sys/contrib/ipfilter/netinet/ip_state.c

   1 /*
   2  * Copyright (C) 1995-2002 by Darren Reed.
   3  *
   4  * See the IPFILTER.LICENCE file for details on licencing.
   5  *
   6  * @(#)ip_state.c   1.8 6/5/96 (C) 1993-2000 Darren Reed
   7  * @(#)$Id: ip_state.c,v 2.30.2.74 2002/07/27 15:58:10 darrenr Exp $
   8  * $FreeBSD: src/sys/contrib/ipfilter/netinet/ip_state.c,v 1.21.2.5 2003/03/01 03:55:54 darrenr Exp $
   9  * $DragonFly: src/sys/contrib/ipfilter/netinet/ip_state.c,v 1.6 2004/02/12 22:35:47 joerg Exp $
  10  */
  11
  12 #if defined(__sgi) && (IRIX > 602)
  13 # include <sys/ptimers.h>
  14 #endif
  15 #include <sys/errno.h>
  16 #include <sys/types.h>
  17 #include <sys/param.h>
  18 #include <sys/file.h>
  19 #if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \
  20     defined(_KERNEL)
  21 # include "opt_ipfilter_log.h"
  22 #endif
  23 #if defined(_KERNEL) && (defined(__DragonFly__) || \
  24     (defined(__FreeBSD_version) && \
  25     __FreeBSD_version >= 400000)) && !defined(KLD_MODULE)
  26 #include "opt_inet6.h"
  27 #endif
  28 #if !defined(_KERNEL) && !defined(KERNEL) && !defined(__KERNEL__)
  29 # include <stdio.h>
  30 # include <stdlib.h>
  31 # include <string.h>
  32 #else
  33 # ifdef linux
  34 #  include <linux/kernel.h>
  35 #  include <linux/module.h>
  36 # endif
  37 #endif
  38 #if (defined(KERNEL) || defined(_KERNEL)) && (defined(__DragonFly__) || __FreeBSD_version >= 220000)
  39 # include <sys/filio.h>
  40 # include <sys/fcntl.h>
  41 # if (defined(__DragonFly__) || __FreeBSD_version >= 300000) && !defined(IPFILTER_LKM)
  42 #  include "opt_ipfilter.h"
  43 # endif
  44 #else
  45 # include <sys/ioctl.h>
  46 #endif
  47 #include <sys/time.h>
  48 #ifndef linux
  49 # include <sys/protosw.h>
  50 #endif
  51 #include <sys/socket.h>
  52 #if (defined(_KERNEL) || defined(KERNEL)) && !defined(linux)
  53 # include <sys/systm.h>
  54 #endif
  55 #if !defined(__SVR4) && !defined(__svr4__)
  56 # ifndef linux
  57 #  include <sys/mbuf.h>
  58 # endif
  59 #else
  60 # include <sys/filio.h>
  61 # include <sys/byteorder.h>
  62 # ifdef _KERNEL
  63 #  include <sys/dditypes.h>
  64 # endif
  65 # include <sys/stream.h>
  66 # include <sys/kmem.h>
  67 #endif
  68
  69 #include <net/if.h>
  70 #ifdef sun
  71 # include <net/af.h>
  72 #endif
  73 #include <net/route.h>
  74 #include <netinet/in.h>
  75 #include <netinet/in_systm.h>
  76 #include <netinet/ip.h>
  77 #include <netinet/tcp.h>
  78 #ifndef linux
  79 # include <netinet/ip_var.h>
  80 # include <netinet/tcp_fsm.h>
  81 #endif
  82 #include <netinet/udp.h>
  83 #include <netinet/ip_icmp.h>
  84 #include "ip_compat.h"
  85 #include <netinet/tcpip.h>
  86 #include "ip_fil.h"
  87 #include "ip_nat.h"
  88 #include "ip_frag.h"
  89 #include "ip_state.h"
  90 #ifdef  USE_INET6
  91 #include <netinet/icmp6.h>
  92 #endif
  93 #if defined(__DragonFly__) || (__FreeBSD_version >= 300000)
  94 # include <sys/malloc.h>
  95 # if (defined(_KERNEL) || defined(KERNEL)) && !defined(IPFILTER_LKM)
  96 #  include <sys/libkern.h>
  97 #  include <sys/systm.h>
  98 # endif
  99 #endif
 100
 101 static const char sccsid[] = "@(#)ip_state.c   1.8 6/5/96 (C) 1993-2000 Darren Reed";
 102
 103 #ifndef MIN
 104 # define        MIN(a,b)        (((a)<(b))?(a):(b))
 105 #endif
 106
 107 #define TCP_CLOSE       (TH_FIN|TH_RST)
 108
 109 static ipstate_t **ips_table = NULL;
 110 static int      ips_num = 0;
 111 static int      ips_wild = 0;
 112 static ips_stat_t ips_stats;
 113 #if     (SOLARIS || defined(__sgi)) && defined(_KERNEL)
 114 extern  KRWLOCK_T       ipf_state, ipf_mutex;
 115 extern  kmutex_t        ipf_rw;
 116 #endif
 117
 118 #ifdef  USE_INET6
 119 static frentry_t *fr_checkicmp6matchingstate (ip6_t *, fr_info_t *);
 120 #endif
 121 static int fr_matchsrcdst (ipstate_t *, union i6addr, union i6addr,
 122                                fr_info_t *, tcphdr_t *);
 123 static frentry_t *fr_checkicmpmatchingstate (ip_t *, fr_info_t *);
 124 static int fr_matchicmpqueryreply (int, ipstate_t *, icmphdr_t *, int);
 125 static int fr_state_flush (int, int);
 126 static ips_stat_t *fr_statetstats (void);
 127 static void fr_delstate (ipstate_t *);
 128 static int fr_state_remove (caddr_t);
 129 static void fr_ipsmove (ipstate_t **, ipstate_t *, u_int);
 130 static int fr_tcpoptions (tcphdr_t *);
 131 int fr_stputent (caddr_t);
 132 int fr_stgetent (caddr_t);
 133 void fr_stinsert (ipstate_t *);
 134
 135
 136 #define FIVE_DAYS       (2 * 5 * 86400) /* 5 days: half closed session */
 137
 138 #define TCP_MSL 240                     /* 2 minutes */
 139 u_long  fr_tcpidletimeout = FIVE_DAYS,
 140         fr_tcpclosewait = 2 * TCP_MSL,
 141         fr_tcplastack = 2 * TCP_MSL,
 142         fr_tcptimeout = 2 * TCP_MSL,
 143         fr_tcpclosed = 120,
 144         fr_tcphalfclosed = 2 * 2 * 3600,    /* 2 hours */
 145         fr_udptimeout = 240,
 146         fr_udpacktimeout = 24,
 147         fr_icmptimeout = 120,
 148         fr_icmpacktimeout = 12;
 149 int     fr_statemax = IPSTATE_MAX,
 150         fr_statesize = IPSTATE_SIZE;
 151 int     fr_state_doflush = 0,
 152         fr_state_lock = 0;
 153 ipstate_t *ips_list = NULL;
 154
 155 static  int icmpreplytype4[ICMP_MAXTYPE + 1];
 156 #ifdef  USE_INET6
 157 static  int icmpreplytype6[ICMP6_MAXTYPE + 1];
 158 #endif
 159
 160 int fr_stateinit()
 161 {
 162         int i;
 163
 164         KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *));
 165         if (ips_table != NULL)
 166                 bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *));
 167         else
 168                 return -1;
 169
 170         /* fill icmp reply type table */
 171         for (i = 0; i <= ICMP_MAXTYPE; i++)
 172                 icmpreplytype4[i] = -1;
 173         icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY;
 174         icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY;
 175         icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY;
 176         icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY;
 177 #ifdef  USE_INET6
 178         /* fill icmp reply type table */
 179         for (i = 0; i <= ICMP6_MAXTYPE; i++)
 180                 icmpreplytype6[i] = -1;
 181         icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY;
 182         icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT;
 183         icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY;
 184         icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT;
 185         icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT;
 186 #endif
 187
 188         return 0;
 189 }
 190
 191
 192 static ips_stat_t *fr_statetstats()
 193 {
 194         ips_stats.iss_active = ips_num;
 195         ips_stats.iss_table = ips_table;
 196         ips_stats.iss_list = ips_list;
 197         return &ips_stats;
 198 }
 199
 200
 201 /*
 202  * flush state tables.  two actions currently defined:
 203  * which == 0 : flush all state table entries
 204  * which == 1 : flush TCP connections which have started to close but are
 205  *              stuck for some reason.
 206  * which == 2 : flush TCP connections which have been idle for a long time,
 207  *              starting at > 4 days idle and working back in successive half-
 208  *              days to at most 12 hours old.
 209  */
 210 static int fr_state_flush(which, proto)
 211 int which, proto;
 212 {
 213         ipstate_t *is, **isp;
 214 #if defined(_KERNEL) && !SOLARIS
 215         int s;
 216 #endif
 217         int delete, removed = 0, try;
 218
 219         SPL_NET(s);
 220         for (isp = &ips_list; (is = *isp); ) {
 221                 delete = 0;
 222
 223                 if ((proto != 0) && (is->is_v != proto))
 224                         continue;
 225
 226                 switch (which)
 227                 {
 228                 case 0 :
 229                         delete = 1;
 230                         break;
 231                 case 1 :
 232                 case 2 :
 233                         if (is->is_p != IPPROTO_TCP)
 234                                 break;
 235                         if ((is->is_state[0] != TCPS_ESTABLISHED) ||
 236                             (is->is_state[1] != TCPS_ESTABLISHED))
 237                                 delete = 1;
 238                         break;
 239                 }
 240
 241                 if (delete) {
 242                         if (is->is_p == IPPROTO_TCP)
 243                                 ips_stats.iss_fin++;
 244                         else
 245                                 ips_stats.iss_expire++;
 246 #ifdef  IPFILTER_LOG
 247                         ipstate_log(is, ISL_FLUSH);
 248 #endif
 249                         fr_delstate(is);
 250                         removed++;
 251                 } else
 252                         isp = &is->is_next;
 253         }
 254
 255         /*
 256          * Asked to remove inactive entries, try again if first attempt
 257          * failed.  In this case, 86400 is half a day because the counter is
 258          * activated every half second.
