ip_state.c revision 89316
1/* 2 * Copyright (C) 1995-2001 by Darren Reed. 3 * 4 * See the IPFILTER.LICENCE file for details on licencing. 5 */ 6 7#include <sys/errno.h> 8#include <sys/types.h> 9#include <sys/param.h> 10#if (__FreeBSD_version >= 500028) 11#include <sys/queue.h> 12#endif 13#include <sys/file.h> 14#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \ 15 defined(_KERNEL) 16# include "opt_ipfilter_log.h" 17#endif 18#if defined(_KERNEL) && defined(__FreeBSD_version) && \ 19 (__FreeBSD_version >= 400000) && !defined(KLD_MODULE) 20#include "opt_inet6.h" 21#endif 22#if !defined(_KERNEL) && !defined(KERNEL) && !defined(__KERNEL__) 23# include <stdio.h> 24# include <stdlib.h> 25# include <string.h> 26#else 27# ifdef linux 28# include <linux/kernel.h> 29# include <linux/module.h> 30# endif 31#endif 32#if (defined(KERNEL) || defined(_KERNEL)) && (__FreeBSD_version >= 220000) 33# include <sys/filio.h> 34# include <sys/fcntl.h> 35# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM) 36# include "opt_ipfilter.h" 37# endif 38#else 39# include <sys/ioctl.h> 40#endif 41#include <sys/time.h> 42#include <sys/uio.h> 43#ifndef linux 44# include <sys/protosw.h> 45#endif 46#include <sys/socket.h> 47#if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux) 48# include <sys/systm.h> 49#endif 50#if !defined(__SVR4) && !defined(__svr4__) 51# ifndef linux 52# include <sys/mbuf.h> 53# endif 54#else 55# include <sys/filio.h> 56# include <sys/byteorder.h> 57# ifdef _KERNEL 58# include <sys/dditypes.h> 59# endif 60# include <sys/stream.h> 61# include <sys/kmem.h> 62#endif 63 64#include <net/if.h> 65#ifdef sun 66# include <net/af.h> 67#endif 68#include <net/route.h> 69#include <netinet/in.h> 70#include <netinet/in_systm.h> 71#include <netinet/ip.h> 72#include <netinet/tcp.h> 73#ifndef linux 74# include <netinet/ip_var.h> 75# include <netinet/tcp_fsm.h> 76#endif 77#include <netinet/udp.h> 78#include <netinet/ip_icmp.h> 79#include "netinet/ip_compat.h" 80#include <netinet/tcpip.h> 81#include "netinet/ip_fil.h" 82#include "netinet/ip_nat.h" 83#include "netinet/ip_frag.h" 84#include "netinet/ip_proxy.h" 85#include "netinet/ip_state.h" 86#ifdef USE_INET6 87#include <netinet/icmp6.h> 88#endif 89#if (__FreeBSD_version >= 300000) 90# include <sys/malloc.h> 91# if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM) 92# include <sys/libkern.h> 93# include <sys/systm.h> 94# endif 95#endif 96 97#if !defined(lint) 98static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed"; 99/* static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.30.2.38 2001/07/23 13:49:46 darrenr Exp $"; */ 100static const char rcsid[] = "@(#)$FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 89316 2002-01-13 21:37:49Z alfred $"; 101#endif 102 103#ifndef MIN 104# define MIN(a,b) (((a)<(b))?(a):(b)) 105#endif 106 107#define TCP_CLOSE (TH_FIN|TH_RST) 108 109static ipstate_t **ips_table = NULL; 110static ipstate_t *ips_list = NULL; 111static int ips_num = 0; 112static int ips_wild = 0; 113static ips_stat_t ips_stats; 114#if (SOLARIS || defined(__sgi)) && defined(_KERNEL) 115extern KRWLOCK_T ipf_state, ipf_mutex; 116extern kmutex_t ipf_rw; 117#endif 118 119#ifdef USE_INET6 120static frentry_t *fr_checkicmp6matchingstate __P((ip6_t *, fr_info_t *)); 121#endif 122static int fr_matchsrcdst __P((ipstate_t *, union i6addr, union i6addr, 123 fr_info_t *, tcphdr_t *)); 124static frentry_t *fr_checkicmpmatchingstate __P((ip_t *, fr_info_t *)); 125static int fr_matchicmpqueryreply __P((int, ipstate_t *, icmphdr_t *)); 126static int fr_state_flush __P((int)); 127static ips_stat_t *fr_statetstats __P((void)); 128static void fr_delstate __P((ipstate_t *)); 129static int fr_state_remove __P((caddr_t)); 130static void fr_ipsmove __P((ipstate_t **, ipstate_t *, u_int)); 131int fr_stputent __P((caddr_t)); 132int fr_stgetent __P((caddr_t)); 133void fr_stinsert __P((ipstate_t *)); 134 135 136#define FIVE_DAYS (2 * 5 * 86400) /* 5 days: half closed session */ 137 138#define TCP_MSL 240 /* 2 minutes */ 139u_long fr_tcpidletimeout = FIVE_DAYS, 140 fr_tcpclosewait = 2 * TCP_MSL, 141 fr_tcplastack = 2 * TCP_MSL, 142 fr_tcptimeout = 2 * TCP_MSL, 143 fr_tcpclosed = 120, 144 fr_tcphalfclosed = 2 * 2 * 3600, /* 2 hours */ 145 fr_udptimeout = 240, 146 fr_udpacktimeout = 24, 147 fr_icmptimeout = 120, 148 fr_icmpacktimeout = 12; 149int fr_statemax = IPSTATE_MAX, 150 fr_statesize = IPSTATE_SIZE; 151int fr_state_doflush = 0, 152 fr_state_lock = 0; 153 154static int icmpreplytype4[ICMP_MAXTYPE + 1]; 155 156int fr_stateinit() 157{ 158 int i; 159 160 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *)); 161 if (ips_table != NULL) 162 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *)); 163 else 164 return -1; 165 166 /* fill icmp reply type table */ 167 for (i = 0; i <= ICMP_MAXTYPE; i++) 168 icmpreplytype4[i] = -1; 169 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 170 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 171 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 172 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 173 174 return 0; 175} 176 177 178static ips_stat_t *fr_statetstats() 179{ 180 ips_stats.iss_active = ips_num; 181 ips_stats.iss_table = ips_table; 182 ips_stats.iss_list = ips_list; 183 return &ips_stats; 184} 185 186 187/* 188 * flush state tables. two actions currently defined: 189 * which == 0 : flush all state table entries 190 * which == 1 : flush TCP connections which have started to close but are 191 * stuck for some reason. 