ip_state.c revision 67853
1/* 2 * Copyright (C) 1995-2000 by Darren Reed. 3 * 4 * Redistribution and use in source and binary forms are permitted 5 * provided that this notice is preserved and due credit is given 6 * to the original author and the contributors. 7 */ 8#if !defined(lint) 9static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-1995 Darren Reed"; 10static const char rcsid[] = "@(#)$FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 67853 2000-10-29 07:53:05Z darrenr $"; 11#endif 12 13#include <sys/errno.h> 14#include <sys/types.h> 15#include <sys/param.h> 16#include <sys/file.h> 17#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \ 18 defined(_KERNEL) 19# include "opt_ipfilter_log.h" 20#endif 21#if defined(_KERNEL) && defined(__FreeBSD_version) && \ 22 (__FreeBSD_version >= 400000) && !defined(KLD_MODULE) 23#include "opt_inet6.h" 24#endif 25#if !defined(_KERNEL) && !defined(KERNEL) && !defined(__KERNEL__) 26# include <stdio.h> 27# include <stdlib.h> 28# include <string.h> 29#else 30# ifdef linux 31# include <linux/kernel.h> 32# include <linux/module.h> 33# endif 34#endif 35#if (defined(KERNEL) || defined(_KERNEL)) && (__FreeBSD_version >= 220000) 36# include <sys/filio.h> 37# include <sys/fcntl.h> 38# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM) 39# include "opt_ipfilter.h" 40# endif 41#else 42# include <sys/ioctl.h> 43#endif 44#include <sys/time.h> 45#include <sys/uio.h> 46#ifndef linux 47# include <sys/protosw.h> 48#endif 49#include <sys/socket.h> 50#if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux) 51# include <sys/systm.h> 52#endif 53#if !defined(__SVR4) && !defined(__svr4__) 54# ifndef linux 55# include <sys/mbuf.h> 56# endif 57#else 58# include <sys/filio.h> 59# include <sys/byteorder.h> 60# ifdef _KERNEL 61# include <sys/dditypes.h> 62# endif 63# include <sys/stream.h> 64# include <sys/kmem.h> 65#endif 66 67#include <net/if.h> 68#ifdef sun 69# include <net/af.h> 70#endif 71#include <net/route.h> 72#include <netinet/in.h> 73#include <netinet/in_systm.h> 74#include <netinet/ip.h> 75#include <netinet/tcp.h> 76#ifndef linux 77# include <netinet/ip_var.h> 78# include <netinet/tcp_fsm.h> 79#endif 80#include <netinet/udp.h> 81#include <netinet/ip_icmp.h> 82#include "netinet/ip_compat.h" 83#include <netinet/tcpip.h> 84#include "netinet/ip_fil.h" 85#include "netinet/ip_nat.h" 86#include "netinet/ip_frag.h" 87#include "netinet/ip_proxy.h" 88#include "netinet/ip_state.h" 89#ifdef USE_INET6 90#include <netinet/icmp6.h> 91#endif 92#if (__FreeBSD_version >= 300000) 93# include <sys/malloc.h> 94# if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM) 95# include <sys/libkern.h> 96# include <sys/systm.h> 97# endif 98#endif 99 100#ifndef MIN 101# define MIN(a,b) (((a)<(b))?(a):(b)) 102#endif 103 104#define TCP_CLOSE (TH_FIN|TH_RST) 105 106static ipstate_t **ips_table = NULL; 107static ipstate_t *ips_list = NULL; 108static int ips_num = 0; 109static int ips_wild = 0; 110static ips_stat_t ips_stats; 111#if (SOLARIS || defined(__sgi)) && defined(_KERNEL) 112extern KRWLOCK_T ipf_state, ipf_mutex; 113extern kmutex_t ipf_rw; 114#endif 115 116#ifdef USE_INET6 117static frentry_t *fr_checkicmp6matchingstate __P((ip6_t *, fr_info_t *)); 118#endif 119static int fr_matchsrcdst __P((ipstate_t *, union i6addr, union i6addr, 120 fr_info_t *, tcphdr_t *)); 121static frentry_t *fr_checkicmpmatchingstate __P((ip_t *, fr_info_t *)); 122static int fr_matchicmpqueryreply __P((int, ipstate_t *, icmphdr_t *)); 123static int fr_state_flush __P((int)); 124static ips_stat_t *fr_statetstats __P((void)); 125static void fr_delstate __P((ipstate_t *)); 126static int fr_state_remove __P((caddr_t)); 127static void fr_ipsmove __P((ipstate_t **, ipstate_t *, u_int)); 128int fr_stputent __P((caddr_t)); 129int fr_stgetent __P((caddr_t)); 130void fr_stinsert __P((ipstate_t *)); 131 132 133#define FIVE_DAYS (2 * 5 * 86400) /* 5 days: half closed session */ 134 135#define TCP_MSL 240 /* 2 minutes */ 136u_long fr_tcpidletimeout = FIVE_DAYS, 137 fr_tcpclosewait = 2 * TCP_MSL, 138 fr_tcplastack = 2 * TCP_MSL, 139 fr_tcptimeout = 2 * TCP_MSL, 140 fr_tcpclosed = 120, 141 fr_tcphalfclosed = 2 * 2 * 3600, /* 2 hours */ 142 fr_udptimeout = 240, 143 fr_icmptimeout = 120; 144int fr_statemax = IPSTATE_MAX, 145 fr_statesize = IPSTATE_SIZE; 146int fr_state_doflush = 0, 147 fr_state_lock = 0; 148 149static int icmpreplytype4[ICMP_MAXTYPE + 1]; 150 151int fr_stateinit() 152{ 153 int i; 154 155 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *)); 156 if (ips_table != NULL) 157 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *)); 158 else 159 return -1; 160 161 /* fill icmp reply type table */ 162 for (i = 0; i <= ICMP_MAXTYPE; i++) 163 icmpreplytype4[i] = -1; 164 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 165 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 166 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 167 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 168 169 return 0; 170} 171 172 173static ips_stat_t *fr_statetstats() 174{ 175 ips_stats.iss_active = ips_num; 176 ips_stats.iss_table = ips_table; 177 ips_stats.iss_list = ips_list; 178 return &ips_stats; 179} 180 181 182/* 183 * flush state tables. two actions currently defined: 184 * which == 0 : flush all state table entries 185 * which == 1 : flush TCP connections which have started to close but are 186 * stuck for some reason. 