 259          */
 260         if ((which == 2) && (removed == 0)) {
 261                 try = 86400;    /* half a day */
 262                 for (; (try < FIVE_DAYS) && (removed == 0); try += 86400) {
 263                         for (isp = &ips_list; (is = *isp); ) {
 264                                 delete = 0;
 265                                 if ((is->is_p == IPPROTO_TCP) &&
 266                                     ((is->is_state[0] == TCPS_ESTABLISHED) ||
 267                                      (is->is_state[1] == TCPS_ESTABLISHED)) &&
 268                                     (is->is_age < try)) {
 269                                         ips_stats.iss_fin++;
 270                                         delete = 1;
 271                                 } else if ((is->is_p != IPPROTO_TCP) &&
 272                                            (is->is_pkts > 1)) {
 273                                         ips_stats.iss_expire++;
 274                                         delete = 1;
 275                                 }
 276                                 if (delete) {
 277 #ifdef  IPFILTER_LOG
 278                                         ipstate_log(is, ISL_FLUSH);
 279 #endif
 280                                         fr_delstate(is);
 281                                         removed++;
 282                                 } else
 283                                         isp = &is->is_next;
 284                         }
 285                 }
 286         }
 287
 288         SPL_X(s);
 289         return removed;
 290 }
 291
 292
 293 static int fr_state_remove(data)
 294 caddr_t data;
 295 {
 296         ipstate_t *sp, st;
 297         int error;
 298
 299         sp = &st;
 300         error = IRCOPYPTR(data, (caddr_t)&st, sizeof(st));
 301         if (error)
 302                 return EFAULT;
 303
 304         WRITE_ENTER(&ipf_state);
 305         for (sp = ips_list; sp; sp = sp->is_next)
 306                 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
 307                     !bcmp((char *)&sp->is_src, (char *)&st.is_src,
 308                           sizeof(st.is_src)) &&
 309                     !bcmp((char *)&sp->is_dst, (char *)&st.is_dst,
 310                           sizeof(st.is_dst)) &&
 311                     !bcmp((char *)&sp->is_ps, (char *)&st.is_ps,
 312                           sizeof(st.is_ps))) {
 313 #ifdef  IPFILTER_LOG
 314                         ipstate_log(sp, ISL_REMOVE);
 315 #endif
 316                         fr_delstate(sp);
 317                         RWLOCK_EXIT(&ipf_state);
 318                         return 0;
 319                 }
 320         RWLOCK_EXIT(&ipf_state);
 321         return ESRCH;
 322 }
 323
 324
 325 int fr_state_ioctl(data, cmd, mode)
 326 caddr_t data;
 327 #if defined(__NetBSD__) || defined(__OpenBSD__)
 328 u_long cmd;
 329 #else
 330 int cmd;
 331 #endif
 332 int mode;
 333 {
 334         int arg, ret, error = 0;
 335
 336         switch (cmd)
 337         {
 338         case SIOCDELST :
 339                 error = fr_state_remove(data);
 340                 break;
 341         case SIOCIPFFL :
 342                 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg));
 343                 if (error)
 344                         break;
 345                 if (arg == 0 || arg == 1) {
 346                         WRITE_ENTER(&ipf_state);
 347                         ret = fr_state_flush(arg, 4);
 348                         RWLOCK_EXIT(&ipf_state);
 349                         error = IWCOPY((caddr_t)&ret, data, sizeof(ret));
 350                 } else
 351                         error = EINVAL;
 352                 break;
 353 #ifdef USE_INET6
 354         case SIOCIPFL6 :
 355                 error = IRCOPY(data, (caddr_t)&arg, sizeof(arg));
 356                 if (error)
 357                         break;
 358                 if (arg == 0 || arg == 1) {
 359                         WRITE_ENTER(&ipf_state);
 360                         ret = fr_state_flush(arg, 6);
 361                         RWLOCK_EXIT(&ipf_state);
 362                         error = IWCOPY((caddr_t)&ret, data, sizeof(ret));
 363                 } else
 364                         error = EINVAL;
 365                 break;
 366 #endif
 367 #ifdef  IPFILTER_LOG
 368         case SIOCIPFFB :
 369                 if (!(mode & FWRITE))
 370                         error = EPERM;
 371                 else {
 372                         int tmp;
 373
 374                         tmp = ipflog_clear(IPL_LOGSTATE);
 375                         IWCOPY((char *)&tmp, data, sizeof(tmp));
 376                 }
 377                 break;
 378 #endif
 379         case SIOCGETFS :
 380                 error = IWCOPYPTR((caddr_t)fr_statetstats(), data,
 381                                   sizeof(ips_stat_t));
 382                 break;
 383         case FIONREAD :
 384 #ifdef  IPFILTER_LOG
 385                 arg = (int)iplused[IPL_LOGSTATE];
 386                 error = IWCOPY((caddr_t)&arg, (caddr_t)data, sizeof(arg));
 387 #endif
 388                 break;
 389         case SIOCSTLCK :
 390                 error = fr_lock(data, &fr_state_lock);
 391                 break;
 392         case SIOCSTPUT :
 393                 if (!fr_state_lock) {
 394                         error = EACCES;
 395                         break;
 396                 }
 397                 error = fr_stputent(data);
 398                 break;
 399         case SIOCSTGET :
 400                 if (!fr_state_lock) {
 401                         error = EACCES;
 402                         break;
 403                 }
 404                 error = fr_stgetent(data);
 405                 break;
 406         default :
 407                 error = EINVAL;
 408                 break;
 409         }
 410         return error;
 411 }
 412
 413
 414 /*
 415  * Copy out state information from the kernel to a user space process.
 416  */
 417 int fr_stgetent(data)
 418 caddr_t data;
 419 {
 420         ipstate_t *is, *isn;
 421         ipstate_save_t ips;
 422         int error;
 423
 424         error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
 425         if (error)
 426                 return error;
 427
 428         isn = ips.ips_next;
 429         if (!isn) {
 430                 isn = ips_list;
 431                 if (isn == NULL) {
 432                         if (ips.ips_next == NULL)
 433                                 return ENOENT;
 434                         return 0;
 435                 }
 436         } else {
 437                 /*
 438                  * Make sure the pointer we're copying from exists in the
 439                  * current list of entries.  Security precaution to prevent
 440                  * copying of random kernel data.
 441                  */
 442                 for (is = ips_list; is; is = is->is_next)
 443                         if (is == isn)
 444                                 break;
 445                 if (!is)
 446                         return ESRCH;
 447         }
 448         ips.ips_next = isn->is_next;
 449         bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
 450         if (isn->is_rule)
 451                 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
 452                       sizeof(ips.ips_fr));
 453         error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
 454         if (error)
 455                 error = EFAULT;
 456         return error;
 457 }
 458
 459
 460 int fr_stputent(data)
 461 caddr_t data;
 462 {
 463         ipstate_t *is, *isn;
 464         ipstate_save_t ips;
 465         int error, out, i;
 466         frentry_t *fr;
 467         char *name;
 468
 469         error = IRCOPYPTR(data, (caddr_t)&ips, sizeof(ips));
 470         if (error)
 471                 return error;
 472
 473         KMALLOC(isn, ipstate_t *);
 474         if (isn == NULL)
 475                 return ENOMEM;
 476
 477         bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
 478         fr = isn->is_rule;
 479         if (fr != NULL) {
 480                 if (isn->is_flags & FI_NEWFR) {
 481                         KMALLOC(fr, frentry_t *);
 482                         if (fr == NULL) {
 483                                 KFREE(isn);
 484                                 return ENOMEM;
 485                         }
 486                         bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
 487                         out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
 488                         isn->is_rule = fr;
 489                         ips.ips_is.is_rule = fr;
 490
 491                         /*
 492                          * Look up all the interface names in the rule.
 493                          */
 494                         for (i = 0; i < 4; i++) {
 495                                 name = fr->fr_ifnames[i];
 496                                 if ((name[1] == '\0') &&
 497                                     ((name[0] == '-') || (name[0] == '*'))) {
 498                                         fr->fr_ifas[i] = NULL;
 499                                 } else if (*name != '\0') {
 500                                         fr->fr_ifas[i] = GETUNIT(name,
 501                                                                  fr->fr_v);
 502                                         if (fr->fr_ifas[i] == NULL)
 503                                                 fr->fr_ifas[i] = (void *)-1;
 504                                         else {
 505                                                 strncpy(isn->is_ifname[i],
 506                                                         IFNAME(fr->fr_ifas[i]),
 507                                                         IFNAMSIZ);
 508                                         }
 509                                 }
 510                                 isn->is_ifp[out] = fr->fr_ifas[i];
 511                         }
 512
 513                         /*
 514                          * send a copy back to userland of what we ended up
 515                          * to allow for verification.
 516                          */
 517                         error = IWCOPYPTR((caddr_t)&ips, data, sizeof(ips));
 518                         if (error) {
 519                                 KFREE(isn);
 520                                 KFREE(fr);
 521                                 return EFAULT;
 522                         }
 523                 } else {
 524                         for (is = ips_list; is; is = is->is_next)
 525                                 if (is->is_rule == fr)
 526                                         break;
 527                         if (!is) {
 528                                 KFREE(isn);
 529                                 return ESRCH;
 530                         }
 531                 }
 532         }
 533         fr_stinsert(isn);
 534         return 0;
 535 }
 536
 537
 538 /*
 539  * Insert a state table entry manually.
 540  */
 541 void fr_stinsert(is)
 542 ipstate_t *is;
 543 {
 544         u_int hv = is->is_hv;
 545         char *name;
 546         int i;
 547
 548         MUTEX_INIT(&is->is_lock, "ipf state entry", NULL);
 549
 550         /*
 551          * Look up all the interface names in the state entry.
 552          */
 553         for (i = 0; i < 4; i++) {
 554                 name = is->is_ifname[i];
 555                 if ((name[1] == '\0') &&
 556                     ((name[0] == '-') || (name[0] == '*'))) {
 557                         is->is_ifp[0] = NULL;
 558                 } else if (*name != '\0') {
 559                         is->is_ifp[i] = GETUNIT(name, is->is_v);
 560                         if (is->is_ifp[i] == NULL)
 561                                 is->is_ifp[i] = (void *)-1;
 562                 }
 563         }
 564
 565
 566         /*
 567          * add into list table.
 568          */
 569         if (ips_list)
 570                 ips_list->is_pnext = &is->is_next;
 571         is->is_pnext = &ips_list;
 572         is->is_next = ips_list;
 573         ips_list = is;
 574         if (ips_table[hv])
 575                 ips_table[hv]->is_phnext = &is->is_hnext;
 576         else
 577                 ips_stats.iss_inuse++;
 578         is->is_phnext = ips_table + hv;
 579         is->is_hnext = ips_table[hv];
 580         ips_table[hv] = is;
 581         ips_num++;
 582 }
 583
 584
 585 /*
 586  * Create a new ipstate structure and hang it off the hash table.
 587  */
 588 ipstate_t *fr_addstate(ip, fin, stsave, flags)
 589 ip_t *ip;
 590 fr_info_t *fin;
 591 ipstate_t **stsave;
 592 u_int flags;
 593 {
 594         tcphdr_t *tcp = NULL;
 595         ipstate_t *is;
 596          u_int hv;
 597         struct icmp *ic;
 598         ipstate_t ips;
 599         int out, ws;
 600         u_int pass;
 601         void *ifp;
 602
 603         if (fr_state_lock || (fin->fin_off != 0) || (fin->fin_fl & FI_SHORT) ||
 604             (fin->fin_misc & FM_BADSTATE))
 605                 return NULL;
 606         if (ips_num == fr_statemax) {
 607                 ips_stats.iss_max++;
 608                 fr_state_doflush = 1;
 609                 return NULL;
 610         }
 611         out = fin->fin_out;
 612         is = &ips;
 613         bzero((char *)is, sizeof(*is));
 614         ips.is_age = 1;
 615         /*
 616          * Copy and calculate...
