ip_state.c revision 170268
1/* $FreeBSD: head/sys/contrib/ipfilter/netinet/ip_state.c 170268 2007-06-04 02:54:36Z darrenr $ */ 2 3/* 4 * Copyright (C) 1995-2003 by Darren Reed. 5 * 6 * See the IPFILTER.LICENCE file for details on licencing. 7 */ 8#if defined(KERNEL) || defined(_KERNEL) 9# undef KERNEL 10# undef _KERNEL 11# define KERNEL 1 12# define _KERNEL 1 13#endif 14#include <sys/errno.h> 15#include <sys/types.h> 16#include <sys/param.h> 17#include <sys/file.h> 18#if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \ 19 defined(_KERNEL) 20# if (__NetBSD_Version__ < 399001400) 21# include "opt_ipfilter_log.h" 22# else 23# include "opt_ipfilter.h" 24# endif 25#endif 26#if defined(_KERNEL) && defined(__FreeBSD_version) && \ 27 (__FreeBSD_version >= 400000) && !defined(KLD_MODULE) 28#include "opt_inet6.h" 29#endif 30#if !defined(_KERNEL) && !defined(__KERNEL__) 31# include <stdio.h> 32# include <stdlib.h> 33# include <string.h> 34# define _KERNEL 35# ifdef __OpenBSD__ 36struct file; 37# endif 38# include <sys/uio.h> 39# undef _KERNEL 40#endif 41#if defined(_KERNEL) && (__FreeBSD_version >= 220000) 42# include <sys/filio.h> 43# include <sys/fcntl.h> 44# if (__FreeBSD_version >= 300000) && !defined(IPFILTER_LKM) 45# include "opt_ipfilter.h" 46# endif 47#else 48# include <sys/ioctl.h> 49#endif 50#include <sys/time.h> 51#if !defined(linux) 52# include <sys/protosw.h> 53#endif 54#include <sys/socket.h> 55#if defined(_KERNEL) 56# include <sys/systm.h> 57# if !defined(__SVR4) && !defined(__svr4__) 58# include <sys/mbuf.h> 59# endif 60#endif 61#if defined(__SVR4) || defined(__svr4__) 62# include <sys/filio.h> 63# include <sys/byteorder.h> 64# ifdef _KERNEL 65# include <sys/dditypes.h> 66# endif 67# include <sys/stream.h> 68# include <sys/kmem.h> 69#endif 70 71#include <net/if.h> 72#ifdef sun 73# include <net/af.h> 74#endif 75#include <net/route.h> 76#include <netinet/in.h> 77#include <netinet/in_systm.h> 78#include <netinet/ip.h> 79#include <netinet/tcp.h> 80#if !defined(linux) 81# include <netinet/ip_var.h> 82#endif 83#if !defined(__hpux) && !defined(linux) 84# include <netinet/tcp_fsm.h> 85#endif 86#include <netinet/udp.h> 87#include <netinet/ip_icmp.h> 88#include "netinet/ip_compat.h" 89#include <netinet/tcpip.h> 90#include "netinet/ip_fil.h" 91#include "netinet/ip_nat.h" 92#include "netinet/ip_frag.h" 93#include "netinet/ip_state.h" 94#include "netinet/ip_proxy.h" 95#ifdef IPFILTER_SYNC 96#include "netinet/ip_sync.h" 97#endif 98#ifdef IPFILTER_SCAN 99#include "netinet/ip_scan.h" 100#endif 101#ifdef USE_INET6 102#include <netinet/icmp6.h> 103#endif 104#if (__FreeBSD_version >= 300000) 105# include <sys/malloc.h> 106# if defined(_KERNEL) && !defined(IPFILTER_LKM) 107# include <sys/libkern.h> 108# include <sys/systm.h> 109# endif 110#endif 111/* END OF INCLUDES */ 112 113 114#if !defined(lint) 115static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed"; 116static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.186.2.69 2007/05/26 13:05:14 darrenr Exp $"; 117#endif 118 119static ipstate_t **ips_table = NULL; 120static u_long *ips_seed = NULL; 121static int ips_num = 0; 122static u_long ips_last_force_flush = 0; 123ips_stat_t ips_stats; 124 125#ifdef USE_INET6 126static ipstate_t *fr_checkicmp6matchingstate __P((fr_info_t *)); 127#endif 128static ipstate_t *fr_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *, 129 i6addr_t *, tcphdr_t *, u_32_t)); 130static ipstate_t *fr_checkicmpmatchingstate __P((fr_info_t *)); 131static int fr_state_flush __P((int, int)); 132static int fr_state_flush_entry __P((void *)); 133static ips_stat_t *fr_statetstats __P((void)); 134static int fr_delstate __P((ipstate_t *, int)); 135static int fr_state_remove __P((caddr_t)); 136static void fr_ipsmove __P((ipstate_t *, u_int)); 137static int fr_tcpstate __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 138static int fr_tcpoptions __P((fr_info_t *, tcphdr_t *, tcpdata_t *)); 139static ipstate_t *fr_stclone __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 140static void fr_fixinisn __P((fr_info_t *, ipstate_t *)); 141static void fr_fixoutisn __P((fr_info_t *, ipstate_t *)); 142static void fr_checknewisn __P((fr_info_t *, ipstate_t *)); 143static int fr_stateiter __P((ipftoken_t *, ipfgeniter_t *)); 144static int fr_stgettable __P((char *)); 145 146int fr_stputent __P((caddr_t)); 147int fr_stgetent __P((caddr_t)); 148 149#define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */ 150#define FIVE_DAYS (5 * ONE_DAY) 151#define DOUBLE_HASH(x) (((x) + ips_seed[(x) % fr_statesize]) % fr_statesize) 152 153u_long fr_tcpidletimeout = FIVE_DAYS, 154 fr_tcpclosewait = IPF_TTLVAL(2 * TCP_MSL), 155 fr_tcplastack = IPF_TTLVAL(30), 156 fr_tcptimeout = IPF_TTLVAL(2 * TCP_MSL), 157 fr_tcptimewait = IPF_TTLVAL(2 * TCP_MSL), 158 fr_tcpclosed = IPF_TTLVAL(30), 159 fr_tcphalfclosed = IPF_TTLVAL(2 * 3600), /* 2 hours */ 160 fr_udptimeout = IPF_TTLVAL(120), 161 fr_udpacktimeout = IPF_TTLVAL(12), 162 fr_icmptimeout = IPF_TTLVAL(60), 163 fr_icmpacktimeout = IPF_TTLVAL(6), 164 fr_iptimeout = IPF_TTLVAL(60); 165int fr_statemax = IPSTATE_MAX, 166 fr_statesize = IPSTATE_SIZE; 167int fr_state_doflush = 0, 168 fr_state_lock = 0, 169 fr_state_maxbucket = 0, 170 fr_state_maxbucket_reset = 1, 171 fr_state_init = 0; 172ipftq_t ips_tqtqb[IPF_TCP_NSTATES], 173 ips_udptq, 174 ips_udpacktq, 175 ips_iptq, 176 ips_icmptq, 177 ips_icmpacktq, 178 ips_deletetq, 179 *ips_utqe = NULL; 180#ifdef IPFILTER_LOG 181int ipstate_logging = 1; 182#else 183int ipstate_logging = 0; 184#endif 185ipstate_t *ips_list = NULL; 186 187 188/* ------------------------------------------------------------------------ */ 189/* Function: fr_stateinit */ 190/* Returns: int - 0 == success, -1 == failure */ 191/* Parameters: Nil */ 192/* */ 193/* Initialise all the global variables used within the state code. */ 194/* This action also includes initiailising locks. */ 195/* ------------------------------------------------------------------------ */ 196int fr_stateinit() 197{ 198 int i; 199 200 KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *)); 201 if (ips_table == NULL) 202 return -1; 203 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *)); 204 205 KMALLOCS(ips_seed, u_long *, fr_statesize * sizeof(*ips_seed)); 206 if (ips_seed == NULL) 207 return -2; 208 for (i = 0; i < fr_statesize; i++) { 209 /* 210 * XXX - ips_seed[X] should be a random number of sorts. 211 */ 212#if (__FreeBSD_version >= 400000) 213 ips_seed[i] = arc4random(); 214#else 215 ips_seed[i] = ((u_long)ips_seed + i) * fr_statesize; 216 ips_seed[i] ^= 0xa5a55a5a; 217 ips_seed[i] *= (u_long)ips_seed; 218 ips_seed[i] ^= 0x5a5aa5a5; 219 ips_seed[i] *= fr_statemax; 220#endif 221 } 222 223 /* fill icmp reply type table */ 224 for (i = 0; i <= ICMP_MAXTYPE; i++) 225 icmpreplytype4[i] = -1; 226 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 227 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 228 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 229 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 230#ifdef USE_INET6 231 /* fill icmp reply type table */ 232 for (i = 0; i <= ICMP6_MAXTYPE; i++) 233 icmpreplytype6[i] = -1; 234 icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY; 235 icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT; 236 icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY; 237 icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT; 238 icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT; 239#endif 240 241 KMALLOCS(ips_stats.iss_bucketlen, u_long *, 242 fr_statesize * sizeof(u_long)); 243 if (ips_stats.iss_bucketlen == NULL) 244 return -1; 245 bzero((char *)ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long)); 246 247 if (fr_state_maxbucket == 0) { 248 for (i = fr_statesize; i > 0; i >>= 1) 249 fr_state_maxbucket++; 250 fr_state_maxbucket *= 2; 251 } 252 253 ips_stats.iss_tcptab = ips_tqtqb; 254 fr_sttab_init(ips_tqtqb); 255 ips_tqtqb[IPF_TCP_NSTATES - 1].ifq_next = &ips_udptq; 256 ips_udptq.ifq_ttl = (u_long)fr_udptimeout; 257 ips_udptq.ifq_ref = 1; 258 ips_udptq.ifq_head = NULL; 259 ips_udptq.ifq_tail = &ips_udptq.ifq_head; 260 MUTEX_INIT(&ips_udptq.ifq_lock, "ipftq udp tab"); 261 ips_udptq.ifq_next = &ips_udpacktq; 262 ips_udpacktq.ifq_ttl = (u_long)fr_udpacktimeout; 263 ips_udpacktq.ifq_ref = 1; 264 ips_udpacktq.ifq_head = NULL; 265 ips_udpacktq.ifq_tail = &ips_udpacktq.ifq_head; 266 MUTEX_INIT(&ips_udpacktq.ifq_lock, "ipftq udpack tab"); 267 ips_udpacktq.ifq_next = &ips_icmptq; 268 ips_icmptq.ifq_ttl = (u_long)fr_icmptimeout; 269 ips_icmptq.ifq_ref = 1; 270 ips_icmptq.ifq_head = NULL; 271 ips_icmptq.ifq_tail = &ips_icmptq.ifq_head; 272 MUTEX_INIT(&ips_icmptq.ifq_lock, "ipftq icmp tab"); 273 ips_icmptq.ifq_next = &ips_icmpacktq; 274 ips_icmpacktq.ifq_ttl = (u_long)fr_icmpacktimeout; 275 ips_icmpacktq.ifq_ref = 1; 276 ips_icmpacktq.ifq_head = NULL; 277 ips_icmpacktq.ifq_tail = &ips_icmpacktq.ifq_head; 278 MUTEX_INIT(&ips_icmpacktq.ifq_lock, "ipftq icmpack tab"); 279 ips_icmpacktq.ifq_next = &ips_iptq; 280 ips_iptq.ifq_ttl = (u_long)fr_iptimeout; 281 ips_iptq.ifq_ref = 1; 282 ips_iptq.ifq_head = NULL; 283 ips_iptq.ifq_tail = &ips_iptq.ifq_head; 284 MUTEX_INIT(&ips_iptq.ifq_lock, "ipftq ip tab"); 285 ips_iptq.ifq_next = &ips_deletetq; 286 ips_deletetq.ifq_ttl = (u_long)1; 287 ips_deletetq.ifq_ref = 1; 288 ips_deletetq.ifq_head = NULL; 289 ips_deletetq.ifq_tail = &ips_deletetq.ifq_head; 290 MUTEX_INIT(&ips_deletetq.ifq_lock, "state delete queue"); 291 ips_deletetq.ifq_next = NULL; 292 293 RWLOCK_INIT(&ipf_state, "ipf IP state rwlock"); 294 MUTEX_INIT(&ipf_stinsert, "ipf state insert mutex"); 295 fr_state_init = 1; 296 297 ips_last_force_flush = fr_ticks; 298 return 0; 299} 300 301 302/* ------------------------------------------------------------------------ */ 303/* Function: fr_stateunload */ 304/* Returns: Nil */ 305/* Parameters: Nil */ 306/* */ 307/* Release and destroy any resources acquired or initialised so that */ 308/* IPFilter can be unloaded or re-initialised. */ 309/* ------------------------------------------------------------------------ */ 310void fr_stateunload() 311{ 312 ipftq_t *ifq, *ifqnext; 313 ipstate_t *is; 314 315 while ((is = ips_list) != NULL) 316 fr_delstate(is, ISL_UNLOAD); 317 318 /* 319 * Proxy timeout queues are not cleaned here because although they 320 * exist on the state list, appr_unload is called after fr_stateunload 321 * and the proxies actually are responsible for them being created. 322 * Should the proxy timeouts have their own list? There's no real 323 * justification as this is the only complicationA 324 */ 325 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 326 ifqnext = ifq->ifq_next; 327 if (((ifq->ifq_flags & IFQF_PROXY) == 0) && 328 (fr_deletetimeoutqueue(ifq) == 0)) 329 fr_freetimeoutqueue(ifq); 330 } 331 332 ips_stats.iss_inuse = 0; 333 ips_num = 0; 334 335 if (fr_state_init == 1) { 336 fr_sttab_destroy(ips_tqtqb); 337 MUTEX_DESTROY(&ips_udptq.ifq_lock); 338 MUTEX_DESTROY(&ips_icmptq.ifq_lock); 339 MUTEX_DESTROY(&ips_udpacktq.ifq_lock); 340 MUTEX_DESTROY(&ips_icmpacktq.ifq_lock); 341 MUTEX_DESTROY(&ips_iptq.ifq_lock); 342 MUTEX_DESTROY(&ips_deletetq.ifq_lock); 343 } 344 345 if (ips_table != NULL) { 346 KFREES(ips_table, fr_statesize * sizeof(*ips_table)); 347 ips_table = NULL; 348 } 349 350 if (ips_seed != NULL) { 351 KFREES(ips_seed, fr_statesize * sizeof(*ips_seed)); 352 ips_seed = NULL; 353 } 354 355 if (ips_stats.iss_bucketlen != NULL) { 356 KFREES(ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long)); 357 ips_stats.iss_bucketlen = NULL; 358 } 359 360 if (fr_state_maxbucket_reset == 1) 361 fr_state_maxbucket = 0; 362 363 if (fr_state_init == 1) { 364 fr_state_init = 0; 365 RW_DESTROY(&ipf_state); 366 MUTEX_DESTROY(&ipf_stinsert); 367 } 368} 369 370 371/* ------------------------------------------------------------------------ */ 372/* Function: fr_statetstats */ 373/* Returns: ips_state_t* - pointer to state stats structure */ 374/* Parameters: Nil */ 375/* */ 376/* Put all the current numbers and pointers into a single struct and return */ 377/* a pointer to it. */ 378/* ------------------------------------------------------------------------ */ 379static ips_stat_t *fr_statetstats() 380{ 381 ips_stats.iss_active = ips_num; 382 ips_stats.iss_statesize = fr_statesize; 383 ips_stats.iss_statemax = fr_statemax; 384 ips_stats.iss_table = ips_table; 385 ips_stats.iss_list = ips_list; 386 ips_stats.iss_ticks = fr_ticks; 387 return &ips_stats; 388} 389 390/* ------------------------------------------------------------------------ */ 391/* Function: fr_state_remove */ 392/* Returns: int - 0 == success, != 0 == failure */ 393/* Parameters: data(I) - pointer to state structure to delete from table */ 394/* */ 395/* Search for a state structure that matches the one passed, according to */ 396/* the IP addresses and other protocol specific information. */ 397/* ------------------------------------------------------------------------ */ 398static int fr_state_remove(data) 399caddr_t data; 400{ 401 ipstate_t *sp, st; 402 int error; 403 404 sp = &st; 405 error = fr_inobj(data, &st, IPFOBJ_IPSTATE); 406 if (error) 407 return EFAULT; 408 409 WRITE_ENTER(&ipf_state); 410 for (sp = ips_list; sp; sp = sp->is_next) 411 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 412 !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src, 413 sizeof(st.is_src)) && 414 !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_src, 415 sizeof(st.is_dst)) && 416 !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps, 417 sizeof(st.is_ps))) { 418 fr_delstate(sp, ISL_REMOVE); 419 RWLOCK_EXIT(&ipf_state); 420 return 0; 421 } 422 RWLOCK_EXIT(&ipf_state); 423 return ESRCH; 424} 425 426 427/* ------------------------------------------------------------------------ */ 428/* Function: fr_state_ioctl */ 429/* Returns: int - 0 == success, != 0 == failure */ 430/* Parameters: data(I) - pointer to ioctl data */ 431/* cmd(I) - ioctl command integer */ 432/* mode(I) - file mode bits used with open */ 433/* */ 434/* Processes an ioctl call made to operate on the IP Filter state device. */ 435/* ------------------------------------------------------------------------ */ 436int fr_state_ioctl(data, cmd, mode, uid, ctx) 437caddr_t data; 438ioctlcmd_t cmd; 439int mode, uid; 440void *ctx; 441{ 442 int arg, ret, error = 0; 443 SPL_INT(s); 444 445 switch (cmd) 446 { 447 /* 448 * Delete an entry from the state table. 449 */ 450 case SIOCDELST : 451 error = fr_state_remove(data); 452 break; 453 454 /* 455 * Flush the state table 456 */ 457 case SIOCIPFFL : 458 error = BCOPYIN(data, (char *)&arg, sizeof(arg)); 459 if (error != 0) { 460 error = EFAULT; 461 } else { 462 WRITE_ENTER(&ipf_state); 463 ret = fr_state_flush(arg, 4); 464 RWLOCK_EXIT(&ipf_state); 465 error = BCOPYOUT((char *)&ret, data, sizeof(ret)); 466 if (error != 0) 467 error = EFAULT; 468 } 469 break; 470 471#ifdef USE_INET6 472 case SIOCIPFL6 : 473 error = BCOPYIN(data, (char *)&arg, sizeof(arg)); 474 if (error != 0) { 475 error = EFAULT; 476 } else { 477 WRITE_ENTER(&ipf_state); 478 ret = fr_state_flush(arg, 6); 479 RWLOCK_EXIT(&ipf_state); 480 error = BCOPYOUT((char *)&ret, data, sizeof(ret)); 481 if (error != 0) 482 error = EFAULT; 483 } 484 break; 485#endif 486#ifdef IPFILTER_LOG 487 /* 488 * Flush the state log. 489 */ 490 case SIOCIPFFB : 491 if (!