_cleanup(void) 1173{ 1174 struct pf_keyhash *kh; 1175 struct pf_idhash *ih; 1176 struct pf_srchash *sh; 1177 struct pf_send_entry *pfse, *next; 1178 u_int i; 1179 1180 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 1181 i++, kh++, ih++) { 1182 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 1183 __func__)); 1184 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 1185 __func__)); 1186 mtx_destroy(&kh->lock); 1187 mtx_destroy(&ih->lock); 1188 } 1189 free(V_pf_keyhash, M_PFHASH); 1190 free(V_pf_idhash, M_PFHASH); 1191 1192 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1193 KASSERT(LIST_EMPTY(&sh->nodes), 1194 ("%s: source node hash not empty", __func__)); 1195 mtx_destroy(&sh->lock); 1196 } 1197 free(V_pf_srchash, M_PFHASH); 1198 1199 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 1200 m_freem(pfse->pfse_m); 1201 free(pfse, M_PFTEMP); 1202 } 1203 MPASS(RB_EMPTY(&V_pf_sctp_endpoints)); 1204 1205 uma_zdestroy(V_pf_sources_z); 1206 uma_zdestroy(V_pf_state_z); 1207 uma_zdestroy(V_pf_state_key_z); 1208} 1209 1210static int 1211pf_mtag_uminit(void *mem, int size, int how) 1212{ 1213 struct m_tag *t; 1214 1215 t = (struct m_tag *)mem; 1216 t->m_tag_cookie = MTAG_ABI_COMPAT; 1217 t->m_tag_id = PACKET_TAG_PF; 1218 t->m_tag_len = sizeof(struct pf_mtag); 1219 t->m_tag_free = pf_mtag_free; 1220 1221 return (0); 1222} 1223 1224static void 1225pf_mtag_free(struct m_tag *t) 1226{ 1227 1228 uma_zfree(pf_mtag_z, t); 1229} 1230 1231struct pf_mtag * 1232pf_get_mtag(struct mbuf *m) 1233{ 1234 struct m_tag *mtag; 1235 1236 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1237 return ((struct pf_mtag *)(mtag + 1)); 1238 1239 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1240 if (mtag == NULL) 1241 return (NULL); 1242 bzero(mtag + 1, sizeof(struct pf_mtag)); 1243 m_tag_prepend(m, mtag); 1244 1245 return ((struct pf_mtag *)(mtag + 1)); 1246} 1247 1248static int 1249pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1250 struct pf_kstate *s) 1251{ 1252 struct pf_keyhash *khs, *khw, *kh; 1253 struct pf_state_key *sk, *cur; 1254 struct pf_kstate *si, *olds = NULL; 1255 int idx; 1256 1257 NET_EPOCH_ASSERT(); 1258 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1259 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1260 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1261 1262 /* 1263 * We need to lock hash slots of both keys. To avoid deadlock 1264 * we always lock the slot with lower address first. Unlock order 1265 * isn't important. 1266 * 1267 * We also need to lock ID hash slot before dropping key 1268 * locks. On success we return with ID hash slot locked. 1269 */ 1270 1271 if (skw == sks) { 1272 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1273 PF_HASHROW_LOCK(khs); 1274 } else { 1275 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1276 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1277 if (khs == khw) { 1278 PF_HASHROW_LOCK(khs); 1279 } else if (khs < khw) { 1280 PF_HASHROW_LOCK(khs); 1281 PF_HASHROW_LOCK(khw); 1282 } else { 1283 PF_HASHROW_LOCK(khw); 1284 PF_HASHROW_LOCK(khs); 1285 } 1286 } 1287 1288#define KEYS_UNLOCK() do { \ 1289 if (khs != khw) { \ 1290 PF_HASHROW_UNLOCK(khs); \ 1291 PF_HASHROW_UNLOCK(khw); \ 1292 } else \ 1293 PF_HASHROW_UNLOCK(khs); \ 1294} while (0) 1295 1296 /* 1297 * First run: start with wire key. 1298 */ 1299 sk = skw; 1300 kh = khw; 1301 idx = PF_SK_WIRE; 1302 1303 MPASS(s->lock == NULL); 1304 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1305 1306keyattach: 1307 LIST_FOREACH(cur, &kh->keys, entry) 1308 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1309 break; 1310 1311 if (cur != NULL) { 1312 /* Key exists. Check for same kif, if none, add to key. */ 1313 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1314 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1315 1316 PF_HASHROW_LOCK(ih); 1317 if (si->kif == s->kif && 1318 si->direction == s->direction) { 1319 if (sk->proto == IPPROTO_TCP && 1320 si->src.state >= TCPS_FIN_WAIT_2 && 1321 si->dst.state >= TCPS_FIN_WAIT_2) { 1322 /* 1323 * New state matches an old >FIN_WAIT_2 1324 * state. We can't drop key hash locks, 1325 * thus we can't unlink it properly. 1326 * 1327 * As a workaround we drop it into 1328 * TCPS_CLOSED state, schedule purge 1329 * ASAP and push it into the very end 1330 * of the slot TAILQ, so that it won't 1331 * conflict with our new state. 1332 */ 1333 pf_set_protostate(si, PF_PEER_BOTH, 1334 TCPS_CLOSED); 1335 si->timeout = PFTM_PURGE; 1336 olds = si; 1337 } else { 1338 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1339 printf("pf: %s key attach " 1340 "failed on %s: ", 1341 (idx == PF_SK_WIRE) ? 1342 "wire" : "stack", 1343 s->kif->pfik_name); 1344 pf_print_state_parts(s, 1345 (idx == PF_SK_WIRE) ? 1346 sk : NULL, 1347 (idx == PF_SK_STACK) ? 1348 sk : NULL); 1349 printf(", existing: "); 1350 pf_print_state_parts(si, 1351 (idx == PF_SK_WIRE) ? 1352 sk : NULL, 1353 (idx == PF_SK_STACK) ? 1354 sk : NULL); 1355 printf("\n"); 1356 } 1357 s->timeout = PFTM_UNLINKED; 1358 PF_HASHROW_UNLOCK(ih); 1359 KEYS_UNLOCK(); 1360 uma_zfree(V_pf_state_key_z, sk); 1361 if (idx == PF_SK_STACK) 1362 pf_detach_state(s); 1363 return (EEXIST); /* collision! */ 1364 } 1365 } 1366 PF_HASHROW_UNLOCK(ih); 1367 } 1368 uma_zfree(V_pf_state_key_z, sk); 1369 s->key[idx] = cur; 1370 } else { 1371 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1372 s->key[idx] = sk; 1373 } 1374 1375stateattach: 1376 /* List is sorted, if-bound states before floating. */ 1377 if (s->kif == V_pfi_all) 1378 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1379 else 1380 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1381 1382 if (olds) { 1383 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1384 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1385 key_list[idx]); 1386 olds = NULL; 1387 } 1388 1389 /* 1390 * Attach done. See how should we (or should not?) 1391 * attach a second key. 1392 */ 1393 if (sks == skw) { 1394 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1395 idx = PF_SK_STACK; 1396 sks = NULL; 1397 goto stateattach; 1398 } else if (sks != NULL) { 1399 /* 1400 * Continue attaching with stack key. 1401 */ 1402 sk = sks; 1403 kh = khs; 1404 idx = PF_SK_STACK; 1405 sks = NULL; 1406 goto keyattach; 1407 } 1408 1409 PF_STATE_LOCK(s); 1410 KEYS_UNLOCK(); 1411 1412 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1413 ("%s failure", __func__)); 1414 1415 return (0); 1416#undef KEYS_UNLOCK 1417} 1418 1419static void 1420pf_detach_state(struct pf_kstate *s) 1421{ 1422 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1423 struct pf_keyhash *kh; 1424 1425 NET_EPOCH_ASSERT(); 1426 MPASS(s->timeout >= PFTM_MAX); 1427 1428 pf_sctp_multihome_detach_addr(s); 1429 1430 if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr) 1431 V_pflow_export_state_ptr(s); 1432 1433 if (sks != NULL) { 1434 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1435 PF_HASHROW_LOCK(kh); 1436 if (s->key[PF_SK_STACK] != NULL) 1437 pf_state_key_detach(s, PF_SK_STACK); 1438 /* 1439 * If both point to same key, then we are done. 1440 */ 1441 if (sks == s->key[PF_SK_WIRE]) { 1442 pf_state_key_detach(s, PF_SK_WIRE); 1443 PF_HASHROW_UNLOCK(kh); 1444 return; 1445 } 1446 PF_HASHROW_UNLOCK(kh); 1447 } 1448 1449 if (s->key[PF_SK_WIRE] != NULL) { 1450 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1451 PF_HASHROW_LOCK(kh); 1452 if (s->key[PF_SK_WIRE] != NULL) 1453 pf_state_key_detach(s, PF_SK_WIRE); 1454 PF_HASHROW_UNLOCK(kh); 1455 } 1456} 1457 1458static void 1459pf_state_key_detach(struct pf_kstate *s, int idx) 1460{ 1461 struct pf_state_key *sk = s->key[idx]; 1462#ifdef INVARIANTS 1463 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1464 1465 PF_HASHROW_ASSERT(kh); 1466#endif 1467 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1468 s->key[idx] = NULL; 1469 1470 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1471 LIST_REMOVE(sk, entry); 1472 uma_zfree(V_pf_state_key_z, sk); 1473 } 1474} 1475 1476static int 1477pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1478{ 1479 struct pf_state_key *sk = mem; 1480 1481 bzero(sk, sizeof(struct pf_state_key_cmp)); 1482 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1483 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1484 1485 return (0); 1486} 1487 1488struct pf_state_key * 1489pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1490 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1491{ 1492 struct pf_state_key *sk; 1493 1494 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1495 if (sk == NULL) 1496 return (NULL); 1497 1498 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1499 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1500 sk->port[pd->sidx] = sport; 1501 sk->port[pd->didx] = dport; 1502 sk->proto = pd->proto; 1503 sk->af = pd->af; 1504 1505 return (sk); 1506} 1507 1508struct pf_state_key * 1509pf_state_key_clone(struct pf_state_key *orig) 1510{ 1511 struct pf_state_key *sk; 1512 1513 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1514 if (sk == NULL) 1515 return (NULL); 1516 1517 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1518 1519 return (sk); 1520} 1521 1522int 1523pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1524 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1525{ 1526 struct pf_idhash *ih; 1527 struct pf_kstate *cur; 1528 int error; 1529 1530 NET_EPOCH_ASSERT(); 1531 1532 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1533 ("%s: sks not pristine", __func__)); 1534 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1535 ("%s: skw not pristine", __func__)); 1536 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1537 1538 s->kif = kif; 1539 s->orig_kif = orig_kif; 1540 1541 if (s->id == 0 && s->creatorid == 0) { 1542 s->id = alloc_unr64(&V_pf_stateid); 1543 s->id = htobe64(s->id); 1544 s->creatorid = V_pf_status.hostid; 1545 } 1546 1547 /* Returns with ID locked on success. */ 1548 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1549 return (error); 1550 1551 ih = &V_pf_idhash[PF_IDHASH(s)]; 1552 PF_HASHROW_ASSERT(ih); 1553 LIST_FOREACH(cur, &ih->states, entry) 1554 if (cur->id == s->id && cur->creatorid == s->creatorid) 1555 break; 1556 1557 if (cur != NULL) { 1558 s->timeout = PFTM_UNLINKED; 1559 PF_HASHROW_UNLOCK(ih); 1560 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1561 printf("pf: state ID collision: " 1562 "id: %016llx creatorid: %08x\n", 1563 (unsigned long long)be64toh(s->id), 1564 ntohl(s->creatorid)); 1565 } 1566 pf_detach_state(s); 1567 return (EEXIST); 1568 } 1569 LIST_INSERT_HEAD(&ih->states, s, entry); 1570 /* One for keys, one for ID hash. */ 1571 refcount_init(&s->refs, 2); 1572 1573 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1574 if (V_pfsync_insert_state_ptr != NULL) 1575 V_pfsync_insert_state_ptr(s); 1576 1577 /* Returns locked. */ 1578 return (0); 1579} 1580 1581/* 1582 * Find state by ID: returns with locked row on success. 1583 */ 1584struct pf_kstate * 1585pf_find_state_byid(uint64_t id, uint32_t creatorid) 1586{ 1587 struct pf_idhash *ih; 1588 struct pf_kstate *s; 1589 1590 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1591 1592 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1593 1594 PF_HASHROW_LOCK(ih); 1595 LIST_FOREACH(s, &ih->states, entry) 1596 if (s->id == id && s->creatorid == creatorid) 1597 break; 1598 1599 if (s == NULL) 1600 PF_HASHROW_UNLOCK(ih); 1601 1602 return (s); 1603} 1604 1605/* 1606 * Find state by key. 1607 * Returns with ID hash slot locked on success. 1608 */ 1609static struct pf_kstate * 1610pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1611{ 1612 struct pf_keyhash *kh; 1613 struct pf_state_key *sk; 1614 struct pf_kstate *s; 1615 int idx; 1616 1617 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1618 1619 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1620 1621 PF_HASHROW_LOCK(kh); 1622 LIST_FOREACH(sk, &kh->keys, entry) 1623 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1624 break; 1625 if (sk == NULL) { 1626 PF_HASHROW_UNLOCK(kh); 1627 return (NULL); 1628 } 1629 1630 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1631 1632 /* List is sorted, if-bound states before floating ones. */ 1633 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1634 if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) { 1635 PF_STATE_LOCK(s); 1636 PF_HASHROW_UNLOCK(kh); 1637 if (__predict_false(s->timeout >= PFTM_MAX)) { 1638 /* 1639 * State is either being processed by 1640 * pf_unlink_state() in an other thread, or 1641 * is scheduled for immediate expiry. 1642 */ 1643 PF_STATE_UNLOCK(s); 1644 return (NULL); 1645 } 1646 return (s); 1647 } 1648 PF_HASHROW_UNLOCK(kh); 1649 1650 return (NULL); 1651} 1652 1653/* 1654 * Returns with ID hash slot locked on success. 1655 */ 1656struct pf_kstate * 1657pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1658{ 1659 struct pf_keyhash *kh; 1660 struct pf_state_key *sk; 1661 struct pf_kstate *s, *ret = NULL; 1662 int idx, inout = 0; 1663 1664 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1665 1666 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1667 1668 PF_HASHROW_LOCK(kh); 1669 LIST_FOREACH(sk, &kh->keys, entry) 1670 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1671 break; 1672 if (sk == NULL) { 1673 PF_HASHROW_UNLOCK(kh); 1674 return (NULL); 1675 } 1676 switch (dir) { 1677 case PF_IN: 1678 idx = PF_SK_WIRE; 1679 break; 1680 case PF_OUT: 1681 idx = PF_SK_STACK; 1682 break; 1683 case PF_INOUT: 1684 idx = PF_SK_WIRE; 1685 inout = 1; 1686 break; 1687 default: 1688 panic("%s: dir %u", __func__, dir); 1689 } 1690second_run: 1691 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1692 if (more == NULL) { 1693 PF_STATE_LOCK(s); 1694 PF_HASHROW_UNLOCK(kh); 1695 return (s); 1696 } 1697 1698 if (ret) 1699 (*more)++; 1700 else { 1701 ret = s; 1702 PF_STATE_LOCK(s); 1703 } 1704 } 1705 if (inout == 1) { 1706 inout = 0; 1707 idx = PF_SK_STACK; 1708 goto second_run; 1709 } 1710 PF_HASHROW_UNLOCK(kh); 1711 1712 return (ret); 1713} 1714 1715/* 1716 * FIXME 1717 * This routine is inefficient -- locks the state only to unlock immediately on 1718 * return. 1719 * It is racy -- after the state is unlocked nothing stops other threads from 1720 * removing it. 1721 */ 1722bool 1723pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1724{ 1725 struct pf_kstate *s; 1726 1727 s = pf_find_state_all(key, dir, NULL); 1728 if (s != NULL) { 1729 PF_STATE_UNLOCK(s); 1730 return (true); 1731 } 1732 return (false); 1733} 1734 1735/* END state table stuff */ 1736 1737static void 1738pf_send(struct pf_send_entry *pfse) 1739{ 1740 1741 PF_SENDQ_LOCK(); 1742 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1743 PF_SENDQ_UNLOCK(); 1744 swi_sched(V_pf_swi_cookie, 0); 1745} 1746 1747static bool 1748pf_isforlocal(struct mbuf *m, int af) 1749{ 1750 switch (af) { 1751#ifdef INET 1752 case AF_INET: { 1753 struct ip *ip = mtod(m, struct ip *); 1754 1755 return (in_localip(ip->ip_dst)); 1756 } 1757#endif 1758#ifdef INET6 1759 case AF_INET6: { 1760 struct ip6_hdr *ip6; 1761 struct in6_ifaddr *ia; 1762 ip6 = mtod(m, struct ip6_hdr *); 1763 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1764 if (ia == NULL) 1765 return (false); 1766 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1767 } 1768#endif 1769 default: 1770 panic("Unsupported af %d", af); 1771 } 1772 1773 return (false); 1774} 1775 1776void 1777pf_intr(void *v) 1778{ 1779 struct epoch_tracker et; 1780 struct pf_send_head queue; 1781 struct pf_send_entry *pfse, *next; 1782 1783 CURVNET_SET((struct vnet *)v); 1784 1785 PF_SENDQ_LOCK(); 1786 queue = V_pf_sendqueue; 1787 STAILQ_INIT(&V_pf_sendqueue); 1788 PF_SENDQ_UNLOCK(); 1789 1790 NET_EPOCH_ENTER(et); 1791 1792 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1793 switch (pfse->pfse_type) { 1794#ifdef INET 1795 case PFSE_IP: { 1796 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1797 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1798 pfse->pfse_m->m_pkthdr.csum_flags |= 1799 CSUM_IP_VALID | CSUM_IP_CHECKED; 1800 ip_input(pfse->pfse_m); 1801 } else { 1802 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1803 NULL); 1804 } 1805 break; 1806 } 1807 case PFSE_ICMP: 1808 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1809 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1810 break; 1811#endif /* INET */ 1812#ifdef INET6 1813 case PFSE_IP6: 1814 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1815 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1816 ip6_input(pfse->pfse_m); 1817 } else { 1818 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1819 NULL, NULL); 1820 } 1821 break; 1822 case PFSE_ICMP6: 1823 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1824 pfse->icmpopts.code, pfse->icmpopts.mtu); 1825 break; 1826#endif /* INET6 */ 1827 default: 1828 panic("%s: unknown type", __func__); 1829 } 1830 free(pfse, M_PFTEMP); 1831 } 1832 NET_EPOCH_EXIT(et); 1833 CURVNET_RESTORE(); 1834} 1835 1836#define pf_purge_thread_period (hz / 10) 1837 1838#ifdef PF_WANT_32_TO_64_COUNTER 1839static void 1840pf_status_counter_u64_periodic(void) 1841{ 1842 1843 PF_RULES_RASSERT(); 1844 1845 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1846 return; 1847 } 1848 1849 for (int i = 0; i < FCNT_MAX; i++) { 1850 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1851 } 1852} 1853 1854static void 1855pf_kif_counter_u64_periodic(void) 1856{ 1857 struct pfi_kkif *kif; 1858 size_t r, run; 1859 1860 PF_RULES_RASSERT(); 1861 1862 if (__predict_false(V_pf_allkifcount == 0)) { 1863 return; 1864 } 1865 1866 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1867 return; 1868 } 1869 1870 run = V_pf_allkifcount / 10; 1871 if (run < 5) 1872 run = 5; 1873 1874 for (r = 0; r < run; r++) { 1875 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1876 if (kif == NULL) { 1877 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1878 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1879 break; 1880 } 1881 1882 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1883 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1884 1885 for (int i = 0; i < 2; i++) { 1886 for (int j = 0; j < 2; j++) { 1887 for (int k = 0; k < 2; k++) { 1888 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1889 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1890 } 1891 } 1892 } 1893 } 1894} 1895 1896static void 1897pf_rule_counter_u64_periodic(void) 1898{ 1899 struct pf_krule *rule; 1900 size_t r, run; 1901 1902 PF_RULES_RASSERT(); 1903 1904 if (__predict_false(V_pf_allrulecount == 0)) { 1905 return; 1906 } 1907 1908 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1909 return; 1910 } 1911 1912 run = V_pf_allrulecount / 10; 1913 if (run < 5) 1914 run = 5; 1915 1916 for (r = 0; r < run; r++) { 1917 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1918 if (rule == NULL) { 1919 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1920 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1921 break; 1922 } 1923 1924 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1925 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1926 1927 pf_counter_u64_periodic(&rule->evaluations); 1928 for (int i = 0; i < 2; i++) { 1929 pf_counter_u64_periodic(&rule->packets[i]); 1930 pf_counter_u64_periodic(&rule->bytes[i]); 1931 } 1932 } 1933} 1934 1935static void 1936pf_counter_u64_periodic_main(void) 1937{ 1938 PF_RULES_RLOCK_TRACKER; 1939 1940 V_pf_counter_periodic_iter++; 1941 1942 PF_RULES_RLOCK(); 1943 pf_counter_u64_critical_enter(); 1944 pf_status_counter_u64_periodic(); 1945 pf_kif_counter_u64_periodic(); 1946 pf_rule_counter_u64_periodic(); 1947 pf_counter_u64_critical_exit(); 1948 PF_RULES_RUNLOCK(); 1949} 1950#else 1951#define pf_counter_u64_periodic_main() do { } while (0) 1952#endif 1953 1954void 1955pf_purge_thread(void *unused __unused) 1956{ 1957 struct epoch_tracker et; 1958 1959 VNET_ITERATOR_DECL(vnet_iter); 1960 1961 sx_xlock(&pf_end_lock); 1962 while (pf_end_threads == 0) { 1963 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1964 1965 VNET_LIST_RLOCK(); 1966 NET_EPOCH_ENTER(et); 1967 VNET_FOREACH(vnet_iter) { 1968 CURVNET_SET(vnet_iter); 1969 1970 /* Wait until V_pf_default_rule is initialized. */ 1971 if (V_pf_vnet_active == 0) { 1972 CURVNET_RESTORE(); 1973 continue; 1974 } 1975 1976 pf_counter_u64_periodic_main(); 1977 1978 /* 1979 * Process 1/interval fraction of the state 1980 * table every run. 1981 */ 1982 V_pf_purge_idx = 1983 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1984 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1985 1986 /* 1987 * Purge other expired types every 1988 * PFTM_INTERVAL seconds. 1989 */ 1990 if (V_pf_purge_idx == 0) { 1991 /* 1992 * Order is important: 1993 * - states and src nodes reference rules 1994 * - states and rules reference kifs 1995 */ 1996 pf_purge_expired_fragments(); 1997 pf_purge_expired_src_nodes(); 1998 pf_purge_unlinked_rules(); 1999 pfi_kkif_purge(); 2000 } 2001 CURVNET_RESTORE(); 2002 } 2003 NET_EPOCH_EXIT(et); 2004 VNET_LIST_RUNLOCK(); 2005 } 2006 2007 pf_end_threads++; 2008 sx_xunlock(&pf_end_lock); 2009 kproc_exit(0); 2010} 2011 2012void 2013pf_unload_vnet_purge(void) 2014{ 2015 2016 /* 2017 * To cleanse up all kifs and rules we need 2018 * two runs: first one clears reference flags, 2019 * then pf_purge_expired_states() doesn't 2020 * raise them, and then second run frees. 2021 */ 2022 pf_purge_unlinked_rules(); 2023 pfi_kkif_purge(); 2024 2025 /* 2026 * Now purge everything. 2027 */ 2028 pf_purge_expired_states(0, pf_hashmask); 2029 pf_purge_fragments(UINT_MAX); 2030 pf_purge_expired_src_nodes(); 2031 2032 /* 2033 * Now all kifs & rules should be unreferenced, 2034 * thus should be successfully freed. 2035 */ 2036 pf_purge_unlinked_rules(); 2037 pfi_kkif_purge(); 2038} 2039 2040u_int32_t 2041pf_state_expires(const struct pf_kstate *state) 2042{ 2043 u_int32_t timeout; 2044 u_int32_t start; 2045 u_int32_t end; 2046 u_int32_t states; 2047 2048 /* handle all PFTM_* > PFTM_MAX here */ 2049 if (state->timeout == PFTM_PURGE) 2050 return (time_uptime); 2051 KASSERT(state->timeout != PFTM_UNLINKED, 2052 ("pf_state_expires: timeout == PFTM_UNLINKED")); 2053 KASSERT((state->timeout < PFTM_MAX), 2054 ("pf_state_expires: timeout > PFTM_MAX")); 2055 timeout = state->rule.ptr->timeout[state->timeout]; 2056 if (!timeout) 2057 timeout = V_pf_default_rule.timeout[state->timeout]; 2058 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 2059 if (start && state->rule.ptr != &V_pf_default_rule) { 2060 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 2061 states = counter_u64_fetch(state->rule.ptr->states_cur); 2062 } else { 2063 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 2064 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 2065 states = V_pf_status.states; 2066 } 2067 if (end && states > start && start < end) { 2068 if (states < end) { 2069 timeout = (u_int64_t)timeout * (end - states) / 2070 (end - start); 2071 return ((state->expire / 1000) + timeout); 2072 } 2073 else 2074 return (time_uptime); 2075 } 2076 return ((state->expire / 1000) + timeout); 2077} 2078 2079void 2080pf_purge_expired_src_nodes(void) 2081{ 2082 struct pf_ksrc_node_list freelist; 2083 struct pf_srchash *sh; 2084 struct pf_ksrc_node *cur, *next; 2085 int i; 2086 2087 LIST_INIT(&freelist); 2088 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 2089 PF_HASHROW_LOCK(sh); 2090 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 2091 if (cur->states == 0 && cur->expire <= time_uptime) { 2092 pf_unlink_src_node(cur); 2093 LIST_INSERT_HEAD(&freelist, cur, entry); 2094 } else if (cur->rule.ptr != NULL) 2095 cur->rule.ptr->rule_ref |= PFRULE_REFS; 2096 PF_HASHROW_UNLOCK(sh); 2097 } 2098 2099 pf_free_src_nodes(&freelist); 2100 2101 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 2102} 2103 2104static void 2105pf_src_tree_remove_state(struct pf_kstate *s) 2106{ 2107 struct pf_ksrc_node *sn; 2108 uint32_t timeout; 2109 2110 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 2111 s->rule.ptr->timeout[PFTM_SRC_NODE] : 2112 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 2113 2114 if (s->src_node != NULL) { 2115 sn = s->src_node; 2116 PF_SRC_NODE_LOCK(sn); 2117 if (s->src.tcp_est) 2118 --sn->conn; 2119 if (--sn->states == 0) 2120 sn->expire = time_uptime + timeout; 2121 PF_SRC_NODE_UNLOCK(sn); 2122 } 2123 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 2124 sn = s->nat_src_node; 2125 PF_SRC_NODE_LOCK(sn); 2126 if (--sn->states == 0) 2127 sn->expire = time_uptime + timeout; 2128 PF_SRC_NODE_UNLOCK(sn); 2129 } 2130 s->src_node = s->nat_src_node = NULL; 2131} 2132 2133/* 2134 * Unlink and potentilly free a state. Function may be 2135 * called with ID hash row locked, but always returns 2136 * unlocked, since it needs to go through key hash locking. 2137 */ 2138int 2139pf_unlink_state(struct pf_kstate *s) 2140{ 2141 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 2142 2143 NET_EPOCH_ASSERT(); 2144 PF_HASHROW_ASSERT(ih); 2145 2146 if (s->timeout == PFTM_UNLINKED) { 2147 /* 2148 * State is being processed 2149 * by pf_unlink_state() in 2150 * an other thread. 2151 */ 2152 PF_HASHROW_UNLOCK(ih); 2153 return (0); /* XXXGL: undefined actually */ 2154 } 2155 2156 if (s->src.state == PF_TCPS_PROXY_DST) { 2157 /* XXX wire key the right one? */ 2158 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 2159 &s->key[PF_SK_WIRE]->addr[1], 2160 &s->key[PF_SK_WIRE]->addr[0], 2161 s->key[PF_SK_WIRE]->port[1], 2162 s->key[PF_SK_WIRE]->port[0], 2163 s->src.seqhi, s->src.seqlo + 1, 2164 TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid); 2165 } 2166 2167 LIST_REMOVE(s, entry); 2168 pf_src_tree_remove_state(s); 2169 2170 if (V_pfsync_delete_state_ptr != NULL) 2171 V_pfsync_delete_state_ptr(s); 2172 2173 STATE_DEC_COUNTERS(s); 2174 2175 s->timeout = PFTM_UNLINKED; 2176 2177 /* Ensure we remove it from the list of halfopen states, if needed. */ 2178 if (s->key[PF_SK_STACK] != NULL && 2179 s->key[PF_SK_STACK]->proto == IPPROTO_TCP) 2180 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 2181 2182 PF_HASHROW_UNLOCK(ih); 2183 2184 pf_detach_state(s); 2185 /* pf_state_insert() initialises refs to 2 */ 2186 return (pf_release_staten(s, 2)); 2187} 2188 2189struct pf_kstate * 2190pf_alloc_state(int flags) 2191{ 2192 2193 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 2194} 2195 2196void 2197pf_free_state(struct pf_kstate *cur) 2198{ 2199 struct pf_krule_item *ri; 2200 2201 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 2202 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 2203 cur->timeout)); 2204 2205 while ((ri = SLIST_FIRST(&cur->match_rules))) { 2206 SLIST_REMOVE_HEAD(&cur->match_rules, entry); 2207 free(ri, M_PF_RULE_ITEM); 2208 } 2209 2210 pf_normalize_tcp_cleanup(cur); 2211 uma_zfree(V_pf_state_z, cur); 2212 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 2213} 2214 2215/* 2216 * Called only from pf_purge_thread(), thus serialized. 2217 */ 2218static u_int 2219pf_purge_expired_states(u_int i, int maxcheck) 2220{ 2221 struct pf_idhash *ih; 2222 struct pf_kstate *s; 2223 struct pf_krule_item *mrm; 2224 size_t count __unused; 2225 2226 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2227 2228 /* 2229 * Go through hash and unlink states that expire now. 2230 */ 2231 while (maxcheck > 0) { 2232 count = 0; 2233 ih = &V_pf_idhash[i]; 2234 2235 /* only take the lock if we expect to do work */ 2236 if (!LIST_EMPTY(&ih->states)) { 2237relock: 2238 PF_HASHROW_LOCK(ih); 2239 LIST_FOREACH(s, &ih->states, entry) { 2240 if (pf_state_expires(s) <= time_uptime) { 2241 V_pf_status.states -= 2242 pf_unlink_state(s); 2243 goto relock; 2244 } 2245 s->rule.ptr->rule_ref |= PFRULE_REFS; 2246 if (s->nat_rule.ptr != NULL) 2247 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 2248 if (s->anchor.ptr != NULL) 2249 s->anchor.ptr->rule_ref |= PFRULE_REFS; 2250 s->kif->pfik_flags |= PFI_IFLAG_REFS; 2251 SLIST_FOREACH(mrm, &s->match_rules, entry) 2252 mrm->r->rule_ref |= PFRULE_REFS; 2253 if (s->rt_kif) 2254 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 2255 count++; 2256 } 2257 PF_HASHROW_UNLOCK(ih); 2258 } 2259 2260 SDT_PROBE2(pf, purge, state, rowcount, i, count); 2261 2262 /* Return when we hit end of hash. */ 2263 if (++i > pf_hashmask) { 2264 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2265 return (0); 2266 } 2267 2268 maxcheck--; 2269 } 2270 2271 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2272 2273 return (i); 2274} 2275 2276static void 2277pf_purge_unlinked_rules(void) 2278{ 2279 struct pf_krulequeue tmpq; 2280 struct pf_krule *r, *r1; 2281 2282 /* 2283 * If we have overloading task pending, then we'd 2284 * better skip purging this time. There is a tiny 2285 * probability that overloading task references 2286 * an already unlinked rule. 2287 */ 2288 PF_OVERLOADQ_LOCK(); 2289 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2290 PF_OVERLOADQ_UNLOCK(); 2291 return; 2292 } 2293 PF_OVERLOADQ_UNLOCK(); 2294 2295 /* 2296 * Do naive mark-and-sweep garbage collecting of old rules. 