 617          */
 618         hv = (is->is_p = fin->fin_fi.fi_p);
 619         is->is_src = fin->fin_fi.fi_src;
 620         hv += is->is_saddr;
 621         is->is_dst = fin->fin_fi.fi_dst;
 622         hv += is->is_daddr;
 623 #ifdef  USE_INET6
 624         if (fin->fin_v == 6) {
 625                 if ((is->is_p == IPPROTO_ICMPV6) &&
 626                     IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
 627                         /*
 628                          * So you can do keep state with neighbour discovery.
 629                          */
 630                         flags |= FI_W_DADDR;
 631                         hv -= is->is_daddr;
 632                 } else {
 633                         hv += is->is_dst.i6[1];
 634                         hv += is->is_dst.i6[2];
 635                         hv += is->is_dst.i6[3];
 636                 }
 637                 hv += is->is_src.i6[1];
 638                 hv += is->is_src.i6[2];
 639                 hv += is->is_src.i6[3];
 640         }
 641 #endif
 642
 643         switch (is->is_p)
 644         {
 645                 int off;
 646
 647 #ifdef  USE_INET6
 648         case IPPROTO_ICMPV6 :
 649                 ic = (struct icmp *)fin->fin_dp;
 650                 if ((ic->icmp_type & ICMP6_INFOMSG_MASK) == 0)
 651                         return NULL;
 652
 653                 switch (ic->icmp_type)
 654                 {
 655                 case ICMP6_ECHO_REQUEST :
 656                         is->is_icmp.ics_type = ic->icmp_type;
 657                         hv += (is->is_icmp.ics_id = ic->icmp_id);
 658                         hv += (is->is_icmp.ics_seq = ic->icmp_seq);
 659                         break;
 660                 case ICMP6_MEMBERSHIP_QUERY :
 661                 case ND_ROUTER_SOLICIT :
 662                 case ND_NEIGHBOR_SOLICIT :
 663                 case ICMP6_NI_QUERY :
 664                         is->is_icmp.ics_type = ic->icmp_type;
 665                         break;
 666                 default :
 667                         return NULL;
 668                 }
 669                 ATOMIC_INCL(ips_stats.iss_icmp);
 670                 is->is_age = fr_icmptimeout;
 671                 break;
 672 #endif
 673         case IPPROTO_ICMP :
 674                 ic = (struct icmp *)fin->fin_dp;
 675
 676                 switch (ic->icmp_type)
 677                 {
 678                 case ICMP_ECHO :
 679                 case ICMP_TSTAMP :
 680                 case ICMP_IREQ :
 681                 case ICMP_MASKREQ :
 682                         is->is_icmp.ics_type = ic->icmp_type;
 683                         hv += (is->is_icmp.ics_id = ic->icmp_id);
 684                         hv += (is->is_icmp.ics_seq = ic->icmp_seq);
 685                         break;
 686                 default :
 687                         return NULL;
 688                 }
 689                 ATOMIC_INCL(ips_stats.iss_icmp);
 690                 is->is_age = fr_icmptimeout;
 691                 break;
 692         case IPPROTO_TCP :
 693                 tcp = (tcphdr_t *)fin->fin_dp;
 694
 695                 if (tcp->th_flags & TH_RST)
 696                         return NULL;
 697                 /*
 698                  * The endian of the ports doesn't matter, but the ack and
 699                  * sequence numbers do as we do mathematics on them later.
 700                  */
 701                 is->is_sport = htons(fin->fin_data[0]);
 702                 is->is_dport = htons(fin->fin_data[1]);
 703                 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
 704                         hv += is->is_sport;
 705                         hv += is->is_dport;
 706                 }
 707                 if ((flags & FI_IGNOREPKT) == 0) {
 708                         is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
 709                                       (off = (tcp->th_off << 2)) +
 710                                       ((tcp->th_flags & TH_SYN) ? 1 : 0) +
 711                                       ((tcp->th_flags & TH_FIN) ? 1 : 0);
 712                         is->is_maxsend = is->is_send;
 713
 714                         if ((tcp->th_flags & TH_SYN) &&
 715                             ((tcp->th_off << 2) >= (sizeof(*tcp) + 4))) {
 716                                 ws = fr_tcpoptions(tcp);
 717                                 if (ws >= 0)
 718                                         is->is_swscale = ws;
 719                         }
 720                 }
 721
 722                 is->is_maxdwin = 1;
 723                 is->is_maxswin = ntohs(tcp->th_win);
 724                 if (is->is_maxswin == 0)
 725                         is->is_maxswin = 1;
 726
 727                 if ((tcp->th_flags & TH_OPENING) == TH_SYN)
 728                         is->is_fsm = 1;
 729
 730                 /*
 731                  * If we're creating state for a starting connection, start the
 732                  * timer on it as we'll never see an error if it fails to
 733                  * connect.
 734                  */
 735                 ATOMIC_INCL(ips_stats.iss_tcp);
 736                 break;
 737
 738         case IPPROTO_UDP :
 739                 tcp = (tcphdr_t *)fin->fin_dp;
 740
 741                 is->is_sport = htons(fin->fin_data[0]);
 742                 is->is_dport = htons(fin->fin_data[1]);
 743                 if ((flags & (FI_W_DPORT|FI_W_SPORT)) == 0) {
 744                         hv += is->is_sport;
 745                         hv += is->is_dport;
 746                 }
 747                 ATOMIC_INCL(ips_stats.iss_udp);
 748                 is->is_age = fr_udptimeout;
 749                 break;
 750         default :
 751                 is->is_age = fr_udptimeout;
 752                 break;
 753         }
 754
 755         KMALLOC(is, ipstate_t *);
 756         if (is == NULL) {
 757                 ATOMIC_INCL(ips_stats.iss_nomem);
 758                 return NULL;
 759         }
 760         bcopy((char *)&ips, (char *)is, sizeof(*is));
 761         hv %= fr_statesize;
 762         is->is_hv = hv;
 763         is->is_rule = fin->fin_fr;
 764         if (is->is_rule != NULL) {
 765                 is->is_group = is->is_rule->fr_group;
 766                 ATOMIC_INC32(is->is_rule->fr_ref);
 767                 pass = is->is_rule->fr_flags;
 768                 is->is_frage[0] = is->is_rule->fr_age[0];
 769                 is->is_frage[1] = is->is_rule->fr_age[1];
 770                 if (is->is_frage[0] != 0)
 771                         is->is_age = is->is_frage[0];
 772
 773                 is->is_ifp[(out << 1) + 1] = is->is_rule->fr_ifas[1];
 774                 is->is_ifp[(1 - out) << 1] = is->is_rule->fr_ifas[2];
 775                 is->is_ifp[((1 - out) << 1) + 1] = is->is_rule->fr_ifas[3];
 776
 777                 if (((ifp = is->is_rule->fr_ifas[1]) != NULL) &&
 778                     (ifp != (void *)-1))
 779                         strncpy(is->is_ifname[(out << 1) + 1],
 780                                 IFNAME(ifp), IFNAMSIZ);
 781                 if (((ifp = is->is_rule->fr_ifas[2]) != NULL) &&
 782                     (ifp != (void *)-1))
 783                         strncpy(is->is_ifname[(1 - out) << 1],
 784                                 IFNAME(ifp), IFNAMSIZ);
 785                 if (((ifp = is->is_rule->fr_ifas[3]) != NULL) &&
 786                     (ifp != (void *)-1))
 787                         strncpy(is->is_ifname[((1 - out) << 1) + 1],
 788                                 IFNAME(ifp), IFNAMSIZ);
 789         } else
 790                 pass = fr_flags;
 791
 792         is->is_ifp[out << 1] = fin->fin_ifp;
 793         strncpy(is->is_ifname[out << 1], IFNAME(fin->fin_ifp), IFNAMSIZ);
 794
 795         WRITE_ENTER(&ipf_state);
 796
 797         is->is_pass = pass;
 798         if ((flags & FI_IGNOREPKT) == 0) {
 799                 is->is_pkts = 1;
 800                 is->is_bytes = fin->fin_dlen + fin->fin_hlen;
 801         }
 802         /*
 803          * We want to check everything that is a property of this packet,
 804          * but we don't (automatically) care about it's fragment status as
 805          * this may change.
 806          */
 807         is->is_v = fin->fin_v;
 808         is->is_rulen = fin->fin_rule;
 809         is->is_opt = fin->fin_fi.fi_optmsk;
 810         is->is_optmsk = 0xffffffff;
 811         is->is_sec = fin->fin_fi.fi_secmsk;
 812         is->is_secmsk = 0xffff;
 813         is->is_auth = fin->fin_fi.fi_auth;
 814         is->is_authmsk = 0xffff;
 815         is->is_flags = fin->fin_fl & FI_CMP;
 816         is->is_flags |= FI_CMP << 4;
 817         is->is_flags |= flags & (FI_WILDP|FI_WILDA);
 818         if (flags & (FI_WILDP|FI_WILDA))
 819                 ips_wild++;
 820
 821         if (pass & FR_LOGFIRST)
 822                 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
 823         fr_stinsert(is);
 824         is->is_me = stsave;
 825         if (is->is_p == IPPROTO_TCP) {
 826                 fr_tcp_age(&is->is_age, is->is_state, fin,
 827                            0, is->is_fsm); /* 0 = packet from the source */
 828         }
 829 #ifdef  IPFILTER_LOG
 830         ipstate_log(is, ISL_NEW);
 831 #endif
 832         RWLOCK_EXIT(&ipf_state);
 833         fin->fin_rev = IP6NEQ(is->is_dst, fin->fin_fi.fi_dst);
 834         if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
 835                 ipfr_newfrag(ip, fin);
 836         return is;
 837 }
 838
 839
 840 static int fr_tcpoptions(tcp)
 841 tcphdr_t *tcp;
 842 {
 843         u_char *opt, *last;
 844         int wscale;
 845
 846         opt = (u_char *) (tcp + 1);
 847         last = ((u_char *)tcp) + (tcp->th_off << 2);
 848
 849         /* If we don't find wscale here, we need to clear it */
 850         wscale = -2;
 851
 852         /* Termination condition picked such that opt[0 .. 2] exist */
 853         while ((opt < last - 2)  && (*opt != TCPOPT_EOL)) {
 854                 switch (*opt) {
 855                 case TCPOPT_NOP:
 856                         opt++;
 857                         continue;
 858                 case TCPOPT_WSCALE:
 859                         /* Proper length ? */
 860                         if (opt[1] == 3) {
 861                                 if (opt[2] > 14)
 862                                         wscale = 14;
 863                                 else
 864                                         wscale = opt[2];
 865                         }
 866                         break;
 867                 default:
 868                         /* Unknown options must be two bytes+ */
 869                         if (opt[1] < 2)
 870                                 break;
 871                         opt += opt[1];
 872                         continue;
 873                 }
 874                 break;
 875         }
 876         return wscale;
 877 }
 878
 879
 880
 881 /*
 882  * check to see if a packet with TCP headers fits within the TCP window.
