sys/contrib/ipfilter/netinet/ip_state.c

   1 /*      $FreeBSD$       */
   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__
  36 struct 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)
 115 static const char sccsid[] = "@(#)ip_state.c    1.8 6/5/96 (C) 1993-2000 Darren Reed";
 116 static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.186.2.80 2007/10/16 09:33:23 darrenr Exp $";
 117 #endif
 118
 119 static  ipstate_t **ips_table = NULL;
 120 static  u_long  *ips_seed = NULL;
 121 static  int     ips_num = 0;
 122 static  u_long ips_last_force_flush = 0;
 123 ips_stat_t ips_stats;
 124
 125 #ifdef  USE_INET6
 126 static ipstate_t *fr_checkicmp6matchingstate __P((fr_info_t *));
 127 #endif
 128 static ipstate_t *fr_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *,
 129                                       i6addr_t *, tcphdr_t *, u_32_t));
 130 static ipstate_t *fr_checkicmpmatchingstate __P((fr_info_t *));
 131 static int fr_state_flush __P((int, int));
 132 static int fr_state_flush_entry __P((void *));
 133 static ips_stat_t *fr_statetstats __P((void));
 134 static int fr_delstate __P((ipstate_t *, int));
 135 static int fr_state_remove __P((caddr_t));
 136 static void fr_ipsmove __P((ipstate_t *, u_int));
 137 static int fr_tcpstate __P((fr_info_t *, tcphdr_t *, ipstate_t *));
 138 static int fr_tcpoptions __P((fr_info_t *, tcphdr_t *, tcpdata_t *));
 139 static ipstate_t *fr_stclone __P((fr_info_t *, tcphdr_t *, ipstate_t *));
 140 static void fr_fixinisn __P((fr_info_t *, ipstate_t *));
 141 static void fr_fixoutisn __P((fr_info_t *, ipstate_t *));
 142 static void fr_checknewisn __P((fr_info_t *, ipstate_t *));
 143 static int fr_stateiter __P((ipftoken_t *, ipfgeniter_t *));
 144 static int fr_stgettable __P((char *));
 145
 146 int fr_stputent __P((caddr_t));
 147 int 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
 153 u_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);
 165 int     fr_statemax = IPSTATE_MAX,
 166         fr_statesize = IPSTATE_SIZE;
 167 int     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;
 172 ipftq_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
 181 int     ipstate_logging = 1;
 182 #else
 183 int     ipstate_logging = 0;
 184 #endif
 185 ipstate_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 /* ------------------------------------------------------------------------ */
 196 int fr_stateinit()
 197 {
 198 #if defined(NEED_LOCAL_RAND) || !defined(_KERNEL)
 199         struct timeval tv;
 200 #endif
 201         int i;
 202
 203         KMALLOCS(ips_table, ipstate_t **, fr_statesize * sizeof(ipstate_t *));
 204         if (ips_table == NULL)
 205                 return -1;
 206         bzero((char *)ips_table, fr_statesize * sizeof(ipstate_t *));
 207
 208         KMALLOCS(ips_seed, u_long *, fr_statesize * sizeof(*ips_seed));
 209         if (ips_seed == NULL)
 210                 return -2;
 211 #if defined(NEED_LOCAL_RAND) || !defined(_KERNEL)
 212         tv.tv_sec = 0;
 213         GETKTIME(&tv);
 214 #endif
 215         for (i = 0; i < fr_statesize; i++) {
 216                 /*
 217                  * XXX - ips_seed[X] should be a random number of sorts.
 218                  */
 219 #if !defined(NEED_LOCAL_RAND) && defined(_KERNEL)
 220                 ips_seed[i] = arc4random();
 221 #else
 222                 ips_seed[i] = ((u_long)ips_seed + i) * fr_statesize;
 223                 ips_seed[i] += tv.tv_sec;
 224                 ips_seed[i] *= (u_long)ips_seed;
 225                 ips_seed[i] ^= 0x5a5aa5a5;
 226                 ips_seed[i] *= fr_statemax;
 227 #endif
 228         }
 229 #if defined(NEED_LOCAL_RAND) && defined(_KERNEL)
 230         ipf_rand_push(ips_seed, fr_statesize * sizeof(*ips_seed));
 231 #endif
 232
 233         /* fill icmp reply type table */
 234         for (i = 0; i <= ICMP_MAXTYPE; i++)
 235                 icmpreplytype4[i] = -1;
 236         icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY;
 237         icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY;
 238         icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY;
 239         icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY;
 240 #ifdef  USE_INET6
 241         /* fill icmp reply type table */
 242         for (i = 0; i <= ICMP6_MAXTYPE; i++)
 243                 icmpreplytype6[i] = -1;
 244         icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY;
 245         icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT;
 246         icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY;
 247         icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT;
 248         icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT;
 249 #endif
 250
 251         KMALLOCS(ips_stats.iss_bucketlen, u_long *,
 252                  fr_statesize * sizeof(u_long));
 253         if (ips_stats.iss_bucketlen == NULL)
 254                 return -1;
 255         bzero((char *)ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long));
 256
 257         if (fr_state_maxbucket == 0) {
 258                 for (i = fr_statesize; i > 0; i >>= 1)
 259                         fr_state_maxbucket++;
 260                 fr_state_maxbucket *= 2;
 261         }
 262
 263         ips_stats.iss_tcptab = ips_tqtqb;
 264         fr_sttab_init(ips_tqtqb);
 265         ips_tqtqb[IPF_TCP_NSTATES - 1].ifq_next = &ips_udptq;
 266         ips_udptq.ifq_ttl = (u_long)fr_udptimeout;
 267         ips_udptq.ifq_ref = 1;
 268         ips_udptq.ifq_head = NULL;
 269         ips_udptq.ifq_tail = &ips_udptq.ifq_head;
 270         MUTEX_INIT(&ips_udptq.ifq_lock, "ipftq udp tab");
 271         ips_udptq.ifq_next = &ips_udpacktq;
 272         ips_udpacktq.ifq_ttl = (u_long)fr_udpacktimeout;
 273         ips_udpacktq.ifq_ref = 1;
 274         ips_udpacktq.ifq_head = NULL;
 275         ips_udpacktq.ifq_tail = &ips_udpacktq.ifq_head;
 276         MUTEX_INIT(&ips_udpacktq.ifq_lock, "ipftq udpack tab");
 277         ips_udpacktq.ifq_next = &ips_icmptq;
 278         ips_icmptq.ifq_ttl = (u_long)fr_icmptimeout;
 279         ips_icmptq.ifq_ref = 1;
 280         ips_icmptq.ifq_head = NULL;
 281         ips_icmptq.ifq_tail = &ips_icmptq.ifq_head;
 282         MUTEX_INIT(&ips_icmptq.ifq_lock, "ipftq icmp tab");
 283         ips_icmptq.ifq_next = &ips_icmpacktq;
 284         ips_icmpacktq.ifq_ttl = (u_long)fr_icmpacktimeout;
 285         ips_icmpacktq.ifq_ref = 1;
 286         ips_icmpacktq.ifq_head = NULL;
 287         ips_icmpacktq.ifq_tail = &ips_icmpacktq.ifq_head;
 288         MUTEX_INIT(&ips_icmpacktq.ifq_lock, "ipftq icmpack tab");
 289         ips_icmpacktq.ifq_next = &ips_iptq;
 290         ips_iptq.ifq_ttl = (u_long)fr_iptimeout;
 291         ips_iptq.ifq_ref = 1;
 292         ips_iptq.ifq_head = NULL;
 293         ips_iptq.ifq_tail = &ips_iptq.ifq_head;
 294         MUTEX_INIT(&ips_iptq.ifq_lock, "ipftq ip tab");
 295         ips_iptq.ifq_next = &ips_deletetq;
 296         ips_deletetq.ifq_ttl = (u_long)1;
 297         ips_deletetq.ifq_ref = 1;
 298         ips_deletetq.ifq_head = NULL;
 299         ips_deletetq.ifq_tail = &ips_deletetq.ifq_head;
 300         MUTEX_INIT(&ips_deletetq.ifq_lock, "state delete queue");