192 */ 193static int fr_state_flush(which) 194int which; 195{ 196 register ipstate_t *is, **isp; 197#if defined(_KERNEL) && !SOLARIS 198 int s; 199#endif 200 int delete, removed = 0; 201 202 SPL_NET(s); 203 for (isp = &ips_list; (is = *isp); ) { 204 delete = 0; 205 206 switch (which) 207 { 208 case 0 : 209 delete = 1; 210 break; 211 case 1 : 212 if (is->is_p != IPPROTO_TCP) 213 break; 214 if ((is->is_state[0] != TCPS_ESTABLISHED) || 215 (is->is_state[1] != TCPS_ESTABLISHED)) 216 delete = 1; 217 break; 218 } 219 220 if (delete) { 221 if (is->is_p == IPPROTO_TCP) 222 ips_stats.iss_fin++; 223 else 224 ips_stats.iss_expire++; 225#ifdef IPFILTER_LOG 226 ipstate_log(is, ISL_FLUSH); 227#endif 228 fr_delstate(is); 229 removed++; 230 } else 231 isp = &is->is_next; 232 } 233 SPL_X(s); 234 return removed; 235} 236 237 238static int fr_state_remove(data) 239caddr_t data; 240{ 241 ipstate_t *sp, st; 242 int error; 243 244 sp = &st; 245 error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st)); 246 if (error) 247 return EFAULT; 248 249 WRITE_ENTER(&ipf_state); 250 for (sp = ips_list; sp; sp = sp->is_next) 251 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 252 !bcmp((char *)&sp->is_src, (char *)&st.is_src, 253 sizeof(st.is_src)) && 254 !bcmp((char *)&sp->is_dst, (char *)&st.is_src, 255 sizeof(st.is_dst)) && 256 !bcmp((char *)&sp->is_ps, (char *)&st.is_ps, 257 sizeof(st.is_ps))) { 258#ifdef IPFILTER_LOG 259 ipstate_log(sp, ISL_REMOVE); 260#endif 261 fr_delstate(sp); 262 RWLOCK_EXIT(&ipf_state); 263 return 0; 264 } 265 RWLOCK_EXIT(&ipf_state); 266 return ESRCH; 267} 268 269 270int fr_state_ioctl(data, cmd, mode) 271caddr_t data; 272#if defined(__NetBSD__) || defined(__OpenBSD__) 273u_long cmd; 274#else 275int cmd; 276#endif 277int mode; 278{ 279 int arg, ret, error = 0; 280 281 switch (cmd) 282 { 283 case SIOCDELST : 284 error = fr_state_remove(data); 285 break; 286 case SIOCIPFFL : 287 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg)); 288 if (error) 289 break; 290 if (arg == 0 || arg == 1) { 291 WRITE_ENTER(&ipf_state); 292 ret = fr_state_flush(arg); 293 RWLOCK_EXIT(&ipf_state); 294 error = IWCOPY((caddr_t)&ret, data, sizeof(ret)); 295 } else 296 error = EINVAL; 297 break; 298#ifdef IPFILTER_LOG 299 case SIOCIPFFB : 300 if (!(mode & FWRITE)) 301 error = EPERM; 302 else { 303 int tmp; 304 305 tmp = ipflog_clear(IPL_LOGSTATE); 306 IWCOPY((char *)&tmp, data, sizeof(tmp)); 307 } 308 break; 309#endif 310 case SIOCGETFS : 311 error = IWCOPYPTR((caddr_t)fr_statetstats(), data, 312 sizeof(ips_stat_t)); 313 break; 314 case FIONREAD : 315#ifdef IPFILTER_LOG 316 arg = (int)iplused[IPL_LOGSTATE]; 317 error = IWCOPY((caddr_t)&arg, (caddr_t)data, sizeof(arg)); 318#endif 319 break; 320 case SIOCSTLCK : 321 error = fr_lock(data, &fr_state_lock); 322 break; 323 case SIOCSTPUT : 324 if (!fr_state_lock) { 325 error = EACCES; 326 break; 327 } 328 error = fr_stputent(data); 329 break; 330 case SIOCSTGET : 331 if (!fr_state_lock) { 332 error = EACCES; 333 break; 334 } 335 error = fr_stgetent(data); 336 break; 337 default : 338 error = EINVAL; 339 break; 340 } 341 return error; 342} 343 344 345int fr_stgetent(data) 346caddr_t data; 347{ 348 register ipstate_t *is, *isn; 349 ipstate_save_t ips, *ipsp; 350 int error; 351 352 error = IRCOPY(data, (caddr_t)&ipsp, sizeof(ipsp)); 353 if (error) 354 return EFAULT; 355 error = IRCOPY((caddr_t)ipsp, (caddr_t)&ips, sizeof(ips)); 356 if (error) 357 return EFAULT; 358 359 isn = ips.ips_next; 360 if (!isn) { 361 isn = ips_list; 362 if (isn == NULL) { 363 if (ips.ips_next == NULL) 364 return ENOENT; 365 return 0; 366 } 367 } else { 368 /* 369 * Make sure the pointer we're copying from exists in the 370 * current list of entries. Security precaution to prevent 371 * copying of random kernel data. 372 */ 373 for (is = ips_list; is; is = is->is_next) 374 if (is == isn) 375 break; 376 if (!is) 377 return ESRCH; 378 } 379 ips.ips_next = isn->is_next; 380 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 381 if (isn->is_rule) 382 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 383 sizeof(ips.ips_fr)); 384 error = IWCOPY((caddr_t)&ips, ipsp, sizeof(ips)); 385 if (error) 386 error = EFAULT; 387 return error; 388} 389 390 391int fr_stputent(data) 392caddr_t data; 393{ 394 register ipstate_t *is, *isn; 395 ipstate_save_t ips, *ipsp; 396 int error, out; 397 frentry_t *fr; 398 399 error = IRCOPY(data, (caddr_t)&ipsp, sizeof(ipsp)); 400 if (error) 401 return EFAULT; 402 error = IRCOPY((caddr_t)ipsp, (caddr_t)&ips, sizeof(ips)); 403 if (error) 404 return EFAULT; 405 406 KMALLOC(isn, ipstate_t *); 407 if (isn == NULL) 408 return ENOMEM; 409 410 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 411 fr = isn->is_rule; 412 if (fr != NULL) { 413 if (isn->is_flags & FI_NEWFR) { 414 KMALLOC(fr, frentry_t *); 415 if (fr == NULL) { 416 KFREE(isn); 417 return ENOMEM; 418 } 419 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 420 out = fr->fr_flags & FR_OUTQUE ? 1 : 0; 421 isn->is_rule = fr; 422 ips.ips_is.is_rule = fr; 423 if (*fr->fr_ifname) { 424 fr->fr_ifa = GETUNIT(fr->fr_ifname, fr->fr_v); 425 if (fr->fr_ifa == NULL) 426 fr->fr_ifa = (void *)-1; 427#ifdef _KERNEL 428 else { 429 strncpy(isn->is_ifname[out], 430 IFNAME(fr->fr_ifa), IFNAMSIZ); 431 isn->is_ifp[out] = fr->fr_ifa; 432 } 433#endif 434 } else 435 fr->fr_ifa = NULL; 436 /* 437 * send a copy back to userland of what we ended up 438 * to allow for verification. 439 */ 440 error = IWCOPY((caddr_t)&ips, ipsp, sizeof(ips)); 441 if (error) { 442 KFREE(isn); 443 KFREE(fr); 444 return EFAULT; 445 } 446 } else { 447 for (is = ips_list; is; is = is->is_next) 448 if (is->is_rule == fr) 449 break; 450 if (!is) { 451 KFREE(isn); 452 return ESRCH; 453 } 454 } 455 } 456 fr_stinsert(isn); 457 return 0; 458} 459 460 461void fr_stinsert(is) 462register ipstate_t *is; 463{ 464 register u_int hv = is->is_hv; 465 466 MUTEX_INIT(&is->is_lock, "ipf state entry", NULL); 467 468 is->is_ifname[0][sizeof(is->is_ifname[0]) - 1] = '\0'; 469 if (is->is_ifname[0][0] != '\0') { 470 is->is_ifp[0] = GETUNIT(is->is_ifname[0], is->is_v); 471 } 472 is->is_ifname[1][sizeof(is->is_ifname[0]) - 1] = '\0'; 473 if (is->is_ifname[1][0] != '\0') { 474 is->is_ifp[1] = GETUNIT(is->is_ifname[1], is->is_v); 475 } 476 477 /* 478 * add into list table. 479 */ 480 if (ips_list) 481 ips_list->is_pnext = &is->is_next; 482 is->is_pnext = &ips_list; 483 is->is_next = ips_list; 484 ips_list = is; 485 if (ips_table[hv]) 486 ips_table[hv]->is_phnext = &is->is_hnext; 487 else 488 ips_stats.iss_inuse++; 489 is->is_phnext = ips_table + hv; 490 is->is_hnext = ips_table[hv]; 491 ips_table[hv] = is; 492 ips_num++; 493} 494 495 496/* 497 * Create a new ipstate structure and hang it off the hash table. 498 */ 499ipstate_t *fr_addstate(ip, fin, flags) 500ip_t *ip; 501fr_info_t *fin; 502u_int flags; 503{ 504 register tcphdr_t *tcp = NULL; 505 register ipstate_t *is; 506 register u_int hv; 507 ipstate_t ips; 508 u_int pass; 509 int out; 510 511 if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT)) 512 return NULL; 513 if (ips_num == fr_statemax) { 514 ips_stats.iss_max++; 515 fr_state_doflush = 1; 516 return NULL; 517 } 518 out = fin->fin_out; 519 is = &ips; 520 bzero((char *)is, sizeof(*is)); 521 ips.is_age = 1; 522 ips.is_state[0] = 0; 523 ips.is_state[1] = 0; 524 /* 525 * Copy and calculate... 