187 */ 188static int fr_state_flush(which) 189int which; 190{ 191 register ipstate_t *is, **isp; 192#if defined(_KERNEL) && !SOLARIS 193 int s; 194#endif 195 int delete, removed = 0; 196 197 SPL_NET(s); 198 for (isp = &ips_list; (is = *isp); ) { 199 delete = 0; 200 201 switch (which) 202 { 203 case 0 : 204 delete = 1; 205 break; 206 case 1 : 207 if (is->is_p != IPPROTO_TCP) 208 break; 209 if ((is->is_state[0] != TCPS_ESTABLISHED) || 210 (is->is_state[1] != TCPS_ESTABLISHED)) 211 delete = 1; 212 break; 213 } 214 215 if (delete) { 216 if (is->is_p == IPPROTO_TCP) 217 ips_stats.iss_fin++; 218 else 219 ips_stats.iss_expire++; 220#ifdef IPFILTER_LOG 221 ipstate_log(is, ISL_FLUSH); 222#endif 223 fr_delstate(is); 224 removed++; 225 } else 226 isp = &is->is_next; 227 } 228 SPL_X(s); 229 return removed; 230} 231 232 233static int fr_state_remove(data) 234caddr_t data; 235{ 236 ipstate_t *sp, st; 237 int error; 238 239 sp = &st; 240 error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st)); 241 if (error) 242 return EFAULT; 243 244 for (sp = ips_list; sp; sp = sp->is_next) 245 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 246 !bcmp((char *)&sp->is_src, (char *)&st.is_src, 247 sizeof(st.is_src)) && 248 !bcmp((char *)&sp->is_dst, (char *)&st.is_src, 249 sizeof(st.is_dst)) && 250 !bcmp((char *)&sp->is_ps, (char *)&st.is_ps, 251 sizeof(st.is_ps))) { 252 WRITE_ENTER(&ipf_state); 253#ifdef IPFILTER_LOG 254 ipstate_log(sp, ISL_REMOVE); 255#endif 256 fr_delstate(sp); 257 RWLOCK_EXIT(&ipf_state); 258 return 0; 259 } 260 return ESRCH; 261} 262 263 264int fr_state_ioctl(data, cmd, mode) 265caddr_t data; 266#if defined(__NetBSD__) || defined(__OpenBSD__) 267u_long cmd; 268#else 269int cmd; 270#endif 271int mode; 272{ 273 int arg, ret, error = 0; 274 275 switch (cmd) 276 { 277 case SIOCDELST : 278 error = fr_state_remove(data); 279 break; 280 case SIOCIPFFL : 281 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg)); 282 if (error) 283 break; 284 if (arg == 0 || arg == 1) { 285 WRITE_ENTER(&ipf_state); 286 ret = fr_state_flush(arg); 287 RWLOCK_EXIT(&ipf_state); 288 error = IWCOPY((caddr_t)&ret, data, sizeof(ret)); 289 } else 290 error = EINVAL; 291 break; 292#ifdef IPFILTER_LOG 293 case SIOCIPFFB : 294 if (!(mode & FWRITE)) 295 error = EPERM; 296 else { 297 int tmp; 298 299 tmp = ipflog_clear(IPL_LOGSTATE); 300 IWCOPY((char *)&tmp, data, sizeof(tmp)); 301 } 302 break; 303#endif 304 case SIOCGETFS : 305 error = IWCOPYPTR((caddr_t)fr_statetstats(), data, 306 sizeof(ips_stat_t)); 307 break; 308 case FIONREAD : 309#ifdef IPFILTER_LOG 310 error = IWCOPY((caddr_t)&iplused[IPL_LOGSTATE], (caddr_t)data, 311 sizeof(iplused[IPL_LOGSTATE])); 312#endif 313 break; 314 case SIOCSTLCK : 315 error = fr_lock(data, &fr_state_lock); 316 break; 317 case SIOCSTPUT : 318 if (!fr_state_lock) { 319 error = EACCES; 320 break; 321 } 322 error = fr_stputent(data); 323 break; 324 case SIOCSTGET : 325 if (!fr_state_lock) { 326 error = EACCES; 327 break; 328 } 329 error = fr_stgetent(data); 330 break; 331 default : 332 error = EINVAL; 333 break; 334 } 335 return error; 336} 337 338 339int fr_stgetent(data) 340caddr_t data; 341{ 342 register ipstate_t *is, *isn; 343 ipstate_save_t ips, *ipsp; 344 int error; 345 346 error = IRCOPY(data, (caddr_t)&ipsp, sizeof(ipsp)); 347 if (error) 348 return EFAULT; 349 error = IRCOPY((caddr_t)ipsp, (caddr_t)&ips, sizeof(ips)); 350 if (error) 351 return EFAULT; 352 353 isn = ips.ips_next; 354 if (!isn) { 355 isn = ips_list; 356 if (isn == NULL) { 357 if (ips.ips_next == NULL) 358 return ENOENT; 359 return 0; 360 } 361 } else { 362 /* 363 * Make sure the pointer we're copying from exists in the 364 * current list of entries. Security precaution to prevent 365 * copying of random kernel data. 366 */ 367 for (is = ips_list; is; is = is->is_next) 368 if (is == isn) 369 break; 370 if (!is) 371 return ESRCH; 372 } 373 ips.ips_next = isn->is_next; 374 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 375 if (isn->is_rule) 376 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 377 sizeof(ips.ips_fr)); 378 error = IWCOPY((caddr_t)&ips, ipsp, sizeof(ips)); 379 if (error) 380 error = EFAULT; 381 return error; 382} 383 384 385int fr_stputent(data) 386caddr_t data; 387{ 388 register ipstate_t *is, *isn; 389 ipstate_save_t ips, *ipsp; 390 int error, out; 391 frentry_t *fr; 392 393 error = IRCOPY(data, (caddr_t)&ipsp, sizeof(ipsp)); 394 if (error) 395 return EFAULT; 396 error = IRCOPY((caddr_t)ipsp, (caddr_t)&ips, sizeof(ips)); 397 if (error) 398 return EFAULT; 399 400 KMALLOC(isn, ipstate_t *); 401 if (isn == NULL) 402 return ENOMEM; 403 404 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 405 fr = isn->is_rule; 406 if (fr != NULL) { 407 if (isn->is_flags & FI_NEWFR) { 408 KMALLOC(fr, frentry_t *); 409 if (fr == NULL) { 410 KFREE(isn); 411 return ENOMEM; 412 } 413 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 414 out = fr->fr_flags & FR_OUTQUE ? 1 : 0; 415 isn->is_rule = fr; 416 ips.ips_is.is_rule = fr; 417 if (*fr->fr_ifname) { 418 fr->fr_ifa = GETUNIT(fr->fr_ifname, fr->fr_v); 419 if (fr->fr_ifa == NULL) 420 fr->fr_ifa = (void *)-1; 421#ifdef _KERNEL 422 else { 423 strncpy(isn->is_ifname[out], 424 IFNAME(fr->fr_ifa), IFNAMSIZ); 425 isn->is_ifp[out] = fr->fr_ifa; 426 } 427#endif 428 } else 429 fr->fr_ifa = NULL; 430 /* 431 * send a copy back to userland of what we ended up 432 * to allow for verification. 433 */ 434 error = IWCOPY((caddr_t)&ips, ipsp, sizeof(ips)); 435 if (error) { 436 KFREE(isn); 437 KFREE(fr); 438 return EFAULT; 439 } 440 } else { 441 for (is = ips_list; is; is = is->is_next) 442 if (is->is_rule == fr) 443 break; 444 if (!is) { 445 KFREE(isn); 446 return ESRCH; 447 } 448 } 449 } 450 fr_stinsert(isn); 451 return 0; 452} 453 454 455void fr_stinsert(is) 456register ipstate_t *is; 457{ 458 register u_int hv = is->is_hv; 459 460 MUTEX_INIT(&is->is_lock, "ipf state entry", NULL); 461 462 is->is_ifname[0][sizeof(is->is_ifname[0]) - 1] = '\0'; 463 if (is->is_ifname[0][0] != '\0') { 464 is->is_ifp[0] = GETUNIT(is->is_ifname[0], is->is_v); 465 } 466 is->is_ifname[1][sizeof(is->is_ifname[0]) - 1] = '\0'; 467 if (is->is_ifname[1][0] != '\0') { 468 is->is_ifp[1] = GETUNIT(is->is_ifname[1], is->is_v); 469 } 470 471 /* 472 * add into list table. 473 */ 474 if (ips_list) 475 ips_list->is_pnext = &is->is_next; 476 is->is_pnext = &ips_list; 477 is->is_next = ips_list; 478 ips_list = is; 479 if (ips_table[hv]) 480 ips_table[hv]->is_phnext = &is->is_hnext; 481 else 482 ips_stats.iss_inuse++; 483 is->is_phnext = ips_table + hv; 484 is->is_hnext = ips_table[hv]; 485 ips_table[hv] = is; 486 ips_num++; 487} 488 489 490/* 491 * Create a new ipstate structure and hang it off the hash table. 492 */ 493ipstate_t *fr_addstate(ip, fin, flags) 494ip_t *ip; 495fr_info_t *fin; 496u_int flags; 497{ 498 register tcphdr_t *tcp = NULL; 499 register ipstate_t *is; 500 register u_int hv; 501 ipstate_t ips; 502 u_int pass; 503 int out; 504 505 if (fr_state_lock || (fin->fin_off & IP_OFFMASK) || 506 (fin->fin_fi.fi_fl & FI_SHORT)) 507 return NULL; 508 if (ips_num == fr_statemax) { 509 ips_stats.iss_max++; 510 fr_state_doflush = 1; 511 return NULL; 512 } 513 out = fin->fin_out; 514 is = &ips; 515 bzero((char *)is, sizeof(*is)); 516 ips.is_age = 1; 517 ips.is_state[0] = 0; 518 ips.is_state[1] = 0; 519 /* 520 * Copy and calculate... 