 301         ips_deletetq.ifq_next = NULL;
 302
 303         RWLOCK_INIT(&ipf_state, "ipf IP state rwlock");
 304         MUTEX_INIT(&ipf_stinsert, "ipf state insert mutex");
 305         fr_state_init = 1;
 306
 307         ips_last_force_flush = fr_ticks;
 308         return 0;
 309 }
 310
 311
 312 /* ------------------------------------------------------------------------ */
 313 /* Function:    fr_stateunload                                              */
 314 /* Returns:     Nil                                                         */
 315 /* Parameters:  Nil                                                         */
 316 /*                                                                          */
 317 /* Release and destroy any resources acquired or initialised so that        */
 318 /* IPFilter can be unloaded or re-initialised.                              */
 319 /* ------------------------------------------------------------------------ */
 320 void fr_stateunload()
 321 {
 322         ipftq_t *ifq, *ifqnext;
 323         ipstate_t *is;
 324
 325         while ((is = ips_list) != NULL)
 326                 fr_delstate(is, ISL_UNLOAD);
 327
 328         /*
 329          * Proxy timeout queues are not cleaned here because although they
 330          * exist on the state list, appr_unload is called after fr_stateunload
 331          * and the proxies actually are responsible for them being created.
 332          * Should the proxy timeouts have their own list?  There's no real
 333          * justification as this is the only complicationA
 334          */
 335         for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) {
 336                 ifqnext = ifq->ifq_next;
 337                 if (((ifq->ifq_flags & IFQF_PROXY) == 0) &&
 338                     (fr_deletetimeoutqueue(ifq) == 0))
 339                         fr_freetimeoutqueue(ifq);
 340         }
 341
 342         ips_stats.iss_inuse = 0;
 343         ips_num = 0;
 344
 345         if (fr_state_init == 1) {
 346                 fr_sttab_destroy(ips_tqtqb);
 347                 MUTEX_DESTROY(&ips_udptq.ifq_lock);
 348                 MUTEX_DESTROY(&ips_icmptq.ifq_lock);
 349                 MUTEX_DESTROY(&ips_udpacktq.ifq_lock);
 350                 MUTEX_DESTROY(&ips_icmpacktq.ifq_lock);
 351                 MUTEX_DESTROY(&ips_iptq.ifq_lock);
 352                 MUTEX_DESTROY(&ips_deletetq.ifq_lock);
 353         }
 354
 355         if (ips_table != NULL) {
 356                 KFREES(ips_table, fr_statesize * sizeof(*ips_table));
 357                 ips_table = NULL;
 358         }
 359
 360         if (ips_seed != NULL) {
 361                 KFREES(ips_seed, fr_statesize * sizeof(*ips_seed));
 362                 ips_seed = NULL;
 363         }
 364
 365         if (ips_stats.iss_bucketlen != NULL) {
 366                 KFREES(ips_stats.iss_bucketlen, fr_statesize * sizeof(u_long));
 367                 ips_stats.iss_bucketlen = NULL;
 368         }
 369
 370         if (fr_state_maxbucket_reset == 1)
 371                 fr_state_maxbucket = 0;
 372
 373         if (fr_state_init == 1) {
 374                 fr_state_init = 0;
 375                 RW_DESTROY(&ipf_state);
 376                 MUTEX_DESTROY(&ipf_stinsert);
 377         }
 378 }
 379
 380
 381 /* ------------------------------------------------------------------------ */
 382 /* Function:    fr_statetstats                                              */
 383 /* Returns:     ips_state_t* - pointer to state stats structure             */
 384 /* Parameters:  Nil                                                         */
 385 /*                                                                          */
 386 /* Put all the current numbers and pointers into a single struct and return */
 387 /* a pointer to it.                                                         */
 388 /* ------------------------------------------------------------------------ */
 389 static ips_stat_t *fr_statetstats()
 390 {
 391         ips_stats.iss_active = ips_num;
 392         ips_stats.iss_statesize = fr_statesize;
 393         ips_stats.iss_statemax = fr_statemax;
 394         ips_stats.iss_table = ips_table;
 395         ips_stats.iss_list = ips_list;
 396         ips_stats.iss_ticks = fr_ticks;
 397         return &ips_stats;
 398 }
 399
 400 /* ------------------------------------------------------------------------ */
 401 /* Function:    fr_state_remove                                             */
 402 /* Returns:     int - 0 == success, != 0 == failure                         */
 403 /* Parameters:  data(I) - pointer to state structure to delete from table   */
 404 /*                                                                          */
 405 /* Search for a state structure that matches the one passed, according to   */
 406 /* the IP addresses and other protocol specific information.                */
 407 /* ------------------------------------------------------------------------ */
 408 static int fr_state_remove(data)
 409 caddr_t data;
 410 {
 411         ipstate_t *sp, st;
 412         int error;
 413
 414         sp = &st;
 415         error = fr_inobj(data, &st, IPFOBJ_IPSTATE);
 416         if (error)
 417                 return EFAULT;
 418
 419         WRITE_ENTER(&ipf_state);
 420         for (sp = ips_list; sp; sp = sp->is_next)
 421                 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) &&
 422                     !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src,
 423                           sizeof(st.is_src)) &&
 424                     !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_src,
 425                           sizeof(st.is_dst)) &&
 426                     !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps,
 427                           sizeof(st.is_ps))) {
 428                         fr_delstate(sp, ISL_REMOVE);
 429                         RWLOCK_EXIT(&ipf_state);
 430                         return 0;
 431                 }
 432         RWLOCK_EXIT(&ipf_state);
 433         return ESRCH;
 434 }
 435
 436
 437 /* ------------------------------------------------------------------------ */
 438 /* Function:    fr_state_ioctl                                              */
 439 /* Returns:     int - 0 == success, != 0 == failure                         */
 440 /* Parameters:  data(I) - pointer to ioctl data                             */
 441 /*              cmd(I)  - ioctl command integer                             */
 442 /*              mode(I) - file mode bits used with open                     */
 443 /*                                                                          */
 444 /* Processes an ioctl call made to operate on the IP Filter state device.   */
 445 /* ------------------------------------------------------------------------ */
 446 int fr_state_ioctl(data, cmd, mode, uid, ctx)
 447 caddr_t data;
 448 ioctlcmd_t cmd;
 449 int mode, uid;
 450 void *ctx;
 451 {
 452         int arg, ret, error = 0;
 453         SPL_INT(s);
 454
 455         switch (cmd)
 456         {
 457         /*
 458          * Delete an entry from the state table.
