ip_state.c revision 98004
1/* 2 * Copyright (C) 1995-2002 by Darren Reed. 3 * 4 * See the IPFILTER.LICENCE file for details on licencing. 5 */ 6 7#ifdef __sgi 8# include <sys/ptimers.h> 9#endif 10#include <sys/errno.h> 11#include <sys/types.h> 12#include <sys/param.h> 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#ifndef linux 43# include <sys/protosw.h> 44#endif 45#include <sys/socket.h> 46#if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux) 47# include <sys/systm.h> 48#endif 49#if !defined(__SVR4) && !defined(__svr4__) 50# ifndef linux 51# include <sys/mbuf.h> 52# endif 53#else 54# include <sys/filio.h> 55# include <sys/byteorder.h> 56# ifdef _KERNEL 57# include <sys/dditypes.h> 58# endif 59# include <sys/stream.h> 60# include <sys/kmem.h> 61#endif 62 63#include <net/if.h> 64#ifdef sun 65# include <net/af.h> 66#endif 67#include <net/route.h> 68#include <netinet/in.h> 69#include <netinet/in_systm.h> 70#include <netinet/ip.h> 71#include <netinet/tcp.h> 72#ifndef linux 73# include <netinet/ip_var.h> 74# include <netinet/tcp_fsm.h> 75#endif 76#include <netinet/udp.h> 77#include <netinet/ip_icmp.h> 78#include "netinet/ip_compat.h" 79#include <netinet/tcpip.h> 80#include "netinet/ip_fil.h" 81#include "netinet/ip_nat.h" 82#include "netinet/ip_frag.h" 83#include "netinet/ip_state.h" 84#ifdef USE_INET6 85#include <netinet/icmp6.h> 86#endif 87#if (__FreeBSD_version >= 300000) 88# include <sys/malloc.h> 89# if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM) 90# include <sys/libkern.h> 91# include <sys/systm.h> 92# endif 93#endif 94 95#if !defined(lint) 96static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed"; 97/* static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.30.2.38 2001/07/23 13:49:46 darrenr Exp $"; */ 98static const char rcsid[] = "@(#)$FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 98004 2002-06-07 08:56:30Z darrenr $"; 99#endif 100 101#ifndef MIN 102# define MIN(a,b) (((a)<(b))?(a):(b)) 103#endif 104 105#define TCP_CLOSE (TH_FIN|TH_RST) 106 107static ipstate_t **ips_table = 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)); 128static int fr_tcpoptions __P((tcphdr_t *)); 129int fr_stputent __P((caddr_t)); 130int fr_stgetent __P((caddr_t)); 131void fr_stinsert __P((ipstate_t *)); 132 133 134#define FIVE_DAYS (2 * 5 * 86400) /* 5 days: half closed session */ 135 136#define TCP_MSL 240 /* 2 minutes */ 137u_long fr_tcpidletimeout = FIVE_DAYS, 138 fr_tcpclosewait = 2 * TCP_MSL, 139 fr_tcplastack = 2 * TCP_MSL, 140 fr_tcptimeout = 2 * TCP_MSL, 141 fr_tcpclosed = 120, 142 fr_tcphalfclosed = 2 * 2 * 3600, /* 2 hours */ 143 fr_udptimeout = 240, 144 fr_udpacktimeout = 24, 145 fr_icmptimeout = 120, 146 fr_icmpacktimeout = 12; 147int fr_statemax = IPSTATE_MAX, 148 fr_statesize = IPSTATE_SIZE; 149int fr_state_doflush = 0, 150 fr_state_lock = 0; 151ipstate_t *ips_list = NULL; 152 153static int icmpreplytype4[ICMP_MAXTYPE + 1]; 154#ifdef USE_INET6 155static int icmpreplytype6[ICMP6_MAXTYPE + 1]; 156#endif 157 158int fr_stateinit() 159{ 160 int i; 161 162 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *)); 163 if (ips_table != NULL) 164 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *)); 165 else 166 return -1; 167 168 /* fill icmp reply type table */ 169 for (i = 0; i <= ICMP_MAXTYPE; i++) 170 icmpreplytype4[i] = -1; 171 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 172 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 173 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 174 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 175#ifdef USE_INET6 176 /* fill icmp reply type table */ 177 for (i = 0; i <= ICMP6_MAXTYPE; i++) 178 icmpreplytype6[i] = -1; 179 icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY; 180 icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT; 181 icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY; 182 icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT; 183 icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT; 184#endif 185 186 return 0; 187} 188 189 190static ips_stat_t *fr_statetstats() 191{ 192 ips_stats.iss_active = ips_num; 193 ips_stats.iss_table = ips_table; 194 ips_stats.iss_list = ips_list; 195 return &ips_stats; 196} 197 198 199/* 200 * flush state tables. two actions currently defined: 201 * which == 0 : flush all state table entries 202 * which == 1 : flush TCP connections which have started to close but are 203 * stuck for some reason. 204 * which == 2 : flush TCP connections which have been idle for a long time, 205 * starting at > 4 days idle and working back in successive half- 206 * days to at most 12 hours old. 207 */ 208static int fr_state_flush(which) 209int which; 210{ 211 ipstate_t *is, **isp; 212#if defined(_KERNEL) && !SOLARIS 213 int s; 214#endif 215 int delete, removed = 0, try; 216 217 SPL_NET(s); 218 for (isp = &ips_list; (is = *isp); ) { 219 delete = 0; 220 221 switch (which) 222 { 223 case 0 : 224 delete = 1; 225 break; 226 case 1 : 227 case 2 : 228 if (is->is_p != IPPROTO_TCP) 229 break; 230 if ((is->is_state[0] != TCPS_ESTABLISHED) || 231 (is->is_state[1] != TCPS_ESTABLISHED)) 232 delete = 1; 233 break; 234 } 235 236 if (delete) { 237 if (is->is_p == IPPROTO_TCP) 238 ips_stats.iss_fin++; 239 else 240 ips_stats.iss_expire++; 241#ifdef IPFILTER_LOG 242 ipstate_log(is, ISL_FLUSH); 243#endif 244 fr_delstate(is); 245 removed++; 246 } else 247 isp = &is->is_next; 248 } 249 250 /* 251 * Asked to remove inactive entries, try again if first attempt 252 * failed. In this case, 86400 is half a day because the counter is 253 * activated every half second. 254 */ 255 if ((which == 2) && (removed == 0)) { 256 try = 86400; /* half a day */ 257 for (; (try < FIVE_DAYS) && (removed == 0); try += 86400) { 258 for (isp = &ips_list; (is = *isp); ) { 259 delete = 0; 260 if ((is->is_p == IPPROTO_TCP) && 261 ((is->is_state[0] == TCPS_ESTABLISHED) || 262 (is->is_state[1] == TCPS_ESTABLISHED)) && 263 (is->is_age < try)) { 264 ips_stats.iss_fin++; 265 delete = 1; 266 } else if ((is->is_p != IPPROTO_TCP) && 267 (is->is_pkts > 1)) { 268 ips_stats.iss_expire++; 269 delete = 1; 270 } 271 if (delete) { 272#ifdef IPFILTER_LOG 273 ipstate_log(is, ISL_FLUSH); 274#endif 275 fr_delstate(is); 276 removed++; 277 } else 278 isp = &is->is_next; 279 } 280 } 281 } 282 283 SPL_X(s); 284 return removed; 285} 286 287 288static int fr_state_remove(data) 289caddr_t data; 290{ 291 ipstate_t *sp, st; 292 int error; 293 294 sp = &st; 295 error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st)); 296 if (error) 297 return EFAULT; 298 299 WRITE_ENTER(&ipf_state); 300 for (sp = ips_list; sp; sp = sp->is_next) 301 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 302 !bcmp((char *)&sp->is_src, (char *)&st.is_src, 303 sizeof(st.is_src)) && 304 !bcmp((char *)&sp->is_dst, (char *)&st.is_dst, 305 sizeof(st.is_dst)) && 306 !bcmp((char *)&sp->is_ps, (char *)&st.is_ps, 307 sizeof(st.is_ps))) { 308#ifdef IPFILTER_LOG 309 ipstate_log(sp, ISL_REMOVE); 310#endif 311 fr_delstate(sp); 312 RWLOCK_EXIT(&ipf_state); 313 return 0; 314 } 315 RWLOCK_EXIT(&ipf_state); 316 return ESRCH; 317} 318 319 320int fr_state_ioctl(data, cmd, mode) 321caddr_t data; 322#if defined(__NetBSD__) || defined(__OpenBSD__) 323u_long cmd; 324#else 325int cmd; 326#endif 327int mode; 328{ 329 int arg, ret, error = 0; 330 331 switch (cmd) 332 { 333 case SIOCDELST : 334 error = fr_state_remove(data); 335 break; 336 case SIOCIPFFL : 337 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg)); 338 if (error) 339 break; 340 if (arg == 0 || arg == 1) { 341 WRITE_ENTER(&ipf_state); 342 ret = fr_state_flush(arg); 343 RWLOCK_EXIT(&ipf_state); 344 error = IWCOPY((caddr_t)&ret, data, sizeof(ret)); 345 } else 346 error = EINVAL; 347 break; 348#ifdef IPFILTER_LOG 349 case SIOCIPFFB : 350 if (!