521 */ 522 hv = (is->is_p = fin->fin_fi.fi_p); 523 is->is_src = fin->fin_fi.fi_src; 524 hv += is->is_saddr; 525 is->is_dst = fin->fin_fi.fi_dst; 526 hv += is->is_daddr; 527#ifdef USE_INET6 528 if (fin->fin_v == 6) { 529 if (is->is_p == IPPROTO_ICMPV6) { 530 if (IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) 531 flags |= FI_W_DADDR; 532 if (out) 533 hv -= is->is_daddr; 534 else 535 hv -= is->is_saddr; 536 } 537 } 538#endif 539 540 switch (is->is_p) 541 { 542#ifdef USE_INET6 543 case IPPROTO_ICMPV6 : 544#endif 545 case IPPROTO_ICMP : 546 { 547 struct icmp *ic = (struct icmp *)fin->fin_dp; 548 549#ifdef USE_INET6 550 if ((is->is_p == IPPROTO_ICMPV6) && 551 ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)) 552 return NULL; 553#endif 554 switch (ic->icmp_type) 555 { 556#ifdef USE_INET6 557 case ICMP6_ECHO_REQUEST : 558 is->is_icmp.ics_type = ICMP6_ECHO_REPLY; 559 hv += (is->is_icmp.ics_id = ic->icmp_id); 560 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 561 break; 562 case ICMP6_MEMBERSHIP_QUERY : 563 case ND_ROUTER_SOLICIT : 564 case ND_NEIGHBOR_SOLICIT : 565 is->is_icmp.ics_type = ic->icmp_type + 1; 566 break; 567#endif 568 case ICMP_ECHO : 569 case ICMP_TSTAMP : 570 case ICMP_IREQ : 571 case ICMP_MASKREQ : 572 is->is_icmp.ics_type = ic->icmp_type; 573 hv += (is->is_icmp.ics_id = ic->icmp_id); 574 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 575 break; 576 default : 577 return NULL; 578 } 579 ATOMIC_INCL(ips_stats.iss_icmp); 580 is->is_age = fr_icmptimeout; 581 break; 582 } 583 case IPPROTO_TCP : 584 { 585 tcp = (tcphdr_t *)fin->fin_dp; 586 587 if (tcp->th_flags & TH_RST) 588 return NULL; 589 /* 590 * The endian of the ports doesn't matter, but the ack and 591 * sequence numbers do as we do mathematics on them later. 592 */ 593 is->is_dport = tcp->th_dport; 594 is->is_sport = tcp->th_sport; 595 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 596 hv += tcp->th_dport; 597 hv += tcp->th_sport; 598 } 599 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 600 (tcp->th_off << 2) + 601 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 602 ((tcp->th_flags & TH_FIN) ? 1 : 0); 603 is->is_maxsend = is->is_send; 604 is->is_dend = 0; 605 is->is_maxdwin = 1; 606 is->is_maxswin = ntohs(tcp->th_win); 607 if (is->is_maxswin == 0) 608 is->is_maxswin = 1; 609 /* 610 * If we're creating state for a starting connection, start the 611 * timer on it as we'll never see an error if it fails to 612 * connect. 613 */ 614 ATOMIC_INCL(ips_stats.iss_tcp); 615 break; 616 } 617 case IPPROTO_UDP : 618 { 619 tcp = (tcphdr_t *)fin->fin_dp; 620 621 is->is_dport = tcp->th_dport; 622 is->is_sport = tcp->th_sport; 623 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 624 hv += tcp->th_dport; 625 hv += tcp->th_sport; 626 } 627 ATOMIC_INCL(ips_stats.iss_udp); 628 is->is_age = fr_udptimeout; 629 break; 630 } 631 default : 632 return NULL; 633 } 634 635 KMALLOC(is, ipstate_t *); 636 if (is == NULL) { 637 ATOMIC_INCL(ips_stats.iss_nomem); 638 return NULL; 639 } 640 bcopy((char *)&ips, (char *)is, sizeof(*is)); 641 hv %= fr_statesize; 642 is->is_hv = hv; 643 is->is_rule = fin->fin_fr; 644 if (is->is_rule != NULL) { 645 ATOMIC_INC32(is->is_rule->fr_ref); 646 pass = is->is_rule->fr_flags; 647 } else 648 pass = fr_flags; 649 WRITE_ENTER(&ipf_state); 650 651 is->is_pass = pass; 652 is->is_pkts = 1; 653 is->is_bytes = fin->fin_dlen + fin->fin_hlen; 654 /* 655 * We want to check everything that is a property of this packet, 656 * but we don't (automatically) care about it's fragment status as 657 * this may change. 658 */ 659 is->is_v = fin->fin_fi.fi_v; 660 is->is_opt = fin->fin_fi.fi_optmsk; 661 is->is_optmsk = 0xffffffff; 662 is->is_sec = fin->fin_fi.fi_secmsk; 663 is->is_secmsk = 0xffff; 664 is->is_auth = fin->fin_fi.fi_auth; 665 is->is_authmsk = 0xffff; 666 is->is_flags = fin->fin_fi.fi_fl & FI_CMP; 667 is->is_flags |= FI_CMP << 4; 668 is->is_flags |= flags & (FI_WILDP|FI_WILDA); 669 if (flags & (FI_WILDP|FI_WILDA)) 670 ips_wild++; 671 is->is_ifp[1 - out] = NULL; 672 is->is_ifp[out] = fin->fin_ifp; 673#ifdef _KERNEL 674 strncpy(is->is_ifname[out], IFNAME(fin->fin_ifp), IFNAMSIZ); 675#endif 676 is->is_ifname[1 - out][0] = '\0'; 677 if (pass & FR_LOGFIRST) 678 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 679 fr_stinsert(is); 680 if (is->is_p == IPPROTO_TCP) { 681 MUTEX_ENTER(&is->is_lock); 682 fr_tcp_age(&is->is_age, is->is_state, fin, 683 0); /* 0 = packet from the source */ 684 MUTEX_EXIT(&is->is_lock); 685 } 686#ifdef IPFILTER_LOG 687 ipstate_log(is, ISL_NEW); 688#endif 689 RWLOCK_EXIT(&ipf_state); 690 fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst); 691 if (fin->fin_fi.fi_fl & FI_FRAG) 692 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 693 return is; 694} 695 696 697 698/* 699 * check to see if a packet with TCP headers fits within the TCP window. 700 * change timeout depending on whether new packet is a SYN-ACK returning for a 701 * SYN or a RST or FIN which indicate time to close up shop. 702 */ 703int fr_tcpstate(is, fin, ip, tcp) 704register ipstate_t *is; 705fr_info_t *fin; 706ip_t *ip; 707tcphdr_t *tcp; 708{ 709 register tcp_seq seq, ack, end; 710 register int ackskew; 711 tcpdata_t *fdata, *tdata; 712 u_short win, maxwin; 713 int ret = 0; 714 int source; 715 716 /* 717 * Find difference between last checked packet and this packet. 718 */ 719 source = IP6EQ(fin->fin_fi.fi_src, is->is_src); 720 fdata = &is->is_tcp.ts_data[!source]; 721 tdata = &is->is_tcp.ts_data[source]; 722 seq = ntohl(tcp->th_seq); 723 ack = ntohl(tcp->th_ack); 724 win = ntohs(tcp->th_win); 725 end = seq + fin->fin_dlen - (tcp->th_off << 2) + 726 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 727 ((tcp->th_flags & TH_FIN) ? 1 : 0); 728 729 MUTEX_ENTER(&is->is_lock); 730 if (fdata->td_end == 0) { 731 /* 732 * Must be a (outgoing) SYN-ACK in reply to a SYN. 733 */ 734 fdata->td_end = end; 735 fdata->td_maxwin = 1; 736 fdata->td_maxend = end + 1; 737 } 738 739 if (!