(mode & FWRITE)) 492 error = EPERM; 493 else { 494 int tmp; 495 496 tmp = ipflog_clear(IPL_LOGSTATE); 497 error = BCOPYOUT((char *)&tmp, data, sizeof(tmp)); 498 if (error != 0) 499 error = EFAULT; 500 } 501 break; 502 503 /* 504 * Turn logging of state information on/off. 505 */ 506 case SIOCSETLG : 507 if (!(mode & FWRITE)) 508 error = EPERM; 509 else { 510 error = BCOPYIN((char *)data, (char *)&ipstate_logging, 511 sizeof(ipstate_logging)); 512 if (error != 0) 513 error = EFAULT; 514 } 515 break; 516 517 /* 518 * Return the current state of logging. 519 */ 520 case SIOCGETLG : 521 error = BCOPYOUT((char *)&ipstate_logging, (char *)data, 522 sizeof(ipstate_logging)); 523 if (error != 0) 524 error = EFAULT; 525 break; 526 527 /* 528 * Return the number of bytes currently waiting to be read. 529 */ 530 case FIONREAD : 531 arg = iplused[IPL_LOGSTATE]; /* returned in an int */ 532 error = BCOPYOUT((char *)&arg, data, sizeof(arg)); 533 if (error != 0) 534 error = EFAULT; 535 break; 536#endif 537 538 /* 539 * Get the current state statistics. 540 */ 541 case SIOCGETFS : 542 error = fr_outobj(data, fr_statetstats(), IPFOBJ_STATESTAT); 543 break; 544 545 /* 546 * Lock/Unlock the state table. (Locking prevents any changes, which 547 * means no packets match). 548 */ 549 case SIOCSTLCK : 550 if (!(mode & FWRITE)) { 551 error = EPERM; 552 } else { 553 fr_lock(data, &fr_state_lock); 554 } 555 break; 556 557 /* 558 * Add an entry to the current state table. 559 */ 560 case SIOCSTPUT : 561 if (!fr_state_lock || !(mode &FWRITE)) { 562 error = EACCES; 563 break; 564 } 565 error = fr_stputent(data); 566 break; 567 568 /* 569 * Get a state table entry. 570 */ 571 case SIOCSTGET : 572 if (!fr_state_lock) { 573 error = EACCES; 574 break; 575 } 576 error = fr_stgetent(data); 577 break; 578 579 /* 580 * Return a copy of the hash table bucket lengths 581 */ 582 case SIOCSTAT1 : 583 error = BCOPYOUT(ips_stats.iss_bucketlen, data, 584 fr_statesize * sizeof(u_long)); 585 if (error != 0) 586 error = EFAULT; 587 break; 588 589 case SIOCGENITER : 590 { 591 ipftoken_t *token; 592 ipfgeniter_t iter; 593 594 error = fr_inobj(data, &iter, IPFOBJ_GENITER); 595 if (error != 0) 596 break; 597 598 SPL_SCHED(s); 599 token = ipf_findtoken(IPFGENITER_STATE, uid, ctx); 600 if (token != NULL) 601 error = fr_stateiter(token, &iter); 602 else 603 error = ESRCH; 604 RWLOCK_EXIT(&ipf_tokens); 605 SPL_X(s); 606 break; 607 } 608 609 case SIOCGTABL : 610 error = fr_stgettable(data); 611 break; 612 613 case SIOCIPFDELTOK : 614 error = BCOPYIN(data, (char *)&arg, sizeof(arg)); 615 if (error != 0) { 616 error = EFAULT; 617 } else { 618 SPL_SCHED(s); 619 error = ipf_deltoken(arg, uid, ctx); 620 SPL_X(s); 621 } 622 break; 623 624 case SIOCGTQTAB : 625 error = fr_outobj(data, ips_tqtqb, IPFOBJ_STATETQTAB); 626 break; 627 628 default : 629 error = EINVAL; 630 break; 631 } 632 return error; 633} 634 635 636/* ------------------------------------------------------------------------ */ 637/* Function: fr_stgetent */ 638/* Returns: int - 0 == success, != 0 == failure */ 639/* Parameters: data(I) - pointer to state structure to retrieve from table */ 640/* */ 641/* Copy out state information from the kernel to a user space process. If */ 642/* there is a filter rule associated with the state entry, copy that out */ 643/* as well. The entry to copy out is taken from the value of "ips_next" in */ 644/* the struct passed in and if not null and not found in the list of current*/ 645/* state entries, the retrieval fails. */ 646/* ------------------------------------------------------------------------ */ 647int fr_stgetent(data) 648caddr_t data; 649{ 650 ipstate_t *is, *isn; 651 ipstate_save_t ips; 652 int error; 653 654 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 655 if (error) 656 return EFAULT; 657 658 isn = ips.ips_next; 659 if (isn == NULL) { 660 isn = ips_list; 661 if (isn == NULL) { 662 if (ips.ips_next == NULL) 663 return ENOENT; 664 return 0; 665 } 666 } else { 667 /* 668 * Make sure the pointer we're copying from exists in the 669 * current list of entries. Security precaution to prevent 670 * copying of random kernel data. 671 */ 672 for (is = ips_list; is; is = is->is_next) 673 if (is == isn) 674 break; 675 if (!is) 676 return ESRCH; 677 } 678 ips.ips_next = isn->is_next; 679 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 680 ips.ips_rule = isn->is_rule; 681 if (isn->is_rule != NULL) 682 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 683 sizeof(ips.ips_fr)); 684 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 685 if (error) 686 return EFAULT; 687 return 0; 688} 689 690 691/* ------------------------------------------------------------------------ */ 692/* Function: fr_stputent */ 693/* Returns: int - 0 == success, != 0 == failure */ 694/* Parameters: data(I) - pointer to state information struct */ 695/* */ 696/* This function implements the SIOCSTPUT ioctl: insert a state entry into */ 697/* the state table. If the state info. includes a pointer to a filter rule */ 698/* then also add in an orphaned rule (will not show up in any "ipfstat -io" */ 699/* output. */ 700/* ------------------------------------------------------------------------ */ 701int fr_stputent(data) 702caddr_t data; 703{ 704 ipstate_t *is, *isn; 705 ipstate_save_t ips; 706 int error, out, i; 707 frentry_t *fr; 708 char *name; 709 710 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 711 if (error) 712 return EFAULT; 713 714 KMALLOC(isn, ipstate_t *); 715 if (isn == NULL) 716 return ENOMEM; 717 718 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 719 bzero((char *)isn, offsetof(struct ipstate, is_pkts)); 720 isn->is_sti.tqe_pnext = NULL; 721 isn->is_sti.tqe_next = NULL; 722 isn->is_sti.tqe_ifq = NULL; 723 isn->is_sti.tqe_parent = isn; 724 isn->is_ifp[0] = NULL; 725 isn->is_ifp[1] = NULL; 726 isn->is_ifp[2] = NULL; 727 isn->is_ifp[3] = NULL; 728 isn->is_sync = NULL; 729 fr = ips.ips_rule; 730 731 if (fr == NULL) { 732 READ_ENTER(&ipf_state); 733 fr_stinsert(isn, 0); 734 MUTEX_EXIT(&isn->is_lock); 735 RWLOCK_EXIT(&ipf_state); 736 return 0; 737 } 738 739 if (isn->is_flags & SI_NEWFR) { 740 KMALLOC(fr, frentry_t *); 741 if (fr == NULL) { 742 KFREE(isn); 743 return ENOMEM; 744 } 745 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 746 out = fr->fr_flags & FR_OUTQUE ? 1 : 0; 747 isn->is_rule = fr; 748 ips.ips_is.is_rule = fr; 749 MUTEX_NUKE(&fr->fr_lock); 750 MUTEX_INIT(&fr->fr_lock, "state filter rule lock"); 751 752 /* 753 * Look up all the interface names in the rule. 754 */ 755 for (i = 0; i < 4; i++) { 756 name = fr->fr_ifnames[i]; 757 fr->fr_ifas[i] = fr_resolvenic(name, fr->fr_v); 758 name = isn->is_ifname[i]; 759 isn->is_ifp[i] = fr_resolvenic(name, isn->is_v); 760 } 761 762 fr->fr_ref = 0; 763 fr->fr_dsize = 0; 764 fr->fr_data = NULL; 765 fr->fr_type = FR_T_NONE; 766 767 fr_resolvedest(&fr->fr_tifs[0], fr->fr_v); 768 fr_resolvedest(&fr->fr_tifs[1], fr->fr_v); 769 fr_resolvedest(&fr->fr_dif, fr->fr_v); 770 771 /* 772 * send a copy back to userland of what we ended up 773 * to allow for verification. 774 */ 775 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 776 if (error) { 777 KFREE(isn); 778 MUTEX_DESTROY(&fr->fr_lock); 779 KFREE(fr); 780 return EFAULT; 781 } 782 READ_ENTER(&ipf_state); 783 fr_stinsert(isn, 0); 784 MUTEX_EXIT(&isn->is_lock); 785 RWLOCK_EXIT(&ipf_state); 786 787 } else { 788 READ_ENTER(&ipf_state); 789 for (is = ips_list; is; is = is->is_next) 790 if (is->is_rule == fr) { 791 fr_stinsert(isn, 0); 792 MUTEX_EXIT(&isn->is_lock); 793 break; 794 } 795 796 if (is == NULL) { 797 KFREE(isn); 798 isn = NULL; 799 } 800 RWLOCK_EXIT(&ipf_state); 801 802 return (isn == NULL) ? ESRCH : 0; 803 } 804 805 return 0; 806} 807 808 809/* ------------------------------------------------------------------------ */ 810/* Function: fr_stinsert */ 811/* Returns: Nil */ 812/* Parameters: is(I) - pointer to state structure */ 813/* rev(I) - flag indicating forward/reverse direction of packet */ 814/* */ 815/* Inserts a state structure into the hash table (for lookups) and the list */ 816/* of state entries (for enumeration). Resolves all of the interface names */ 817/* to pointers and adjusts running stats for the hash table as appropriate. */ 818/* */ 819/* Locking: it is assumed that some kind of lock on ipf_state is held. */ 820/* Exits with is_lock initialised and held. */ 821/* ------------------------------------------------------------------------ */ 822void fr_stinsert(is, rev) 823ipstate_t *is; 824int rev; 825{ 826 frentry_t *fr; 827 u_int hv; 828 int i; 829 830 MUTEX_INIT(&is->is_lock, "ipf state entry"); 831 832 fr = is->is_rule; 833 if (fr != NULL) { 834 MUTEX_ENTER(&fr->fr_lock); 835 fr->fr_ref++; 836 fr->fr_statecnt++; 837 MUTEX_EXIT(&fr->fr_lock); 838 } 839 840 /* 841 * Look up all the interface names in the state entry. 842 */ 843 for (i = 0; i < 4; i++) { 844 if (is->is_ifp[i] != NULL) 845 continue; 846 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], is->is_v); 847 } 848 849 /* 850 * If we could trust is_hv, then the modulous would not be needed, but 851 * when running with IPFILTER_SYNC, this stops bad values. 852 */ 853 hv = is->is_hv % fr_statesize; 854 is->is_hv = hv; 855 856 /* 857 * We need to get both of these locks...the first because it is 858 * possible that once the insert is complete another packet might 859 * come along, match the entry and want to update it. 860 */ 861 MUTEX_ENTER(&is->is_lock); 862 MUTEX_ENTER(&ipf_stinsert); 863 864 /* 865 * add into list table. 866 */ 867 if (ips_list != NULL) 868 ips_list->is_pnext = &is->is_next; 869 is->is_pnext = &ips_list; 870 is->is_next = ips_list; 871 ips_list = is; 872 873 if (ips_table[hv] != NULL) 874 ips_table[hv]->is_phnext = &is->is_hnext; 875 else 876 ips_stats.iss_inuse++; 877 is->is_phnext = ips_table + hv; 878 is->is_hnext = ips_table[hv]; 879 ips_table[hv] = is; 880 ips_stats.iss_bucketlen[hv]++; 881 ips_num++; 882 MUTEX_EXIT(&ipf_stinsert); 883 884 fr_setstatequeue(is, rev); 885} 886 887 888/* ------------------------------------------------------------------------ */ 889/* Function: fr_addstate */ 890/* Returns: ipstate_t* - NULL == failure, else pointer to new state */ 891/* Parameters: fin(I) - pointer to packet information */ 892/* stsave(O) - pointer to place to save pointer to created */ 893/* state structure. */ 894/* flags(I) - flags to use when creating the structure */ 895/* */ 896/* Creates a new IP state structure from the packet information collected. */ 897/* Inserts it into the state table and appends to the bottom of the active */ 898/* list. If the capacity of the table has reached the maximum allowed then */ 899/* the call will fail and a flush is scheduled for the next timeout call. */ 900/* */ 901/* NOTE: The use of stsave to point to nat_state will result in memory */ 902/* corruption. It should only be used to point to objects that will */ 903/* either outlive this (not expired) or will deref the ip_state_t */ 904/* when they are deleted. */ 905/* ------------------------------------------------------------------------ */ 906ipstate_t *fr_addstate(fin, stsave, flags) 907fr_info_t *fin; 908ipstate_t **stsave; 909u_int flags; 910{ 911 ipstate_t *is, ips; 912 struct icmp *ic; 913 u_int pass, hv; 914 frentry_t *fr; 915 tcphdr_t *tcp; 916 grehdr_t *gre; 917 int out; 918 919 if (fr_state_lock || 920 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 921 return NULL; 922 923 if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN)) 924 return NULL; 925 926 /* 927 * If a "keep state" rule has reached the maximum number of references 928 * to it, then schedule an automatic flush in case we can clear out 929 * some "dead old wood". Note that because the lock isn't held on 930 * fr it is possible that we could overflow. The cost of overflowing 931 * is being ignored here as the number by which it can overflow is 932 * a product of the number of simultaneous threads that could be 933 * executing in here, so a limit of 100 won't result in 200, but could 934 * result in 101 or 102. 935 */ 936 fr = fin->fin_fr; 937 if (fr != NULL) { 938 if ((ips_num >= fr_statemax) && (fr->fr_statemax == 0)) { 939 ATOMIC_INCL(ips_stats.iss_max); 940 fr_state_doflush = 1; 941 return NULL; 942 } 943 if ((fr->fr_statemax != 0) && 944 (fr->fr_statecnt >= fr->fr_statemax)) { 945 ATOMIC_INCL(ips_stats.iss_maxref); 946 return NULL; 947 } 948 } 949 950 pass = (fr == NULL) ? 0 : fr->fr_flags; 951 952 ic = NULL; 953 tcp = NULL; 954 out = fin->fin_out; 955 is = &ips; 956 bzero((char *)is, sizeof(*is)); 957 is->is_die = 1 + fr_ticks; 958 959 /* 960 * Copy and calculate... 961 */ 962 hv = (is->is_p = fin->fin_fi.fi_p); 963 is->is_src = fin->fin_fi.fi_src; 964 hv += is->is_saddr; 965 is->is_dst = fin->fin_fi.fi_dst; 966 hv += is->is_daddr; 967#ifdef USE_INET6 968 if (fin->fin_v == 6) { 969 /* 970 * For ICMPv6, we check to see if the destination address is 971 * a multicast address. If it is, do not include it in the 972 * calculation of the hash because the correct reply will come 973 * back from a real address, not a multicast address. 974 */ 975 if ((is->is_p == IPPROTO_ICMPV6) && 976 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) { 977 /* 978 * So you can do keep state with neighbour discovery. 979 * 980 * Here we could use the address from the neighbour 981 * solicit message to put in the state structure and 982 * we could use that without a wildcard flag too... 983 */ 984 flags |= SI_W_DADDR; 985 hv -= is->is_daddr; 986 } else { 987 hv += is->is_dst.i6[1]; 988 hv += is->is_dst.i6[2]; 989 hv += is->is_dst.i6[3]; 990 } 991 hv += is->is_src.i6[1]; 992 hv += is->is_src.i6[2]; 993 hv += is->is_src.i6[3]; 994 } 995#endif 996 997 switch (is->is_p) 998 { 999#ifdef USE_INET6 1000 case IPPROTO_ICMPV6 : 1001 ic = fin->fin_dp; 1002 1003 switch (ic->icmp_type) 1004 { 1005 case ICMP6_ECHO_REQUEST : 1006 is->is_icmp.ici_type = ic->icmp_type; 1007 hv += (is->is_icmp.ici_id = ic->icmp_id); 1008 break; 1009 case ICMP6_MEMBERSHIP_QUERY : 1010 case ND_ROUTER_SOLICIT : 1011 case ND_NEIGHBOR_SOLICIT : 1012 case ICMP6_NI_QUERY : 1013 is->is_icmp.ici_type = ic->icmp_type; 1014 break; 1015 default : 1016 return NULL; 1017 } 1018 ATOMIC_INCL(ips_stats.iss_icmp); 1019 break; 1020#endif 1021 case IPPROTO_ICMP : 1022 ic = fin->fin_dp; 1023 1024 switch (ic->icmp_type) 1025 { 1026 case ICMP_ECHO : 1027 case ICMP_TSTAMP : 1028 case ICMP_IREQ : 1029 case ICMP_MASKREQ : 1030 is->is_icmp.ici_type = ic->icmp_type; 1031 hv += (is->is_icmp.ici_id = ic->icmp_id); 1032 break; 1033 default : 1034 return NULL; 1035 } 1036 ATOMIC_INCL(ips_stats.iss_icmp); 1037 break; 1038 1039 case IPPROTO_GRE : 1040 gre = fin->fin_dp; 1041 1042 is->is_gre.gs_flags = gre->gr_flags; 1043 is->is_gre.gs_ptype = gre->gr_ptype; 1044 if (GRE_REV(is->is_gre.gs_flags) == 1) { 1045 is->is_call[0] = fin->fin_data[0]; 1046 is->is_call[1] = fin->fin_data[1]; 1047 } 1048 break; 1049 1050 case IPPROTO_TCP : 1051 tcp = fin->fin_dp; 1052 1053 if (tcp->th_flags & TH_RST) 1054 return NULL; 1055 /* 1056 * The endian of the ports doesn't matter, but the ack and 1057 * sequence numbers do as we do mathematics on them later. 