526 */ 527 hv = (is->is_p = fin->fin_fi.fi_p); 528 is->is_src = fin->fin_fi.fi_src; 529 hv += is->is_saddr; 530 is->is_dst = fin->fin_fi.fi_dst; 531 hv += is->is_daddr; 532#ifdef USE_INET6 533 if (fin->fin_v == 6) { 534 if (is->is_p == IPPROTO_ICMPV6) { 535 if (IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) 536 flags |= FI_W_DADDR; 537 if (out) 538 hv -= is->is_daddr; 539 else 540 hv -= is->is_saddr; 541 } 542 } 543#endif 544 545 switch (is->is_p) 546 { 547#ifdef USE_INET6 548 case IPPROTO_ICMPV6 : 549#endif 550 case IPPROTO_ICMP : 551 { 552 struct icmp *ic = (struct icmp *)fin->fin_dp; 553 554#ifdef USE_INET6 555 if ((is->is_p == IPPROTO_ICMPV6) && 556 ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)) 557 return NULL; 558#endif 559 switch (ic->icmp_type) 560 { 561#ifdef USE_INET6 562 case ICMP6_ECHO_REQUEST : 563 is->is_icmp.ics_type = ICMP6_ECHO_REPLY; 564 hv += (is->is_icmp.ics_id = ic->icmp_id); 565 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 566 break; 567 case ICMP6_MEMBERSHIP_QUERY : 568 case ND_ROUTER_SOLICIT : 569 case ND_NEIGHBOR_SOLICIT : 570 is->is_icmp.ics_type = ic->icmp_type + 1; 571 break; 572#endif 573 case ICMP_ECHO : 574 case ICMP_TSTAMP : 575 case ICMP_IREQ : 576 case ICMP_MASKREQ : 577 is->is_icmp.ics_type = ic->icmp_type; 578 hv += (is->is_icmp.ics_id = ic->icmp_id); 579 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 580 break; 581 default : 582 return NULL; 583 } 584 ATOMIC_INCL(ips_stats.iss_icmp); 585 is->is_age = fr_icmptimeout; 586 break; 587 } 588 case IPPROTO_TCP : 589 { 590 tcp = (tcphdr_t *)fin->fin_dp; 591 592 if (tcp->th_flags & TH_RST) 593 return NULL; 594 /* 595 * The endian of the ports doesn't matter, but the ack and 596 * sequence numbers do as we do mathematics on them later. 597 */ 598 is->is_dport = tcp->th_dport; 599 is->is_sport = tcp->th_sport; 600 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 601 hv += tcp->th_dport; 602 hv += tcp->th_sport; 603 } 604 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 605 (tcp->th_off << 2) + 606 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 607 ((tcp->th_flags & TH_FIN) ? 1 : 0); 608 is->is_maxsend = is->is_send; 609 is->is_dend = 0; 610 is->is_maxdwin = 1; 611 is->is_maxswin = ntohs(tcp->th_win); 612 if (is->is_maxswin == 0) 613 is->is_maxswin = 1; 614 /* 615 * If we're creating state for a starting connection, start the 616 * timer on it as we'll never see an error if it fails to 617 * connect. 618 */ 619 ATOMIC_INCL(ips_stats.iss_tcp); 620 break; 621 } 622 case IPPROTO_UDP : 623 { 624 tcp = (tcphdr_t *)fin->fin_dp; 625 626 is->is_dport = tcp->th_dport; 627 is->is_sport = tcp->th_sport; 628 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 629 hv += tcp->th_dport; 630 hv += tcp->th_sport; 631 } 632 ATOMIC_INCL(ips_stats.iss_udp); 633 is->is_age = fr_udptimeout; 634 break; 635 } 636 default : 637 return NULL; 638 } 639 640 KMALLOC(is, ipstate_t *); 641 if (is == NULL) { 642 ATOMIC_INCL(ips_stats.iss_nomem); 643 return NULL; 644 } 645 bcopy((char *)&ips, (char *)is, sizeof(*is)); 646 hv %= fr_statesize; 647 is->is_hv = hv; 648 is->is_rule = fin->fin_fr; 649 if (is->is_rule != NULL) { 650 ATOMIC_INC32(is->is_rule->fr_ref); 651 pass = is->is_rule->fr_flags; 652 } else 653 pass = fr_flags; 654 WRITE_ENTER(&ipf_state); 655 656 is->is_pass = pass; 657 is->is_pkts = 1; 658 is->is_bytes = fin->fin_dlen + fin->fin_hlen; 659 /* 660 * We want to check everything that is a property of this packet, 661 * but we don't (automatically) care about it's fragment status as 662 * this may change. 663 */ 664 is->is_v = fin->fin_fi.fi_v; 665 is->is_opt = fin->fin_fi.fi_optmsk; 666 is->is_optmsk = 0xffffffff; 667 is->is_sec = fin->fin_fi.fi_secmsk; 668 is->is_secmsk = 0xffff; 669 is->is_auth = fin->fin_fi.fi_auth; 670 is->is_authmsk = 0xffff; 671 is->is_flags = fin->fin_fl & FI_CMP; 672 is->is_flags |= FI_CMP << 4; 673 is->is_flags |= flags & (FI_WILDP|FI_WILDA); 674 if (flags & (FI_WILDP|FI_WILDA)) 675 ips_wild++; 676 is->is_ifp[1 - out] = NULL; 677 is->is_ifp[out] = fin->fin_ifp; 678#ifdef _KERNEL 679 strncpy(is->is_ifname[out], IFNAME(fin->fin_ifp), IFNAMSIZ); 680#endif 681 is->is_ifname[1 - out][0] = '\0'; 682 if (pass & FR_LOGFIRST) 683 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 684 fr_stinsert(is); 685 if (is->is_p == IPPROTO_TCP) { 686 MUTEX_ENTER(&is->is_lock); 687 fr_tcp_age(&is->is_age, is->is_state, fin, 688 0); /* 0 = packet from the source */ 689 MUTEX_EXIT(&is->is_lock); 690 } 691#ifdef IPFILTER_LOG 692 ipstate_log(is, ISL_NEW); 693#endif 694 RWLOCK_EXIT(&ipf_state); 695 fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst); 696 if ((fin->fin_fi.fi_fl & FI_FRAG) && (pass & FR_KEEPFRAG)) 697 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 698 return is; 699} 700 701 702 703/* 704 * check to see if a packet with TCP headers fits within the TCP window. 705 * change timeout depending on whether new packet is a SYN-ACK returning for a 706 * SYN or a RST or FIN which indicate time to close up shop. 707 */ 708int fr_tcpstate(is, fin, ip, tcp) 709register ipstate_t *is; 710fr_info_t *fin; 711ip_t *ip; 712tcphdr_t *tcp; 713{ 714 register tcp_seq seq, ack, end; 715 register int ackskew; 716 tcpdata_t *fdata, *tdata; 717 u_short win, maxwin; 718 int ret = 0; 719 int source; 720 721 /* 722 * Find difference between last checked packet and this packet. 723 */ 724 source = IP6EQ(fin->fin_fi.fi_src, is->is_src); 725 if (source && (ntohs(is->is_sport) != fin->fin_data[0])) 726 source = 0; 727 fdata = &is->is_tcp.ts_data[!source]; 728 tdata = &is->is_tcp.ts_data[source]; 729 seq = ntohl(tcp->th_seq); 730 ack = ntohl(tcp->th_ack); 731 win = ntohs(tcp->th_win); 732 end = seq + fin->fin_dlen - (tcp->th_off << 2) + 733 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 734 ((tcp->th_flags & TH_FIN) ? 1 : 0); 735 736 MUTEX_ENTER(&is->is_lock); 737 if (fdata->td_end == 0) { 738 /* 739 * Must be a (outgoing) SYN-ACK in reply to a SYN. 740 */ 741 fdata->td_end = end; 742 fdata->td_maxwin = 1; 743 fdata->td_maxend = end + 1; 744 } 745 746 if (!