 459          */
 460         case SIOCDELST :
 461                 error = fr_state_remove(data);
 462                 break;
 463
 464         /*
 465          * Flush the state table
 466          */
 467         case SIOCIPFFL :
 468                 error = BCOPYIN(data, (char *)&arg, sizeof(arg));
 469                 if (error != 0) {
 470                         error = EFAULT;
 471                 } else {
 472                         WRITE_ENTER(&ipf_state);
 473                         ret = fr_state_flush(arg, 4);
 474                         RWLOCK_EXIT(&ipf_state);
 475                         error = BCOPYOUT((char *)&ret, data, sizeof(ret));
 476                         if (error != 0)
 477                                 error = EFAULT;
 478                 }
 479                 break;
 480
 481 #ifdef  USE_INET6
 482         case SIOCIPFL6 :
 483                 error = BCOPYIN(data, (char *)&arg, sizeof(arg));
 484                 if (error != 0) {
 485                         error = EFAULT;
 486                 } else {
 487                         WRITE_ENTER(&ipf_state);
 488                         ret = fr_state_flush(arg, 6);
 489                         RWLOCK_EXIT(&ipf_state);
 490                         error = BCOPYOUT((char *)&ret, data, sizeof(ret));
 491                         if (error != 0)
 492                                 error = EFAULT;
 493                 }
 494                 break;
 495 #endif
 496 #ifdef  IPFILTER_LOG
 497         /*
 498          * Flush the state log.
 499          */
 500         case SIOCIPFFB :
 501                 if (!(mode & FWRITE))
 502                         error = EPERM;
 503                 else {
 504                         int tmp;
 505
 506                         tmp = ipflog_clear(IPL_LOGSTATE);
 507                         error = BCOPYOUT((char *)&tmp, data, sizeof(tmp));
 508                         if (error != 0)
 509                                 error = EFAULT;
 510                 }
 511                 break;
 512
 513         /*
 514          * Turn logging of state information on/off.
 515          */
 516         case SIOCSETLG :
 517                 if (!(mode & FWRITE))
 518                         error = EPERM;
 519                 else {
 520                         error = BCOPYIN((char *)data, (char *)&ipstate_logging,
 521                                         sizeof(ipstate_logging));
 522                         if (error != 0)
 523                                 error = EFAULT;
 524                 }
 525                 break;
 526
 527         /*
 528          * Return the current state of logging.
 529          */
 530         case SIOCGETLG :
 531                 error = BCOPYOUT((char *)&ipstate_logging, (char *)data,
 532                                  sizeof(ipstate_logging));
 533                 if (error != 0)
 534                         error = EFAULT;
 535                 break;
 536
 537         /*
 538          * Return the number of bytes currently waiting to be read.
 539          */
 540         case FIONREAD :
 541                 arg = iplused[IPL_LOGSTATE];    /* returned in an int */
 542                 error = BCOPYOUT((char *)&arg, data, sizeof(arg));
 543                 if (error != 0)
 544                         error = EFAULT;
 545                 break;
 546 #endif
 547
 548         /*
 549          * Get the current state statistics.
 550          */
 551         case SIOCGETFS :
 552                 error = fr_outobj(data, fr_statetstats(), IPFOBJ_STATESTAT);
 553                 break;
 554
 555         /*
 556          * Lock/Unlock the state table.  (Locking prevents any changes, which
 557          * means no packets match).
 558          */
 559         case SIOCSTLCK :
 560                 if (!(mode & FWRITE)) {
 561                         error = EPERM;
 562                 } else {
 563                         error = fr_lock(data, &fr_state_lock);
 564                 }
 565                 break;
 566
 567         /*
 568          * Add an entry to the current state table.
 569          */
 570         case SIOCSTPUT :
 571                 if (!fr_state_lock || !(mode &FWRITE)) {
 572                         error = EACCES;
 573                         break;
 574                 }
 575                 error = fr_stputent(data);
 576                 break;
 577
 578         /*
 579          * Get a state table entry.
 580          */
 581         case SIOCSTGET :
 582                 if (!fr_state_lock) {
 583                         error = EACCES;
 584                         break;
 585                 }
 586                 error = fr_stgetent(data);
 587                 break;
 588
 589         /*
 590          * Return a copy of the hash table bucket lengths
 591          */
 592         case SIOCSTAT1 :
 593                 error = BCOPYOUT(ips_stats.iss_bucketlen, data,
 594                                  fr_statesize * sizeof(u_long));
 595                 if (error != 0)
 596                         error = EFAULT;
 597                 break;
 598
 599         case SIOCGENITER :
 600             {
 601                 ipftoken_t *token;
 602                 ipfgeniter_t iter;
 603
 604                 error = fr_inobj(data, &iter, IPFOBJ_GENITER);
 605                 if (error != 0)
 606                         break;
 607
 608                 SPL_SCHED(s);
 609                 token = ipf_findtoken(IPFGENITER_STATE, uid, ctx);
 610                 if (token != NULL)
 611                         error = fr_stateiter(token, &iter);
 612                 else
 613                         error = ESRCH;
 614                 RWLOCK_EXIT(&ipf_tokens);
 615                 SPL_X(s);
 616                 break;
 617             }
 618
 619         case SIOCGTABL :
 620                 error = fr_stgettable(data);
 621                 break;
 622
 623         case SIOCIPFDELTOK :
 624                 error = BCOPYIN(data, (char *)&arg, sizeof(arg));
 625                 if (error != 0) {
 626                         error = EFAULT;
 627                 } else {
 628                         SPL_SCHED(s);
 629                         error = ipf_deltoken(arg, uid, ctx);
 630                         SPL_X(s);
 631                 }
 632                 break;
 633
 634         case SIOCGTQTAB :
 635                 error = fr_outobj(data, ips_tqtqb, IPFOBJ_STATETQTAB);
 636                 break;
 637
 638         default :
 639                 error = EINVAL;
 640                 break;
 641         }
 642         return error;
 643 }
 644
 645
 646 /* ------------------------------------------------------------------------ */
 647 /* Function:    fr_stgetent                                                 */
 648 /* Returns:     int - 0 == success, != 0 == failure                         */
 649 /* Parameters:  data(I) - pointer to state structure to retrieve from table */
 650 /*                                                                          */
 651 /* Copy out state information from the kernel to a user space process.  If  */
 652 /* there is a filter rule associated with the state entry, copy that out    */
 653 /* as well.  The entry to copy out is taken from the value of "ips_next" in */
 654 /* the struct passed in and if not null and not found in the list of current*/
 655 /* state entries, the retrieval fails.                                      */
 656 /* ------------------------------------------------------------------------ */
 657 int fr_stgetent(data)
 658 caddr_t data;
 659 {
 660         ipstate_t *is, *isn;
 661         ipstate_save_t ips;
 662         int error;
 663
 664         error = fr_inobj(data, &ips, IPFOBJ_STATESAVE);
 665         if (error != 0)
 666                 return error;
 667
 668         READ_ENTER(&ipf_state);
 669         isn = ips.ips_next;
 670         if (isn == NULL) {
 671                 isn = ips_list;
 672                 if (isn == NULL) {
 673                         RWLOCK_EXIT(&ipf_state);
 674                         if (ips.ips_next == NULL)
 675                                 return ENOENT;
 676                         return 0;
 677                 }
 678         } else {
 679                 /*
 680                  * Make sure the pointer we're copying from exists in the
 681                  * current list of entries.  Security precaution to prevent
 682                  * copying of random kernel data.