1058 */ 1059 is->is_sport = htons(fin->fin_data[0]); 1060 is->is_dport = htons(fin->fin_data[1]); 1061 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 1062 hv += is->is_sport; 1063 hv += is->is_dport; 1064 } 1065 1066 /* 1067 * If this is a real packet then initialise fields in the 1068 * state information structure from the TCP header information. 1069 */ 1070 1071 is->is_maxdwin = 1; 1072 is->is_maxswin = ntohs(tcp->th_win); 1073 if (is->is_maxswin == 0) 1074 is->is_maxswin = 1; 1075 1076 if ((fin->fin_flx & FI_IGNORE) == 0) { 1077 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 1078 (TCP_OFF(tcp) << 2) + 1079 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 1080 ((tcp->th_flags & TH_FIN) ? 1 : 0); 1081 is->is_maxsend = is->is_send; 1082 1083 /* 1084 * Window scale option is only present in 1085 * SYN/SYN-ACK packet. 1086 */ 1087 if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) == 1088 TH_SYN && 1089 (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 1090 if (fr_tcpoptions(fin, tcp, 1091 &is->is_tcp.ts_data[0]) == -1) { 1092 fin->fin_flx |= FI_BAD; 1093 } 1094 } 1095 1096 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) { 1097 fr_checknewisn(fin, is); 1098 fr_fixoutisn(fin, is); 1099 } 1100 1101 if ((tcp->th_flags & TH_OPENING) == TH_SYN) 1102 flags |= IS_TCPFSM; 1103 else { 1104 is->is_maxdwin = is->is_maxswin * 2; 1105 is->is_dend = ntohl(tcp->th_ack); 1106 is->is_maxdend = ntohl(tcp->th_ack); 1107 is->is_maxdwin *= 2; 1108 } 1109 } 1110 1111 /* 1112 * If we're creating state for a starting connection, start the 1113 * timer on it as we'll never see an error if it fails to 1114 * connect. 1115 */ 1116 ATOMIC_INCL(ips_stats.iss_tcp); 1117 break; 1118 1119 case IPPROTO_UDP : 1120 tcp = fin->fin_dp; 1121 1122 is->is_sport = htons(fin->fin_data[0]); 1123 is->is_dport = htons(fin->fin_data[1]); 1124 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 1125 hv += tcp->th_dport; 1126 hv += tcp->th_sport; 1127 } 1128 ATOMIC_INCL(ips_stats.iss_udp); 1129 break; 1130 1131 default : 1132 break; 1133 } 1134 hv = DOUBLE_HASH(hv); 1135 is->is_hv = hv; 1136 is->is_rule = fr; 1137 is->is_flags = flags & IS_INHERITED; 1138 1139 /* 1140 * Look for identical state. 1141 */ 1142 for (is = ips_table[is->is_hv % fr_statesize]; is != NULL; 1143 is = is->is_hnext) { 1144 if (bcmp(&ips.is_src, &is->is_src, 1145 offsetof(struct ipstate, is_ps) - 1146 offsetof(struct ipstate, is_src)) == 0) 1147 break; 1148 } 1149 if (is != NULL) 1150 return NULL; 1151 1152 if (ips_stats.iss_bucketlen[hv] >= fr_state_maxbucket) { 1153 ATOMIC_INCL(ips_stats.iss_bucketfull); 1154 return NULL; 1155 } 1156 KMALLOC(is, ipstate_t *); 1157 if (is == NULL) { 1158 ATOMIC_INCL(ips_stats.iss_nomem); 1159 return NULL; 1160 } 1161 bcopy((char *)&ips, (char *)is, sizeof(*is)); 1162 /* 1163 * Do not do the modulous here, it is done in fr_stinsert(). 1164 */ 1165 if (fr != NULL) { 1166 (void) strncpy(is->is_group, fr->fr_group, FR_GROUPLEN); 1167 if (fr->fr_age[0] != 0) { 1168 is->is_tqehead[0] = fr_addtimeoutqueue(&ips_utqe, 1169 fr->fr_age[0]); 1170 is->is_sti.tqe_flags |= TQE_RULEBASED; 1171 } 1172 if (fr->fr_age[1] != 0) { 1173 is->is_tqehead[1] = fr_addtimeoutqueue(&ips_utqe, 1174 fr->fr_age[1]); 1175 is->is_sti.tqe_flags |= TQE_RULEBASED; 1176 } 1177 1178 is->is_tag = fr->fr_logtag; 1179 1180 /* 1181 * The name '-' is special for network interfaces and causes 1182 * a NULL name to be present, always, allowing packets to 1183 * match it, regardless of their interface. 1184 */ 1185 if ((fin->fin_ifp == NULL) || 1186 (fr->fr_ifnames[out << 1][0] == '-' && 1187 fr->fr_ifnames[out << 1][1] == '\0')) { 1188 is->is_ifp[out << 1] = fr->fr_ifas[0]; 1189 strncpy(is->is_ifname[out << 1], fr->fr_ifnames[0], 1190 sizeof(fr->fr_ifnames[0])); 1191 } else { 1192 is->is_ifp[out << 1] = fin->fin_ifp; 1193 COPYIFNAME(fin->fin_ifp, is->is_ifname[out << 1]); 1194 } 1195 1196 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1]; 1197 strncpy(is->is_ifname[(out << 1) + 1], fr->fr_ifnames[1], 1198 sizeof(fr->fr_ifnames[1])); 1199 1200 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2]; 1201 strncpy(is->is_ifname[((1 - out) << 1)], fr->fr_ifnames[2], 1202 sizeof(fr->fr_ifnames[2])); 1203 1204 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3]; 1205 strncpy(is->is_ifname[((1 - out) << 1) + 1], fr->fr_ifnames[3], 1206 sizeof(fr->fr_ifnames[3])); 1207 } else { 1208 pass = fr_flags; 1209 is->is_tag = FR_NOLOGTAG; 1210 1211 if (fin->fin_ifp != NULL) { 1212 is->is_ifp[out << 1] = fin->fin_ifp; 1213 COPYIFNAME(fin->fin_ifp, is->is_ifname[out << 1]); 1214 } 1215 } 1216 1217 /* 1218 * It may seem strange to set is_ref to 2, but fr_check() will call 1219 * fr_statederef() after calling fr_addstate() and the idea is to 1220 * have it exist at the end of fr_check() with is_ref == 1. 1221 */ 1222 is->is_ref = 2; 1223 is->is_pass = pass; 1224 is->is_pkts[0] = 0, is->is_bytes[0] = 0; 1225 is->is_pkts[1] = 0, is->is_bytes[1] = 0; 1226 is->is_pkts[2] = 0, is->is_bytes[2] = 0; 1227 is->is_pkts[3] = 0, is->is_bytes[3] = 0; 1228 if ((fin->fin_flx & FI_IGNORE) == 0) { 1229 is->is_pkts[out] = 1; 1230 is->is_bytes[out] = fin->fin_plen; 1231 is->is_flx[out][0] = fin->fin_flx & FI_CMP; 1232 is->is_flx[out][0] &= ~FI_OOW; 1233 } 1234 1235 if (pass & FR_STSTRICT) 1236 is->is_flags |= IS_STRICT; 1237 1238 if (pass & FR_STATESYNC) 1239 is->is_flags |= IS_STATESYNC; 1240 1241 /* 1242 * We want to check everything that is a property of this packet, 1243 * but we don't (automatically) care about it's fragment status as 1244 * this may change. 1245 */ 1246 is->is_v = fin->fin_v; 1247 is->is_opt[0] = fin->fin_optmsk; 1248 is->is_optmsk[0] = 0xffffffff; 1249 is->is_optmsk[1] = 0xffffffff; 1250 if (is->is_v == 6) { 1251 is->is_opt[0] &= ~0x8; 1252 is->is_optmsk[0] &= ~0x8; 1253 is->is_optmsk[1] &= ~0x8; 1254 } 1255 is->is_me = stsave; 1256 is->is_sec = fin->fin_secmsk; 1257 is->is_secmsk = 0xffff; 1258 is->is_auth = fin->fin_auth; 1259 is->is_authmsk = 0xffff; 1260 if (flags & (SI_WILDP|SI_WILDA)) { 1261 ATOMIC_INCL(ips_stats.iss_wild); 1262 } 1263 is->is_rulen = fin->fin_rule; 1264 1265 1266 if (pass & FR_LOGFIRST) 1267 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 1268 1269 READ_ENTER(&ipf_state); 1270 1271 fr_stinsert(is, fin->fin_rev); 1272 1273 if (fin->fin_p == IPPROTO_TCP) { 1274 /* 1275 * If we're creating state for a starting connection, start the 1276 * timer on it as we'll never see an error if it fails to 1277 * connect. 1278 */ 1279 (void) fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags); 1280 MUTEX_EXIT(&is->is_lock); 1281#ifdef IPFILTER_SCAN 1282 if ((is->is_flags & SI_CLONE) == 0) 1283 (void) ipsc_attachis(is); 1284#endif 1285 } else { 1286 MUTEX_EXIT(&is->is_lock); 1287 } 1288#ifdef IPFILTER_SYNC 1289 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0)) 1290 is->is_sync = ipfsync_new(SMC_STATE, fin, is); 1291#endif 1292 if (ipstate_logging) 1293 ipstate_log(is, ISL_NEW); 1294 1295 RWLOCK_EXIT(&ipf_state); 1296 fin->fin_state = is; 1297 fin->fin_rev = IP6_NEQ(&is->is_dst, &fin->fin_daddr); 1298 fin->fin_flx |= FI_STATE; 1299 if (fin->fin_flx & FI_FRAG) 1300 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 1301 1302 return is; 1303} 1304 1305 1306/* ------------------------------------------------------------------------ */ 1307/* Function: fr_tcpoptions */ 1308/* Returns: int - 1 == packet matches state entry, 0 == it does not, */ 1309/* -1 == packet has bad TCP options data */ 1310/* Parameters: fin(I) - pointer to packet information */ 1311/* tcp(I) - pointer to TCP packet header */ 1312/* td(I) - pointer to TCP data held as part of the state */ 1313/* */ 1314/* Look after the TCP header for any options and deal with those that are */ 1315/* present. Record details about those that we recogise. */ 1316/* ------------------------------------------------------------------------ */ 1317static int fr_tcpoptions(fin, tcp, td) 1318fr_info_t *fin; 1319tcphdr_t *tcp; 1320tcpdata_t *td; 1321{ 1322 int off, mlen, ol, i, len, retval; 1323 char buf[64], *s, opt; 1324 mb_t *m = NULL; 1325 1326 len = (TCP_OFF(tcp) << 2); 1327 if (fin->fin_dlen < len) 1328 return 0; 1329 len -= sizeof(*tcp); 1330 1331 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff; 1332 1333 m = fin->fin_m; 1334 mlen = MSGDSIZE(m) - off; 1335 if (len > mlen) { 1336 len = mlen; 1337 retval = 0; 1338 } else { 1339 retval = 1; 1340 } 1341 1342 COPYDATA(m, off, len, buf); 1343 1344 for (s = buf; len > 0; ) { 1345 opt = *s; 1346 if (opt == TCPOPT_EOL) 1347 break; 1348 else if (opt == TCPOPT_NOP) 1349 ol = 1; 1350 else { 1351 if (len < 2) 1352 break; 1353 ol = (int)*(s + 1); 1354 if (ol < 2 || ol > len) 1355 break; 1356 1357 /* 1358 * Extract the TCP options we are interested in out of 1359 * the header and store them in the the tcpdata struct. 1360 */ 1361 switch (opt) 1362 { 1363 case TCPOPT_WINDOW : 1364 if (ol == TCPOLEN_WINDOW) { 1365 i = (int)*(s + 2); 1366 if (i > TCP_WSCALE_MAX) 1367 i = TCP_WSCALE_MAX; 1368 else if (i < 0) 1369 i = 0; 1370 td->td_winscale = i; 1371 td->td_winflags |= TCP_WSCALE_SEEN| 1372 TCP_WSCALE_FIRST; 1373 } else 1374 retval = -1; 1375 break; 1376 case TCPOPT_MAXSEG : 1377 /* 1378 * So, if we wanted to set the TCP MAXSEG, 1379 * it should be done here... 1380 */ 1381 if (ol == TCPOLEN_MAXSEG) { 1382 i = (int)*(s + 2); 1383 i <<= 8; 1384 i += (int)*(s + 3); 1385 td->td_maxseg = i; 1386 } else 1387 retval = -1; 1388 break; 1389 case TCPOPT_SACK_PERMITTED : 1390 if (ol == TCPOLEN_SACK_PERMITTED) 1391 td->td_winflags |= TCP_SACK_PERMIT; 1392 else 1393 retval = -1; 1394 break; 1395 } 1396 } 1397 len -= ol; 1398 s += ol; 1399 } 1400 return retval; 1401} 1402 1403 1404/* ------------------------------------------------------------------------ */ 1405/* Function: fr_tcpstate */ 1406/* Returns: int - 1 == packet matches state entry, 0 == it does not */ 1407/* Parameters: fin(I) - pointer to packet information */ 1408/* tcp(I) - pointer to TCP packet header */ 1409/* is(I) - pointer to master state structure */ 1410/* */ 1411/* Check to see if a packet with TCP headers fits within the TCP window. */ 1412/* Change timeout depending on whether new packet is a SYN-ACK returning */ 1413/* for a SYN or a RST or FIN which indicate time to close up shop. */ 1414/* ------------------------------------------------------------------------ */ 1415static int fr_tcpstate(fin, tcp, is) 1416fr_info_t *fin; 1417tcphdr_t *tcp; 1418ipstate_t *is; 1419{ 1420 int source, ret = 0, flags; 1421 tcpdata_t *fdata, *tdata; 1422 1423 source = !fin->fin_rev; 1424 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) && 1425 (ntohs(is->is_sport) != fin->fin_data[0])) 1426 source = 0; 1427 fdata = &is->is_tcp.ts_data[!source]; 1428 tdata = &is->is_tcp.ts_data[source]; 1429 1430 MUTEX_ENTER(&is->is_lock); 1431 1432 /* 1433 * If a SYN packet is received for a connection that is on the way out 1434 * but hasn't yet departed then advance this session along the way. 1435 */ 1436 if ((tcp->th_flags & TH_OPENING) == TH_SYN) { 1437 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) && 1438 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) { 1439 is->is_state[!source] = IPF_TCPS_CLOSED; 1440 fr_movequeue(&is->is_sti, is->is_sti.tqe_ifq, 1441 &ips_deletetq); 1442 MUTEX_ENTER(&is->is_lock); 1443 return 0; 1444 } 1445 } 1446 1447 if (fr_tcpinwindow(fin, fdata, tdata, tcp, is->is_flags)) { 1448#ifdef IPFILTER_SCAN 1449 if (is->is_flags & (IS_SC_CLIENT|IS_SC_SERVER)) { 1450 ipsc_packet(fin, is); 1451 if (FR_ISBLOCK(is->is_pass)) { 1452 MUTEX_EXIT(&is->is_lock); 1453 return 1; 1454 } 1455 } 1456#endif 1457 1458 /* 1459 * Nearing end of connection, start timeout. 1460 */ 1461 ret = fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags); 1462 if (ret == 0) { 1463 MUTEX_EXIT(&is->is_lock); 1464 return 0; 1465 } 1466 1467 /* 1468 * set s0's as appropriate. Use syn-ack packet as it 1469 * contains both pieces of required information. 1470 */ 1471 /* 1472 * Window scale option is only present in SYN/SYN-ACK packet. 1473 * Compare with ~TH_FIN to mask out T/TCP setups. 1474 */ 1475 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL); 1476 if (flags == (TH_SYN|TH_ACK)) { 1477 is->is_s0[source] = ntohl(tcp->th_ack); 1478 is->is_s0[!source] = ntohl(tcp->th_seq) + 1; 1479 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 1480 if (fr_tcpoptions(fin, tcp, fdata) == -1) 1481 fin->fin_flx |= FI_BAD; 1482 } 1483 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1484 fr_checknewisn(fin, is); 1485 } else if (flags == TH_SYN) { 1486 is->is_s0[source] = ntohl(tcp->th_seq) + 1; 1487 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 1488 if (fr_tcpoptions(fin, tcp, fdata) == -1) 1489 fin->fin_flx |= FI_BAD; 1490 } 1491 1492 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1493 fr_checknewisn(fin, is); 1494 1495 } 1496 ret = 1; 1497 } else { 1498 fin->fin_flx |= FI_OOW; 1499 } 1500 MUTEX_EXIT(&is->is_lock); 1501 return ret; 1502} 1503 1504 1505/* ------------------------------------------------------------------------ */ 1506/* Function: fr_checknewisn */ 1507/* Returns: Nil */ 1508/* Parameters: fin(I) - pointer to packet information */ 1509/* is(I) - pointer to master state structure */ 1510/* */ 1511/* Check to see if this TCP connection is expecting and needs a new */ 1512/* sequence number for a particular direction of the connection. */ 1513/* */ 1514/* NOTE: This does not actually change the sequence numbers, only gets new */ 1515/* one ready. */ 1516/* ------------------------------------------------------------------------ */ 1517static void fr_checknewisn(fin, is) 1518fr_info_t *fin; 1519ipstate_t *is; 1520{ 1521 u_32_t sumd, old, new; 1522 tcphdr_t *tcp; 1523 int i; 1524 1525 i = fin->fin_rev; 1526 tcp = fin->fin_dp; 1527 1528 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) || 1529 ((i == 1) && !(is->is_flags & IS_ISNACK))) { 1530 old = ntohl(tcp->th_seq); 1531 new = fr_newisn(fin); 1532 is->is_isninc[i] = new - old; 1533 CALC_SUMD(old, new, sumd); 1534 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16); 1535 1536 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK); 1537 } 1538} 1539 1540 1541/* ------------------------------------------------------------------------ */ 1542/* Function: fr_tcpinwindow */ 1543/* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */ 1544/* Parameters: fin(I) - pointer to packet information */ 1545/* fdata(I) - pointer to tcp state informatio (forward) */ 1546/* tdata(I) - pointer to tcp state informatio (reverse) */ 1547/* tcp(I) - pointer to TCP packet header */ 1548/* */ 1549/* Given a packet has matched addresses and ports, check to see if it is */ 1550/* within the TCP data window. In a show of generosity, allow packets that */ 1551/* are within the window space behind the current sequence # as well. */ 1552/* ------------------------------------------------------------------------ */ 1553int fr_tcpinwindow(fin, fdata, tdata, tcp, flags) 1554fr_info_t *fin; 1555tcpdata_t *fdata, *tdata; 1556tcphdr_t *tcp; 1557int flags; 1558{ 1559 tcp_seq seq, ack, end; 1560 int ackskew, tcpflags; 1561 u_32_t win, maxwin; 1562 int dsize, inseq; 1563 1564 /* 1565 * Find difference between last checked packet and this packet. 