(mode & FWRITE)) 351 error = EPERM; 352 else { 353 int tmp; 354 355 tmp = ipflog_clear(IPL_LOGSTATE); 356 IWCOPY((char *)&tmp, data, sizeof(tmp)); 357 } 358 break; 359#endif 360 case SIOCGETFS : 361 error = IWCOPYPTR((caddr_t)fr_statetstats(), data, 362 sizeof(ips_stat_t)); 363 break; 364 case FIONREAD : 365#ifdef IPFILTER_LOG 366 arg = (int)iplused[IPL_LOGSTATE]; 367 error = IWCOPY((caddr_t)&arg, (caddr_t)data, sizeof(arg)); 368#endif 369 break; 370 case SIOCSTLCK : 371 error = fr_lock(data, &fr_state_lock); 372 break; 373 case SIOCSTPUT : 374 if (!fr_state_lock) { 375 error = EACCES; 376 break; 377 } 378 error = fr_stputent(data); 379 break; 380 case SIOCSTGET : 381 if (!fr_state_lock) { 382 error = EACCES; 383 break; 384 } 385 error = fr_stgetent(data); 386 break; 387 default : 388 error = EINVAL; 389 break; 390 } 391 return error; 392} 393 394 395/* 396 * Copy out state information from the kernel to a user space process. 397 */ 398int fr_stgetent(data) 399caddr_t data; 400{ 401 register ipstate_t *is, *isn; 402 ipstate_save_t ips; 403 int error; 404 405 error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips)); 406 if (error) 407 return error; 408 409 isn = ips.ips_next; 410 if (!isn) { 411 isn = ips_list; 412 if (isn == NULL) { 413 if (ips.ips_next == NULL) 414 return ENOENT; 415 return 0; 416 } 417 } else { 418 /* 419 * Make sure the pointer we're copying from exists in the 420 * current list of entries. Security precaution to prevent 421 * copying of random kernel data. 422 */ 423 for (is = ips_list; is; is = is->is_next) 424 if (is == isn) 425 break; 426 if (!is) 427 return ESRCH; 428 } 429 ips.ips_next = isn->is_next; 430 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 431 if (isn->is_rule) 432 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 433 sizeof(ips.ips_fr)); 434 error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips)); 435 if (error) 436 error = EFAULT; 437 return error; 438} 439 440 441int fr_stputent(data) 442caddr_t data; 443{ 444 register ipstate_t *is, *isn; 445 ipstate_save_t ips; 446 int error, out, i; 447 frentry_t *fr; 448 char *name; 449 450 error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips)); 451 if (error) 452 return error; 453 454 KMALLOC(isn, ipstate_t *); 455 if (isn == NULL) 456 return ENOMEM; 457 458 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 459 fr = isn->is_rule; 460 if (fr != NULL) { 461 if (isn->is_flags & FI_NEWFR) { 462 KMALLOC(fr, frentry_t *); 463 if (fr == NULL) { 464 KFREE(isn); 465 return ENOMEM; 466 } 467 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 468 out = fr->fr_flags & FR_OUTQUE ? 1 : 0; 469 isn->is_rule = fr; 470 ips.ips_is.is_rule = fr; 471 472 /* 473 * Look up all the interface names in the rule. 474 */ 475 for (i = 0; i < 4; i++) { 476 name = fr->fr_ifnames[i]; 477 if ((name[1] == '\0') && 478 ((name[0] == '-') || (name[0] == '*'))) { 479 fr->fr_ifas[i] = NULL; 480 } else if (*name != '\0') { 481 fr->fr_ifas[i] = GETUNIT(name, 482 fr->fr_v); 483 if (fr->fr_ifas[i] == NULL) 484 fr->fr_ifas[i] = (void *)-1; 485 else { 486 strncpy(isn->is_ifname[i], 487 IFNAME(fr->fr_ifas[i]), 488 IFNAMSIZ); 489 } 490 } 491 isn->is_ifp[out] = fr->fr_ifas[i]; 492 } 493 494 /* 495 * send a copy back to userland of what we ended up 496 * to allow for verification. 497 */ 498 error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips)); 499 if (error) { 500 KFREE(isn); 501 KFREE(fr); 502 return EFAULT; 503 } 504 } else { 505 for (is = ips_list; is; is = is->is_next) 506 if (is->is_rule == fr) 507 break; 508 if (!is) { 509 KFREE(isn); 510 return ESRCH; 511 } 512 } 513 } 514 fr_stinsert(isn); 515 return 0; 516} 517 518 519/* 520 * Insert a state table entry manually. 521 */ 522void fr_stinsert(is) 523register ipstate_t *is; 524{ 525 register u_int hv = is->is_hv; 526 char *name; 527 int i; 528 529 MUTEX_INIT(&is->is_lock, "ipf state entry", NULL); 530 531 /* 532 * Look up all the interface names in the state entry. 533 */ 534 for (i = 0; i < 4; i++) { 535 name = is->is_ifname[i]; 536 if ((name[1] == '\0') && 537 ((name[0] == '-') || (name[0] == '*'))) { 538 is->is_ifp[0] = NULL; 539 } else if (*name != '\0') { 540 is->is_ifp[i] = GETUNIT(name, is->is_v); 541 if (is->is_ifp[i] == NULL) 542 is->is_ifp[i] = (void *)-1; 543 } 544 } 545 546 547 /* 548 * add into list table. 549 */ 550 if (ips_list) 551 ips_list->is_pnext = &is->is_next; 552 is->is_pnext = &ips_list; 553 is->is_next = ips_list; 554 ips_list = is; 555 if (ips_table[hv]) 556 ips_table[hv]->is_phnext = &is->is_hnext; 557 else 558 ips_stats.iss_inuse++; 559 is->is_phnext = ips_table + hv; 560 is->is_hnext = ips_table[hv]; 561 ips_table[hv] = is; 562 ips_num++; 563} 564 565 566/* 567 * Create a new ipstate structure and hang it off the hash table. 568 */ 569ipstate_t *fr_addstate(ip, fin, stsave, flags) 570ip_t *ip; 571fr_info_t *fin; 572ipstate_t **stsave; 573u_int flags; 574{ 575 register tcphdr_t *tcp = NULL; 576 register ipstate_t *is; 577 register u_int hv; 578 struct icmp *ic; 579 ipstate_t ips; 580 int out, ws; 581 u_int pass; 582 void *ifp; 583 584 if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT) || 585 (fin->fin_misc & FM_BADSTATE)) 586 return NULL; 587 if (ips_num == fr_statemax) { 588 ips_stats.iss_max++; 589 fr_state_doflush = 1; 590 return NULL; 591 } 592 out = fin->fin_out; 593 is = &ips; 594 bzero((char *)is, sizeof(*is)); 595 ips.is_age = 1; 596 /* 597 * Copy and calculate... 598 */ 599 hv = (is->is_p = fin->fin_fi.fi_p); 600 is->is_src = fin->fin_fi.fi_src; 601 hv += is->is_saddr; 602 is->is_dst = fin->fin_fi.fi_dst; 603 hv += is->is_daddr; 604#ifdef USE_INET6 605 if (fin->fin_v == 6) { 606 if ((is->is_p == IPPROTO_ICMPV6) && 607 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) { 608 /* 609 * So you can do keep state with neighbour discovery. 610 */ 611 flags |= FI_W_DADDR; 612 hv -= is->is_daddr; 613 } else { 614 hv += is->is_dst.i6[1]; 615 hv += is->is_dst.i6[2]; 616 hv += is->is_dst.i6[3]; 617 } 618 hv += is->is_src.i6[1]; 619 hv += is->is_src.i6[2]; 620 hv += is->is_src.i6[3]; 621 } 622#endif 623 624 switch (is->is_p) 625 { 626 int off; 627 628#ifdef USE_INET6 629 case IPPROTO_ICMPV6 : 630 ic = (struct icmp *)fin->fin_dp; 631 if ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0) 632 return NULL; 633 634 switch (ic->icmp_type) 635 { 636 case ICMP6_ECHO_REQUEST : 637 is->is_icmp.ics_type = ic->icmp_type; 638 hv += (is->is_icmp.ics_id = ic->icmp_id); 639 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 640 break; 641 case ICMP6_MEMBERSHIP_QUERY : 642 case ND_ROUTER_SOLICIT : 643 case ND_NEIGHBOR_SOLICIT : 644 case ICMP6_NI_QUERY : 645 is->is_icmp.ics_type = ic->icmp_type; 646 break; 647 default : 648 return NULL; 649 } 650 ATOMIC_INCL(ips_stats.iss_icmp); 651 is->is_age = fr_icmptimeout; 652 break; 653#endif 654 case IPPROTO_ICMP : 655 ic = (struct icmp *)fin->fin_dp; 656 657 switch (ic->icmp_type) 658 { 659 case ICMP_ECHO : 660 case ICMP_TSTAMP : 661 case ICMP_IREQ : 662 case ICMP_MASKREQ : 663 is->is_icmp.ics_type = ic->icmp_type; 664 hv += (is->is_icmp.ics_id = ic->icmp_id); 665 hv += (is->is_icmp.ics_seq = ic->icmp_seq); 666 break; 667 default : 668 return NULL; 669 } 670 ATOMIC_INCL(ips_stats.iss_icmp); 671 is->is_age = fr_icmptimeout; 672 break; 673 case IPPROTO_TCP : 674 tcp = (tcphdr_t *)fin->fin_dp; 675 676 if (tcp->th_flags & TH_RST) 677 return NULL; 678 /* 679 * The endian of the ports doesn't matter, but the ack and 680 * sequence numbers do as we do mathematics on them later. 