(tcp->th_flags & TH_ACK)) { /* Pretend an ack was sent */ 747 ack = tdata->td_end; 748 } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 749 (ack == 0)) { 750 /* gross hack to get around certain broken tcp stacks */ 751 ack = tdata->td_end; 752 } 753 754 if (seq == end) 755 seq = end = fdata->td_end; 756 757 maxwin = tdata->td_maxwin; 758 ackskew = tdata->td_end - ack; 759 760#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 761#define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 762 if ((SEQ_GE(fdata->td_maxend, end)) && 763 (SEQ_GE(seq, fdata->td_end - maxwin)) && 764/* XXX what about big packets */ 765#define MAXACKWINDOW 66000 766 (ackskew >= -MAXACKWINDOW) && 767 (ackskew <= MAXACKWINDOW)) { 768 /* if ackskew < 0 then this should be due to fragented 769 * packets. There is no way to know the length of the 770 * total packet in advance. 771 * We do know the total length from the fragment cache though. 772 * Note however that there might be more sessions with 773 * exactly the same source and destination paramters in the 774 * state cache (and source and destination is the only stuff 775 * that is saved in the fragment cache). Note further that 776 * some TCP connections in the state cache are hashed with 777 * sport and dport as well which makes it not worthwhile to 778 * look for them. 779 * Thus, when ackskew is negative but still seems to belong 780 * to this session, we bump up the destinations end value. 781 */ 782 if (ackskew < 0) 783 tdata->td_end = ack; 784 785 /* update max window seen */ 786 if (fdata->td_maxwin < win) 787 fdata->td_maxwin = win; 788 if (SEQ_GT(end, fdata->td_end)) 789 fdata->td_end = end; 790 if (SEQ_GE(ack + win, tdata->td_maxend)) { 791 tdata->td_maxend = ack + win; 792 if (win == 0) 793 tdata->td_maxend++; 794 } 795 796 ATOMIC_INCL(ips_stats.iss_hits); 797 /* 798 * Nearing end of connection, start timeout. 799 */ 800 /* source ? 0 : 1 -> !source */ 801 fr_tcp_age(&is->is_age, is->is_state, fin, !source); 802 ret = 1; 803 } 804 MUTEX_EXIT(&is->is_lock); 805 return ret; 806} 807 808 809static int fr_matchsrcdst(is, src, dst, fin, tcp) 810ipstate_t *is; 811union i6addr src, dst; 812fr_info_t *fin; 813tcphdr_t *tcp; 814{ 815 int ret = 0, rev, out, flags; 816 u_short sp, dp; 817 void *ifp; 818 819 rev = IP6NEQ(is->is_dst, dst); 820 ifp = fin->fin_ifp; 821 out = fin->fin_out; 822 823 if (tcp != NULL) { 824 flags = is->is_flags; 825 sp = tcp->th_sport; 826 dp = tcp->th_dport; 827 if (!rev) { 828 if (!(flags & FI_W_SPORT) && (sp != is->is_sport)) 829 rev = 1; 830 else if (!(flags & FI_W_DPORT) && (dp != is->is_dport)) 831 rev = 1; 832 } 833 } else { 834 flags = is->is_flags & FI_WILDA; 835 sp = 0; 836 dp = 0; 837 } 838 839 if (rev == 0) { 840 if (!out) { 841 if (is->is_ifpin == NULL || is->is_ifpin == ifp) 842 ret = 1; 843 } else { 844 if (is->is_ifpout == NULL || is->is_ifpout == ifp) 845 ret = 1; 846 } 847 } else { 848 if (out) { 849 if (is->is_ifpin == NULL || is->is_ifpin == ifp) 850 ret = 1; 851 } else { 852 if (is->is_ifpout == NULL || is->is_ifpout == ifp) 853 ret = 1; 854 } 855 } 856 if (ret == 0) 857 return 0; 858 ret = 0; 859 860 if (rev == 0) { 861 if ( 862 (IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) && 863 (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) && 864 (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) && 865 (dp == is->is_dport || flags & FI_W_DPORT)))) { 866 ret = 1; 867 } 868 } else { 869 if ( 870 (IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) && 871 (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) && 872 (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) && 873 (dp == is->is_sport || flags & FI_W_SPORT)))) { 874 ret = 1; 875 } 876 } 877 if (ret == 0) 878 return 0; 879 880 /* 881 * Whether or not this should be here, is questionable, but the aim 882 * is to get this out of the main line. 883 */ 884 if (tcp == NULL) 885 flags = is->is_flags & (FI_CMP|(FI_CMP<<4)); 886 887 if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) || 888 (fin->fin_fi.fi_optmsk != is->is_opt) || 889 (fin->fin_fi.fi_secmsk != is->is_sec) || 890 (fin->fin_fi.fi_auth != is->is_auth)) 891 return 0; 892 893 if ((flags & (FI_W_SPORT|FI_W_DPORT))) { 894 if ((flags & FI_W_SPORT) != 0) { 895 if (rev == 0) { 896 is->is_sport = sp; 897 is->is_send = htonl(tcp->th_seq); 898 } else { 899 is->is_sport = dp; 900 is->is_send = htonl(tcp->th_ack); 901 } 902 is->is_maxsend = is->is_send + 1; 903 } else if ((flags & FI_W_DPORT) != 0) { 904 if (rev == 0) { 905 is->is_dport = dp; 906 is->is_dend = htonl(tcp->th_ack); 907 } else { 908 is->is_dport = sp; 909 is->is_dend = htonl(tcp->th_seq); 910 } 911 is->is_maxdend = is->is_dend + 1; 912 } 913 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT); 914 ips_wild--; 915 } 916 917 ret = -1; 918 919 if (!rev) { 920 if (out) { 921 if (!is->is_ifpout) 922 ret = 1; 923 } else { 924 if (!is->is_ifpin) 925 ret = 0; 926 } 927 } else { 928 if (out) { 929 if (!is->is_ifpin) 930 ret = 0; 931 } else { 932 if (!is->is_ifpout) 933 ret = 1; 934 } 935 } 936 937 if (ret >= 0) { 938 is->is_ifp[ret] = ifp; 939#ifdef _KERNEL 940 strncpy(is->is_ifname[ret], IFNAME(fin->fin_ifp), 941 sizeof(is->is_ifname[ret])); 942#endif 943 } 944 fin->fin_rev = rev; 945 return 1; 946} 947 948static int fr_matchicmpqueryreply(v, is, icmp) 949int v; 950ipstate_t *is; 951icmphdr_t *icmp; 952{ 953 if (v == 4) { 954 /* 955 * If we matched its type on the way in, then when going out 956 * it will still be the same type. 957 */ 958 if (((icmp->icmp_type == is->is_type) || 959 (icmpreplytype4[is->is_type] == icmp->icmp_type)) && 960 (icmp->icmp_id == is->is_icmp.ics_id) && 961 (icmp->icmp_seq == is->is_icmp.ics_seq)) { 962 return 1; 963 }; 964 } 965#ifdef USE_INET6 966 else if (is->is_v == 6) { 967 if ((is->is_type == ICMP6_ECHO_REPLY) && 968 (icmp->icmp_type == ICMP6_ECHO_REQUEST) && 969 (icmp->icmp_id == is->is_icmp.ics_id) && 970 (icmp->icmp_seq == is->is_icmp.ics_seq)) { 971 return 1; 972 }; 973 } 974#endif 975 return 0; 976} 977 978static frentry_t *fr_checkicmpmatchingstate(ip, fin) 979ip_t *ip; 980fr_info_t *fin; 981{ 982 register ipstate_t *is, **isp; 983 register u_short sport, dport; 984 register u_char pr; 985 union i6addr dst, src; 986 struct icmp *ic; 987 u_short savelen; 988 icmphdr_t *icmp; 989 fr_info_t ofin; 990 int type, len; 991 tcphdr_t *tcp; 992 frentry_t *fr; 993 ip_t *oip; 994 u_int hv; 995 996 /* 997 * Does it at least have the return (basic) IP header ? 998 * Only a basic IP header (no options) should be with 999 * an ICMP error header. 