 883  * change timeout depending on whether new packet is a SYN-ACK returning for a
 884  * SYN or a RST or FIN which indicate time to close up shop.
 885  */
 886 int fr_tcpstate(is, fin, ip, tcp)
 887 ipstate_t *is;
 888 fr_info_t *fin;
 889 ip_t *ip;
 890 tcphdr_t *tcp;
 891 {
 892         tcp_seq seq, ack, end;
 893         int ackskew;
 894         tcpdata_t  *fdata, *tdata;
 895         u_32_t  win, maxwin;
 896         int ret = 0, off;
 897         int source;
 898         int wscale;
 899
 900         /*
 901          * Find difference between last checked packet and this packet.
 902          */
 903         source = IP6EQ(fin->fin_fi.fi_src, is->is_src);
 904         if (source && (ntohs(is->is_sport) != fin->fin_data[0]))
 905                 source = 0;
 906         fdata = &is->is_tcp.ts_data[!source];
 907         tdata = &is->is_tcp.ts_data[source];
 908         off = tcp->th_off << 2;
 909         seq = ntohl(tcp->th_seq);
 910         ack = ntohl(tcp->th_ack);
 911         win = ntohs(tcp->th_win);
 912         end = seq + fin->fin_dlen - off +
 913                ((tcp->th_flags & TH_SYN) ? 1 : 0) +
 914                ((tcp->th_flags & TH_FIN) ? 1 : 0);
 915
 916
 917         if ((tcp->th_flags & TH_SYN) && (off >= sizeof(*tcp) + 4))
 918                 wscale = fr_tcpoptions(tcp);
 919         else
 920                 wscale = -1;
 921
 922         MUTEX_ENTER(&is->is_lock);
 923
 924         if (wscale >= 0)
 925                 fdata->td_wscale = wscale;
 926         else if (wscale == -2)
 927                 fdata->td_wscale = tdata->td_wscale = 0;
 928         win <<= fdata->td_wscale;
 929
 930         if ((fdata->td_end == 0) &&
 931             (!is->is_fsm || ((tcp->th_flags & TH_OPENING) == TH_OPENING))) {
 932                 /*
 933                  * Must be a (outgoing) SYN-ACK in reply to a SYN.
 934                  */
 935                 fdata->td_end = end;
 936                 fdata->td_maxwin = 1;
 937                 fdata->td_maxend = end + win;
 938                 if (win == 0)
 939                         fdata->td_maxend++;
 940         }
 941
 942         if (!(tcp->th_flags & TH_ACK)) {  /* Pretend an ack was sent */
 943                 ack = tdata->td_end;
 944         } else if (((tcp->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
 945                    (ack == 0)) {
 946                 /* gross hack to get around certain broken tcp stacks */
 947                 ack = tdata->td_end;
 948         }
 949
 950         if (seq == end)
 951                 seq = end = fdata->td_end;
 952
 953         maxwin = tdata->td_maxwin;
 954         ackskew = tdata->td_end - ack;
 955
 956 #define SEQ_GE(a,b)     ((int)((a) - (b)) >= 0)
 957 #define SEQ_GT(a,b)     ((int)((a) - (b)) > 0)
 958         if ((SEQ_GE(fdata->td_maxend, end)) &&
 959             (SEQ_GE(seq, fdata->td_end - maxwin)) &&
 960 /* XXX what about big packets */
 961 #define MAXACKWINDOW 66000
 962             (ackskew >= -MAXACKWINDOW) &&
 963             (ackskew <= MAXACKWINDOW)) {
 964                 /* if ackskew < 0 then this should be due to fragented
 965                  * packets. There is no way to know the length of the
 966                  * total packet in advance.
 967                  * We do know the total length from the fragment cache though.
 968                  * Note however that there might be more sessions with
 969                  * exactly the same source and destination paramters in the
 970                  * state cache (and source and destination is the only stuff
 971                  * that is saved in the fragment cache). Note further that
 972                  * some TCP connections in the state cache are hashed with
 973                  * sport and dport as well which makes it not worthwhile to
 974                  * look for them.
 975                  * Thus, when ackskew is negative but still seems to belong
 976                  * to this session, we bump up the destinations end value.
 977                  */
 978                 /*
 979                  * Nearing end of connection, start timeout.
 980                  */
 981                 /* source ? 0 : 1 -> !source */
 982                 if (fr_tcp_age(&is->is_age, is->is_state, fin, !source,
 983                                (int)is->is_fsm) == 0) {
 984                         if (ackskew < 0)
 985                                 tdata->td_end = ack;
 986
 987                         /* update max window seen */
 988                         if (fdata->td_maxwin < win)
 989                                 fdata->td_maxwin = win;
 990                         if (SEQ_GT(end, fdata->td_end))
 991                                 fdata->td_end = end;
 992                         if (SEQ_GE(ack + win, tdata->td_maxend)) {
 993                                 tdata->td_maxend = ack + win;
 994                                 if (win == 0)
 995                                         tdata->td_maxend++;
 996                         }
 997
 998                         ATOMIC_INCL(ips_stats.iss_hits);
 999                         ret = 1;
1000                 }
1001         }
1002         MUTEX_EXIT(&is->is_lock);
1003         if ((ret == 0) && ((tcp->th_flags & TH_OPENING) != TH_SYN))
1004                 fin->fin_misc |= FM_BADSTATE;
1005         return ret;
1006 }
1007
1008
1009 /*
1010  * Match a state table entry against an IP packet.
1011  */
1012 static int fr_matchsrcdst(is, src, dst, fin, tcp)
1013 ipstate_t *is;
1014 union i6addr src, dst;
1015 fr_info_t *fin;
1016 tcphdr_t *tcp;
1017 {
1018         int ret = 0, rev, out, flags, idx;
1019         u_short sp, dp;
1020         void *ifp;
1021
1022         rev = IP6NEQ(is->is_dst, dst);
1023         ifp = fin->fin_ifp;
1024         out = fin->fin_out;
1025         flags = is->is_flags & (FI_WILDA|FI_WILDP);
1026         sp = 0;
1027         dp = 0;
1028
1029         if (tcp != NULL) {
1030                 flags = is->is_flags;
1031                 sp = tcp->th_sport;
1032                 dp = tcp->th_dport;
1033                 if (!rev) {
1034                         if (!(flags & FI_W_SPORT) && (sp != is->is_sport))
1035                                 rev = 1;
1036                         else if (!(flags & FI_W_DPORT) && (dp != is->is_dport))
1037                                 rev = 1;
1038                 }
1039         }
1040
1041         idx = (out << 1) + rev;
1042
1043         if ((is->is_ifp[idx] == NULL &&
1044              (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) ||
1045             is->is_ifp[idx] == ifp)
1046                 ret = 1;
1047
1048         if (ret == 0)
1049                 return 0;
1050         ret = 0;
1051
1052         if (rev == 0) {
1053                 if ((IP6EQ(is->is_dst, dst) || (flags & FI_W_DADDR)) &&
1054                     (IP6EQ(is->is_src, src) || (flags & FI_W_SADDR)) &&
1055                     (!tcp || ((sp == is->is_sport || flags & FI_W_SPORT) &&
1056                      (dp == is->is_dport || flags & FI_W_DPORT)))) {
1057                         ret = 1;
1058                 }
1059         } else {
1060                 if ((IP6EQ(is->is_dst, src) || (flags & FI_W_DADDR)) &&
1061                     (IP6EQ(is->is_src, dst) || (flags & FI_W_SADDR)) &&
1062                     (!tcp || ((sp == is->is_dport || flags & FI_W_DPORT) &&
1063                      (dp == is->is_sport || flags & FI_W_SPORT)))) {
1064                         ret = 1;
1065                 }
1066         }
1067         if (ret == 0)
1068                 return 0;
1069
1070         /*
1071          * Whether or not this should be here, is questionable, but the aim
1072          * is to get this out of the main line.
1073          */
1074         if (tcp == NULL)
1075                 flags = is->is_flags & (FI_CMP|(FI_CMP<<4));
1076
1077         if (((fin->fin_fl & (flags >> 4)) != (flags & FI_CMP)) ||
1078             (fin->fin_fi.fi_optmsk != is->is_opt) ||
1079             (fin->fin_fi.fi_secmsk != is->is_sec) ||
1080             (fin->fin_fi.fi_auth != is->is_auth))
1081                 return 0;
1082
1083         flags = is->is_flags & (FI_WILDA|FI_WILDP);
1084         if ((flags & (FI_W_SADDR|FI_W_DADDR))) {
1085                 if ((flags & FI_W_SADDR) != 0) {
1086                         if (rev == 0) {
1087                                 is->is_src = fin->fin_fi.fi_src;
1088                         } else {
1089                                 is->is_src = fin->fin_fi.fi_dst;
1090                         }
1091                 } else if ((flags & FI_W_DADDR) != 0) {
1092                         if (rev == 0) {
1093                                 is->is_dst = fin->fin_fi.fi_dst;
1094                         } else {
1095                                 is->is_dst = fin->fin_fi.fi_src;
1096                         }
1097                 }
1098                 is->is_flags &= ~(FI_W_SADDR|FI_W_DADDR);
1099                 if ((is->is_flags & (FI_WILDA|FI_WILDP)) == 0)
1100                         ips_wild--;
1101         }
1102
1103         if ((flags & (FI_W_SPORT|FI_W_DPORT))) {
1104                 if ((flags & FI_W_SPORT) != 0) {
1105                         if (rev == 0) {
1106                                 is->is_sport = sp;
1107                                 is->is_send = htonl(tcp->th_seq);
1108                         } else {
1109                                 is->is_sport = dp;
1110                                 is->is_send = htonl(tcp->th_ack);
1111                         }
1112                         is->is_maxsend = is->is_send + 1;
1113                 } else if ((flags & FI_W_DPORT) != 0) {
1114                         if (rev == 0) {
1115                                 is->is_dport = dp;
1116                                 is->is_dend = htonl(tcp->th_ack);
1117                         } else {
1118                                 is->is_dport = sp;
1119                                 is->is_dend = htonl(tcp->th_seq);
1120                         }
1121                         is->is_maxdend = is->is_dend + 1;
1122                 }
1123                 is->is_flags &= ~(FI_W_SPORT|FI_W_DPORT);
1124                 ips_wild--;
1125         }
1126
1127         ret = -1;
1128
1129         if (is->is_ifp[idx] == NULL &&
1130             (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*'))
1131                 ret = idx;
1132
1133         if (ret >= 0) {
1134                 is->is_ifp[ret] = ifp;
1135                 strncpy(is->is_ifname[ret], IFNAME(ifp),
1136                         sizeof(is->is_ifname[ret]));
1137         }
1138         fin->fin_rev = rev;
1139         return 1;
1140 }
1141
1142 static int fr_matchicmpqueryreply(v, is, icmp, rev)
1143 int v;
1144 ipstate_t *is;
1145 icmphdr_t *icmp;
1146 {
1147         if (v == 4) {
1148                 /*
1149                  * If we matched its type on the way in, then when going out
1150                  * it will still be the same type.