2297 * Reference flag is raised by pf_purge_expired_states() 2298 * and pf_purge_expired_src_nodes(). 2299 * 2300 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2301 * use a temporary queue. 2302 */ 2303 TAILQ_INIT(&tmpq); 2304 PF_UNLNKDRULES_LOCK(); 2305 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2306 if (!(r->rule_ref & PFRULE_REFS)) { 2307 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2308 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2309 } else 2310 r->rule_ref &= ~PFRULE_REFS; 2311 } 2312 PF_UNLNKDRULES_UNLOCK(); 2313 2314 if (!TAILQ_EMPTY(&tmpq)) { 2315 PF_CONFIG_LOCK(); 2316 PF_RULES_WLOCK(); 2317 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2318 TAILQ_REMOVE(&tmpq, r, entries); 2319 pf_free_rule(r); 2320 } 2321 PF_RULES_WUNLOCK(); 2322 PF_CONFIG_UNLOCK(); 2323 } 2324} 2325 2326void 2327pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2328{ 2329 switch (af) { 2330#ifdef INET 2331 case AF_INET: { 2332 u_int32_t a = ntohl(addr->addr32[0]); 2333 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2334 (a>>8)&255, a&255); 2335 if (p) { 2336 p = ntohs(p); 2337 printf(":%u", p); 2338 } 2339 break; 2340 } 2341#endif /* INET */ 2342#ifdef INET6 2343 case AF_INET6: { 2344 u_int16_t b; 2345 u_int8_t i, curstart, curend, maxstart, maxend; 2346 curstart = curend = maxstart = maxend = 255; 2347 for (i = 0; i < 8; i++) { 2348 if (!addr->addr16[i]) { 2349 if (curstart == 255) 2350 curstart = i; 2351 curend = i; 2352 } else { 2353 if ((curend - curstart) > 2354 (maxend - maxstart)) { 2355 maxstart = curstart; 2356 maxend = curend; 2357 } 2358 curstart = curend = 255; 2359 } 2360 } 2361 if ((curend - curstart) > 2362 (maxend - maxstart)) { 2363 maxstart = curstart; 2364 maxend = curend; 2365 } 2366 for (i = 0; i < 8; i++) { 2367 if (i >= maxstart && i <= maxend) { 2368 if (i == 0) 2369 printf(":"); 2370 if (i == maxend) 2371 printf(":"); 2372 } else { 2373 b = ntohs(addr->addr16[i]); 2374 printf("%x", b); 2375 if (i < 7) 2376 printf(":"); 2377 } 2378 } 2379 if (p) { 2380 p = ntohs(p); 2381 printf("[%u]", p); 2382 } 2383 break; 2384 } 2385#endif /* INET6 */ 2386 } 2387} 2388 2389void 2390pf_print_state(struct pf_kstate *s) 2391{ 2392 pf_print_state_parts(s, NULL, NULL); 2393} 2394 2395static void 2396pf_print_state_parts(struct pf_kstate *s, 2397 struct pf_state_key *skwp, struct pf_state_key *sksp) 2398{ 2399 struct pf_state_key *skw, *sks; 2400 u_int8_t proto, dir; 2401 2402 /* Do our best to fill these, but they're skipped if NULL */ 2403 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2404 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2405 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2406 dir = s ? s->direction : 0; 2407 2408 switch (proto) { 2409 case IPPROTO_IPV4: 2410 printf("IPv4"); 2411 break; 2412 case IPPROTO_IPV6: 2413 printf("IPv6"); 2414 break; 2415 case IPPROTO_TCP: 2416 printf("TCP"); 2417 break; 2418 case IPPROTO_UDP: 2419 printf("UDP"); 2420 break; 2421 case IPPROTO_ICMP: 2422 printf("ICMP"); 2423 break; 2424 case IPPROTO_ICMPV6: 2425 printf("ICMPv6"); 2426 break; 2427 default: 2428 printf("%u", proto); 2429 break; 2430 } 2431 switch (dir) { 2432 case PF_IN: 2433 printf(" in"); 2434 break; 2435 case PF_OUT: 2436 printf(" out"); 2437 break; 2438 } 2439 if (skw) { 2440 printf(" wire: "); 2441 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2442 printf(" "); 2443 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2444 } 2445 if (sks) { 2446 printf(" stack: "); 2447 if (sks != skw) { 2448 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2449 printf(" "); 2450 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2451 } else 2452 printf("-"); 2453 } 2454 if (s) { 2455 if (proto == IPPROTO_TCP) { 2456 printf(" [lo=%u high=%u win=%u modulator=%u", 2457 s->src.seqlo, s->src.seqhi, 2458 s->src.max_win, s->src.seqdiff); 2459 if (s->src.wscale && s->dst.wscale) 2460 printf(" wscale=%u", 2461 s->src.wscale & PF_WSCALE_MASK); 2462 printf("]"); 2463 printf(" [lo=%u high=%u win=%u modulator=%u", 2464 s->dst.seqlo, s->dst.seqhi, 2465 s->dst.max_win, s->dst.seqdiff); 2466 if (s->src.wscale && s->dst.wscale) 2467 printf(" wscale=%u", 2468 s->dst.wscale & PF_WSCALE_MASK); 2469 printf("]"); 2470 } 2471 printf(" %u:%u", s->src.state, s->dst.state); 2472 } 2473} 2474 2475void 2476pf_print_flags(u_int8_t f) 2477{ 2478 if (f) 2479 printf(" "); 2480 if (f & TH_FIN) 2481 printf("F"); 2482 if (f & TH_SYN) 2483 printf("S"); 2484 if (f & TH_RST) 2485 printf("R"); 2486 if (f & TH_PUSH) 2487 printf("P"); 2488 if (f & TH_ACK) 2489 printf("A"); 2490 if (f & TH_URG) 2491 printf("U"); 2492 if (f & TH_ECE) 2493 printf("E"); 2494 if (f & TH_CWR) 2495 printf("W"); 2496} 2497 2498#define PF_SET_SKIP_STEPS(i) \ 2499 do { \ 2500 while (head[i] != cur) { \ 2501 head[i]->skip[i].ptr = cur; \ 2502 head[i] = TAILQ_NEXT(head[i], entries); \ 2503 } \ 2504 } while (0) 2505 2506void 2507pf_calc_skip_steps(struct pf_krulequeue *rules) 2508{ 2509 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2510 int i; 2511 2512 cur = TAILQ_FIRST(rules); 2513 prev = cur; 2514 for (i = 0; i < PF_SKIP_COUNT; ++i) 2515 head[i] = cur; 2516 while (cur != NULL) { 2517 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2518 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2519 if (cur->direction != prev->direction) 2520 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2521 if (cur->af != prev->af) 2522 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2523 if (cur->proto != prev->proto) 2524 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2525 if (cur->src.neg != prev->src.neg || 2526 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2527 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2528 if (cur->src.port[0] != prev->src.port[0] || 2529 cur->src.port[1] != prev->src.port[1] || 2530 cur->src.port_op != prev->src.port_op) 2531 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2532 if (cur->dst.neg != prev->dst.neg || 2533 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2534 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2535 if (cur->dst.port[0] != prev->dst.port[0] || 2536 cur->dst.port[1] != prev->dst.port[1] || 2537 cur->dst.port_op != prev->dst.port_op) 2538 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2539 2540 prev = cur; 2541 cur = TAILQ_NEXT(cur, entries); 2542 } 2543 for (i = 0; i < PF_SKIP_COUNT; ++i) 2544 PF_SET_SKIP_STEPS(i); 2545} 2546 2547int 2548pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2549{ 2550 if (aw1->type != aw2->type) 2551 return (1); 2552 switch (aw1->type) { 2553 case PF_ADDR_ADDRMASK: 2554 case PF_ADDR_RANGE: 2555 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2556 return (1); 2557 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2558 return (1); 2559 return (0); 2560 case PF_ADDR_DYNIFTL: 2561 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2562 case PF_ADDR_NOROUTE: 2563 case PF_ADDR_URPFFAILED: 2564 return (0); 2565 case PF_ADDR_TABLE: 2566 return (aw1->p.tbl != aw2->p.tbl); 2567 default: 2568 printf("invalid address type: %d\n", aw1->type); 2569 return (1); 2570 } 2571} 2572 2573/** 2574 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2575 * header isn't always a full checksum. In some cases (i.e. output) it's a 2576 * pseudo-header checksum, which is a partial checksum over src/dst IP 2577 * addresses, protocol number and length. 2578 * 2579 * That means we have the following cases: 2580 * * Input or forwarding: we don't have TSO, the checksum fields are full 2581 * checksums, we need to update the checksum whenever we change anything. 2582 * * Output (i.e. the checksum is a pseudo-header checksum): 2583 * x The field being updated is src/dst address or affects the length of 2584 * the packet. We need to update the pseudo-header checksum (note that this 2585 * checksum is not ones' complement). 2586 * x Some other field is being modified (e.g. src/dst port numbers): We 2587 * don't have to update anything. 2588 **/ 2589u_int16_t 2590pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2591{ 2592 u_int32_t x; 2593 2594 x = cksum + old - new; 2595 x = (x + (x >> 16)) & 0xffff; 2596 2597 /* optimise: eliminate a branch when not udp */ 2598 if (udp && cksum == 0x0000) 2599 return cksum; 2600 if (udp && x == 0x0000) 2601 x = 0xffff; 2602 2603 return (u_int16_t)(x); 2604} 2605 2606static void 2607pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2608 u_int8_t udp) 2609{ 2610 u_int16_t old = htons(hi ? (*f << 8) : *f); 2611 u_int16_t new = htons(hi ? ( v << 8) : v); 2612 2613 if (*f == v) 2614 return; 2615 2616 *f = v; 2617 2618 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2619 return; 2620 2621 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2622} 2623 2624void 2625pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2626 bool hi, u_int8_t udp) 2627{ 2628 u_int8_t *fb = (u_int8_t *)f; 2629 u_int8_t *vb = (u_int8_t *)&v; 2630 2631 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2632 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2633} 2634 2635void 2636pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2637 bool hi, u_int8_t udp) 2638{ 2639 u_int8_t *fb = (u_int8_t *)f; 2640 u_int8_t *vb = (u_int8_t *)&v; 2641 2642 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2643 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2644 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2645 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2646} 2647 2648u_int16_t 2649pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2650 u_int16_t new, u_int8_t udp) 2651{ 2652 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2653 return (cksum); 2654 2655 return (pf_cksum_fixup(cksum, old, new, udp)); 2656} 2657 2658static void 2659pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2660 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2661 sa_family_t af) 2662{ 2663 struct pf_addr ao; 2664 u_int16_t po = *p; 2665 2666 PF_ACPY(&ao, a, af); 2667 PF_ACPY(a, an, af); 2668 2669 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2670 *pc = ~*pc; 2671 2672 *p = pn; 2673 2674 switch (af) { 2675#ifdef INET 2676 case AF_INET: 2677 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2678 ao.addr16[0], an->addr16[0], 0), 2679 ao.addr16[1], an->addr16[1], 0); 2680 *p = pn; 2681 2682 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2683 ao.addr16[0], an->addr16[0], u), 2684 ao.addr16[1], an->addr16[1], u); 2685 2686 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2687 break; 2688#endif /* INET */ 2689#ifdef INET6 2690 case AF_INET6: 2691 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2692 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2693 pf_cksum_fixup(pf_cksum_fixup(*pc, 2694 ao.addr16[0], an->addr16[0], u), 2695 ao.addr16[1], an->addr16[1], u), 2696 ao.addr16[2], an->addr16[2], u), 2697 ao.addr16[3], an->addr16[3], u), 2698 ao.addr16[4], an->addr16[4], u), 2699 ao.addr16[5], an->addr16[5], u), 2700 ao.addr16[6], an->addr16[6], u), 2701 ao.addr16[7], an->addr16[7], u); 2702 2703 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2704 break; 2705#endif /* INET6 */ 2706 } 2707 2708 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2709 CSUM_DELAY_DATA_IPV6)) { 2710 *pc = ~*pc; 2711 if (! *pc) 2712 *pc = 0xffff; 2713 } 2714} 2715 2716/* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2717void 2718pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2719{ 2720 u_int32_t ao; 2721 2722 memcpy(&ao, a, sizeof(ao)); 2723 memcpy(a, &an, sizeof(u_int32_t)); 2724 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2725 ao % 65536, an % 65536, u); 2726} 2727 2728void 2729pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2730{ 2731 u_int32_t ao; 2732 2733 memcpy(&ao, a, sizeof(ao)); 2734 memcpy(a, &an, sizeof(u_int32_t)); 2735 2736 *c = pf_proto_cksum_fixup(m, 2737 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2738 ao % 65536, an % 65536, udp); 2739} 2740 2741#ifdef INET6 2742static void 2743pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2744{ 2745 struct pf_addr ao; 2746 2747 PF_ACPY(&ao, a, AF_INET6); 2748 PF_ACPY(a, an, AF_INET6); 2749 2750 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2751 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2752 pf_cksum_fixup(pf_cksum_fixup(*c, 2753 ao.addr16[0], an->addr16[0], u), 2754 ao.addr16[1], an->addr16[1], u), 2755 ao.addr16[2], an->addr16[2], u), 2756 ao.addr16[3], an->addr16[3], u), 2757 ao.addr16[4], an->addr16[4], u), 2758 ao.addr16[5], an->addr16[5], u), 2759 ao.addr16[6], an->addr16[6], u), 2760 ao.addr16[7], an->addr16[7], u); 2761} 2762#endif /* INET6 */ 2763 2764static void 2765pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2766 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2767 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2768{ 2769 struct pf_addr oia, ooa; 2770 2771 PF_ACPY(&oia, ia, af); 2772 if (oa) 2773 PF_ACPY(&ooa, oa, af); 2774 2775 /* Change inner protocol port, fix inner protocol checksum. */ 2776 if (ip != NULL) { 2777 u_int16_t oip = *ip; 2778 u_int32_t opc; 2779 2780 if (pc != NULL) 2781 opc = *pc; 2782 *ip = np; 2783 if (pc != NULL) 2784 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2785 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2786 if (pc != NULL) 2787 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2788 } 2789 /* Change inner ip address, fix inner ip and icmp checksums. */ 2790 PF_ACPY(ia, na, af); 2791 switch (af) { 2792#ifdef INET 2793 case AF_INET: { 2794 u_int32_t oh2c = *h2c; 2795 2796 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2797 oia.addr16[0], ia->addr16[0], 0), 2798 oia.addr16[1], ia->addr16[1], 0); 2799 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2800 oia.addr16[0], ia->addr16[0], 0), 2801 oia.addr16[1], ia->addr16[1], 0); 2802 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2803 break; 2804 } 2805#endif /* INET */ 2806#ifdef INET6 2807 case AF_INET6: 2808 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2809 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2810 pf_cksum_fixup(pf_cksum_fixup(*ic, 2811 oia.addr16[0], ia->addr16[0], u), 2812 oia.addr16[1], ia->addr16[1], u), 2813 oia.addr16[2], ia->addr16[2], u), 2814 oia.addr16[3], ia->addr16[3], u), 2815 oia.addr16[4], ia->addr16[4], u), 2816 oia.addr16[5], ia->addr16[5], u), 2817 oia.addr16[6], ia->addr16[6], u), 2818 oia.addr16[7], ia->addr16[7], u); 2819 break; 2820#endif /* INET6 */ 2821 } 2822 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2823 if (oa) { 2824 PF_ACPY(oa, na, af); 2825 switch (af) { 2826#ifdef INET 2827 case AF_INET: 2828 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2829 ooa.addr16[0], oa->addr16[0], 0), 2830 ooa.addr16[1], oa->addr16[1], 0); 2831 break; 2832#endif /* INET */ 2833#ifdef INET6 2834 case AF_INET6: 2835 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2836 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2837 pf_cksum_fixup(pf_cksum_fixup(*ic, 2838 ooa.addr16[0], oa->addr16[0], u), 2839 ooa.addr16[1], oa->addr16[1], u), 2840 ooa.addr16[2], oa->addr16[2], u), 2841 ooa.addr16[3], oa->addr16[3], u), 2842 ooa.addr16[4], oa->addr16[4], u), 2843 ooa.addr16[5], oa->addr16[5], u), 2844 ooa.addr16[6], oa->addr16[6], u), 2845 ooa.addr16[7], oa->addr16[7], u); 2846 break; 2847#endif /* INET6 */ 2848 } 2849 } 2850} 2851 2852/* 2853 * Need to modulate the sequence numbers in the TCP SACK option 2854 * (credits to Krzysztof Pfaff for report and patch) 2855 */ 2856static int 2857pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2858 struct tcphdr *th, struct pf_state_peer *dst) 2859{ 2860 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2861 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2862 int copyback = 0, i, olen; 2863 struct sackblk sack; 2864 2865#define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2866 if (hlen < TCPOLEN_SACKLEN || 2867 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2868 return 0; 2869 2870 while (hlen >= TCPOLEN_SACKLEN) { 2871 size_t startoff = opt - opts; 2872 olen = opt[1]; 2873 switch (*opt) { 2874 case TCPOPT_EOL: /* FALLTHROUGH */ 2875 case TCPOPT_NOP: 2876 opt++; 2877 hlen--; 2878 break; 2879 case TCPOPT_SACK: 2880 if (olen > hlen) 2881 olen = hlen; 2882 if (olen >= TCPOLEN_SACKLEN) { 2883 for (i = 2; i + TCPOLEN_SACK <= olen; 2884 i += TCPOLEN_SACK) { 2885 memcpy(&sack, &opt[i], sizeof(sack)); 2886 pf_patch_32_unaligned(m, 2887 &th->th_sum, &sack.start, 2888 htonl(ntohl(sack.start) - dst->seqdiff), 2889 PF_ALGNMNT(startoff), 2890 0); 2891 pf_patch_32_unaligned(m, &th->th_sum, 2892 &sack.end, 2893 htonl(ntohl(sack.end) - dst->seqdiff), 2894 PF_ALGNMNT(startoff), 2895 0); 2896 memcpy(&opt[i], &sack, sizeof(sack)); 2897 } 2898 copyback = 1; 2899 } 2900 /* FALLTHROUGH */ 2901 default: 2902 if (olen < 2) 2903 olen = 2; 2904 hlen -= olen; 2905 opt += olen; 2906 } 2907 } 2908 2909 if (copyback) 2910 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2911 return (copyback); 2912} 2913 2914struct mbuf * 2915pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2916 const struct pf_addr *saddr, const struct pf_addr *daddr, 2917 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2918 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 2919 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 2920{ 2921 struct mbuf *m; 2922 int len, tlen; 2923#ifdef INET 2924 struct ip *h = NULL; 2925#endif /* INET */ 2926#ifdef INET6 2927 struct ip6_hdr *h6 = NULL; 2928#endif /* INET6 */ 2929 struct tcphdr *th; 2930 char *opt; 2931 struct pf_mtag *pf_mtag; 2932 2933 len = 0; 2934 th = NULL; 2935 2936 /* maximum segment size tcp option */ 2937 tlen = sizeof(struct tcphdr); 2938 if (mss) 2939 tlen += 4; 2940 2941 switch (af) { 2942#ifdef INET 2943 case AF_INET: 2944 len = sizeof(struct ip) + tlen; 2945 break; 2946#endif /* INET */ 2947#ifdef INET6 2948 case AF_INET6: 2949 len = sizeof(struct ip6_hdr) + tlen; 2950 break; 2951#endif /* INET6 */ 2952 default: 2953 panic("%s: unsupported af %d", __func__, af); 2954 } 2955 2956 m = m_gethdr(M_NOWAIT, MT_DATA); 2957 if (m == NULL) 2958 return (NULL); 2959 2960#ifdef MAC 2961 mac_netinet_firewall_send(m); 2962#endif 2963 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2964 m_freem(m); 2965 return (NULL); 2966 } 2967 if (skip_firewall) 2968 m->m_flags |= M_SKIP_FIREWALL; 2969 pf_mtag->tag = mtag_tag; 2970 pf_mtag->flags = mtag_flags; 2971 2972 if (rtableid >= 0) 2973 M_SETFIB(m, rtableid); 2974 2975#ifdef ALTQ 2976 if (r != NULL && r->qid) { 2977 pf_mtag->qid = r->qid; 2978 2979 /* add hints for ecn */ 2980 pf_mtag->hdr = mtod(m, struct ip *); 2981 } 2982#endif /* ALTQ */ 2983 m->m_data += max_linkhdr; 2984 m->m_pkthdr.len = m->m_len = len; 2985 /* The rest of the stack assumes a rcvif, so provide one. 2986 * This is a locally generated packet, so .. close enough. */ 2987 m->m_pkthdr.rcvif = V_loif; 2988 bzero(m->m_data, len); 2989 switch (af) { 2990#ifdef INET 2991 case AF_INET: 2992 h = mtod(m, struct ip *); 2993 2994 /* IP header fields included in the TCP checksum */ 2995 h->ip_p = IPPROTO_TCP; 2996 h->ip_len = htons(tlen); 2997 h->ip_src.s_addr = saddr->v4.s_addr; 2998 h->ip_dst.s_addr = daddr->v4.s_addr; 2999 3000 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 3001 break; 3002#endif /* INET */ 3003#ifdef INET6 3004 case AF_INET6: 3005 h6 = mtod(m, struct ip6_hdr *); 3006 3007 /* IP header fields included in the TCP checksum */ 3008 h6->ip6_nxt = IPPROTO_TCP; 3009 h6->ip6_plen = htons(tlen); 3010 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 3011 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 3012 3013 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 3014 break; 3015#endif /* INET6 */ 3016 } 3017 3018 /* TCP header */ 3019 th->th_sport = sport; 3020 th->th_dport = dport; 3021 th->th_seq = htonl(seq); 3022 th->th_ack = htonl(ack); 3023 th->th_off = tlen >> 2; 3024 th->th_flags = tcp_flags; 3025 th->th_win = htons(win); 3026 3027 if (mss) { 3028 opt = (char *)(th + 1); 3029 opt[0] = TCPOPT_MAXSEG; 3030 opt[1] = 4; 3031 HTONS(mss); 3032 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 3033 } 3034 3035 switch (af) { 3036#ifdef INET 3037 case AF_INET: 3038 /* TCP checksum */ 3039 th->th_sum = in_cksum(m, len); 3040 3041 /* Finish the IP header */ 3042 h->ip_v = 4; 3043 h->ip_hl = sizeof(*h) >> 2; 3044 h->ip_tos = IPTOS_LOWDELAY; 3045 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 3046 h->ip_len = htons(len); 3047 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3048 h->ip_sum = 0; 3049 break; 3050#endif /* INET */ 3051#ifdef INET6 3052 case AF_INET6: 3053 /* TCP checksum */ 3054 th->th_sum = in6_cksum(m, IPPROTO_TCP, 3055 sizeof(struct ip6_hdr), tlen); 3056 3057 h6->ip6_vfc |= IPV6_VERSION; 3058 h6->ip6_hlim = IPV6_DEFHLIM; 3059 break; 3060#endif /* INET6 */ 3061 } 3062 3063 return (m); 3064} 3065 3066static void 3067pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd, 3068 uint8_t ttl, int rtableid) 3069{ 3070 struct mbuf *m; 3071#ifdef INET 3072 struct ip *h = NULL; 3073#endif /* INET */ 3074#ifdef INET6 3075 struct ip6_hdr *h6 = NULL; 3076#endif /* INET6 */ 3077 struct sctphdr *hdr; 3078 struct sctp_chunkhdr *chunk; 3079 struct pf_send_entry *pfse; 3080 int off = 0; 3081 3082 MPASS(af == pd->af); 3083 3084 m = m_gethdr(M_NOWAIT, MT_DATA); 3085 if (m == NULL) 3086 return; 3087 3088 m->m_data += max_linkhdr; 3089 m->m_flags |= M_SKIP_FIREWALL; 3090 /* The rest of the stack assumes a rcvif, so provide one. 3091 * This is a locally generated packet, so .. close enough. */ 3092 m->m_pkthdr.rcvif = V_loif; 3093 3094 /* IPv4|6 header */ 3095 switch (af) { 3096#ifdef INET 3097 case AF_INET: 3098 bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk)); 3099 3100 h = mtod(m, struct ip *); 3101 3102 /* IP header fields included in the TCP checksum */ 3103 3104 h->ip_p = IPPROTO_SCTP; 3105 h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk)); 3106 h->ip_ttl = ttl ? ttl : V_ip_defttl; 3107 h->ip_src = pd->dst->v4; 3108 h->ip_dst = pd->src->v4; 3109 3110 off += sizeof(struct ip); 3111 break; 3112#endif /* INET */ 3113#ifdef INET6 3114 case AF_INET6: 3115 bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk)); 3116 3117 h6 = mtod(m, struct ip6_hdr *); 3118 3119 /* IP header fields included in the TCP checksum */ 3120 h6->ip6_vfc |= IPV6_VERSION; 3121 h6->ip6_nxt = IPPROTO_SCTP; 3122 h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk)); 3123 h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim; 3124 memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr)); 3125 memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr)); 3126 3127 off += sizeof(struct ip6_hdr); 3128 break; 3129#endif /* INET6 */ 3130 } 3131 3132 /* SCTP header */ 3133 hdr = mtodo(m, off); 3134 3135 hdr->src_port = pd->hdr.sctp.dest_port; 3136 hdr->dest_port = pd->hdr.sctp.src_port; 3137 hdr->v_tag = pd->sctp_initiate_tag; 3138 hdr->checksum = 0; 3139 3140 /* Abort chunk. */ 3141 off += sizeof(struct sctphdr); 3142 chunk = mtodo(m, off); 3143 3144 chunk->chunk_type = SCTP_ABORT_ASSOCIATION; 3145 chunk->chunk_length = htons(sizeof(*chunk)); 3146 3147 /* SCTP checksum */ 3148 off += sizeof(*chunk); 3149 m->m_pkthdr.len = m->m_len = off; 3150 3151 pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk)); 3152 3153 if (rtableid >= 0) 3154 M_SETFIB(m, rtableid); 3155 3156 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3157 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3158 if (pfse == NULL) { 3159 m_freem(m); 3160 return; 3161 } 3162 3163 switch (af) { 3164#ifdef INET 3165 case AF_INET: 3166 pfse->pfse_type = PFSE_IP; 3167 break; 3168#endif /* INET */ 3169#ifdef INET6 3170 case AF_INET6: 3171 pfse->pfse_type = PFSE_IP6; 3172 break; 3173#endif /* INET6 */ 3174 } 3175 3176 pfse->pfse_m = m; 3177 pf_send(pfse); 3178} 3179 3180void 3181pf_send_tcp(const struct pf_krule *r, sa_family_t af, 3182 const struct pf_addr *saddr, const struct pf_addr *daddr, 3183 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 3184 u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, 3185 bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid) 3186{ 3187 struct pf_send_entry *pfse; 3188 struct mbuf *m; 3189 3190 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags, 3191 win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid); 3192 if (m == NULL) 3193 return; 3194 3195 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3196 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3197 if (pfse == NULL) { 3198 m_freem(m); 3199 return; 3200 } 3201 3202 switch (af) { 3203#ifdef INET 3204 case AF_INET: 3205 pfse->pfse_type = PFSE_IP; 3206 break; 3207#endif /* INET */ 3208#ifdef INET6 3209 case AF_INET6: 3210 pfse->pfse_type = PFSE_IP6; 3211 break; 3212#endif /* INET6 */ 3213 } 3214 3215 pfse->pfse_m = m; 3216 pf_send(pfse); 3217} 3218 3219static void 3220pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 3221 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 3222 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 3223 u_short *reason, int rtableid) 3224{ 3225 struct pf_addr * const saddr = pd->src; 3226 struct pf_addr * const daddr = pd->dst; 3227 sa_family_t af = pd->af; 3228 3229 /* undo NAT changes, if they have taken place */ 3230 if (nr != NULL) { 3231 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 3232 PF_ACPY(daddr, &sk->addr[pd->didx], af); 3233 if (pd->sport) 3234 *pd->sport = sk->port[pd->sidx]; 3235 if (pd->dport) 3236 *pd->dport = sk->port[pd->didx]; 3237 if (pd->proto_sum) 3238 *pd->proto_sum = bproto_sum; 3239 if (pd->ip_sum) 3240 *pd->ip_sum = bip_sum; 3241 m_copyback(m, off, hdrlen, pd->hdr.any); 3242 } 3243 if (pd->proto == IPPROTO_TCP && 3244 ((r->rule_flag & PFRULE_RETURNRST) || 3245 (r->rule_flag & PFRULE_RETURN)) && 3246 !