1566 */ 1567 tcpflags = tcp->th_flags; 1568 seq = ntohl(tcp->th_seq); 1569 ack = ntohl(tcp->th_ack); 1570 if (tcpflags & TH_SYN) 1571 win = ntohs(tcp->th_win); 1572 else 1573 win = ntohs(tcp->th_win) << fdata->td_winscale; 1574 1575 /* 1576 * A window of 0 produces undesirable behaviour from this function. 1577 */ 1578 if (win == 0) 1579 win = 1; 1580 1581 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1582 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0); 1583 1584 /* 1585 * if window scaling is present, the scaling is only allowed 1586 * for windows not in the first SYN packet. In that packet the 1587 * window is 65535 to specify the largest window possible 1588 * for receivers not implementing the window scale option. 1589 * Currently, we do not assume TTCP here. That means that 1590 * if we see a second packet from a host (after the initial 1591 * SYN), we can assume that the receiver of the SYN did 1592 * already send back the SYN/ACK (and thus that we know if 1593 * the receiver also does window scaling) 1594 */ 1595 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) { 1596 fdata->td_winflags &= ~TCP_WSCALE_FIRST; 1597 fdata->td_maxwin = win; 1598 } 1599 1600 end = seq + dsize; 1601 1602 if ((fdata->td_end == 0) && 1603 (!(flags & IS_TCPFSM) || 1604 ((tcpflags & TH_OPENING) == TH_OPENING))) { 1605 /* 1606 * Must be a (outgoing) SYN-ACK in reply to a SYN. 1607 */ 1608 fdata->td_end = end - 1; 1609 fdata->td_maxwin = 1; 1610 fdata->td_maxend = end + win; 1611 } 1612 1613 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */ 1614 ack = tdata->td_end; 1615 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 1616 (ack == 0)) { 1617 /* gross hack to get around certain broken tcp stacks */ 1618 ack = tdata->td_end; 1619 } 1620 1621 maxwin = tdata->td_maxwin; 1622 ackskew = tdata->td_end - ack; 1623 1624 /* 1625 * Strict sequencing only allows in-order delivery. 1626 */ 1627 if ((flags & IS_STRICT) != 0) { 1628 if (seq != fdata->td_end) { 1629 return 0; 1630 } 1631 } 1632 1633#define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 1634#define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 1635 inseq = 0; 1636 if ((SEQ_GE(fdata->td_maxend, end)) && 1637 (SEQ_GE(seq, fdata->td_end - maxwin)) && 1638/* XXX what about big packets */ 1639#define MAXACKWINDOW 66000 1640 (-ackskew <= (MAXACKWINDOW)) && 1641 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) { 1642 inseq = 1; 1643 /* 1644 * Microsoft Windows will send the next packet to the right of the 1645 * window if SACK is in use. 1646 */ 1647 } else if ((seq == fdata->td_maxend) && (ackskew == 0) && 1648 (fdata->td_winflags & TCP_SACK_PERMIT) && 1649 (tdata->td_winflags & TCP_SACK_PERMIT)) { 1650 inseq = 1; 1651 /* 1652 * Sometimes a TCP RST will be generated with only the ACK field 1653 * set to non-zero. 1654 */ 1655 } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) && 1656 (ackskew >= -1) && (ackskew <= 1)) { 1657 inseq = 1; 1658 } else if (!(flags & IS_TCPFSM)) { 1659 int i; 1660 1661 i = (fin->fin_rev << 1) + fin->fin_out; 1662 1663#if 0 1664 if (is_pkts[i]0 == 0) { 1665 /* 1666 * Picking up a connection in the middle, the "next" 1667 * packet seen from a direction that is new should be 1668 * accepted, even if it appears out of sequence. 1669 */ 1670 inseq = 1; 1671 } else 1672#endif 1673 if (!(fdata->td_winflags & 1674 (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) { 1675 /* 1676 * No TCPFSM and no window scaling, so make some 1677 * extra guesses. 1678 */ 1679 if ((seq == fdata->td_maxend) && (ackskew == 0)) 1680 inseq = 1; 1681 else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin)) 1682 inseq = 1; 1683 } 1684 } 1685 1686 /* TRACE(inseq, fdata, tdata, seq, end, ack, ackskew, win, maxwin) */ 1687 1688 if (inseq) { 1689 /* if ackskew < 0 then this should be due to fragmented 1690 * packets. There is no way to know the length of the 1691 * total packet in advance. 1692 * We do know the total length from the fragment cache though. 1693 * Note however that there might be more sessions with 1694 * exactly the same source and destination parameters in the 1695 * state cache (and source and destination is the only stuff 1696 * that is saved in the fragment cache). Note further that 1697 * some TCP connections in the state cache are hashed with 1698 * sport and dport as well which makes it not worthwhile to 1699 * look for them. 1700 * Thus, when ackskew is negative but still seems to belong 1701 * to this session, we bump up the destinations end value. 1702 */ 1703 if (ackskew < 0) 1704 tdata->td_end = ack; 1705 1706 /* update max window seen */ 1707 if (fdata->td_maxwin < win) 1708 fdata->td_maxwin = win; 1709 if (SEQ_GT(end, fdata->td_end)) 1710 fdata->td_end = end; 1711 if (SEQ_GE(ack + win, tdata->td_maxend)) 1712 tdata->td_maxend = ack + win; 1713 return 1; 1714 } 1715 return 0; 1716} 1717 1718 1719/* ------------------------------------------------------------------------ */ 1720/* Function: fr_stclone */ 1721/* Returns: ipstate_t* - NULL == cloning failed, */ 1722/* else pointer to new state structure */ 1723/* Parameters: fin(I) - pointer to packet information */ 1724/* tcp(I) - pointer to TCP/UDP header */ 1725/* is(I) - pointer to master state structure */ 1726/* */ 1727/* Create a "duplcate" state table entry from the master. */ 1728/* ------------------------------------------------------------------------ */ 1729static ipstate_t *fr_stclone(fin, tcp, is) 1730fr_info_t *fin; 1731tcphdr_t *tcp; 1732ipstate_t *is; 1733{ 1734 ipstate_t *clone; 1735 u_32_t send; 1736 1737 if (ips_num == fr_statemax) { 1738 ATOMIC_INCL(ips_stats.iss_max); 1739 fr_state_doflush = 1; 1740 return NULL; 1741 } 1742 KMALLOC(clone, ipstate_t *); 1743 if (clone == NULL) 1744 return NULL; 1745 bcopy((char *)is, (char *)clone, sizeof(*clone)); 1746 1747 MUTEX_NUKE(&clone->is_lock); 1748 1749 clone->is_die = ONE_DAY + fr_ticks; 1750 clone->is_state[0] = 0; 1751 clone->is_state[1] = 0; 1752 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1753 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 1754 ((tcp->th_flags & TH_FIN) ? 1 : 0); 1755 1756 if (fin->fin_rev == 1) { 1757 clone->is_dend = send; 1758 clone->is_maxdend = send; 1759 clone->is_send = 0; 1760 clone->is_maxswin = 1; 1761 clone->is_maxdwin = ntohs(tcp->th_win); 1762 if (clone->is_maxdwin == 0) 1763 clone->is_maxdwin = 1; 1764 } else { 1765 clone->is_send = send; 1766 clone->is_maxsend = send; 1767 clone->is_dend = 0; 1768 clone->is_maxdwin = 1; 1769 clone->is_maxswin = ntohs(tcp->th_win); 1770 if (clone->is_maxswin == 0) 1771 clone->is_maxswin = 1; 1772 } 1773 1774 clone->is_flags &= ~SI_CLONE; 1775 clone->is_flags |= SI_CLONED; 1776 fr_stinsert(clone, fin->fin_rev); 1777 clone->is_ref = 2; 1778 if (clone->is_p == IPPROTO_TCP) { 1779 (void) fr_tcp_age(&clone->is_sti, fin, ips_tqtqb, 1780 clone->is_flags); 1781 } 1782 MUTEX_EXIT(&clone->is_lock); 1783#ifdef IPFILTER_SCAN 1784 (void) ipsc_attachis(is); 1785#endif 1786#ifdef IPFILTER_SYNC 1787 if (is->is_flags & IS_STATESYNC) 1788 clone->is_sync = ipfsync_new(SMC_STATE, fin, clone); 1789#endif 1790 return clone; 1791} 1792 1793 1794/* ------------------------------------------------------------------------ */ 1795/* Function: fr_matchsrcdst */ 1796/* Returns: Nil */ 1797/* Parameters: fin(I) - pointer to packet information */ 1798/* is(I) - pointer to state structure */ 1799/* src(I) - pointer to source address */ 1800/* dst(I) - pointer to destination address */ 1801/* tcp(I) - pointer to TCP/UDP header */ 1802/* */ 1803/* Match a state table entry against an IP packet. The logic below is that */ 1804/* ret gets set to one if the match succeeds, else remains 0. If it is */ 1805/* still 0 after the test. no match. */ 1806/* ------------------------------------------------------------------------ */ 1807static ipstate_t *fr_matchsrcdst(fin, is, src, dst, tcp, cmask) 1808fr_info_t *fin; 1809ipstate_t *is; 1810i6addr_t *src, *dst; 1811tcphdr_t *tcp; 1812u_32_t cmask; 1813{ 1814 int ret = 0, rev, out, flags, flx = 0, idx; 1815 u_short sp, dp; 1816 u_32_t cflx; 1817 void *ifp; 1818 1819 rev = IP6_NEQ(&is->is_dst, dst); 1820 ifp = fin->fin_ifp; 1821 out = fin->fin_out; 1822 flags = is->is_flags; 1823 sp = 0; 1824 dp = 0; 1825 1826 if (tcp != NULL) { 1827 sp = htons(fin->fin_sport); 1828 dp = ntohs(fin->fin_dport); 1829 } 1830 if (!rev) { 1831 if (tcp != NULL) { 1832 if (!(flags & SI_W_SPORT) && (sp != is->is_sport)) 1833 rev = 1; 1834 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport)) 1835 rev = 1; 1836 } 1837 } 1838 1839 idx = (out << 1) + rev; 1840 1841 /* 1842 * If the interface for this 'direction' is set, make sure it matches. 1843 * An interface name that is not set matches any, as does a name of *. 1844 */ 1845 if ((is->is_ifp[idx] == ifp) || (is->is_ifp[idx] == NULL && 1846 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '-' || 1847 *is->is_ifname[idx] == '*'))) 1848 ret = 1; 1849 1850 if (ret == 0) 1851 return NULL; 1852 ret = 0; 1853 1854 /* 1855 * Match addresses and ports. 1856 */ 1857 if (rev == 0) { 1858 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) && 1859 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) { 1860 if (tcp) { 1861 if ((sp == is->is_sport || flags & SI_W_SPORT)&& 1862 (dp == is->is_dport || flags & SI_W_DPORT)) 1863 ret = 1; 1864 } else { 1865 ret = 1; 1866 } 1867 } 1868 } else { 1869 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) && 1870 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) { 1871 if (tcp) { 1872 if ((dp == is->is_sport || flags & SI_W_SPORT)&& 1873 (sp == is->is_dport || flags & SI_W_DPORT)) 1874 ret = 1; 1875 } else { 1876 ret = 1; 1877 } 1878 } 1879 } 1880 1881 if (ret == 0) 1882 return NULL; 1883 1884 /* 1885 * Whether or not this should be here, is questionable, but the aim 1886 * is to get this out of the main line. 1887 */ 1888 if (tcp == NULL) 1889 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED); 1890 1891 /* 1892 * Only one of the source or destination address can be flaged as a 1893 * wildcard. Fill in the missing address, if set. 1894 * For IPv6, if the address being copied in is multicast, then 1895 * don't reset the wild flag - multicast causes it to be set in the 1896 * first place! 1897 */ 1898 if ((flags & (SI_W_SADDR|SI_W_DADDR))) { 1899 fr_ip_t *fi = &fin->fin_fi; 1900 1901 if ((flags & SI_W_SADDR) != 0) { 1902 if (rev == 0) { 1903#ifdef USE_INET6 1904 if (is->is_v == 6 && 1905 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 1906 /*EMPTY*/; 1907 else 1908#endif 1909 { 1910 is->is_src = fi->fi_src; 1911 is->is_flags &= ~SI_W_SADDR; 1912 } 1913 } else { 1914#ifdef USE_INET6 1915 if (is->is_v == 6 && 1916 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 1917 /*EMPTY*/; 1918 else 1919#endif 1920 { 1921 is->is_src = fi->fi_dst; 1922 is->is_flags &= ~SI_W_SADDR; 1923 } 1924 } 1925 } else if ((flags & SI_W_DADDR) != 0) { 1926 if (rev == 0) { 1927#ifdef USE_INET6 1928 if (is->is_v == 6 && 1929 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 1930 /*EMPTY*/; 1931 else 1932#endif 1933 { 1934 is->is_dst = fi->fi_dst; 1935 is->is_flags &= ~SI_W_DADDR; 1936 } 1937 } else { 1938#ifdef USE_INET6 1939 if (is->is_v == 6 && 1940 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 1941 /*EMPTY*/; 1942 else 1943#endif 1944 { 1945 is->is_dst = fi->fi_src; 1946 is->is_flags &= ~SI_W_DADDR; 1947 } 1948 } 1949 } 1950 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) { 1951 ATOMIC_DECL(ips_stats.iss_wild); 1952 } 1953 } 1954 1955 flx = fin->fin_flx & cmask; 1956 cflx = is->is_flx[out][rev]; 1957 1958 /* 1959 * Match up any flags set from IP options. 1960 */ 1961 if ((cflx && (flx != (cflx & cmask))) || 1962 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) || 1963 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) || 1964 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) 1965 return NULL; 1966 1967 /* 1968 * Only one of the source or destination port can be flagged as a 1969 * wildcard. When filling it in, fill in a copy of the matched entry 1970 * if it has the cloning flag set. 1971 */ 1972 if ((fin->fin_flx & FI_IGNORE) != 0) { 1973 fin->fin_rev = rev; 1974 return is; 1975 } 1976 1977 if ((flags & (SI_W_SPORT|SI_W_DPORT))) { 1978 if ((flags & SI_CLONE) != 0) { 1979 ipstate_t *clone; 1980 1981 clone = fr_stclone(fin, tcp, is); 1982 if (clone == NULL) 1983 return NULL; 1984 is = clone; 1985 } else { 1986 ATOMIC_DECL(ips_stats.iss_wild); 1987 } 1988 1989 if ((flags & SI_W_SPORT) != 0) { 1990 if (rev == 0) { 1991 is->is_sport = sp; 1992 is->is_send = ntohl(tcp->th_seq); 1993 } else { 1994 is->is_sport = dp; 1995 is->is_send = ntohl(tcp->th_ack); 1996 } 1997 is->is_maxsend = is->is_send + 1; 1998 } else if ((flags & SI_W_DPORT) != 0) { 1999 if (rev == 0) { 2000 is->is_dport = dp; 2001 is->is_dend = ntohl(tcp->th_ack); 2002 } else { 2003 is->is_dport = sp; 2004 is->is_dend = ntohl(tcp->th_seq); 2005 } 2006 is->is_maxdend = is->is_dend + 1; 2007 } 2008 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT); 2009 if ((flags & SI_CLONED) && ipstate_logging) 2010 ipstate_log(is, ISL_CLONE); 2011 } 2012 2013 ret = -1; 2014 2015 if (is->is_flx[out][rev] == 0) { 2016 is->is_flx[out][rev] = flx; 2017 is->is_opt[rev] = fin->fin_optmsk; 2018 if (is->is_v == 6) { 2019 is->is_opt[rev] &= ~0x8; 2020 is->is_optmsk[rev] &= ~0x8; 2021 } 2022 } 2023 2024 /* 2025 * Check if the interface name for this "direction" is set and if not, 2026 * fill it in. 2027 */ 2028 if (is->is_ifp[idx] == NULL && 2029 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) { 2030 is->is_ifp[idx] = ifp; 2031 COPYIFNAME(ifp, is->is_ifname[idx]); 2032 } 2033 fin->fin_rev = rev; 2034 return is; 2035} 2036 2037 2038/* ------------------------------------------------------------------------ */ 2039/* Function: fr_checkicmpmatchingstate */ 2040/* Returns: Nil */ 2041/* Parameters: fin(I) - pointer to packet information */ 2042/* */ 2043/* If we've got an ICMP error message, using the information stored in the */ 2044/* ICMP packet, look for a matching state table entry. */ 2045/* */ 2046/* If we return NULL then no lock on ipf_state is held. */ 2047/* If we return non-null then a read-lock on ipf_state is held. */ 2048/* ------------------------------------------------------------------------ */ 2049static ipstate_t *fr_checkicmpmatchingstate(fin) 2050fr_info_t *fin; 2051{ 2052 ipstate_t *is, **isp; 2053 u_short sport, dport; 2054 u_char pr; 2055 int backward, i, oi; 2056 i6addr_t dst, src; 2057 struct icmp *ic; 2058 u_short savelen; 2059 icmphdr_t *icmp; 2060 fr_info_t ofin; 2061 tcphdr_t *tcp; 2062 int type, len; 2063 ip_t *oip; 2064 u_int hv; 2065 2066 /* 2067 * Does it at least have the return (basic) IP header ? 2068 * Is it an actual recognised ICMP error type? 2069 * Only a basic IP header (no options) should be with 2070 * an ICMP error header. 2071 */ 2072 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) || 2073 (fin->fin_plen < ICMPERR_MINPKTLEN) || 2074 !