1000 */ 1001 if (((ip->ip_v != 4) || (ip->ip_hl != 5)) || 1002 (fin->fin_plen < ICMPERR_MINPKTLEN)) 1003 return NULL; 1004 ic = (struct icmp *)fin->fin_dp; 1005 type = ic->icmp_type; 1006 /* 1007 * If it's not an error type, then return 1008 */ 1009 if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) && 1010 (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) && 1011 (type != ICMP_PARAMPROB)) 1012 return NULL; 1013 1014 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 1015 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((oip->ip_hl - 5) << 2)) 1016 return NULL; 1017 1018 /* 1019 * Sanity checks. 1020 */ 1021 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 1022 if ((len <= 0) || ((oip->ip_hl << 2) > len)) 1023 return NULL; 1024 1025 /* 1026 * Is the buffer big enough for all of it ? It's the size of the IP 1027 * header claimed in the encapsulated part which is of concern. It 1028 * may be too big to be in this buffer but not so big that it's 1029 * outside the ICMP packet, leading to TCP deref's causing problems. 1030 * This is possible because we don't know how big oip_hl is when we 1031 * do the pullup early in fr_check() and thus can't gaurantee it is 1032 * all here now. 1033 */ 1034#ifdef _KERNEL 1035 { 1036 mb_t *m; 1037 1038# if SOLARIS 1039 m = fin->fin_qfm; 1040 if ((char *)oip + len > (char *)m->b_wptr) 1041 return NULL; 1042# else 1043 m = *(mb_t **)fin->fin_mp; 1044 if ((char *)oip + len > (char *)ip + m->m_len) 1045 return NULL; 1046# endif 1047 } 1048#endif 1049 1050 /* 1051 * in the IPv4 case we must zero the i6addr union otherwise 1052 * the IP6EQ and IP6NEQ macros produce the wrong results because 1053 * of the 'junk' in the unused part of the union 1054 */ 1055 bzero((char *)&src, sizeof(src)); 1056 bzero((char *)&dst, sizeof(dst)); 1057 1058 if (oip->ip_p == IPPROTO_ICMP) { 1059 icmp = (icmphdr_t *)((char *)oip + (oip->ip_hl << 2)); 1060 1061 /* 1062 * a ICMP error can only be generated as a result of an 1063 * ICMP query, not as the response on an ICMP error 1064 * 1065 * XXX theoretically ICMP_ECHOREP and the other reply's are 1066 * ICMP query's as well, but adding them here seems strange XXX 1067 */ 1068 if ((icmp->icmp_type != ICMP_ECHO) && 1069 (icmp->icmp_type != ICMP_TSTAMP) && 1070 (icmp->icmp_type != ICMP_IREQ) && 1071 (icmp->icmp_type != ICMP_MASKREQ)) 1072 return NULL; 1073 1074 /* 1075 * perform a lookup of the ICMP packet in the state table 1076 */ 1077 hv = (pr = oip->ip_p); 1078 src.in4 = oip->ip_src; 1079 hv += src.in4.s_addr; 1080 dst.in4 = oip->ip_dst; 1081 hv += dst.in4.s_addr; 1082 hv += icmp->icmp_id; 1083 hv += icmp->icmp_seq; 1084 hv %= fr_statesize; 1085 1086 savelen = oip->ip_len; 1087 oip->ip_len = len; 1088 ofin.fin_v = 4; 1089 fr_makefrip(oip->ip_hl << 2, oip, &ofin); 1090 oip->ip_len = savelen; 1091 ofin.fin_ifp = fin->fin_ifp; 1092 ofin.fin_out = !fin->fin_out; 1093 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1094 1095 READ_ENTER(&ipf_state); 1096 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 1097 if ((is->is_p == pr) && (is->is_v == 4) && 1098 fr_matchsrcdst(is, src, dst, &ofin, NULL) && 1099 fr_matchicmpqueryreply(is->is_v, is, icmp)) { 1100 ips_stats.iss_hits++; 1101 is->is_pkts++; 1102 is->is_bytes += ip->ip_len; 1103 fr = is->is_rule; 1104 RWLOCK_EXIT(&ipf_state); 1105 return fr; 1106 } 1107 RWLOCK_EXIT(&ipf_state); 1108 return NULL; 1109 }; 1110 1111 if ((oip->ip_p != IPPROTO_TCP) && (oip->ip_p != IPPROTO_UDP)) 1112 return NULL; 1113 1114 tcp = (tcphdr_t *)((char *)oip + (oip->ip_hl << 2)); 1115 dport = tcp->th_dport; 1116 sport = tcp->th_sport; 1117 1118 hv = (pr = oip->ip_p); 1119 src.in4 = oip->ip_src; 1120 hv += src.in4.s_addr; 1121 dst.in4 = oip->ip_dst; 1122 hv += dst.in4.s_addr; 1123 hv += dport; 1124 hv += sport; 1125 hv %= fr_statesize; 1126 /* 1127 * we make an fin entry to be able to feed it to 1128 * matchsrcdst note that not all fields are encessary 1129 * but this is the cleanest way. Note further we fill 1130 * in fin_mp such that if someone uses it we'll get 1131 * a kernel panic. fr_matchsrcdst does not use this. 1132 * 1133 * watch out here, as ip is in host order and oip in network 1134 * order. Any change we make must be undone afterwards. 1135 */ 1136 savelen = oip->ip_len; 1137 oip->ip_len = len; 1138 ofin.fin_v = 4; 1139 fr_makefrip(oip->ip_hl << 2, oip, &ofin); 1140 oip->ip_len = savelen; 1141 ofin.fin_ifp = fin->fin_ifp; 1142 ofin.fin_out = !fin->fin_out; 1143 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1144 READ_ENTER(&ipf_state); 1145 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1146 /* 1147 * Only allow this icmp though if the 1148 * encapsulated packet was allowed through the 1149 * other way around. Note that the minimal amount 1150 * of info present does not allow for checking against 1151 * tcp internals such as seq and ack numbers. 1152 */ 1153 if ((is->is_p == pr) && (is->is_v == 4) && 1154 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 1155 fr = is->is_rule; 1156 ips_stats.iss_hits++; 1157 is->is_pkts++; 1158 is->is_bytes += fin->fin_plen; 1159 /* 1160 * we deliberately do not touch the timeouts 1161 * for the accompanying state table entry. 1162 * It remains to be seen if that is correct. XXX 1163 */ 1164 RWLOCK_EXIT(&ipf_state); 1165 return fr; 1166 } 1167 } 1168 RWLOCK_EXIT(&ipf_state); 1169 return NULL; 1170} 1171 1172 1173static void fr_ipsmove(isp, is, hv) 1174ipstate_t **isp, *is; 1175u_int hv; 1176{ 1177 u_int hvm; 1178 1179 hvm = is->is_hv; 1180 /* 1181 * Remove the hash from the old location... 1182 */ 1183 if (is->is_hnext) 1184 is->is_hnext->is_phnext = isp; 1185 *isp = is->is_hnext; 1186 if (ips_table[hvm] == NULL) 1187 ips_stats.iss_inuse--; 1188 1189 /* 1190 * ...and put the hash in the new one. 1191 */ 1192 hvm = hv % fr_statesize; 1193 is->is_hv = hvm; 1194 isp = &ips_table[hvm]; 1195 if (*isp) 1196 (*isp)->is_phnext = &is->is_hnext; 1197 else 1198 ips_stats.iss_inuse++; 1199 is->is_phnext = isp; 1200 is->is_hnext = *isp; 1201 *isp = is; 1202} 1203 1204 1205/* 1206 * Check if a packet has a registered state. 