(th->th_flags & TH_RST)) { 3247 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 3248 int len = 0; 3249#ifdef INET 3250 struct ip *h4; 3251#endif 3252#ifdef INET6 3253 struct ip6_hdr *h6; 3254#endif 3255 3256 switch (af) { 3257#ifdef INET 3258 case AF_INET: 3259 h4 = mtod(m, struct ip *); 3260 len = ntohs(h4->ip_len) - off; 3261 break; 3262#endif 3263#ifdef INET6 3264 case AF_INET6: 3265 h6 = mtod(m, struct ip6_hdr *); 3266 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 3267 break; 3268#endif 3269 } 3270 3271 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 3272 REASON_SET(reason, PFRES_PROTCKSUM); 3273 else { 3274 if (th->th_flags & TH_SYN) 3275 ack++; 3276 if (th->th_flags & TH_FIN) 3277 ack++; 3278 pf_send_tcp(r, af, pd->dst, 3279 pd->src, th->th_dport, th->th_sport, 3280 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 3281 r->return_ttl, true, 0, 0, rtableid); 3282 } 3283 } else if (pd->proto == IPPROTO_SCTP && 3284 (r->rule_flag & PFRULE_RETURN)) { 3285 pf_send_sctp_abort(af, pd, r->return_ttl, rtableid); 3286 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 3287 r->return_icmp) 3288 pf_send_icmp(m, r->return_icmp >> 8, 3289 r->return_icmp & 255, af, r, rtableid); 3290 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 3291 r->return_icmp6) 3292 pf_send_icmp(m, r->return_icmp6 >> 8, 3293 r->return_icmp6 & 255, af, r, rtableid); 3294} 3295 3296static int 3297pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 3298{ 3299 struct m_tag *mtag; 3300 u_int8_t mpcp; 3301 3302 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 3303 if (mtag == NULL) 3304 return (0); 3305 3306 if (prio == PF_PRIO_ZERO) 3307 prio = 0; 3308 3309 mpcp = *(uint8_t *)(mtag + 1); 3310 3311 return (mpcp == prio); 3312} 3313 3314static int 3315pf_icmp_to_bandlim(uint8_t type) 3316{ 3317 switch (type) { 3318 case ICMP_ECHO: 3319 case ICMP_ECHOREPLY: 3320 return (BANDLIM_ICMP_ECHO); 3321 case ICMP_TSTAMP: 3322 case ICMP_TSTAMPREPLY: 3323 return (BANDLIM_ICMP_TSTAMP); 3324 case ICMP_UNREACH: 3325 default: 3326 return (BANDLIM_ICMP_UNREACH); 3327 } 3328} 3329 3330static void 3331pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 3332 struct pf_krule *r, int rtableid) 3333{ 3334 struct pf_send_entry *pfse; 3335 struct mbuf *m0; 3336 struct pf_mtag *pf_mtag; 3337 3338 /* ICMP packet rate limitation. */ 3339#ifdef INET6 3340 if (af == AF_INET6) { 3341 if (icmp6_ratelimit(NULL, type, code)) 3342 return; 3343 } 3344#endif 3345#ifdef INET 3346 if (af == AF_INET) { 3347 if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0) 3348 return; 3349 } 3350#endif 3351 3352 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 3353 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 3354 if (pfse == NULL) 3355 return; 3356 3357 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 3358 free(pfse, M_PFTEMP); 3359 return; 3360 } 3361 3362 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 3363 free(pfse, M_PFTEMP); 3364 return; 3365 } 3366 /* XXX: revisit */ 3367 m0->m_flags |= M_SKIP_FIREWALL; 3368 3369 if (rtableid >= 0) 3370 M_SETFIB(m0, rtableid); 3371 3372#ifdef ALTQ 3373 if (r->qid) { 3374 pf_mtag->qid = r->qid; 3375 /* add hints for ecn */ 3376 pf_mtag->hdr = mtod(m0, struct ip *); 3377 } 3378#endif /* ALTQ */ 3379 3380 switch (af) { 3381#ifdef INET 3382 case AF_INET: 3383 pfse->pfse_type = PFSE_ICMP; 3384 break; 3385#endif /* INET */ 3386#ifdef INET6 3387 case AF_INET6: 3388 pfse->pfse_type = PFSE_ICMP6; 3389 break; 3390#endif /* INET6 */ 3391 } 3392 pfse->pfse_m = m0; 3393 pfse->icmpopts.type = type; 3394 pfse->icmpopts.code = code; 3395 pf_send(pfse); 3396} 3397 3398/* 3399 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 3400 * If n is 0, they match if they are equal. If n is != 0, they match if they 3401 * are different. 3402 */ 3403int 3404pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 3405 struct pf_addr *b, sa_family_t af) 3406{ 3407 int match = 0; 3408 3409 switch (af) { 3410#ifdef INET 3411 case AF_INET: 3412 if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4)) 3413 match++; 3414 break; 3415#endif /* INET */ 3416#ifdef INET6 3417 case AF_INET6: 3418 if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6)) 3419 match++; 3420 break; 3421#endif /* INET6 */ 3422 } 3423 if (match) { 3424 if (n) 3425 return (0); 3426 else 3427 return (1); 3428 } else { 3429 if (n) 3430 return (1); 3431 else 3432 return (0); 3433 } 3434} 3435 3436/* 3437 * Return 1 if b <= a <= e, otherwise return 0. 3438 */ 3439int 3440pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3441 struct pf_addr *a, sa_family_t af) 3442{ 3443 switch (af) { 3444#ifdef INET 3445 case AF_INET: 3446 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3447 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3448 return (0); 3449 break; 3450#endif /* INET */ 3451#ifdef INET6 3452 case AF_INET6: { 3453 int i; 3454 3455 /* check a >= b */ 3456 for (i = 0; i < 4; ++i) 3457 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3458 break; 3459 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3460 return (0); 3461 /* check a <= e */ 3462 for (i = 0; i < 4; ++i) 3463 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3464 break; 3465 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3466 return (0); 3467 break; 3468 } 3469#endif /* INET6 */ 3470 } 3471 return (1); 3472} 3473 3474static int 3475pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3476{ 3477 switch (op) { 3478 case PF_OP_IRG: 3479 return ((p > a1) && (p < a2)); 3480 case PF_OP_XRG: 3481 return ((p < a1) || (p > a2)); 3482 case PF_OP_RRG: 3483 return ((p >= a1) && (p <= a2)); 3484 case PF_OP_EQ: 3485 return (p == a1); 3486 case PF_OP_NE: 3487 return (p != a1); 3488 case PF_OP_LT: 3489 return (p < a1); 3490 case PF_OP_LE: 3491 return (p <= a1); 3492 case PF_OP_GT: 3493 return (p > a1); 3494 case PF_OP_GE: 3495 return (p >= a1); 3496 } 3497 return (0); /* never reached */ 3498} 3499 3500int 3501pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3502{ 3503 NTOHS(a1); 3504 NTOHS(a2); 3505 NTOHS(p); 3506 return (pf_match(op, a1, a2, p)); 3507} 3508 3509static int 3510pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3511{ 3512 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3513 return (0); 3514 return (pf_match(op, a1, a2, u)); 3515} 3516 3517static int 3518pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3519{ 3520 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3521 return (0); 3522 return (pf_match(op, a1, a2, g)); 3523} 3524 3525int 3526pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3527{ 3528 if (*tag == -1) 3529 *tag = mtag; 3530 3531 return ((!r->match_tag_not && r->match_tag == *tag) || 3532 (r->match_tag_not && r->match_tag != *tag)); 3533} 3534 3535int 3536pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3537{ 3538 3539 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3540 3541 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3542 return (ENOMEM); 3543 3544 pd->pf_mtag->tag = tag; 3545 3546 return (0); 3547} 3548 3549#define PF_ANCHOR_STACKSIZE 32 3550struct pf_kanchor_stackframe { 3551 struct pf_kruleset *rs; 3552 struct pf_krule *r; /* XXX: + match bit */ 3553 struct pf_kanchor *child; 3554}; 3555 3556/* 3557 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3558 */ 3559#define PF_ANCHORSTACK_MATCH 0x00000001 3560#define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3561 3562#define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3563#define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3564 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3565#define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3566 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3567} while (0) 3568 3569void 3570pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3571 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3572 int *match) 3573{ 3574 struct pf_kanchor_stackframe *f; 3575 3576 PF_RULES_RASSERT(); 3577 3578 if (match) 3579 *match = 0; 3580 if (*depth >= PF_ANCHOR_STACKSIZE) { 3581 printf("%s: anchor stack overflow on %s\n", 3582 __func__, (*r)->anchor->name); 3583 *r = TAILQ_NEXT(*r, entries); 3584 return; 3585 } else if (*depth == 0 && a != NULL) 3586 *a = *r; 3587 f = stack + (*depth)++; 3588 f->rs = *rs; 3589 f->r = *r; 3590 if ((*r)->anchor_wildcard) { 3591 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3592 3593 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3594 *r = NULL; 3595 return; 3596 } 3597 *rs = &f->child->ruleset; 3598 } else { 3599 f->child = NULL; 3600 *rs = &(*r)->anchor->ruleset; 3601 } 3602 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3603} 3604 3605int 3606pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3607 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3608 int *match) 3609{ 3610 struct pf_kanchor_stackframe *f; 3611 struct pf_krule *fr; 3612 int quick = 0; 3613 3614 PF_RULES_RASSERT(); 3615 3616 do { 3617 if (*depth <= 0) 3618 break; 3619 f = stack + *depth - 1; 3620 fr = PF_ANCHOR_RULE(f); 3621 if (f->child != NULL) { 3622 /* 3623 * This block traverses through 3624 * a wildcard anchor. 3625 */ 3626 if (match != NULL && *match) { 3627 /* 3628 * If any of "*" matched, then 3629 * "foo/ *" matched, mark frame 3630 * appropriately. 3631 */ 3632 PF_ANCHOR_SET_MATCH(f); 3633 *match = 0; 3634 } 3635 f->child = RB_NEXT(pf_kanchor_node, 3636 &fr->anchor->children, f->child); 3637 if (f->child != NULL) { 3638 *rs = &f->child->ruleset; 3639 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3640 if (*r == NULL) 3641 continue; 3642 else 3643 break; 3644 } 3645 } 3646 (*depth)--; 3647 if (*depth == 0 && a != NULL) 3648 *a = NULL; 3649 *rs = f->rs; 3650 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3651 quick = fr->quick; 3652 *r = TAILQ_NEXT(fr, entries); 3653 } while (*r == NULL); 3654 3655 return (quick); 3656} 3657 3658struct pf_keth_anchor_stackframe { 3659 struct pf_keth_ruleset *rs; 3660 struct pf_keth_rule *r; /* XXX: + match bit */ 3661 struct pf_keth_anchor *child; 3662}; 3663 3664#define PF_ETH_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3665#define PF_ETH_ANCHOR_RULE(f) (struct pf_keth_rule *) \ 3666 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3667#define PF_ETH_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3668 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3669} while (0) 3670 3671void 3672pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3673 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3674 struct pf_keth_rule **a, int *match) 3675{ 3676 struct pf_keth_anchor_stackframe *f; 3677 3678 NET_EPOCH_ASSERT(); 3679 3680 if (match) 3681 *match = 0; 3682 if (*depth >= PF_ANCHOR_STACKSIZE) { 3683 printf("%s: anchor stack overflow on %s\n", 3684 __func__, (*r)->anchor->name); 3685 *r = TAILQ_NEXT(*r, entries); 3686 return; 3687 } else if (*depth == 0 && a != NULL) 3688 *a = *r; 3689 f = stack + (*depth)++; 3690 f->rs = *rs; 3691 f->r = *r; 3692 if ((*r)->anchor_wildcard) { 3693 struct pf_keth_anchor_node *parent = &(*r)->anchor->children; 3694 3695 if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) { 3696 *r = NULL; 3697 return; 3698 } 3699 *rs = &f->child->ruleset; 3700 } else { 3701 f->child = NULL; 3702 *rs = &(*r)->anchor->ruleset; 3703 } 3704 *r = TAILQ_FIRST((*rs)->active.rules); 3705} 3706 3707int 3708pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth, 3709 struct pf_keth_ruleset **rs, struct pf_keth_rule **r, 3710 struct pf_keth_rule **a, int *match) 3711{ 3712 struct pf_keth_anchor_stackframe *f; 3713 struct pf_keth_rule *fr; 3714 int quick = 0; 3715 3716 NET_EPOCH_ASSERT(); 3717 3718 do { 3719 if (*depth <= 0) 3720 break; 3721 f = stack + *depth - 1; 3722 fr = PF_ETH_ANCHOR_RULE(f); 3723 if (f->child != NULL) { 3724 /* 3725 * This block traverses through 3726 * a wildcard anchor. 3727 */ 3728 if (match != NULL && *match) { 3729 /* 3730 * If any of "*" matched, then 3731 * "foo/ *" matched, mark frame 3732 * appropriately. 3733 */ 3734 PF_ETH_ANCHOR_SET_MATCH(f); 3735 *match = 0; 3736 } 3737 f->child = RB_NEXT(pf_keth_anchor_node, 3738 &fr->anchor->children, f->child); 3739 if (f->child != NULL) { 3740 *rs = &f->child->ruleset; 3741 *r = TAILQ_FIRST((*rs)->active.rules); 3742 if (*r == NULL) 3743 continue; 3744 else 3745 break; 3746 } 3747 } 3748 (*depth)--; 3749 if (*depth == 0 && a != NULL) 3750 *a = NULL; 3751 *rs = f->rs; 3752 if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match)) 3753 quick = fr->quick; 3754 *r = TAILQ_NEXT(fr, entries); 3755 } while (*r == NULL); 3756 3757 return (quick); 3758} 3759 3760#ifdef INET6 3761void 3762pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3763 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3764{ 3765 switch (af) { 3766#ifdef INET 3767 case AF_INET: 3768 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3769 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3770 break; 3771#endif /* INET */ 3772 case AF_INET6: 3773 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3774 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3775 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3776 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3777 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3778 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3779 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3780 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3781 break; 3782 } 3783} 3784 3785void 3786pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3787{ 3788 switch (af) { 3789#ifdef INET 3790 case AF_INET: 3791 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3792 break; 3793#endif /* INET */ 3794 case AF_INET6: 3795 if (addr->addr32[3] == 0xffffffff) { 3796 addr->addr32[3] = 0; 3797 if (addr->addr32[2] == 0xffffffff) { 3798 addr->addr32[2] = 0; 3799 if (addr->addr32[1] == 0xffffffff) { 3800 addr->addr32[1] = 0; 3801 addr->addr32[0] = 3802 htonl(ntohl(addr->addr32[0]) + 1); 3803 } else 3804 addr->addr32[1] = 3805 htonl(ntohl(addr->addr32[1]) + 1); 3806 } else 3807 addr->addr32[2] = 3808 htonl(ntohl(addr->addr32[2]) + 1); 3809 } else 3810 addr->addr32[3] = 3811 htonl(ntohl(addr->addr32[3]) + 1); 3812 break; 3813 } 3814} 3815#endif /* INET6 */ 3816 3817void 3818pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3819{ 3820 /* 3821 * Modern rules use the same flags in rules as they do in states. 3822 */ 3823 a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID| 3824 PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO)); 3825 3826 /* 3827 * Old-style scrub rules have different flags which need to be translated. 3828 */ 3829 if (r->rule_flag & PFRULE_RANDOMID) 3830 a->flags |= PFSTATE_RANDOMID; 3831 if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) { 3832 a->flags |= PFSTATE_SETTOS; 3833 a->set_tos = r->set_tos; 3834 } 3835 3836 if (r->qid) 3837 a->qid = r->qid; 3838 if (r->pqid) 3839 a->pqid = r->pqid; 3840 if (r->rtableid >= 0) 3841 a->rtableid = r->rtableid; 3842 a->log |= r->log; 3843 if (r->min_ttl) 3844 a->min_ttl = r->min_ttl; 3845 if (r->max_mss) 3846 a->max_mss = r->max_mss; 3847 if (r->dnpipe) 3848 a->dnpipe = r->dnpipe; 3849 if (r->dnrpipe) 3850 a->dnrpipe = r->dnrpipe; 3851 if (r->dnpipe || r->dnrpipe) { 3852 if (r->free_flags & PFRULE_DN_IS_PIPE) 3853 a->flags |= PFSTATE_DN_IS_PIPE; 3854 else 3855 a->flags &= ~PFSTATE_DN_IS_PIPE; 3856 } 3857 if (r->scrub_flags & PFSTATE_SETPRIO) { 3858 a->set_prio[0] = r->set_prio[0]; 3859 a->set_prio[1] = r->set_prio[1]; 3860 } 3861} 3862 3863int 3864pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m) 3865{ 3866 struct pf_addr *saddr, *daddr; 3867 u_int16_t sport, dport; 3868 struct inpcbinfo *pi; 3869 struct inpcb *inp; 3870 3871 pd->lookup.uid = UID_MAX; 3872 pd->lookup.gid = GID_MAX; 3873 3874 switch (pd->proto) { 3875 case IPPROTO_TCP: 3876 sport = pd->hdr.tcp.th_sport; 3877 dport = pd->hdr.tcp.th_dport; 3878 pi = &V_tcbinfo; 3879 break; 3880 case IPPROTO_UDP: 3881 sport = pd->hdr.udp.uh_sport; 3882 dport = pd->hdr.udp.uh_dport; 3883 pi = &V_udbinfo; 3884 break; 3885 default: 3886 return (-1); 3887 } 3888 if (pd->dir == PF_IN) { 3889 saddr = pd->src; 3890 daddr = pd->dst; 3891 } else { 3892 u_int16_t p; 3893 3894 p = sport; 3895 sport = dport; 3896 dport = p; 3897 saddr = pd->dst; 3898 daddr = pd->src; 3899 } 3900 switch (pd->af) { 3901#ifdef INET 3902 case AF_INET: 3903 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3904 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3905 if (inp == NULL) { 3906 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3907 daddr->v4, dport, INPLOOKUP_WILDCARD | 3908 INPLOOKUP_RLOCKPCB, NULL, m); 3909 if (inp == NULL) 3910 return (-1); 3911 } 3912 break; 3913#endif /* INET */ 3914#ifdef INET6 3915 case AF_INET6: 3916 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3917 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3918 if (inp == NULL) { 3919 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3920 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3921 INPLOOKUP_RLOCKPCB, NULL, m); 3922 if (inp == NULL) 3923 return (-1); 3924 } 3925 break; 3926#endif /* INET6 */ 3927 3928 default: 3929 return (-1); 3930 } 3931 INP_RLOCK_ASSERT(inp); 3932 pd->lookup.uid = inp->inp_cred->cr_uid; 3933 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3934 INP_RUNLOCK(inp); 3935 3936 return (1); 3937} 3938 3939u_int8_t 3940pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3941{ 3942 int hlen; 3943 u_int8_t hdr[60]; 3944 u_int8_t *opt, optlen; 3945 u_int8_t wscale = 0; 3946 3947 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3948 if (hlen <= sizeof(struct tcphdr)) 3949 return (0); 3950 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3951 return (0); 3952 opt = hdr + sizeof(struct tcphdr); 3953 hlen -= sizeof(struct tcphdr); 3954 while (hlen >= 3) { 3955 switch (*opt) { 3956 case TCPOPT_EOL: 3957 case TCPOPT_NOP: 3958 ++opt; 3959 --hlen; 3960 break; 3961 case TCPOPT_WINDOW: 3962 wscale = opt[2]; 3963 if (wscale > TCP_MAX_WINSHIFT) 3964 wscale = TCP_MAX_WINSHIFT; 3965 wscale |= PF_WSCALE_FLAG; 3966 /* FALLTHROUGH */ 3967 default: 3968 optlen = opt[1]; 3969 if (optlen < 2) 3970 optlen = 2; 3971 hlen -= optlen; 3972 opt += optlen; 3973 break; 3974 } 3975 } 3976 return (wscale); 3977} 3978 3979u_int16_t 3980pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3981{ 3982 int hlen; 3983 u_int8_t hdr[60]; 3984 u_int8_t *opt, optlen; 3985 u_int16_t mss = V_tcp_mssdflt; 3986 3987 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3988 if (hlen <= sizeof(struct tcphdr)) 3989 return (0); 3990 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3991 return (0); 3992 opt = hdr + sizeof(struct tcphdr); 3993 hlen -= sizeof(struct tcphdr); 3994 while (hlen >= TCPOLEN_MAXSEG) { 3995 switch (*opt) { 3996 case TCPOPT_EOL: 3997 case TCPOPT_NOP: 3998 ++opt; 3999 --hlen; 4000 break; 4001 case TCPOPT_MAXSEG: 4002 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 4003 NTOHS(mss); 4004 /* FALLTHROUGH */ 4005 default: 4006 optlen = opt[1]; 4007 if (optlen < 2) 4008 optlen = 2; 4009 hlen -= optlen; 4010 opt += optlen; 4011 break; 4012 } 4013 } 4014 return (mss); 4015} 4016 4017static u_int16_t 4018pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 4019{ 4020 struct nhop_object *nh; 4021#ifdef INET6 4022 struct in6_addr dst6; 4023 uint32_t scopeid; 4024#endif /* INET6 */ 4025 int hlen = 0; 4026 uint16_t mss = 0; 4027 4028 NET_EPOCH_ASSERT(); 4029 4030 switch (af) { 4031#ifdef INET 4032 case AF_INET: 4033 hlen = sizeof(struct ip); 4034 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 4035 if (nh != NULL) 4036 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4037 break; 4038#endif /* INET */ 4039#ifdef INET6 4040 case AF_INET6: 4041 hlen = sizeof(struct ip6_hdr); 4042 in6_splitscope(&addr->v6, &dst6, &scopeid); 4043 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 4044 if (nh != NULL) 4045 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 4046 break; 4047#endif /* INET6 */ 4048 } 4049 4050 mss = max(V_tcp_mssdflt, mss); 4051 mss = min(mss, offer); 4052 mss = max(mss, 64); /* sanity - at least max opt space */ 4053 return (mss); 4054} 4055 4056static u_int32_t 4057pf_tcp_iss(struct pf_pdesc *pd) 4058{ 4059 MD5_CTX ctx; 4060 u_int32_t digest[4]; 4061 4062 if (V_pf_tcp_secret_init == 0) { 4063 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 4064 MD5Init(&V_pf_tcp_secret_ctx); 4065 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 4066 sizeof(V_pf_tcp_secret)); 4067 V_pf_tcp_secret_init = 1; 4068 } 4069 4070 ctx = V_pf_tcp_secret_ctx; 4071 4072 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 4073 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 4074 if (pd->af == AF_INET6) { 4075 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 4076 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 4077 } else { 4078 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 4079 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 4080 } 4081 MD5Final((u_char *)digest, &ctx); 4082 V_pf_tcp_iss_off += 4096; 4083#define ISN_RANDOM_INCREMENT (4096 - 1) 4084 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 4085 V_pf_tcp_iss_off); 4086#undef ISN_RANDOM_INCREMENT 4087} 4088 4089static bool 4090pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r) 4091{ 4092 bool match = true; 4093 4094 /* Always matches if not set */ 4095 if (! r->isset) 4096 return (!r->neg); 4097 4098 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4099 if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) { 4100 match = false; 4101 break; 4102 } 4103 } 4104 4105 return (match ^ r->neg); 4106} 4107 4108static int 4109pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag) 4110{ 4111 if (*tag == -1) 4112 *tag = mtag; 4113 4114 return ((!r->match_tag_not && r->match_tag == *tag) || 4115 (r->match_tag_not && r->match_tag != *tag)); 4116} 4117 4118static void 4119pf_bridge_to(struct ifnet *ifp, struct mbuf *m) 4120{ 4121 /* If we don't have the interface drop the packet. */ 4122 if (ifp == NULL) { 4123 m_freem(m); 4124 return; 4125 } 4126 4127 switch (ifp->if_type) { 4128 case IFT_ETHER: 4129 case IFT_XETHER: 4130 case IFT_L2VLAN: 4131 case IFT_BRIDGE: 4132 case IFT_IEEE8023ADLAG: 4133 break; 4134 default: 4135 m_freem(m); 4136 return; 4137 } 4138 4139 ifp->if_transmit(ifp, m); 4140} 4141 4142static int 4143pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0) 4144{ 4145#ifdef INET 4146 struct ip ip; 4147#endif 4148#ifdef INET6 4149 struct ip6_hdr ip6; 4150#endif 4151 struct mbuf *m = *m0; 4152 struct ether_header *e; 4153 struct pf_keth_rule *r, *rm, *a = NULL; 4154 struct pf_keth_ruleset *ruleset = NULL; 4155 struct pf_mtag *mtag; 4156 struct pf_keth_ruleq *rules; 4157 struct pf_addr *src = NULL, *dst = NULL; 4158 struct pfi_kkif *bridge_to; 4159 sa_family_t af = 0; 4160 uint16_t proto; 4161 int asd = 0, match = 0; 4162 int tag = -1; 4163 uint8_t action; 4164 struct pf_keth_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4165 4166 MPASS(kif->pfik_ifp->if_vnet == curvnet); 4167 NET_EPOCH_ASSERT(); 4168 4169 PF_RULES_RLOCK_TRACKER; 4170 4171 SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m); 4172 4173 mtag = pf_find_mtag(m); 4174 if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 4175 /* Dummynet re-injects packets after they've 4176 * completed their delay. We've already 4177 * processed them, so pass unconditionally. */ 4178 4179 /* But only once. We may see the packet multiple times (e.g. 4180 * PFIL_IN/PFIL_OUT). */ 4181 pf_dummynet_flag_remove(m, mtag); 4182 4183 return (PF_PASS); 4184 } 4185 4186 ruleset = V_pf_keth; 4187 rules = ck_pr_load_ptr(&ruleset->active.rules); 4188 r = TAILQ_FIRST(rules); 4189 rm = NULL; 4190 4191 e = mtod(m, struct ether_header *); 4192 proto = ntohs(e->ether_type); 4193 4194 switch (proto) { 4195#ifdef INET 4196 case ETHERTYPE_IP: { 4197 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4198 sizeof(ip))) 4199 return (PF_DROP); 4200 4201 af = AF_INET; 4202 m_copydata(m, sizeof(struct ether_header), sizeof(ip), 4203 (caddr_t)&ip); 4204 src = (struct pf_addr *)&ip.ip_src; 4205 dst = (struct pf_addr *)&ip.ip_dst; 4206 break; 4207 } 4208#endif /* INET */ 4209#ifdef INET6 4210 case ETHERTYPE_IPV6: { 4211 if (m_length(m, NULL) < (sizeof(struct ether_header) + 4212 sizeof(ip6))) 4213 return (PF_DROP); 4214 4215 af = AF_INET6; 4216 m_copydata(m, sizeof(struct ether_header), sizeof(ip6), 4217 (caddr_t)&ip6); 4218 src = (struct pf_addr *)&ip6.ip6_src; 4219 dst = (struct pf_addr *)&ip6.ip6_dst; 4220 break; 4221 } 4222#endif /* INET6 */ 4223 } 4224 4225 PF_RULES_RLOCK(); 4226 4227 while (r != NULL) { 4228 counter_u64_add(r->evaluations, 1); 4229 SDT_PROBE2(pf, eth, test_rule, test, r->nr, r); 4230 4231 if (pfi_kkif_match(r->kif, kif) == r->ifnot) { 4232 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4233 "kif"); 4234 r = r->skip[PFE_SKIP_IFP].ptr; 4235 } 4236 else if (r->direction && r->direction != dir) { 4237 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4238 "dir"); 4239 r = r->skip[PFE_SKIP_DIR].ptr; 4240 } 4241 else if (r->proto && r->proto != proto) { 4242 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4243 "proto"); 4244 r = r->skip[PFE_SKIP_PROTO].ptr; 4245 } 4246 else if (! pf_match_eth_addr(e->ether_shost, &r->src)) { 4247 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4248 "src"); 4249 r = r->skip[PFE_SKIP_SRC_ADDR].ptr; 4250 } 4251 else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) { 4252 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4253 "dst"); 4254 r = r->skip[PFE_SKIP_DST_ADDR].ptr; 4255 } 4256 else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af, 4257 r->ipsrc.neg, kif, M_GETFIB(m))) { 4258 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4259 "ip_src"); 4260 r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr; 4261 } 4262 else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af, 4263 r->ipdst.neg, kif, M_GETFIB(m))) { 4264 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4265 "ip_dst"); 4266 r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr; 4267 } 4268 else if (r->match_tag && !pf_match_eth_tag(m, r, &tag, 4269 mtag ? mtag->tag : 0)) { 4270 SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r, 4271 "match_tag"); 4272 r = TAILQ_NEXT(r, entries); 4273 } 4274 else { 4275 if (r->tag) 4276 tag = r->tag; 4277 if (r->anchor == NULL) { 4278 /* Rule matches */ 4279 rm = r; 4280 4281 SDT_PROBE2(pf, eth, test_rule, match, r->nr, r); 4282 4283 if (r->quick) 4284 break; 4285 4286 r = TAILQ_NEXT(r, entries); 4287 } else { 4288 pf_step_into_keth_anchor(anchor_stack, &asd, 4289 &ruleset, &r, &a, &match); 4290 } 4291 } 4292 if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd, 4293 &ruleset, &r, &a, &match)) 4294 break; 4295 } 4296 4297 r = rm; 4298 4299 SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r); 4300 4301 /* Default to pass. */ 4302 if (r == NULL) { 4303 PF_RULES_RUNLOCK(); 4304 return (PF_PASS); 4305 } 4306 4307 /* Execute action. */ 4308 counter_u64_add(r->packets[dir == PF_OUT], 1); 4309 counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL)); 4310 pf_update_timestamp(r); 4311 4312 /* Shortcut. Don't tag if we're just going to drop anyway. */ 4313 if (r->action == PF_DROP) { 4314 PF_RULES_RUNLOCK(); 4315 return (PF_DROP); 4316 } 4317 4318 if (tag > 0) { 4319 if (mtag == NULL) 4320 mtag = pf_get_mtag(m); 4321 if (mtag == NULL) { 4322 PF_RULES_RUNLOCK(); 4323 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4324 return (PF_DROP); 4325 } 4326 mtag->tag = tag; 4327 } 4328 4329 if (r->qid != 0) { 4330 if (mtag == NULL) 4331 mtag = pf_get_mtag(m); 4332 if (mtag == NULL) { 4333 PF_RULES_RUNLOCK(); 4334 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4335 return (PF_DROP); 4336 } 4337 mtag->qid = r->qid; 4338 } 4339 4340 action = r->action; 4341 bridge_to = r->bridge_to; 4342 4343 /* Dummynet */ 4344 if (r->dnpipe) { 4345 struct ip_fw_args dnflow; 4346 4347 /* Drop packet if dummynet is not loaded. */ 4348 if (ip_dn_io_ptr == NULL) { 4349 PF_RULES_RUNLOCK(); 4350 m_freem(m); 4351 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4352 return (PF_DROP); 4353 } 4354 if (mtag == NULL) 4355 mtag = pf_get_mtag(m); 4356 if (mtag == NULL) { 4357 PF_RULES_RUNLOCK(); 4358 counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1); 4359 return (PF_DROP); 4360 } 4361 4362 bzero(&dnflow, sizeof(dnflow)); 4363 4364 /* We don't have port numbers here, so we set 0. That means 4365 * that we'll be somewhat limited in distinguishing flows (i.e. 4366 * only based on IP addresses, not based on port numbers), but 4367 * it's better than nothing. */ 4368 dnflow.f_id.dst_port = 0; 4369 dnflow.f_id.src_port = 0; 4370 dnflow.f_id.proto = 0; 4371 4372 dnflow.rule.info = r->dnpipe; 4373 dnflow.rule.info |= IPFW_IS_DUMMYNET; 4374 if (r->dnflags & PFRULE_DN_IS_PIPE) 4375 dnflow.rule.info |= IPFW_IS_PIPE; 4376 4377 dnflow.f_id.extra = dnflow.rule.info; 4378 4379 dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT; 4380 dnflow.flags |= IPFW_ARGS_ETHER; 4381 dnflow.ifp = kif->pfik_ifp; 4382 4383 switch (af) { 4384 case AF_INET: 4385 dnflow.f_id.addr_type = 4; 4386 dnflow.f_id.src_ip = src->v4.s_addr; 4387 dnflow.f_id.dst_ip = dst->v4.s_addr; 4388 break; 4389 case AF_INET6: 4390 dnflow.flags |= IPFW_ARGS_IP6; 4391 dnflow.f_id.addr_type = 6; 4392 dnflow.f_id.src_ip6 = src->v6; 4393 dnflow.f_id.dst_ip6 = dst->v6; 4394 break; 4395 } 4396 4397 PF_RULES_RUNLOCK(); 4398 4399 mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 4400 ip_dn_io_ptr(m0, &dnflow); 4401 if (*m0 != NULL) 4402 pf_dummynet_flag_remove(m, mtag); 4403 } else { 4404 PF_RULES_RUNLOCK(); 4405 } 4406 4407 if (action == PF_PASS && bridge_to) { 4408 pf_bridge_to(bridge_to->pfik_ifp, *m0); 4409 *m0 = NULL; /* We've eaten the packet. */ 4410 } 4411 4412 return (action); 4413} 4414 4415static int 4416pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif, 4417 struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am, 4418 struct pf_kruleset **rsm, struct inpcb *inp) 4419{ 4420 struct pf_krule *nr = NULL; 4421 struct pf_addr * const saddr = pd->src; 4422 struct pf_addr * const daddr = pd->dst; 4423 sa_family_t af = pd->af; 4424 struct pf_krule *r, *a = NULL; 4425 struct pf_kruleset *ruleset = NULL; 4426 struct pf_krule_slist match_rules; 4427 struct pf_krule_item *ri; 4428 struct pf_ksrc_node *nsn = NULL; 4429 struct tcphdr *th = &pd->hdr.tcp; 4430 struct pf_state_key *sk = NULL, *nk = NULL; 4431 u_short reason; 4432 int rewrite = 0, hdrlen = 0; 4433 int tag = -1; 4434 int asd = 0; 4435 int match = 0; 4436 int state_icmp = 0; 4437 u_int16_t sport = 0, dport = 0; 4438 u_int16_t bproto_sum = 0, bip_sum = 0; 4439 u_int8_t icmptype = 0, icmpcode = 0; 4440 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4441 4442 PF_RULES_RASSERT(); 4443 4444 if (inp != NULL) { 4445 INP_LOCK_ASSERT(inp); 4446 pd->lookup.uid = inp->inp_cred->cr_uid; 4447 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 4448 pd->lookup.done = 1; 4449 } 4450 4451 switch (pd->proto) { 4452 case IPPROTO_TCP: 4453 sport = th->th_sport; 4454 dport = th->th_dport; 4455 hdrlen = sizeof(*th); 4456 break; 4457 case IPPROTO_UDP: 4458 sport = pd->hdr.udp.uh_sport; 4459 dport = pd->hdr.udp.uh_dport; 4460 hdrlen = sizeof(pd->hdr.udp); 4461 break; 4462 case IPPROTO_SCTP: 4463 sport = pd->hdr.sctp.src_port; 4464 dport = pd->hdr.sctp.dest_port; 4465 hdrlen = sizeof(pd->hdr.sctp); 4466 break; 4467#ifdef INET 4468 case IPPROTO_ICMP: 4469 if (pd->af != AF_INET) 4470 break; 4471 sport = dport = pd->hdr.icmp.icmp_id; 4472 hdrlen = sizeof(pd->hdr.icmp); 4473 icmptype = pd->hdr.icmp.icmp_type; 4474 icmpcode = pd->hdr.icmp.icmp_code; 4475 4476 if (icmptype == ICMP_UNREACH || 4477 icmptype == ICMP_SOURCEQUENCH || 4478 icmptype == ICMP_REDIRECT || 4479 icmptype == ICMP_TIMXCEED || 4480 icmptype == ICMP_PARAMPROB) 4481 state_icmp++; 4482 break; 4483#endif /* INET */ 4484#ifdef INET6 4485 case IPPROTO_ICMPV6: 4486 if (af != AF_INET6) 4487 break; 4488 sport = dport = pd->hdr.icmp6.icmp6_id; 4489 hdrlen = sizeof(pd->hdr.icmp6); 4490 icmptype = pd->hdr.icmp6.icmp6_type; 4491 icmpcode = pd->hdr.icmp6.icmp6_code; 4492 4493 if (icmptype == ICMP6_DST_UNREACH || 4494 icmptype == ICMP6_PACKET_TOO_BIG || 4495 icmptype == ICMP6_TIME_EXCEEDED || 4496 icmptype == ICMP6_PARAM_PROB) 4497 state_icmp++; 4498 break; 4499#endif /* INET6 */ 4500 default: 4501 sport = dport = hdrlen = 0; 4502 break; 4503 } 4504 4505 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4506 4507 /* check packet for BINAT/NAT/RDR */ 4508 if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk, 4509 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 4510 KASSERT(sk != NULL, ("%s: null sk", __func__)); 4511 KASSERT(nk != NULL, ("%s: null nk", __func__)); 4512 4513 if (nr->log) { 4514 PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a, 4515 ruleset, pd, 1); 4516 } 4517 4518 if (pd->ip_sum) 4519 bip_sum = *pd->ip_sum; 4520 4521 switch (pd->proto) { 4522 case IPPROTO_TCP: 4523 bproto_sum = th->th_sum; 4524 pd->proto_sum = &th->th_sum; 4525 4526 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4527 nk->port[pd->sidx] != sport) { 4528 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 4529 &th->th_sum, &nk->addr[pd->sidx], 4530 nk->port[pd->sidx], 0, af); 4531 pd->sport = &th->th_sport; 4532 sport = th->th_sport; 4533 } 4534 4535 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4536 nk->port[pd->didx] != dport) { 4537 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 4538 &th->th_sum, &nk->addr[pd->didx], 4539 nk->port[pd->didx], 0, af); 4540 dport = th->th_dport; 4541 pd->dport = &th->th_dport; 4542 } 4543 rewrite++; 4544 break; 4545 case IPPROTO_UDP: 4546 bproto_sum = pd->hdr.udp.uh_sum; 4547 pd->proto_sum = &pd->hdr.udp.uh_sum; 4548 4549 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4550 nk->port[pd->sidx] != sport) { 4551 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 4552 pd->ip_sum, &pd->hdr.udp.uh_sum, 4553 &nk->addr[pd->sidx], 4554 nk->port[pd->sidx], 1, af); 4555 sport = pd->hdr.udp.uh_sport; 4556 pd->sport = &pd->hdr.udp.uh_sport; 4557 } 4558 4559 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4560 nk->port[pd->didx] != dport) { 4561 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 4562 pd->ip_sum, &pd->hdr.udp.uh_sum, 4563 &nk->addr[pd->didx], 4564 nk->port[pd->didx], 1, af); 4565 dport = pd->hdr.udp.uh_dport; 4566 pd->dport = &pd->hdr.udp.uh_dport; 4567 } 4568 rewrite++; 4569 break; 4570 case IPPROTO_SCTP: { 4571 uint16_t checksum = 0; 4572 4573 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 4574 nk->port[pd->sidx] != sport) { 4575 pf_change_ap(m, saddr, &pd->hdr.sctp.src_port, 4576 pd->ip_sum, &checksum, 4577 &nk->addr[pd->sidx], 4578 nk->port[pd->sidx], 1, af); 4579 } 4580 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 4581 nk->port[pd->didx] != dport) { 4582 pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port, 4583 pd->ip_sum, &checksum, 4584 &nk->addr[pd->didx], 4585 nk->port[pd->didx], 1, af); 4586 } 4587 break; 4588 } 4589#ifdef INET 4590 case IPPROTO_ICMP: 4591 nk->port[0] = nk->port[1]; 4592 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 4593 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 4594 nk->addr[pd->sidx].v4.s_addr, 0); 4595 4596 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 4597 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 4598 nk->addr[pd->didx].v4.s_addr, 0); 4599 4600 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 4601 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 4602 pd->hdr.icmp.icmp_cksum, sport, 4603 nk->port[1], 0); 4604 pd->hdr.icmp.icmp_id = nk->port[1]; 4605 pd->sport = &pd->hdr.icmp.icmp_id; 4606 } 4607 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 4608 break; 4609#endif /* INET */ 4610#ifdef INET6 4611 case IPPROTO_ICMPV6: 4612 nk->port[0] = nk->port[1]; 4613 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 4614 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 4615 &nk->addr[pd->sidx], 0); 4616 4617 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 4618 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 4619 &nk->addr[pd->didx], 0); 4620 rewrite++; 4621 break; 4622#endif /* INET */ 4623 default: 4624 switch (af) { 4625#ifdef INET 4626 case AF_INET: 4627 if (PF_ANEQ(saddr, 4628 &nk->addr[pd->sidx], AF_INET)) 4629 pf_change_a(&saddr->v4.s_addr, 4630 pd->ip_sum, 4631 nk->addr[pd->sidx].v4.s_addr, 0); 4632 4633 if (PF_ANEQ(daddr, 4634 &nk->addr[pd->didx], AF_INET)) 4635 pf_change_a(&daddr->v4.s_addr, 4636 pd->ip_sum, 4637 nk->addr[pd->didx].v4.s_addr, 0); 4638 break; 4639#endif /* INET */ 4640#ifdef INET6 4641 case AF_INET6: 4642 if (PF_ANEQ(saddr, 4643 &nk->addr[pd->sidx], AF_INET6)) 4644 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 4645 4646 if (PF_ANEQ(daddr, 4647 &nk->addr[pd->didx], AF_INET6)) 4648 PF_ACPY(daddr, &nk->addr[pd->didx], af); 4649 break; 4650#endif /* INET */ 4651 } 4652 break; 4653 } 4654 if (nr->natpass) 4655 r = NULL; 4656 pd->nat_rule = nr; 4657 } 4658 4659 SLIST_INIT(&match_rules); 4660 while (r != NULL) { 4661 pf_counter_u64_add(&r->evaluations, 1); 4662 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4663 r = r->skip[PF_SKIP_IFP].ptr; 4664 else if (r->direction && r->direction != pd->dir) 4665 r = r->skip[PF_SKIP_DIR].ptr; 4666 else if (r->af && r->af != af) 4667 r = r->skip[PF_SKIP_AF].ptr; 4668 else if (r->proto && r->proto != pd->proto) 4669 r = r->skip[PF_SKIP_PROTO].ptr; 4670 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 4671 r->src.neg, kif, M_GETFIB(m))) 4672 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4673 /* tcp/udp only. port_op always 0 in other cases */ 4674 else if (r->src.port_op && !pf_match_port(r->src.port_op, 4675 r->src.port[0], r->src.port[1], sport)) 4676 r = r->skip[PF_SKIP_SRC_PORT].ptr; 4677 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 4678 r->dst.neg, NULL, M_GETFIB(m))) 4679 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4680 /* tcp/udp only. port_op always 0 in other cases */ 4681 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 4682 r->dst.port[0], r->dst.port[1], dport)) 4683 r = r->skip[PF_SKIP_DST_PORT].ptr; 4684 /* icmp only. type always 0 in other cases */ 4685 else if (r->type && r->type != icmptype + 1) 4686 r = TAILQ_NEXT(r, entries); 4687 /* icmp only. type always 0 in other cases */ 4688 else if (r->code && r->code != icmpcode + 1) 4689 r = TAILQ_NEXT(r, entries); 4690 else if (r->tos && !(r->tos == pd->tos)) 4691 r = TAILQ_NEXT(r, entries); 4692 else if (r->rule_flag & PFRULE_FRAGMENT) 4693 r = TAILQ_NEXT(r, entries); 4694 else if (pd->proto == IPPROTO_TCP && 4695 (r->flagset & th->th_flags) != r->flags) 4696 r = TAILQ_NEXT(r, entries); 4697 /* tcp/udp only. uid.op always 0 in other cases */ 4698 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 4699 pf_socket_lookup(pd, m), 1)) && 4700 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 4701 pd->lookup.uid)) 4702 r = TAILQ_NEXT(r, entries); 4703 /* tcp/udp only. gid.op always 0 in other cases */ 4704 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 4705 pf_socket_lookup(pd, m), 1)) && 4706 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 4707 pd->lookup.gid)) 4708 r = TAILQ_NEXT(r, entries); 4709 else if (r->prio && 4710 !pf_match_ieee8021q_pcp(r->prio, m)) 4711 r = TAILQ_NEXT(r, entries); 4712 else if (r->prob && 4713 r->prob <= arc4random()) 4714 r = TAILQ_NEXT(r, entries); 4715 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4716 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4717 r = TAILQ_NEXT(r, entries); 4718 else if (r->os_fingerprint != PF_OSFP_ANY && 4719 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 4720 pf_osfp_fingerprint(pd, m, off, th), 4721 r->os_fingerprint))) 4722 r = TAILQ_NEXT(r, entries); 4723 else { 4724 if (r->tag) 4725 tag = r->tag; 4726 if (r->anchor == NULL) { 4727 if (r->action == PF_MATCH) { 4728 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 4729 if (ri == NULL) { 4730 REASON_SET(&reason, PFRES_MEMORY); 4731 goto cleanup; 4732 } 4733 ri->r = r; 4734 SLIST_INSERT_HEAD(&match_rules, ri, entry); 4735 pf_counter_u64_critical_enter(); 4736 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 4737 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 4738 pf_counter_u64_critical_exit(); 4739 pf_rule_to_actions(r, &pd->act); 4740 if (r->log) 4741 PFLOG_PACKET(kif, m, af, 4742 r->action, PFRES_MATCH, r, 4743 a, ruleset, pd, 1); 4744 } else { 4745 match = 1; 4746 *rm = r; 4747 *am = a; 4748 *rsm = ruleset; 4749 } 4750 if ((*rm)->quick) 4751 break; 4752 r = TAILQ_NEXT(r, entries); 4753 } else 4754 pf_step_into_anchor(anchor_stack, &asd, 4755 &ruleset, PF_RULESET_FILTER, &r, &a, 4756 &match); 4757 } 4758 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4759 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4760 break; 4761 } 4762 r = *rm; 4763 a = *am; 4764 ruleset = *rsm; 4765 4766 REASON_SET(&reason, PFRES_MATCH); 4767 4768 /* apply actions for last matching pass/block rule */ 4769 pf_rule_to_actions(r, &pd->act); 4770 4771 if (r->log) { 4772 if (rewrite) 4773 m_copyback(m, off, hdrlen, pd->hdr.any); 4774 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 4775 } 4776 4777 if ((r->action == PF_DROP) && 4778 ((r->rule_flag & PFRULE_RETURNRST) || 4779 (r->rule_flag & PFRULE_RETURNICMP) || 4780 (r->rule_flag & PFRULE_RETURN))) { 4781 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 4782 bip_sum, hdrlen, &reason, r->rtableid); 4783 } 4784 4785 if (r->action == PF_DROP) 4786 goto cleanup; 4787 4788 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4789 REASON_SET(&reason, PFRES_MEMORY); 4790 goto cleanup; 4791 } 4792 if (pd->act.rtableid >= 0) 4793 M_SETFIB(m, pd->act.rtableid); 4794 4795 if (!state_icmp && (r->keep_state || nr != NULL || 4796 (pd->flags & PFDESC_TCP_NORM))) { 4797 int action; 4798 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 4799 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 4800 hdrlen, &match_rules); 4801 if (action != PF_PASS) { 4802 if (action == PF_DROP && 4803 (r->rule_flag & PFRULE_RETURN)) 4804 pf_return(r, nr, pd, sk, off, m, th, kif, 4805 bproto_sum, bip_sum, hdrlen, &reason, 4806 pd->act.rtableid); 4807 return (action); 4808 } 4809 } else { 4810 while ((ri = SLIST_FIRST(&match_rules))) { 4811 SLIST_REMOVE_HEAD(&match_rules, entry); 4812 free(ri, M_PF_RULE_ITEM); 4813 } 4814 4815 uma_zfree(V_pf_state_key_z, sk); 4816 uma_zfree(V_pf_state_key_z, nk); 4817 } 4818 4819 /* copy back packet headers if we performed NAT operations */ 4820 if (rewrite) 4821 m_copyback(m, off, hdrlen, pd->hdr.any); 4822 4823 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 4824 pd->dir == PF_OUT && 4825 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 4826 /* 4827 * We want the state created, but we dont 4828 * want to send this in case a partner 4829 * firewall has to know about it to allow 4830 * replies through it. 4831 */ 4832 return (PF_DEFER); 4833 4834 return (PF_PASS); 4835 4836cleanup: 4837 while ((ri = SLIST_FIRST(&match_rules))) { 4838 SLIST_REMOVE_HEAD(&match_rules, entry); 4839 free(ri, M_PF_RULE_ITEM); 4840 } 4841 4842 uma_zfree(V_pf_state_key_z, sk); 4843 uma_zfree(V_pf_state_key_z, nk); 4844 return (PF_DROP); 4845} 4846 4847static int 4848pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 4849 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 4850 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 4851 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 4852 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 4853 struct pf_krule_slist *match_rules) 4854{ 4855 struct pf_kstate *s = NULL; 4856 struct pf_ksrc_node *sn = NULL; 4857 struct tcphdr *th = &pd->hdr.tcp; 4858 u_int16_t mss = V_tcp_mssdflt; 4859 u_short reason, sn_reason; 4860 struct pf_krule_item *ri; 4861 4862 /* check maximums */ 4863 if (r->max_states && 4864 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 4865 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 4866 REASON_SET(&reason, PFRES_MAXSTATES); 4867 goto csfailed; 4868 } 4869 /* src node for filter rule */ 4870 if ((r->rule_flag & PFRULE_SRCTRACK || 4871 r->rpool.opts & PF_POOL_STICKYADDR) && 4872 (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) { 4873 REASON_SET(&reason, sn_reason); 4874 goto csfailed; 4875 } 4876 /* src node for translation rule */ 4877 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 4878 (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], 4879 pd->af)) != 0 ) { 4880 REASON_SET(&reason, sn_reason); 4881 goto csfailed; 4882 } 4883 s = pf_alloc_state(M_NOWAIT); 4884 if (s == NULL) { 4885 REASON_SET(&reason, PFRES_MEMORY); 4886 goto csfailed; 4887 } 4888 s->rule.ptr = r; 4889 s->nat_rule.ptr = nr; 4890 s->anchor.ptr = a; 4891 bcopy(match_rules, &s->match_rules, sizeof(s->match_rules)); 4892 memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions)); 4893 4894 STATE_INC_COUNTERS(s); 4895 if (r->allow_opts) 4896 s->state_flags |= PFSTATE_ALLOWOPTS; 4897 if (r->rule_flag & PFRULE_STATESLOPPY) 4898 s->state_flags |= PFSTATE_SLOPPY; 4899 if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */ 4900 s->state_flags |= PFSTATE_SCRUB_TCP; 4901 if ((r->rule_flag & PFRULE_PFLOW) || 4902 (nr != NULL && nr->rule_flag & PFRULE_PFLOW)) 4903 s->state_flags |= PFSTATE_PFLOW; 4904 4905 s->act.log = pd->act.log & PF_LOG_ALL; 4906 s->sync_state = PFSYNC_S_NONE; 4907 s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */ 4908 4909 if (nr != NULL) 4910 s->act.log |= nr->log & PF_LOG_ALL; 4911 switch (pd->proto) { 4912 case IPPROTO_TCP: 4913 s->src.seqlo = ntohl(th->th_seq); 4914 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4915 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4916 r->keep_state == PF_STATE_MODULATE) { 4917 /* Generate sequence number modulator */ 4918 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4919 0) 4920 s->src.seqdiff = 1; 4921 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4922 htonl(s->src.seqlo + s->src.seqdiff), 0); 4923 *rewrite = 1; 4924 } else 4925 s->src.seqdiff = 0; 4926 if (th->th_flags & TH_SYN) { 4927 s->src.seqhi++; 4928 s->src.wscale = pf_get_wscale(m, off, 4929 th->th_off, pd->af); 4930 } 4931 s->src.max_win = MAX(ntohs(th->th_win), 1); 4932 if (s->src.wscale & PF_WSCALE_MASK) { 4933 /* Remove scale factor from initial window */ 4934 int win = s->src.max_win; 4935 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4936 s->src.max_win = (win - 1) >> 4937 (s->src.wscale & PF_WSCALE_MASK); 4938 } 4939 if (th->th_flags & TH_FIN) 4940 s->src.seqhi++; 4941 s->dst.seqhi = 1; 4942 s->dst.max_win = 1; 4943 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4944 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4945 s->timeout = PFTM_TCP_FIRST_PACKET; 4946 atomic_add_32(&V_pf_status.states_halfopen, 1); 4947 break; 4948 case IPPROTO_UDP: 4949 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4950 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4951 s->timeout = PFTM_UDP_FIRST_PACKET; 4952 break; 4953 case IPPROTO_SCTP: 4954 pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT); 4955 pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED); 4956 s->timeout = PFTM_SCTP_FIRST_PACKET; 4957 break; 4958 case IPPROTO_ICMP: 4959#ifdef INET6 4960 case IPPROTO_ICMPV6: 4961#endif 4962 s->timeout = PFTM_ICMP_FIRST_PACKET; 4963 break; 4964 default: 4965 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4966 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4967 s->timeout = PFTM_OTHER_FIRST_PACKET; 4968 } 4969 4970 if (r->rt) { 4971 /* pf_map_addr increases the reason counters */ 4972 if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr, 4973 &s->rt_kif, NULL, &sn)) != 0) 4974 goto csfailed; 4975 s->rt = r->rt; 4976 } 4977 4978 s->creation = s->expire = pf_get_uptime(); 4979 4980 if (sn != NULL) 4981 s->src_node = sn; 4982 if (nsn != NULL) { 4983 /* XXX We only modify one side for now. */ 4984 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4985 s->nat_src_node = nsn; 4986 } 4987 if (pd->proto == IPPROTO_TCP) { 4988 if (s->state_flags & PFSTATE_SCRUB_TCP && 4989 pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { 4990 REASON_SET(&reason, PFRES_MEMORY); 4991 goto drop; 4992 } 4993 if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub && 4994 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4995 &s->src, &s->dst, rewrite)) { 4996 /* This really shouldn't happen!!! */ 4997 DPFPRINTF(PF_DEBUG_URGENT, 4998 ("pf_normalize_tcp_stateful failed on first " 4999 "pkt\n")); 5000 goto drop; 5001 } 5002 } else if (pd->proto == IPPROTO_SCTP) { 5003 if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst)) 5004 goto drop; 5005 if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP))) 5006 goto drop; 5007 } 5008 s->direction = pd->dir; 5009 5010 /* 5011 * sk/nk could already been setup by pf_get_translation(). 5012 */ 5013 if (nr == NULL) { 5014 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 5015 __func__, nr, sk, nk)); 5016 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 5017 if (sk == NULL) 5018 goto csfailed; 5019 nk = sk; 5020 } else 5021 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 5022 __func__, nr, sk, nk)); 5023 5024 /* Swap sk/nk for PF_OUT. */ 5025 if (pf_state_insert(BOUND_IFACE(s, kif), kif, 5026 (pd->dir == PF_IN) ? sk : nk, 5027 (pd->dir == PF_IN) ? nk : sk, s)) { 5028 REASON_SET(&reason, PFRES_STATEINS); 5029 goto drop; 5030 } else 5031 *sm = s; 5032 5033 if (tag > 0) 5034 s->tag = tag; 5035 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 5036 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 5037 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 5038 /* undo NAT changes, if they have taken place */ 5039 if (nr != NULL) { 5040 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 5041 if (pd->dir == PF_OUT) 5042 skt = s->key[PF_SK_STACK]; 5043 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 5044 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 5045 if (pd->sport) 5046 *pd->sport = skt->port[pd->sidx]; 5047 if (pd->dport) 5048 *pd->dport = skt->port[pd->didx]; 5049 if (pd->proto_sum) 5050 *pd->proto_sum = bproto_sum; 5051 if (pd->ip_sum) 5052 *pd->ip_sum = bip_sum; 5053 m_copyback(m, off, hdrlen, pd->hdr.any); 5054 } 5055 s->src.seqhi = htonl(arc4random()); 5056 /* Find mss option */ 5057 int rtid = M_GETFIB(m); 5058 mss = pf_get_mss(m, off, th->th_off, pd->af); 5059 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 5060 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 5061 s->src.mss = mss; 5062 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 5063 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 5064 TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0, 5065 pd->act.rtableid); 5066 REASON_SET(&reason, PFRES_SYNPROXY); 5067 return (PF_SYNPROXY_DROP); 5068 } 5069 5070 return (PF_PASS); 5071 5072csfailed: 5073 while ((ri = SLIST_FIRST(match_rules))) { 5074 SLIST_REMOVE_HEAD(match_rules, entry); 5075 free(ri, M_PF_RULE_ITEM); 5076 } 5077 5078 uma_zfree(V_pf_state_key_z, sk); 5079 uma_zfree(V_pf_state_key_z, nk); 5080 5081 if (sn != NULL) { 5082 PF_SRC_NODE_LOCK(sn); 5083 if (--sn->states == 0 && sn->expire == 0) { 5084 pf_unlink_src_node(sn); 5085 uma_zfree(V_pf_sources_z, sn); 5086 counter_u64_add( 5087 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5088 } 5089 PF_SRC_NODE_UNLOCK(sn); 5090 } 5091 5092 if (nsn != sn && nsn != NULL) { 5093 PF_SRC_NODE_LOCK(nsn); 5094 if (--nsn->states == 0 && nsn->expire == 0) { 5095 pf_unlink_src_node(nsn); 5096 uma_zfree(V_pf_sources_z, nsn); 5097 counter_u64_add( 5098 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 5099 } 5100 PF_SRC_NODE_UNLOCK(nsn); 5101 } 5102 5103drop: 5104 if (s != NULL) { 5105 pf_src_tree_remove_state(s); 5106 s->timeout = PFTM_UNLINKED; 5107 STATE_DEC_COUNTERS(s); 5108 pf_free_state(s); 5109 } 5110 5111 return (PF_DROP); 5112} 5113 5114static int 5115pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif, 5116 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 5117 struct pf_kruleset **rsm) 5118{ 5119 struct pf_krule *r, *a = NULL; 5120 struct pf_kruleset *ruleset = NULL; 5121 struct pf_krule_slist match_rules; 5122 struct pf_krule_item *ri; 5123 sa_family_t af = pd->af; 5124 u_short reason; 5125 int tag = -1; 5126 int asd = 0; 5127 int match = 0; 5128 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 5129 5130 PF_RULES_RASSERT(); 5131 5132 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 5133 SLIST_INIT(&match_rules); 5134 while (r != NULL) { 5135 pf_counter_u64_add(&r->evaluations, 1); 5136 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 5137 r = r->skip[PF_SKIP_IFP].ptr; 5138 else if (r->direction && r->direction != pd->dir) 5139 r = r->skip[PF_SKIP_DIR].ptr; 5140 else if (r->af && r->af != af) 5141 r = r->skip[PF_SKIP_AF].ptr; 5142 else if (r->proto && r->proto != pd->proto) 5143 r = r->skip[PF_SKIP_PROTO].ptr; 5144 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 5145 r->src.neg, kif, M_GETFIB(m))) 5146 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 5147 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 5148 r->dst.neg, NULL, M_GETFIB(m))) 5149 r = r->skip[PF_SKIP_DST_ADDR].ptr; 5150 else if (r->tos && !(r->tos == pd->tos)) 5151 r = TAILQ_NEXT(r, entries); 5152 else if (r->os_fingerprint != PF_OSFP_ANY) 5153 r = TAILQ_NEXT(r, entries); 5154 else if (pd->proto == IPPROTO_UDP && 5155 (r->src.port_op || r->dst.port_op)) 5156 r = TAILQ_NEXT(r, entries); 5157 else if (pd->proto == IPPROTO_TCP && 5158 (r->src.port_op || r->dst.port_op || r->flagset)) 5159 r = TAILQ_NEXT(r, entries); 5160 else if ((pd->proto == IPPROTO_ICMP || 5161 pd->proto == IPPROTO_ICMPV6) && 5162 (r->type || r->code)) 5163 r = TAILQ_NEXT(r, entries); 5164 else if (r->prio && 5165 !pf_match_ieee8021q_pcp(r->prio, m)) 5166 r = TAILQ_NEXT(r, entries); 5167 else if (r->prob && r->prob <= 5168 (arc4random() % (UINT_MAX - 1) + 1)) 5169 r = TAILQ_NEXT(r, entries); 5170 else if (r->match_tag && !pf_match_tag(m, r, &tag, 5171 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 5172 r = TAILQ_NEXT(r, entries); 5173 else { 5174 if (r->anchor == NULL) { 5175 if (r->action == PF_MATCH) { 5176 ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO); 5177 if (ri == NULL) { 5178 REASON_SET(&reason, PFRES_MEMORY); 5179 goto cleanup; 5180 } 5181 ri->r = r; 5182 SLIST_INSERT_HEAD(&match_rules, ri, entry); 5183 pf_counter_u64_critical_enter(); 5184 pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1); 5185 pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len); 5186 pf_counter_u64_critical_exit(); 5187 pf_rule_to_actions(r, &pd->act); 5188 if (r->log) 5189 PFLOG_PACKET(kif, m, af, 5190 r->action, PFRES_MATCH, r, 5191 a, ruleset, pd, 1); 5192 } else { 5193 match = 1; 5194 *rm = r; 5195 *am = a; 5196 *rsm = ruleset; 5197 } 5198 if ((*rm)->quick) 5199 break; 5200 r = TAILQ_NEXT(r, entries); 5201 } else 5202 pf_step_into_anchor(anchor_stack, &asd, 5203 &ruleset, PF_RULESET_FILTER, &r, &a, 5204 &match); 5205 } 5206 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 5207 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 5208 break; 5209 } 5210 r = *rm; 5211 a = *am; 5212 ruleset = *rsm; 5213 5214 REASON_SET(&reason, PFRES_MATCH); 5215 5216 /* apply actions for last matching pass/block rule */ 5217 pf_rule_to_actions(r, &pd->act); 5218 5219 if (r->log) 5220 PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1); 5221 5222 if (r->action != PF_PASS) 5223 return (PF_DROP); 5224 5225 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 5226 REASON_SET(&reason, PFRES_MEMORY); 5227 goto cleanup; 5228 } 5229 5230 return (PF_PASS); 5231 5232cleanup: 5233 while ((ri = SLIST_FIRST(&match_rules))) { 5234 SLIST_REMOVE_HEAD(&match_rules, entry); 5235 free(ri, M_PF_RULE_ITEM); 5236 } 5237 5238 return (PF_DROP); 5239} 5240 5241static int 5242pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 5243 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 5244 int *copyback) 5245{ 5246 struct tcphdr *th = &pd->hdr.tcp; 5247 struct pf_state_peer *src, *dst; 5248 u_int16_t win = ntohs(th->th_win); 5249 u_int32_t ack, end, seq, orig_seq; 5250 u_int8_t sws, dws, psrc, pdst; 5251 int ackskew; 5252 5253 if (pd->dir == (*state)->direction) { 5254 src = &(*state)->src; 5255 dst = &(*state)->dst; 5256 psrc = PF_PEER_SRC; 5257 pdst = PF_PEER_DST; 5258 } else { 5259 src = &(*state)->dst; 5260 dst = &(*state)->src; 5261 psrc = PF_PEER_DST; 5262 pdst = PF_PEER_SRC; 5263 } 5264 5265 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 5266 sws = src->wscale & PF_WSCALE_MASK; 5267 dws = dst->wscale & PF_WSCALE_MASK; 5268 } else 5269 sws = dws = 0; 5270 5271 /* 5272 * Sequence tracking algorithm from Guido van Rooij's paper: 5273 * http://www.madison-gurkha.com/publications/tcp_filtering/ 5274 * tcp_filtering.ps 5275 */ 5276 5277 orig_seq = seq = ntohl(th->th_seq); 5278 if (src->seqlo == 0) { 5279 /* First packet from this end. Set its state */ 5280 5281 if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) && 5282 src->scrub == NULL) { 5283 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 5284 REASON_SET(reason, PFRES_MEMORY); 5285 return (PF_DROP); 5286 } 5287 } 5288 5289 /* Deferred generation of sequence number modulator */ 5290 if (dst->seqdiff && !src->seqdiff) { 5291 /* use random iss for the TCP server */ 5292 while ((src->seqdiff = arc4random() - seq) == 0) 5293 ; 5294 ack = ntohl(th->th_ack) - dst->seqdiff; 5295 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5296 src->seqdiff), 0); 5297 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5298 *copyback = 1; 5299 } else { 5300 ack = ntohl(th->th_ack); 5301 } 5302 5303 end = seq + pd->p_len; 5304 if (th->th_flags & TH_SYN) { 5305 end++; 5306 if (dst->wscale & PF_WSCALE_FLAG) { 5307 src->wscale = pf_get_wscale(m, off, th->th_off, 5308 pd->af); 5309 if (src->wscale & PF_WSCALE_FLAG) { 5310 /* Remove scale factor from initial 5311 * window */ 5312 sws = src->wscale & PF_WSCALE_MASK; 5313 win = ((u_int32_t)win + (1 << sws) - 1) 5314 >> sws; 5315 dws = dst->wscale & PF_WSCALE_MASK; 5316 } else { 5317 /* fixup other window */ 5318 dst->max_win <<= dst->wscale & 5319 PF_WSCALE_MASK; 5320 /* in case of a retrans SYN|ACK */ 5321 dst->wscale = 0; 5322 } 5323 } 5324 } 5325 if (th->th_flags & TH_FIN) 5326 end++; 5327 5328 src->seqlo = seq; 5329 if (src->state < TCPS_SYN_SENT) 5330 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5331 5332 /* 5333 * May need to slide the window (seqhi may have been set by 5334 * the crappy stack check or if we picked up the connection 5335 * after establishment) 5336 */ 5337 if (src->seqhi == 1 || 5338 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 5339 src->seqhi = end + MAX(1, dst->max_win << dws); 5340 if (win > src->max_win) 5341 src->max_win = win; 5342 5343 } else { 5344 ack = ntohl(th->th_ack) - dst->seqdiff; 5345 if (src->seqdiff) { 5346 /* Modulate sequence numbers */ 5347 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 5348 src->seqdiff), 0); 5349 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 5350 *copyback = 1; 5351 } 5352 end = seq + pd->p_len; 5353 if (th->th_flags & TH_SYN) 5354 end++; 5355 if (th->th_flags & TH_FIN) 5356 end++; 5357 } 5358 5359 if ((th->th_flags & TH_ACK) == 0) { 5360 /* Let it pass through the ack skew check */ 5361 ack = dst->seqlo; 5362 } else if ((ack == 0 && 5363 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 5364 /* broken tcp stacks do not set ack */ 5365 (dst->state < TCPS_SYN_SENT)) { 5366 /* 5367 * Many stacks (ours included) will set the ACK number in an 5368 * FIN|ACK if the SYN times out -- no sequence to ACK. 5369 */ 5370 ack = dst->seqlo; 5371 } 5372 5373 if (seq == end) { 5374 /* Ease sequencing restrictions on no data packets */ 5375 seq = src->seqlo; 5376 end = seq; 5377 } 5378 5379 ackskew = dst->seqlo - ack; 5380 5381 /* 5382 * Need to demodulate the sequence numbers in any TCP SACK options 5383 * (Selective ACK). We could optionally validate the SACK values 5384 * against the current ACK window, either forwards or backwards, but 5385 * I'm not confident that SACK has been implemented properly 5386 * everywhere. It wouldn't surprise me if several stacks accidentally 5387 * SACK too far backwards of previously ACKed data. There really aren't 5388 * any security implications of bad SACKing unless the target stack 5389 * doesn't validate the option length correctly. Someone trying to 5390 * spoof into a TCP connection won't bother blindly sending SACK 5391 * options anyway. 