681 */ 682 is->is_sport = htons(fin->fin_data[0]); 683 is->is_dport = htons(fin->fin_data[1]); 684 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 685 hv += is->is_sport; 686 hv += is->is_dport; 687 } 688 if ((flags & FI_IGNOREPKT) == 0) { 689 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 690 (off = (tcp->th_off << 2)) + 691 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 692 ((tcp->th_flags & TH_FIN) ? 1 : 0); 693 is->is_maxsend = is->is_send; 694 695 if ((tcp->th_flags & TH_SYN) && 696 ((tcp->th_off << 2) >= (sizeof(*tcp) + 4))) { 697 ws = fr_tcpoptions(tcp); 698 if (ws >= 0) 699 is->is_swscale = ws; 700 } 701 } 702 703 is->is_maxdwin = 1; 704 is->is_maxswin = ntohs(tcp->th_win); 705 if (is->is_maxswin == 0) 706 is->is_maxswin = 1; 707 708 if ((tcp->th_flags & TH_OPENING) == TH_SYN) 709 is->is_fsm = 1; 710 711 /* 712 * If we're creating state for a starting connection, start the 713 * timer on it as we'll never see an error if it fails to 714 * connect. 715 */ 716 ATOMIC_INCL(ips_stats.iss_tcp); 717 break; 718 719 case IPPROTO_UDP : 720 tcp = (tcphdr_t *)fin->fin_dp; 721 722 is->is_sport = htons(fin->fin_data[0]); 723 is->is_dport = htons(fin->fin_data[1]); 724 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) { 725 hv += is->is_sport; 726 hv += is->is_dport; 727 } 728 ATOMIC_INCL(ips_stats.iss_udp); 729 is->is_age = fr_udptimeout; 730 break; 731 default : 732 is->is_age = fr_udptimeout; 733 break; 734 } 735 736 KMALLOC(is, ipstate_t *); 737 if (is == NULL) { 738 ATOMIC_INCL(ips_stats.iss_nomem); 739 return NULL; 740 } 741 bcopy((char *)&ips, (char *)is, sizeof(*is)); 742 hv %= fr_statesize; 743 is->is_hv = hv; 744 is->is_rule = fin->fin_fr; 745 if (is->is_rule != NULL) { 746 ATOMIC_INC32(is->is_rule->fr_ref); 747 pass = is->is_rule->fr_flags; 748 is->is_frage[0] = is->is_rule->fr_age[0]; 749 is->is_frage[1] = is->is_rule->fr_age[1]; 750 if (is->is_frage[0] != 0) 751 is->is_age = is->is_frage[0]; 752 753 is->is_ifp[(out << 1) + 1] = is->is_rule->fr_ifas[1]; 754 is->is_ifp[(1 - out) << 1] = is->is_rule->fr_ifas[2]; 755 is->is_ifp[((1 - out) << 1) + 1] = is->is_rule->fr_ifas[3]; 756 757 if (((ifp = is->is_rule->fr_ifas[1]) != NULL) && 758 (ifp != (void *)-1)) 759 strncpy(is->is_ifname[(out << 1) + 1], 760 IFNAME(ifp), IFNAMSIZ); 761 if (((ifp = is->is_rule->fr_ifas[2]) != NULL) && 762 (ifp != (void *)-1)) 763 strncpy(is->is_ifname[(1 - out) << 1], 764 IFNAME(ifp), IFNAMSIZ); 765 if (((ifp = is->is_rule->fr_ifas[3]) != NULL) && 766 (ifp != (void *)-1)) 767 strncpy(is->is_ifname[((1 - out) << 1) + 1], 768 IFNAME(ifp), IFNAMSIZ); 769 } else 770 pass = fr_flags; 771 772 is->is_ifp[out << 1] = fin->fin_ifp; 773 strncpy(is->is_ifname[out << 1], IFNAME(fin->fin_ifp), IFNAMSIZ); 774 775 WRITE_ENTER(&ipf_state); 776 777 is->is_pass = pass; 778 if ((flags & FI_IGNOREPKT) == 0) { 779 is->is_pkts = 1; 780 is->is_bytes = fin->fin_dlen + fin->fin_hlen; 781 } 782 /* 783 * We want to check everything that is a property of this packet, 784 * but we don't (automatically) care about it's fragment status as 785 * this may change. 786 */ 787 is->is_v = fin->fin_v; 788 is->is_rulen = fin->fin_rule; 789 is->is_opt = fin->fin_fi.fi_optmsk; 790 is->is_optmsk = 0xffffffff; 791 is->is_sec = fin->fin_fi.fi_secmsk; 792 is->is_secmsk = 0xffff; 793 is->is_auth = fin->fin_fi.fi_auth; 794 is->is_authmsk = 0xffff; 795 is->is_flags = fin->fin_fl & FI_CMP; 796 is->is_flags |= FI_CMP << 4; 797 is->is_flags |= flags & (FI_WILDP|FI_WILDA); 798 if (flags & (FI_WILDP|FI_WILDA)) 799 ips_wild++; 800 801 if (pass & FR_LOGFIRST) 802 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 803 fr_stinsert(is); 804 is->is_me = stsave; 805 if (is->is_p == IPPROTO_TCP) { 806 fr_tcp_age(&is->is_age, is->is_state, fin, 807 0, is->is_fsm); /* 0 = packet from the source */ 808 } 809#ifdef IPFILTER_LOG 810 ipstate_log(is, ISL_NEW); 811#endif 812 RWLOCK_EXIT(&ipf_state); 813 fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst); 814 if ((fin->fin_fi.fi_fl & FI_FRAG) && (pass & FR_KEEPFRAG)) 815 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 816 return is; 817} 818 819 820static int fr_tcpoptions(tcp) 821tcphdr_t *tcp; 822{ 823 u_char *opt, *last; 824 int wscale; 825 826 opt = (u_char *) (tcp + 1); 827 last = ((u_char *)tcp) + (tcp->th_off << 2); 828 829 /* If we don't find wscale here, we need to clear it */ 830 wscale = -2; 831 832 /* Termination condition picked such that opt[0 .. 2] exist */ 833 while ((opt < last - 2) && (*opt != TCPOPT_EOL)) { 834 switch (*opt) { 835 case TCPOPT_NOP: 836 opt++; 837 continue; 838 case TCPOPT_WSCALE: 839 /* Proper length ? */ 840 if (opt[1] == 3) { 841 if (opt[2] > 14) 842 wscale = 14; 843 else 844 wscale = opt[2]; 845 } 846 break; 847 default: 848 /* Unknown options must be two bytes+ */ 849 if (opt[1] < 2) 850 break; 851 opt += opt[1]; 852 continue; 853 } 854 break; 855 } 856 return wscale; 857} 858 859 860 861/* 862 * check to see if a packet with TCP headers fits within the TCP window. 863 * change timeout depending on whether new packet is a SYN-ACK returning for a 864 * SYN or a RST or FIN which indicate time to close up shop. 865 */ 866int fr_tcpstate(is, fin, ip, tcp) 867register ipstate_t *is; 868fr_info_t *fin; 869ip_t *ip; 870tcphdr_t *tcp; 871{ 872 register tcp_seq seq, ack, end; 873 register int ackskew; 874 tcpdata_t *fdata, *tdata; 875 u_32_t win, maxwin; 876 int ret = 0, off; 877 int source; 878 int wscale; 879 880 /* 881 * Find difference between last checked packet and this packet. 882 */ 883 source = IP6EQ(fin->fin_fi.fi_src, is->is_src); 884 if (source && (ntohs(is->is_sport) != fin->fin_data[0])) 885 source = 0; 886 fdata = &is->is_tcp.ts_data[!source]; 887 tdata = &is->is_tcp.ts_data[source]; 888 off = tcp->th_off << 2; 889 seq = ntohl(tcp->th_seq); 890 ack = ntohl(tcp->th_ack); 891 win = ntohs(tcp->th_win); 892 end = seq + fin->fin_dlen - off + 893 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 894 ((tcp->th_flags & TH_FIN) ? 1 : 0); 895 896 897 if ((tcp->th_flags & TH_SYN) && (off >= sizeof(*tcp) + 4)) 898 wscale = fr_tcpoptions(tcp); 899 else 900 wscale = -1; 901 902 MUTEX_ENTER(&is->is_lock); 903 904 if (wscale >= 0) 905 fdata->td_wscale = wscale; 906 else if (wscale == -2) 907 fdata->td_wscale = tdata->td_wscale = 0; 908 win <<= fdata->td_wscale; 909 910 if ((fdata->td_end == 0) && 911 (!is->is_fsm || ((tcp->th_flags & TH_OPENING) == TH_OPENING))) { 912 /* 913 * Must be a (outgoing) SYN-ACK in reply to a SYN. 914 */ 915 fdata->td_end = end; 916 fdata->td_maxwin = 1; 917 fdata->td_maxend = end + win; 918 if (win == 0) 919 fdata->td_maxend++; 920 } 921 922 if (!