(fin->fin_flx & FI_ICMPERR)) 2075 return NULL; 2076 ic = fin->fin_dp; 2077 type = ic->icmp_type; 2078 2079 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 2080 /* 2081 * Check if the at least the old IP header (with options) and 2082 * 8 bytes of payload is present. 2083 */ 2084 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) 2085 return NULL; 2086 2087 /* 2088 * Sanity Checks. 2089 */ 2090 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 2091 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) 2092 return NULL; 2093 2094 /* 2095 * Is the buffer big enough for all of it ? It's the size of the IP 2096 * header claimed in the encapsulated part which is of concern. It 2097 * may be too big to be in this buffer but not so big that it's 2098 * outside the ICMP packet, leading to TCP deref's causing problems. 2099 * This is possible because we don't know how big oip_hl is when we 2100 * do the pullup early in fr_check() and thus can't guarantee it is 2101 * all here now. 2102 */ 2103#ifdef _KERNEL 2104 { 2105 mb_t *m; 2106 2107 m = fin->fin_m; 2108# if defined(MENTAT) 2109 if ((char *)oip + len > (char *)m->b_wptr) 2110 return NULL; 2111# else 2112 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) 2113 return NULL; 2114# endif 2115 } 2116#endif 2117 bcopy((char *)fin, (char *)&ofin, sizeof(*fin)); 2118 2119 /* 2120 * in the IPv4 case we must zero the i6addr union otherwise 2121 * the IP6_EQ and IP6_NEQ macros produce the wrong results because 2122 * of the 'junk' in the unused part of the union 2123 */ 2124 bzero((char *)&src, sizeof(src)); 2125 bzero((char *)&dst, sizeof(dst)); 2126 2127 /* 2128 * we make an fin entry to be able to feed it to 2129 * matchsrcdst note that not all fields are encessary 2130 * but this is the cleanest way. Note further we fill 2131 * in fin_mp such that if someone uses it we'll get 2132 * a kernel panic. fr_matchsrcdst does not use this. 2133 * 2134 * watch out here, as ip is in host order and oip in network 2135 * order. Any change we make must be undone afterwards, like 2136 * oip->ip_off - it is still in network byte order so fix it. 2137 */ 2138 savelen = oip->ip_len; 2139 oip->ip_len = len; 2140 oip->ip_off = ntohs(oip->ip_off); 2141 2142 ofin.fin_flx = FI_NOCKSUM; 2143 ofin.fin_v = 4; 2144 ofin.fin_ip = oip; 2145 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 2146 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 2147 (void) fr_makefrip(IP_HL(oip) << 2, oip, &ofin); 2148 ofin.fin_ifp = fin->fin_ifp; 2149 ofin.fin_out = !fin->fin_out; 2150 /* 2151 * Reset the short and bad flag here because in fr_matchsrcdst() 2152 * the flags for the current packet (fin_flx) are compared against 2153 * those for the existing session. 2154 */ 2155 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 2156 2157 /* 2158 * Put old values of ip_len and ip_off back as we don't know 2159 * if we have to forward the packet (or process it again. 2160 */ 2161 oip->ip_len = savelen; 2162 oip->ip_off = htons(oip->ip_off); 2163 2164 switch (oip->ip_p) 2165 { 2166 case IPPROTO_ICMP : 2167 /* 2168 * an ICMP error can only be generated as a result of an 2169 * ICMP query, not as the response on an ICMP error 2170 * 2171 * XXX theoretically ICMP_ECHOREP and the other reply's are 2172 * ICMP query's as well, but adding them here seems strange XXX 2173 */ 2174 if ((ofin.fin_flx & FI_ICMPERR) != 0) 2175 return NULL; 2176 2177 /* 2178 * perform a lookup of the ICMP packet in the state table 2179 */ 2180 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2181 hv = (pr = oip->ip_p); 2182 src.in4 = oip->ip_src; 2183 hv += src.in4.s_addr; 2184 dst.in4 = oip->ip_dst; 2185 hv += dst.in4.s_addr; 2186 hv += icmp->icmp_id; 2187 hv = DOUBLE_HASH(hv); 2188 2189 READ_ENTER(&ipf_state); 2190 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2191 isp = &is->is_hnext; 2192 if ((is->is_p != pr) || (is->is_v != 4)) 2193 continue; 2194 if (is->is_pass & FR_NOICMPERR) 2195 continue; 2196 is = fr_matchsrcdst(&ofin, is, &src, &dst, 2197 NULL, FI_ICMPCMP); 2198 if (is != NULL) { 2199 /* 2200 * i : the index of this packet (the icmp 2201 * unreachable) 2202 * oi : the index of the original packet found 2203 * in the icmp header (i.e. the packet 2204 * causing this icmp) 2205 * backward : original packet was backward 2206 * compared to the state 2207 */ 2208 backward = IP6_NEQ(&is->is_src, &src); 2209 fin->fin_rev = !backward; 2210 i = (!backward << 1) + fin->fin_out; 2211 oi = (backward << 1) + ofin.fin_out; 2212 if (is->is_icmppkts[i] > is->is_pkts[oi]) 2213 continue; 2214 ips_stats.iss_hits++; 2215 is->is_icmppkts[i]++; 2216 return is; 2217 } 2218 } 2219 RWLOCK_EXIT(&ipf_state); 2220 return NULL; 2221 case IPPROTO_TCP : 2222 case IPPROTO_UDP : 2223 break; 2224 default : 2225 return NULL; 2226 } 2227 2228 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2229 dport = tcp->th_dport; 2230 sport = tcp->th_sport; 2231 2232 hv = (pr = oip->ip_p); 2233 src.in4 = oip->ip_src; 2234 hv += src.in4.s_addr; 2235 dst.in4 = oip->ip_dst; 2236 hv += dst.in4.s_addr; 2237 hv += dport; 2238 hv += sport; 2239 hv = DOUBLE_HASH(hv); 2240 2241 READ_ENTER(&ipf_state); 2242 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2243 isp = &is->is_hnext; 2244 /* 2245 * Only allow this icmp though if the 2246 * encapsulated packet was allowed through the 2247 * other way around. Note that the minimal amount 2248 * of info present does not allow for checking against 2249 * tcp internals such as seq and ack numbers. Only the 2250 * ports are known to be present and can be even if the 2251 * short flag is set. 2252 */ 2253 if ((is->is_p == pr) && (is->is_v == 4) && 2254 (is = fr_matchsrcdst(&ofin, is, &src, &dst, 2255 tcp, FI_ICMPCMP))) { 2256 /* 2257 * i : the index of this packet (the icmp unreachable) 2258 * oi : the index of the original packet found in the 2259 * icmp header (i.e. the packet causing this icmp) 2260 * backward : original packet was backward compared to 2261 * the state 2262 */ 2263 backward = IP6_NEQ(&is->is_src, &src); 2264 fin->fin_rev = !backward; 2265 i = (!backward << 1) + fin->fin_out; 2266 oi = (backward << 1) + ofin.fin_out; 2267 2268 if (((is->is_pass & FR_NOICMPERR) != 0) || 2269 (is->is_icmppkts[i] > is->is_pkts[oi])) 2270 break; 2271 ips_stats.iss_hits++; 2272 is->is_icmppkts[i]++; 2273 /* 2274 * we deliberately do not touch the timeouts 2275 * for the accompanying state table entry. 2276 * It remains to be seen if that is correct. XXX 2277 */ 2278 return is; 2279 } 2280 } 2281 RWLOCK_EXIT(&ipf_state); 2282 return NULL; 2283} 2284 2285 2286/* ------------------------------------------------------------------------ */ 2287/* Function: fr_ipsmove */ 2288/* Returns: Nil */ 2289/* Parameters: is(I) - pointer to state table entry */ 2290/* hv(I) - new hash value for state table entry */ 2291/* Write Locks: ipf_state */ 2292/* */ 2293/* Move a state entry from one position in the hash table to another. */ 2294/* ------------------------------------------------------------------------ */ 2295static void fr_ipsmove(is, hv) 2296ipstate_t *is; 2297u_int hv; 2298{ 2299 ipstate_t **isp; 2300 u_int hvm; 2301 2302 ASSERT(rw_read_locked(&ipf_state.ipf_lk) == 0); 2303 2304 hvm = is->is_hv; 2305 /* 2306 * Remove the hash from the old location... 2307 */ 2308 isp = is->is_phnext; 2309 if (is->is_hnext) 2310 is->is_hnext->is_phnext = isp; 2311 *isp = is->is_hnext; 2312 if (ips_table[hvm] == NULL) 2313 ips_stats.iss_inuse--; 2314 ips_stats.iss_bucketlen[hvm]--; 2315 2316 /* 2317 * ...and put the hash in the new one. 2318 */ 2319 hvm = DOUBLE_HASH(hv); 2320 is->is_hv = hvm; 2321 isp = &ips_table[hvm]; 2322 if (*isp) 2323 (*isp)->is_phnext = &is->is_hnext; 2324 else 2325 ips_stats.iss_inuse++; 2326 ips_stats.iss_bucketlen[hvm]++; 2327 is->is_phnext = isp; 2328 is->is_hnext = *isp; 2329 *isp = is; 2330} 2331 2332 2333/* ------------------------------------------------------------------------ */ 2334/* Function: fr_stlookup */ 2335/* Returns: ipstate_t* - NULL == no matching state found, */ 2336/* else pointer to state information is returned */ 2337/* Parameters: fin(I) - pointer to packet information */ 2338/* tcp(I) - pointer to TCP/UDP header. */ 2339/* */ 2340/* Search the state table for a matching entry to the packet described by */ 2341/* the contents of *fin. */ 2342/* */ 2343/* If we return NULL then no lock on ipf_state is held. */ 2344/* If we return non-null then a read-lock on ipf_state is held. */ 2345/* ------------------------------------------------------------------------ */ 2346ipstate_t *fr_stlookup(fin, tcp, ifqp) 2347fr_info_t *fin; 2348tcphdr_t *tcp; 2349ipftq_t **ifqp; 2350{ 2351 u_int hv, hvm, pr, v, tryagain; 2352 ipstate_t *is, **isp; 2353 u_short dport, sport; 2354 i6addr_t src, dst; 2355 struct icmp *ic; 2356 ipftq_t *ifq; 2357 int oow; 2358 2359 is = NULL; 2360 ifq = NULL; 2361 tcp = fin->fin_dp; 2362 ic = (struct icmp *)tcp; 2363 hv = (pr = fin->fin_fi.fi_p); 2364 src = fin->fin_fi.fi_src; 2365 dst = fin->fin_fi.fi_dst; 2366 hv += src.in4.s_addr; 2367 hv += dst.in4.s_addr; 2368 2369 v = fin->fin_fi.fi_v; 2370#ifdef USE_INET6 2371 if (v == 6) { 2372 hv += fin->fin_fi.fi_src.i6[1]; 2373 hv += fin->fin_fi.fi_src.i6[2]; 2374 hv += fin->fin_fi.fi_src.i6[3]; 2375 2376 if ((fin->fin_p == IPPROTO_ICMPV6) && 2377 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) { 2378 hv -= dst.in4.s_addr; 2379 } else { 2380 hv += fin->fin_fi.fi_dst.i6[1]; 2381 hv += fin->fin_fi.fi_dst.i6[2]; 2382 hv += fin->fin_fi.fi_dst.i6[3]; 2383 } 2384 } 2385#endif 2386 2387 /* 2388 * Search the hash table for matching packet header info. 2389 */ 2390 switch (pr) 2391 { 2392#ifdef USE_INET6 2393 case IPPROTO_ICMPV6 : 2394 tryagain = 0; 2395 if (v == 6) { 2396 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 2397 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 2398 hv += ic->icmp_id; 2399 } 2400 } 2401 READ_ENTER(&ipf_state); 2402icmp6again: 2403 hvm = DOUBLE_HASH(hv); 2404 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2405 isp = &is->is_hnext; 2406 /* 2407 * If a connection is about to be deleted, no packets 2408 * are allowed to match it. 2409 */ 2410 if (is->is_sti.tqe_ifq == &ips_deletetq) 2411 continue; 2412 2413 if ((is->is_p != pr) || (is->is_v != v)) 2414 continue; 2415 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2416 if (is != NULL && 2417 fr_matchicmpqueryreply(v, &is->is_icmp, 2418 ic, fin->fin_rev)) { 2419 if (fin->fin_rev) 2420 ifq = &ips_icmpacktq; 2421 else 2422 ifq = &ips_icmptq; 2423 break; 2424 } 2425 } 2426 2427 if (is != NULL) { 2428 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) { 2429 hv += fin->fin_fi.fi_src.i6[0]; 2430 hv += fin->fin_fi.fi_src.i6[1]; 2431 hv += fin->fin_fi.fi_src.i6[2]; 2432 hv += fin->fin_fi.fi_src.i6[3]; 2433 fr_ipsmove(is, hv); 2434 MUTEX_DOWNGRADE(&ipf_state); 2435 } 2436 break; 2437 } 2438 RWLOCK_EXIT(&ipf_state); 2439 2440 /* 2441 * No matching icmp state entry. Perhaps this is a 2442 * response to another state entry. 2443 * 2444 * XXX With some ICMP6 packets, the "other" address is already 2445 * in the packet, after the ICMP6 header, and this could be 2446 * used in place of the multicast address. However, taking 2447 * advantage of this requires some significant code changes 2448 * to handle the specific types where that is the case. 2449 */ 2450 if ((ips_stats.iss_wild != 0) && (v == 6) && (tryagain == 0) && 2451 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) { 2452 hv -= fin->fin_fi.fi_src.i6[0]; 2453 hv -= fin->fin_fi.fi_src.i6[1]; 2454 hv -= fin->fin_fi.fi_src.i6[2]; 2455 hv -= fin->fin_fi.fi_src.i6[3]; 2456 tryagain = 1; 2457 WRITE_ENTER(&ipf_state); 2458 goto icmp6again; 2459 } 2460 2461 is = fr_checkicmp6matchingstate(fin); 2462 if (is != NULL) 2463 return is; 2464 break; 2465#endif 2466 2467 case IPPROTO_ICMP : 2468 if (v == 4) { 2469 hv += ic->icmp_id; 2470 } 2471 hv = DOUBLE_HASH(hv); 2472 READ_ENTER(&ipf_state); 2473 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 2474 isp = &is->is_hnext; 2475 if ((is->is_p != pr) || (is->is_v != v)) 2476 continue; 2477 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2478 if ((is != NULL) && 2479 (ic->icmp_id == is->is_icmp.ici_id) && 2480 fr_matchicmpqueryreply(v, &is->is_icmp, 2481 ic, fin->fin_rev)) { 2482 if (fin->fin_rev) 2483 ifq = &ips_icmpacktq; 2484 else 2485 ifq = &ips_icmptq; 2486 break; 2487 } 2488 } 2489 if (is == NULL) { 2490 RWLOCK_EXIT(&ipf_state); 2491 } 2492 break; 2493 2494 case IPPROTO_TCP : 2495 case IPPROTO_UDP : 2496 ifqp = NULL; 2497 sport = htons(fin->fin_data[0]); 2498 hv += sport; 2499 dport = htons(fin->fin_data[1]); 2500 hv += dport; 2501 oow = 0; 2502 tryagain = 0; 2503 READ_ENTER(&ipf_state); 2504retry_tcpudp: 2505 hvm = DOUBLE_HASH(hv); 2506 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2507 isp = &is->is_hnext; 2508 if ((is->is_p != pr) || (is->is_v != v)) 2509 continue; 2510 fin->fin_flx &= ~FI_OOW; 2511 is = fr_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP); 2512 if (is != NULL) { 2513 if (pr == IPPROTO_TCP) { 2514 if (!fr_tcpstate(fin, tcp, is)) { 2515 oow |= fin->fin_flx & FI_OOW; 2516 continue; 2517 } 2518 } 2519 break; 2520 } 2521 } 2522 if (is != NULL) { 2523 if (tryagain && 2524 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) { 2525 hv += dport; 2526 hv += sport; 2527 fr_ipsmove(is, hv); 2528 MUTEX_DOWNGRADE(&ipf_state); 2529 } 2530 break; 2531 } 2532 RWLOCK_EXIT(&ipf_state); 2533 2534 if (!tryagain && ips_stats.iss_wild) { 2535 hv -= dport; 2536 hv -= sport; 2537 tryagain = 1; 2538 WRITE_ENTER(&ipf_state); 2539 goto retry_tcpudp; 2540 } 2541 fin->fin_flx |= oow; 2542 break; 2543 2544#if 0 2545 case IPPROTO_GRE : 2546 gre = fin->fin_dp; 2547 if (GRE_REV(gre->gr_flags) == 1) { 2548 hv += gre->gr_call; 2549 } 2550 /* FALLTHROUGH */ 2551#endif 2552 default : 2553 ifqp = NULL; 2554 hvm = DOUBLE_HASH(hv); 2555 READ_ENTER(&ipf_state); 2556 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) { 2557 isp = &is->is_hnext; 2558 if ((is->is_p != pr) || (is->is_v != v)) 2559 continue; 2560 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2561 if (is != NULL) { 2562 ifq = &ips_iptq; 2563 break; 2564 } 2565 } 2566 if (is == NULL) { 2567 RWLOCK_EXIT(&ipf_state); 2568 } 2569 break; 2570 } 2571 2572 if (is != NULL) { 2573 if (((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) && 2574 (is->is_tqehead[fin->fin_rev] != NULL)) 2575 ifq = is->is_tqehead[fin->fin_rev]; 2576 if (ifq != NULL && ifqp != NULL) 2577 *ifqp = ifq; 2578 } 2579 return is; 2580} 2581 2582 2583/* ------------------------------------------------------------------------ */ 2584/* Function: fr_updatestate */ 2585/* Returns: Nil */ 2586/* Parameters: fin(I) - pointer to packet information */ 2587/* is(I) - pointer to state table entry */ 2588/* Read Locks: ipf_state */ 2589/* */ 2590/* Updates packet and byte counters for a newly received packet. Seeds the */ 2591/* fragment cache with a new entry as required. */ 2592/* ------------------------------------------------------------------------ */ 2593void fr_updatestate(fin, is, ifq) 2594fr_info_t *fin; 2595ipstate_t *is; 2596ipftq_t *ifq; 2597{ 2598 ipftqent_t *tqe; 2599 int i, pass; 2600 2601 i = (fin->fin_rev << 1) + fin->fin_out; 2602 2603 /* 2604 * For TCP packets, ifq == NULL. For all others, check if this new 2605 * queue is different to the last one it was on and move it if so. 2606 */ 2607 tqe = &is->is_sti; 2608 MUTEX_ENTER(&is->is_lock); 2609 if ((tqe->tqe_flags & TQE_RULEBASED) != 0) 2610 ifq = is->is_tqehead[fin->fin_rev]; 2611 2612 if (ifq != NULL) 2613 fr_movequeue(tqe, tqe->tqe_ifq, ifq); 2614 2615 is->is_pkts[i]++; 2616 is->is_bytes[i] += fin->fin_plen; 2617 MUTEX_EXIT(&is->is_lock); 2618 2619#ifdef IPFILTER_SYNC 2620 if (is->is_flags & IS_STATESYNC) 2621 ipfsync_update(SMC_STATE, fin, is->is_sync); 2622#endif 2623 2624 ATOMIC_INCL(ips_stats.iss_hits); 2625 2626 fin->fin_fr = is->is_rule; 2627 2628 /* 2629 * If this packet is a fragment and the rule says to track fragments, 2630 * then create a new fragment cache entry. 