5392 */ 5393 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 5394 if (pf_modulate_sack(m, off, pd, th, dst)) 5395 *copyback = 1; 5396 } 5397 5398#define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 5399 if (SEQ_GEQ(src->seqhi, end) && 5400 /* Last octet inside other's window space */ 5401 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 5402 /* Retrans: not more than one window back */ 5403 (ackskew >= -MAXACKWINDOW) && 5404 /* Acking not more than one reassembled fragment backwards */ 5405 (ackskew <= (MAXACKWINDOW << sws)) && 5406 /* Acking not more than one window forward */ 5407 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 5408 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) { 5409 /* Require an exact/+1 sequence match on resets when possible */ 5410 5411 if (dst->scrub || src->scrub) { 5412 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5413 *state, src, dst, copyback)) 5414 return (PF_DROP); 5415 } 5416 5417 /* update max window */ 5418 if (src->max_win < win) 5419 src->max_win = win; 5420 /* synchronize sequencing */ 5421 if (SEQ_GT(end, src->seqlo)) 5422 src->seqlo = end; 5423 /* slide the window of what the other end can send */ 5424 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5425 dst->seqhi = ack + MAX((win << sws), 1); 5426 5427 /* update states */ 5428 if (th->th_flags & TH_SYN) 5429 if (src->state < TCPS_SYN_SENT) 5430 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5431 if (th->th_flags & TH_FIN) 5432 if (src->state < TCPS_CLOSING) 5433 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5434 if (th->th_flags & TH_ACK) { 5435 if (dst->state == TCPS_SYN_SENT) { 5436 pf_set_protostate(*state, pdst, 5437 TCPS_ESTABLISHED); 5438 if (src->state == TCPS_ESTABLISHED && 5439 (*state)->src_node != NULL && 5440 pf_src_connlimit(state)) { 5441 REASON_SET(reason, PFRES_SRCLIMIT); 5442 return (PF_DROP); 5443 } 5444 } else if (dst->state == TCPS_CLOSING) 5445 pf_set_protostate(*state, pdst, 5446 TCPS_FIN_WAIT_2); 5447 } 5448 if (th->th_flags & TH_RST) 5449 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5450 5451 /* update expire time */ 5452 (*state)->expire = pf_get_uptime(); 5453 if (src->state >= TCPS_FIN_WAIT_2 && 5454 dst->state >= TCPS_FIN_WAIT_2) 5455 (*state)->timeout = PFTM_TCP_CLOSED; 5456 else if (src->state >= TCPS_CLOSING && 5457 dst->state >= TCPS_CLOSING) 5458 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5459 else if (src->state < TCPS_ESTABLISHED || 5460 dst->state < TCPS_ESTABLISHED) 5461 (*state)->timeout = PFTM_TCP_OPENING; 5462 else if (src->state >= TCPS_CLOSING || 5463 dst->state >= TCPS_CLOSING) 5464 (*state)->timeout = PFTM_TCP_CLOSING; 5465 else 5466 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5467 5468 /* Fall through to PASS packet */ 5469 5470 } else if ((dst->state < TCPS_SYN_SENT || 5471 dst->state >= TCPS_FIN_WAIT_2 || 5472 src->state >= TCPS_FIN_WAIT_2) && 5473 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 5474 /* Within a window forward of the originating packet */ 5475 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 5476 /* Within a window backward of the originating packet */ 5477 5478 /* 5479 * This currently handles three situations: 5480 * 1) Stupid stacks will shotgun SYNs before their peer 5481 * replies. 5482 * 2) When PF catches an already established stream (the 5483 * firewall rebooted, the state table was flushed, routes 5484 * changed...) 5485 * 3) Packets get funky immediately after the connection 5486 * closes (this should catch Solaris spurious ACK|FINs 5487 * that web servers like to spew after a close) 5488 * 5489 * This must be a little more careful than the above code 5490 * since packet floods will also be caught here. We don't 5491 * update the TTL here to mitigate the damage of a packet 5492 * flood and so the same code can handle awkward establishment 5493 * and a loosened connection close. 5494 * In the establishment case, a correct peer response will 5495 * validate the connection, go through the normal state code 5496 * and keep updating the state TTL. 5497 */ 5498 5499 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5500 printf("pf: loose state match: "); 5501 pf_print_state(*state); 5502 pf_print_flags(th->th_flags); 5503 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5504 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 5505 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 5506 (unsigned long long)(*state)->packets[1], 5507 pd->dir == PF_IN ? "in" : "out", 5508 pd->dir == (*state)->direction ? "fwd" : "rev"); 5509 } 5510 5511 if (dst->scrub || src->scrub) { 5512 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 5513 *state, src, dst, copyback)) 5514 return (PF_DROP); 5515 } 5516 5517 /* update max window */ 5518 if (src->max_win < win) 5519 src->max_win = win; 5520 /* synchronize sequencing */ 5521 if (SEQ_GT(end, src->seqlo)) 5522 src->seqlo = end; 5523 /* slide the window of what the other end can send */ 5524 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 5525 dst->seqhi = ack + MAX((win << sws), 1); 5526 5527 /* 5528 * Cannot set dst->seqhi here since this could be a shotgunned 5529 * SYN and not an already established connection. 5530 */ 5531 5532 if (th->th_flags & TH_FIN) 5533 if (src->state < TCPS_CLOSING) 5534 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5535 if (th->th_flags & TH_RST) 5536 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5537 5538 /* Fall through to PASS packet */ 5539 5540 } else { 5541 if ((*state)->dst.state == TCPS_SYN_SENT && 5542 (*state)->src.state == TCPS_SYN_SENT) { 5543 /* Send RST for state mismatches during handshake */ 5544 if (!(th->th_flags & TH_RST)) 5545 pf_send_tcp((*state)->rule.ptr, pd->af, 5546 pd->dst, pd->src, th->th_dport, 5547 th->th_sport, ntohl(th->th_ack), 0, 5548 TH_RST, 0, 0, 5549 (*state)->rule.ptr->return_ttl, true, 0, 0, 5550 (*state)->act.rtableid); 5551 src->seqlo = 0; 5552 src->seqhi = 1; 5553 src->max_win = 1; 5554 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 5555 printf("pf: BAD state: "); 5556 pf_print_state(*state); 5557 pf_print_flags(th->th_flags); 5558 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 5559 "pkts=%llu:%llu dir=%s,%s\n", 5560 seq, orig_seq, ack, pd->p_len, ackskew, 5561 (unsigned long long)(*state)->packets[0], 5562 (unsigned long long)(*state)->packets[1], 5563 pd->dir == PF_IN ? "in" : "out", 5564 pd->dir == (*state)->direction ? "fwd" : "rev"); 5565 printf("pf: State failure on: %c %c %c %c | %c %c\n", 5566 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 5567 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 5568 ' ': '2', 5569 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 5570 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 5571 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 5572 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 5573 } 5574 REASON_SET(reason, PFRES_BADSTATE); 5575 return (PF_DROP); 5576 } 5577 5578 return (PF_PASS); 5579} 5580 5581static int 5582pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 5583{ 5584 struct tcphdr *th = &pd->hdr.tcp; 5585 struct pf_state_peer *src, *dst; 5586 u_int8_t psrc, pdst; 5587 5588 if (pd->dir == (*state)->direction) { 5589 src = &(*state)->src; 5590 dst = &(*state)->dst; 5591 psrc = PF_PEER_SRC; 5592 pdst = PF_PEER_DST; 5593 } else { 5594 src = &(*state)->dst; 5595 dst = &(*state)->src; 5596 psrc = PF_PEER_DST; 5597 pdst = PF_PEER_SRC; 5598 } 5599 5600 if (th->th_flags & TH_SYN) 5601 if (src->state < TCPS_SYN_SENT) 5602 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 5603 if (th->th_flags & TH_FIN) 5604 if (src->state < TCPS_CLOSING) 5605 pf_set_protostate(*state, psrc, TCPS_CLOSING); 5606 if (th->th_flags & TH_ACK) { 5607 if (dst->state == TCPS_SYN_SENT) { 5608 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 5609 if (src->state == TCPS_ESTABLISHED && 5610 (*state)->src_node != NULL && 5611 pf_src_connlimit(state)) { 5612 REASON_SET(reason, PFRES_SRCLIMIT); 5613 return (PF_DROP); 5614 } 5615 } else if (dst->state == TCPS_CLOSING) { 5616 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 5617 } else if (src->state == TCPS_SYN_SENT && 5618 dst->state < TCPS_SYN_SENT) { 5619 /* 5620 * Handle a special sloppy case where we only see one 5621 * half of the connection. If there is a ACK after 5622 * the initial SYN without ever seeing a packet from 5623 * the destination, set the connection to established. 5624 */ 5625 pf_set_protostate(*state, PF_PEER_BOTH, 5626 TCPS_ESTABLISHED); 5627 dst->state = src->state = TCPS_ESTABLISHED; 5628 if ((*state)->src_node != NULL && 5629 pf_src_connlimit(state)) { 5630 REASON_SET(reason, PFRES_SRCLIMIT); 5631 return (PF_DROP); 5632 } 5633 } else if (src->state == TCPS_CLOSING && 5634 dst->state == TCPS_ESTABLISHED && 5635 dst->seqlo == 0) { 5636 /* 5637 * Handle the closing of half connections where we 5638 * don't see the full bidirectional FIN/ACK+ACK 5639 * handshake. 5640 */ 5641 pf_set_protostate(*state, pdst, TCPS_CLOSING); 5642 } 5643 } 5644 if (th->th_flags & TH_RST) 5645 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 5646 5647 /* update expire time */ 5648 (*state)->expire = pf_get_uptime(); 5649 if (src->state >= TCPS_FIN_WAIT_2 && 5650 dst->state >= TCPS_FIN_WAIT_2) 5651 (*state)->timeout = PFTM_TCP_CLOSED; 5652 else if (src->state >= TCPS_CLOSING && 5653 dst->state >= TCPS_CLOSING) 5654 (*state)->timeout = PFTM_TCP_FIN_WAIT; 5655 else if (src->state < TCPS_ESTABLISHED || 5656 dst->state < TCPS_ESTABLISHED) 5657 (*state)->timeout = PFTM_TCP_OPENING; 5658 else if (src->state >= TCPS_CLOSING || 5659 dst->state >= TCPS_CLOSING) 5660 (*state)->timeout = PFTM_TCP_CLOSING; 5661 else 5662 (*state)->timeout = PFTM_TCP_ESTABLISHED; 5663 5664 return (PF_PASS); 5665} 5666 5667static int 5668pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 5669{ 5670 struct pf_state_key *sk = (*state)->key[pd->didx]; 5671 struct tcphdr *th = &pd->hdr.tcp; 5672 5673 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 5674 if (pd->dir != (*state)->direction) { 5675 REASON_SET(reason, PFRES_SYNPROXY); 5676 return (PF_SYNPROXY_DROP); 5677 } 5678 if (th->th_flags & TH_SYN) { 5679 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 5680 REASON_SET(reason, PFRES_SYNPROXY); 5681 return (PF_DROP); 5682 } 5683 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5684 pd->src, th->th_dport, th->th_sport, 5685 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 5686 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0, 5687 (*state)->act.rtableid); 5688 REASON_SET(reason, PFRES_SYNPROXY); 5689 return (PF_SYNPROXY_DROP); 5690 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 5691 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5692 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5693 REASON_SET(reason, PFRES_SYNPROXY); 5694 return (PF_DROP); 5695 } else if ((*state)->src_node != NULL && 5696 pf_src_connlimit(state)) { 5697 REASON_SET(reason, PFRES_SRCLIMIT); 5698 return (PF_DROP); 5699 } else 5700 pf_set_protostate(*state, PF_PEER_SRC, 5701 PF_TCPS_PROXY_DST); 5702 } 5703 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 5704 if (pd->dir == (*state)->direction) { 5705 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 5706 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 5707 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 5708 REASON_SET(reason, PFRES_SYNPROXY); 5709 return (PF_DROP); 5710 } 5711 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 5712 if ((*state)->dst.seqhi == 1) 5713 (*state)->dst.seqhi = htonl(arc4random()); 5714 pf_send_tcp((*state)->rule.ptr, pd->af, 5715 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5716 sk->port[pd->sidx], sk->port[pd->didx], 5717 (*state)->dst.seqhi, 0, TH_SYN, 0, 5718 (*state)->src.mss, 0, false, (*state)->tag, 0, 5719 (*state)->act.rtableid); 5720 REASON_SET(reason, PFRES_SYNPROXY); 5721 return (PF_SYNPROXY_DROP); 5722 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 5723 (TH_SYN|TH_ACK)) || 5724 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 5725 REASON_SET(reason, PFRES_SYNPROXY); 5726 return (PF_DROP); 5727 } else { 5728 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 5729 (*state)->dst.seqlo = ntohl(th->th_seq); 5730 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 5731 pd->src, th->th_dport, th->th_sport, 5732 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 5733 TH_ACK, (*state)->src.max_win, 0, 0, false, 5734 (*state)->tag, 0, (*state)->act.rtableid); 5735 pf_send_tcp((*state)->rule.ptr, pd->af, 5736 &sk->addr[pd->sidx], &sk->addr[pd->didx], 5737 sk->port[pd->sidx], sk->port[pd->didx], 5738 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 5739 TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0, 5740 (*state)->act.rtableid); 5741 (*state)->src.seqdiff = (*state)->dst.seqhi - 5742 (*state)->src.seqlo; 5743 (*state)->dst.seqdiff = (*state)->src.seqhi - 5744 (*state)->dst.seqlo; 5745 (*state)->src.seqhi = (*state)->src.seqlo + 5746 (*state)->dst.max_win; 5747 (*state)->dst.seqhi = (*state)->dst.seqlo + 5748 (*state)->src.max_win; 5749 (*state)->src.wscale = (*state)->dst.wscale = 0; 5750 pf_set_protostate(*state, PF_PEER_BOTH, 5751 TCPS_ESTABLISHED); 5752 REASON_SET(reason, PFRES_SYNPROXY); 5753 return (PF_SYNPROXY_DROP); 5754 } 5755 } 5756 5757 return (PF_PASS); 5758} 5759 5760static int 5761pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif, 5762 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 5763 u_short *reason) 5764{ 5765 struct pf_state_key_cmp key; 5766 struct tcphdr *th = &pd->hdr.tcp; 5767 int copyback = 0; 5768 int action; 5769 struct pf_state_peer *src, *dst; 5770 5771 bzero(&key, sizeof(key)); 5772 key.af = pd->af; 5773 key.proto = IPPROTO_TCP; 5774 if (pd->dir == PF_IN) { /* wire side, straight */ 5775 PF_ACPY(&key.addr[0], pd->src, key.af); 5776 PF_ACPY(&key.addr[1], pd->dst, key.af); 5777 key.port[0] = th->th_sport; 5778 key.port[1] = th->th_dport; 5779 } else { /* stack side, reverse */ 5780 PF_ACPY(&key.addr[1], pd->src, key.af); 5781 PF_ACPY(&key.addr[0], pd->dst, key.af); 5782 key.port[1] = th->th_sport; 5783 key.port[0] = th->th_dport; 5784 } 5785 5786 STATE_LOOKUP(kif, &key, *state, pd); 5787 5788 if (pd->dir == (*state)->direction) { 5789 src = &(*state)->src; 5790 dst = &(*state)->dst; 5791 } else { 5792 src = &(*state)->dst; 5793 dst = &(*state)->src; 5794 } 5795 5796 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 5797 return (action); 5798 5799 if (dst->state >= TCPS_FIN_WAIT_2 && 5800 src->state >= TCPS_FIN_WAIT_2 && 5801 (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) || 5802 ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK && 5803 pf_syncookie_check(pd) && pd->dir == PF_IN))) { 5804 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5805 printf("pf: state reuse "); 5806 pf_print_state(*state); 5807 pf_print_flags(th->th_flags); 5808 printf("\n"); 5809 } 5810 /* XXX make sure it's the same direction ?? */ 5811 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 5812 pf_unlink_state(*state); 5813 *state = NULL; 5814 return (PF_DROP); 5815 } 5816 5817 if ((*state)->state_flags & PFSTATE_SLOPPY) { 5818 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 5819 return (PF_DROP); 5820 } else { 5821 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 5822 ©back) == PF_DROP) 5823 return (PF_DROP); 5824 } 5825 5826 /* translate source/destination address, if necessary */ 5827 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5828 struct pf_state_key *nk = (*state)->key[pd->didx]; 5829 5830 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5831 nk->port[pd->sidx] != th->th_sport) 5832 pf_change_ap(m, pd->src, &th->th_sport, 5833 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 5834 nk->port[pd->sidx], 0, pd->af); 5835 5836 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5837 nk->port[pd->didx] != th->th_dport) 5838 pf_change_ap(m, pd->dst, &th->th_dport, 5839 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 5840 nk->port[pd->didx], 0, pd->af); 5841 copyback = 1; 5842 } 5843 5844 /* Copyback sequence modulation or stateful scrub changes if needed */ 5845 if (copyback) 5846 m_copyback(m, off, sizeof(*th), (caddr_t)th); 5847 5848 return (PF_PASS); 5849} 5850 5851static int 5852pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif, 5853 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 5854{ 5855 struct pf_state_peer *src, *dst; 5856 struct pf_state_key_cmp key; 5857 struct udphdr *uh = &pd->hdr.udp; 5858 uint8_t psrc, pdst; 5859 5860 bzero(&key, sizeof(key)); 5861 key.af = pd->af; 5862 key.proto = IPPROTO_UDP; 5863 if (pd->dir == PF_IN) { /* wire side, straight */ 5864 PF_ACPY(&key.addr[0], pd->src, key.af); 5865 PF_ACPY(&key.addr[1], pd->dst, key.af); 5866 key.port[0] = uh->uh_sport; 5867 key.port[1] = uh->uh_dport; 5868 } else { /* stack side, reverse */ 5869 PF_ACPY(&key.addr[1], pd->src, key.af); 5870 PF_ACPY(&key.addr[0], pd->dst, key.af); 5871 key.port[1] = uh->uh_sport; 5872 key.port[0] = uh->uh_dport; 5873 } 5874 5875 STATE_LOOKUP(kif, &key, *state, pd); 5876 5877 if (pd->dir == (*state)->direction) { 5878 src = &(*state)->src; 5879 dst = &(*state)->dst; 5880 psrc = PF_PEER_SRC; 5881 pdst = PF_PEER_DST; 5882 } else { 5883 src = &(*state)->dst; 5884 dst = &(*state)->src; 5885 psrc = PF_PEER_DST; 5886 pdst = PF_PEER_SRC; 5887 } 5888 5889 /* update states */ 5890 if (src->state < PFUDPS_SINGLE) 5891 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 5892 if (dst->state == PFUDPS_SINGLE) 5893 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 5894 5895 /* update expire time */ 5896 (*state)->expire = pf_get_uptime(); 5897 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 5898 (*state)->timeout = PFTM_UDP_MULTIPLE; 5899 else 5900 (*state)->timeout = PFTM_UDP_SINGLE; 5901 5902 /* translate source/destination address, if necessary */ 5903 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5904 struct pf_state_key *nk = (*state)->key[pd->didx]; 5905 5906 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 5907 nk->port[pd->sidx] != uh->uh_sport) 5908 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 5909 &uh->uh_sum, &nk->addr[pd->sidx], 5910 nk->port[pd->sidx], 1, pd->af); 5911 5912 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 5913 nk->port[pd->didx] != uh->uh_dport) 5914 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 5915 &uh->uh_sum, &nk->addr[pd->didx], 5916 nk->port[pd->didx], 1, pd->af); 5917 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 5918 } 5919 5920 return (PF_PASS); 5921} 5922 5923static int 5924pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif, 5925 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 5926{ 5927 struct pf_state_key_cmp key; 5928 struct pf_state_peer *src, *dst; 5929 struct sctphdr *sh = &pd->hdr.sctp; 5930 u_int8_t psrc; //, pdst; 5931 5932 bzero(&key, sizeof(key)); 5933 key.af = pd->af; 5934 key.proto = IPPROTO_SCTP; 5935 if (pd->dir == PF_IN) { /* wire side, straight */ 5936 PF_ACPY(&key.addr[0], pd->src, key.af); 5937 PF_ACPY(&key.addr[1], pd->dst, key.af); 5938 key.port[0] = sh->src_port; 5939 key.port[1] = sh->dest_port; 5940 } else { /* stack side, reverse */ 5941 PF_ACPY(&key.addr[1], pd->src, key.af); 5942 PF_ACPY(&key.addr[0], pd->dst, key.af); 5943 key.port[1] = sh->src_port; 5944 key.port[0] = sh->dest_port; 5945 } 5946 5947 STATE_LOOKUP(kif, &key, *state, pd); 5948 5949 if (pd->dir == (*state)->direction) { 5950 src = &(*state)->src; 5951 dst = &(*state)->dst; 5952 psrc = PF_PEER_SRC; 5953 } else { 5954 src = &(*state)->dst; 5955 dst = &(*state)->src; 5956 psrc = PF_PEER_DST; 5957 } 5958 5959 /* Track state. */ 5960 if (pd->sctp_flags & PFDESC_SCTP_INIT) { 5961 if (src->state < SCTP_COOKIE_WAIT) { 5962 pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT); 5963 (*state)->timeout = PFTM_SCTP_OPENING; 5964 } 5965 } 5966 if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) { 5967 MPASS(dst->scrub != NULL); 5968 if (dst->scrub->pfss_v_tag == 0) 5969 dst->scrub->pfss_v_tag = pd->sctp_initiate_tag; 5970 } 5971 5972 if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) { 5973 if (src->state < SCTP_ESTABLISHED) { 5974 pf_set_protostate(*state, psrc, SCTP_ESTABLISHED); 5975 (*state)->timeout = PFTM_SCTP_ESTABLISHED; 5976 } 5977 } 5978 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT | 5979 PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5980 if (src->state < SCTP_SHUTDOWN_PENDING) { 5981 pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING); 5982 (*state)->timeout = PFTM_SCTP_CLOSING; 5983 } 5984 } 5985 if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) { 5986 pf_set_protostate(*state, psrc, SCTP_CLOSED); 5987 (*state)->timeout = PFTM_SCTP_CLOSED; 5988 } 5989 5990 if (src->scrub != NULL) { 5991 if (src->scrub->pfss_v_tag == 0) { 5992 src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag; 5993 } else if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag) 5994 return (PF_DROP); 5995 } 5996 5997 (*state)->expire = pf_get_uptime(); 5998 5999 /* translate source/destination address, if necessary */ 6000 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6001 uint16_t checksum = 0; 6002 struct pf_state_key *nk = (*state)->key[pd->didx]; 6003 6004 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 6005 nk->port[pd->sidx] != pd->hdr.sctp.src_port) { 6006 pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port, 6007 pd->ip_sum, &checksum, &nk->addr[pd->sidx], 6008 nk->port[pd->sidx], 1, pd->af); 6009 } 6010 6011 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 6012 nk->port[pd->didx] != pd->hdr.sctp.dest_port) { 6013 pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port, 6014 pd->ip_sum, &checksum, &nk->addr[pd->didx], 6015 nk->port[pd->didx], 1, pd->af); 6016 } 6017 } 6018 6019 return (PF_PASS); 6020} 6021 6022static void 6023pf_sctp_multihome_detach_addr(const struct pf_kstate *s) 6024{ 6025 struct pf_sctp_endpoint key; 6026 struct pf_sctp_endpoint *ep; 6027 struct pf_state_key *sks = s->key[PF_SK_STACK]; 6028 struct pf_sctp_source *i, *tmp; 6029 6030 if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL) 6031 return; 6032 6033 PF_SCTP_ENDPOINTS_LOCK(); 6034 6035 key.v_tag = s->dst.scrub->pfss_v_tag; 6036 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6037 if (ep != NULL) { 6038 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6039 if (pf_addr_cmp(&i->addr, 6040 &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT], 6041 s->key[PF_SK_WIRE]->af) == 0) { 6042 SDT_PROBE3(pf, sctp, multihome, remove, 6043 key.v_tag, s, i); 6044 TAILQ_REMOVE(&ep->sources, i, entry); 6045 free(i, M_PFTEMP); 6046 break; 6047 } 6048 } 6049 6050 if (TAILQ_EMPTY(&ep->sources)) { 6051 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6052 free(ep, M_PFTEMP); 6053 } 6054 } 6055 6056 /* Other direction. */ 6057 key.v_tag = s->src.scrub->pfss_v_tag; 6058 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6059 if (ep != NULL) { 6060 TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) { 6061 if (pf_addr_cmp(&i->addr, 6062 &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN], 6063 s->key[PF_SK_WIRE]->af) == 0) { 6064 SDT_PROBE3(pf, sctp, multihome, remove, 6065 key.v_tag, s, i); 6066 TAILQ_REMOVE(&ep->sources, i, entry); 6067 free(i, M_PFTEMP); 6068 break; 6069 } 6070 } 6071 6072 if (TAILQ_EMPTY(&ep->sources)) { 6073 RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6074 free(ep, M_PFTEMP); 6075 } 6076 } 6077 6078 PF_SCTP_ENDPOINTS_UNLOCK(); 6079} 6080 6081static void 6082pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag) 6083{ 6084 struct pf_sctp_endpoint key = { 6085 .v_tag = v_tag, 6086 }; 6087 struct pf_sctp_source *i; 6088 struct pf_sctp_endpoint *ep; 6089 6090 PF_SCTP_ENDPOINTS_LOCK(); 6091 6092 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6093 if (ep == NULL) { 6094 ep = malloc(sizeof(struct pf_sctp_endpoint), 6095 M_PFTEMP, M_NOWAIT); 6096 if (ep == NULL) { 6097 PF_SCTP_ENDPOINTS_UNLOCK(); 6098 return; 6099 } 6100 6101 ep->v_tag = v_tag; 6102 TAILQ_INIT(&ep->sources); 6103 RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep); 6104 } 6105 6106 /* Avoid inserting duplicates. */ 6107 TAILQ_FOREACH(i, &ep->sources, entry) { 6108 if (pf_addr_cmp(&i->addr, a, pd->af) == 0) { 6109 PF_SCTP_ENDPOINTS_UNLOCK(); 6110 return; 6111 } 6112 } 6113 6114 i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT); 6115 if (i == NULL) { 6116 PF_SCTP_ENDPOINTS_UNLOCK(); 6117 return; 6118 } 6119 6120 i->af = pd->af; 6121 memcpy(&i->addr, a, sizeof(*a)); 6122 TAILQ_INSERT_TAIL(&ep->sources, i, entry); 6123 SDT_PROBE2(pf, sctp, multihome, add, v_tag, i); 6124 6125 PF_SCTP_ENDPOINTS_UNLOCK(); 6126} 6127 6128static void 6129pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif, 6130 struct pf_kstate *s, int action) 6131{ 6132 struct pf_sctp_multihome_job *j, *tmp; 6133 struct pf_sctp_source *i; 6134 int ret __unused; 6135 struct pf_kstate *sm = NULL; 6136 struct pf_krule *ra = NULL; 6137 struct pf_krule *r = &V_pf_default_rule; 6138 struct pf_kruleset *rs = NULL; 6139 bool do_extra = true; 6140 6141 PF_RULES_RLOCK_TRACKER; 6142 6143again: 6144 TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) { 6145 if (s == NULL || action != PF_PASS) 6146 goto free; 6147 6148 /* Confirm we don't recurse here. */ 6149 MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP)); 6150 6151 switch (j->op) { 6152 case SCTP_ADD_IP_ADDRESS: { 6153 uint32_t v_tag = pd->sctp_initiate_tag; 6154 6155 if (v_tag == 0) { 6156 if (s->direction == pd->dir) 6157 v_tag = s->src.scrub->pfss_v_tag; 6158 else 6159 v_tag = s->dst.scrub->pfss_v_tag; 6160 } 6161 6162 /* 6163 * Avoid duplicating states. We'll already have 6164 * created a state based on the source address of 6165 * the packet, but SCTP endpoints may also list this 6166 * address again in the INIT(_ACK) parameters. 