(tcp->th_flags & TH_ACK)) { /* Pretend an ack was sent */ 740 ack = tdata->td_end; 741 } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 742 (ack == 0)) { 743 /* gross hack to get around certain broken tcp stacks */ 744 ack = tdata->td_end; 745 } 746 747 if (seq == end) 748 seq = end = fdata->td_end; 749 750 maxwin = tdata->td_maxwin; 751 ackskew = tdata->td_end - ack; 752 753#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 754#define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 755 if ((SEQ_GE(fdata->td_maxend, end)) && 756 (SEQ_GE(seq, fdata->td_end - maxwin)) && 757/* XXX what about big packets */ 758#define MAXACKWINDOW 66000 759 (ackskew >= -MAXACKWINDOW) && 760 (ackskew <= MAXACKWINDOW)) { 761 /* if ackskew < 0 then this should be due to fragented 762 * packets. There is no way to know the length of the 763 * total packet in advance. 764 * We do know the total length from the fragment cache though. 765 * Note however that there might be more sessions with 766 * exactly the same source and destination paramters in the 767 * state cache (and source and destination is the only stuff 768 * that is saved in the fragment cache). Note further that 769 * some TCP connections in the state cache are hashed with 770 * sport and dport as well which makes it not worthwhile to 771 * look for them. 772 * Thus, when ackskew is negative but still seems to belong 773 * to this session, we bump up the destinations end value. 774 */ 775 if (ackskew < 0) 776 tdata->td_end = ack; 777 778 /* update max window seen */ 779 if (fdata->td_maxwin < win) 780 fdata->td_maxwin = win; 781 if (SEQ_GT(end, fdata->td_end)) 782 fdata->td_end = end; 783 if (SEQ_GE(ack + win, tdata->td_maxend)) { 784 tdata->td_maxend = ack + win; 785 if (win == 0) 786 tdata->td_maxend++; 787 } 788 789 ATOMIC_INCL(ips_stats.iss_hits); 790 is->is_pkts++; 791 is->is_bytes += fin->fin_dlen + fin->fin_hlen; 792 /* 793 * Nearing end of connection, start timeout. 794 */ 795 /* source ? 0 : 1 -> !source */ 796 fr_tcp_age(&is->is_age, is->is_state, fin, !source); 797 ret = 1; 798 } 799 MUTEX_EXIT(&is->is_lock); 800 return ret; 801} 802 803 804static int fr_matchsrcdst(is, src, dst, fin, tcp) 805ipstate_t *is; 806union i6addr src, dst; 807fr_info_t *fin; 808tcphdr_t *tcp; 809{ 810 int ret = 0, rev, out, flags; 811 u_short sp, dp; 812 void *ifp; 813 814 rev = fin->fin_rev = IP6NEQ(is->is_dst, dst); 815 ifp = fin->fin_ifp; 816 out = fin->fin_out; 817 818 if (tcp != NULL) { 819 flags = is->is_flags; 820 sp = tcp->th_sport; 821 dp = tcp->th_dport; 822 } else { 823 flags = is->is_flags & FI_WILDA; 824 sp = 0; 825 dp = 0; 826 } 827 828 if (rev == 0) { 829 if (!out) { 830 if (is->is_ifpin == NULL || is->is_ifpin == ifp) 831 ret = 1; 832 } else { 833 if (is->is_ifpout == NULL || is->is_ifpout == ifp) 834 ret = 1; 835 } 836 } else { 837 if (out) { 838 if (is->is_ifpin == NULL || is->is_ifpin == ifp) 839 ret = 1; 840 } else { 841 if (is->is_ifpout == NULL || is->is_ifpout == ifp) 842 ret = 1; 843 } 844 } 845 if (ret == 0) 846 return 0; 847 ret = 0; 848 849 if (rev == 0) { 850 if ( 851 (IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) && 852 (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) && 853 (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) && 854 (dp == is->is_dport || flags & FI_W_DPORT)))) { 855 ret = 1; 856 } 857 } else { 858 if ( 859 (IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) && 860 (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) && 861 (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) && 862 (dp == is->is_sport || flags & FI_W_SPORT)))) { 863 ret = 1; 864 } 865 } 866 if (ret == 0) 867 return 0; 868 869 /* 870 * Whether or not this should be here, is questionable, but the aim 871 * is to get this out of the main line. 872 */ 873 if (tcp == NULL) 874 flags = is->is_flags & (FI_CMP|(FI_CMP<<4)); 875 876 if (((fin->fin_fi.fi_fl & (flags >> 4)) != (flags & FI_CMP)) || 877 ((fin->fin_fi.fi_optmsk & is->is_optmsk) != is->is_opt) || 878 ((fin->fin_fi.fi_secmsk & is->is_secmsk) != is->is_sec) || 879 ((fin->fin_fi.fi_auth & is->is_authmsk) != is->is_auth)) 880 return 0; 881 882 if ((flags & (FI_W_SPORT|FI_W_DPORT))) { 883 if ((flags & FI_W_SPORT) != 0) { 884 if (rev == 0) { 885 is->is_sport = sp; 886 is->is_send = htonl(tcp->th_seq); 887 } else { 888 is->is_sport = dp; 889 is->is_send = htonl(tcp->th_ack); 890 } 891 is->is_maxsend = is->is_send + 1; 892 } else if ((flags & FI_W_DPORT) != 0) { 893 if (rev == 0) { 894 is->is_dport = dp; 895 is->is_dend = htonl(tcp->th_ack); 896 } else { 897 is->is_dport = sp; 898 is->is_dend = htonl(tcp->th_seq); 899 } 900 is->is_maxdend = is->is_dend + 1; 901 } 902 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT); 903 ips_wild--; 904 } 905 906 ret = -1; 907 908 if (!rev) { 909 if (out) { 910 if (!is->is_ifpout) 911 ret = 1; 912 } else { 913 if (!is->is_ifpin) 914 ret = 0; 915 } 916 } else { 917 if (out) { 918 if (!is->is_ifpin) 919 ret = 0; 920 } else { 921 if (!is->is_ifpout) 922 ret = 1; 923 } 924 } 925 926 if (ret >= 0) { 927 is->is_ifp[ret] = ifp; 928#ifdef _KERNEL 929 strncpy(is->is_ifname[out], IFNAME(fin->fin_ifp), 930 sizeof(is->is_ifname[1])); 931#endif 932 } 933#ifdef _KERNEL 934 if (ret >= 0) { 935 strncpy(is->is_ifname[out], IFNAME(fin->fin_ifp), 936 sizeof(is->is_ifname[1])); 937 } 938#endif 939 return 1; 940} 941 942static int fr_matchicmpqueryreply(v, is, icmp) 943int v; 944ipstate_t *is; 945icmphdr_t *icmp; 946{ 947 if (v == 4) { 948 /* 949 * If we matched its type on the way in, then when going out 950 * it will still be the same type. 951 */ 952 if (((icmp->icmp_type == is->is_type) || 953 (icmpreplytype4[is->is_type] == icmp->icmp_type)) && 954 (icmp->icmp_id == is->is_icmp.ics_id) && 955 (icmp->icmp_seq == is->is_icmp.ics_seq)) { 956 return 1; 957 }; 958 } 959#ifdef USE_INET6 960 else if (is->is_v == 6) { 961 if ((is->is_type == ICMP6_ECHO_REPLY) && 962 (icmp->icmp_type == ICMP6_ECHO_REQUEST) && 963 (icmp->icmp_id == is->is_icmp.ics_id) && 964 (icmp->icmp_seq == is->is_icmp.ics_seq)) { 965 return 1; 966 }; 967 } 968#endif 969 return 0; 970} 971 972static frentry_t *fr_checkicmpmatchingstate(ip, fin) 973ip_t *ip; 974fr_info_t *fin; 975{ 976 register ipstate_t *is, **isp; 977 register u_short sport, dport; 978 register u_char pr; 979 union i6addr dst, src; 980 struct icmp *ic; 981 u_short savelen; 982 icmphdr_t *icmp; 983 fr_info_t ofin; 984 int type, len; 985 tcphdr_t *tcp; 986 frentry_t *fr; 987 ip_t *oip; 988 u_int hv; 989 990 /* 991 * Does it at least have the return (basic) IP header ? 992 * Only a basic IP header (no options) should be with 993 * an ICMP error header. 