1151                  */
1152                 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1153                     (rev && (icmpreplytype4[is->is_type] == icmp->icmp_type))) {
1154                         if (icmp->icmp_type != ICMP_ECHOREPLY)
1155                                 return 1;
1156                         if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1157                             (icmp->icmp_seq == is->is_icmp.ics_seq))
1158                                 return 1;
1159                 }
1160         }
1161 #ifdef  USE_INET6
1162         else if (is->is_v == 6) {
1163                 if ((!rev && (icmp->icmp_type == is->is_type)) ||
1164                     (rev && (icmpreplytype6[is->is_type] == icmp->icmp_type))) {
1165                         if (icmp->icmp_type != ICMP6_ECHO_REPLY)
1166                                 return 1;
1167                         if ((icmp->icmp_id == is->is_icmp.ics_id) &&
1168                             (icmp->icmp_seq == is->is_icmp.ics_seq))
1169                                 return 1;
1170                 }
1171         }
1172 #endif
1173         return 0;
1174 }
1175
1176 static frentry_t *fr_checkicmpmatchingstate(ip, fin)
1177 ip_t *ip;
1178 fr_info_t *fin;
1179 {
1180         ipstate_t *is, **isp;
1181         u_short sport, dport;
1182         u_char  pr;
1183         u_short savelen, ohlen;
1184         union i6addr dst, src;
1185         struct icmp *ic;
1186         icmphdr_t *icmp;
1187         fr_info_t ofin;
1188         int type, len;
1189         tcphdr_t *tcp;
1190         frentry_t *fr;
1191         ip_t *oip;
1192         u_int hv;
1193
1194         /*
1195          * Does it at least have the return (basic) IP header ?
1196          * Only a basic IP header (no options) should be with
1197          * an ICMP error header.
1198          */
1199         if (((ip->ip_v != 4) || (ip->ip_hl != 5)) ||
1200             (fin->fin_plen < ICMPERR_MINPKTLEN))
1201                 return NULL;
1202
1203         ic = (struct icmp *)fin->fin_dp;
1204         type = ic->icmp_type;
1205         /*
1206          * If it's not an error type, then return
1207          */
1208         if ((type != ICMP_UNREACH) && (type != ICMP_SOURCEQUENCH) &&
1209             (type != ICMP_REDIRECT) && (type != ICMP_TIMXCEED) &&
1210             (type != ICMP_PARAMPROB))
1211                 return NULL;
1212
1213         oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
1214         ohlen = oip->ip_hl << 2;
1215         if (fin->fin_plen < ICMPERR_MAXPKTLEN + ohlen - sizeof(*oip))
1216                 return NULL;
1217
1218         /*
1219          * Sanity checks.
1220          */
1221         len = fin->fin_dlen - ICMPERR_ICMPHLEN;
1222         if ((len <= 0) || (ohlen > len))
1223                 return NULL;
1224
1225         /*
1226          * Is the buffer big enough for all of it ?  It's the size of the IP
1227          * header claimed in the encapsulated part which is of concern.  It
1228          * may be too big to be in this buffer but not so big that it's
1229          * outside the ICMP packet, leading to TCP deref's causing problems.
1230          * This is possible because we don't know how big oip_hl is when we
1231          * do the pullup early in fr_check() and thus can't gaurantee it is
1232          * all here now.
1233          */
1234 #ifdef  _KERNEL
1235         {
1236         mb_t *m;
1237
1238 # if SOLARIS
1239         m = fin->fin_qfm;
1240         if ((char *)oip + len > (char *)m->b_wptr)
1241                 return NULL;
1242 # else
1243         m = *(mb_t **)fin->fin_mp;
1244         if ((char *)oip + len > (char *)ip + m->m_len)
1245                 return NULL;
1246 # endif
1247         }
1248 #endif
1249
1250         /*
1251          * in the IPv4 case we must zero the i6addr union otherwise
1252          * the IP6EQ and IP6NEQ macros produce the wrong results because
1253          * of the 'junk' in the unused part of the union
1254          */
1255         bzero((char *)&src, sizeof(src));
1256         bzero((char *)&dst, sizeof(dst));
1257         bzero((char *)&ofin, sizeof(ofin));
1258         ofin.fin_ifp = fin->fin_ifp;
1259         ofin.fin_out = !fin->fin_out;
1260         ofin.fin_v = 4;
1261         fr = NULL;
1262
1263         switch (oip->ip_p)
1264         {
1265         case IPPROTO_ICMP :
1266                 icmp = (icmphdr_t *)((char *)oip + ohlen);
1267
1268                 /*
1269                  * a ICMP error can only be generated as a result of an
1270                  * ICMP query, not as the response on an ICMP error
1271                  *
1272                  * XXX theoretically ICMP_ECHOREP and the other reply's are
1273                  * ICMP query's as well, but adding them here seems strange XXX
1274                  */
1275                  if ((icmp->icmp_type != ICMP_ECHO) &&
1276                      (icmp->icmp_type != ICMP_TSTAMP) &&
1277                      (icmp->icmp_type != ICMP_IREQ) &&
1278                      (icmp->icmp_type != ICMP_MASKREQ))
1279                         return NULL;
1280
1281                 /*
1282                  * perform a lookup of the ICMP packet in the state table
1283                  */
1284                 hv = (pr = oip->ip_p);
1285                 src.in4 = oip->ip_src;
1286                 hv += src.in4.s_addr;
1287                 dst.in4 = oip->ip_dst;
1288                 hv += dst.in4.s_addr;
1289                 hv += icmp->icmp_id;
1290                 hv += icmp->icmp_seq;
1291                 hv %= fr_statesize;
1292
1293                 savelen = oip->ip_len;
1294                 oip->ip_len = len;
1295                 fr_makefrip(ohlen, oip, &ofin);
1296                 oip->ip_len = savelen;
1297
1298                 READ_ENTER(&ipf_state);
1299                 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
1300                         if ((is->is_p == pr) && (is->is_v == 4) &&
1301                             fr_matchsrcdst(is, src, dst, &ofin, NULL) &&
1302                             fr_matchicmpqueryreply(is->is_v, is, icmp, fin->fin_rev)) {
1303                                 ips_stats.iss_hits++;
1304                                 is->is_pkts++;
1305                                 is->is_bytes += ip->ip_len;
1306                                 fr = is->is_rule;
1307                                 break;
1308                         }
1309                 RWLOCK_EXIT(&ipf_state);
1310                 return fr;
1311
1312         case IPPROTO_TCP :
1313         case IPPROTO_UDP :
1314                 if (fin->fin_plen < ICMPERR_MAXPKTLEN)
1315                         return NULL;
1316                 break;
1317         default :
1318                 return NULL;
1319         }
1320
1321         tcp = (tcphdr_t *)((char *)oip + ohlen);
1322         dport = tcp->th_dport;
1323         sport = tcp->th_sport;
1324
1325         hv = (pr = oip->ip_p);
1326         src.in4 = oip->ip_src;
1327         hv += src.in4.s_addr;
1328         dst.in4 = oip->ip_dst;
1329         hv += dst.in4.s_addr;
1330         hv += dport;
1331         hv += sport;
1332         hv %= fr_statesize;
1333         /*
1334          * we make an fin entry to be able to feed it to
1335          * matchsrcdst note that not all fields are encessary
1336          * but this is the cleanest way. Note further we fill
1337          * in fin_mp such that if someone uses it we'll get
1338          * a kernel panic. fr_matchsrcdst does not use this.
1339          *
1340          * watch out here, as ip is in host order and oip in network
1341          * order. Any change we make must be undone afterwards.
1342          */
1343         savelen = oip->ip_len;
1344         oip->ip_len = len;
1345         fr_makefrip(ohlen, oip, &ofin);
1346         oip->ip_len = savelen;
1347         READ_ENTER(&ipf_state);
1348         for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1349                 /*
1350                  * Only allow this icmp though if the
1351                  * encapsulated packet was allowed through the
1352                  * other way around. Note that the minimal amount
1353                  * of info present does not allow for checking against
1354                  * tcp internals such as seq and ack numbers.
1355                  */
1356                 if ((is->is_p == pr) && (is->is_v == 4) &&
1357                     fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
1358                         fr = is->is_rule;
1359                         ips_stats.iss_hits++;
1360                         is->is_pkts++;
1361                         is->is_bytes += fin->fin_plen;
1362                         /*
1363                          * we deliberately do not touch the timeouts
1364                          * for the accompanying state table entry.
1365                          * It remains to be seen if that is correct. XXX
1366                          */
1367                         break;
1368                 }
1369         }
1370         RWLOCK_EXIT(&ipf_state);
1371         return fr;
1372 }
1373
1374
1375 /*
1376  * Move a state hash table entry from its old location at is->is_hv to
1377  * its new location, indexed by hv % fr_statesize.
1378  */
1379 static void fr_ipsmove(isp, is, hv)
1380 ipstate_t **isp, *is;
1381 u_int hv;
1382 {
1383         u_int hvm;
1384
1385         hvm = is->is_hv;
1386         /*
1387          * Remove the hash from the old location...
1388          */
1389         if (is->is_hnext)
1390                 is->is_hnext->is_phnext = isp;
1391         *isp = is->is_hnext;
1392         if (ips_table[hvm] == NULL)
1393                 ips_stats.iss_inuse--;
1394
1395         /*
1396          * ...and put the hash in the new one.
1397          */
1398         hvm = hv % fr_statesize;
1399         is->is_hv = hvm;
1400         isp = &ips_table[hvm];
1401         if (*isp)
1402                 (*isp)->is_phnext = &is->is_hnext;
1403         else
1404                 ips_stats.iss_inuse++;
1405         is->is_phnext = isp;
1406         is->is_hnext = *isp;
1407         *isp = is;
1408 }
1409
1410
1411 /*
1412  * Check if a packet has a registered state.
1413  */
1414 frentry_t *fr_checkstate(ip, fin)
1415 ip_t *ip;
1416 fr_info_t *fin;
1417 {
1418         union i6addr dst, src;
1419         ipstate_t *is, **isp;
1420         u_char pr;
1421         u_int hv, hvm, hlen, tryagain, pass, v;
1422         struct icmp *ic;
1423         frentry_t *fr;
1424         tcphdr_t *tcp;
1425         int rev;
1426
1427         if ((ips_list == NULL) || (fin->fin_off != 0) || fr_state_lock ||
1428             (fin->fin_fl & FI_SHORT))
1429                 return NULL;
1430
1431         is = NULL;
1432         hlen = fin->fin_hlen;
1433         tcp = (tcphdr_t *)((char *)ip + hlen);
1434         ic = (struct icmp *)tcp;
1435         hv = (pr = fin->fin_fi.fi_p);
1436         src = fin->fin_fi.fi_src;
1437         dst = fin->fin_fi.fi_dst;
1438         hv += src.in4.s_addr;
1439         hv += dst.in4.s_addr;
1440
1441         /*
1442          * Search the hash table for matching packet header info.