(tcp->th_flags & TH_ACK)) { /* Pretend an ack was sent */ 923 ack = tdata->td_end; 924 } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 925 (ack == 0)) { 926 /* gross hack to get around certain broken tcp stacks */ 927 ack = tdata->td_end; 928 } 929 930 if (seq == end) 931 seq = end = fdata->td_end; 932 933 maxwin = tdata->td_maxwin; 934 ackskew = tdata->td_end - ack; 935 936#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 937#define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 938 if ((SEQ_GE(fdata->td_maxend, end)) && 939 (SEQ_GE(seq, fdata->td_end - maxwin)) && 940/* XXX what about big packets */ 941#define MAXACKWINDOW 66000 942 (ackskew >= -MAXACKWINDOW) && 943 (ackskew <= MAXACKWINDOW)) { 944 /* if ackskew < 0 then this should be due to fragented 945 * packets. There is no way to know the length of the 946 * total packet in advance. 947 * We do know the total length from the fragment cache though. 948 * Note however that there might be more sessions with 949 * exactly the same source and destination paramters in the 950 * state cache (and source and destination is the only stuff 951 * that is saved in the fragment cache). Note further that 952 * some TCP connections in the state cache are hashed with 953 * sport and dport as well which makes it not worthwhile to 954 * look for them. 955 * Thus, when ackskew is negative but still seems to belong 956 * to this session, we bump up the destinations end value. 957 */ 958 /* 959 * Nearing end of connection, start timeout. 960 */ 961 /* source ? 0 : 1 -> !source */ 962 if (fr_tcp_age(&is->is_age, is->is_state, fin, !source, 963 (int)is->is_fsm) == 0) { 964 if (ackskew < 0) 965 tdata->td_end = ack; 966 967 /* update max window seen */ 968 if (fdata->td_maxwin < win) 969 fdata->td_maxwin = win; 970 if (SEQ_GT(end, fdata->td_end)) 971 fdata->td_end = end; 972 if (SEQ_GE(ack + win, tdata->td_maxend)) { 973 tdata->td_maxend = ack + win; 974 if (win == 0) 975 tdata->td_maxend++; 976 } 977 978 ATOMIC_INCL(ips_stats.iss_hits); 979 ret = 1; 980 } 981 } 982 MUTEX_EXIT(&is->is_lock); 983 if ((ret == 0) && ((tcp->th_flags & TH_OPENING) != TH_SYN)) 984 fin->fin_misc |= FM_BADSTATE; 985 return ret; 986} 987 988 989/* 990 * Match a state table entry against an IP packet. 991 */ 992static int fr_matchsrcdst(is, src, dst, fin, tcp) 993ipstate_t *is; 994union i6addr src, dst; 995fr_info_t *fin; 996tcphdr_t *tcp; 997{ 998 int ret = 0, rev, out, flags, idx; 999 u_short sp, dp; 1000 void *ifp; 1001 1002 rev = IP6NEQ(is->is_dst, dst); 1003 ifp = fin->fin_ifp; 1004 out = fin->fin_out; 1005 flags = is->is_flags & (FI_WILDA|FI_WILDP); 1006 sp = 0; 1007 dp = 0; 1008 1009 if (tcp != NULL) { 1010 flags = is->is_flags; 1011 sp = tcp->th_sport; 1012 dp = tcp->th_dport; 1013 if (!rev) { 1014 if (!(flags & FI_W_SPORT) && (sp != is->is_sport)) 1015 rev = 1; 1016 else if (!(flags & FI_W_DPORT) && (dp != is->is_dport)) 1017 rev = 1; 1018 } 1019 } 1020 1021 idx = (out << 1) + rev; 1022 1023 if ((is->is_ifp[idx] == NULL && 1024 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) || 1025 is->is_ifp[idx] == ifp) 1026 ret = 1; 1027 1028 if (ret == 0) 1029 return 0; 1030 ret = 0; 1031 1032 if (rev == 0) { 1033 if ((IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) && 1034 (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) && 1035 (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) && 1036 (dp == is->is_dport || flags & FI_W_DPORT)))) { 1037 ret = 1; 1038 } 1039 } else { 1040 if ((IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) && 1041 (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) && 1042 (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) && 1043 (dp == is->is_sport || flags & FI_W_SPORT)))) { 1044 ret = 1; 1045 } 1046 } 1047 if (ret == 0) 1048 return 0; 1049 1050 /* 1051 * Whether or not this should be here, is questionable, but the aim 1052 * is to get this out of the main line. 1053 */ 1054 if (tcp == NULL) 1055 flags = is->is_flags & (FI_CMP|(FI_CMP<<4)); 1056 1057 if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) || 1058 (fin->fin_fi.fi_optmsk != is->is_opt) || 1059 (fin->fin_fi.fi_secmsk != is->is_sec) || 1060 (fin->fin_fi.fi_auth != is->is_auth)) 1061 return 0; 1062 1063 flags = is->is_flags & (FI_WILDA|FI_WILDP); 1064 if ((flags & (FI_W_SADDR|FI_W_DADDR))) { 1065 if ((flags & FI_W_SADDR) != 0) { 1066 if (rev == 0) { 1067 is->is_src = fin->fin_fi.fi_src; 1068 } else { 1069 is->is_src = fin->fin_fi.fi_dst; 1070 } 1071 } else if ((flags & FI_W_DPORT) != 0) { 1072 if (rev == 0) { 1073 is->is_dst = fin->fin_fi.fi_dst; 1074 } else { 1075 is->is_dst = fin->fin_fi.fi_src; 1076 } 1077 } 1078 is->is_flags &= ~(FI_W_SADDR|FI_W_DADDR); 1079 if ((is->is_flags & (FI_WILDA|FI_WILDP)) == 0) 1080 ips_wild--; 1081 } 1082 1083 if ((flags & (FI_W_SPORT|FI_W_DPORT))) { 1084 if ((flags & FI_W_SPORT) != 0) { 1085 if (rev == 0) { 1086 is->is_sport = sp; 1087 is->is_send = htonl(tcp->th_seq); 1088 } else { 1089 is->is_sport = dp; 1090 is->is_send = htonl(tcp->th_ack); 1091 } 1092 is->is_maxsend = is->is_send + 1; 1093 } else if ((flags & FI_W_DPORT) != 0) { 1094 if (rev == 0) { 1095 is->is_dport = dp; 1096 is->is_dend = htonl(tcp->th_ack); 1097 } else { 1098 is->is_dport = sp; 1099 is->is_dend = htonl(tcp->th_seq); 1100 } 1101 is->is_maxdend = is->is_dend + 1; 1102 } 1103 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT); 1104 ips_wild--; 1105 } 1106 1107 ret = -1; 1108 1109 if (is->is_ifp[idx] == NULL && 1110 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) 1111 ret = idx; 1112 1113 if (ret >= 0) { 1114 is->is_ifp[ret] = ifp; 1115 strncpy(is->is_ifname[ret], IFNAME(ifp), 1116 sizeof(is->is_ifname[ret])); 1117 } 1118 fin->fin_rev = rev; 1119 return 1; 1120} 1121 1122static int fr_matchicmpqueryreply(v, is, icmp) 1123int v; 1124ipstate_t *is; 1125icmphdr_t *icmp; 1126{ 1127 if (v == 4) { 1128 /* 1129 * If we matched its type on the way in, then when going out 1130 * it will still be the same type. 1131 */ 1132 if (((icmp->icmp_type == is->is_type) || 1133 (icmpreplytype4[is->is_type] == icmp->icmp_type))) { 1134 if (icmp->icmp_type != ICMP_ECHOREPLY) 1135 return 1; 1136 if ((icmp->icmp_id == is->is_icmp.ics_id) && 1137 (icmp->icmp_seq == is->is_icmp.ics_seq)) 1138 return 1; 1139 } 1140 } 1141#ifdef USE_INET6 1142 else if (is->is_v == 6) { 1143 if (((icmp->icmp_type == is->is_type) || 1144 (icmpreplytype6[is->is_type] == icmp->icmp_type))) { 1145 if (icmp->icmp_type != ICMP6_ECHO_REPLY) 1146 return 1; 1147 if ((icmp->icmp_id == is->is_icmp.ics_id) && 1148 (icmp->icmp_seq == is->is_icmp.ics_seq)) 1149 return 1; 1150 } 1151 } 1152#endif 1153 return 0; 1154} 1155 1156static frentry_t *fr_checkicmpmatchingstate(ip, fin) 1157ip_t *ip; 1158fr_info_t *fin; 1159{ 1160 register ipstate_t *is, **isp; 1161 register u_short sport, dport; 1162 register u_char pr; 1163 u_short savelen, ohlen; 1164 union i6addr dst, src; 1165 struct icmp *ic; 1166 icmphdr_t *icmp; 1167 fr_info_t ofin; 1168 int type, len; 1169 tcphdr_t *tcp; 1170 frentry_t *fr; 1171 ip_t *oip; 1172 u_int hv; 1173 1174 /* 1175 * Does it at least have the return (basic) IP header ? 1176 * Only a basic IP header (no options) should be with 1177 * an ICMP error header. 