994 */ 995 if (((ip->ip_v != 4) || (ip->ip_hl != 5)) || 996 (fin->fin_plen < ICMPERR_MINPKTLEN)) 997 return NULL; 998 ic = (struct icmp *)fin->fin_dp; 999 type = ic->icmp_type; 1000 /* 1001 * If it's not an error type, then return 1002 */ 1003 if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) && 1004 (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) && 1005 (type != ICMP_PARAMPROB)) 1006 return NULL; 1007 1008 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 1009 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((oip->ip_hl - 5) << 2)) 1010 return NULL; 1011 1012 /* 1013 * Sanity checks. 1014 */ 1015 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 1016 if ((len <= 0) || ((oip->ip_hl << 2) > len)) 1017 return NULL; 1018 1019 /* 1020 * Is the buffer big enough for all of it ? It's the size of the IP 1021 * header claimed in the encapsulated part which is of concern. It 1022 * may be too big to be in this buffer but not so big that it's 1023 * outside the ICMP packet, leading to TCP deref's causing problems. 1024 * This is possible because we don't know how big oip_hl is when we 1025 * do the pullup early in fr_check() and thus can't gaurantee it is 1026 * all here now. 1027 */ 1028#ifdef _KERNEL 1029 { 1030 mb_t *m; 1031 1032# if SOLARIS 1033 m = fin->fin_qfm; 1034 if ((char *)oip + len > (char *)m->b_wptr) 1035 return NULL; 1036# else 1037 m = *(mb_t **)fin->fin_mp; 1038 if ((char *)oip + len > (char *)ip + m->m_len) 1039 return NULL; 1040# endif 1041 } 1042#endif 1043 1044 /* 1045 * in the IPv4 case we must zero the i6addr union otherwise 1046 * the IP6EQ and IP6NEQ macros produce the wrong results because 1047 * of the 'junk' in the unused part of the union 1048 */ 1049 bzero((char *)&src, sizeof(src)); 1050 bzero((char *)&dst, sizeof(dst)); 1051 1052 if (oip->ip_p == IPPROTO_ICMP) { 1053 icmp = (icmphdr_t *)((char *)oip + (oip->ip_hl << 2)); 1054 1055 /* 1056 * a ICMP error can only be generated as a result of an 1057 * ICMP query, not as the response on an ICMP error 1058 * 1059 * XXX theoretically ICMP_ECHOREP and the other reply's are 1060 * ICMP query's as well, but adding them here seems strange XXX 1061 */ 1062 if ((icmp->icmp_type != ICMP_ECHO) && 1063 (icmp->icmp_type != ICMP_TSTAMP) && 1064 (icmp->icmp_type != ICMP_IREQ) && 1065 (icmp->icmp_type != ICMP_MASKREQ)) 1066 return NULL; 1067 1068 /* 1069 * perform a lookup of the ICMP packet in the state table 1070 */ 1071 hv = (pr = oip->ip_p); 1072 src.in4 = oip->ip_src; 1073 hv += src.in4.s_addr; 1074 dst.in4 = oip->ip_dst; 1075 hv += dst.in4.s_addr; 1076 hv += icmp->icmp_id; 1077 hv += icmp->icmp_seq; 1078 hv %= fr_statesize; 1079 1080 savelen = oip->ip_len; 1081 oip->ip_len = len; 1082 ofin.fin_v = 4; 1083 fr_makefrip(oip->ip_hl << 2, oip, &ofin); 1084 oip->ip_len = savelen; 1085 ofin.fin_ifp = fin->fin_ifp; 1086 ofin.fin_out = !fin->fin_out; 1087 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1088 1089 READ_ENTER(&ipf_state); 1090 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 1091 if ((is->is_p == pr) && (is->is_v == 4) && 1092 fr_matchsrcdst(is, src, dst, &ofin, NULL) && 1093 fr_matchicmpqueryreply(is->is_v, is, icmp)) { 1094 ips_stats.iss_hits++; 1095 is->is_pkts++; 1096 is->is_bytes += ip->ip_len; 1097 fr = is->is_rule; 1098 RWLOCK_EXIT(&ipf_state); 1099 return fr; 1100 } 1101 RWLOCK_EXIT(&ipf_state); 1102 return NULL; 1103 }; 1104 1105 if ((oip->ip_p != IPPROTO_TCP) && (oip->ip_p != IPPROTO_UDP)) 1106 return NULL; 1107 1108 tcp = (tcphdr_t *)((char *)oip + (oip->ip_hl << 2)); 1109 dport = tcp->th_dport; 1110 sport = tcp->th_sport; 1111 1112 hv = (pr = oip->ip_p); 1113 src.in4 = oip->ip_src; 1114 hv += src.in4.s_addr; 1115 dst.in4 = oip->ip_dst; 1116 hv += dst.in4.s_addr; 1117 hv += dport; 1118 hv += sport; 1119 hv %= fr_statesize; 1120 /* 1121 * we make an fin entry to be able to feed it to 1122 * matchsrcdst note that not all fields are encessary 1123 * but this is the cleanest way. Note further we fill 1124 * in fin_mp such that if someone uses it we'll get 1125 * a kernel panic. fr_matchsrcdst does not use this. 1126 * 1127 * watch out here, as ip is in host order and oip in network 1128 * order. Any change we make must be undone afterwards. 1129 */ 1130 savelen = oip->ip_len; 1131 oip->ip_len = len; 1132 ofin.fin_v = 4; 1133 fr_makefrip(oip->ip_hl << 2, oip, &ofin); 1134 oip->ip_len = savelen; 1135 ofin.fin_ifp = fin->fin_ifp; 1136 ofin.fin_out = !fin->fin_out; 1137 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1138 READ_ENTER(&ipf_state); 1139 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1140 /* 1141 * Only allow this icmp though if the 1142 * encapsulated packet was allowed through the 1143 * other way around. Note that the minimal amount 1144 * of info present does not allow for checking against 1145 * tcp internals such as seq and ack numbers. 1146 */ 1147 if ((is->is_p == pr) && (is->is_v == 4) && 1148 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 1149 fr = is->is_rule; 1150 ips_stats.iss_hits++; 1151 /* 1152 * we must swap src and dst here because the icmp 1153 * comes the other way around 1154 */ 1155 is->is_pkts++; 1156 is->is_bytes += fin->fin_plen; 1157 /* 1158 * we deliberately do not touch the timeouts 1159 * for the accompanying state table entry. 1160 * It remains to be seen if that is correct. XXX 1161 */ 1162 RWLOCK_EXIT(&ipf_state); 1163 return fr; 1164 } 1165 } 1166 RWLOCK_EXIT(&ipf_state); 1167 return NULL; 1168} 1169 1170 1171static void fr_ipsmove(isp, is, hv) 1172ipstate_t **isp, *is; 1173u_int hv; 1174{ 1175 u_int hvm; 1176 1177 hvm = is->is_hv; 1178 /* 1179 * Remove the hash from the old location... 1180 */ 1181 if (is->is_hnext) 1182 is->is_hnext->is_phnext = isp; 1183 *isp = is->is_hnext; 1184 if (ips_table[hvm] == NULL) 1185 ips_stats.iss_inuse--; 1186 1187 /* 1188 * ...and put the hash in the new one. 1189 */ 1190 hvm = hv % fr_statesize; 1191 is->is_hv = hvm; 1192 isp = &ips_table[hvm]; 1193 if (*isp) 1194 (*isp)->is_phnext = &is->is_hnext; 1195 else 1196 ips_stats.iss_inuse++; 1197 is->is_phnext = isp; 1198 is->is_hnext = *isp; 1199 *isp = is; 1200} 1201 1202 1203/* 1204 * Check if a packet has a registered state. 