2631 */ 2632 pass = is->is_pass; 2633 if ((fin->fin_flx & FI_FRAG) && FR_ISPASS(pass)) 2634 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 2635} 2636 2637 2638/* ------------------------------------------------------------------------ */ 2639/* Function: fr_checkstate */ 2640/* Returns: frentry_t* - NULL == search failed, */ 2641/* else pointer to rule for matching state */ 2642/* Parameters: ifp(I) - pointer to interface */ 2643/* passp(I) - pointer to filtering result flags */ 2644/* */ 2645/* Check if a packet is associated with an entry in the state table. */ 2646/* ------------------------------------------------------------------------ */ 2647frentry_t *fr_checkstate(fin, passp) 2648fr_info_t *fin; 2649u_32_t *passp; 2650{ 2651 ipstate_t *is; 2652 frentry_t *fr; 2653 tcphdr_t *tcp; 2654 ipftq_t *ifq; 2655 u_int pass; 2656 2657 if (fr_state_lock || (ips_list == NULL) || 2658 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 2659 return NULL; 2660 2661 is = NULL; 2662 if ((fin->fin_flx & FI_TCPUDP) || 2663 (fin->fin_fi.fi_p == IPPROTO_ICMP) 2664#ifdef USE_INET6 2665 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6) 2666#endif 2667 ) 2668 tcp = fin->fin_dp; 2669 else 2670 tcp = NULL; 2671 2672 /* 2673 * Search the hash table for matching packet header info. 2674 */ 2675 ifq = NULL; 2676 is = fin->fin_state; 2677 if (is == NULL) 2678 is = fr_stlookup(fin, tcp, &ifq); 2679 switch (fin->fin_p) 2680 { 2681#ifdef USE_INET6 2682 case IPPROTO_ICMPV6 : 2683 if (is != NULL) 2684 break; 2685 if (fin->fin_v == 6) { 2686 is = fr_checkicmp6matchingstate(fin); 2687 if (is != NULL) 2688 goto matched; 2689 } 2690 break; 2691#endif 2692 case IPPROTO_ICMP : 2693 if (is != NULL) 2694 break; 2695 /* 2696 * No matching icmp state entry. Perhaps this is a 2697 * response to another state entry. 2698 */ 2699 is = fr_checkicmpmatchingstate(fin); 2700 if (is != NULL) 2701 goto matched; 2702 break; 2703 case IPPROTO_TCP : 2704 if (is == NULL) 2705 break; 2706 2707 if (is->is_pass & FR_NEWISN) { 2708 if (fin->fin_out == 0) 2709 fr_fixinisn(fin, is); 2710 else if (fin->fin_out == 1) 2711 fr_fixoutisn(fin, is); 2712 } 2713 break; 2714 default : 2715 if (fin->fin_rev) 2716 ifq = &ips_udpacktq; 2717 else 2718 ifq = &ips_udptq; 2719 break; 2720 } 2721 if (is == NULL) { 2722 ATOMIC_INCL(ips_stats.iss_miss); 2723 return NULL; 2724 } 2725 2726matched: 2727 fr = is->is_rule; 2728 if (fr != NULL) { 2729 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) { 2730 if (fin->fin_nattag == NULL) 2731 return NULL; 2732 if (fr_matchtag(&fr->fr_nattag, fin->fin_nattag) != 0) 2733 return NULL; 2734 } 2735 (void) strncpy(fin->fin_group, fr->fr_group, FR_GROUPLEN); 2736 fin->fin_icode = fr->fr_icode; 2737 } 2738 2739 fin->fin_rule = is->is_rulen; 2740 pass = is->is_pass; 2741 fr_updatestate(fin, is, ifq); 2742 2743 fin->fin_state = is; 2744 is->is_touched = fr_ticks; 2745 MUTEX_ENTER(&is->is_lock); 2746 is->is_ref++; 2747 MUTEX_EXIT(&is->is_lock); 2748 RWLOCK_EXIT(&ipf_state); 2749 fin->fin_flx |= FI_STATE; 2750 if ((pass & FR_LOGFIRST) != 0) 2751 pass &= ~(FR_LOGFIRST|FR_LOG); 2752 *passp = pass; 2753 return fr; 2754} 2755 2756 2757/* ------------------------------------------------------------------------ */ 2758/* Function: fr_fixoutisn */ 2759/* Returns: Nil */ 2760/* Parameters: fin(I) - pointer to packet information */ 2761/* is(I) - pointer to master state structure */ 2762/* */ 2763/* Called only for outbound packets, adjusts the sequence number and the */ 2764/* TCP checksum to match that change. */ 2765/* ------------------------------------------------------------------------ */ 2766static void fr_fixoutisn(fin, is) 2767fr_info_t *fin; 2768ipstate_t *is; 2769{ 2770 tcphdr_t *tcp; 2771 int rev; 2772 u_32_t seq; 2773 2774 tcp = fin->fin_dp; 2775 rev = fin->fin_rev; 2776 if ((is->is_flags & IS_ISNSYN) != 0) { 2777 if (rev == 0) { 2778 seq = ntohl(tcp->th_seq); 2779 seq += is->is_isninc[0]; 2780 tcp->th_seq = htonl(seq); 2781 fix_outcksum(fin, &tcp->th_sum, is->is_sumd[0]); 2782 } 2783 } 2784 if ((is->is_flags & IS_ISNACK) != 0) { 2785 if (rev == 1) { 2786 seq = ntohl(tcp->th_seq); 2787 seq += is->is_isninc[1]; 2788 tcp->th_seq = htonl(seq); 2789 fix_outcksum(fin, &tcp->th_sum, is->is_sumd[1]); 2790 } 2791 } 2792} 2793 2794 2795/* ------------------------------------------------------------------------ */ 2796/* Function: fr_fixinisn */ 2797/* Returns: Nil */ 2798/* Parameters: fin(I) - pointer to packet information */ 2799/* is(I) - pointer to master state structure */ 2800/* */ 2801/* Called only for inbound packets, adjusts the acknowledge number and the */ 2802/* TCP checksum to match that change. */ 2803/* ------------------------------------------------------------------------ */ 2804static void fr_fixinisn(fin, is) 2805fr_info_t *fin; 2806ipstate_t *is; 2807{ 2808 tcphdr_t *tcp; 2809 int rev; 2810 u_32_t ack; 2811 2812 tcp = fin->fin_dp; 2813 rev = fin->fin_rev; 2814 if ((is->is_flags & IS_ISNSYN) != 0) { 2815 if (rev == 1) { 2816 ack = ntohl(tcp->th_ack); 2817 ack -= is->is_isninc[0]; 2818 tcp->th_ack = htonl(ack); 2819 fix_incksum(fin, &tcp->th_sum, is->is_sumd[0]); 2820 } 2821 } 2822 if ((is->is_flags & IS_ISNACK) != 0) { 2823 if (rev == 0) { 2824 ack = ntohl(tcp->th_ack); 2825 ack -= is->is_isninc[1]; 2826 tcp->th_ack = htonl(ack); 2827 fix_incksum(fin, &tcp->th_sum, is->is_sumd[1]); 2828 } 2829 } 2830} 2831 2832 2833/* ------------------------------------------------------------------------ */ 2834/* Function: fr_statesync */ 2835/* Returns: Nil */ 2836/* Parameters: ifp(I) - pointer to interface */ 2837/* */ 2838/* Walk through all state entries and if an interface pointer match is */ 2839/* found then look it up again, based on its name in case the pointer has */ 2840/* changed since last time. */ 2841/* */ 2842/* If ifp is passed in as being non-null then we are only doing updates for */ 2843/* existing, matching, uses of it. */ 2844/* ------------------------------------------------------------------------ */ 2845void fr_statesync(ifp) 2846void *ifp; 2847{ 2848 ipstate_t *is; 2849 int i; 2850 2851 if (fr_running <= 0) 2852 return; 2853 2854 WRITE_ENTER(&ipf_state); 2855 2856 if (fr_running <= 0) { 2857 RWLOCK_EXIT(&ipf_state); 2858 return; 2859 } 2860 2861 for (is = ips_list; is; is = is->is_next) { 2862 /* 2863 * Look up all the interface names in the state entry. 2864 */ 2865 for (i = 0; i < 4; i++) { 2866 if (ifp == NULL || ifp == is->is_ifp[i]) 2867 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], 2868 is->is_v); 2869 } 2870 } 2871 RWLOCK_EXIT(&ipf_state); 2872} 2873 2874 2875/* ------------------------------------------------------------------------ */ 2876/* Function: fr_delstate */ 2877/* Returns: int - 0 = entry deleted, else reference count on struct */ 2878/* Parameters: is(I) - pointer to state structure to delete */ 2879/* why(I) - if not 0, log reason why it was deleted */ 2880/* Write Locks: ipf_state */ 2881/* */ 2882/* Deletes a state entry from the enumerated list as well as the hash table */ 2883/* and timeout queue lists. Make adjustments to hash table statistics and */ 2884/* global counters as required. */ 2885/* ------------------------------------------------------------------------ */ 2886static int fr_delstate(is, why) 2887ipstate_t *is; 2888int why; 2889{ 2890 2891 ASSERT(rw_read_locked(&ipf_state.ipf_lk) == 0); 2892 2893 /* 2894 * Since we want to delete this, remove it from the state table, 2895 * where it can be found & used, first. 2896 */ 2897 if (is->is_phnext != NULL) { 2898 *is->is_phnext = is->is_hnext; 2899 if (is->is_hnext != NULL) 2900 is->is_hnext->is_phnext = is->is_phnext; 2901 if (ips_table[is->is_hv] == NULL) 2902 ips_stats.iss_inuse--; 2903 ips_stats.iss_bucketlen[is->is_hv]--; 2904 2905 is->is_phnext = NULL; 2906 is->is_hnext = NULL; 2907 } 2908 2909 /* 2910 * Because ips_stats.iss_wild is a count of entries in the state 2911 * table that have wildcard flags set, only decerement it once 2912 * and do it here. 2913 */ 2914 if (is->is_flags & (SI_WILDP|SI_WILDA)) { 2915 if (!(is->is_flags & SI_CLONED)) { 2916 ATOMIC_DECL(ips_stats.iss_wild); 2917 } 2918 is->is_flags &= ~(SI_WILDP|SI_WILDA); 2919 } 2920 2921 /* 2922 * Next, remove it from the timeout queue it is in. 2923 */ 2924 if (is->is_sti.tqe_ifq != NULL) 2925 fr_deletequeueentry(&is->is_sti); 2926 2927 if (is->is_me != NULL) { 2928 *is->is_me = NULL; 2929 is->is_me = NULL; 2930 } 2931 2932 /* 2933 * If it is still in use by something else, do not go any further, 2934 * but note that at this point it is now an orphan. How can this 2935 * be? fr_state_flush() calls fr_delete() directly because it wants 2936 * to empty the table out and if something has a hold on a state 2937 * entry (such as ipfstat), it'll do the deref path that'll bring 2938 * us back here to do the real delete & free. 2939 */ 2940 is->is_ref--; 2941 if (is->is_ref > 0) 2942 return is->is_ref; 2943 2944 if (is->is_tqehead[0] != NULL) { 2945 if (fr_deletetimeoutqueue(is->is_tqehead[0]) == 0) 2946 fr_freetimeoutqueue(is->is_tqehead[0]); 2947 } 2948 if (is->is_tqehead[1] != NULL) { 2949 if (fr_deletetimeoutqueue(is->is_tqehead[1]) == 0) 2950 fr_freetimeoutqueue(is->is_tqehead[1]); 2951 } 2952 2953#ifdef IPFILTER_SYNC 2954 if (is->is_sync) 2955 ipfsync_del(is->is_sync); 2956#endif 2957#ifdef IPFILTER_SCAN 2958 (void) ipsc_detachis(is); 2959#endif 2960 2961 /* 2962 * Now remove it from the linked list of known states 2963 */ 2964 if (is->is_pnext != NULL) { 2965 *is->is_pnext = is->is_next; 2966 2967 if (is->is_next != NULL) 2968 is->is_next->is_pnext = is->is_pnext; 2969 2970 is->is_pnext = NULL; 2971 is->is_next = NULL; 2972 } 2973 2974 if (ipstate_logging != 0 && why != 0) 2975 ipstate_log(is, why); 2976 2977 if (is->is_p == IPPROTO_TCP) 2978 ips_stats.iss_fin++; 2979 else 2980 ips_stats.iss_expire++; 2981 2982 if (is->is_rule != NULL) { 2983 is->is_rule->fr_statecnt--; 2984 (void) fr_derefrule(&is->is_rule); 2985 } 2986 2987 MUTEX_DESTROY(&is->is_lock); 2988 KFREE(is); 2989 ips_num--; 2990 2991 return 0; 2992} 2993 2994 2995/* ------------------------------------------------------------------------ */ 2996/* Function: fr_timeoutstate */ 2997/* Returns: Nil */ 2998/* Parameters: Nil */ 2999/* */ 3000/* Slowly expire held state for thingslike UDP and ICMP. The algorithm */ 3001/* used here is to keep the queue sorted with the oldest things at the top */ 3002/* and the youngest at the bottom. So if the top one doesn't need to be */ 3003/* expired then neither will any under it. */ 3004/* ------------------------------------------------------------------------ */ 3005void fr_timeoutstate() 3006{ 3007 ipftq_t *ifq, *ifqnext; 3008 ipftqent_t *tqe, *tqn; 3009 ipstate_t *is; 3010 SPL_INT(s); 3011 3012 SPL_NET(s); 3013 WRITE_ENTER(&ipf_state); 3014 for (ifq = ips_tqtqb; ifq != NULL; ifq = ifq->ifq_next) 3015 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3016 if (tqe->tqe_die > fr_ticks) 3017 break; 3018 tqn = tqe->tqe_next; 3019 is = tqe->tqe_parent; 3020 fr_delstate(is, ISL_EXPIRE); 3021 } 3022 3023 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 3024 ifqnext = ifq->ifq_next; 3025 3026 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3027 if (tqe->tqe_die > fr_ticks) 3028 break; 3029 tqn = tqe->tqe_next; 3030 is = tqe->tqe_parent; 3031 fr_delstate(is, ISL_EXPIRE); 3032 } 3033 } 3034 3035 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 3036 ifqnext = ifq->ifq_next; 3037 3038 if (((ifq->ifq_flags & IFQF_DELETE) != 0) && 3039 (ifq->ifq_ref == 0)) { 3040 fr_freetimeoutqueue(ifq); 3041 } 3042 } 3043 3044 if (fr_state_doflush) { 3045 (void) fr_state_flush(2, 0); 3046 fr_state_doflush = 0; 3047 } 3048 3049 RWLOCK_EXIT(&ipf_state); 3050 SPL_X(s); 3051} 3052 3053 3054/* ------------------------------------------------------------------------ */ 3055/* Function: fr_state_flush */ 3056/* Returns: int - 0 == success, -1 == failure */ 3057/* Parameters: Nil */ 3058/* Write Locks: ipf_state */ 3059/* */ 3060/* Flush state tables. Three actions currently defined: */ 3061/* which == 0 : flush all state table entries */ 3062/* which == 1 : flush TCP connections which have started to close but are */ 3063/* stuck for some reason. */ 3064/* which == 2 : flush TCP connections which have been idle for a long time, */ 3065/* starting at > 4 days idle and working back in successive half-*/ 3066/* days to at most 12 hours old. If this fails to free enough */ 3067/* slots then work backwards in half hour slots to 30 minutes. */ 3068/* If that too fails, then work backwards in 30 second intervals */ 3069/* for the last 30 minutes to at worst 30 seconds idle. */ 3070/* ------------------------------------------------------------------------ */ 3071static int fr_state_flush(which, proto) 3072int which, proto; 3073{ 3074 ipftq_t *ifq, *ifqnext; 3075 ipftqent_t *tqe, *tqn; 3076 ipstate_t *is, **isp; 3077 int removed; 3078 SPL_INT(s); 3079 3080 removed = 0; 3081 3082 SPL_NET(s); 3083 3084 switch (which) 3085 { 3086 case 0 : 3087 /* 3088 * Style 0 flush removes everything... 3089 */ 3090 for (isp = &ips_list; ((is = *isp) != NULL); ) { 3091 if ((proto != 0) && (is->is_v != proto)) { 3092 isp = &is->is_next; 3093 continue; 3094 } 3095 if (fr_delstate(is, ISL_FLUSH) == 0) 3096 removed++; 3097 else 3098 isp = &is->is_next; 3099 } 3100 break; 3101 3102 case 1 : 3103 /* 3104 * Since we're only interested in things that are closing, 3105 * we can start with the appropriate timeout queue. 3106 */ 3107 for (ifq = ips_tqtqb + IPF_TCPS_CLOSE_WAIT; ifq != NULL; 3108 ifq = ifq->ifq_next) { 3109 3110 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3111 tqn = tqe->tqe_next; 3112 is = tqe->tqe_parent; 3113 if (is->is_p != IPPROTO_TCP) 3114 break; 3115 if (fr_delstate(is, ISL_EXPIRE) == 0) 3116 removed++; 3117 } 3118 } 3119 3120 /* 3121 * Also need to look through the user defined queues. 3122 */ 3123 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) { 3124 ifqnext = ifq->ifq_next; 3125 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3126 tqn = tqe->tqe_next; 3127 is = tqe->tqe_parent; 3128 if (is->is_p != IPPROTO_TCP) 3129 continue; 3130 3131 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) && 3132 (is->is_state[1] > IPF_TCPS_ESTABLISHED)) { 3133 if (fr_delstate(is, ISL_EXPIRE) == 0) 3134 removed++; 3135 } 3136 } 3137 } 3138 break; 3139 3140 case 2 : 3141 break; 3142 3143 /* 3144 * Args 5-11 correspond to flushing those particular states 3145 * for TCP connections. 