1178 */ 1179 if (((ip->ip_v != 4) || (ip->ip_hl != 5)) || 1180 (fin->fin_plen < ICMPERR_MINPKTLEN)) 1181 return NULL; 1182 1183 ic = (struct icmp *)fin->fin_dp; 1184 type = ic->icmp_type; 1185 /* 1186 * If it's not an error type, then return 1187 */ 1188 if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) && 1189 (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) && 1190 (type != ICMP_PARAMPROB)) 1191 return NULL; 1192 1193 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 1194 ohlen = oip->ip_hl << 2; 1195 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ohlen - sizeof(*oip)) 1196 return NULL; 1197 1198 /* 1199 * Sanity checks. 1200 */ 1201 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 1202 if ((len <= 0) || (ohlen > len)) 1203 return NULL; 1204 1205 /* 1206 * Is the buffer big enough for all of it ? It's the size of the IP 1207 * header claimed in the encapsulated part which is of concern. It 1208 * may be too big to be in this buffer but not so big that it's 1209 * outside the ICMP packet, leading to TCP deref's causing problems. 1210 * This is possible because we don't know how big oip_hl is when we 1211 * do the pullup early in fr_check() and thus can't gaurantee it is 1212 * all here now. 1213 */ 1214#ifdef _KERNEL 1215 { 1216 mb_t *m; 1217 1218# if SOLARIS 1219 m = fin->fin_qfm; 1220 if ((char *)oip + len > (char *)m->b_wptr) 1221 return NULL; 1222# else 1223 m = *(mb_t **)fin->fin_mp; 1224 if ((char *)oip + len > (char *)ip + m->m_len) 1225 return NULL; 1226# endif 1227 } 1228#endif 1229 1230 /* 1231 * in the IPv4 case we must zero the i6addr union otherwise 1232 * the IP6EQ and IP6NEQ macros produce the wrong results because 1233 * of the 'junk' in the unused part of the union 1234 */ 1235 bzero((char *)&src, sizeof(src)); 1236 bzero((char *)&dst, sizeof(dst)); 1237 bzero((char *)&ofin, sizeof(ofin)); 1238 ofin.fin_ifp = fin->fin_ifp; 1239 ofin.fin_out = !fin->fin_out; 1240 ofin.fin_v = 4; 1241 fr = NULL; 1242 1243 switch (oip->ip_p) 1244 { 1245 case IPPROTO_ICMP : 1246 icmp = (icmphdr_t *)((char *)oip + ohlen); 1247 1248 /* 1249 * a ICMP error can only be generated as a result of an 1250 * ICMP query, not as the response on an ICMP error 1251 * 1252 * XXX theoretically ICMP_ECHOREP and the other reply's are 1253 * ICMP query's as well, but adding them here seems strange XXX 1254 */ 1255 if ((icmp->icmp_type != ICMP_ECHO) && 1256 (icmp->icmp_type != ICMP_TSTAMP) && 1257 (icmp->icmp_type != ICMP_IREQ) && 1258 (icmp->icmp_type != ICMP_MASKREQ)) 1259 return NULL; 1260 1261 /* 1262 * perform a lookup of the ICMP packet in the state table 1263 */ 1264 hv = (pr = oip->ip_p); 1265 src.in4 = oip->ip_src; 1266 hv += src.in4.s_addr; 1267 dst.in4 = oip->ip_dst; 1268 hv += dst.in4.s_addr; 1269 hv += icmp->icmp_id; 1270 hv += icmp->icmp_seq; 1271 hv %= fr_statesize; 1272 1273 savelen = oip->ip_len; 1274 oip->ip_len = len; 1275 fr_makefrip(ohlen, oip, &ofin); 1276 oip->ip_len = savelen; 1277 1278 READ_ENTER(&ipf_state); 1279 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 1280 if ((is->is_p == pr) && (is->is_v == 4) && 1281 fr_matchsrcdst(is, src, dst, &ofin, NULL) && 1282 fr_matchicmpqueryreply(is->is_v, is, icmp)) { 1283 ips_stats.iss_hits++; 1284 is->is_pkts++; 1285 is->is_bytes += ip->ip_len; 1286 fr = is->is_rule; 1287 break; 1288 } 1289 RWLOCK_EXIT(&ipf_state); 1290 return fr; 1291 1292 case IPPROTO_TCP : 1293 case IPPROTO_UDP : 1294 if (fin->fin_plen < ICMPERR_MAXPKTLEN) 1295 return NULL; 1296 break; 1297 default : 1298 return NULL; 1299 } 1300 1301 tcp = (tcphdr_t *)((char *)oip + ohlen); 1302 dport = tcp->th_dport; 1303 sport = tcp->th_sport; 1304 1305 hv = (pr = oip->ip_p); 1306 src.in4 = oip->ip_src; 1307 hv += src.in4.s_addr; 1308 dst.in4 = oip->ip_dst; 1309 hv += dst.in4.s_addr; 1310 hv += dport; 1311 hv += sport; 1312 hv %= fr_statesize; 1313 /* 1314 * we make an fin entry to be able to feed it to 1315 * matchsrcdst note that not all fields are encessary 1316 * but this is the cleanest way. Note further we fill 1317 * in fin_mp such that if someone uses it we'll get 1318 * a kernel panic. fr_matchsrcdst does not use this. 1319 * 1320 * watch out here, as ip is in host order and oip in network 1321 * order. Any change we make must be undone afterwards. 1322 */ 1323 savelen = oip->ip_len; 1324 oip->ip_len = len; 1325 fr_makefrip(ohlen, oip, &ofin); 1326 oip->ip_len = savelen; 1327 READ_ENTER(&ipf_state); 1328 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1329 /* 1330 * Only allow this icmp though if the 1331 * encapsulated packet was allowed through the 1332 * other way around. Note that the minimal amount 1333 * of info present does not allow for checking against 1334 * tcp internals such as seq and ack numbers. 1335 */ 1336 if ((is->is_p == pr) && (is->is_v == 4) && 1337 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 1338 fr = is->is_rule; 1339 ips_stats.iss_hits++; 1340 is->is_pkts++; 1341 is->is_bytes += fin->fin_plen; 1342 /* 1343 * we deliberately do not touch the timeouts 1344 * for the accompanying state table entry. 1345 * It remains to be seen if that is correct. XXX 1346 */ 1347 break; 1348 } 1349 } 1350 RWLOCK_EXIT(&ipf_state); 1351 return fr; 1352} 1353 1354 1355/* 1356 * Move a state hash table entry from its old location at is->is_hv to 1357 * its new location, indexed by hv % fr_statesize. 1358 */ 1359static void fr_ipsmove(isp, is, hv) 1360ipstate_t **isp, *is; 1361u_int hv; 1362{ 1363 u_int hvm; 1364 1365 hvm = is->is_hv; 1366 /* 1367 * Remove the hash from the old location... 1368 */ 1369 if (is->is_hnext) 1370 is->is_hnext->is_phnext = isp; 1371 *isp = is->is_hnext; 1372 if (ips_table[hvm] == NULL) 1373 ips_stats.iss_inuse--; 1374 1375 /* 1376 * ...and put the hash in the new one. 1377 */ 1378 hvm = hv % fr_statesize; 1379 is->is_hv = hvm; 1380 isp = &ips_table[hvm]; 1381 if (*isp) 1382 (*isp)->is_phnext = &is->is_hnext; 1383 else 1384 ips_stats.iss_inuse++; 1385 is->is_phnext = isp; 1386 is->is_hnext = *isp; 1387 *isp = is; 1388} 1389 1390 1391/* 1392 * Check if a packet has a registered state. 1393 */ 1394frentry_t *fr_checkstate(ip, fin) 1395ip_t *ip; 1396fr_info_t *fin; 1397{ 1398 union i6addr dst, src; 1399 register ipstate_t *is, **isp; 1400 register u_char pr; 1401 u_int hv, hvm, hlen, tryagain, pass, v; 1402 struct icmp *ic; 1403 frentry_t *fr; 1404 tcphdr_t *tcp; 1405 int rev; 1406 1407 if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT)) 1408 return NULL; 1409 1410 is = NULL; 1411 hlen = fin->fin_hlen; 1412 tcp = (tcphdr_t *)((char *)ip + hlen); 1413 ic = (struct icmp *)tcp; 1414 hv = (pr = fin->fin_fi.fi_p); 1415 src = fin->fin_fi.fi_src; 1416 dst = fin->fin_fi.fi_dst; 1417 hv += src.in4.s_addr; 1418 hv += dst.in4.s_addr; 1419 1420 /* 1421 * Search the hash table for matching packet header info. 1422 * At the bottom of this switch statement, the following is expected: 1423 * is == NULL, no lock on ipf_state is held. 1424 * is != NULL, a lock on ipf_state is held. 1425 */ 1426 v = fin->fin_fi.fi_v; 1427#ifdef USE_INET6 1428 if (v == 6) { 1429 hv += fin->fin_fi.fi_src.i6[1]; 1430 hv += fin->fin_fi.fi_src.i6[2]; 1431 hv += fin->fin_fi.fi_src.i6[3]; 1432 1433 if ((fin->fin_p == IPPROTO_ICMPV6) && 1434 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) { 1435 hv -= dst.in4.s_addr; 1436 } else { 1437 hv += fin->fin_fi.fi_dst.i6[1]; 1438 hv += fin->fin_fi.fi_dst.i6[2]; 1439 hv += fin->fin_fi.fi_dst.i6[3]; 1440 } 1441 } 1442#endif 1443 1444 switch (fin->fin_p) 1445 { 1446#ifdef USE_INET6 1447 case IPPROTO_ICMPV6 : 1448 tcp = NULL; 1449 tryagain = 0; 1450 if (v == 6) { 1451 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 1452 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 1453 hv += ic->icmp_id; 1454 hv += ic->icmp_seq; 1455 } 1456 } 1457 READ_ENTER(&ipf_state); 1458icmp6again: 1459 hvm = hv % fr_statesize; 1460 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext) 1461 if ((is->is_p == pr) && (is->is_v == v) && 1462 fr_matchsrcdst(is, src, dst, fin, NULL) && 1463 fr_matchicmpqueryreply(v, is, ic)) { 1464 rev = fin->fin_rev; 1465 if (is->is_frage[rev] != 0) 1466 is->is_age = is->is_frage[rev]; 1467 else if (rev != 0) 1468 is->is_age = fr_icmpacktimeout; 1469 else 1470 is->is_age = fr_icmptimeout; 1471 break; 1472 } 1473 1474 if (is != NULL) { 1475 if (tryagain && !