1205 */ 1206frentry_t *fr_checkstate(ip, fin) 1207ip_t *ip; 1208fr_info_t *fin; 1209{ 1210 union i6addr dst, src; 1211 register ipstate_t *is, **isp; 1212 register u_char pr; 1213 u_int hv, hvm, hlen, tryagain, pass, v; 1214 struct icmp *ic; 1215 frentry_t *fr; 1216 tcphdr_t *tcp; 1217 1218 if (fr_state_lock || (fin->fin_off & IP_OFFMASK) || 1219 (fin->fin_fi.fi_fl & FI_SHORT)) 1220 return NULL; 1221 1222 is = NULL; 1223 hlen = fin->fin_hlen; 1224 tcp = (tcphdr_t *)((char *)ip + hlen); 1225 ic = (struct icmp *)tcp; 1226 hv = (pr = fin->fin_fi.fi_p); 1227 src = fin->fin_fi.fi_src; 1228 dst = fin->fin_fi.fi_dst; 1229 hv += src.in4.s_addr; 1230 hv += dst.in4.s_addr; 1231 1232 /* 1233 * Search the hash table for matching packet header info. 1234 */ 1235 v = fin->fin_fi.fi_v; 1236 switch (fin->fin_fi.fi_p) 1237 { 1238#ifdef USE_INET6 1239 case IPPROTO_ICMPV6 : 1240 if (v == 6) { 1241 if (fin->fin_out) 1242 hv -= dst.in4.s_addr; 1243 else 1244 hv -= src.in4.s_addr; 1245 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 1246 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 1247 hv += ic->icmp_id; 1248 hv += ic->icmp_seq; 1249 } 1250 } 1251#endif 1252 case IPPROTO_ICMP : 1253 if (v == 4) { 1254 hv += ic->icmp_id; 1255 hv += ic->icmp_seq; 1256 } 1257 hv %= fr_statesize; 1258 READ_ENTER(&ipf_state); 1259 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1260 if ((is->is_p == pr) && (is->is_v == v) && 1261 fr_matchsrcdst(is, src, dst, fin, NULL) && 1262 fr_matchicmpqueryreply(v, is, ic)) { 1263 is->is_age = fr_icmptimeout; 1264 break; 1265 } 1266 } 1267 if (is != NULL) 1268 break; 1269 RWLOCK_EXIT(&ipf_state); 1270 /* 1271 * No matching icmp state entry. Perhaps this is a 1272 * response to another state entry. 1273 */ 1274#ifdef USE_INET6 1275 if (v == 6) 1276 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin); 1277 else 1278#endif 1279 fr = fr_checkicmpmatchingstate(ip, fin); 1280 if (fr) 1281 return fr; 1282 break; 1283 case IPPROTO_TCP : 1284 { 1285 register u_short dport, sport; 1286 register int i; 1287 1288 i = tcp->th_flags; 1289 /* 1290 * Just plain ignore RST flag set with either FIN or SYN. 1291 */ 1292 if ((i & TH_RST) && 1293 ((i & (TH_FIN|TH_SYN|TH_RST)) != TH_RST)) 1294 break; 1295 case IPPROTO_UDP : 1296 dport = tcp->th_dport; 1297 sport = tcp->th_sport; 1298 tryagain = 0; 1299 hv += dport; 1300 hv += sport; 1301 READ_ENTER(&ipf_state); 1302retry_tcpudp: 1303 hvm = hv % fr_statesize; 1304 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext) 1305 if ((is->is_p == pr) && (is->is_v == v) && 1306 fr_matchsrcdst(is, src, dst, fin, tcp)) { 1307 if ((pr == IPPROTO_TCP)) { 1308 if (!fr_tcpstate(is, fin, ip, tcp)) { 1309 continue; 1310 } 1311 } 1312 break; 1313 } 1314 if (is != NULL) { 1315 if (tryagain && 1316 !(is->is_flags & (FI_WILDP|FI_WILDA))) { 1317 hv += dport; 1318 hv += sport; 1319 fr_ipsmove(isp, is, hv); 1320 MUTEX_DOWNGRADE(&ipf_state); 1321 } 1322 break; 1323 } 1324 RWLOCK_EXIT(&ipf_state); 1325 if (!tryagain && ips_wild) { 1326 hv -= dport; 1327 hv -= sport; 1328 tryagain = 1; 1329 WRITE_ENTER(&ipf_state); 1330 goto retry_tcpudp; 1331 } 1332 break; 1333 } 1334 default : 1335 break; 1336 } 1337 if (is == NULL) { 1338 ATOMIC_INCL(ips_stats.iss_miss); 1339 return NULL; 1340 } 1341 MUTEX_ENTER(&is->is_lock); 1342 is->is_bytes += fin->fin_plen; 1343 ips_stats.iss_hits++; 1344 is->is_pkts++; 1345 MUTEX_EXIT(&is->is_lock); 1346 fr = is->is_rule; 1347 fin->fin_fr = fr; 1348 pass = is->is_pass; 1349#ifndef _KERNEL 1350 if (tcp->th_flags & TCP_CLOSE) 1351 fr_delstate(is); 1352#endif 1353 RWLOCK_EXIT(&ipf_state); 1354 if (fin->fin_fi.fi_fl & FI_FRAG) 1355 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 1356 return fr; 1357} 1358 1359 1360void ip_statesync(ifp) 1361void *ifp; 1362{ 1363 register ipstate_t *is; 1364 1365 WRITE_ENTER(&ipf_state); 1366 for (is = ips_list; is; is = is->is_next) { 1367 if (is->is_ifpin == ifp) { 1368 is->is_ifpin = GETUNIT(is->is_ifname[0], is->is_v); 1369 if (!is->is_ifpin) 1370 is->is_ifpin = (void *)-1; 1371 } 1372 if (is->is_ifpout == ifp) { 1373 is->is_ifpout = GETUNIT(is->is_ifname[1], is->is_v); 1374 if (!is->is_ifpout) 1375 is->is_ifpout = (void *)-1; 1376 } 1377 } 1378 RWLOCK_EXIT(&ipf_state); 1379} 1380 1381 1382static void fr_delstate(is) 1383ipstate_t *is; 1384{ 1385 frentry_t *fr; 1386 1387 if (is->is_flags & (FI_WILDP|FI_WILDA)) 1388 ips_wild--; 1389 if (is->is_next) 1390 is->is_next->is_pnext = is->is_pnext; 1391 *is->is_pnext = is->is_next; 1392 if (is->is_hnext) 1393 is->is_hnext->is_phnext = is->is_phnext; 1394 *is->is_phnext = is->is_hnext; 1395 if (ips_table[is->is_hv] == NULL) 1396 ips_stats.iss_inuse--; 1397 1398 fr = is->is_rule; 1399 if (fr != NULL) { 1400 ATOMIC_DEC32(fr->fr_ref); 1401 if (fr->fr_ref == 0) 1402 KFREE(fr); 1403 } 1404#ifdef _KERNEL 1405 MUTEX_DESTROY(&is->is_lock); 1406#endif 1407 KFREE(is); 1408 ips_num--; 1409} 1410 1411 1412/* 1413 * Free memory in use by all state info. kept. 1414 */ 1415void fr_stateunload() 1416{ 1417 register ipstate_t *is; 1418 1419 WRITE_ENTER(&ipf_state); 1420 while ((is = ips_list)) 1421 fr_delstate(is); 1422 ips_stats.iss_inuse = 0; 1423 ips_num = 0; 1424 RWLOCK_EXIT(&ipf_state); 1425 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *)); 1426 ips_table = NULL; 1427} 1428 1429 1430/* 1431 * Slowly expire held state for thingslike UDP and ICMP. Timeouts are set 1432 * in expectation of this being called twice per second. 1433 */ 1434void fr_timeoutstate() 1435{ 1436 register ipstate_t *is, **isp; 1437#if defined(_KERNEL) && !SOLARIS 1438 int s; 1439#endif 1440 1441 SPL_NET(s); 1442 WRITE_ENTER(&ipf_state); 1443 for (isp = &ips_list; (is = *isp); ) 1444 if (is->is_age && !--is->is_age) { 1445 if (is->is_p == IPPROTO_TCP) 1446 ips_stats.iss_fin++; 1447 else 1448 ips_stats.iss_expire++; 1449#ifdef IPFILTER_LOG 1450 ipstate_log(is, ISL_EXPIRE); 1451#endif 1452 fr_delstate(is); 1453 } else 1454 isp = &is->is_next; 1455 RWLOCK_EXIT(&ipf_state); 1456 SPL_X(s); 1457 if (fr_state_doflush) { 1458 (void) fr_state_flush(1); 1459 fr_state_doflush = 0; 1460 } 1461} 1462 1463 1464/* 1465 * Original idea freom Pradeep Krishnan for use primarily with NAT code. 1466 * (pkrishna@netcom.com) 1467 * 1468 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: 1469 * 1470 * - (try to) base state transitions on real evidence only, 1471 * i.e. packets that are sent and have been received by ipfilter; 1472 * diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. 