1207 */ 1208frentry_t *fr_checkstate(ip, fin) 1209ip_t *ip; 1210fr_info_t *fin; 1211{ 1212 union i6addr dst, src; 1213 register ipstate_t *is, **isp; 1214 register u_char pr; 1215 u_int hv, hvm, hlen, tryagain, pass, v; 1216 struct icmp *ic; 1217 frentry_t *fr; 1218 tcphdr_t *tcp; 1219 1220 if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT)) 1221 return NULL; 1222 1223 is = NULL; 1224 hlen = fin->fin_hlen; 1225 tcp = (tcphdr_t *)((char *)ip + hlen); 1226 ic = (struct icmp *)tcp; 1227 hv = (pr = fin->fin_fi.fi_p); 1228 src = fin->fin_fi.fi_src; 1229 dst = fin->fin_fi.fi_dst; 1230 hv += src.in4.s_addr; 1231 hv += dst.in4.s_addr; 1232 1233 /* 1234 * Search the hash table for matching packet header info. 1235 */ 1236 v = fin->fin_fi.fi_v; 1237 switch (fin->fin_fi.fi_p) 1238 { 1239#ifdef USE_INET6 1240 case IPPROTO_ICMPV6 : 1241 if (v == 6) { 1242 if (fin->fin_out) 1243 hv -= dst.in4.s_addr; 1244 else 1245 hv -= src.in4.s_addr; 1246 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 1247 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 1248 hv += ic->icmp_id; 1249 hv += ic->icmp_seq; 1250 } 1251 } 1252#endif 1253 case IPPROTO_ICMP : 1254 if (v == 4) { 1255 hv += ic->icmp_id; 1256 hv += ic->icmp_seq; 1257 } 1258 hv %= fr_statesize; 1259 READ_ENTER(&ipf_state); 1260 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1261 if ((is->is_p == pr) && (is->is_v == v) && 1262 fr_matchsrcdst(is, src, dst, fin, NULL) && 1263 fr_matchicmpqueryreply(v, is, ic)) { 1264 if (fin->fin_rev) 1265 is->is_age = fr_icmpacktimeout; 1266 else 1267 is->is_age = fr_icmptimeout; 1268 break; 1269 } 1270 } 1271 if (is != NULL) 1272 break; 1273 RWLOCK_EXIT(&ipf_state); 1274 /* 1275 * No matching icmp state entry. Perhaps this is a 1276 * response to another state entry. 1277 */ 1278#ifdef USE_INET6 1279 if (v == 6) 1280 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin); 1281 else 1282#endif 1283 fr = fr_checkicmpmatchingstate(ip, fin); 1284 if (fr) 1285 return fr; 1286 break; 1287 case IPPROTO_TCP : 1288 { 1289 register u_short dport, sport; 1290 register int i; 1291 1292 i = tcp->th_flags; 1293 /* 1294 * Just plain ignore RST flag set with either FIN or SYN. 1295 */ 1296 if ((i & TH_RST) && 1297 ((i & (TH_FIN|TH_SYN|TH_RST)) != TH_RST)) 1298 break; 1299 case IPPROTO_UDP : 1300 dport = tcp->th_dport; 1301 sport = tcp->th_sport; 1302 tryagain = 0; 1303 hv += dport; 1304 hv += sport; 1305 READ_ENTER(&ipf_state); 1306retry_tcpudp: 1307 hvm = hv % fr_statesize; 1308 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext) 1309 if ((is->is_p == pr) && (is->is_v == v) && 1310 fr_matchsrcdst(is, src, dst, fin, tcp)) { 1311 if ((pr == IPPROTO_TCP)) { 1312 if (!fr_tcpstate(is, fin, ip, tcp)) { 1313 continue; 1314 } 1315 } else if ((pr == IPPROTO_UDP)) { 1316 if (fin->fin_rev) 1317 is->is_age = fr_udpacktimeout; 1318 else 1319 is->is_age = fr_udptimeout; 1320 } 1321 break; 1322 } 1323 if (is != NULL) { 1324 if (tryagain && 1325 !(is->is_flags & (FI_WILDP|FI_WILDA))) { 1326 hv += dport; 1327 hv += sport; 1328 fr_ipsmove(isp, is, hv); 1329 MUTEX_DOWNGRADE(&ipf_state); 1330 } 1331 break; 1332 } 1333 RWLOCK_EXIT(&ipf_state); 1334 if (!tryagain && ips_wild) { 1335 hv -= dport; 1336 hv -= sport; 1337 tryagain = 1; 1338 WRITE_ENTER(&ipf_state); 1339 goto retry_tcpudp; 1340 } 1341 break; 1342 } 1343 default : 1344 break; 1345 } 1346 if (is == NULL) { 1347 ATOMIC_INCL(ips_stats.iss_miss); 1348 return NULL; 1349 } 1350 MUTEX_ENTER(&is->is_lock); 1351 is->is_bytes += fin->fin_plen; 1352 ips_stats.iss_hits++; 1353 is->is_pkts++; 1354 MUTEX_EXIT(&is->is_lock); 1355 fr = is->is_rule; 1356 fin->fin_fr = fr; 1357 pass = is->is_pass; 1358#ifndef _KERNEL 1359 if (tcp->th_flags & TCP_CLOSE) 1360 fr_delstate(is); 1361#endif 1362 RWLOCK_EXIT(&ipf_state); 1363 if ((fin->fin_fi.fi_fl & FI_FRAG) && (pass & FR_KEEPFRAG)) 1364 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 1365 return fr; 1366} 1367 1368 1369void ip_statesync(ifp) 1370void *ifp; 1371{ 1372 register ipstate_t *is; 1373 1374 WRITE_ENTER(&ipf_state); 1375 for (is = ips_list; is; is = is->is_next) { 1376 if (is->is_ifpin == ifp) { 1377 is->is_ifpin = GETUNIT(is->is_ifname[0], is->is_v); 1378 if (!is->is_ifpin) 1379 is->is_ifpin = (void *)-1; 1380 } 1381 if (is->is_ifpout == ifp) { 1382 is->is_ifpout = GETUNIT(is->is_ifname[1], is->is_v); 1383 if (!is->is_ifpout) 1384 is->is_ifpout = (void *)-1; 1385 } 1386 } 1387 RWLOCK_EXIT(&ipf_state); 1388} 1389 1390 1391/* 1392 * Must always be called with fr_ipfstate held as a write lock. 1393 */ 1394static void fr_delstate(is) 1395ipstate_t *is; 1396{ 1397 frentry_t *fr; 1398 1399 if (is->is_flags & (FI_WILDP|FI_WILDA)) 1400 ips_wild--; 1401 if (is->is_next) 1402 is->is_next->is_pnext = is->is_pnext; 1403 *is->is_pnext = is->is_next; 1404 if (is->is_hnext) 1405 is->is_hnext->is_phnext = is->is_phnext; 1406 *is->is_phnext = is->is_hnext; 1407 if (ips_table[is->is_hv] == NULL) 1408 ips_stats.iss_inuse--; 1409 1410 fr = is->is_rule; 1411 if (fr != NULL) { 1412 fr->fr_ref--; 1413 if (fr->fr_ref == 0) { 1414 KFREE(fr); 1415 } 1416 } 1417#ifdef _KERNEL 1418 MUTEX_DESTROY(&is->is_lock); 1419#endif 1420 KFREE(is); 1421 ips_num--; 1422} 1423 1424 1425/* 1426 * Free memory in use by all state info. kept. 1427 */ 1428void fr_stateunload() 1429{ 1430 register ipstate_t *is; 1431 1432 WRITE_ENTER(&ipf_state); 1433 while ((is = ips_list)) 1434 fr_delstate(is); 1435 ips_stats.iss_inuse = 0; 1436 ips_num = 0; 1437 RWLOCK_EXIT(&ipf_state); 1438 if (ips_table) 1439 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *)); 1440 ips_table = NULL; 1441} 1442 1443 1444/* 1445 * Slowly expire held state for thingslike UDP and ICMP. Timeouts are set 1446 * in expectation of this being called twice per second. 1447 */ 1448void fr_timeoutstate() 1449{ 1450 register ipstate_t *is, **isp; 1451#if defined(_KERNEL) && !SOLARIS 1452 int s; 1453#endif 1454 1455 SPL_NET(s); 1456 WRITE_ENTER(&ipf_state); 1457 for (isp = &ips_list; (is = *isp); ) 1458 if (is->is_age && !--is->is_age) { 1459 if (is->is_p == IPPROTO_TCP) 1460 ips_stats.iss_fin++; 1461 else 1462 ips_stats.iss_expire++; 1463#ifdef IPFILTER_LOG 1464 ipstate_log(is, ISL_EXPIRE); 1465#endif 1466 fr_delstate(is); 1467 } else 1468 isp = &is->is_next; 1469 if (fr_state_doflush) { 1470 (void) fr_state_flush(1); 1471 fr_state_doflush = 0; 1472 } 1473 RWLOCK_EXIT(&ipf_state); 1474 SPL_X(s); 1475} 1476 1477 1478/* 1479 * Original idea freom Pradeep Krishnan for use primarily with NAT code. 1480 * (pkrishna@netcom.com) 1481 * 1482 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: 1483 * 1484 * - (try to) base state transitions on real evidence only, 1485 * i.e. packets that are sent and have been received by ipfilter; 1486 * diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. 