3146 */ 3147 case IPF_TCPS_CLOSE_WAIT : 3148 case IPF_TCPS_FIN_WAIT_1 : 3149 case IPF_TCPS_CLOSING : 3150 case IPF_TCPS_LAST_ACK : 3151 case IPF_TCPS_FIN_WAIT_2 : 3152 case IPF_TCPS_TIME_WAIT : 3153 case IPF_TCPS_CLOSED : 3154 tqn = ips_tqtqb[which].ifq_head; 3155 while (tqn != NULL) { 3156 tqe = tqn; 3157 tqn = tqe->tqe_next; 3158 is = tqe->tqe_parent; 3159 if (fr_delstate(is, ISL_FLUSH) == 0) 3160 removed++; 3161 } 3162 break; 3163 3164 default : 3165 if (which < 30) 3166 break; 3167 3168 /* 3169 * Take a large arbitrary number to mean the number of seconds 3170 * for which which consider to be the maximum value we'll allow 3171 * the expiration to be. 3172 */ 3173 which = IPF_TTLVAL(which); 3174 for (isp = &ips_list; ((is = *isp) != NULL); ) { 3175 if ((proto == 0) || (is->is_v == proto)) { 3176 if (fr_ticks - is->is_touched > which) { 3177 if (fr_delstate(is, ISL_FLUSH) == 0) { 3178 removed++; 3179 continue; 3180 } 3181 } 3182 } 3183 isp = &is->is_next; 3184 } 3185 break; 3186 } 3187 3188 if (which != 2) { 3189 SPL_X(s); 3190 return removed; 3191 } 3192 3193 /* 3194 * Asked to remove inactive entries because the table is full. 3195 */ 3196 if (fr_ticks - ips_last_force_flush > IPF_TTLVAL(5)) { 3197 ips_last_force_flush = fr_ticks; 3198 removed = ipf_queueflush(fr_state_flush_entry, ips_tqtqb, 3199 ips_utqe); 3200 } 3201 3202 SPL_X(s); 3203 return removed; 3204} 3205 3206 3207/* ------------------------------------------------------------------------ */ 3208/* Function: fr_state_flush_entry */ 3209/* Returns: int - 0 = entry deleted, else not deleted */ 3210/* Parameters: entry(I) - pointer to state structure to delete */ 3211/* Write Locks: ipf_state */ 3212/* */ 3213/* This function is a stepping stone between ipf_queueflush() and */ 3214/* fr_delstate(). It is used so we can provide a uniform interface via the */ 3215/* ipf_queueflush() function. */ 3216/* ------------------------------------------------------------------------ */ 3217static int fr_state_flush_entry(entry) 3218void *entry; 3219{ 3220 return fr_delstate(entry, ISL_FLUSH); 3221} 3222 3223 3224/* ------------------------------------------------------------------------ */ 3225/* Function: fr_tcp_age */ 3226/* Returns: int - 1 == state transition made, 0 == no change (rejected) */ 3227/* Parameters: tq(I) - pointer to timeout queue information */ 3228/* fin(I) - pointer to packet information */ 3229/* tqtab(I) - TCP timeout queue table this is in */ 3230/* flags(I) - flags from state/NAT entry */ 3231/* */ 3232/* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */ 3233/* */ 3234/* - (try to) base state transitions on real evidence only, */ 3235/* i.e. packets that are sent and have been received by ipfilter; */ 3236/* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */ 3237/* */ 3238/* - deal with half-closed connections correctly; */ 3239/* */ 3240/* - store the state of the source in state[0] such that ipfstat */ 3241/* displays the state as source/dest instead of dest/source; the calls */ 3242/* to fr_tcp_age have been changed accordingly. */ 3243/* */ 3244/* Internal Parameters: */ 3245/* */ 3246/* state[0] = state of source (host that initiated connection) */ 3247/* state[1] = state of dest (host that accepted the connection) */ 3248/* */ 3249/* dir == 0 : a packet from source to dest */ 3250/* dir == 1 : a packet from dest to source */ 3251/* */ 3252/* A typical procession for a connection is as follows: */ 3253/* */ 3254/* +--------------+-------------------+ */ 3255/* | Side '0' | Side '1' | */ 3256/* +--------------+-------------------+ */ 3257/* | 0 -> 1 (SYN) | | */ 3258/* | | 0 -> 2 (SYN-ACK) | */ 3259/* | 1 -> 3 (ACK) | | */ 3260/* | | 2 -> 4 (ACK-PUSH) | */ 3261/* | 3 -> 4 (ACK) | | */ 3262/* | ... | ... | */ 3263/* | | 4 -> 6 (FIN-ACK) | */ 3264/* | 4 -> 5 (ACK) | | */ 3265/* | | 6 -> 6 (ACK-PUSH) | */ 3266/* | 5 -> 5 (ACK) | | */ 3267/* | 5 -> 8 (FIN) | | */ 3268/* | | 6 -> 10 (ACK) | */ 3269/* +--------------+-------------------+ */ 3270/* */ 3271/* Locking: it is assumed that the parent of the tqe structure is locked. */ 3272/* ------------------------------------------------------------------------ */ 3273int fr_tcp_age(tqe, fin, tqtab, flags) 3274ipftqent_t *tqe; 3275fr_info_t *fin; 3276ipftq_t *tqtab; 3277int flags; 3278{ 3279 int dlen, ostate, nstate, rval, dir; 3280 u_char tcpflags; 3281 tcphdr_t *tcp; 3282 3283 tcp = fin->fin_dp; 3284 3285 rval = 0; 3286 dir = fin->fin_rev; 3287 tcpflags = tcp->th_flags; 3288 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2); 3289 3290 if (tcpflags & TH_RST) { 3291 if (!(tcpflags & TH_PUSH) && !dlen) 3292 nstate = IPF_TCPS_CLOSED; 3293 else 3294 nstate = IPF_TCPS_CLOSE_WAIT; 3295 rval = 1; 3296 } else { 3297 ostate = tqe->tqe_state[1 - dir]; 3298 nstate = tqe->tqe_state[dir]; 3299 3300 switch (nstate) 3301 { 3302 case IPF_TCPS_LISTEN: /* 0 */ 3303 if ((tcpflags & TH_OPENING) == TH_OPENING) { 3304 /* 3305 * 'dir' received an S and sends SA in 3306 * response, LISTEN -> SYN_RECEIVED 3307 */ 3308 nstate = IPF_TCPS_SYN_RECEIVED; 3309 rval = 1; 3310 } else if ((tcpflags & TH_OPENING) == TH_SYN) { 3311 /* 'dir' sent S, LISTEN -> SYN_SENT */ 3312 nstate = IPF_TCPS_SYN_SENT; 3313 rval = 1; 3314 } 3315 /* 3316 * the next piece of code makes it possible to get 3317 * already established connections into the state table 3318 * after a restart or reload of the filter rules; this 3319 * does not work when a strict 'flags S keep state' is 3320 * used for tcp connections of course 3321 */ 3322 if (((flags & IS_TCPFSM) == 0) && 3323 ((tcpflags & TH_ACKMASK) == TH_ACK)) { 3324 /* 3325 * we saw an A, guess 'dir' is in ESTABLISHED 3326 * mode 3327 */ 3328 switch (ostate) 3329 { 3330 case IPF_TCPS_LISTEN : 3331 case IPF_TCPS_SYN_RECEIVED : 3332 nstate = IPF_TCPS_HALF_ESTAB; 3333 rval = 1; 3334 break; 3335 case IPF_TCPS_HALF_ESTAB : 3336 case IPF_TCPS_ESTABLISHED : 3337 nstate = IPF_TCPS_ESTABLISHED; 3338 rval = 1; 3339 break; 3340 default : 3341 break; 3342 } 3343 } 3344 /* 3345 * TODO: besides regular ACK packets we can have other 3346 * packets as well; it is yet to be determined how we 3347 * should initialize the states in those cases 3348 */ 3349 break; 3350 3351 case IPF_TCPS_SYN_SENT: /* 1 */ 3352 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) { 3353 /* 3354 * A retransmitted SYN packet. We do not reset 3355 * the timeout here to fr_tcptimeout because a 3356 * connection connect timeout does not renew 3357 * after every packet that is sent. We need to 3358 * set rval so as to indicate the packet has 3359 * passed the check for its flags being valid 3360 * in the TCP FSM. Setting rval to 2 has the 3361 * result of not resetting the timeout. 3362 */ 3363 rval = 2; 3364 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == 3365 TH_ACK) { 3366 /* 3367 * we see an A from 'dir' which is in SYN_SENT 3368 * state: 'dir' sent an A in response to an SA 3369 * which it received, SYN_SENT -> ESTABLISHED 3370 */ 3371 nstate = IPF_TCPS_ESTABLISHED; 3372 rval = 1; 3373 } else if (tcpflags & TH_FIN) { 3374 /* 3375 * we see an F from 'dir' which is in SYN_SENT 3376 * state and wants to close its side of the 3377 * connection; SYN_SENT -> FIN_WAIT_1 3378 */ 3379 nstate = IPF_TCPS_FIN_WAIT_1; 3380 rval = 1; 3381 } else if ((tcpflags & TH_OPENING) == TH_OPENING) { 3382 /* 3383 * we see an SA from 'dir' which is already in 3384 * SYN_SENT state, this means we have a 3385 * simultaneous open; SYN_SENT -> SYN_RECEIVED 3386 */ 3387 nstate = IPF_TCPS_SYN_RECEIVED; 3388 rval = 1; 3389 } 3390 break; 3391 3392 case IPF_TCPS_SYN_RECEIVED: /* 2 */ 3393 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 3394 /* 3395 * we see an A from 'dir' which was in 3396 * SYN_RECEIVED state so it must now be in 3397 * established state, SYN_RECEIVED -> 3398 * ESTABLISHED 3399 */ 3400 nstate = IPF_TCPS_ESTABLISHED; 3401 rval = 1; 3402 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) == 3403 TH_OPENING) { 3404 /* 3405 * We see an SA from 'dir' which is already in 3406 * SYN_RECEIVED state. 3407 */ 3408 rval = 2; 3409 } else if (tcpflags & TH_FIN) { 3410 /* 3411 * we see an F from 'dir' which is in 3412 * SYN_RECEIVED state and wants to close its 3413 * side of the connection; SYN_RECEIVED -> 3414 * FIN_WAIT_1 3415 */ 3416 nstate = IPF_TCPS_FIN_WAIT_1; 3417 rval = 1; 3418 } 3419 break; 3420 3421 case IPF_TCPS_HALF_ESTAB: /* 3 */ 3422 if (tcpflags & TH_FIN) { 3423 nstate = IPF_TCPS_FIN_WAIT_1; 3424 rval = 1; 3425 } else if ((tcpflags & TH_ACKMASK) == TH_ACK) { 3426 /* 3427 * If we've picked up a connection in mid 3428 * flight, we could be looking at a follow on 3429 * packet from the same direction as the one 3430 * that created this state. Recognise it but 3431 * do not advance the entire connection's 3432 * state. 3433 */ 3434 switch (ostate) 3435 { 3436 case IPF_TCPS_LISTEN : 3437 case IPF_TCPS_SYN_SENT : 3438 case IPF_TCPS_SYN_RECEIVED : 3439 rval = 1; 3440 break; 3441 case IPF_TCPS_HALF_ESTAB : 3442 case IPF_TCPS_ESTABLISHED : 3443 nstate = IPF_TCPS_ESTABLISHED; 3444 rval = 1; 3445 break; 3446 default : 3447 break; 3448 } 3449 } 3450 break; 3451 3452 case IPF_TCPS_ESTABLISHED: /* 4 */ 3453 rval = 1; 3454 if (tcpflags & TH_FIN) { 3455 /* 3456 * 'dir' closed its side of the connection; 3457 * this gives us a half-closed connection; 3458 * ESTABLISHED -> FIN_WAIT_1 3459 */ 3460 if (ostate == IPF_TCPS_FIN_WAIT_1) { 3461 nstate = IPF_TCPS_CLOSING; 3462 } else { 3463 nstate = IPF_TCPS_FIN_WAIT_1; 3464 } 3465 } else if (tcpflags & TH_ACK) { 3466 /* 3467 * an ACK, should we exclude other flags here? 3468 */ 3469 if (ostate == IPF_TCPS_FIN_WAIT_1) { 3470 /* 3471 * We know the other side did an active 3472 * close, so we are ACKing the recvd 3473 * FIN packet (does the window matching 3474 * code guarantee this?) and go into 3475 * CLOSE_WAIT state; this gives us a 3476 * half-closed connection 3477 */ 3478 nstate = IPF_TCPS_CLOSE_WAIT; 3479 } else if (ostate < IPF_TCPS_CLOSE_WAIT) { 3480 /* 3481 * still a fully established 3482 * connection reset timeout 3483 */ 3484 nstate = IPF_TCPS_ESTABLISHED; 3485 } 3486 } 3487 break; 3488 3489 case IPF_TCPS_CLOSE_WAIT: /* 5 */ 3490 rval = 1; 3491 if (tcpflags & TH_FIN) { 3492 /* 3493 * application closed and 'dir' sent a FIN, 3494 * we're now going into LAST_ACK state 3495 */ 3496 nstate = IPF_TCPS_LAST_ACK; 3497 } else { 3498 /* 3499 * we remain in CLOSE_WAIT because the other 3500 * side has closed already and we did not 3501 * close our side yet; reset timeout 3502 */ 3503 nstate = IPF_TCPS_CLOSE_WAIT; 3504 } 3505 break; 3506 3507 case IPF_TCPS_FIN_WAIT_1: /* 6 */ 3508 rval = 1; 3509 if ((tcpflags & TH_ACK) && 3510 ostate > IPF_TCPS_CLOSE_WAIT) { 3511 /* 3512 * if the other side is not active anymore 3513 * it has sent us a FIN packet that we are 3514 * ack'ing now with an ACK; this means both 3515 * sides have now closed the connection and 3516 * we go into TIME_WAIT 3517 */ 3518 /* 3519 * XXX: how do we know we really are ACKing 3520 * the FIN packet here? does the window code 3521 * guarantee that? 3522 */ 3523 nstate = IPF_TCPS_TIME_WAIT; 3524 } else { 3525 /* 3526 * we closed our side of the connection 3527 * already but the other side is still active 3528 * (ESTABLISHED/CLOSE_WAIT); continue with 3529 * this half-closed connection 3530 */ 3531 nstate = IPF_TCPS_FIN_WAIT_1; 3532 } 3533 break; 3534 3535 case IPF_TCPS_CLOSING: /* 7 */ 3536 if ((tcpflags & (TH_FIN|TH_ACK)) == TH_ACK) { 3537 nstate = IPF_TCPS_TIME_WAIT; 3538 } 3539 rval = 1; 3540 break; 3541 3542 case IPF_TCPS_LAST_ACK: /* 8 */ 3543 if (tcpflags & TH_ACK) { 3544 if ((tcpflags & TH_PUSH) || dlen) 3545 /* 3546 * there is still data to be delivered, 3547 * reset timeout 3548 */ 3549 rval = 1; 3550 else 3551 rval = 2; 3552 } 3553 /* 3554 * we cannot detect when we go out of LAST_ACK state to 3555 * CLOSED because that is based on the reception of ACK 3556 * packets; ipfilter can only detect that a packet 3557 * has been sent by a host 3558 */ 3559 break; 3560 3561 case IPF_TCPS_FIN_WAIT_2: /* 9 */ 3562 /* NOT USED */ 3563#if 0 3564 rval = 1; 3565 if ((tcpflags & TH_OPENING) == TH_OPENING) { 3566 nstate = IPF_TCPS_SYN_RECEIVED; 3567 } else if (tcpflags & TH_SYN) { 3568 nstate = IPF_TCPS_SYN_SENT; 3569 } else if ((tcpflags & (TH_FIN|TH_ACK)) != 0) { 3570 nstate = IPF_TCPS_TIME_WAIT; 3571 } 3572#endif 3573 break; 3574 3575 case IPF_TCPS_TIME_WAIT: /* 10 */ 3576 /* we're in 2MSL timeout now */ 3577 rval = 2; 3578 if (ostate == IPF_TCPS_LAST_ACK) { 3579 nstate = IPF_TCPS_CLOSED; 3580 } 3581 break; 3582 3583 case IPF_TCPS_CLOSED: /* 11 */ 3584 rval = 2; 3585 break; 3586 3587 default : 3588#if defined(_KERNEL) 3589# if SOLARIS 3590 cmn_err(CE_NOTE, 3591 "tcp %lx flags %x si %lx nstate %d ostate %d\n", 3592 (u_long)tcp, tcpflags, (u_long)tqe, 3593 nstate, ostate); 3594# else 3595 printf("tcp %lx flags %x si %lx nstate %d ostate %d\n", 3596 (u_long)tcp, tcpflags, (u_long)tqe, 3597 nstate, ostate); 3598# endif 3599#else 3600 abort(); 3601#endif 3602 break; 3603 } 3604 } 3605 3606 /* 3607 * If rval == 2 then do not update the queue position, but treat the 3608 * packet as being ok. 3609 */ 3610 if (rval == 2) 3611 rval = 1; 3612 else if (rval == 1) { 3613 tqe->tqe_state[dir] = nstate; 3614 if ((tqe->tqe_flags & TQE_RULEBASED) == 0) 3615 fr_movequeue(tqe, tqe->tqe_ifq, tqtab + nstate); 3616 } 3617 3618 return rval; 3619} 3620 3621 3622/* ------------------------------------------------------------------------ */ 3623/* Function: ipstate_log */ 3624/* Returns: Nil */ 3625/* Parameters: is(I) - pointer to state structure */ 3626/* type(I) - type of log entry to create */ 3627/* */ 3628/* Creates a state table log entry using the state structure and type info. */ 3629/* passed in. Log packet/byte counts, source/destination address and other */ 3630/* protocol specific information. */ 3631/* ------------------------------------------------------------------------ */ 3632void ipstate_log(is, type) 3633struct ipstate *is; 3634u_int type; 3635{ 3636#ifdef IPFILTER_LOG 3637 struct ipslog ipsl; 3638 size_t sizes[1]; 3639 void *items[1]; 3640 int types[1]; 3641 3642 /* 3643 * Copy information out of the ipstate_t structure and into the 3644 * structure used for logging. 