 683                  */
 684                 for (is = ips_list; is; is = is->is_next)
 685                         if (is == isn)
 686                                 break;
 687                 if (is == NULL) {
 688                         RWLOCK_EXIT(&ipf_state);
 689                         return ESRCH;
 690                 }
 691         }
 692         ips.ips_next = isn->is_next;
 693         bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is));
 694         ips.ips_rule = isn->is_rule;
 695         if (isn->is_rule != NULL)
 696                 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr,
 697                       sizeof(ips.ips_fr));
 698         RWLOCK_EXIT(&ipf_state);
 699         error = fr_outobj(data, &ips, IPFOBJ_STATESAVE);
 700         return error;
 701 }
 702
 703
 704 /* ------------------------------------------------------------------------ */
 705 /* Function:    fr_stputent                                                 */
 706 /* Returns:     int - 0 == success, != 0 == failure                         */
 707 /* Parameters:  data(I) - pointer to state information struct               */
 708 /*                                                                          */
 709 /* This function implements the SIOCSTPUT ioctl: insert a state entry into  */
 710 /* the state table.  If the state info. includes a pointer to a filter rule */
 711 /* then also add in an orphaned rule (will not show up in any "ipfstat -io" */
 712 /* output.                                                                  */
 713 /* ------------------------------------------------------------------------ */
 714 int fr_stputent(data)
 715 caddr_t data;
 716 {
 717         ipstate_t *is, *isn;
 718         ipstate_save_t ips;
 719         int error, out, i;
 720         frentry_t *fr;
 721         char *name;
 722
 723         error = fr_inobj(data, &ips, IPFOBJ_STATESAVE);
 724         if (error)
 725                 return EFAULT;
 726
 727         KMALLOC(isn, ipstate_t *);
 728         if (isn == NULL)
 729                 return ENOMEM;
 730
 731         bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn));
 732         bzero((char *)isn, offsetof(struct ipstate, is_pkts));
 733         isn->is_sti.tqe_pnext = NULL;
 734         isn->is_sti.tqe_next = NULL;
 735         isn->is_sti.tqe_ifq = NULL;
 736         isn->is_sti.tqe_parent = isn;
 737         isn->is_ifp[0] = NULL;
 738         isn->is_ifp[1] = NULL;
 739         isn->is_ifp[2] = NULL;
 740         isn->is_ifp[3] = NULL;
 741         isn->is_sync = NULL;
 742         fr = ips.ips_rule;
 743
 744         if (fr == NULL) {
 745                 READ_ENTER(&ipf_state);
 746                 fr_stinsert(isn, 0);
 747                 MUTEX_EXIT(&isn->is_lock);
 748                 RWLOCK_EXIT(&ipf_state);
 749                 return 0;
 750         }
 751
 752         if (isn->is_flags & SI_NEWFR) {
 753                 KMALLOC(fr, frentry_t *);
 754                 if (fr == NULL) {
 755                         KFREE(isn);
 756                         return ENOMEM;
 757                 }
 758                 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr));
 759                 out = fr->fr_flags & FR_OUTQUE ? 1 : 0;
 760                 isn->is_rule = fr;
 761                 ips.ips_is.is_rule = fr;
 762                 MUTEX_NUKE(&fr->fr_lock);
 763                 MUTEX_INIT(&fr->fr_lock, "state filter rule lock");
 764
 765                 /*
 766                  * Look up all the interface names in the rule.
 767                  */
 768                 for (i = 0; i < 4; i++) {
 769                         name = fr->fr_ifnames[i];
 770                         fr->fr_ifas[i] = fr_resolvenic(name, fr->fr_v);
 771                         name = isn->is_ifname[i];
 772                         isn->is_ifp[i] = fr_resolvenic(name, isn->is_v);
 773                 }
 774
 775                 fr->fr_ref = 0;
 776                 fr->fr_dsize = 0;
 777                 fr->fr_data = NULL;
 778                 fr->fr_type = FR_T_NONE;
 779
 780                 fr_resolvedest(&fr->fr_tifs[0], fr->fr_v);
 781                 fr_resolvedest(&fr->fr_tifs[1], fr->fr_v);
 782                 fr_resolvedest(&fr->fr_dif, fr->fr_v);
 783
 784                 /*
 785                  * send a copy back to userland of what we ended up
 786                  * to allow for verification.