(is->is_flags & FI_W_DADDR)) { 1476 hv += fin->fin_fi.fi_src.i6[0]; 1477 hv += fin->fin_fi.fi_src.i6[1]; 1478 hv += fin->fin_fi.fi_src.i6[2]; 1479 hv += fin->fin_fi.fi_src.i6[3]; 1480 fr_ipsmove(isp, is, hv); 1481 MUTEX_DOWNGRADE(&ipf_state); 1482 } 1483 break; 1484 } 1485 RWLOCK_EXIT(&ipf_state); 1486 1487 /* 1488 * No matching icmp state entry. Perhaps this is a 1489 * response to another state entry. 1490 */ 1491 if ((ips_wild != 0) && (v == 6) && (tryagain == 0) && 1492 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) { 1493 hv -= fin->fin_fi.fi_src.i6[0]; 1494 hv -= fin->fin_fi.fi_src.i6[1]; 1495 hv -= fin->fin_fi.fi_src.i6[2]; 1496 hv -= fin->fin_fi.fi_src.i6[3]; 1497 tryagain = 1; 1498 WRITE_ENTER(&ipf_state); 1499 goto icmp6again; 1500 } 1501 1502 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin); 1503 if (fr) 1504 return fr; 1505 break; 1506#endif 1507 case IPPROTO_ICMP : 1508 tcp = NULL; 1509 if (v == 4) { 1510 hv += ic->icmp_id; 1511 hv += ic->icmp_seq; 1512 } 1513 hvm = hv % fr_statesize; 1514 READ_ENTER(&ipf_state); 1515 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext) 1516 if ((is->is_p == pr) && (is->is_v == v) && 1517 fr_matchsrcdst(is, src, dst, fin, NULL) && 1518 fr_matchicmpqueryreply(v, is, ic)) { 1519 rev = fin->fin_rev; 1520 if (is->is_frage[rev] != 0) 1521 is->is_age = is->is_frage[rev]; 1522 else if (fin->fin_rev) 1523 is->is_age = fr_icmpacktimeout; 1524 else 1525 is->is_age = fr_icmptimeout; 1526 break; 1527 } 1528 1529 if (is != NULL) 1530 break; 1531 RWLOCK_EXIT(&ipf_state); 1532 /* 1533 * No matching icmp state entry. Perhaps this is a 1534 * response to another state entry. 1535 */ 1536 fr = fr_checkicmpmatchingstate(ip, fin); 1537 if (fr) 1538 return fr; 1539 break; 1540 case IPPROTO_TCP : 1541 /* 1542 * Just plain ignore RST flag set with either FIN or SYN. 1543 */ 1544 if ((tcp->th_flags & TH_RST) && 1545 ((tcp->th_flags & (TH_FIN|TH_SYN|TH_RST)) != TH_RST)) 1546 break; 1547 case IPPROTO_UDP : 1548 { 1549 register u_short dport, sport; 1550 1551 dport = tcp->th_dport; 1552 sport = tcp->th_sport; 1553 tryagain = 0; 1554 hv += dport; 1555 hv += sport; 1556 READ_ENTER(&ipf_state); 1557retry_tcpudp: 1558 hvm = hv % fr_statesize; 1559 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext) 1560 if ((is->is_p == pr) && (is->is_v == v) && 1561 fr_matchsrcdst(is, src, dst, fin, tcp)) { 1562 rev = fin->fin_rev; 1563 if ((pr == IPPROTO_TCP)) { 1564 if (!fr_tcpstate(is, fin, ip, tcp)) 1565 is = NULL; 1566 } else if ((pr == IPPROTO_UDP)) { 1567 if (is->is_frage[rev] != 0) 1568 is->is_age = is->is_frage[rev]; 1569 else if (fin->fin_rev) 1570 is->is_age = fr_udpacktimeout; 1571 else 1572 is->is_age = fr_udptimeout; 1573 } 1574 break; 1575 } 1576 if (is != NULL) { 1577 if (tryagain && 1578 !(is->is_flags & (FI_WILDP|FI_WILDA))) { 1579 hv += dport; 1580 hv += sport; 1581 fr_ipsmove(isp, is, hv); 1582 MUTEX_DOWNGRADE(&ipf_state); 1583 } 1584 break; 1585 } 1586 1587 RWLOCK_EXIT(&ipf_state); 1588 if (!tryagain && ips_wild) { 1589 hv -= dport; 1590 hv -= sport; 1591 tryagain = 1; 1592 WRITE_ENTER(&ipf_state); 1593 goto retry_tcpudp; 1594 } 1595 break; 1596 } 1597 default : 1598 tcp = NULL; 1599 hv %= fr_statesize; 1600 READ_ENTER(&ipf_state); 1601 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 1602 if ((is->is_p == pr) && (is->is_v == v) && 1603 fr_matchsrcdst(is, src, dst, fin, NULL)) { 1604 rev = fin->fin_rev; 1605 if (is->is_frage[rev] != 0) 1606 is->is_age = is->is_frage[rev]; 1607 else 1608 is->is_age = fr_udptimeout; 1609 break; 1610 } 1611 } 1612 if (is == NULL) { 1613 RWLOCK_EXIT(&ipf_state); 1614 } 1615 break; 1616 } 1617 1618 if (is == NULL) { 1619 ATOMIC_INCL(ips_stats.iss_miss); 1620 return NULL; 1621 } 1622 1623 MUTEX_ENTER(&is->is_lock); 1624 is->is_bytes += fin->fin_plen; 1625 ips_stats.iss_hits++; 1626 is->is_pkts++; 1627 MUTEX_EXIT(&is->is_lock); 1628 fr = is->is_rule; 1629 fin->fin_rule = is->is_rulen; 1630 if (fr != NULL) { 1631 fin->fin_group = fr->fr_group; 1632 fin->fin_icode = fr->fr_icode; 1633 } 1634 fin->fin_fr = fr; 1635 pass = is->is_pass; 1636 RWLOCK_EXIT(&ipf_state); 1637 if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG)) 1638 ipfr_newfrag(ip, fin, pass ^ FR_KEEPSTATE); 1639#ifndef _KERNEL 1640 if ((tcp != NULL) && (tcp->th_flags & TCP_CLOSE)) 1641 fr_delstate(is); 1642#endif 1643 return fr; 1644} 1645 1646 1647/* 1648 * Sync. state entries. If interfaces come or go or just change position, 1649 * this is needed. 1650 */ 1651void ip_statesync(ifp) 1652void *ifp; 1653{ 1654 register ipstate_t *is; 1655 int i; 1656 1657 WRITE_ENTER(&ipf_state); 1658 for (is = ips_list; is; is = is->is_next) { 1659 for (i = 0; i < 4; i++) { 1660 if (is->is_ifp[i] == ifp) { 1661 is->is_ifpin = GETUNIT(is->is_ifname[i], 1662 is->is_v); 1663 if (!is->is_ifp[i]) 1664 is->is_ifp[i] = (void *)-1; 1665 } 1666 } 1667 } 1668 RWLOCK_EXIT(&ipf_state); 1669} 1670 1671 1672/* 1673 * Must always be called with fr_ipfstate held as a write lock. 1674 */ 1675static void fr_delstate(is) 1676ipstate_t *is; 1677{ 1678 frentry_t *fr; 1679 1680 if (is->is_flags & (FI_WILDP|FI_WILDA)) 1681 ips_wild--; 1682 if (is->is_next) 1683 is->is_next->is_pnext = is->is_pnext; 1684 *is->is_pnext = is->is_next; 1685 if (is->is_hnext) 1686 is->is_hnext->is_phnext = is->is_phnext; 1687 *is->is_phnext = is->is_hnext; 1688 if (ips_table[is->is_hv] == NULL) 1689 ips_stats.iss_inuse--; 1690 if (is->is_me) 1691 *is->is_me = NULL; 1692 1693 fr = is->is_rule; 1694 if (fr != NULL) { 1695 fr->fr_ref--; 1696 if (fr->fr_ref == 0) { 1697 KFREE(fr); 1698 } 1699 } 1700#ifdef _KERNEL 1701 MUTEX_DESTROY(&is->is_lock); 1702#endif 1703 KFREE(is); 1704 ips_num--; 1705} 1706 1707 1708/* 1709 * Free memory in use by all state info. kept. 1710 */ 1711void fr_stateunload() 1712{ 1713 register ipstate_t *is; 1714 1715 WRITE_ENTER(&ipf_state); 1716 while ((is = ips_list)) 1717 fr_delstate(is); 1718 ips_stats.iss_inuse = 0; 1719 ips_num = 0; 1720 RWLOCK_EXIT(&ipf_state); 1721 if (ips_table) 1722 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *)); 1723 ips_table = NULL; 1724} 1725 1726 1727/* 1728 * Slowly expire held state for thingslike UDP and ICMP. Timeouts are set 1729 * in expectation of this being called twice per second. 