6167 */ 6168 if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) { 6169 break; 6170 } 6171 6172 j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP; 6173 PF_RULES_RLOCK(); 6174 sm = NULL; 6175 /* 6176 * New connections need to be floating, because 6177 * we cannot know what interfaces it will use. 6178 * That's why we pass V_pfi_all rather than kif. 6179 */ 6180 ret = pf_test_rule(&r, &sm, V_pfi_all, 6181 j->m, off, &j->pd, &ra, &rs, NULL); 6182 PF_RULES_RUNLOCK(); 6183 SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret); 6184 if (ret != PF_DROP && sm != NULL) { 6185 /* Inherit v_tag values. */ 6186 if (sm->direction == s->direction) { 6187 sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6188 sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6189 } else { 6190 sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag; 6191 sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag; 6192 } 6193 PF_STATE_UNLOCK(sm); 6194 } else { 6195 /* If we try duplicate inserts? */ 6196 break; 6197 } 6198 6199 /* Only add the address if we've actually allowed the state. */ 6200 pf_sctp_multihome_add_addr(pd, &j->src, v_tag); 6201 6202 if (! do_extra) { 6203 break; 6204 } 6205 /* 6206 * We need to do this for each of our source addresses. 6207 * Find those based on the verification tag. 6208 */ 6209 struct pf_sctp_endpoint key = { 6210 .v_tag = pd->hdr.sctp.v_tag, 6211 }; 6212 struct pf_sctp_endpoint *ep; 6213 6214 PF_SCTP_ENDPOINTS_LOCK(); 6215 ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key); 6216 if (ep == NULL) { 6217 PF_SCTP_ENDPOINTS_UNLOCK(); 6218 break; 6219 } 6220 MPASS(ep != NULL); 6221 6222 TAILQ_FOREACH(i, &ep->sources, entry) { 6223 struct pf_sctp_multihome_job *nj; 6224 6225 /* SCTP can intermingle IPv4 and IPv6. */ 6226 if (i->af != pd->af) 6227 continue; 6228 6229 nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO); 6230 if (! nj) { 6231 continue; 6232 } 6233 memcpy(&nj->pd, &j->pd, sizeof(j->pd)); 6234 memcpy(&nj->src, &j->src, sizeof(nj->src)); 6235 nj->pd.src = &nj->src; 6236 // New destination address! 6237 memcpy(&nj->dst, &i->addr, sizeof(nj->dst)); 6238 nj->pd.dst = &nj->dst; 6239 nj->m = j->m; 6240 nj->op = j->op; 6241 6242 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next); 6243 } 6244 PF_SCTP_ENDPOINTS_UNLOCK(); 6245 6246 break; 6247 } 6248 case SCTP_DEL_IP_ADDRESS: { 6249 struct pf_state_key_cmp key; 6250 uint8_t psrc; 6251 6252 bzero(&key, sizeof(key)); 6253 key.af = j->pd.af; 6254 key.proto = IPPROTO_SCTP; 6255 if (j->pd.dir == PF_IN) { /* wire side, straight */ 6256 PF_ACPY(&key.addr[0], j->pd.src, key.af); 6257 PF_ACPY(&key.addr[1], j->pd.dst, key.af); 6258 key.port[0] = j->pd.hdr.sctp.src_port; 6259 key.port[1] = j->pd.hdr.sctp.dest_port; 6260 } else { /* stack side, reverse */ 6261 PF_ACPY(&key.addr[1], j->pd.src, key.af); 6262 PF_ACPY(&key.addr[0], j->pd.dst, key.af); 6263 key.port[1] = j->pd.hdr.sctp.src_port; 6264 key.port[0] = j->pd.hdr.sctp.dest_port; 6265 } 6266 6267 sm = pf_find_state(kif, &key, j->pd.dir); 6268 if (sm != NULL) { 6269 PF_STATE_LOCK_ASSERT(sm); 6270 if (j->pd.dir == sm->direction) { 6271 psrc = PF_PEER_SRC; 6272 } else { 6273 psrc = PF_PEER_DST; 6274 } 6275 pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING); 6276 sm->timeout = PFTM_SCTP_CLOSING; 6277 PF_STATE_UNLOCK(sm); 6278 } 6279 break; 6280 default: 6281 panic("Unknown op %#x", j->op); 6282 } 6283 } 6284 6285 free: 6286 TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next); 6287 free(j, M_PFTEMP); 6288 } 6289 6290 /* We may have inserted extra work while processing the list. */ 6291 if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) { 6292 do_extra = false; 6293 goto again; 6294 } 6295} 6296 6297static int 6298pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6299 struct pfi_kkif *kif, int op) 6300{ 6301 int off = 0; 6302 struct pf_sctp_multihome_job *job; 6303 6304 while (off < len) { 6305 struct sctp_paramhdr h; 6306 6307 if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL, 6308 pd->af)) 6309 return (PF_DROP); 6310 6311 /* Parameters are at least 4 bytes. */ 6312 if (ntohs(h.param_length) < 4) 6313 return (PF_DROP); 6314 6315 switch (ntohs(h.param_type)) { 6316 case SCTP_IPV4_ADDRESS: { 6317 struct in_addr t; 6318 6319 if (ntohs(h.param_length) != 6320 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6321 return (PF_DROP); 6322 6323 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6324 NULL, NULL, pd->af)) 6325 return (PF_DROP); 6326 6327 if (in_nullhost(t)) 6328 t.s_addr = pd->src->v4.s_addr; 6329 6330 /* 6331 * We hold the state lock (idhash) here, which means 6332 * that we can't acquire the keyhash, or we'll get a 6333 * LOR (and potentially double-lock things too). We also 6334 * can't release the state lock here, so instead we'll 6335 * enqueue this for async handling. 6336 * There's a relatively small race here, in that a 6337 * packet using the new addresses could arrive already, 6338 * but that's just though luck for it. 6339 */ 6340 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6341 if (! job) 6342 return (PF_DROP); 6343 6344 memcpy(&job->pd, pd, sizeof(*pd)); 6345 6346 // New source address! 6347 memcpy(&job->src, &t, sizeof(t)); 6348 job->pd.src = &job->src; 6349 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6350 job->pd.dst = &job->dst; 6351 job->m = m; 6352 job->op = op; 6353 6354 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6355 break; 6356 } 6357#ifdef INET6 6358 case SCTP_IPV6_ADDRESS: { 6359 struct in6_addr t; 6360 6361 if (ntohs(h.param_length) != 6362 (sizeof(struct sctp_paramhdr) + sizeof(t))) 6363 return (PF_DROP); 6364 6365 if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t), 6366 NULL, NULL, pd->af)) 6367 return (PF_DROP); 6368 if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0) 6369 break; 6370 if (memcmp(&t, &in6addr_any, sizeof(t)) == 0) 6371 memcpy(&t, &pd->src->v6, sizeof(t)); 6372 6373 job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO); 6374 if (! job) 6375 return (PF_DROP); 6376 6377 memcpy(&job->pd, pd, sizeof(*pd)); 6378 memcpy(&job->src, &t, sizeof(t)); 6379 job->pd.src = &job->src; 6380 memcpy(&job->dst, pd->dst, sizeof(job->dst)); 6381 job->pd.dst = &job->dst; 6382 job->m = m; 6383 job->op = op; 6384 6385 TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next); 6386 break; 6387 } 6388#endif 6389 case SCTP_ADD_IP_ADDRESS: { 6390 int ret; 6391 struct sctp_asconf_paramhdr ah; 6392 6393 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6394 NULL, NULL, pd->af)) 6395 return (PF_DROP); 6396 6397 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6398 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6399 SCTP_ADD_IP_ADDRESS); 6400 if (ret != PF_PASS) 6401 return (ret); 6402 break; 6403 } 6404 case SCTP_DEL_IP_ADDRESS: { 6405 int ret; 6406 struct sctp_asconf_paramhdr ah; 6407 6408 if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah), 6409 NULL, NULL, pd->af)) 6410 return (PF_DROP); 6411 ret = pf_multihome_scan(m, start + off + sizeof(ah), 6412 ntohs(ah.ph.param_length) - sizeof(ah), pd, kif, 6413 SCTP_DEL_IP_ADDRESS); 6414 if (ret != PF_PASS) 6415 return (ret); 6416 break; 6417 } 6418 default: 6419 break; 6420 } 6421 6422 off += roundup(ntohs(h.param_length), 4); 6423 } 6424 6425 return (PF_PASS); 6426} 6427int 6428pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd, 6429 struct pfi_kkif *kif) 6430{ 6431 start += sizeof(struct sctp_init_chunk); 6432 len -= sizeof(struct sctp_init_chunk); 6433 6434 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6435} 6436 6437int 6438pf_multihome_scan_asconf(struct mbuf *m, int start, int len, 6439 struct pf_pdesc *pd, struct pfi_kkif *kif) 6440{ 6441 start += sizeof(struct sctp_asconf_chunk); 6442 len -= sizeof(struct sctp_asconf_chunk); 6443 6444 return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS)); 6445} 6446 6447static int 6448pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif, 6449 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 6450{ 6451 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 6452 u_int16_t icmpid = 0, *icmpsum; 6453 u_int8_t icmptype, icmpcode; 6454 int state_icmp = 0; 6455 struct pf_state_key_cmp key; 6456 6457 bzero(&key, sizeof(key)); 6458 switch (pd->proto) { 6459#ifdef INET 6460 case IPPROTO_ICMP: 6461 icmptype = pd->hdr.icmp.icmp_type; 6462 icmpcode = pd->hdr.icmp.icmp_code; 6463 icmpid = pd->hdr.icmp.icmp_id; 6464 icmpsum = &pd->hdr.icmp.icmp_cksum; 6465 6466 if (icmptype == ICMP_UNREACH || 6467 icmptype == ICMP_SOURCEQUENCH || 6468 icmptype == ICMP_REDIRECT || 6469 icmptype == ICMP_TIMXCEED || 6470 icmptype == ICMP_PARAMPROB) 6471 state_icmp++; 6472 break; 6473#endif /* INET */ 6474#ifdef INET6 6475 case IPPROTO_ICMPV6: 6476 icmptype = pd->hdr.icmp6.icmp6_type; 6477 icmpcode = pd->hdr.icmp6.icmp6_code; 6478 icmpid = pd->hdr.icmp6.icmp6_id; 6479 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 6480 6481 if (icmptype == ICMP6_DST_UNREACH || 6482 icmptype == ICMP6_PACKET_TOO_BIG || 6483 icmptype == ICMP6_TIME_EXCEEDED || 6484 icmptype == ICMP6_PARAM_PROB) 6485 state_icmp++; 6486 break; 6487#endif /* INET6 */ 6488 } 6489 6490 if (!state_icmp) { 6491 /* 6492 * ICMP query/reply message not related to a TCP/UDP packet. 6493 * Search for an ICMP state. 6494 */ 6495 key.af = pd->af; 6496 key.proto = pd->proto; 6497 key.port[0] = key.port[1] = icmpid; 6498 if (pd->dir == PF_IN) { /* wire side, straight */ 6499 PF_ACPY(&key.addr[0], pd->src, key.af); 6500 PF_ACPY(&key.addr[1], pd->dst, key.af); 6501 } else { /* stack side, reverse */ 6502 PF_ACPY(&key.addr[1], pd->src, key.af); 6503 PF_ACPY(&key.addr[0], pd->dst, key.af); 6504 } 6505 6506 STATE_LOOKUP(kif, &key, *state, pd); 6507 6508 (*state)->expire = pf_get_uptime(); 6509 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 6510 6511 /* translate source/destination address, if necessary */ 6512 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 6513 struct pf_state_key *nk = (*state)->key[pd->didx]; 6514 6515 switch (pd->af) { 6516#ifdef INET 6517 case AF_INET: 6518 if (PF_ANEQ(pd->src, 6519 &nk->addr[pd->sidx], AF_INET)) 6520 pf_change_a(&saddr->v4.s_addr, 6521 pd->ip_sum, 6522 nk->addr[pd->sidx].v4.s_addr, 0); 6523 6524 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 6525 AF_INET)) 6526 pf_change_a(&daddr->v4.s_addr, 6527 pd->ip_sum, 6528 nk->addr[pd->didx].v4.s_addr, 0); 6529 6530 if (nk->port[0] != 6531 pd->hdr.icmp.icmp_id) { 6532 pd->hdr.icmp.icmp_cksum = 6533 pf_cksum_fixup( 6534 pd->hdr.icmp.icmp_cksum, icmpid, 6535 nk->port[pd->sidx], 0); 6536 pd->hdr.icmp.icmp_id = 6537 nk->port[pd->sidx]; 6538 } 6539 6540 m_copyback(m, off, ICMP_MINLEN, 6541 (caddr_t )&pd->hdr.icmp); 6542 break; 6543#endif /* INET */ 6544#ifdef INET6 6545 case AF_INET6: 6546 if (PF_ANEQ(pd->src, 6547 &nk->addr[pd->sidx], AF_INET6)) 6548 pf_change_a6(saddr, 6549 &pd->hdr.icmp6.icmp6_cksum, 6550 &nk->addr[pd->sidx], 0); 6551 6552 if (PF_ANEQ(pd->dst, 6553 &nk->addr[pd->didx], AF_INET6)) 6554 pf_change_a6(daddr, 6555 &pd->hdr.icmp6.icmp6_cksum, 6556 &nk->addr[pd->didx], 0); 6557 6558 m_copyback(m, off, sizeof(struct icmp6_hdr), 6559 (caddr_t )&pd->hdr.icmp6); 6560 break; 6561#endif /* INET6 */ 6562 } 6563 } 6564 return (PF_PASS); 6565 6566 } else { 6567 /* 6568 * ICMP error message in response to a TCP/UDP packet. 6569 * Extract the inner TCP/UDP header and search for that state. 6570 */ 6571 6572 struct pf_pdesc pd2; 6573 bzero(&pd2, sizeof pd2); 6574#ifdef INET 6575 struct ip h2; 6576#endif /* INET */ 6577#ifdef INET6 6578 struct ip6_hdr h2_6; 6579 int terminal = 0; 6580#endif /* INET6 */ 6581 int ipoff2 = 0; 6582 int off2 = 0; 6583 6584 pd2.af = pd->af; 6585 /* Payload packet is from the opposite direction. */ 6586 pd2.sidx = (pd->dir == PF_IN) ? 1 : 0; 6587 pd2.didx = (pd->dir == PF_IN) ? 0 : 1; 6588 switch (pd->af) { 6589#ifdef INET 6590 case AF_INET: 6591 /* offset of h2 in mbuf chain */ 6592 ipoff2 = off + ICMP_MINLEN; 6593 6594 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 6595 NULL, reason, pd2.af)) { 6596 DPFPRINTF(PF_DEBUG_MISC, 6597 ("pf: ICMP error message too short " 6598 "(ip)\n")); 6599 return (PF_DROP); 6600 } 6601 /* 6602 * ICMP error messages don't refer to non-first 6603 * fragments 6604 */ 6605 if (h2.ip_off & htons(IP_OFFMASK)) { 6606 REASON_SET(reason, PFRES_FRAG); 6607 return (PF_DROP); 6608 } 6609 6610 /* offset of protocol header that follows h2 */ 6611 off2 = ipoff2 + (h2.ip_hl << 2); 6612 6613 pd2.proto = h2.ip_p; 6614 pd2.src = (struct pf_addr *)&h2.ip_src; 6615 pd2.dst = (struct pf_addr *)&h2.ip_dst; 6616 pd2.ip_sum = &h2.ip_sum; 6617 break; 6618#endif /* INET */ 6619#ifdef INET6 6620 case AF_INET6: 6621 ipoff2 = off + sizeof(struct icmp6_hdr); 6622 6623 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 6624 NULL, reason, pd2.af)) { 6625 DPFPRINTF(PF_DEBUG_MISC, 6626 ("pf: ICMP error message too short " 6627 "(ip6)\n")); 6628 return (PF_DROP); 6629 } 6630 pd2.proto = h2_6.ip6_nxt; 6631 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 6632 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 6633 pd2.ip_sum = NULL; 6634 off2 = ipoff2 + sizeof(h2_6); 6635 do { 6636 switch (pd2.proto) { 6637 case IPPROTO_FRAGMENT: 6638 /* 6639 * ICMPv6 error messages for 6640 * non-first fragments 6641 */ 6642 REASON_SET(reason, PFRES_FRAG); 6643 return (PF_DROP); 6644 case IPPROTO_AH: 6645 case IPPROTO_HOPOPTS: 6646 case IPPROTO_ROUTING: 6647 case IPPROTO_DSTOPTS: { 6648 /* get next header and header length */ 6649 struct ip6_ext opt6; 6650 6651 if (!pf_pull_hdr(m, off2, &opt6, 6652 sizeof(opt6), NULL, reason, 6653 pd2.af)) { 6654 DPFPRINTF(PF_DEBUG_MISC, 6655 ("pf: ICMPv6 short opt\n")); 6656 return (PF_DROP); 6657 } 6658 if (pd2.proto == IPPROTO_AH) 6659 off2 += (opt6.ip6e_len + 2) * 4; 6660 else 6661 off2 += (opt6.ip6e_len + 1) * 8; 6662 pd2.proto = opt6.ip6e_nxt; 6663 /* goto the next header */ 6664 break; 6665 } 6666 default: 6667 terminal++; 6668 break; 6669 } 6670 } while (!terminal); 6671 break; 6672#endif /* INET6 */ 6673 } 6674 6675 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 6676 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6677 printf("pf: BAD ICMP %d:%d outer dst: ", 6678 icmptype, icmpcode); 6679 pf_print_host(pd->src, 0, pd->af); 6680 printf(" -> "); 6681 pf_print_host(pd->dst, 0, pd->af); 6682 printf(" inner src: "); 6683 pf_print_host(pd2.src, 0, pd2.af); 6684 printf(" -> "); 6685 pf_print_host(pd2.dst, 0, pd2.af); 6686 printf("\n"); 6687 } 6688 REASON_SET(reason, PFRES_BADSTATE); 6689 return (PF_DROP); 6690 } 6691 6692 switch (pd2.proto) { 6693 case IPPROTO_TCP: { 6694 struct tcphdr th; 6695 u_int32_t seq; 6696 struct pf_state_peer *src, *dst; 6697 u_int8_t dws; 6698 int copyback = 0; 6699 6700 /* 6701 * Only the first 8 bytes of the TCP header can be 6702 * expected. Don't access any TCP header fields after 6703 * th_seq, an ackskew test is not possible. 6704 */ 6705 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 6706 pd2.af)) { 6707 DPFPRINTF(PF_DEBUG_MISC, 6708 ("pf: ICMP error message too short " 6709 "(tcp)\n")); 6710 return (PF_DROP); 6711 } 6712 6713 key.af = pd2.af; 6714 key.proto = IPPROTO_TCP; 6715 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6716 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6717 key.port[pd2.sidx] = th.th_sport; 6718 key.port[pd2.didx] = th.th_dport; 6719 6720 STATE_LOOKUP(kif, &key, *state, pd); 6721 6722 if (pd->dir == (*state)->direction) { 6723 src = &(*state)->dst; 6724 dst = &(*state)->src; 6725 } else { 6726 src = &(*state)->src; 6727 dst = &(*state)->dst; 6728 } 6729 6730 if (src->wscale && dst->wscale) 6731 dws = dst->wscale & PF_WSCALE_MASK; 6732 else 6733 dws = 0; 6734 6735 /* Demodulate sequence number */ 6736 seq = ntohl(th.th_seq) - src->seqdiff; 6737 if (src->seqdiff) { 6738 pf_change_a(&th.th_seq, icmpsum, 6739 htonl(seq), 0); 6740 copyback = 1; 6741 } 6742 6743 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 6744 (!SEQ_GEQ(src->seqhi, seq) || 6745 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 6746 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6747 printf("pf: BAD ICMP %d:%d ", 6748 icmptype, icmpcode); 6749 pf_print_host(pd->src, 0, pd->af); 6750 printf(" -> "); 6751 pf_print_host(pd->dst, 0, pd->af); 6752 printf(" state: "); 6753 pf_print_state(*state); 6754 printf(" seq=%u\n", seq); 6755 } 6756 REASON_SET(reason, PFRES_BADSTATE); 6757 return (PF_DROP); 6758 } else { 6759 if (V_pf_status.debug >= PF_DEBUG_MISC) { 6760 printf("pf: OK ICMP %d:%d ", 6761 icmptype, icmpcode); 6762 pf_print_host(pd->src, 0, pd->af); 6763 printf(" -> "); 6764 pf_print_host(pd->dst, 0, pd->af); 6765 printf(" state: "); 6766 pf_print_state(*state); 6767 printf(" seq=%u\n", seq); 6768 } 6769 } 6770 6771 /* translate source/destination address, if necessary */ 6772 if ((*state)->key[PF_SK_WIRE] != 6773 (*state)->key[PF_SK_STACK]) { 6774 struct pf_state_key *nk = 6775 (*state)->key[pd->didx]; 6776 6777 if (PF_ANEQ(pd2.src, 6778 &nk->addr[pd2.sidx], pd2.af) || 6779 nk->port[pd2.sidx] != th.th_sport) 6780 pf_change_icmp(pd2.src, &th.th_sport, 6781 daddr, &nk->addr[pd2.sidx], 6782 nk->port[pd2.sidx], NULL, 6783 pd2.ip_sum, icmpsum, 6784 pd->ip_sum, 0, pd2.af); 6785 6786 if (PF_ANEQ(pd2.dst, 6787 &nk->addr[pd2.didx], pd2.af) || 6788 nk->port[pd2.didx] != th.th_dport) 6789 pf_change_icmp(pd2.dst, &th.th_dport, 6790 saddr, &nk->addr[pd2.didx], 6791 nk->port[pd2.didx], NULL, 6792 pd2.ip_sum, icmpsum, 6793 pd->ip_sum, 0, pd2.af); 6794 copyback = 1; 6795 } 6796 6797 if (copyback) { 6798 switch (pd2.af) { 6799#ifdef INET 6800 case AF_INET: 6801 m_copyback(m, off, ICMP_MINLEN, 6802 (caddr_t )&pd->hdr.icmp); 6803 m_copyback(m, ipoff2, sizeof(h2), 6804 (caddr_t )&h2); 6805 break; 6806#endif /* INET */ 6807#ifdef INET6 6808 case AF_INET6: 6809 m_copyback(m, off, 6810 sizeof(struct icmp6_hdr), 6811 (caddr_t )&pd->hdr.icmp6); 6812 m_copyback(m, ipoff2, sizeof(h2_6), 6813 (caddr_t )&h2_6); 6814 break; 6815#endif /* INET6 */ 6816 } 6817 m_copyback(m, off2, 8, (caddr_t)&th); 6818 } 6819 6820 return (PF_PASS); 6821 break; 6822 } 6823 case IPPROTO_UDP: { 6824 struct udphdr uh; 6825 6826 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 6827 NULL, reason, pd2.af)) { 6828 DPFPRINTF(PF_DEBUG_MISC, 6829 ("pf: ICMP error message too short " 6830 "(udp)\n")); 6831 return (PF_DROP); 6832 } 6833 6834 key.af = pd2.af; 6835 key.proto = IPPROTO_UDP; 6836 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6837 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6838 key.port[pd2.sidx] = uh.uh_sport; 6839 key.port[pd2.didx] = uh.uh_dport; 6840 6841 STATE_LOOKUP(kif, &key, *state, pd); 6842 6843 /* translate source/destination address, if necessary */ 6844 if ((*state)->key[PF_SK_WIRE] != 6845 (*state)->key[PF_SK_STACK]) { 6846 struct pf_state_key *nk = 6847 (*state)->key[pd->didx]; 6848 6849 if (PF_ANEQ(pd2.src, 6850 &nk->addr[pd2.sidx], pd2.af) || 6851 nk->port[pd2.sidx] != uh.uh_sport) 6852 pf_change_icmp(pd2.src, &uh.uh_sport, 6853 daddr, &nk->addr[pd2.sidx], 6854 nk->port[pd2.sidx], &uh.uh_sum, 6855 pd2.ip_sum, icmpsum, 6856 pd->ip_sum, 1, pd2.af); 6857 6858 if (PF_ANEQ(pd2.dst, 6859 &nk->addr[pd2.didx], pd2.af) || 6860 nk->port[pd2.didx] != uh.uh_dport) 6861 pf_change_icmp(pd2.dst, &uh.uh_dport, 6862 saddr, &nk->addr[pd2.didx], 6863 nk->port[pd2.didx], &uh.uh_sum, 6864 pd2.ip_sum, icmpsum, 6865 pd->ip_sum, 1, pd2.af); 6866 6867 switch (pd2.af) { 6868#ifdef INET 6869 case AF_INET: 6870 m_copyback(m, off, ICMP_MINLEN, 6871 (caddr_t )&pd->hdr.icmp); 6872 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6873 break; 6874#endif /* INET */ 6875#ifdef INET6 6876 case AF_INET6: 6877 m_copyback(m, off, 6878 sizeof(struct icmp6_hdr), 6879 (caddr_t )&pd->hdr.icmp6); 6880 m_copyback(m, ipoff2, sizeof(h2_6), 6881 (caddr_t )&h2_6); 6882 break; 6883#endif /* INET6 */ 6884 } 6885 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 6886 } 6887 return (PF_PASS); 6888 break; 6889 } 6890#ifdef INET 6891 case IPPROTO_ICMP: { 6892 struct icmp iih; 6893 6894 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 6895 NULL, reason, pd2.af)) { 6896 DPFPRINTF(PF_DEBUG_MISC, 6897 ("pf: ICMP error message too short i" 6898 "(icmp)\n")); 6899 return (PF_DROP); 6900 } 6901 6902 key.af = pd2.af; 6903 key.proto = IPPROTO_ICMP; 6904 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6905 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6906 key.port[0] = key.port[1] = iih.icmp_id; 6907 6908 STATE_LOOKUP(kif, &key, *state, pd); 6909 6910 /* translate source/destination address, if necessary */ 6911 if ((*state)->key[PF_SK_WIRE] != 6912 (*state)->key[PF_SK_STACK]) { 6913 struct pf_state_key *nk = 6914 (*state)->key[pd->didx]; 6915 6916 if (PF_ANEQ(pd2.src, 6917 &nk->addr[pd2.sidx], pd2.af) || 6918 nk->port[pd2.sidx] != iih.icmp_id) 6919 pf_change_icmp(pd2.src, &iih.icmp_id, 6920 daddr, &nk->addr[pd2.sidx], 6921 nk->port[pd2.sidx], NULL, 6922 pd2.ip_sum, icmpsum, 6923 pd->ip_sum, 0, AF_INET); 6924 6925 if (PF_ANEQ(pd2.dst, 6926 &nk->addr[pd2.didx], pd2.af) || 6927 nk->port[pd2.didx] != iih.icmp_id) 6928 pf_change_icmp(pd2.dst, &iih.icmp_id, 6929 saddr, &nk->addr[pd2.didx], 6930 nk->port[pd2.didx], NULL, 6931 pd2.ip_sum, icmpsum, 6932 pd->ip_sum, 0, AF_INET); 6933 6934 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 6935 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 6936 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 6937 } 6938 return (PF_PASS); 6939 break; 6940 } 6941#endif /* INET */ 6942#ifdef INET6 6943 case IPPROTO_ICMPV6: { 6944 struct icmp6_hdr iih; 6945 6946 if (!pf_pull_hdr(m, off2, &iih, 6947 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 6948 DPFPRINTF(PF_DEBUG_MISC, 6949 ("pf: ICMP error message too short " 6950 "(icmp6)\n")); 6951 return (PF_DROP); 6952 } 6953 6954 key.af = pd2.af; 6955 key.proto = IPPROTO_ICMPV6; 6956 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 6957 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 6958 key.port[0] = key.port[1] = iih.icmp6_id; 6959 6960 STATE_LOOKUP(kif, &key, *state, pd); 6961 6962 /* translate source/destination address, if necessary */ 6963 if ((*state)->key[PF_SK_WIRE] != 6964 (*state)->key[PF_SK_STACK]) { 6965 struct pf_state_key *nk = 6966 (*state)->key[pd->didx]; 6967 6968 if (PF_ANEQ(pd2.src, 6969 &nk->addr[pd2.sidx], pd2.af) || 6970 nk->port[pd2.sidx] != iih.icmp6_id) 6971 pf_change_icmp(pd2.src, &iih.icmp6_id, 6972 daddr, &nk->addr[pd2.sidx], 6973 nk->port[pd2.sidx], NULL, 6974 pd2.ip_sum, icmpsum, 6975 pd->ip_sum, 0, AF_INET6); 6976 6977 if (PF_ANEQ(pd2.dst, 6978 &nk->addr[pd2.didx], pd2.af) || 6979 nk->port[pd2.didx] != iih.icmp6_id) 6980 pf_change_icmp(pd2.dst, &iih.icmp6_id, 6981 saddr, &nk->addr[pd2.didx], 6982 nk->port[pd2.didx], NULL, 6983 pd2.ip_sum, icmpsum, 6984 pd->ip_sum, 0, AF_INET6); 6985 6986 m_copyback(m, off, sizeof(struct icmp6_hdr), 6987 (caddr_t)&pd->hdr.icmp6); 6988 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 6989 m_copyback(m, off2, sizeof(struct icmp6_hdr), 6990 (caddr_t)&iih); 6991 } 6992 return (PF_PASS); 6993 break; 6994 } 6995#endif /* INET6 */ 6996 default: { 6997 key.af = pd2.af; 6998 key.proto = pd2.proto; 6999 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 7000 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 7001 key.port[0] = key.port[1] = 0; 7002 7003 STATE_LOOKUP(kif, &key, *state, pd); 7004 7005 /* translate source/destination address, if necessary */ 7006 if ((*state)->key[PF_SK_WIRE] != 7007 (*state)->key[PF_SK_STACK]) { 7008 struct pf_state_key *nk = 7009 (*state)->key[pd->didx]; 7010 7011 if (PF_ANEQ(pd2.src, 7012 &nk->addr[pd2.sidx], pd2.af)) 7013 pf_change_icmp(pd2.src, NULL, daddr, 7014 &nk->addr[pd2.sidx], 0, NULL, 7015 pd2.ip_sum, icmpsum, 7016 pd->ip_sum, 0, pd2.af); 7017 7018 if (PF_ANEQ(pd2.dst, 7019 &nk->addr[pd2.didx], pd2.af)) 7020 pf_change_icmp(pd2.dst, NULL, saddr, 7021 &nk->addr[pd2.didx], 0, NULL, 7022 pd2.ip_sum, icmpsum, 7023 pd->ip_sum, 0, pd2.af); 7024 7025 switch (pd2.af) { 7026#ifdef INET 7027 case AF_INET: 7028 m_copyback(m, off, ICMP_MINLEN, 7029 (caddr_t)&pd->hdr.icmp); 7030 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 7031 break; 7032#endif /* INET */ 7033#ifdef INET6 7034 case AF_INET6: 7035 m_copyback(m, off, 7036 sizeof(struct icmp6_hdr), 7037 (caddr_t )&pd->hdr.icmp6); 7038 m_copyback(m, ipoff2, sizeof(h2_6), 7039 (caddr_t )&h2_6); 7040 break; 7041#endif /* INET6 */ 7042 } 7043 } 7044 return (PF_PASS); 7045 break; 7046 } 7047 } 7048 } 7049} 7050 7051static int 7052pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif, 7053 struct mbuf *m, struct pf_pdesc *pd) 7054{ 7055 struct pf_state_peer *src, *dst; 7056 struct pf_state_key_cmp key; 7057 uint8_t psrc, pdst; 7058 7059 bzero(&key, sizeof(key)); 7060 key.af = pd->af; 7061 key.proto = pd->proto; 7062 if (pd->dir == PF_IN) { 7063 PF_ACPY(&key.addr[0], pd->src, key.af); 7064 PF_ACPY(&key.addr[1], pd->dst, key.af); 7065 key.port[0] = key.port[1] = 0; 7066 } else { 7067 PF_ACPY(&key.addr[1], pd->src, key.af); 7068 PF_ACPY(&key.addr[0], pd->dst, key.af); 7069 key.port[1] = key.port[0] = 0; 7070 } 7071 7072 STATE_LOOKUP(kif, &key, *state, pd); 7073 7074 if (pd->dir == (*state)->direction) { 7075 src = &(*state)->src; 7076 dst = &(*state)->dst; 7077 psrc = PF_PEER_SRC; 7078 pdst = PF_PEER_DST; 7079 } else { 7080 src = &(*state)->dst; 7081 dst = &(*state)->src; 7082 psrc = PF_PEER_DST; 7083 pdst = PF_PEER_SRC; 7084 } 7085 7086 /* update states */ 7087 if (src->state < PFOTHERS_SINGLE) 7088 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 7089 if (dst->state == PFOTHERS_SINGLE) 7090 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 7091 7092 /* update expire time */ 7093 (*state)->expire = pf_get_uptime(); 7094 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 7095 (*state)->timeout = PFTM_OTHER_MULTIPLE; 7096 else 7097 (*state)->timeout = PFTM_OTHER_SINGLE; 7098 7099 /* translate source/destination address, if necessary */ 7100 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 7101 struct pf_state_key *nk = (*state)->key[pd->didx]; 7102 7103 KASSERT(nk, ("%s: nk is null", __func__)); 7104 KASSERT(pd, ("%s: pd is null", __func__)); 7105 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 7106 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 7107 switch (pd->af) { 7108#ifdef INET 7109 case AF_INET: 7110 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7111 pf_change_a(&pd->src->v4.s_addr, 7112 pd->ip_sum, 7113 nk->addr[pd->sidx].v4.s_addr, 7114 0); 7115 7116 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7117 pf_change_a(&pd->dst->v4.s_addr, 7118 pd->ip_sum, 7119 nk->addr[pd->didx].v4.s_addr, 7120 0); 7121 7122 break; 7123#endif /* INET */ 7124#ifdef INET6 7125 case AF_INET6: 7126 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 7127 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 7128 7129 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 7130 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 7131#endif /* INET6 */ 7132 } 7133 } 7134 return (PF_PASS); 7135} 7136 7137/* 7138 * ipoff and off are measured from the start of the mbuf chain. 