3645 */ 3646 ipsl.isl_type = type; 3647 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0]; 3648 ipsl.isl_bytes[0] = is->is_bytes[0]; 3649 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1]; 3650 ipsl.isl_bytes[1] = is->is_bytes[1]; 3651 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2]; 3652 ipsl.isl_bytes[2] = is->is_bytes[2]; 3653 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3]; 3654 ipsl.isl_bytes[3] = is->is_bytes[3]; 3655 ipsl.isl_src = is->is_src; 3656 ipsl.isl_dst = is->is_dst; 3657 ipsl.isl_p = is->is_p; 3658 ipsl.isl_v = is->is_v; 3659 ipsl.isl_flags = is->is_flags; 3660 ipsl.isl_tag = is->is_tag; 3661 ipsl.isl_rulen = is->is_rulen; 3662 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN); 3663 3664 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 3665 ipsl.isl_sport = is->is_sport; 3666 ipsl.isl_dport = is->is_dport; 3667 if (ipsl.isl_p == IPPROTO_TCP) { 3668 ipsl.isl_state[0] = is->is_state[0]; 3669 ipsl.isl_state[1] = is->is_state[1]; 3670 } 3671 } else if (ipsl.isl_p == IPPROTO_ICMP) { 3672 ipsl.isl_itype = is->is_icmp.ici_type; 3673 } else if (ipsl.isl_p == IPPROTO_ICMPV6) { 3674 ipsl.isl_itype = is->is_icmp.ici_type; 3675 } else { 3676 ipsl.isl_ps.isl_filler[0] = 0; 3677 ipsl.isl_ps.isl_filler[1] = 0; 3678 } 3679 3680 items[0] = &ipsl; 3681 sizes[0] = sizeof(ipsl); 3682 types[0] = 0; 3683 3684 if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1)) { 3685 ATOMIC_INCL(ips_stats.iss_logged); 3686 } else { 3687 ATOMIC_INCL(ips_stats.iss_logfail); 3688 } 3689#endif 3690} 3691 3692 3693#ifdef USE_INET6 3694/* ------------------------------------------------------------------------ */ 3695/* Function: fr_checkicmp6matchingstate */ 3696/* Returns: ipstate_t* - NULL == no match found, */ 3697/* else pointer to matching state entry */ 3698/* Parameters: fin(I) - pointer to packet information */ 3699/* Locks: NULL == no locks, else Read Lock on ipf_state */ 3700/* */ 3701/* If we've got an ICMPv6 error message, using the information stored in */ 3702/* the ICMPv6 packet, look for a matching state table entry. */ 3703/* ------------------------------------------------------------------------ */ 3704static ipstate_t *fr_checkicmp6matchingstate(fin) 3705fr_info_t *fin; 3706{ 3707 struct icmp6_hdr *ic6, *oic; 3708 int type, backward, i; 3709 ipstate_t *is, **isp; 3710 u_short sport, dport; 3711 i6addr_t dst, src; 3712 u_short savelen; 3713 icmpinfo_t *ic; 3714 fr_info_t ofin; 3715 tcphdr_t *tcp; 3716 ip6_t *oip6; 3717 u_char pr; 3718 u_int hv; 3719 3720 /* 3721 * Does it at least have the return (basic) IP header ? 3722 * Is it an actual recognised ICMP error type? 3723 * Only a basic IP header (no options) should be with 3724 * an ICMP error header. 3725 */ 3726 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) || 3727 !(fin->fin_flx & FI_ICMPERR)) 3728 return NULL; 3729 3730 ic6 = fin->fin_dp; 3731 type = ic6->icmp6_type; 3732 3733 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN); 3734 if (fin->fin_plen < sizeof(*oip6)) 3735 return NULL; 3736 3737 bcopy((char *)fin, (char *)&ofin, sizeof(*fin)); 3738 ofin.fin_v = 6; 3739 ofin.fin_ifp = fin->fin_ifp; 3740 ofin.fin_out = !fin->fin_out; 3741 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 3742 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 3743 3744 /* 3745 * We make a fin entry to be able to feed it to 3746 * matchsrcdst. Note that not all fields are necessary 3747 * but this is the cleanest way. Note further we fill 3748 * in fin_mp such that if someone uses it we'll get 3749 * a kernel panic. fr_matchsrcdst does not use this. 3750 * 3751 * watch out here, as ip is in host order and oip6 in network 3752 * order. Any change we make must be undone afterwards. 3753 */ 3754 savelen = oip6->ip6_plen; 3755 oip6->ip6_plen = fin->fin_dlen - ICMPERR_ICMPHLEN; 3756 ofin.fin_flx = FI_NOCKSUM; 3757 ofin.fin_ip = (ip_t *)oip6; 3758 (void) fr_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin); 3759 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 3760 oip6->ip6_plen = savelen; 3761 3762 if (oip6->ip6_nxt == IPPROTO_ICMPV6) { 3763 oic = (struct icmp6_hdr *)(oip6 + 1); 3764 /* 3765 * an ICMP error can only be generated as a result of an 3766 * ICMP query, not as the response on an ICMP error 3767 * 3768 * XXX theoretically ICMP_ECHOREP and the other reply's are 3769 * ICMP query's as well, but adding them here seems strange XXX 3770 */ 3771 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 3772 return NULL; 3773 3774 /* 3775 * perform a lookup of the ICMP packet in the state table 3776 */ 3777 hv = (pr = oip6->ip6_nxt); 3778 src.in6 = oip6->ip6_src; 3779 hv += src.in4.s_addr; 3780 dst.in6 = oip6->ip6_dst; 3781 hv += dst.in4.s_addr; 3782 hv += oic->icmp6_id; 3783 hv += oic->icmp6_seq; 3784 hv = DOUBLE_HASH(hv); 3785 3786 READ_ENTER(&ipf_state); 3787 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 3788 ic = &is->is_icmp; 3789 isp = &is->is_hnext; 3790 if ((is->is_p == pr) && 3791 !(is->is_pass & FR_NOICMPERR) && 3792 (oic->icmp6_id == ic->ici_id) && 3793 (oic->icmp6_seq == ic->ici_seq) && 3794 (is = fr_matchsrcdst(&ofin, is, &src, 3795 &dst, NULL, FI_ICMPCMP))) { 3796 /* 3797 * in the state table ICMP query's are stored 3798 * with the type of the corresponding ICMP 3799 * response. Correct here 3800 */ 3801 if (((ic->ici_type == ICMP6_ECHO_REPLY) && 3802 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 3803 (ic->ici_type - 1 == oic->icmp6_type )) { 3804 ips_stats.iss_hits++; 3805 backward = IP6_NEQ(&is->is_dst, &src); 3806 fin->fin_rev = !backward; 3807 i = (backward << 1) + fin->fin_out; 3808 is->is_icmppkts[i]++; 3809 return is; 3810 } 3811 } 3812 } 3813 RWLOCK_EXIT(&ipf_state); 3814 return NULL; 3815 } 3816 3817 hv = (pr = oip6->ip6_nxt); 3818 src.in6 = oip6->ip6_src; 3819 hv += src.i6[0]; 3820 hv += src.i6[1]; 3821 hv += src.i6[2]; 3822 hv += src.i6[3]; 3823 dst.in6 = oip6->ip6_dst; 3824 hv += dst.i6[0]; 3825 hv += dst.i6[1]; 3826 hv += dst.i6[2]; 3827 hv += dst.i6[3]; 3828 3829 if ((oip6->ip6_nxt == IPPROTO_TCP) || (oip6->ip6_nxt == IPPROTO_UDP)) { 3830 tcp = (tcphdr_t *)(oip6 + 1); 3831 dport = tcp->th_dport; 3832 sport = tcp->th_sport; 3833 hv += dport; 3834 hv += sport; 3835 } else 3836 tcp = NULL; 3837 hv = DOUBLE_HASH(hv); 3838 3839 READ_ENTER(&ipf_state); 3840 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) { 3841 isp = &is->is_hnext; 3842 /* 3843 * Only allow this icmp though if the 3844 * encapsulated packet was allowed through the 3845 * other way around. Note that the minimal amount 3846 * of info present does not allow for checking against 3847 * tcp internals such as seq and ack numbers. 3848 */ 3849 if ((is->is_p != pr) || (is->is_v != 6) || 3850 (is->is_pass & FR_NOICMPERR)) 3851 continue; 3852 is = fr_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP); 3853 if (is != NULL) { 3854 ips_stats.iss_hits++; 3855 backward = IP6_NEQ(&is->is_dst, &src); 3856 fin->fin_rev = !backward; 3857 i = (backward << 1) + fin->fin_out; 3858 is->is_icmppkts[i]++; 3859 /* 3860 * we deliberately do not touch the timeouts 3861 * for the accompanying state table entry. 3862 * It remains to be seen if that is correct. XXX 3863 */ 3864 return is; 3865 } 3866 } 3867 RWLOCK_EXIT(&ipf_state); 3868 return NULL; 3869} 3870#endif 3871 3872 3873/* ------------------------------------------------------------------------ */ 3874/* Function: fr_sttab_init */ 3875/* Returns: Nil */ 3876/* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 3877/* */ 3878/* Initialise the array of timeout queues for TCP. */ 3879/* ------------------------------------------------------------------------ */ 3880void fr_sttab_init(tqp) 3881ipftq_t *tqp; 3882{ 3883 int i; 3884 3885 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) { 3886 tqp[i].ifq_ttl = 0; 3887 tqp[i].ifq_ref = 1; 3888 tqp[i].ifq_head = NULL; 3889 tqp[i].ifq_tail = &tqp[i].ifq_head; 3890 tqp[i].ifq_next = tqp + i + 1; 3891 MUTEX_INIT(&tqp[i].ifq_lock, "ipftq tcp tab"); 3892 } 3893 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL; 3894 tqp[IPF_TCPS_CLOSED].ifq_ttl = fr_tcpclosed; 3895 tqp[IPF_TCPS_LISTEN].ifq_ttl = fr_tcptimeout; 3896 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = fr_tcptimeout; 3897 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = fr_tcptimeout; 3898 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = fr_tcpidletimeout; 3899 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = fr_tcphalfclosed; 3900 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = fr_tcphalfclosed; 3901 tqp[IPF_TCPS_CLOSING].ifq_ttl = fr_tcptimeout; 3902 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = fr_tcplastack; 3903 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = fr_tcpclosewait; 3904 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = fr_tcptimewait; 3905 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = fr_tcptimeout; 3906} 3907 3908 3909/* ------------------------------------------------------------------------ */ 3910/* Function: fr_sttab_destroy */ 3911/* Returns: Nil */ 3912/* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 3913/* */ 3914/* Do whatever is necessary to "destroy" each of the entries in the array */ 3915/* of timeout queues for TCP. */ 3916/* ------------------------------------------------------------------------ */ 3917void fr_sttab_destroy(tqp) 3918ipftq_t *tqp; 3919{ 3920 int i; 3921 3922 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) 3923 MUTEX_DESTROY(&tqp[i].ifq_lock); 3924} 3925 3926 3927/* ------------------------------------------------------------------------ */ 3928/* Function: fr_statederef */ 3929/* Returns: Nil */ 3930/* Parameters: isp(I) - pointer to pointer to state table entry */ 3931/* */ 3932/* Decrement the reference counter for this state table entry and free it */ 3933/* if there are no more things using it. */ 3934/* */ 3935/* When operating in userland (ipftest), we have no timers to clear a state */ 3936/* entry. Therefore, we make a few simple tests before deleting an entry */ 3937/* outright. We compare states on each side looking for a combination of */ 3938/* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK. Then we factor */ 3939/* in packet direction with the interface list to make sure we don't */ 3940/* prematurely delete an entry on a final inbound packet that's we're also */ 3941/* supposed to route elsewhere. */ 3942/* */ 3943/* Internal parameters: */ 3944/* state[0] = state of source (host that initiated connection) */ 3945/* state[1] = state of dest (host that accepted the connection) */ 3946/* */ 3947/* dir == 0 : a packet from source to dest */ 3948/* dir == 1 : a packet from dest to source */ 3949/* ------------------------------------------------------------------------ */ 3950void fr_statederef(isp) 3951ipstate_t **isp; 3952{ 3953 ipstate_t *is; 3954 3955 is = *isp; 3956 *isp = NULL; 3957 WRITE_ENTER(&ipf_state); 3958 is->is_ref--; 3959 if (is->is_ref == 0) { 3960 is->is_ref++; /* To counter ref-- in fr_delstate() */ 3961 fr_delstate(is, ISL_EXPIRE); 3962#ifndef _KERNEL 3963 } else if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) || 3964 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) { 3965 fr_delstate(is, ISL_ORPHAN); 3966#endif 3967 } 3968 RWLOCK_EXIT(&ipf_state); 3969} 3970 3971 3972/* ------------------------------------------------------------------------ */ 3973/* Function: fr_setstatequeue */ 3974/* Returns: Nil */ 3975/* Parameters: is(I) - pointer to state structure */ 3976/* rev(I) - forward(0) or reverse(1) direction */ 3977/* Locks: ipf_state (read or write) */ 3978/* */ 3979/* Put the state entry on its default queue entry, using rev as a helped in */ 3980/* determining which queue it should be placed on. */ 3981/* ------------------------------------------------------------------------ */ 3982void fr_setstatequeue(is, rev) 3983ipstate_t *is; 3984int rev; 3985{ 3986 ipftq_t *oifq, *nifq; 3987 3988 3989 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) 3990 nifq = is->is_tqehead[rev]; 3991 else 3992 nifq = NULL; 3993 3994 if (nifq == NULL) { 3995 switch (is->is_p) 3996 { 3997#ifdef USE_INET6 3998 case IPPROTO_ICMPV6 : 3999 if (rev == 1) 4000 nifq = &ips_icmpacktq; 4001 else 4002 nifq = &ips_icmptq; 4003 break; 4004#endif 4005 case IPPROTO_ICMP : 4006 if (rev == 1) 4007 nifq = &ips_icmpacktq; 4008 else 4009 nifq = &ips_icmptq; 4010 break; 4011 case IPPROTO_TCP : 4012 nifq = ips_tqtqb + is->is_state[rev]; 4013 break; 4014 4015 case IPPROTO_UDP : 4016 if (rev == 1) 4017 nifq = &ips_udpacktq; 4018 else 4019 nifq = &ips_udptq; 4020 break; 4021 4022 default : 4023 nifq = &ips_iptq; 4024 break; 4025 } 4026 } 4027 4028 oifq = is->is_sti.tqe_ifq; 4029 /* 4030 * If it's currently on a timeout queue, move it from one queue to 4031 * another, else put it on the end of the newly determined queue. 4032 */ 4033 if (oifq != NULL) 4034 fr_movequeue(&is->is_sti, oifq, nifq); 4035 else 4036 fr_queueappend(&is->is_sti, nifq, is); 4037 return; 4038} 4039 4040 4041/* ------------------------------------------------------------------------ */ 4042/* Function: fr_stateiter */ 4043/* Returns: int - 0 == success, else error */ 4044/* Parameters: token(I) - pointer to ipftoken structure */ 4045/* itp(I) - pointer to ipfgeniter structure */ 4046/* */ 4047/* This function handles the SIOCGENITER ioctl for the state tables and */ 4048/* walks through the list of entries in the state table list (ips_list.) */ 4049/* ------------------------------------------------------------------------ */ 4050static int fr_stateiter(token, itp) 4051ipftoken_t *token; 4052ipfgeniter_t *itp; 4053{ 4054 ipstate_t *is, *next, zero; 4055 int error, count; 4056 char *dst; 4057 4058 if (itp->igi_data == NULL) 4059 return EFAULT; 4060 4061 if (itp->igi_nitems == 0) 4062 return ENOSPC; 4063 4064 if (itp->igi_type != IPFGENITER_STATE) 4065 return EINVAL; 4066 4067 is = token->ipt_data; 4068 if (is == (void *)-1) { 4069 ipf_freetoken(token); 4070 return ESRCH; 4071 } 4072 4073 error = 0; 4074 dst = itp->igi_data; 4075 4076 READ_ENTER(&ipf_state); 4077 if (is == NULL) { 4078 next = ips_list; 4079 } else { 4080 next = is->is_next; 4081 } 4082 4083 for (count = itp->igi_nitems; count > 0; count--) { 4084 if (next != NULL) { 4085 /* 4086 * If we find a state entry to use, bump its 4087 * reference count so that it can be used for 4088 * is_next when we come back. 4089 */ 4090 MUTEX_ENTER(&next->is_lock); 4091 next->is_ref++; 4092 MUTEX_EXIT(&next->is_lock); 4093 token->ipt_data = next; 4094 } else { 4095 bzero(&zero, sizeof(zero)); 4096 next = &zero; 4097 token->ipt_data = (void *)-1; 4098 count = 1; 4099 } 4100 RWLOCK_EXIT(&ipf_state); 4101 4102 /* 4103 * If we had a prior pointer to a state entry, release it. 4104 */ 4105 if (is != NULL) { 4106 fr_statederef(&is); 4107 } 4108 4109 /* 4110 * This should arguably be via fr_outobj() so that the state 4111 * structure can (if required) be massaged going out. 4112 */ 4113 error = COPYOUT(next, dst, sizeof(*next)); 4114 if (error != 0) 4115 error = EFAULT; 4116 if ((count == 1) || (error != 0)) 4117 break; 4118 4119 dst += sizeof(*next); 4120 READ_ENTER(&ipf_state); 4121 is = next; 4122 next = is->is_next; 4123 } 4124 4125 return error; 4126} 4127 4128 4129/* ------------------------------------------------------------------------ */ 4130/* Function: fr_stgettable */ 4131/* Returns: int - 0 = success, else error */ 4132/* Parameters: data(I) - pointer to ioctl data */ 4133/* */ 4134/* This function handles ioctl requests for tables of state information. */ 4135/* At present the only table it deals with is the hash bucket statistics. */ 4136/* ------------------------------------------------------------------------ */ 4137static int fr_stgettable(data) 4138char *data; 4139{ 4140 ipftable_t table; 4141 int error; 4142 4143 error = fr_inobj(data, &table, IPFOBJ_GTABLE); 4144 if (error != 0) 4145 return error; 4146 4147 if (table.ita_type != IPFTABLE_BUCKETS) 4148 return EINVAL; 4149 4150 error = COPYOUT(ips_stats.iss_bucketlen, table.ita_table, 4151 fr_statesize * sizeof(u_long)); 4152 if (error != 0) 4153 error = EFAULT; 4154 return error; 4155} 4156