1443          * At the bottom of this switch statement, the following is expected:
1444          * is == NULL, no lock on ipf_state is held.
1445          * is != NULL, a lock on ipf_state is held.
1446          */
1447         v = fin->fin_fi.fi_v;
1448 #ifdef  USE_INET6
1449         if (v == 6) {
1450                 hv += fin->fin_fi.fi_src.i6[1];
1451                 hv += fin->fin_fi.fi_src.i6[2];
1452                 hv += fin->fin_fi.fi_src.i6[3];
1453
1454                 if ((fin->fin_p == IPPROTO_ICMPV6) &&
1455                     IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
1456                         hv -= dst.in4.s_addr;
1457                 } else {
1458                         hv += fin->fin_fi.fi_dst.i6[1];
1459                         hv += fin->fin_fi.fi_dst.i6[2];
1460                         hv += fin->fin_fi.fi_dst.i6[3];
1461                 }
1462         }
1463 #endif
1464
1465         switch (fin->fin_p)
1466         {
1467 #ifdef  USE_INET6
1468         case IPPROTO_ICMPV6 :
1469                 tcp = NULL;
1470                 tryagain = 0;
1471                 if (v == 6) {
1472                         if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
1473                             (ic->icmp_type == ICMP6_ECHO_REPLY)) {
1474                                 hv += ic->icmp_id;
1475                                 hv += ic->icmp_seq;
1476                         }
1477                 }
1478                 READ_ENTER(&ipf_state);
1479 icmp6again:
1480                 hvm = hv % fr_statesize;
1481                 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1482                         if ((is->is_p == pr) && (is->is_v == v) &&
1483                             fr_matchsrcdst(is, src, dst, fin, NULL) &&
1484                             fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1485                                 rev = fin->fin_rev;
1486                                 if (is->is_frage[rev] != 0)
1487                                         is->is_age = is->is_frage[rev];
1488                                 else if (rev != 0)
1489                                         is->is_age = fr_icmpacktimeout;
1490                                 else
1491                                         is->is_age = fr_icmptimeout;
1492                                 break;
1493                         }
1494
1495                 if (is != NULL) {
1496                         if (tryagain && !(is->is_flags & FI_W_DADDR)) {
1497                                 hv += fin->fin_fi.fi_src.i6[0];
1498                                 hv += fin->fin_fi.fi_src.i6[1];
1499                                 hv += fin->fin_fi.fi_src.i6[2];
1500                                 hv += fin->fin_fi.fi_src.i6[3];
1501                                 fr_ipsmove(isp, is, hv);
1502                                 MUTEX_DOWNGRADE(&ipf_state);
1503                         }
1504                         break;
1505                 }
1506                 RWLOCK_EXIT(&ipf_state);
1507
1508                 /*
1509                  * No matching icmp state entry. Perhaps this is a
1510                  * response to another state entry.
1511                  */
1512                 if ((ips_wild != 0) && (v == 6) && (tryagain == 0) &&
1513                     !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
1514                         hv -= fin->fin_fi.fi_src.i6[0];
1515                         hv -= fin->fin_fi.fi_src.i6[1];
1516                         hv -= fin->fin_fi.fi_src.i6[2];
1517                         hv -= fin->fin_fi.fi_src.i6[3];
1518                         tryagain = 1;
1519                         WRITE_ENTER(&ipf_state);
1520                         goto icmp6again;
1521                 }
1522
1523                 fr = fr_checkicmp6matchingstate((ip6_t *)ip, fin);
1524                 if (fr)
1525                         return fr;
1526                 break;
1527 #endif
1528         case IPPROTO_ICMP :
1529                 tcp = NULL;
1530                 if (v == 4) {
1531                         hv += ic->icmp_id;
1532                         hv += ic->icmp_seq;
1533                 }
1534                 hvm = hv % fr_statesize;
1535                 READ_ENTER(&ipf_state);
1536                 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1537                         if ((is->is_p == pr) && (is->is_v == v) &&
1538                             fr_matchsrcdst(is, src, dst, fin, NULL) &&
1539                             fr_matchicmpqueryreply(v, is, ic, fin->fin_rev)) {
1540                                 rev = fin->fin_rev;
1541                                 if (is->is_frage[rev] != 0)
1542                                         is->is_age = is->is_frage[rev];
1543                                 else if (fin->fin_rev)
1544                                         is->is_age = fr_icmpacktimeout;
1545                                 else
1546                                         is->is_age = fr_icmptimeout;
1547                                 break;
1548                         }
1549
1550                 if (is != NULL)
1551                         break;
1552                 RWLOCK_EXIT(&ipf_state);
1553                 /*
1554                  * No matching icmp state entry. Perhaps this is a
1555                  * response to another state entry.
1556                  */
1557                 fr = fr_checkicmpmatchingstate(ip, fin);
1558                 if (fr)
1559                         return fr;
1560                 break;
1561         case IPPROTO_TCP :
1562                 /*
1563                  * Just plain ignore RST flag set with either FIN or SYN.
1564                  */
1565                 if ((tcp->th_flags & TH_RST) &&
1566                     ((tcp->th_flags & (TH_FIN|TH_SYN|TH_RST)) != TH_RST))
1567                         break;
1568         case IPPROTO_UDP :
1569             {
1570                 u_short dport, sport;
1571
1572                 dport = tcp->th_dport;
1573                 sport = tcp->th_sport;
1574                 tryagain = 0;
1575                 hv += dport;
1576                 hv += sport;
1577                 READ_ENTER(&ipf_state);
1578 retry_tcpudp:
1579                 hvm = hv % fr_statesize;
1580                 for (isp = &ips_table[hvm]; (is = *isp); isp = &is->is_hnext)
1581                         if ((is->is_p == pr) && (is->is_v == v) &&
1582                             fr_matchsrcdst(is, src, dst, fin, tcp)) {
1583                                 rev = fin->fin_rev;
1584                                 if ((pr == IPPROTO_TCP)) {
1585                                         if (!fr_tcpstate(is, fin, ip, tcp))
1586                                                 is = NULL;
1587                                 } else if ((pr == IPPROTO_UDP)) {
1588                                         if (is->is_frage[rev] != 0)
1589                                                 is->is_age = is->is_frage[rev];
1590                                         else if (fin->fin_rev)
1591                                                 is->is_age = fr_udpacktimeout;
1592                                         else
1593                                                 is->is_age = fr_udptimeout;
1594                                 }
1595                                 break;
1596                         }
1597                 if (is != NULL) {
1598                         if (tryagain &&
1599                             !(is->is_flags & (FI_WILDP|FI_WILDA))) {
1600                                 hv += dport;
1601                                 hv += sport;
1602                                 fr_ipsmove(isp, is, hv);
1603                                 MUTEX_DOWNGRADE(&ipf_state);
1604                         }
1605                         break;
1606                 }
1607
1608                 RWLOCK_EXIT(&ipf_state);
1609                 if (!tryagain && ips_wild) {
1610                         hv -= dport;
1611                         hv -= sport;
1612                         tryagain = 1;
1613                         WRITE_ENTER(&ipf_state);
1614                         goto retry_tcpudp;
1615                 }
1616                 break;
1617             }
1618         default :
1619                 tcp = NULL;
1620                 hv %= fr_statesize;
1621                 READ_ENTER(&ipf_state);
1622                 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
1623                         if ((is->is_p == pr) && (is->is_v == v) &&
1624                             fr_matchsrcdst(is, src, dst, fin, NULL)) {
1625                                 rev = fin->fin_rev;
1626                                 if (is->is_frage[rev] != 0)
1627                                         is->is_age = is->is_frage[rev];
1628                                 else
1629                                         is->is_age = fr_udptimeout;
1630                                 break;
1631                         }
1632                 }
1633                 if (is == NULL) {
1634                         RWLOCK_EXIT(&ipf_state);
1635                 }
1636                 break;
1637         }
1638
1639         if (is == NULL) {
1640                 ATOMIC_INCL(ips_stats.iss_miss);
1641                 return NULL;
1642         }
1643
1644         MUTEX_ENTER(&is->is_lock);
1645         is->is_bytes += fin->fin_plen;
1646         ips_stats.iss_hits++;
1647         is->is_pkts++;
1648         MUTEX_EXIT(&is->is_lock);
1649         fr = is->is_rule;
1650         fin->fin_rule = is->is_rulen;
1651         if (fr != NULL) {
1652                 fin->fin_group = fr->fr_group;
1653                 fin->fin_icode = fr->fr_icode;
1654         }
1655         fin->fin_fr = fr;
1656         pass = is->is_pass;
1657         RWLOCK_EXIT(&ipf_state);
1658         if ((fin->fin_fl & FI_FRAG) && (pass & FR_KEEPFRAG))
1659                 ipfr_newfrag(ip, fin);
1660 #ifndef _KERNEL
1661         if ((tcp != NULL) && (tcp->th_flags & TCP_CLOSE))
1662                 fr_delstate(is);
1663 #endif
1664         return fr;
1665 }
1666
1667
1668 /*
1669  * Sync. state entries.  If interfaces come or go or just change position,
1670  * this is needed.
1671  */
1672 void ip_statesync(ifp)
1673 void *ifp;
1674 {
1675         ipstate_t *is;
1676         int i;
1677
1678         WRITE_ENTER(&ipf_state);
1679         for (is = ips_list; is; is = is->is_next) {
1680                 for (i = 0; i < 4; i++) {
1681                         if (is->is_ifp[i] == ifp) {
1682                                 is->is_ifp[i] = GETUNIT(is->is_ifname[i],
1683                                                         is->is_v);
1684                                 if (!is->is_ifp[i])
1685                                         is->is_ifp[i] = (void *)-1;
1686                         }
1687                 }
1688         }
1689         RWLOCK_EXIT(&ipf_state);
1690 }
1691
1692
1693 /*
1694  * Must always be called with fr_ipfstate held as a write lock.
1695  */
1696 static void fr_delstate(is)
1697 ipstate_t *is;
1698 {
1699         frentry_t *fr;
1700
1701         if (is->is_flags & (FI_WILDP|FI_WILDA))
1702                 ips_wild--;
1703         if (is->is_next)
1704                 is->is_next->is_pnext = is->is_pnext;
1705         *is->is_pnext = is->is_next;
1706         if (is->is_hnext)
1707                 is->is_hnext->is_phnext = is->is_phnext;
1708         *is->is_phnext = is->is_hnext;
1709         if (ips_table[is->is_hv] == NULL)
1710                 ips_stats.iss_inuse--;
1711         if (is->is_me)
1712                 *is->is_me = NULL;
1713
1714         fr = is->is_rule;
1715         if (fr != NULL) {
1716                 fr->fr_ref--;
1717                 if (fr->fr_ref == 0) {
1718                         KFREE(fr);
1719                 }
1720         }
1721 #ifdef  _KERNEL
1722         MUTEX_DESTROY(&is->is_lock);
1723 #endif
1724         KFREE(is);
1725         ips_num--;
1726 }
1727
1728
1729 /*
1730  * Free memory in use by all state info. kept.