 787                  */
 788                 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE);
 789                 if (error) {
 790                         KFREE(isn);
 791                         MUTEX_DESTROY(&fr->fr_lock);
 792                         KFREE(fr);
 793                         return EFAULT;
 794                 }
 795                 READ_ENTER(&ipf_state);
 796                 fr_stinsert(isn, 0);
 797                 MUTEX_EXIT(&isn->is_lock);
 798                 RWLOCK_EXIT(&ipf_state);
 799
 800         } else {
 801                 READ_ENTER(&ipf_state);
 802                 for (is = ips_list; is; is = is->is_next)
 803                         if (is->is_rule == fr) {
 804                                 fr_stinsert(isn, 0);
 805                                 MUTEX_EXIT(&isn->is_lock);
 806                                 break;
 807                         }
 808
 809                 if (is == NULL) {
 810                         KFREE(isn);
 811                         isn = NULL;
 812                 }
 813                 RWLOCK_EXIT(&ipf_state);
 814
 815                 return (isn == NULL) ? ESRCH : 0;
 816         }
 817
 818         return 0;
 819 }
 820
 821
 822 /* ------------------------------------------------------------------------ */
 823 /* Function:   fr_stinsert                                                  */
 824 /* Returns:    Nil                                                          */
 825 /* Parameters: is(I)  - pointer to state structure                          */
 826 /*             rev(I) - flag indicating forward/reverse direction of packet */
 827 /*                                                                          */
 828 /* Inserts a state structure into the hash table (for lookups) and the list */
 829 /* of state entries (for enumeration).  Resolves all of the interface names */
 830 /* to pointers and adjusts running stats for the hash table as appropriate. */
 831 /*                                                                          */
 832 /* Locking: it is assumed that some kind of lock on ipf_state is held.      */
 833 /*          Exits with is_lock initialised and held.                        */
 834 /* ------------------------------------------------------------------------ */
 835 void fr_stinsert(is, rev)
 836 ipstate_t *is;
 837 int rev;
 838 {
 839         frentry_t *fr;
 840         u_int hv;
 841         int i;
 842
 843         MUTEX_INIT(&is->is_lock, "ipf state entry");
 844
 845         fr = is->is_rule;
 846         if (fr != NULL) {
 847                 MUTEX_ENTER(&fr->fr_lock);
 848                 fr->fr_ref++;
 849                 fr->fr_statecnt++;
 850                 MUTEX_EXIT(&fr->fr_lock);
 851         }
 852
 853         /*
 854          * Look up all the interface names in the state entry.
 855          */
 856         for (i = 0; i < 4; i++) {
 857                 if (is->is_ifp[i] != NULL)
 858                         continue;
 859                 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], is->is_v);
 860         }
 861
 862         /*
 863          * If we could trust is_hv, then the modulous would not be needed, but
 864          * when running with IPFILTER_SYNC, this stops bad values.
 865          */
 866         hv = is->is_hv % fr_statesize;
 867         is->is_hv = hv;
 868
 869         /*
 870          * We need to get both of these locks...the first because it is
 871          * possible that once the insert is complete another packet might
 872          * come along, match the entry and want to update it.
 873          */
 874         MUTEX_ENTER(&is->is_lock);
 875         MUTEX_ENTER(&ipf_stinsert);
 876
 877         /*
 878          * add into list table.
 879          */
 880         if (ips_list != NULL)
 881                 ips_list->is_pnext = &is->is_next;
 882         is->is_pnext = &ips_list;
 883         is->is_next = ips_list;
 884         ips_list = is;
 885
 886         if (ips_table[hv] != NULL)
 887                 ips_table[hv]->is_phnext = &is->is_hnext;
 888         else
 889                 ips_stats.iss_inuse++;
 890         is->is_phnext = ips_table + hv;
 891         is->is_hnext = ips_table[hv];
 892         ips_table[hv] = is;
 893         ips_stats.iss_bucketlen[hv]++;
 894         ips_num++;
 895         MUTEX_EXIT(&ipf_stinsert);
 896
 897         fr_setstatequeue(is, rev);
 898 }
 899
 900
 901 /* ------------------------------------------------------------------------ */
 902 /* Function:    fr_addstate                                                 */
 903 /* Returns:     ipstate_t* - NULL == failure, else pointer to new state     */
 904 /* Parameters:  fin(I)    - pointer to packet information                   */
 905 /*              stsave(O) - pointer to place to save pointer to created     */
 906 /*                          state structure.                                */
 907 /*              flags(I)  - flags to use when creating the structure        */
 908 /*                                                                          */
 909 /* Creates a new IP state structure from the packet information collected.  */
 910 /* Inserts it into the state table and appends to the bottom of the active  */
 911 /* list.  If the capacity of the table has reached the maximum allowed then */
 912 /* the call will fail and a flush is scheduled for the next timeout call.   */
 913 /*                                                                          */
 914 /* NOTE: The use of stsave to point to nat_state will result in memory      */
 915 /*       corruption.  It should only be used to point to objects that will  */
 916 /*       either outlive this (not expired) or will deref the ip_state_t     */
 917 /*       when they are deleted.                                             */
 918 /* ------------------------------------------------------------------------ */
 919 ipstate_t *fr_addstate(fin, stsave, flags)
 920 fr_info_t *fin;
 921 ipstate_t **stsave;
 922 u_int flags;
 923 {
 924         ipstate_t *is, ips;
 925         struct icmp *ic;
 926         u_int pass, hv;
 927         frentry_t *fr;
 928         tcphdr_t *tcp;
 929         grehdr_t *gre;
 930         int out;
 931
 932         if (fr_state_lock ||
 933             (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)))
 934                 return NULL;
 935
 936         if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN))
 937                 return NULL;
 938
 939         /*
 940          * If a "keep state" rule has reached the maximum number of references
 941          * to it, then schedule an automatic flush in case we can clear out
 942          * some "dead old wood".  Note that because the lock isn't held on
 943          * fr it is possible that we could overflow.  The cost of overflowing
 944          * is being ignored here as the number by which it can overflow is
 945          * a product of the number of simultaneous threads that could be
 946          * executing in here, so a limit of 100 won't result in 200, but could
 947          * result in 101 or 102.
 948          */
 949         fr = fin->fin_fr;
 950         if (fr != NULL) {
 951                 if ((ips_num >= fr_statemax) && (fr->fr_statemax == 0)) {
 952                         ATOMIC_INCL(ips_stats.iss_max);
 953                         fr_state_doflush = 1;
 954                         return NULL;
 955                 }
 956                 if ((fr->fr_statemax != 0) &&
 957                     (fr->fr_statecnt >= fr->fr_statemax)) {
 958                         ATOMIC_INCL(ips_stats.iss_maxref);
 959                         return NULL;
 960                 }
 961         }
 962
 963         pass = (fr == NULL) ? 0 : fr->fr_flags;
 964
 965         ic = NULL;
 966         tcp = NULL;
 967         out = fin->fin_out;
 968         is = &ips;
 969         bzero((char *)is, sizeof(*is));
 970         is->is_die = 1 + fr_ticks;
 971
 972         /*
 973          * Copy and calculate...
 974          */
 975         hv = (is->is_p = fin->fin_fi.fi_p);
 976         is->is_src = fin->fin_fi.fi_src;
 977         hv += is->is_saddr;
 978         is->is_dst = fin->fin_fi.fi_dst;
 979         hv += is->is_daddr;
 980 #ifdef  USE_INET6
 981         if (fin->fin_v == 6) {
 982                 /*
 983                  * For ICMPv6, we check to see if the destination address is
 984                  * a multicast address.  If it is, do not include it in the
 985                  * calculation of the hash because the correct reply will come
 986                  * back from a real address, not a multicast address.