1730 */ 1731void fr_timeoutstate() 1732{ 1733 register ipstate_t *is, **isp; 1734#if defined(_KERNEL) && !SOLARIS 1735 int s; 1736#endif 1737 1738 SPL_NET(s); 1739 WRITE_ENTER(&ipf_state); 1740 for (isp = &ips_list; (is = *isp); ) 1741 if (is->is_age && !--is->is_age) { 1742 if (is->is_p == IPPROTO_TCP) 1743 ips_stats.iss_fin++; 1744 else 1745 ips_stats.iss_expire++; 1746#ifdef IPFILTER_LOG 1747 ipstate_log(is, ISL_EXPIRE); 1748#endif 1749 fr_delstate(is); 1750 } else 1751 isp = &is->is_next; 1752 if (fr_state_doflush) { 1753 (void) fr_state_flush(2); 1754 fr_state_doflush = 0; 1755 } 1756 RWLOCK_EXIT(&ipf_state); 1757 SPL_X(s); 1758} 1759 1760 1761/* 1762 * Original idea freom Pradeep Krishnan for use primarily with NAT code. 1763 * (pkrishna@netcom.com) 1764 * 1765 * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: 1766 * 1767 * - (try to) base state transitions on real evidence only, 1768 * i.e. packets that are sent and have been received by ipfilter; 1769 * diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. 1770 * 1771 * - deal with half-closed connections correctly; 1772 * 1773 * - store the state of the source in state[0] such that ipfstat 1774 * displays the state as source/dest instead of dest/source; the calls 1775 * to fr_tcp_age have been changed accordingly. 1776 * 1777 * Parameters: 1778 * 1779 * state[0] = state of source (host that initiated connection) 1780 * state[1] = state of dest (host that accepted the connection) 1781 * 1782 * dir == 0 : a packet from source to dest 1783 * dir == 1 : a packet from dest to source 1784 * 1785 */ 1786int fr_tcp_age(age, state, fin, dir, fsm) 1787u_long *age; 1788u_char *state; 1789fr_info_t *fin; 1790int dir, fsm; 1791{ 1792 tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp; 1793 u_char flags = tcp->th_flags; 1794 int dlen, ostate; 1795 u_long newage; 1796 1797 ostate = state[1 - dir]; 1798 1799 dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2); 1800 1801 if (flags & TH_RST) { 1802 if (!(tcp->th_flags & TH_PUSH) && !dlen) { 1803 *age = fr_tcpclosed; 1804 state[dir] = TCPS_CLOSED; 1805 } else { 1806 *age = fr_tcpclosewait; 1807 state[dir] = TCPS_CLOSE_WAIT; 1808 } 1809 return 0; 1810 } 1811 1812 newage = 0; 1813 1814 switch(state[dir]) 1815 { 1816 case TCPS_CLOSED: /* 0 */ 1817 if ((flags & TH_OPENING) == TH_OPENING) { 1818 /* 1819 * 'dir' received an S and sends SA in response, 1820 * CLOSED -> SYN_RECEIVED 1821 */ 1822 state[dir] = TCPS_SYN_RECEIVED; 1823 newage = fr_tcptimeout; 1824 } else if ((flags & TH_OPENING) == TH_SYN) { 1825 /* 'dir' sent S, CLOSED -> SYN_SENT */ 1826 state[dir] = TCPS_SYN_SENT; 1827 newage = fr_tcptimeout; 1828 } 1829 /* 1830 * The next piece of code makes it possible to get 1831 * already established connections into the state table 1832 * after a restart or reload of the filter rules; this 1833 * does not work when a strict 'flags S keep state' is 1834 * used for tcp connections of course 1835 */ 1836 if (!fsm && (flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) { 1837 /* we saw an A, guess 'dir' is in ESTABLISHED mode */ 1838 if (state[1 - dir] == TCPS_CLOSED || 1839 state[1 - dir] == TCPS_ESTABLISHED) { 1840 state[dir] = TCPS_ESTABLISHED; 1841 newage = fr_tcpidletimeout; 1842 } 1843 } 1844 /* 1845 * TODO: besides regular ACK packets we can have other 1846 * packets as well; it is yet to be determined how we 1847 * should initialize the states in those cases 1848 */ 1849 break; 1850 1851 case TCPS_LISTEN: /* 1 */ 1852 /* NOT USED */ 1853 break; 1854 1855 case TCPS_SYN_SENT: /* 2 */ 1856 if ((flags & ~(TH_ECN|TH_CWR)) == TH_SYN) { 1857 /* 1858 * A retransmitted SYN packet. We do not reset the 1859 * timeout here to fr_tcptimeout because a connection 1860 * connect timeout does not renew after every packet 1861 * that is sent. We need to set newage to something 1862 * to indicate the packet has passed the check for its 1863 * flags being valid in the TCP FSM. 1864 */ 1865 newage = *age; 1866 } else if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1867 /* 1868 * We see an A from 'dir' which is in SYN_SENT 1869 * state: 'dir' sent an A in response to an SA 1870 * which it received, SYN_SENT -> ESTABLISHED 1871 */ 1872 state[dir] = TCPS_ESTABLISHED; 1873 newage = fr_tcpidletimeout; 1874 } else if (flags & TH_FIN) { 1875 /* 1876 * We see an F from 'dir' which is in SYN_SENT 1877 * state and wants to close its side of the 1878 * connection; SYN_SENT -> FIN_WAIT_1 1879 */ 1880 state[dir] = TCPS_FIN_WAIT_1; 1881 newage = fr_tcpidletimeout; /* or fr_tcptimeout? */ 1882 } else if ((flags & TH_OPENING) == TH_OPENING) { 1883 /* 1884 * We see an SA from 'dir' which is already in 1885 * SYN_SENT state, this means we have a 1886 * simultaneous open; SYN_SENT -> SYN_RECEIVED 1887 */ 1888 state[dir] = TCPS_SYN_RECEIVED; 1889 newage = fr_tcptimeout; 1890 } 1891 break; 1892 1893 case TCPS_SYN_RECEIVED: /* 3 */ 1894 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 1895 /* 1896 * We see an A from 'dir' which was in SYN_RECEIVED 1897 * state so it must now be in established state, 1898 * SYN_RECEIVED -> ESTABLISHED 1899 */ 1900 state[dir] = TCPS_ESTABLISHED; 1901 newage = fr_tcpidletimeout; 1902 } else if ((flags & ~(TH_ECN|TH_CWR)) == TH_OPENING) { 1903 /* 1904 * We see an SA from 'dir' which is already in 1905 * SYN_RECEIVED state. 1906 */ 1907 newage = fr_tcptimeout; 1908 } else if (flags & TH_FIN) { 1909 /* 1910 * We see an F from 'dir' which is in SYN_RECEIVED 1911 * state and wants to close its side of the connection; 1912 * SYN_RECEIVED -> FIN_WAIT_1 1913 */ 1914 state[dir] = TCPS_FIN_WAIT_1; 1915 newage = fr_tcpidletimeout; 1916 } 1917 break; 1918 1919 case TCPS_ESTABLISHED: /* 4 */ 1920 if (flags & TH_FIN) { 1921 /* 1922 * 'dir' closed its side of the connection; this 1923 * gives us a half-closed connection; 1924 * ESTABLISHED -> FIN_WAIT_1 1925 */ 1926 state[dir] = TCPS_FIN_WAIT_1; 1927 newage = fr_tcphalfclosed; 1928 } else if (flags & TH_ACK) { 1929 /* an ACK, should we exclude other flags here? */ 1930 if (ostate == TCPS_FIN_WAIT_1) { 1931 /* 1932 * We know the other side did an active close, 1933 * so we are ACKing the recvd FIN packet (does 1934 * the window matching code guarantee this?) 1935 * and go into CLOSE_WAIT state; this gives us 1936 * a half-closed connection 1937 */ 1938 state[dir] = TCPS_CLOSE_WAIT; 1939 newage = fr_tcphalfclosed; 1940 } else if (ostate < TCPS_CLOSE_WAIT) 1941 /* 1942 * Still a fully established connection, 1943 * reset timeout 1944 */ 1945 newage = fr_tcpidletimeout; 1946 } 1947 break; 1948 1949 case TCPS_CLOSE_WAIT: /* 5 */ 1950 if (flags & TH_FIN) { 1951 /* 1952 * Application closed and 'dir' sent a FIN, we're now 1953 * going into LAST_ACK state 1954 */ 1955 newage = fr_tcplastack; 1956 state[dir] = TCPS_LAST_ACK; 1957 } else { 1958 /* 1959 * We remain in CLOSE_WAIT because the other side has 1960 * closed already and we did not close our side yet; 1961 * reset timeout 1962 */ 1963 newage = fr_tcphalfclosed; 1964 } 1965 break; 1966 1967 case TCPS_FIN_WAIT_1: /* 6 */ 1968 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) { 1969 /* 1970 * If the other side is not active anymore it has sent 1971 * us a FIN packet that we are ack'ing now with an ACK; 1972 * this means both sides have now closed the connection 1973 * and we go into TIME_WAIT 1974 */ 1975 /* 1976 * XXX: how