1487 * 1488 * - deal with half-closed connections correctly; 1489 * 1490 * - store the state of the source in state[0] such that ipfstat 1491 * displays the state as source/dest instead of dest/source; the calls 1492 * to fr_tcp_age have been changed accordingly. 1493 * 1494 * Parameters: 1495 * 1496 * state[0] = state of source (host that initiated connection) 1497 * state[1] = state of dest (host that accepted the connection) 1498 * 1499 * dir == 0 : a packet from source to dest 1500 * dir == 1 : a packet from dest to source 1501 * 1502 */ 1503void fr_tcp_age(age, state, fin, dir) 1504u_long *age; 1505u_char *state; 1506fr_info_t *fin; 1507int dir; 1508{ 1509 tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp; 1510 u_char flags = tcp->th_flags; 1511 int dlen, ostate; 1512 1513 ostate = state[1 - dir]; 1514 1515 dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2); 1516 1517 if (flags & TH_RST) { 1518 if (!(tcp->th_flags & TH_PUSH) && !dlen) { 1519 *age = fr_tcpclosed; 1520 state[dir] = TCPS_CLOSED; 1521 } else { 1522 *age = fr_tcpclosewait; 1523 state[dir] = TCPS_CLOSE_WAIT; 1524 } 1525 return; 1526 } 1527 1528 *age = fr_tcptimeout; /* default 4 mins */ 1529 1530 switch(state[dir]) 1531 { 1532 case TCPS_CLOSED: /* 0 */ 1533 if ((flags & TH_OPENING) == TH_OPENING) { 1534 /* 1535 * 'dir' received an S and sends SA in response, 1536 * CLOSED -> SYN_RECEIVED 1537 */ 1538 state[dir] = TCPS_SYN_RECEIVED; 1539 *age = fr_tcptimeout; 1540 } else if ((flags & (TH_SYN|TH_ACK)) == TH_SYN) { 1541 /* 'dir' sent S, CLOSED -> SYN_SENT */ 1542 state[dir] = TCPS_SYN_SENT; 1543 *age = fr_tcptimeout; 1544 } 1545 /* 1546 * The next piece of code makes it possible to get 1547 * already established connections into the state table 1548 * after a restart or reload of the filter rules; this 1549 * does not work when a strict 'flags S keep state' is 1550 * used for tcp connections of course 1551 */ 1552 if ((flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) { 1553 /* we saw an A, guess 'dir' is in ESTABLISHED mode */ 1554 state[dir] = TCPS_ESTABLISHED; 1555 *age = fr_tcpidletimeout; 1556 } 1557 /* 1558 * TODO: besides regular ACK packets we can have other 1559 * packets as well; it is yet to be determined how we 1560 * should initialize the states in those cases 1561 */ 1562 break; 1563 1564 case TCPS_LISTEN: /* 1 */ 1565 /* NOT USED */ 1566 break; 1567 1568 case TCPS_SYN_SENT: /* 2 */ 1569 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1570 /* 1571 * We see an A from 'dir' which is in SYN_SENT 1572 * state: 'dir' sent an A in response to an SA 1573 * which it received, SYN_SENT -> ESTABLISHED 1574 */ 1575 state[dir] = TCPS_ESTABLISHED; 1576 *age = fr_tcpidletimeout; 1577 } else if (flags & TH_FIN) { 1578 /* 1579 * We see an F from 'dir' which is in SYN_SENT 1580 * state and wants to close its side of the 1581 * connection; SYN_SENT -> FIN_WAIT_1 1582 */ 1583 state[dir] = TCPS_FIN_WAIT_1; 1584 *age = fr_tcpidletimeout; /* or fr_tcptimeout? */ 1585 } else if ((flags & TH_OPENING) == TH_OPENING) { 1586 /* 1587 * We see an SA from 'dir' which is already in 1588 * SYN_SENT state, this means we have a 1589 * simultaneous open; SYN_SENT -> SYN_RECEIVED 1590 */ 1591 state[dir] = TCPS_SYN_RECEIVED; 1592 *age = fr_tcptimeout; 1593 } 1594 break; 1595 1596 case TCPS_SYN_RECEIVED: /* 3 */ 1597 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1598 /* 1599 * We see an A from 'dir' which was in SYN_RECEIVED 1600 * state so it must now be in established state, 1601 * SYN_RECEIVED -> ESTABLISHED 1602 */ 1603 state[dir] = TCPS_ESTABLISHED; 1604 *age = fr_tcpidletimeout; 1605 } else if (flags & TH_FIN) { 1606 /* 1607 * We see an F from 'dir' which is in SYN_RECEIVED 1608 * state and wants to close its side of the connection; 1609 * SYN_RECEIVED -> FIN_WAIT_1 1610 */ 1611 state[dir] = TCPS_FIN_WAIT_1; 1612 *age = fr_tcpidletimeout; 1613 } 1614 break; 1615 1616 case TCPS_ESTABLISHED: /* 4 */ 1617 if (flags & TH_FIN) { 1618 /* 1619 * 'dir' closed its side of the connection; this 1620 * gives us a half-closed connection; 1621 * ESTABLISHED -> FIN_WAIT_1 1622 */ 1623 state[dir] = TCPS_FIN_WAIT_1; 1624 *age = fr_tcphalfclosed; 1625 } else if (flags & TH_ACK) { 1626 /* an ACK, should we exclude other flags here? */ 1627 if (ostate == TCPS_FIN_WAIT_1) { 1628 /* 1629 * We know the other side did an active close, 1630 * so we are ACKing the recvd FIN packet (does 1631 * the window matching code guarantee this?) 1632 * and go into CLOSE_WAIT state; this gives us 1633 * a half-closed connection 1634 */ 1635 state[dir] = TCPS_CLOSE_WAIT; 1636 *age = fr_tcphalfclosed; 1637 } else if (ostate < TCPS_CLOSE_WAIT) 1638 /* 1639 * Still a fully established connection, 1640 * reset timeout 1641 */ 1642 *age = fr_tcpidletimeout; 1643 } 1644 break; 1645 1646 case TCPS_CLOSE_WAIT: /* 5 */ 1647 if (flags & TH_FIN) { 1648 /* 1649 * Application closed and 'dir' sent a FIN, we're now 1650 * going into LAST_ACK state 1651 */ 1652 *age = fr_tcplastack; 1653 state[dir] = TCPS_LAST_ACK; 1654 } else { 1655 /* 1656 * We remain in CLOSE_WAIT because the other side has 1657 * closed already and we did not close our side yet; 1658 * reset timeout 1659 */ 1660 *age = fr_tcphalfclosed; 1661 } 1662 break; 1663 1664 case TCPS_FIN_WAIT_1: /* 6 */ 1665 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) { 1666 /* 1667 * If the other side is not active anymore it has sent 1668 * us a FIN packet that we are ack'ing now with an ACK; 1669 * this means both sides have now closed the connection 1670 * and we go into TIME_WAIT 1671 */ 1672 /* 1673 * XXX: how do we know we really are ACKing the FIN 1674 * packet here? does the window code guarantee that? 1675 */ 1676 state[dir] = TCPS_TIME_WAIT; 1677 *age = fr_tcptimeout; 1678 } else 1679 /* 1680 * We closed our side of the connection already but the 1681 * other side is still active (ESTABLISHED/CLOSE_WAIT); 1682 * continue with this half-closed connection 1683 */ 1684 *age = fr_tcphalfclosed; 1685 break; 1686 1687 case TCPS_CLOSING: /* 7 */ 1688 /* NOT USED */ 1689 break; 1690 1691 case TCPS_LAST_ACK: /* 8 */ 1692 if (flags & TH_ACK) { 1693 if ((flags & TH_PUSH) || dlen) 1694 /* 1695 * There is still data to be delivered, reset 1696 * timeout 1697 */ 1698 *age = fr_tcplastack; 1699 } 1700 /* 1701 * We cannot detect when we