 987                  */
 988                 if ((is->is_p == IPPROTO_ICMPV6) &&
 989                     IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) {
 990                         /*
 991                          * So you can do keep state with neighbour discovery.
 992                          *
 993                          * Here we could use the address from the neighbour
 994                          * solicit message to put in the state structure and
 995                          * we could use that without a wildcard flag too...
 996                          */
 997                         flags |= SI_W_DADDR;
 998                         hv -= is->is_daddr;
 999                 } else {
1000                         hv += is->is_dst.i6[1];
1001                         hv += is->is_dst.i6[2];
1002                         hv += is->is_dst.i6[3];
1003                 }
1004                 hv += is->is_src.i6[1];
1005                 hv += is->is_src.i6[2];
1006                 hv += is->is_src.i6[3];
1007         }
1008 #endif
1009         if ((fin->fin_v == 4) &&
1010             (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
1011                 if (fin->fin_out == 0) {
1012                         flags |= SI_W_DADDR|SI_CLONE;
1013                         hv -= is->is_daddr;
1014                 } else {
1015                         flags |= SI_W_SADDR|SI_CLONE;
1016                         hv -= is->is_saddr;
1017                 }
1018         }
1019
1020         switch (is->is_p)
1021         {
1022 #ifdef  USE_INET6
1023         case IPPROTO_ICMPV6 :
1024                 ic = fin->fin_dp;
1025
1026                 switch (ic->icmp_type)
1027                 {
1028                 case ICMP6_ECHO_REQUEST :
1029                         is->is_icmp.ici_type = ic->icmp_type;
1030                         hv += (is->is_icmp.ici_id = ic->icmp_id);
1031                         break;
1032                 case ICMP6_MEMBERSHIP_QUERY :
1033                 case ND_ROUTER_SOLICIT :
1034                 case ND_NEIGHBOR_SOLICIT :
1035                 case ICMP6_NI_QUERY :
1036                         is->is_icmp.ici_type = ic->icmp_type;
1037                         break;
1038                 default :
1039                         return NULL;
1040                 }
1041                 ATOMIC_INCL(ips_stats.iss_icmp);
1042                 break;
1043 #endif
1044         case IPPROTO_ICMP :
1045                 ic = fin->fin_dp;
1046
1047                 switch (ic->icmp_type)
1048                 {
1049                 case ICMP_ECHO :
1050                 case ICMP_TSTAMP :
1051                 case ICMP_IREQ :
1052                 case ICMP_MASKREQ :
1053                         is->is_icmp.ici_type = ic->icmp_type;
1054                         hv += (is->is_icmp.ici_id = ic->icmp_id);
1055                         break;
1056                 default :
1057                         return NULL;
1058                 }
1059                 ATOMIC_INCL(ips_stats.iss_icmp);
1060                 break;
1061
1062         case IPPROTO_GRE :
1063                 gre = fin->fin_dp;
1064
1065                 is->is_gre.gs_flags = gre->gr_flags;
1066                 is->is_gre.gs_ptype = gre->gr_ptype;
1067                 if (GRE_REV(is->is_gre.gs_flags) == 1) {
1068                         is->is_call[0] = fin->fin_data[0];
1069                         is->is_call[1] = fin->fin_data[1];
1070                 }
1071                 break;
1072
1073         case IPPROTO_TCP :
1074                 tcp = fin->fin_dp;
1075
1076                 if (tcp->th_flags & TH_RST)
1077                         return NULL;
1078                 /*
1079                  * The endian of the ports doesn't matter, but the ack and
1080                  * sequence numbers do as we do mathematics on them later.
1081                  */
1082                 is->is_sport = htons(fin->fin_data[0]);
1083                 is->is_dport = htons(fin->fin_data[1]);
1084                 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1085                         hv += is->is_sport;
1086                         hv += is->is_dport;
1087                 }
1088
1089                 /*
1090                  * If this is a real packet then initialise fields in the
1091                  * state information structure from the TCP header information.
1092                  */
1093
1094                 is->is_maxdwin = 1;
1095                 is->is_maxswin = ntohs(tcp->th_win);
1096                 if (is->is_maxswin == 0)
1097                         is->is_maxswin = 1;
1098
1099                 if ((fin->fin_flx & FI_IGNORE) == 0) {
1100                         is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen -
1101                                       (TCP_OFF(tcp) << 2) +
1102                                       ((tcp->th_flags & TH_SYN) ? 1 : 0) +
1103                                       ((tcp->th_flags & TH_FIN) ? 1 : 0);
1104                         is->is_maxsend = is->is_send;
1105
1106                         /*
1107                          * Window scale option is only present in
1108                          * SYN/SYN-ACK packet.
1109                          */
1110                         if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) ==
1111                             TH_SYN &&
1112                             (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
1113                                 if (fr_tcpoptions(fin, tcp,
1114                                               &is->is_tcp.ts_data[0]) == -1) {
1115                                         fin->fin_flx |= FI_BAD;
1116                                 }
1117                         }
1118
1119                         if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) {
1120                                 fr_checknewisn(fin, is);
1121                                 fr_fixoutisn(fin, is);
1122                         }
1123
1124                         if ((tcp->th_flags & TH_OPENING) == TH_SYN)
1125                                 flags |= IS_TCPFSM;
1126                         else {
1127                                 is->is_maxdwin = is->is_maxswin * 2;
1128                                 is->is_dend = ntohl(tcp->th_ack);
1129                                 is->is_maxdend = ntohl(tcp->th_ack);
1130                                 is->is_maxdwin *= 2;
1131                         }
1132                 }
1133
1134                 /*
1135                  * If we're creating state for a starting connection, start the
1136                  * timer on it as we'll never see an error if it fails to
1137                  * connect.
1138                  */
1139                 ATOMIC_INCL(ips_stats.iss_tcp);
1140                 break;
1141
1142         case IPPROTO_UDP :
1143                 tcp = fin->fin_dp;
1144
1145                 is->is_sport = htons(fin->fin_data[0]);
1146                 is->is_dport = htons(fin->fin_data[1]);
1147                 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) {
1148                         hv += tcp->th_dport;
1149                         hv += tcp->th_sport;
1150                 }
1151                 ATOMIC_INCL(ips_stats.iss_udp);
1152                 break;
1153
1154         default :
1155                 break;
1156         }
1157         hv = DOUBLE_HASH(hv);
1158         is->is_hv = hv;
1159         is->is_rule = fr;
1160         is->is_flags = flags & IS_INHERITED;
1161
1162         /*
1163          * Look for identical state.