1731  */
1732 void fr_stateunload()
1733 {
1734         ipstate_t *is;
1735
1736         WRITE_ENTER(&ipf_state);
1737         while ((is = ips_list))
1738                 fr_delstate(is);
1739         ips_stats.iss_inuse = 0;
1740         ips_num = 0;
1741         RWLOCK_EXIT(&ipf_state);
1742         if (ips_table)
1743                 KFREES(ips_table, fr_statesize * sizeof(ipstate_t *));
1744         ips_table = NULL;
1745 }
1746
1747
1748 /*
1749  * Slowly expire held state for thingslike UDP and ICMP.  Timeouts are set
1750  * in expectation of this being called twice per second.
1751  */
1752 void fr_timeoutstate()
1753 {
1754         ipstate_t *is, **isp;
1755 #if defined(_KERNEL) && !SOLARIS
1756         int s;
1757 #endif
1758
1759         SPL_NET(s);
1760         WRITE_ENTER(&ipf_state);
1761         for (isp = &ips_list; (is = *isp); )
1762                 if (is->is_age && !--is->is_age) {
1763                         if (is->is_p == IPPROTO_TCP)
1764                                 ips_stats.iss_fin++;
1765                         else
1766                                 ips_stats.iss_expire++;
1767 #ifdef  IPFILTER_LOG
1768                         ipstate_log(is, ISL_EXPIRE);
1769 #endif
1770                         fr_delstate(is);
1771                 } else
1772                         isp = &is->is_next;
1773         if (fr_state_doflush) {
1774                 (void) fr_state_flush(2, 0);
1775                 fr_state_doflush = 0;
1776         }
1777         RWLOCK_EXIT(&ipf_state);
1778         SPL_X(s);
1779 }
1780
1781
1782 /*
1783  * Original idea freom Pradeep Krishnan for use primarily with NAT code.
1784  * (pkrishna@netcom.com)
1785  *
1786  * Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29:
1787  *
1788  * - (try to) base state transitions on real evidence only,
1789  *   i.e. packets that are sent and have been received by ipfilter;
1790  *   diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used.
1791  *
1792  * - deal with half-closed connections correctly;
1793  *
1794  * - store the state of the source in state[0] such that ipfstat
1795  *   displays the state as source/dest instead of dest/source; the calls
1796  *   to fr_tcp_age have been changed accordingly.
1797  *
1798  * Parameters:
1799  *
1800  *    state[0] = state of source (host that initiated connection)
1801  *    state[1] = state of dest   (host that accepted the connection)
1802  *
1803  *    dir == 0 : a packet from source to dest
1804  *    dir == 1 : a packet from dest to source
1805  *
1806  */
1807 int fr_tcp_age(age, state, fin, dir, fsm)
1808 u_long *age;
1809 u_char *state;
1810 fr_info_t *fin;
1811 int dir, fsm;
1812 {
1813         tcphdr_t *tcp = (tcphdr_t *)fin->fin_dp;
1814         u_char flags = tcp->th_flags;
1815         int dlen, ostate;
1816         u_long newage;
1817
1818         ostate = state[1 - dir];
1819
1820         dlen = fin->fin_plen - fin->fin_hlen - (tcp->th_off << 2);
1821
1822         if (flags & TH_RST) {
1823                 if (!(tcp->th_flags & TH_PUSH) && !dlen) {
1824                         *age = fr_tcpclosed;
1825                         state[dir] = TCPS_CLOSED;
1826                 } else {
1827                         *age = fr_tcpclosewait;
1828                         state[dir] = TCPS_CLOSE_WAIT;
1829                 }
1830                 return 0;
1831         }
1832
1833         newage = 0;
1834
1835         switch(state[dir])
1836         {
1837         case TCPS_CLOSED: /* 0 */
1838                 if ((flags & TH_OPENING) == TH_OPENING) {
1839                         /*
1840                          * 'dir' received an S and sends SA in response,
1841                          * CLOSED -> SYN_RECEIVED
1842                          */
1843                         state[dir] = TCPS_SYN_RECEIVED;
1844                         newage = fr_tcptimeout;
1845                 } else if ((flags & TH_OPENING) == TH_SYN) {
1846                         /* 'dir' sent S, CLOSED -> SYN_SENT */
1847                         state[dir] = TCPS_SYN_SENT;
1848                         newage = fr_tcptimeout;
1849                 }
1850
1851                 /*
1852                  * It is apparently possible that a hosts sends two syncs
1853                  * before the remote party is able to respond with a SA. In
1854                  * such a case the remote server sometimes ACK's the second
1855                  * sync, and then responds with a SA. The following code
1856                  * is used to prevent this ack from being blocked.
1857                  *
1858                  * We do not reset the timeout here to fr_tcptimeout because
1859                  * a connection connect timeout does not renew after every
1860                  * packet that is sent.  We need to set newage to something
1861                  * to indicate the packet has passed the check for its flags
1862                  * being valid in the TCP FSM.
1863                  */
1864                 else if ((ostate == TCPS_SYN_SENT) &&
1865                          ((flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK)) {
1866                         newage = *age;
1867                 }
1868
1869                 /*
1870                  * The next piece of code makes it possible to get
1871                  * already established connections into the state table
1872                  * after a restart or reload of the filter rules; this
1873                  * does not work when a strict 'flags S keep state' is
1874                  * used for tcp connections of course, however, use a
1875                  * lower time-out so the state disappears quickly if
1876                  * the other side does not pick it up.
1877                  */
1878                 else if (!fsm &&
1879                          (flags & (TH_FIN|TH_SYN|TH_RST|TH_ACK)) == TH_ACK) {
1880                         /* we saw an A, guess 'dir' is in ESTABLISHED mode */
1881                         if (ostate == TCPS_CLOSED) {
1882                                 state[dir] = TCPS_ESTABLISHED;
1883                                 newage = fr_tcptimeout;
1884                         } else if (ostate == TCPS_ESTABLISHED) {
1885                                 state[dir] = TCPS_ESTABLISHED;
1886                                 newage = fr_tcpidletimeout;
1887                         }
1888                 }
1889                 /*
1890                  * TODO: besides regular ACK packets we can have other
1891                  * packets as well; it is yet to be determined how we
1892                  * should initialize the states in those cases
1893                  */
1894                 break;
1895
1896         case TCPS_LISTEN: /* 1 */
1897                 /* NOT USED */
1898                 break;
1899
1900         case TCPS_SYN_SENT: /* 2 */
1901                 if ((flags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
1902                         /*
1903                          * A retransmitted SYN packet.  We do not reset the
1904                          * timeout here to fr_tcptimeout because a connection
1905                          * connect timeout does not renew after every packet
1906                          * that is sent.  We need to set newage to something
1907                          * to indicate the packet has passed the check for its
1908                          * flags being valid in the TCP FSM.
1909                          */
1910                         newage = *age;
1911                 } else if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1912                         /*
1913                          * We see an A from 'dir' which is in SYN_SENT
1914                          * state: 'dir' sent an A in response to an SA
1915                          * which it received, SYN_SENT -> ESTABLISHED
1916                          */
1917                         state[dir] = TCPS_ESTABLISHED;
1918                         newage = fr_tcpidletimeout;
1919                 } else if (flags & TH_FIN) {
1920                         /*
1921                          * We see an F from 'dir' which is in SYN_SENT
1922                          * state and wants to close its side of the
1923                          * connection; SYN_SENT -> FIN_WAIT_1
1924                          */
1925                         state[dir] = TCPS_FIN_WAIT_1;
1926                         newage = fr_tcpidletimeout; /* or fr_tcptimeout? */
1927                 } else if ((flags & TH_OPENING) == TH_OPENING) {
1928                         /*
1929                          * We see an SA from 'dir' which is already in
1930                          * SYN_SENT state, this means we have a
1931                          * simultaneous open; SYN_SENT -> SYN_RECEIVED
1932                          */
1933                         state[dir] = TCPS_SYN_RECEIVED;
1934                         newage = fr_tcptimeout;
1935                 }
1936                 break;
1937
1938         case TCPS_SYN_RECEIVED: /* 3 */
1939                 if ((flags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
1940                         /*
1941                          * We see an A from 'dir' which was in SYN_RECEIVED
1942                          * state so it must now be in established state,
1943                          * SYN_RECEIVED -> ESTABLISHED
1944                          */
1945                         state[dir] = TCPS_ESTABLISHED;
1946                         newage = fr_tcpidletimeout;
1947                 } else if ((flags & ~(TH_ECN|TH_CWR)) == TH_OPENING) {
1948                         /*
1949                          * We see an SA from 'dir' which is already in
1950                          * SYN_RECEIVED state.
1951                          */
1952                         newage = fr_tcptimeout;
1953                 } else if (flags & TH_FIN) {
1954                         /*
1955                          * We see an F from 'dir' which is in SYN_RECEIVED
1956                          * state and wants to close its side of the connection;
1957                          * SYN_RECEIVED -> FIN_WAIT_1
1958                          */
1959                         state[dir] = TCPS_FIN_WAIT_1;
1960                         newage = fr_tcpidletimeout;
1961                 }
1962                 break;
1963
1964         case TCPS_ESTABLISHED: /* 4 */
1965                 if (flags & TH_FIN) {
1966                         /*
1967                          * 'dir' closed its side of the connection; this
1968                          * gives us a half-closed connection;
1969                          * ESTABLISHED -> FIN_WAIT_1
1970                          */
1971                         state[dir] = TCPS_FIN_WAIT_1;
1972                         newage = fr_tcphalfclosed;
1973                 } else if (flags & TH_ACK) {
1974                         /* an ACK, should we exclude other flags here? */
1975                         if (ostate == TCPS_FIN_WAIT_1) {
1976                                 /*
1977                                  * We know the other side did an active close,
1978                                  * so we are ACKing the recvd FIN packet (does
1979                                  * the window matching code guarantee this?)