7139 * h must be at "ipoff" on the mbuf chain. 7140 */ 7141void * 7142pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 7143 u_short *actionp, u_short *reasonp, sa_family_t af) 7144{ 7145 switch (af) { 7146#ifdef INET 7147 case AF_INET: { 7148 struct ip *h = mtod(m, struct ip *); 7149 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 7150 7151 if (fragoff) { 7152 if (fragoff >= len) 7153 ACTION_SET(actionp, PF_PASS); 7154 else { 7155 ACTION_SET(actionp, PF_DROP); 7156 REASON_SET(reasonp, PFRES_FRAG); 7157 } 7158 return (NULL); 7159 } 7160 if (m->m_pkthdr.len < off + len || 7161 ntohs(h->ip_len) < off + len) { 7162 ACTION_SET(actionp, PF_DROP); 7163 REASON_SET(reasonp, PFRES_SHORT); 7164 return (NULL); 7165 } 7166 break; 7167 } 7168#endif /* INET */ 7169#ifdef INET6 7170 case AF_INET6: { 7171 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 7172 7173 if (m->m_pkthdr.len < off + len || 7174 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 7175 (unsigned)(off + len)) { 7176 ACTION_SET(actionp, PF_DROP); 7177 REASON_SET(reasonp, PFRES_SHORT); 7178 return (NULL); 7179 } 7180 break; 7181 } 7182#endif /* INET6 */ 7183 } 7184 m_copydata(m, off, len, p); 7185 return (p); 7186} 7187 7188int 7189pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 7190 int rtableid) 7191{ 7192 struct ifnet *ifp; 7193 7194 /* 7195 * Skip check for addresses with embedded interface scope, 7196 * as they would always match anyway. 7197 */ 7198 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 7199 return (1); 7200 7201 if (af != AF_INET && af != AF_INET6) 7202 return (0); 7203 7204 if (kif == V_pfi_all) 7205 return (1); 7206 7207 /* Skip checks for ipsec interfaces */ 7208 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 7209 return (1); 7210 7211 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 7212 7213 switch (af) { 7214#ifdef INET6 7215 case AF_INET6: 7216 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 7217 ifp)); 7218#endif 7219#ifdef INET 7220 case AF_INET: 7221 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 7222 ifp)); 7223#endif 7224 } 7225 7226 return (0); 7227} 7228 7229#ifdef INET 7230static void 7231pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7232 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7233{ 7234 struct mbuf *m0, *m1, *md; 7235 struct sockaddr_in dst; 7236 struct ip *ip; 7237 struct pfi_kkif *nkif = NULL; 7238 struct ifnet *ifp = NULL; 7239 struct pf_addr naddr; 7240 struct pf_ksrc_node *sn = NULL; 7241 int error = 0; 7242 uint16_t ip_len, ip_off; 7243 uint16_t tmp; 7244 int r_rt, r_dir; 7245 7246 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7247 7248 if (s) { 7249 r_rt = s->rt; 7250 r_dir = s->direction; 7251 } else { 7252 r_rt = r->rt; 7253 r_dir = r->direction; 7254 } 7255 7256 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7257 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7258 __func__)); 7259 7260 if ((pd->pf_mtag == NULL && 7261 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7262 pd->pf_mtag->routed++ > 3) { 7263 m0 = *m; 7264 *m = NULL; 7265 goto bad_locked; 7266 } 7267 7268 if (r_rt == PF_DUPTO) { 7269 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7270 if (s == NULL) { 7271 ifp = r->rpool.cur->kif ? 7272 r->rpool.cur->kif->pfik_ifp : NULL; 7273 } else { 7274 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7275 /* If pfsync'd */ 7276 if (ifp == NULL && r->rpool.cur != NULL) 7277 ifp = r->rpool.cur->kif ? 7278 r->rpool.cur->kif->pfik_ifp : NULL; 7279 PF_STATE_UNLOCK(s); 7280 } 7281 if (ifp == oifp) { 7282 /* When the 2nd interface is not skipped */ 7283 return; 7284 } else { 7285 m0 = *m; 7286 *m = NULL; 7287 goto bad; 7288 } 7289 } else { 7290 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7291 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7292 if (s) 7293 PF_STATE_UNLOCK(s); 7294 return; 7295 } 7296 } 7297 } else { 7298 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7299 pf_dummynet(pd, s, r, m); 7300 if (s) 7301 PF_STATE_UNLOCK(s); 7302 return; 7303 } 7304 m0 = *m; 7305 } 7306 7307 ip = mtod(m0, struct ip *); 7308 7309 bzero(&dst, sizeof(dst)); 7310 dst.sin_family = AF_INET; 7311 dst.sin_len = sizeof(dst); 7312 dst.sin_addr = ip->ip_dst; 7313 7314 bzero(&naddr, sizeof(naddr)); 7315 7316 if (s == NULL) { 7317 if (TAILQ_EMPTY(&r->rpool.list)) { 7318 DPFPRINTF(PF_DEBUG_URGENT, 7319 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7320 goto bad_locked; 7321 } 7322 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 7323 &naddr, &nkif, NULL, &sn); 7324 if (!PF_AZERO(&naddr, AF_INET)) 7325 dst.sin_addr.s_addr = naddr.v4.s_addr; 7326 ifp = nkif ? nkif->pfik_ifp : NULL; 7327 } else { 7328 struct pfi_kkif *kif; 7329 7330 if (!PF_AZERO(&s->rt_addr, AF_INET)) 7331 dst.sin_addr.s_addr = 7332 s->rt_addr.v4.s_addr; 7333 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7334 kif = s->rt_kif; 7335 /* If pfsync'd */ 7336 if (ifp == NULL && r->rpool.cur != NULL) { 7337 ifp = r->rpool.cur->kif ? 7338 r->rpool.cur->kif->pfik_ifp : NULL; 7339 kif = r->rpool.cur->kif; 7340 } 7341 if (ifp != NULL && kif != NULL && 7342 r->rule_flag & PFRULE_IFBOUND && 7343 r->rt == PF_REPLYTO && 7344 s->kif == V_pfi_all) { 7345 s->kif = kif; 7346 s->orig_kif = oifp->if_pf_kif; 7347 } 7348 7349 PF_STATE_UNLOCK(s); 7350 } 7351 7352 if (ifp == NULL) 7353 goto bad; 7354 7355 if (pd->dir == PF_IN) { 7356 if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7357 goto bad; 7358 else if (m0 == NULL) 7359 goto done; 7360 if (m0->m_len < sizeof(struct ip)) { 7361 DPFPRINTF(PF_DEBUG_URGENT, 7362 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 7363 goto bad; 7364 } 7365 ip = mtod(m0, struct ip *); 7366 } 7367 7368 if (ifp->if_flags & IFF_LOOPBACK) 7369 m0->m_flags |= M_SKIP_FIREWALL; 7370 7371 ip_len = ntohs(ip->ip_len); 7372 ip_off = ntohs(ip->ip_off); 7373 7374 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 7375 m0->m_pkthdr.csum_flags |= CSUM_IP; 7376 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 7377 in_delayed_cksum(m0); 7378 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 7379 } 7380 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 7381 pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2)); 7382 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 7383 } 7384 7385 if (pd->dir == PF_IN) { 7386 /* 7387 * Make sure dummynet gets the correct direction, in case it needs to 7388 * re-inject later. 7389 */ 7390 pd->dir = PF_OUT; 7391 7392 /* 7393 * The following processing is actually the rest of the inbound processing, even 7394 * though we've marked it as outbound (so we don't look through dummynet) and it 7395 * happens after the outbound processing (pf_test(PF_OUT) above). 7396 * Swap the dummynet pipe numbers, because it's going to come to the wrong 7397 * conclusion about what direction it's processing, and we can't fix it or it 7398 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect 7399 * decision will pick the right pipe, and everything will mostly work as expected. 7400 */ 7401 tmp = pd->act.dnrpipe; 7402 pd->act.dnrpipe = pd->act.dnpipe; 7403 pd->act.dnpipe = tmp; 7404 } 7405 7406 /* 7407 * If small enough for interface, or the interface will take 7408 * care of the fragmentation for us, we can just send directly. 7409 */ 7410 if (ip_len <= ifp->if_mtu || 7411 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 7412 ip->ip_sum = 0; 7413 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 7414 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 7415 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 7416 } 7417 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 7418 7419 md = m0; 7420 error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7421 if (md != NULL) 7422 error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL); 7423 goto done; 7424 } 7425 7426 /* Balk when DF bit is set or the interface didn't support TSO. */ 7427 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 7428 error = EMSGSIZE; 7429 KMOD_IPSTAT_INC(ips_cantfrag); 7430 if (r_rt != PF_DUPTO) { 7431 if (s && pd->nat_rule != NULL) 7432 PACKET_UNDO_NAT(m0, pd, 7433 (ip->ip_hl << 2) + (ip_off & IP_OFFMASK), 7434 s); 7435 7436 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 7437 ifp->if_mtu); 7438 goto done; 7439 } else 7440 goto bad; 7441 } 7442 7443 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 7444 if (error) 7445 goto bad; 7446 7447 for (; m0; m0 = m1) { 7448 m1 = m0->m_nextpkt; 7449 m0->m_nextpkt = NULL; 7450 if (error == 0) { 7451 m_clrprotoflags(m0); 7452 md = m0; 7453 pd->pf_mtag = pf_find_mtag(md); 7454 error = pf_dummynet_route(pd, s, r, ifp, 7455 sintosa(&dst), &md); 7456 if (md != NULL) 7457 error = (*ifp->if_output)(ifp, md, 7458 sintosa(&dst), NULL); 7459 } else 7460 m_freem(m0); 7461 } 7462 7463 if (error == 0) 7464 KMOD_IPSTAT_INC(ips_fragmented); 7465 7466done: 7467 if (r_rt != PF_DUPTO) 7468 *m = NULL; 7469 return; 7470 7471bad_locked: 7472 if (s) 7473 PF_STATE_UNLOCK(s); 7474bad: 7475 m_freem(m0); 7476 goto done; 7477} 7478#endif /* INET */ 7479 7480#ifdef INET6 7481static void 7482pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp, 7483 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 7484{ 7485 struct mbuf *m0, *md; 7486 struct sockaddr_in6 dst; 7487 struct ip6_hdr *ip6; 7488 struct pfi_kkif *nkif = NULL; 7489 struct ifnet *ifp = NULL; 7490 struct pf_addr naddr; 7491 struct pf_ksrc_node *sn = NULL; 7492 int r_rt, r_dir; 7493 7494 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 7495 7496 if (s) { 7497 r_rt = s->rt; 7498 r_dir = s->direction; 7499 } else { 7500 r_rt = r->rt; 7501 r_dir = r->direction; 7502 } 7503 7504 KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT || 7505 r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction", 7506 __func__)); 7507 7508 if ((pd->pf_mtag == NULL && 7509 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 7510 pd->pf_mtag->routed++ > 3) { 7511 m0 = *m; 7512 *m = NULL; 7513 goto bad_locked; 7514 } 7515 7516 if (r_rt == PF_DUPTO) { 7517 if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) { 7518 if (s == NULL) { 7519 ifp = r->rpool.cur->kif ? 7520 r->rpool.cur->kif->pfik_ifp : NULL; 7521 } else { 7522 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7523 /* If pfsync'd */ 7524 if (ifp == NULL && r->rpool.cur != NULL) 7525 ifp = r->rpool.cur->kif ? 7526 r->rpool.cur->kif->pfik_ifp : NULL; 7527 PF_STATE_UNLOCK(s); 7528 } 7529 if (ifp == oifp) { 7530 /* When the 2nd interface is not skipped */ 7531 return; 7532 } else { 7533 m0 = *m; 7534 *m = NULL; 7535 goto bad; 7536 } 7537 } else { 7538 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED; 7539 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 7540 if (s) 7541 PF_STATE_UNLOCK(s); 7542 return; 7543 } 7544 } 7545 } else { 7546 if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) { 7547 pf_dummynet(pd, s, r, m); 7548 if (s) 7549 PF_STATE_UNLOCK(s); 7550 return; 7551 } 7552 m0 = *m; 7553 } 7554 7555 ip6 = mtod(m0, struct ip6_hdr *); 7556 7557 bzero(&dst, sizeof(dst)); 7558 dst.sin6_family = AF_INET6; 7559 dst.sin6_len = sizeof(dst); 7560 dst.sin6_addr = ip6->ip6_dst; 7561 7562 bzero(&naddr, sizeof(naddr)); 7563 7564 if (s == NULL) { 7565 if (TAILQ_EMPTY(&r->rpool.list)) { 7566 DPFPRINTF(PF_DEBUG_URGENT, 7567 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 7568 goto bad_locked; 7569 } 7570 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 7571 &naddr, &nkif, NULL, &sn); 7572 if (!PF_AZERO(&naddr, AF_INET6)) 7573 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7574 &naddr, AF_INET6); 7575 ifp = nkif ? nkif->pfik_ifp : NULL; 7576 } else { 7577 struct pfi_kkif *kif; 7578 7579 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 7580 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 7581 &s->rt_addr, AF_INET6); 7582 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 7583 kif = s->rt_kif; 7584 /* If pfsync'd */ 7585 if (ifp == NULL && r->rpool.cur != NULL) { 7586 ifp = r->rpool.cur->kif ? 7587 r->rpool.cur->kif->pfik_ifp : NULL; 7588 kif = r->rpool.cur->kif; 7589 } 7590 if (ifp != NULL && kif != NULL && 7591 r->rule_flag & PFRULE_IFBOUND && 7592 r->rt == PF_REPLYTO && 7593 s->kif == V_pfi_all) { 7594 s->kif = kif; 7595 s->orig_kif = oifp->if_pf_kif; 7596 } 7597 } 7598 7599 if (s) 7600 PF_STATE_UNLOCK(s); 7601 7602 if (ifp == NULL) 7603 goto bad; 7604 7605 if (pd->dir == PF_IN) { 7606 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) 7607 goto bad; 7608 else if (m0 == NULL) 7609 goto done; 7610 if (m0->m_len < sizeof(struct ip6_hdr)) { 7611 DPFPRINTF(PF_DEBUG_URGENT, 7612 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 7613 __func__)); 7614 goto bad; 7615 } 7616 ip6 = mtod(m0, struct ip6_hdr *); 7617 } 7618 7619 if (ifp->if_flags & IFF_LOOPBACK) 7620 m0->m_flags |= M_SKIP_FIREWALL; 7621 7622 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 7623 ~ifp->if_hwassist) { 7624 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 7625 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 7626 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 7627 } 7628 7629 /* 7630 * If the packet is too large for the outgoing interface, 7631 * send back an icmp6 error. 7632 */ 7633 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 7634 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 7635 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) { 7636 md = m0; 7637 pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md); 7638 if (md != NULL) 7639 nd6_output_ifp(ifp, ifp, md, &dst, NULL); 7640 } 7641 else { 7642 in6_ifstat_inc(ifp, ifs6_in_toobig); 7643 if (r_rt != PF_DUPTO) { 7644 if (s && pd->nat_rule != NULL) 7645 PACKET_UNDO_NAT(m0, pd, 7646 ((caddr_t)ip6 - m0->m_data) + 7647 sizeof(struct ip6_hdr), s); 7648 7649 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 7650 } else 7651 goto bad; 7652 } 7653 7654done: 7655 if (r_rt != PF_DUPTO) 7656 *m = NULL; 7657 return; 7658 7659bad_locked: 7660 if (s) 7661 PF_STATE_UNLOCK(s); 7662bad: 7663 m_freem(m0); 7664 goto done; 7665} 7666#endif /* INET6 */ 7667 7668/* 7669 * FreeBSD supports cksum offloads for the following drivers. 7670 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 7671 * 7672 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 7673 * network driver performed cksum including pseudo header, need to verify 7674 * csum_data 7675 * CSUM_DATA_VALID : 7676 * network driver performed cksum, needs to additional pseudo header 7677 * cksum computation with partial csum_data(i.e. lack of H/W support for 7678 * pseudo header, for instance sk(4) and possibly gem(4)) 7679 * 7680 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 7681 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 7682 * TCP/UDP layer. 7683 * Also, set csum_data to 0xffff to force cksum validation. 7684 */ 7685static int 7686pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 7687{ 7688 u_int16_t sum = 0; 7689 int hw_assist = 0; 7690 struct ip *ip; 7691 7692 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 7693 return (1); 7694 if (m->m_pkthdr.len < off + len) 7695 return (1); 7696 7697 switch (p) { 7698 case IPPROTO_TCP: 7699 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7700 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7701 sum = m->m_pkthdr.csum_data; 7702 } else { 7703 ip = mtod(m, struct ip *); 7704 sum = in_pseudo(ip->ip_src.s_addr, 7705 ip->ip_dst.s_addr, htonl((u_short)len + 7706 m->m_pkthdr.csum_data + IPPROTO_TCP)); 7707 } 7708 sum ^= 0xffff; 7709 ++hw_assist; 7710 } 7711 break; 7712 case IPPROTO_UDP: 7713 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 7714 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 7715 sum = m->m_pkthdr.csum_data; 7716 } else { 7717 ip = mtod(m, struct ip *); 7718 sum = in_pseudo(ip->ip_src.s_addr, 7719 ip->ip_dst.s_addr, htonl((u_short)len + 7720 m->m_pkthdr.csum_data + IPPROTO_UDP)); 7721 } 7722 sum ^= 0xffff; 7723 ++hw_assist; 7724 } 7725 break; 7726 case IPPROTO_ICMP: 7727#ifdef INET6 7728 case IPPROTO_ICMPV6: 7729#endif /* INET6 */ 7730 break; 7731 default: 7732 return (1); 7733 } 7734 7735 if (!hw_assist) { 7736 switch (af) { 7737 case AF_INET: 7738 if (p == IPPROTO_ICMP) { 7739 if (m->m_len < off) 7740 return (1); 7741 m->m_data += off; 7742 m->m_len -= off; 7743 sum = in_cksum(m, len); 7744 m->m_data -= off; 7745 m->m_len += off; 7746 } else { 7747 if (m->m_len < sizeof(struct ip)) 7748 return (1); 7749 sum = in4_cksum(m, p, off, len); 7750 } 7751 break; 7752#ifdef INET6 7753 case AF_INET6: 7754 if (m->m_len < sizeof(struct ip6_hdr)) 7755 return (1); 7756 sum = in6_cksum(m, p, off, len); 7757 break; 7758#endif /* INET6 */ 7759 default: 7760 return (1); 7761 } 7762 } 7763 if (sum) { 7764 switch (p) { 7765 case IPPROTO_TCP: 7766 { 7767 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 7768 break; 7769 } 7770 case IPPROTO_UDP: 7771 { 7772 KMOD_UDPSTAT_INC(udps_badsum); 7773 break; 7774 } 7775#ifdef INET 7776 case IPPROTO_ICMP: 7777 { 7778 KMOD_ICMPSTAT_INC(icps_checksum); 7779 break; 7780 } 7781#endif 7782#ifdef INET6 7783 case IPPROTO_ICMPV6: 7784 { 7785 KMOD_ICMP6STAT_INC(icp6s_checksum); 7786 break; 7787 } 7788#endif /* INET6 */ 7789 } 7790 return (1); 7791 } else { 7792 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 7793 m->m_pkthdr.csum_flags |= 7794 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 7795 m->m_pkthdr.csum_data = 0xffff; 7796 } 7797 } 7798 return (0); 7799} 7800 7801static bool 7802pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r, 7803 const struct pf_kstate *s, struct ip_fw_args *dnflow) 7804{ 7805 int dndir = r->direction; 7806 7807 if (s && dndir == PF_INOUT) { 7808 dndir = s->direction; 7809 } else if (dndir == PF_INOUT) { 7810 /* Assume primary direction. Happens when we've set dnpipe in 7811 * the ethernet level code. */ 7812 dndir = pd->dir; 7813 } 7814 7815 if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED) 7816 return (false); 7817 7818 memset(dnflow, 0, sizeof(*dnflow)); 7819 7820 if (pd->dport != NULL) 7821 dnflow->f_id.dst_port = ntohs(*pd->dport); 7822 if (pd->sport != NULL) 7823 dnflow->f_id.src_port = ntohs(*pd->sport); 7824 7825 if (pd->dir == PF_IN) 7826 dnflow->flags |= IPFW_ARGS_IN; 7827 else 7828 dnflow->flags |= IPFW_ARGS_OUT; 7829 7830 if (pd->dir != dndir && pd->act.dnrpipe) { 7831 dnflow->rule.info = pd->act.dnrpipe; 7832 } 7833 else if (pd->dir == dndir && pd->act.dnpipe) { 7834 dnflow->rule.info = pd->act.dnpipe; 7835 } 7836 else { 7837 return (false); 7838 } 7839 7840 dnflow->rule.info |= IPFW_IS_DUMMYNET; 7841 if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE) 7842 dnflow->rule.info |= IPFW_IS_PIPE; 7843 7844 dnflow->f_id.proto = pd->proto; 7845 dnflow->f_id.extra = dnflow->rule.info; 7846 switch (pd->af) { 7847 case AF_INET: 7848 dnflow->f_id.addr_type = 4; 7849 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 7850 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 7851 break; 7852 case AF_INET6: 7853 dnflow->flags |= IPFW_ARGS_IP6; 7854 dnflow->f_id.addr_type = 6; 7855 dnflow->f_id.src_ip6 = pd->src->v6; 7856 dnflow->f_id.dst_ip6 = pd->dst->v6; 7857 break; 7858 default: 7859 panic("Invalid AF"); 7860 break; 7861 } 7862 7863 return (true); 7864} 7865 7866int 7867pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7868 struct inpcb *inp) 7869{ 7870 struct pfi_kkif *kif; 7871 struct mbuf *m = *m0; 7872 7873 M_ASSERTPKTHDR(m); 7874 MPASS(ifp->if_vnet == curvnet); 7875 NET_EPOCH_ASSERT(); 7876 7877 if (!V_pf_status.running) 7878 return (PF_PASS); 7879 7880 kif = (struct pfi_kkif *)ifp->if_pf_kif; 7881 7882 if (kif == NULL) { 7883 DPFPRINTF(PF_DEBUG_URGENT, 7884 ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname)); 7885 return (PF_DROP); 7886 } 7887 if (kif->pfik_flags & PFI_IFLAG_SKIP) 7888 return (PF_PASS); 7889 7890 if (m->m_flags & M_SKIP_FIREWALL) 7891 return (PF_PASS); 7892 7893 /* Stateless! */ 7894 return (pf_test_eth_rule(dir, kif, m0)); 7895} 7896 7897static __inline void 7898pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag) 7899{ 7900 struct m_tag *mtag; 7901 7902 pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET; 7903 7904 /* dummynet adds this tag, but pf does not need it, 7905 * and keeping it creates unexpected behavior, 7906 * e.g. in case of divert(4) usage right after dummynet. */ 7907 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 7908 if (mtag != NULL) 7909 m_tag_delete(m, mtag); 7910} 7911 7912static int 7913pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s, 7914 struct pf_krule *r, struct mbuf **m0) 7915{ 7916 return (pf_dummynet_route(pd, s, r, NULL, NULL, m0)); 7917} 7918 7919static int 7920pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s, 7921 struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa, 7922 struct mbuf **m0) 7923{ 7924 NET_EPOCH_ASSERT(); 7925 7926 if (pd->act.dnpipe || pd->act.dnrpipe) { 7927 struct ip_fw_args dnflow; 7928 if (ip_dn_io_ptr == NULL) { 7929 m_freem(*m0); 7930 *m0 = NULL; 7931 return (ENOMEM); 7932 } 7933 7934 if (pd->pf_mtag == NULL && 7935 ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) { 7936 m_freem(*m0); 7937 *m0 = NULL; 7938 return (ENOMEM); 7939 } 7940 7941 if (ifp != NULL) { 7942 pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO; 7943 7944 pd->pf_mtag->if_index = ifp->if_index; 7945 pd->pf_mtag->if_idxgen = ifp->if_idxgen; 7946 7947 MPASS(sa != NULL); 7948 7949 if (pd->af == AF_INET) 7950 memcpy(&pd->pf_mtag->dst, sa, 7951 sizeof(struct sockaddr_in)); 7952 else 7953 memcpy(&pd->pf_mtag->dst, sa, 7954 sizeof(struct sockaddr_in6)); 7955 } 7956 7957 if (s != NULL && s->nat_rule.ptr != NULL && 7958 s->nat_rule.ptr->action == PF_RDR && 7959 ( 7960#ifdef INET 7961 (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) || 7962#endif 7963 (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) { 7964 /* 7965 * If we're redirecting to loopback mark this packet 7966 * as being local. Otherwise it might get dropped 7967 * if dummynet re-injects. 7968 */ 7969 (*m0)->m_pkthdr.rcvif = V_loif; 7970 } 7971 7972 if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) { 7973 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET; 7974 pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED; 7975 ip_dn_io_ptr(m0, &dnflow); 7976 if (*m0 != NULL) { 7977 pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 7978 pf_dummynet_flag_remove(*m0, pd->pf_mtag); 7979 } 7980 } 7981 } 7982 7983 return (0); 7984} 7985 7986#ifdef INET 7987int 7988pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, 7989 struct inpcb *inp, struct pf_rule_actions *default_actions) 7990{ 7991 struct pfi_kkif *kif; 7992 u_short action, reason = 0; 7993 struct mbuf *m = *m0; 7994 struct ip *h = NULL; 7995 struct m_tag *mtag; 7996 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 7997 struct pf_kstate *s = NULL; 7998 struct pf_kruleset *ruleset = NULL; 7999 struct pf_pdesc pd; 8000 int off, dirndx, use_2nd_queue = 0; 8001 uint16_t tag; 8002 uint8_t rt; 8003 8004 PF_RULES_RLOCK_TRACKER; 8005 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8006 M_ASSERTPKTHDR(m); 8007 8008 if (!V_pf_status.running) 8009 return (PF_PASS); 8010 8011 PF_RULES_RLOCK(); 8012 8013 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8014 8015 if (__predict_false(kif == NULL)) { 8016 DPFPRINTF(PF_DEBUG_URGENT, 8017 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 8018 PF_RULES_RUNLOCK(); 8019 return (PF_DROP); 8020 } 8021 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8022 PF_RULES_RUNLOCK(); 8023 return (PF_PASS); 8024 } 8025 8026 if (m->m_flags & M_SKIP_FIREWALL) { 8027 PF_RULES_RUNLOCK(); 8028 return (PF_PASS); 8029 } 8030 8031 memset(&pd, 0, sizeof(pd)); 8032 TAILQ_INIT(&pd.sctp_multihome_jobs); 8033 if (default_actions != NULL) 8034 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8035 pd.pf_mtag = pf_find_mtag(m); 8036 8037 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8038 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8039 8040 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8041 pd.pf_mtag->if_idxgen); 8042 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8043 PF_RULES_RUNLOCK(); 8044 m_freem(*m0); 8045 *m0 = NULL; 8046 return (PF_PASS); 8047 } 8048 PF_RULES_RUNLOCK(); 8049 (ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL); 8050 *m0 = NULL; 8051 return (PF_PASS); 8052 } 8053 8054 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8055 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8056 pd.act.flags = pd.pf_mtag->dnflags; 8057 } 8058 8059 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8060 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8061 /* Dummynet re-injects packets after they've 8062 * completed their delay. We've already 8063 * processed them, so pass unconditionally. */ 8064 8065 /* But only once. We may see the packet multiple times (e.g. 8066 * PFIL_IN/PFIL_OUT). */ 8067 pf_dummynet_flag_remove(m, pd.pf_mtag); 8068 PF_RULES_RUNLOCK(); 8069 8070 return (PF_PASS); 8071 } 8072 8073 pd.sport = pd.dport = NULL; 8074 pd.proto_sum = NULL; 8075 pd.dir = dir; 8076 pd.sidx = (dir == PF_IN) ? 0 : 1; 8077 pd.didx = (dir == PF_IN) ? 