1164          */
1165         for (is = ips_table[is->is_hv % fr_statesize]; is != NULL;
1166              is = is->is_hnext) {
1167                 if (bcmp(&ips.is_src, &is->is_src,
1168                          offsetof(struct ipstate, is_ps) -
1169                          offsetof(struct ipstate, is_src)) == 0)
1170                         break;
1171         }
1172         if (is != NULL)
1173                 return NULL;
1174
1175         if (ips_stats.iss_bucketlen[hv] >= fr_state_maxbucket) {
1176                 ATOMIC_INCL(ips_stats.iss_bucketfull);
1177                 return NULL;
1178         }
1179         KMALLOC(is, ipstate_t *);
1180         if (is == NULL) {
1181                 ATOMIC_INCL(ips_stats.iss_nomem);
1182                 return NULL;
1183         }
1184         bcopy((char *)&ips, (char *)is, sizeof(*is));
1185         /*
1186          * Do not do the modulous here, it is done in fr_stinsert().
1187          */
1188         if (fr != NULL) {
1189                 (void) strncpy(is->is_group, fr->fr_group, FR_GROUPLEN);
1190                 if (fr->fr_age[0] != 0) {
1191                         is->is_tqehead[0] = fr_addtimeoutqueue(&ips_utqe,
1192                                                                fr->fr_age[0]);
1193                         is->is_sti.tqe_flags |= TQE_RULEBASED;
1194                 }
1195                 if (fr->fr_age[1] != 0) {
1196                         is->is_tqehead[1] = fr_addtimeoutqueue(&ips_utqe,
1197                                                                fr->fr_age[1]);
1198                         is->is_sti.tqe_flags |= TQE_RULEBASED;
1199                 }
1200
1201                 is->is_tag = fr->fr_logtag;
1202
1203                 /*
1204                  * The name '-' is special for network interfaces and causes
1205                  * a NULL name to be present, always, allowing packets to
1206                  * match it, regardless of their interface.
1207                  */
1208                 if ((fin->fin_ifp == NULL) ||
1209                     (fr->fr_ifnames[out << 1][0] == '-' &&
1210                      fr->fr_ifnames[out << 1][1] == '\0')) {
1211                         is->is_ifp[out << 1] = fr->fr_ifas[0];
1212                         strncpy(is->is_ifname[out << 1], fr->fr_ifnames[0],
1213                                 sizeof(fr->fr_ifnames[0]));
1214                 } else {
1215                         is->is_ifp[out << 1] = fin->fin_ifp;
1216                         COPYIFNAME(is->is_v, fin->fin_ifp,
1217                                    is->is_ifname[out << 1]);
1218                 }
1219
1220                 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1];
1221                 strncpy(is->is_ifname[(out << 1) + 1], fr->fr_ifnames[1],
1222                         sizeof(fr->fr_ifnames[1]));
1223
1224                 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2];
1225                 strncpy(is->is_ifname[((1 - out) << 1)], fr->fr_ifnames[2],
1226                         sizeof(fr->fr_ifnames[2]));
1227
1228                 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3];
1229                 strncpy(is->is_ifname[((1 - out) << 1) + 1], fr->fr_ifnames[3],
1230                         sizeof(fr->fr_ifnames[3]));
1231         } else {
1232                 pass = fr_flags;
1233                 is->is_tag = FR_NOLOGTAG;
1234
1235                 if (fin->fin_ifp != NULL) {
1236                         is->is_ifp[out << 1] = fin->fin_ifp;
1237                         COPYIFNAME(is->is_v, fin->fin_ifp,
1238                                    is->is_ifname[out << 1]);
1239                 }
1240         }
1241
1242         /*
1243          * It may seem strange to set is_ref to 2, but fr_check() will call
1244          * fr_statederef() after calling fr_addstate() and the idea is to
1245          * have it exist at the end of fr_check() with is_ref == 1.
1246          */
1247         is->is_ref = 2;
1248         is->is_pass = pass;
1249         is->is_pkts[0] = 0, is->is_bytes[0] = 0;
1250         is->is_pkts[1] = 0, is->is_bytes[1] = 0;
1251         is->is_pkts[2] = 0, is->is_bytes[2] = 0;
1252         is->is_pkts[3] = 0, is->is_bytes[3] = 0;
1253         if ((fin->fin_flx & FI_IGNORE) == 0) {
1254                 is->is_pkts[out] = 1;
1255                 is->is_bytes[out] = fin->fin_plen;
1256                 is->is_flx[out][0] = fin->fin_flx & FI_CMP;
1257                 is->is_flx[out][0] &= ~FI_OOW;
1258         }
1259
1260         if (pass & FR_STSTRICT)
1261                 is->is_flags |= IS_STRICT;
1262
1263         if (pass & FR_STATESYNC)
1264                 is->is_flags |= IS_STATESYNC;
1265
1266         /*
1267          * We want to check everything that is a property of this packet,
1268          * but we don't (automatically) care about it's fragment status as
1269          * this may change.
1270          */
1271         is->is_v = fin->fin_v;
1272         is->is_opt[0] = fin->fin_optmsk;
1273         is->is_optmsk[0] = 0xffffffff;
1274         is->is_optmsk[1] = 0xffffffff;
1275         if (is->is_v == 6) {
1276                 is->is_opt[0] &= ~0x8;
1277                 is->is_optmsk[0] &= ~0x8;
1278                 is->is_optmsk[1] &= ~0x8;
1279         }
1280         is->is_me = stsave;
1281         is->is_sec = fin->fin_secmsk;
1282         is->is_secmsk = 0xffff;
1283         is->is_auth = fin->fin_auth;
1284         is->is_authmsk = 0xffff;
1285         if (flags & (SI_WILDP|SI_WILDA)) {
1286                 ATOMIC_INCL(ips_stats.iss_wild);
1287         }
1288         is->is_rulen = fin->fin_rule;
1289
1290
1291         if (pass & FR_LOGFIRST)
1292                 is->is_pass &= ~(FR_LOGFIRST|FR_LOG);
1293
1294         READ_ENTER(&ipf_state);
1295
1296         fr_stinsert(is, fin->fin_rev);
1297
1298         if (fin->fin_p == IPPROTO_TCP) {
1299                 /*
1300                 * If we're creating state for a starting connection, start the
1301                 * timer on it as we'll never see an error if it fails to
1302                 * connect.