do we know we really are ACKing the FIN 1977 * packet here? does the window code guarantee that? 1978 */ 1979 state[dir] = TCPS_TIME_WAIT; 1980 newage = fr_tcptimeout; 1981 } else 1982 /* 1983 * We closed our side of the connection already but the 1984 * other side is still active (ESTABLISHED/CLOSE_WAIT); 1985 * continue with this half-closed connection 1986 */ 1987 newage = fr_tcphalfclosed; 1988 break; 1989 1990 case TCPS_CLOSING: /* 7 */ 1991 /* NOT USED */ 1992 break; 1993 1994 case TCPS_LAST_ACK: /* 8 */ 1995 if (flags & TH_ACK) { 1996 if ((flags & TH_PUSH) || dlen) 1997 /* 1998 * There is still data to be delivered, reset 1999 * timeout 2000 */ 2001 newage = fr_tcplastack; 2002 else 2003 newage = *age; 2004 } 2005 /* 2006 * We cannot detect when we go out of LAST_ACK state to CLOSED 2007 * because that is based on the reception of ACK packets; 2008 * ipfilter can only detect that a packet has been sent by a 2009 * host 2010 */ 2011 break; 2012 2013 case TCPS_FIN_WAIT_2: /* 9 */ 2014 /* NOT USED */ 2015 break; 2016 2017 case TCPS_TIME_WAIT: /* 10 */ 2018 newage = fr_tcptimeout; /* default 4 mins */ 2019 /* we're in 2MSL timeout now */ 2020 break; 2021 } 2022 2023 if (newage != 0) { 2024 *age = newage; 2025 return 0; 2026 } 2027 return -1; 2028} 2029 2030 2031#ifdef IPFILTER_LOG 2032void ipstate_log(is, type) 2033struct ipstate *is; 2034u_int type; 2035{ 2036 struct ipslog ipsl; 2037 void *items[1]; 2038 size_t sizes[1]; 2039 int types[1]; 2040 2041 ipsl.isl_type = type; 2042 ipsl.isl_pkts = is->is_pkts; 2043 ipsl.isl_bytes = is->is_bytes; 2044 ipsl.isl_src = is->is_src; 2045 ipsl.isl_dst = is->is_dst; 2046 ipsl.isl_p = is->is_p; 2047 ipsl.isl_v = is->is_v; 2048 ipsl.isl_flags = is->is_flags; 2049 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 2050 ipsl.isl_sport = is->is_sport; 2051 ipsl.isl_dport = is->is_dport; 2052 if (ipsl.isl_p == IPPROTO_TCP) { 2053 ipsl.isl_state[0] = is->is_state[0]; 2054 ipsl.isl_state[1] = is->is_state[1]; 2055 } 2056 } else if (ipsl.isl_p == IPPROTO_ICMP) { 2057 ipsl.isl_itype = is->is_icmp.ics_type; 2058 } else if (ipsl.isl_p == IPPROTO_ICMPV6) { 2059 ipsl.isl_itype = is->is_icmp.ics_type; 2060 } else { 2061 ipsl.isl_ps.isl_filler[0] = 0; 2062 ipsl.isl_ps.isl_filler[1] = 0; 2063 } 2064 items[0] = &ipsl; 2065 sizes[0] = sizeof(ipsl); 2066 types[0] = 0; 2067 2068 (void) ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1); 2069} 2070#endif 2071 2072 2073#ifdef USE_INET6 2074frentry_t *fr_checkicmp6matchingstate(ip, fin) 2075ip6_t *ip; 2076fr_info_t *fin; 2077{ 2078 register ipstate_t *is, **isp; 2079 register u_short sport, dport; 2080 register u_char pr; 2081 struct icmp6_hdr *ic, *oic; 2082 union i6addr dst, src; 2083 u_short savelen; 2084 fr_info_t ofin; 2085 tcphdr_t *tcp; 2086 frentry_t *fr; 2087 ip6_t *oip; 2088 int type; 2089 u_int hv; 2090 2091 /* 2092 * Does it at least have the return (basic) IP header ? 2093 * Only a basic IP header (no options) should be with 2094 * an ICMP error header. 2095 */ 2096 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN)) 2097 return NULL; 2098 ic = (struct icmp6_hdr *)fin->fin_dp; 2099 type = ic->icmp6_type; 2100 /* 2101 * If it's not an error type, then return 2102 */ 2103 if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) && 2104 (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB)) 2105 return NULL; 2106 2107 oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN); 2108 if (fin->fin_plen < sizeof(*oip)) 2109 return NULL; 2110 2111 if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP) && 2112 (oip->ip6_nxt != IPPROTO_ICMPV6)) 2113 return NULL; 2114 2115 bzero((char *)&ofin, sizeof(ofin)); 2116 ofin.fin_out = !fin->fin_out; 2117 ofin.fin_ifp = fin->fin_ifp; 2118 ofin.fin_v = 6; 2119 2120 if (oip->ip6_nxt == IPPROTO_ICMPV6) { 2121 oic = (struct icmp6_hdr *)(oip + 1); 2122 /* 2123 * a ICMP error can only be generated as a result of an 2124 * ICMP query, not as the response on an ICMP error 2125 * 2126 * XXX theoretically ICMP_ECHOREP and the other reply's are 2127 * ICMP query's as well, but adding them here seems strange XXX 2128 */ 2129 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 2130 return NULL; 2131 2132 /* 2133 * perform a lookup of the ICMP packet in the state table 2134 */ 2135 hv = (pr = oip->ip6_nxt); 2136 src.in6 = oip->ip6_src; 2137 hv += src.in4.s_addr; 2138 dst.in6 = oip->ip6_dst; 2139 hv += dst.in4.s_addr; 2140 hv += oic->icmp6_id; 2141 hv += oic->icmp6_seq; 2142 hv %= fr_statesize; 2143 2144 oip->ip6_plen = ntohs(oip->ip6_plen); 2145 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 2146 oip->ip6_plen = htons(oip->ip6_plen); 2147 2148 READ_ENTER(&ipf_state); 2149 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) 2150 if ((is->is_p == pr) && 2151 (oic->icmp6_id == is->is_icmp.ics_id) && 2152 (oic->icmp6_seq == is->is_icmp.ics_seq) && 2153 fr_matchsrcdst(is, src, dst, &ofin, NULL)) { 2154 /* 2155 * in the state table ICMP query's are stored 2156 * with the type of the corresponding ICMP 2157 * response. Correct here 2158 */ 2159 if (((is->is_type == ICMP6_ECHO_REPLY) && 2160 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 2161 (is->is_type - 1 == oic->icmp6_type )) { 2162 ips_stats.iss_hits++; 2163 is->is_pkts++; 2164 is->is_bytes += fin->fin_plen; 2165 return is->is_rule; 2166 } 2167 } 2168 RWLOCK_EXIT(&ipf_state); 2169 2170 return NULL; 2171 } 2172 2173 tcp = (tcphdr_t *)(oip + 1); 2174 dport = tcp->th_dport; 2175 sport = tcp->th_sport; 2176 2177 hv = (pr = oip->ip6_nxt); 2178 src.in6 = oip->ip6_src; 2179 hv += src.in4.s_addr; 2180 hv += src.i6[1]; 2181 hv += src.i6[2]; 2182 hv += src.i6[3]; 2183 dst.in6 = oip->ip6_dst; 2184 hv += dst.in4.s_addr; 2185 hv += dst.i6[1]; 2186 hv += dst.i6[2]; 2187 hv += dst.i6[3]; 2188 hv += dport; 2189 hv += sport; 2190 hv %= fr_statesize; 2191 /* 2192 * we make an fin entry to be able to feed it to 2193 * matchsrcdst note that not all fields are encessary 2194 * but this is the cleanest way. Note further we fill 2195 * in fin_mp such that if someone uses it we'll get 2196 * a kernel panic. fr_matchsrcdst does not use this. 2197 * 2198 * watch out here, as ip is in host order and oip in network 2199 * order. Any change we make must be undone afterwards. 2200 */ 2201 savelen = oip->ip6_plen; 2202 oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN; 2203 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin); 2204 oip->ip6_plen = savelen; 2205 READ_ENTER(&ipf_state); 2206 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) { 2207 /* 2208 * Only allow this icmp though if the 2209 * encapsulated packet was allowed through the 2210 * other way around. Note that the minimal amount 2211 * of info present does not allow for checking against 2212 * tcp internals such as seq and ack numbers. 2213 */ 2214 if ((is->is_p == pr) && (is->is_v == 6) && 2215 fr_matchsrcdst(is, src, dst, &ofin, tcp)) { 2216 fr = is->is_rule; 2217 ips_stats.iss_hits++; 2218 is->is_pkts++; 2219 is->is_bytes += fin->fin_plen; 2220 /* 2221 * we deliberately do not touch the timeouts 2222 * for the accompanying state table entry. 2223 * It remains to be seen if that is correct. XXX 2224 */ 2225 RWLOCK_EXIT(&ipf_state); 2226 return fr; 2227 } 2228 } 2229 RWLOCK_EXIT(&ipf_state); 2230 return NULL; 2231} 2232#endif 2233