1 : 0; 8078 pd.af = AF_INET; 8079 pd.act.rtableid = -1; 8080 8081 h = mtod(m, struct ip *); 8082 off = h->ip_hl << 2; 8083 8084 if (__predict_false(ip_divert_ptr != NULL) && 8085 ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) { 8086 struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1); 8087 if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) || 8088 (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) { 8089 if (pd.pf_mtag == NULL && 8090 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8091 action = PF_DROP; 8092 goto done; 8093 } 8094 pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED; 8095 } 8096 if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) { 8097 m->m_flags |= M_FASTFWD_OURS; 8098 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8099 } 8100 m_tag_delete(m, mtag); 8101 8102 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 8103 if (mtag != NULL) 8104 m_tag_delete(m, mtag); 8105 } else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) { 8106 /* We do IP header normalization and packet reassembly here */ 8107 action = PF_DROP; 8108 goto done; 8109 } 8110 m = *m0; /* pf_normalize messes with m0 */ 8111 h = mtod(m, struct ip *); 8112 8113 off = h->ip_hl << 2; 8114 if (off < (int)sizeof(struct ip)) { 8115 action = PF_DROP; 8116 REASON_SET(&reason, PFRES_SHORT); 8117 pd.act.log = PF_LOG_FORCE; 8118 goto done; 8119 } 8120 8121 pd.src = (struct pf_addr *)&h->ip_src; 8122 pd.dst = (struct pf_addr *)&h->ip_dst; 8123 pd.ip_sum = &h->ip_sum; 8124 pd.proto = h->ip_p; 8125 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 8126 pd.tot_len = ntohs(h->ip_len); 8127 8128 /* handle fragments that didn't get reassembled by normalization */ 8129 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 8130 action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset); 8131 goto done; 8132 } 8133 8134 switch (h->ip_p) { 8135 case IPPROTO_TCP: { 8136 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8137 &action, &reason, AF_INET)) { 8138 if (action != PF_PASS) 8139 pd.act.log = PF_LOG_FORCE; 8140 goto done; 8141 } 8142 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8143 8144 pd.sport = &pd.hdr.tcp.th_sport; 8145 pd.dport = &pd.hdr.tcp.th_dport; 8146 8147 /* Respond to SYN with a syncookie. */ 8148 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8149 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8150 pf_syncookie_send(m, off, &pd); 8151 action = PF_DROP; 8152 break; 8153 } 8154 8155 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 8156 use_2nd_queue = 1; 8157 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8158 if (action == PF_DROP) 8159 goto done; 8160 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8161 if (action == PF_PASS) { 8162 if (V_pfsync_update_state_ptr != NULL) 8163 V_pfsync_update_state_ptr(s); 8164 r = s->rule.ptr; 8165 a = s->anchor.ptr; 8166 } else if (s == NULL) { 8167 /* Validate remote SYN|ACK, re-create original SYN if 8168 * valid. */ 8169 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8170 TH_ACK && pf_syncookie_validate(&pd) && 8171 pd.dir == PF_IN) { 8172 struct mbuf *msyn; 8173 8174 msyn = pf_syncookie_recreate_syn(h->ip_ttl, off, 8175 &pd); 8176 if (msyn == NULL) { 8177 action = PF_DROP; 8178 break; 8179 } 8180 8181 action = pf_test(dir, pflags, ifp, &msyn, inp, 8182 &pd.act); 8183 m_freem(msyn); 8184 if (action != PF_PASS) 8185 break; 8186 8187 action = pf_test_state_tcp(&s, kif, m, off, h, 8188 &pd, &reason); 8189 if (action != PF_PASS || s == NULL) { 8190 action = PF_DROP; 8191 break; 8192 } 8193 8194 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8195 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8196 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8197 action = pf_synproxy(&pd, &s, &reason); 8198 break; 8199 } else { 8200 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8201 &a, &ruleset, inp); 8202 } 8203 } 8204 break; 8205 } 8206 8207 case IPPROTO_UDP: { 8208 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8209 &action, &reason, AF_INET)) { 8210 if (action != PF_PASS) 8211 pd.act.log = PF_LOG_FORCE; 8212 goto done; 8213 } 8214 pd.sport = &pd.hdr.udp.uh_sport; 8215 pd.dport = &pd.hdr.udp.uh_dport; 8216 if (pd.hdr.udp.uh_dport == 0 || 8217 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8218 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8219 action = PF_DROP; 8220 REASON_SET(&reason, PFRES_SHORT); 8221 goto done; 8222 } 8223 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8224 if (action == PF_PASS) { 8225 if (V_pfsync_update_state_ptr != NULL) 8226 V_pfsync_update_state_ptr(s); 8227 r = s->rule.ptr; 8228 a = s->anchor.ptr; 8229 } else if (s == NULL) 8230 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8231 &a, &ruleset, inp); 8232 break; 8233 } 8234 8235 case IPPROTO_SCTP: { 8236 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8237 &action, &reason, AF_INET)) { 8238 if (action != PF_PASS) 8239 pd.act.log |= PF_LOG_FORCE; 8240 goto done; 8241 } 8242 pd.p_len = pd.tot_len - off; 8243 8244 pd.sport = &pd.hdr.sctp.src_port; 8245 pd.dport = &pd.hdr.sctp.dest_port; 8246 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8247 action = PF_DROP; 8248 REASON_SET(&reason, PFRES_SHORT); 8249 goto done; 8250 } 8251 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8252 if (action == PF_DROP) 8253 goto done; 8254 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8255 &reason); 8256 if (action == PF_PASS) { 8257 if (V_pfsync_update_state_ptr != NULL) 8258 V_pfsync_update_state_ptr(s); 8259 r = s->rule.ptr; 8260 a = s->anchor.ptr; 8261 } else { 8262 action = pf_test_rule(&r, &s, kif, m, off, 8263 &pd, &a, &ruleset, inp); 8264 } 8265 break; 8266 } 8267 8268 case IPPROTO_ICMP: { 8269 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 8270 &action, &reason, AF_INET)) { 8271 if (action != PF_PASS) 8272 pd.act.log = PF_LOG_FORCE; 8273 goto done; 8274 } 8275 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8276 if (action == PF_PASS) { 8277 if (V_pfsync_update_state_ptr != NULL) 8278 V_pfsync_update_state_ptr(s); 8279 r = s->rule.ptr; 8280 a = s->anchor.ptr; 8281 } else if (s == NULL) 8282 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8283 &a, &ruleset, inp); 8284 break; 8285 } 8286 8287#ifdef INET6 8288 case IPPROTO_ICMPV6: { 8289 action = PF_DROP; 8290 DPFPRINTF(PF_DEBUG_MISC, 8291 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 8292 goto done; 8293 } 8294#endif 8295 8296 default: 8297 action = pf_test_state_other(&s, kif, m, &pd); 8298 if (action == PF_PASS) { 8299 if (V_pfsync_update_state_ptr != NULL) 8300 V_pfsync_update_state_ptr(s); 8301 r = s->rule.ptr; 8302 a = s->anchor.ptr; 8303 } else if (s == NULL) 8304 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8305 &a, &ruleset, inp); 8306 break; 8307 } 8308 8309done: 8310 PF_RULES_RUNLOCK(); 8311 if (action == PF_PASS && h->ip_hl > 5 && 8312 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8313 action = PF_DROP; 8314 REASON_SET(&reason, PFRES_IPOPTIONS); 8315 pd.act.log = PF_LOG_FORCE; 8316 DPFPRINTF(PF_DEBUG_MISC, 8317 ("pf: dropping packet with ip options\n")); 8318 } 8319 8320 if (s) { 8321 uint8_t log = pd.act.log; 8322 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8323 pd.act.log |= log; 8324 tag = s->tag; 8325 rt = s->rt; 8326 } else { 8327 tag = r->tag; 8328 rt = r->rt; 8329 } 8330 8331 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8332 action = PF_DROP; 8333 REASON_SET(&reason, PFRES_MEMORY); 8334 } 8335 8336 pf_scrub_ip(&m, &pd); 8337 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8338 pf_normalize_mss(m, off, &pd); 8339 8340 if (pd.act.rtableid >= 0) 8341 M_SETFIB(m, pd.act.rtableid); 8342 8343 if (pd.act.flags & PFSTATE_SETPRIO) { 8344 if (pd.tos & IPTOS_LOWDELAY) 8345 use_2nd_queue = 1; 8346 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8347 action = PF_DROP; 8348 REASON_SET(&reason, PFRES_MEMORY); 8349 pd.act.log = PF_LOG_FORCE; 8350 DPFPRINTF(PF_DEBUG_MISC, 8351 ("pf: failed to allocate 802.1q mtag\n")); 8352 } 8353 } 8354 8355#ifdef ALTQ 8356 if (action == PF_PASS && pd.act.qid) { 8357 if (pd.pf_mtag == NULL && 8358 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8359 action = PF_DROP; 8360 REASON_SET(&reason, PFRES_MEMORY); 8361 } else { 8362 if (s != NULL) 8363 pd.pf_mtag->qid_hash = pf_state_hash(s); 8364 if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY)) 8365 pd.pf_mtag->qid = pd.act.pqid; 8366 else 8367 pd.pf_mtag->qid = pd.act.qid; 8368 /* Add hints for ecn. */ 8369 pd.pf_mtag->hdr = h; 8370 } 8371 } 8372#endif /* ALTQ */ 8373 8374 /* 8375 * connections redirected to loopback should not match sockets 8376 * bound specifically to loopback due to security implications, 8377 * see tcp_input() and in_pcblookup_listen(). 8378 */ 8379 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 8380 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 8381 (s->nat_rule.ptr->action == PF_RDR || 8382 s->nat_rule.ptr->action == PF_BINAT) && 8383 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 8384 m->m_flags |= M_SKIP_FIREWALL; 8385 8386 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 8387 r->divert.port && !PACKET_LOOPED(&pd)) { 8388 mtag = m_tag_alloc(MTAG_PF_DIVERT, 0, 8389 sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO); 8390 if (mtag != NULL) { 8391 ((struct pf_divert_mtag *)(mtag+1))->port = 8392 ntohs(r->divert.port); 8393 ((struct pf_divert_mtag *)(mtag+1))->idir = 8394 (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN : 8395 PF_DIVERT_MTAG_DIR_OUT; 8396 8397 if (s) 8398 PF_STATE_UNLOCK(s); 8399 8400 m_tag_prepend(m, mtag); 8401 if (m->m_flags & M_FASTFWD_OURS) { 8402 if (pd.pf_mtag == NULL && 8403 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8404 action = PF_DROP; 8405 REASON_SET(&reason, PFRES_MEMORY); 8406 pd.act.log = PF_LOG_FORCE; 8407 DPFPRINTF(PF_DEBUG_MISC, 8408 ("pf: failed to allocate tag\n")); 8409 } else { 8410 pd.pf_mtag->flags |= 8411 PF_MTAG_FLAG_FASTFWD_OURS_PRESENT; 8412 m->m_flags &= ~M_FASTFWD_OURS; 8413 } 8414 } 8415 ip_divert_ptr(*m0, dir == PF_IN); 8416 *m0 = NULL; 8417 8418 return (action); 8419 } else { 8420 /* XXX: ipfw has the same behaviour! */ 8421 action = PF_DROP; 8422 REASON_SET(&reason, PFRES_MEMORY); 8423 pd.act.log = PF_LOG_FORCE; 8424 DPFPRINTF(PF_DEBUG_MISC, 8425 ("pf: failed to allocate divert tag\n")); 8426 } 8427 } 8428 /* this flag will need revising if the pkt is forwarded */ 8429 if (pd.pf_mtag) 8430 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED; 8431 8432 if (pd.act.log) { 8433 struct pf_krule *lr; 8434 struct pf_krule_item *ri; 8435 8436 if (s != NULL && s->nat_rule.ptr != NULL && 8437 s->nat_rule.ptr->log & PF_LOG_ALL) 8438 lr = s->nat_rule.ptr; 8439 else 8440 lr = r; 8441 8442 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 8443 PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a, 8444 ruleset, &pd, (s == NULL)); 8445 if (s) { 8446 SLIST_FOREACH(ri, &s->match_rules, entry) 8447 if (ri->r->log & PF_LOG_ALL) 8448 PFLOG_PACKET(kif, m, AF_INET, action, 8449 reason, ri->r, a, ruleset, &pd, 0); 8450 } 8451 } 8452 8453 pf_counter_u64_critical_enter(); 8454 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 8455 pd.tot_len); 8456 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 8457 1); 8458 8459 if (action == PF_PASS || r->action == PF_DROP) { 8460 dirndx = (dir == PF_OUT); 8461 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 8462 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 8463 pf_update_timestamp(r); 8464 8465 if (a != NULL) { 8466 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 8467 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 8468 } 8469 if (s != NULL) { 8470 struct pf_krule_item *ri; 8471 8472 if (s->nat_rule.ptr != NULL) { 8473 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 8474 1); 8475 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 8476 pd.tot_len); 8477 } 8478 if (s->src_node != NULL) { 8479 counter_u64_add(s->src_node->packets[dirndx], 8480 1); 8481 counter_u64_add(s->src_node->bytes[dirndx], 8482 pd.tot_len); 8483 } 8484 if (s->nat_src_node != NULL) { 8485 counter_u64_add(s->nat_src_node->packets[dirndx], 8486 1); 8487 counter_u64_add(s->nat_src_node->bytes[dirndx], 8488 pd.tot_len); 8489 } 8490 dirndx = (dir == s->direction) ? 0 : 1; 8491 s->packets[dirndx]++; 8492 s->bytes[dirndx] += pd.tot_len; 8493 SLIST_FOREACH(ri, &s->match_rules, entry) { 8494 pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1); 8495 pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len); 8496 } 8497 } 8498 tr = r; 8499 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 8500 if (nr != NULL && r == &V_pf_default_rule) 8501 tr = nr; 8502 if (tr->src.addr.type == PF_ADDR_TABLE) 8503 pfr_update_stats(tr->src.addr.p.tbl, 8504 (s == NULL) ? pd.src : 8505 &s->key[(s->direction == PF_IN)]-> 8506 addr[(s->direction == PF_OUT)], 8507 pd.af, pd.tot_len, dir == PF_OUT, 8508 r->action == PF_PASS, tr->src.neg); 8509 if (tr->dst.addr.type == PF_ADDR_TABLE) 8510 pfr_update_stats(tr->dst.addr.p.tbl, 8511 (s == NULL) ? pd.dst : 8512 &s->key[(s->direction == PF_IN)]-> 8513 addr[(s->direction == PF_IN)], 8514 pd.af, pd.tot_len, dir == PF_OUT, 8515 r->action == PF_PASS, tr->dst.neg); 8516 } 8517 pf_counter_u64_critical_exit(); 8518 8519 switch (action) { 8520 case PF_SYNPROXY_DROP: 8521 m_freem(*m0); 8522 case PF_DEFER: 8523 *m0 = NULL; 8524 action = PF_PASS; 8525 break; 8526 case PF_DROP: 8527 m_freem(*m0); 8528 *m0 = NULL; 8529 break; 8530 default: 8531 /* pf_route() returns unlocked. */ 8532 if (rt) { 8533 pf_route(m0, r, kif->pfik_ifp, s, &pd, inp); 8534 goto out; 8535 } 8536 if (pf_dummynet(&pd, s, r, m0) != 0) { 8537 action = PF_DROP; 8538 REASON_SET(&reason, PFRES_MEMORY); 8539 } 8540 break; 8541 } 8542 8543 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 8544 8545 if (s && action != PF_DROP) { 8546 if (!s->if_index_in && dir == PF_IN) 8547 s->if_index_in = ifp->if_index; 8548 else if (!s->if_index_out && dir == PF_OUT) 8549 s->if_index_out = ifp->if_index; 8550 } 8551 8552 if (s) 8553 PF_STATE_UNLOCK(s); 8554 8555out: 8556 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 8557 8558 return (action); 8559} 8560#endif /* INET */ 8561 8562#ifdef INET6 8563int 8564pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp, 8565 struct pf_rule_actions *default_actions) 8566{ 8567 struct pfi_kkif *kif; 8568 u_short action, reason = 0; 8569 struct mbuf *m = *m0, *n = NULL; 8570 struct m_tag *mtag; 8571 struct ip6_hdr *h = NULL; 8572 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 8573 struct pf_kstate *s = NULL; 8574 struct pf_kruleset *ruleset = NULL; 8575 struct pf_pdesc pd; 8576 int off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0; 8577 uint16_t tag; 8578 uint8_t rt; 8579 8580 PF_RULES_RLOCK_TRACKER; 8581 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 8582 M_ASSERTPKTHDR(m); 8583 8584 if (!V_pf_status.running) 8585 return (PF_PASS); 8586 8587 PF_RULES_RLOCK(); 8588 8589 kif = (struct pfi_kkif *)ifp->if_pf_kif; 8590 if (__predict_false(kif == NULL)) { 8591 DPFPRINTF(PF_DEBUG_URGENT, 8592 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 8593 PF_RULES_RUNLOCK(); 8594 return (PF_DROP); 8595 } 8596 if (kif->pfik_flags & PFI_IFLAG_SKIP) { 8597 PF_RULES_RUNLOCK(); 8598 return (PF_PASS); 8599 } 8600 8601 if (m->m_flags & M_SKIP_FIREWALL) { 8602 PF_RULES_RUNLOCK(); 8603 return (PF_PASS); 8604 } 8605 8606 /* 8607 * If we end up changing IP addresses (e.g. binat) the stack may get 8608 * confused and fail to send the icmp6 packet too big error. Just send 8609 * it here, before we do any NAT. 8610 */ 8611 if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) { 8612 PF_RULES_RUNLOCK(); 8613 *m0 = NULL; 8614 icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp)); 8615 return (PF_DROP); 8616 } 8617 8618 memset(&pd, 0, sizeof(pd)); 8619 TAILQ_INIT(&pd.sctp_multihome_jobs); 8620 if (default_actions != NULL) 8621 memcpy(&pd.act, default_actions, sizeof(pd.act)); 8622 pd.pf_mtag = pf_find_mtag(m); 8623 8624 if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) { 8625 pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO; 8626 8627 ifp = ifnet_byindexgen(pd.pf_mtag->if_index, 8628 pd.pf_mtag->if_idxgen); 8629 if (ifp == NULL || ifp->if_flags & IFF_DYING) { 8630 PF_RULES_RUNLOCK(); 8631 m_freem(*m0); 8632 *m0 = NULL; 8633 return (PF_PASS); 8634 } 8635 PF_RULES_RUNLOCK(); 8636 nd6_output_ifp(ifp, ifp, m, 8637 (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL); 8638 *m0 = NULL; 8639 return (PF_PASS); 8640 } 8641 8642 if (pd.pf_mtag && pd.pf_mtag->dnpipe) { 8643 pd.act.dnpipe = pd.pf_mtag->dnpipe; 8644 pd.act.flags = pd.pf_mtag->dnflags; 8645 } 8646 8647 if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL && 8648 pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) { 8649 pf_dummynet_flag_remove(m, pd.pf_mtag); 8650 /* Dummynet re-injects packets after they've 8651 * completed their delay. We've already 8652 * processed them, so pass unconditionally. */ 8653 PF_RULES_RUNLOCK(); 8654 return (PF_PASS); 8655 } 8656 8657 pd.sport = pd.dport = NULL; 8658 pd.ip_sum = NULL; 8659 pd.proto_sum = NULL; 8660 pd.dir = dir; 8661 pd.sidx = (dir == PF_IN) ? 0 : 1; 8662 pd.didx = (dir == PF_IN) ? 1 : 0; 8663 pd.af = AF_INET6; 8664 pd.act.rtableid = -1; 8665 8666 h = mtod(m, struct ip6_hdr *); 8667 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8668 8669 /* We do IP header normalization and packet reassembly here */ 8670 if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) { 8671 action = PF_DROP; 8672 goto done; 8673 } 8674 m = *m0; /* pf_normalize messes with m0 */ 8675 h = mtod(m, struct ip6_hdr *); 8676 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 8677 8678 /* 8679 * we do not support jumbogram. if we keep going, zero ip6_plen 8680 * will do something bad, so drop the packet for now. 8681 */ 8682 if (htons(h->ip6_plen) == 0) { 8683 action = PF_DROP; 8684 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 8685 goto done; 8686 } 8687 8688 pd.src = (struct pf_addr *)&h->ip6_src; 8689 pd.dst = (struct pf_addr *)&h->ip6_dst; 8690 pd.tos = IPV6_DSCP(h); 8691 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 8692 8693 pd.proto = h->ip6_nxt; 8694 do { 8695 switch (pd.proto) { 8696 case IPPROTO_FRAGMENT: 8697 action = pf_test_fragment(&r, kif, m, h, &pd, &a, 8698 &ruleset); 8699 if (action == PF_DROP) 8700 REASON_SET(&reason, PFRES_FRAG); 8701 goto done; 8702 case IPPROTO_ROUTING: { 8703 struct ip6_rthdr rthdr; 8704 8705 if (rh_cnt++) { 8706 DPFPRINTF(PF_DEBUG_MISC, 8707 ("pf: IPv6 more than one rthdr\n")); 8708 action = PF_DROP; 8709 REASON_SET(&reason, PFRES_IPOPTIONS); 8710 pd.act.log = PF_LOG_FORCE; 8711 goto done; 8712 } 8713 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 8714 &reason, pd.af)) { 8715 DPFPRINTF(PF_DEBUG_MISC, 8716 ("pf: IPv6 short rthdr\n")); 8717 action = PF_DROP; 8718 REASON_SET(&reason, PFRES_SHORT); 8719 pd.act.log = PF_LOG_FORCE; 8720 goto done; 8721 } 8722 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 8723 DPFPRINTF(PF_DEBUG_MISC, 8724 ("pf: IPv6 rthdr0\n")); 8725 action = PF_DROP; 8726 REASON_SET(&reason, PFRES_IPOPTIONS); 8727 pd.act.log = PF_LOG_FORCE; 8728 goto done; 8729 } 8730 /* FALLTHROUGH */ 8731 } 8732 case IPPROTO_AH: 8733 case IPPROTO_HOPOPTS: 8734 case IPPROTO_DSTOPTS: { 8735 /* get next header and header length */ 8736 struct ip6_ext opt6; 8737 8738 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 8739 NULL, &reason, pd.af)) { 8740 DPFPRINTF(PF_DEBUG_MISC, 8741 ("pf: IPv6 short opt\n")); 8742 action = PF_DROP; 8743 pd.act.log = PF_LOG_FORCE; 8744 goto done; 8745 } 8746 if (pd.proto == IPPROTO_AH) 8747 off += (opt6.ip6e_len + 2) * 4; 8748 else 8749 off += (opt6.ip6e_len + 1) * 8; 8750 pd.proto = opt6.ip6e_nxt; 8751 /* goto the next header */ 8752 break; 8753 } 8754 default: 8755 terminal++; 8756 break; 8757 } 8758 } while (!terminal); 8759 8760 /* if there's no routing header, use unmodified mbuf for checksumming */ 8761 if (!n) 8762 n = m; 8763 8764 switch (pd.proto) { 8765 case IPPROTO_TCP: { 8766 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 8767 &action, &reason, AF_INET6)) { 8768 if (action != PF_PASS) 8769 pd.act.log |= PF_LOG_FORCE; 8770 goto done; 8771 } 8772 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 8773 pd.sport = &pd.hdr.tcp.th_sport; 8774 pd.dport = &pd.hdr.tcp.th_dport; 8775 8776 /* Respond to SYN with a syncookie. */ 8777 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 8778 pd.dir == PF_IN && pf_synflood_check(&pd)) { 8779 pf_syncookie_send(m, off, &pd); 8780 action = PF_DROP; 8781 break; 8782 } 8783 8784 action = pf_normalize_tcp(kif, m, 0, off, h, &pd); 8785 if (action == PF_DROP) 8786 goto done; 8787 action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason); 8788 if (action == PF_PASS) { 8789 if (V_pfsync_update_state_ptr != NULL) 8790 V_pfsync_update_state_ptr(s); 8791 r = s->rule.ptr; 8792 a = s->anchor.ptr; 8793 } else if (s == NULL) { 8794 /* Validate remote SYN|ACK, re-create original SYN if 8795 * valid. */ 8796 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 8797 TH_ACK && pf_syncookie_validate(&pd) && 8798 pd.dir == PF_IN) { 8799 struct mbuf *msyn; 8800 8801 msyn = pf_syncookie_recreate_syn(h->ip6_hlim, 8802 off, &pd); 8803 if (msyn == NULL) { 8804 action = PF_DROP; 8805 break; 8806 } 8807 8808 action = pf_test6(dir, pflags, ifp, &msyn, inp, 8809 &pd.act); 8810 m_freem(msyn); 8811 if (action != PF_PASS) 8812 break; 8813 8814 action = pf_test_state_tcp(&s, kif, m, off, h, 8815 &pd, &reason); 8816 if (action != PF_PASS || s == NULL) { 8817 action = PF_DROP; 8818 break; 8819 } 8820 8821 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1; 8822 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1; 8823 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST); 8824 8825 action = pf_synproxy(&pd, &s, &reason); 8826 break; 8827 } else { 8828 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8829 &a, &ruleset, inp); 8830 } 8831 } 8832 break; 8833 } 8834 8835 case IPPROTO_UDP: { 8836 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 8837 &action, &reason, AF_INET6)) { 8838 if (action != PF_PASS) 8839 pd.act.log |= PF_LOG_FORCE; 8840 goto done; 8841 } 8842 pd.sport = &pd.hdr.udp.uh_sport; 8843 pd.dport = &pd.hdr.udp.uh_dport; 8844 if (pd.hdr.udp.uh_dport == 0 || 8845 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 8846 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 8847 action = PF_DROP; 8848 REASON_SET(&reason, PFRES_SHORT); 8849 goto done; 8850 } 8851 action = pf_test_state_udp(&s, kif, m, off, h, &pd); 8852 if (action == PF_PASS) { 8853 if (V_pfsync_update_state_ptr != NULL) 8854 V_pfsync_update_state_ptr(s); 8855 r = s->rule.ptr; 8856 a = s->anchor.ptr; 8857 } else if (s == NULL) 8858 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8859 &a, &ruleset, inp); 8860 break; 8861 } 8862 8863 case IPPROTO_SCTP: { 8864 if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp), 8865 &action, &reason, AF_INET6)) { 8866 if (action != PF_PASS) 8867 pd.act.log |= PF_LOG_FORCE; 8868 goto done; 8869 } 8870 pd.sport = &pd.hdr.sctp.src_port; 8871 pd.dport = &pd.hdr.sctp.dest_port; 8872 if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) { 8873 action = PF_DROP; 8874 REASON_SET(&reason, PFRES_SHORT); 8875 goto done; 8876 } 8877 action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd); 8878 if (action == PF_DROP) 8879 goto done; 8880 action = pf_test_state_sctp(&s, kif, m, off, h, &pd, 8881 &reason); 8882 if (action == PF_PASS) { 8883 if (V_pfsync_update_state_ptr != NULL) 8884 V_pfsync_update_state_ptr(s); 8885 r = s->rule.ptr; 8886 a = s->anchor.ptr; 8887 } else { 8888 action = pf_test_rule(&r, &s, kif, m, off, 8889 &pd, &a, &ruleset, inp); 8890 } 8891 break; 8892 } 8893 8894 case IPPROTO_ICMP: { 8895 action = PF_DROP; 8896 DPFPRINTF(PF_DEBUG_MISC, 8897 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 8898 goto done; 8899 } 8900 8901 case IPPROTO_ICMPV6: { 8902 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 8903 &action, &reason, AF_INET6)) { 8904 if (action != PF_PASS) 8905 pd.act.log |= PF_LOG_FORCE; 8906 goto done; 8907 } 8908 action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason); 8909 if (action == PF_PASS) { 8910 if (V_pfsync_update_state_ptr != NULL) 8911 V_pfsync_update_state_ptr(s); 8912 r = s->rule.ptr; 8913 a = s->anchor.ptr; 8914 } else if (s == NULL) 8915 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8916 &a, &ruleset, inp); 8917 break; 8918 } 8919 8920 default: 8921 action = pf_test_state_other(&s, kif, m, &pd); 8922 if (action == PF_PASS) { 8923 if (V_pfsync_update_state_ptr != NULL) 8924 V_pfsync_update_state_ptr(s); 8925 r = s->rule.ptr; 8926 a = s->anchor.ptr; 8927 } else if (s == NULL) 8928 action = pf_test_rule(&r, &s, kif, m, off, &pd, 8929 &a, &ruleset, inp); 8930 break; 8931 } 8932 8933done: 8934 PF_RULES_RUNLOCK(); 8935 if (n != m) { 8936 m_freem(n); 8937 n = NULL; 8938 } 8939 8940 /* handle dangerous IPv6 extension headers. */ 8941 if (action == PF_PASS && rh_cnt && 8942 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 8943 action = PF_DROP; 8944 REASON_SET(&reason, PFRES_IPOPTIONS); 8945 pd.act.log = r->log; 8946 DPFPRINTF(PF_DEBUG_MISC, 8947 ("pf: dropping packet with dangerous v6 headers\n")); 8948 } 8949 8950 if (s) { 8951 uint8_t log = pd.act.log; 8952 memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions)); 8953 pd.act.log |= log; 8954 tag = s->tag; 8955 rt = s->rt; 8956 } else { 8957 tag = r->tag; 8958 rt = r->rt; 8959 } 8960 8961 if (tag > 0 && pf_tag_packet(m, &pd, tag)) { 8962 action = PF_DROP; 8963 REASON_SET(&reason, PFRES_MEMORY); 8964 } 8965 8966 pf_scrub_ip6(&m, &pd); 8967 if (pd.proto == IPPROTO_TCP && pd.act.max_mss) 8968 pf_normalize_mss(m, off, &pd); 8969 8970 if (pd.act.rtableid >= 0) 8971 M_SETFIB(m, pd.act.rtableid); 8972 8973 if (pd.act.flags & PFSTATE_SETPRIO) { 8974 if (pd.tos & IPTOS_LOWDELAY) 8975 use_2nd_queue = 1; 8976 if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) { 8977 action = PF_DROP; 8978 REASON_SET(&reason, PFRES_MEMORY); 8979 pd.act.log = PF_LOG_FORCE; 8980 DPFPRINTF(PF_DEBUG_MISC, 8981 ("pf: failed to allocate 802.1q mtag\n")); 8982 } 8983 } 8984 8985#ifdef ALTQ 8986 if (action == PF_PASS && pd.act.qid) { 8987 if (pd.pf_mtag == NULL && 8988 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 8989 action = PF_DROP; 8990 REASON_SET(&reason, PFRES_MEMORY); 8991 } else { 8992 if (s != NULL) 8993 pd.pf_mtag->qid_hash = pf_state_hash(s); 8994 if (pd.tos & IPTOS_LOWDELAY) 8995 pd.pf_mtag->qid = pd.act.pqid; 8996 else 8997 pd.pf_mtag->qid = pd.act.qid; 8998 /* Add hints for ecn. */ 8999 pd.pf_mtag->hdr = h; 9000 } 9001 } 9002#endif /* ALTQ */ 9003 9004 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 9005 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 9006 (s->nat_rule.ptr->action == PF_RDR || 9007 s->nat_rule.ptr->action == PF_BINAT) && 9008 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 9009 m->m_flags |= M_SKIP_FIREWALL; 9010 9011 /* XXX: Anybody working on it?! */ 9012 if (r->divert.port) 9013 printf("pf: divert(9) is not supported for IPv6\n"); 9014 9015 if (pd.act.log) { 9016 struct pf_krule *lr; 9017 struct pf_krule_item *ri; 9018 9019 if (s != NULL && s->nat_rule.ptr != NULL && 9020 s->nat_rule.ptr->log & PF_LOG_ALL) 9021 lr = s->nat_rule.ptr; 9022 else 9023 lr = r; 9024 9025 if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL) 9026 PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset, 9027 &pd, (s == NULL)); 9028 if (s) { 9029 SLIST_FOREACH(ri, &s->match_rules, entry) 9030 if (ri->r->log & PF_LOG_ALL) 9031 PFLOG_PACKET(kif, m, AF_INET6, action, reason, 9032 ri->r, a, ruleset, &pd, 0); 9033 } 9034 } 9035 9036 pf_counter_u64_critical_enter(); 9037 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 9038 pd.tot_len); 9039 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 9040 1); 9041 9042 if (action == PF_PASS || r->action == PF_DROP) { 9043 dirndx = (dir == PF_OUT); 9044 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 9045 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 9046 if (a != NULL) { 9047 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 9048 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 9049 } 9050 if (s != NULL) { 9051 if (s->nat_rule.ptr != NULL) { 9052 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 9053 1); 9054 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 9055 pd.tot_len); 9056 } 9057 if (s->src_node != NULL) { 9058 counter_u64_add(s->src_node->packets[dirndx], 9059 1); 9060 counter_u64_add(s->src_node->bytes[dirndx], 9061 pd.tot_len); 9062 } 9063 if (s->nat_src_node != NULL) { 9064 counter_u64_add(s->nat_src_node->packets[dirndx], 9065 1); 9066 counter_u64_add(s->nat_src_node->bytes[dirndx], 9067 pd.tot_len); 9068 } 9069 dirndx = (dir == s->direction) ? 0 : 1; 9070 s->packets[dirndx]++; 9071 s->bytes[dirndx] += pd.tot_len; 9072 } 9073 tr = r; 9074 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 9075 if (nr != NULL && r == &V_pf_default_rule) 9076 tr = nr; 9077 if (tr->src.addr.type == PF_ADDR_TABLE) 9078 pfr_update_stats(tr->src.addr.p.tbl, 9079 (s == NULL) ? pd.src : 9080 &s->key[(s->direction == PF_IN)]->addr[0], 9081 pd.af, pd.tot_len, dir == PF_OUT, 9082 r->action == PF_PASS, tr->src.neg); 9083 if (tr->dst.addr.type == PF_ADDR_TABLE) 9084 pfr_update_stats(tr->dst.addr.p.tbl, 9085 (s == NULL) ? pd.dst : 9086 &s->key[(s->direction == PF_IN)]->addr[1], 9087 pd.af, pd.tot_len, dir == PF_OUT, 9088 r->action == PF_PASS, tr->dst.neg); 9089 } 9090 pf_counter_u64_critical_exit(); 9091 9092 switch (action) { 9093 case PF_SYNPROXY_DROP: 9094 m_freem(*m0); 9095 case PF_DEFER: 9096 *m0 = NULL; 9097 action = PF_PASS; 9098 break; 9099 case PF_DROP: 9100 m_freem(*m0); 9101 *m0 = NULL; 9102 break; 9103 default: 9104 /* pf_route6() returns unlocked. */ 9105 if (rt) { 9106 pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp); 9107 goto out; 9108 } 9109 if (pf_dummynet(&pd, s, r, m0) != 0) { 9110 action = PF_DROP; 9111 REASON_SET(&reason, PFRES_MEMORY); 9112 } 9113 break; 9114 } 9115 9116 if (s && action != PF_DROP) { 9117 if (!s->if_index_in && dir == PF_IN) 9118 s->if_index_in = ifp->if_index; 9119 else if (!s->if_index_out && dir == PF_OUT) 9120 s->if_index_out = ifp->if_index; 9121 } 9122 9123 if (s) 9124 PF_STATE_UNLOCK(s); 9125 9126 /* If reassembled packet passed, create new fragments. */ 9127 if (action == PF_PASS && *m0 && dir == PF_OUT && 9128 (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL) 9129 action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD); 9130 9131out: 9132 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 9133 9134 pf_sctp_multihome_delayed(&pd, off, kif, s, action); 9135 9136 return (action); 9137} 9138#endif /* INET6 */ 9139