go out of LAST_ACK state to CLOSED 1702 * because that is based on the reception of ACK packets; 1703 * ipfilter can only detect that a packet has been sent by a 1704 * host 1705 */ 1706 break; 1707 1708 case TCPS_FIN_WAIT_2: /* 9 */ 1709 /* NOT USED */ 1710 break; 1711 1712 case TCPS_TIME_WAIT: /* 10 */ 1713 /* we're in 2MSL timeout now */ 1714 break; 1715 } 1716} 1717 1718 1719#ifdef IPFILTER_LOG 1720void ipstate_log(is, type) 1721struct ipstate *is; 1722u_int type; 1723{ 1724 struct ipslog ipsl; 1725 void *items[1]; 1726 size_t sizes[1]; 1727 int types[1]; 1728 1729 ipsl.isl_type = type; 1730 ipsl.isl_pkts = is->is_pkts; 1731 ipsl.isl_bytes = is->is_bytes; 1732 ipsl.isl_src = is->is_src; 1733 ipsl.isl_dst = is->is_dst; 1734 ipsl.isl_p = is->is_p; 1735 ipsl.isl_v = is->is_v; 1736 ipsl.isl_flags = is->is_flags; 1737 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 1738 ipsl.isl_sport = is->is_sport; 1739 ipsl.isl_dport = is->is_dport; 1740 if (ipsl.isl_p == IPPROTO_TCP) { 1741 ipsl.isl_state[0] = is->is_state[0]; 1742 ipsl.isl_state[1] = is->is_state[1]; 1743 } 1744 } else if (ipsl.isl_p == IPPROTO_ICMP) 1745 ipsl.isl_itype = is->is_icmp.ics_type; 1746 else { 1747 ipsl.isl_ps.isl_filler[0] = 0; 1748 ipsl.isl_ps.isl_filler[1] = 0; 1749 } 1750 items[0] = &ipsl; 1751 sizes[0] = sizeof(ipsl); 1752 types[0] = 0; 1753 1754 (void) ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1); 1755} 1756#endif 1757 1758 1759#ifdef USE_INET6 1760frentry_t *fr_checkicmp6matchingstate(ip, fin) 1761ip6_t *ip; 1762fr_info_t *fin; 1763{ 1764 register ipstate_t *is, **isp; 1765 register u_short sport, dport; 1766 register u_char pr; 1767 struct icmp6_hdr *ic, *oic; 1768 union i6addr dst, src; 1769 u_short savelen; 1770 fr_info_t ofin; 1771 tcphdr_t *tcp; 1772 frentry_t *fr; 1773 ip6_t *oip; 1774 int type; 1775 u_int hv; 1776 1777 /* 1778 * Does it at least have the return (basic) IP header ? 1779 * Only a basic IP header (no options) should be with 1780 * an ICMP error header. 1781 */ 1782 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN)) 1783 return NULL; 1784 ic = (struct icmp6_hdr *)fin->fin_dp; 1785 type = ic->icmp6_type; 1786 /* 1787 * If it's not an error type, then return 1788 */ 1789 if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) && 1790 (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB)) 1791 return NULL; 1792 1793 oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN); 1794 if (fin->fin_plen < sizeof(*oip)) 1795 return NULL; 1796 1797 if (oip->ip6_nxt == IPPROTO_ICMPV6) { 1798 oic = (struct icmp6_hdr *)(oip + 1); 1799 /* 1800 * a ICMP error can only be generated as a result of an 1801 * ICMP query, not as the response on an ICMP error 1802 * 1803 * XXX theoretically ICMP_ECHOREP and the other reply's are 1804 * ICMP query's as well, but adding them here seems strange XXX 1805 */ 1806 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 1807 return NULL; 1808 1809 /* 1810 * perform a lookup of the ICMP packet in the state table 1811 */ 1812 hv = (pr = oip->ip6_nxt); 1813 src.in6 = oip->ip6_src; 1814 hv += src.in4.s_addr; 1815 dst.in6 = oip->ip6_dst; 1816 hv += dst.in4.s_addr; 1817 hv += oic->icmp6_id; 1818 hv += oic->icmp6_seq; 1819 hv %= fr_statesize; 1820 1821 oip->ip6_plen = ntohs(oip->ip6_plen); 1822 ofin.fin_v = 6; 1823 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 1824 oip->ip6_plen = htons(oip->ip6_plen); 1825 ofin.fin_ifp = fin->fin_ifp; 1826 ofin.fin_out = !fin->fin_out; 1827 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1828 1829 READ_ENTER(&ipf_state); 1830 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 1831 if ((is->is_p == pr) && 1832 (oic->icmp6_id == is->is_icmp.ics_id) && 1833 (oic->icmp6_seq == is->is_icmp.ics_seq) && 1834 fr_matchsrcdst(is, src, dst, &ofin, NULL)) { 1835 /* 1836 * in the state table ICMP query's are stored 1837 * with the type of the corresponding ICMP 1838 * response. Correct here 1839 */ 1840 if (((is->is_type == ICMP6_ECHO_REPLY) && 1841 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 1842 (is->is_type - 1 == oic->icmp6_type )) { 1843 ips_stats.iss_hits++; 1844 is->is_pkts++; 1845 is->is_bytes += fin->fin_plen; 1846 return is->is_rule; 1847 } 1848 } 1849 RWLOCK_EXIT(&ipf_state); 1850 1851 return NULL; 1852 }; 1853 1854 if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP)) 1855 return NULL; 1856 tcp = (tcphdr_t *)(oip + 1); 1857 dport = tcp->th_dport; 1858 sport = tcp->th_sport; 1859 1860 hv = (pr = oip->ip6_nxt); 1861 src.in6 = oip->ip6_src; 1862 hv += src.in4.s_addr; 1863 dst.in6 = oip->ip6_dst; 1864 hv += dst.in4.s_addr; 1865 hv += dport; 1866 hv += sport; 1867 hv %= fr_statesize; 1868 /* 1869 * we make an fin entry to be able to feed it to 1870 * matchsrcdst note that not all fields are encessary 1871 * but this is the cleanest way. Note further we fill 1872 * in fin_mp such that if someone uses it we'll get 1873 * a kernel panic. fr_matchsrcdst does not use this. 1874 * 1875 * watch out here, as ip is in host order and oip in network 1876 * order. Any change we make must be undone afterwards. 1877 */ 1878 savelen = oip->ip6_plen; 1879 oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN; 1880 ofin.fin_v = 6; 1881 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 1882 oip->ip6_plen = savelen; 1883 ofin.fin_ifp = fin->fin_ifp; 1884 ofin.fin_out = !fin->fin_out; 1885 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1886 READ_ENTER(&ipf_state); 1887 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1888 /* 1889 * Only allow this icmp though if the 1890 * encapsulated packet was allowed through the 1891 * other way around. Note that the minimal amount 1892 * of info present does not allow for checking against 1893 * tcp internals such as seq and ack numbers. 1894 */ 1895 if ((is->is_p == pr) && (is->is_v == 6) && 1896 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 1897 fr = is->is_rule; 1898 ips_stats.iss_hits++; 1899 /* 1900 * we must swap src and dst here because the icmp 1901 * comes the other way around 1902 */ 1903 is->is_pkts++; 1904 is->is_bytes += fin->fin_plen; 1905 /* 1906 * we deliberately do not touch the timeouts 1907 * for the accompanying state table entry. 1908 * It remains to be seen if that is correct. XXX 1909 */ 1910 RWLOCK_EXIT(&ipf_state); 1911 return fr; 1912 } 1913 } 1914 RWLOCK_EXIT(&ipf_state); 1915 return NULL; 1916} 1917#endif 1918