1473 * 1474 * - deal with half-closed connections correctly; 1475 * 1476 * - store the state of the source in state[0] such that ipfstat 1477 * displays the state as source/dest instead of dest/source; the calls 1478 * to fr_tcp_age have been changed accordingly. 1479 * 1480 * Parameters: 1481 * 1482 * state[0] = state of source (host that initiated connection) 1483 * state[1] = state of dest (host that accepted the connection) 1484 * 1485 * dir == 0 : a packet from source to dest 1486 * dir == 1 : a packet from dest to source 1487 * 1488 */ 1489void fr_tcp_age(age, state, fin, dir) 1490u_long *age; 1491u_char *state; 1492fr_info_t *fin; 1493int dir; 1494{ 1495 tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp; 1496 u_char flags = tcp->th_flags; 1497 int dlen, ostate; 1498 1499 ostate = state[1 - dir]; 1500 1501 dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2); 1502 1503 if (flags & TH_RST) { 1504 if (!(tcp->th_flags & TH_PUSH) && !dlen) { 1505 *age = fr_tcpclosed; 1506 state[dir] = TCPS_CLOSED; 1507 } else { 1508 *age = fr_tcpclosewait; 1509 state[dir] = TCPS_CLOSE_WAIT; 1510 } 1511 return; 1512 } 1513 1514 *age = fr_tcptimeout; /* default 4 mins */ 1515 1516 switch(state[dir]) 1517 { 1518 case TCPS_CLOSED: /* 0 */ 1519 if ((flags & TH_OPENING) == TH_OPENING) { 1520 /* 1521 * 'dir' received an S and sends SA in response, 1522 * CLOSED -> SYN_RECEIVED 1523 */ 1524 state[dir] = TCPS_SYN_RECEIVED; 1525 *age = fr_tcptimeout; 1526 } else if ((flags & (TH_SYN|TH_ACK)) == TH_SYN) { 1527 /* 'dir' sent S, CLOSED -> SYN_SENT */ 1528 state[dir] = TCPS_SYN_SENT; 1529 *age = fr_tcptimeout; 1530 } 1531 /* 1532 * The next piece of code makes it possible to get 1533 * already established connections into the state table 1534 * after a restart or reload of the filter rules; this 1535 * does not work when a strict 'flags S keep state' is 1536 * used for tcp connections of course 1537 */ 1538 if ((flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) { 1539 /* we saw an A, guess 'dir' is in ESTABLISHED mode */ 1540 state[dir] = TCPS_ESTABLISHED; 1541 *age = fr_tcpidletimeout; 1542 } 1543 /* 1544 * TODO: besides regular ACK packets we can have other 1545 * packets as well; it is yet to be determined how we 1546 * should initialize the states in those cases 1547 */ 1548 break; 1549 1550 case TCPS_LISTEN: /* 1 */ 1551 /* NOT USED */ 1552 break; 1553 1554 case TCPS_SYN_SENT: /* 2 */ 1555 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1556 /* 1557 * We see an A from 'dir' which is in SYN_SENT 1558 * state: 'dir' sent an A in response to an SA 1559 * which it received, SYN_SENT -> ESTABLISHED 1560 */ 1561 state[dir] = TCPS_ESTABLISHED; 1562 *age = fr_tcpidletimeout; 1563 } else if (flags & TH_FIN) { 1564 /* 1565 * We see an F from 'dir' which is in SYN_SENT 1566 * state and wants to close its side of the 1567 * connection; SYN_SENT -> FIN_WAIT_1 1568 */ 1569 state[dir] = TCPS_FIN_WAIT_1; 1570 *age = fr_tcpidletimeout; /* or fr_tcptimeout? */ 1571 } else if ((flags & TH_OPENING) == TH_OPENING) { 1572 /* 1573 * We see an SA from 'dir' which is already in 1574 * SYN_SENT state, this means we have a 1575 * simultaneous open; SYN_SENT -> SYN_RECEIVED 1576 */ 1577 state[dir] = TCPS_SYN_RECEIVED; 1578 *age = fr_tcptimeout; 1579 } 1580 break; 1581 1582 case TCPS_SYN_RECEIVED: /* 3 */ 1583 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1584 /* 1585 * We see an A from 'dir' which was in SYN_RECEIVED 1586 * state so it must now be in established state, 1587 * SYN_RECEIVED -> ESTABLISHED 1588 */ 1589 state[dir] = TCPS_ESTABLISHED; 1590 *age = fr_tcpidletimeout; 1591 } else if (flags & TH_FIN) { 1592 /* 1593 * We see an F from 'dir' which is in SYN_RECEIVED 1594 * state and wants to close its side of the connection; 1595 * SYN_RECEIVED -> FIN_WAIT_1 1596 */ 1597 state[dir] = TCPS_FIN_WAIT_1; 1598 *age = fr_tcpidletimeout; 1599 } 1600 break; 1601 1602 case TCPS_ESTABLISHED: /* 4 */ 1603 if (flags & TH_FIN) { 1604 /* 1605 * 'dir' closed its side of the connection; this 1606 * gives us a half-closed connection; 1607 * ESTABLISHED -> FIN_WAIT_1 1608 */ 1609 state[dir] = TCPS_FIN_WAIT_1; 1610 *age = fr_tcphalfclosed; 1611 } else if (flags & TH_ACK) { 1612 /* an ACK, should we exclude other flags here? */ 1613 if (ostate == TCPS_FIN_WAIT_1) { 1614 /* 1615 * We know the other side did an active close, 1616 * so we are ACKing the recvd FIN packet (does 1617 * the window matching code guarantee this?) 1618 * and go into CLOSE_WAIT state; this gives us 1619 * a half-closed connection 1620 */ 1621 state[dir] = TCPS_CLOSE_WAIT; 1622 *age = fr_tcphalfclosed; 1623 } else if (ostate < TCPS_CLOSE_WAIT) 1624 /* 1625 * Still a fully established connection, 1626 * reset timeout 1627 */ 1628 *age = fr_tcpidletimeout; 1629 } 1630 break; 1631 1632 case TCPS_CLOSE_WAIT: /* 5 */ 1633 if (flags & TH_FIN) { 1634 /* 1635 * Application closed and 'dir' sent a FIN, we're now 1636 * going into LAST_ACK state 1637 */ 1638 *age = fr_tcplastack; 1639 state[dir] = TCPS_LAST_ACK; 1640 } else { 1641 /* 1642 * We remain in CLOSE_WAIT because the other side has 1643 * closed already and we did not close our side yet; 1644 * reset timeout 1645 */ 1646 *age = fr_tcphalfclosed; 1647 } 1648 break; 1649 1650 case TCPS_FIN_WAIT_1: /* 6 */ 1651 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) { 1652 /* 1653 * If the other side is not active anymore it has sent 1654 * us a FIN packet that we are ack'ing now with an ACK; 1655 * this means both sides have now closed the connection 1656 * and we go into TIME_WAIT 1657 */ 1658 /* 1659 * XXX: how do we know we really are ACKing the FIN 1660 * packet here? does the window code guarantee that? 1661 */ 1662 state[dir] = TCPS_TIME_WAIT; 1663 *age = fr_tcptimeout; 1664 } else 1665 /* 1666 * We closed our side of the connection already but the 1667 * other side is still active (ESTABLISHED/CLOSE_WAIT); 1668 * continue with this half-closed connection 1669 */ 1670 *age = fr_tcphalfclosed; 1671 break; 1672 1673 case TCPS_CLOSING: /* 7 */ 1674 /* NOT USED */ 1675 break; 1676 1677 case TCPS_LAST_ACK: /* 8 */ 1678 if (flags & TH_ACK) { 1679 if ((flags & TH_PUSH) || dlen) 1680 /* 1681 * There is still data to be delivered, reset 1682 * timeout 1683 */ 1684 *age = fr_tcplastack; 1685 } 1686 /* 1687 * We cannot detect when we go out of LAST_ACK state to CLOSED 1688 * because that is based on the reception of ACK packets; 1689 * ipfilter can only detect that a packet has been sent by a 1690 * host 1691 */ 1692 break; 1693 1694 case TCPS_FIN_WAIT_2: /* 9 */ 1695 /* NOT USED */ 1696 break; 1697 1698 case TCPS_TIME_WAIT: /* 10 */ 1699 /* we're in 2MSL timeout now */ 1700 break; 1701 } 1702} 1703 1704 1705#ifdef IPFILTER_LOG 1706void ipstate_log(is, type) 1707struct ipstate *is; 1708u_int type; 1709{ 1710 struct ipslog ipsl; 1711 void *items[1]; 1712 size_t sizes[1]; 1713 int types[1]; 1714 1715 ipsl.isl_type = type; 1716 ipsl.isl_pkts = is->is_pkts; 1717 ipsl.isl_bytes = is->is_bytes; 1718 ipsl.isl_src = is->is_src; 1719 ipsl.isl_dst = is->is_dst; 1720 ipsl.isl_p = is->is_p; 1721 ipsl.isl_v = is->is_v; 1722 ipsl.isl_flags = is->is_flags; 1723 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 1724 ipsl.isl_sport = is->is_sport; 1725 ipsl.isl_dport = is->is_dport; 1726 if (ipsl.isl_p == IPPROTO_TCP) { 1727 ipsl.isl_state[0] = is->is_state[0]; 1728 ipsl.isl_state[1] = is->is_state[1]; 1729 } 1730 } else if (ipsl.isl_p == IPPROTO_ICMP) 1731 ipsl.isl_itype = is->is_icmp.ics_type; 1732 else { 1733 ipsl.isl_ps.isl_filler[0] = 0; 1734 ipsl.isl_ps.isl_filler[1] = 0; 1735 } 1736 items[0] = &ipsl; 1737 sizes[0] = sizeof(ipsl); 1738 types[0] = 0; 1739 1740 (void) ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1); 1741} 1742#endif 1743 1744 1745#ifdef USE_INET6 1746frentry_t *fr_checkicmp6matchingstate(ip, fin) 1747ip6_t *ip; 1748fr_info_t *fin; 1749{ 1750 register ipstate_t *is, **isp; 1751 register u_short sport, dport; 1752 register u_char pr; 1753 struct icmp6_hdr *ic, *oic; 1754 union i6addr dst, src; 1755 u_short savelen; 1756 fr_info_t ofin; 1757 tcphdr_t *tcp; 1758 frentry_t *fr; 1759 ip6_t *oip; 1760 int type; 1761 u_int hv; 1762 1763 /* 1764 * Does it at least have the return (basic) IP header ? 1765 * Only a basic IP header (no options) should be with 1766 * an ICMP error header. 1767 */ 1768 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN)) 1769 return NULL; 1770 ic = (struct icmp6_hdr *)fin->fin_dp; 1771 type = ic->icmp6_type; 1772 /* 1773 * If it's not an error type, then return 1774 */ 1775 if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) && 1776 (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB)) 1777 return NULL; 1778 1779 oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN); 1780 if (fin->fin_plen < sizeof(*oip)) 1781 return NULL; 1782 1783 if (oip->ip6_nxt == IPPROTO_ICMPV6) { 1784 oic = (struct icmp6_hdr *)(oip + 1); 1785 /* 1786 * a ICMP error can only be generated as a result of an 1787 * ICMP query, not as the response on an ICMP error 1788 * 1789 * XXX theoretically ICMP_ECHOREP and the other reply's are 1790 * ICMP query's as well, but adding them here seems strange XXX 1791 */ 1792 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 1793 return NULL; 1794 1795 /* 1796 * perform a lookup of the ICMP packet in the state table 1797 */ 1798 hv = (pr = oip->ip6_nxt); 1799 src.in6 = oip->ip6_src; 1800 hv += src.in4.s_addr; 1801 dst.in6 = oip->ip6_dst; 1802 hv += dst.in4.s_addr; 1803 hv += oic->icmp6_id; 1804 hv += oic->icmp6_seq; 1805 hv %= fr_statesize; 1806 1807 oip->ip6_plen = ntohs(oip->ip6_plen); 1808 ofin.fin_v = 6; 1809 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 1810 oip->ip6_plen = htons(oip->ip6_plen); 1811 ofin.fin_ifp = fin->fin_ifp; 1812 ofin.fin_out = !fin->fin_out; 1813 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1814 1815 READ_ENTER(&ipf_state); 1816 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 1817 if ((is->is_p == pr) && 1818 (oic->icmp6_id == is->is_icmp.ics_id) && 1819 (oic->icmp6_seq == is->is_icmp.ics_seq) && 1820 fr_matchsrcdst(is, src, dst, &ofin, NULL)) { 1821 /* 1822 * in the state table ICMP query's are stored 1823 * with the type of the corresponding ICMP 1824 * response. Correct here 1825 */ 1826 if (((is->is_type == ICMP6_ECHO_REPLY) && 1827 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 1828 (is->is_type - 1 == oic->icmp6_type )) { 1829 ips_stats.iss_hits++; 1830 is->is_pkts++; 1831 is->is_bytes += fin->fin_plen; 1832 return is->is_rule; 1833 } 1834 } 1835 RWLOCK_EXIT(&ipf_state); 1836 1837 return NULL; 1838 }; 1839 1840 if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP)) 1841 return NULL; 1842 tcp = (tcphdr_t *)(oip + 1); 1843 dport = tcp->th_dport; 1844 sport = tcp->th_sport; 1845 1846 hv = (pr = oip->ip6_nxt); 1847 src.in6 = oip->ip6_src; 1848 hv += src.in4.s_addr; 1849 dst.in6 = oip->ip6_dst; 1850 hv += dst.in4.s_addr; 1851 hv += dport; 1852 hv += sport; 1853 hv %= fr_statesize; 1854 /* 1855 * we make an fin entry to be able to feed it to 1856 * matchsrcdst note that not all fields are encessary 1857 * but this is the cleanest way. Note further we fill 1858 * in fin_mp such that if someone uses it we'll get 1859 * a kernel panic. fr_matchsrcdst does not use this. 1860 * 1861 * watch out here, as ip is in host order and oip in network 1862 * order. Any change we make must be undone afterwards. 1863 */ 1864 savelen = oip->ip6_plen; 1865 oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN; 1866 ofin.fin_v = 6; 1867 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 1868 oip->ip6_plen = savelen; 1869 ofin.fin_ifp = fin->fin_ifp; 1870 ofin.fin_out = !fin->fin_out; 1871 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 1872 READ_ENTER(&ipf_state); 1873 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1874 /* 1875 * Only allow this icmp though if the 1876 * encapsulated packet was allowed through the 1877 * other way around. Note that the minimal amount 1878 * of info present does not allow for checking against 1879 * tcp internals such as seq and ack numbers. 1880 */ 1881 if ((is->is_p == pr) && (is->is_v == 6) && 1882 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 1883 fr = is->is_rule; 1884 ips_stats.iss_hits++; 1885 /* 1886 * we must swap src and dst here because the icmp 1887 * comes the other way around 1888 */ 1889 is->is_pkts++; 1890 is->is_bytes += fin->fin_plen; 1891 /* 1892 * we deliberately do not touch the timeouts 1893 * for the accompanying state table entry. 1894 * It remains to be seen if that is correct. XXX 1895 */ 1896 RWLOCK_EXIT(&ipf_state); 1897 return fr; 1898 } 1899 } 1900 RWLOCK_EXIT(&ipf_state); 1901 return NULL; 1902} 1903#endif 1904