1980                                  * and go into CLOSE_WAIT state; this gives us
1981                                  * a half-closed connection
1982                                  */
1983                                 state[dir] = TCPS_CLOSE_WAIT;
1984                                 newage = fr_tcphalfclosed;
1985                         } else if (ostate < TCPS_CLOSE_WAIT)
1986                                 /*
1987                                  * Still a fully established connection,
1988                                  * reset timeout
1989                                  */
1990                                 newage = fr_tcpidletimeout;
1991                 }
1992                 break;
1993
1994         case TCPS_CLOSE_WAIT: /* 5 */
1995                 if (flags & TH_FIN) {
1996                         /*
1997                          * Application closed and 'dir' sent a FIN, we're now
1998                          * going into LAST_ACK state
1999                          */
2000                         newage  = fr_tcplastack;
2001                         state[dir] = TCPS_LAST_ACK;
2002                 } else {
2003                         /*
2004                          * We remain in CLOSE_WAIT because the other side has
2005                          * closed already and we did not close our side yet;
2006                          * reset timeout
2007                          */
2008                         newage  = fr_tcphalfclosed;
2009                 }
2010                 break;
2011
2012         case TCPS_FIN_WAIT_1: /* 6 */
2013                 if ((flags & TH_ACK) && ostate > TCPS_CLOSE_WAIT) {
2014                         /*
2015                          * If the other side is not active anymore it has sent
2016                          * us a FIN packet that we are ack'ing now with an ACK;
2017                          * this means both sides have now closed the connection
2018                          * and we go into TIME_WAIT
2019                          */
2020                         /*
2021                          * XXX: how do we know we really are ACKing the FIN
2022                          * packet here? does the window code guarantee that?
2023                          */
2024                         state[dir] = TCPS_TIME_WAIT;
2025                         newage = fr_tcptimeout;
2026                 } else
2027                         /*
2028                          * We closed our side of the connection already but the
2029                          * other side is still active (ESTABLISHED/CLOSE_WAIT);
2030                          * continue with this half-closed connection
2031                          */
2032                         newage = fr_tcphalfclosed;
2033                 break;
2034
2035         case TCPS_CLOSING: /* 7 */
2036                 /* NOT USED */
2037                 break;
2038
2039         case TCPS_LAST_ACK: /* 8 */
2040                 if (flags & TH_ACK) {
2041                         if ((flags & TH_PUSH) || dlen)
2042                                 /*
2043                                  * There is still data to be delivered, reset
2044                                  * timeout
2045                                  */
2046                                 newage  = fr_tcplastack;
2047                         else
2048                                 newage = *age;
2049                 }
2050                 /*
2051                  * We cannot detect when we go out of LAST_ACK state to CLOSED
2052                  * because that is based on the reception of ACK packets;
2053                  * ipfilter can only detect that a packet has been sent by a
2054                  * host
2055                  */
2056                 break;
2057
2058         case TCPS_FIN_WAIT_2: /* 9 */
2059                 /* NOT USED */
2060                 break;
2061
2062         case TCPS_TIME_WAIT: /* 10 */
2063                 newage = fr_tcptimeout; /* default 4 mins */
2064                 /* we're in 2MSL timeout now */
2065                 break;
2066         }
2067
2068         if (newage != 0) {
2069                 *age = newage;
2070                 return 0;
2071         }
2072         return -1;
2073 }
2074
2075
2076 #ifdef  IPFILTER_LOG
2077 void ipstate_log(is, type)
2078 struct ipstate *is;
2079 u_int type;
2080 {
2081         struct  ipslog  ipsl;
2082         void *items[1];
2083         size_t sizes[1];
2084         int types[1];
2085
2086         ipsl.isl_type = type;
2087         ipsl.isl_pkts = is->is_pkts;
2088         ipsl.isl_bytes = is->is_bytes;
2089         ipsl.isl_src = is->is_src;
2090         ipsl.isl_dst = is->is_dst;
2091         ipsl.isl_p = is->is_p;
2092         ipsl.isl_v = is->is_v;
2093         ipsl.isl_flags = is->is_flags;
2094         ipsl.isl_rulen = is->is_rulen;
2095         ipsl.isl_group = is->is_group;
2096         if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
2097                 ipsl.isl_sport = is->is_sport;
2098                 ipsl.isl_dport = is->is_dport;
2099                 if (ipsl.isl_p == IPPROTO_TCP) {
2100                         ipsl.isl_state[0] = is->is_state[0];
2101                         ipsl.isl_state[1] = is->is_state[1];
2102                 }
2103         } else if (ipsl.isl_p == IPPROTO_ICMP) {
2104                 ipsl.isl_itype = is->is_icmp.ics_type;
2105         } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
2106                 ipsl.isl_itype = is->is_icmp.ics_type;
2107         } else {
2108                 ipsl.isl_ps.isl_filler[0] = 0;
2109                 ipsl.isl_ps.isl_filler[1] = 0;
2110         }
2111         items[0] = &ipsl;
2112         sizes[0] = sizeof(ipsl);
2113         types[0] = 0;
2114
2115         (void) ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1);
2116 }
2117 #endif
2118
2119
2120 #ifdef  USE_INET6
2121 frentry_t *fr_checkicmp6matchingstate(ip, fin)
2122 ip6_t *ip;
2123 fr_info_t *fin;
2124 {
2125         ipstate_t *is, **isp;
2126         u_short sport, dport;
2127         u_char  pr;
2128         struct icmp6_hdr *ic, *oic;
2129         union i6addr dst, src;
2130         u_short savelen;
2131         fr_info_t ofin;
2132         tcphdr_t *tcp;
2133         frentry_t *fr;
2134         ip6_t *oip;
2135         int type;
2136         u_int hv;
2137
2138         /*
2139          * Does it at least have the return (basic) IP header ?
2140          * Only a basic IP header (no options) should be with
2141          * an ICMP error header.
2142          */
2143         if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN))
2144                 return NULL;
2145         ic = (struct icmp6_hdr *)fin->fin_dp;
2146         type = ic->icmp6_type;
2147         /*
2148          * If it's not an error type, then return
2149          */
2150         if ((type != ICMP6_DST_UNREACH) && (type != ICMP6_PACKET_TOO_BIG) &&
2151             (type != ICMP6_TIME_EXCEEDED) && (type != ICMP6_PARAM_PROB))
2152                 return NULL;
2153
2154         oip = (ip6_t *)((char *)ic + ICMPERR_ICMPHLEN);
2155         if (fin->fin_plen < sizeof(*oip))
2156                 return NULL;
2157
2158         if ((oip->ip6_nxt != IPPROTO_TCP) && (oip->ip6_nxt != IPPROTO_UDP) &&
2159             (oip->ip6_nxt != IPPROTO_ICMPV6))
2160                 return NULL;
2161
2162         bzero((char *)&ofin, sizeof(ofin));
2163         ofin.fin_out = !fin->fin_out;
2164         ofin.fin_ifp = fin->fin_ifp;
2165         ofin.fin_v = 6;
2166
2167         if (oip->ip6_nxt == IPPROTO_ICMPV6) {
2168                 oic = (struct icmp6_hdr *)(oip + 1);
2169                 /*
2170                  * a ICMP error can only be generated as a result of an
2171                  * ICMP query, not as the response on an ICMP error
2172                  *
2173                  * XXX theoretically ICMP_ECHOREP and the other reply's are
2174                  * ICMP query's as well, but adding them here seems strange XXX
2175                  */
2176                  if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK))
2177                         return NULL;
2178
2179                 /*
2180                  * perform a lookup of the ICMP packet in the state table
2181                  */
2182                 hv = (pr = oip->ip6_nxt);
2183                 src.in6 = oip->ip6_src;
2184                 hv += src.in4.s_addr;
2185                 dst.in6 = oip->ip6_dst;
2186                 hv += dst.in4.s_addr;
2187                 hv += oic->icmp6_id;
2188                 hv += oic->icmp6_seq;
2189                 hv %= fr_statesize;
2190
2191                 oip->ip6_plen = ntohs(oip->ip6_plen);
2192                 fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2193                 oip->ip6_plen = htons(oip->ip6_plen);
2194
2195                 READ_ENTER(&ipf_state);
2196                 for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext)
2197                         if ((is->is_p == pr) &&
2198                             (oic->icmp6_id == is->is_icmp.ics_id) &&
2199                             (oic->icmp6_seq == is->is_icmp.ics_seq) &&
2200                             fr_matchsrcdst(is, src, dst, &ofin, NULL)) {
2201                                 /*
2202                                  * in the state table ICMP query's are stored
2203                                  * with the type of the corresponding ICMP
2204                                  * response. Correct here
2205                                  */
2206                                 if (((is->is_type == ICMP6_ECHO_REPLY) &&
2207                                      (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
2208                                      (is->is_type - 1 == oic->icmp6_type )) {
2209                                         ips_stats.iss_hits++;
2210                                         is->is_pkts++;
2211                                         is->is_bytes += fin->fin_plen;
2212                                         return is->is_rule;
2213                                 }
2214                         }
2215                 RWLOCK_EXIT(&ipf_state);
2216
2217                 return NULL;
2218         }
2219
2220         tcp = (tcphdr_t *)(oip + 1);
2221         dport = tcp->th_dport;
2222         sport = tcp->th_sport;
2223
2224         hv = (pr = oip->ip6_nxt);
2225         src.in6 = oip->ip6_src;
2226         hv += src.in4.s_addr;
2227         hv += src.i6[1];
2228         hv += src.i6[2];
2229         hv += src.i6[3];
2230         dst.in6 = oip->ip6_dst;
2231         hv += dst.in4.s_addr;
2232         hv += dst.i6[1];
2233         hv += dst.i6[2];
2234         hv += dst.i6[3];
2235         hv += dport;
2236         hv += sport;
2237         hv %= fr_statesize;
2238         /*
2239          * we make an fin entry to be able to feed it to
2240          * matchsrcdst note that not all fields are encessary
2241          * but this is the cleanest way. Note further we fill
2242          * in fin_mp such that if someone uses it we'll get
2243          * a kernel panic. fr_matchsrcdst does not use this.
2244          *
2245          * watch out here, as ip is in host order and oip in network
2246          * order. Any change we make must be undone afterwards.
2247          */
2248         savelen = oip->ip6_plen;
2249         oip->ip6_plen = ip->ip6_plen - sizeof(*ip) - ICMPERR_ICMPHLEN;
2250         fr_makefrip(sizeof(*oip), (ip_t *)oip, &ofin);
2251         oip->ip6_plen = savelen;
2252         READ_ENTER(&ipf_state);
2253         for (isp = &ips_table[hv]; (is = *isp); isp = &is->is_hnext) {
2254                 /*
2255                  * Only allow this icmp though if the
2256                  * encapsulated packet was allowed through the
2257                  * other way around. Note that the minimal amount
2258                  * of info present does not allow for checking against
2259                  * tcp internals such as seq and ack numbers.
2260                  */
2261                 if ((is->is_p == pr) && (is->is_v == 6) &&
2262                     fr_matchsrcdst(is, src, dst, &ofin, tcp)) {
2263                         fr = is->is_rule;
2264                         ips_stats.iss_hits++;
2265                         is->is_pkts++;
2266                         is->is_bytes += fin->fin_plen;
2267                         /*
2268                          * we deliberately do not touch the timeouts
2269                          * for the accompanying state table entry.
2270                          * It remains to be seen if that is correct. XXX
2271                          */
2272                         RWLOCK_EXIT(&ipf_state);
2273                         return fr;
2274                 }
2275         }
2276         RWLOCK_EXIT(&ipf_state);
2277         return NULL;
2278 }
2279 #endif