1303                 */
1304                 (void) fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags);
1305                 MUTEX_EXIT(&is->is_lock);
1306 #ifdef  IPFILTER_SCAN
1307                 if ((is->is_flags & SI_CLONE) == 0)
1308                         (void) ipsc_attachis(is);
1309 #endif
1310         } else {
1311                 MUTEX_EXIT(&is->is_lock);
1312         }
1313 #ifdef  IPFILTER_SYNC
1314         if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0))
1315                 is->is_sync = ipfsync_new(SMC_STATE, fin, is);
1316 #endif
1317         if (ipstate_logging)
1318                 ipstate_log(is, ISL_NEW);
1319
1320         RWLOCK_EXIT(&ipf_state);
1321         fin->fin_state = is;
1322         fin->fin_rev = IP6_NEQ(&is->is_dst, &fin->fin_daddr);
1323         fin->fin_flx |= FI_STATE;
1324         if (fin->fin_flx & FI_FRAG)
1325                 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE);
1326
1327         return is;
1328 }
1329
1330
1331 /* ------------------------------------------------------------------------ */
1332 /* Function:    fr_tcpoptions                                               */
1333 /* Returns:     int - 1 == packet matches state entry, 0 == it does not,    */
1334 /*                   -1 == packet has bad TCP options data                  */
1335 /* Parameters:  fin(I) - pointer to packet information                      */
1336 /*              tcp(I) - pointer to TCP packet header                       */
1337 /*              td(I)  - pointer to TCP data held as part of the state      */
1338 /*                                                                          */
1339 /* Look after the TCP header for any options and deal with those that are   */
1340 /* present.  Record details about those that we recogise.                   */
1341 /* ------------------------------------------------------------------------ */
1342 static int fr_tcpoptions(fin, tcp, td)
1343 fr_info_t *fin;
1344 tcphdr_t *tcp;
1345 tcpdata_t *td;
1346 {
1347         int off, mlen, ol, i, len, retval;
1348         char buf[64], *s, opt;
1349         mb_t *m = NULL;
1350
1351         len = (TCP_OFF(tcp) << 2);
1352         if (fin->fin_dlen < len)
1353                 return 0;
1354         len -= sizeof(*tcp);
1355
1356         off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff;
1357
1358         m = fin->fin_m;
1359         mlen = MSGDSIZE(m) - off;
1360         if (len > mlen) {
1361                 len = mlen;
1362                 retval = 0;
1363         } else {
1364                 retval = 1;
1365         }
1366
1367         COPYDATA(m, off, len, buf);
1368
1369         for (s = buf; len > 0; ) {
1370                 opt = *s;
1371                 if (opt == TCPOPT_EOL)
1372                         break;
1373                 else if (opt == TCPOPT_NOP)
1374                         ol = 1;
1375                 else {
1376                         if (len < 2)
1377                                 break;
1378                         ol = (int)*(s + 1);
1379                         if (ol < 2 || ol > len)
1380                                 break;
1381
1382                         /*
1383                          * Extract the TCP options we are interested in out of
1384                          * the header and store them in the the tcpdata struct.
1385                          */
1386                         switch (opt)
1387                         {
1388                         case TCPOPT_WINDOW :
1389                                 if (ol == TCPOLEN_WINDOW) {
1390                                         i = (int)*(s + 2);
1391                                         if (i > TCP_WSCALE_MAX)
1392                                                 i = TCP_WSCALE_MAX;
1393                                         else if (i < 0)
1394                                                 i = 0;
1395                                         td->td_winscale = i;
1396                                         td->td_winflags |= TCP_WSCALE_SEEN|
1397                                                            TCP_WSCALE_FIRST;
1398                                 } else
1399                                         retval = -1;
1400                                 break;
1401                         case TCPOPT_MAXSEG :
1402                                 /*
1403                                  * So, if we wanted to set the TCP MAXSEG,
1404                                  * it should be done here...
1405                                  */
1406                                 if (ol == TCPOLEN_MAXSEG) {
1407                                         i = (int)*(s + 2);
1408                                         i <<= 8;
1409                                         i += (int)*(s + 3);
1410                                         td->td_maxseg = i;
1411                                 } else
1412                                         retval = -1;
1413                                 break;
1414                         case TCPOPT_SACK_PERMITTED :
1415                                 if (ol == TCPOLEN_SACK_PERMITTED)
1416                                         td->td_winflags |= TCP_SACK_PERMIT;
1417                                 else
1418                                         retval = -1;
1419                                 break;
1420                         }
1421                 }
1422                 len -= ol;
1423                 s += ol;
1424         }
1425         return retval;
1426 }
1427
1428
1429 /* ------------------------------------------------------------------------ */
1430 /* Function:    fr_tcpstate                                                 */
1431 /* Returns:     int - 1 == packet matches state entry, 0 == it does not     */
1432 /* Parameters:  fin(I)   - pointer to packet information                    */
1433 /*              tcp(I)   - pointer to TCP packet header                     */
1434 /*              is(I)  - pointer to master state structure                  */
1435 /*                                                                          */
1436 /* Check to see if a packet with TCP headers fits within the TCP window.    */
1437 /* Change timeout depending on whether new packet is a SYN-ACK returning    */
1438 /* for a SYN or a RST or FIN which indicate time to close up shop.          */
1439 /* ------------------------------------------------------------------------ */
1440 static int fr_tcpstate(fin, tcp, is)
1441 fr_info_t *fin;
1442 tcphdr_t *tcp;
1443 ipstate_t *is;
1444 {
1445         int source, ret = 0, flags;
1446         tcpdata_t  *fdata, *tdata;
1447
1448         source = !fin->fin_rev;
1449         if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) &&
1450             (ntohs(is->is_sport) != fin->fin_data[0]))
1451                 source = 0;
1452         fdata = &is->is_tcp.ts_data[!source];
1453         tdata = &is->is_tcp.ts_data[source];
1454
1455         MUTEX_ENTER(&is->is_lock);
1456
1457         /*
1458          * If a SYN packet is received for a connection that is on the way out
1459          * but hasn't yet departed then advance this session along the way.
1460          */
1461         if ((tcp->th_flags & TH_OPENING) == TH_SYN) {
1462                 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
1463                     (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
1464                         is->is_state[!source] = IPF_TCPS_CLOSED;
1465                         fr_movequeue(&is->is_sti, is->is_sti.tqe_ifq,
1466                                      &ips_deletetq);
1467                         MUTEX_EXIT(&is->is_lock);
1468                         return 0;
1469                 }
1470         }
1471
1472         ret = fr_tcpinwindow(fin, fdata, tdata, tcp, is->is_flags);
1473         if (ret > 0) {
1474 #ifdef  IPFILTER_SCAN
1475                 if (is->is_flags & (IS_SC_CLIENT|IS_SC_SERVER)) {
1476                         ipsc_packet(fin, is);
1477                         if (FR_ISBLOCK(is->is_pass)) {
1478                                 MUTEX_EXIT(&is->is_lock);
1479                                 return 1;
1480                         }
1481                 }
1482 #endif
1483
1484                 /*
1485                  * Nearing end of connection, start timeout.
1486                  */
1487                 ret = fr_tcp_age(&is->is_sti, fin, ips_tqtqb, is->is_flags);
1488                 if (ret == 0) {
1489                         MUTEX_EXIT(&is->is_lock);
1490                         return 0;
1491                 }
1492
1493                 /*
1494                  * set s0's as appropriate.  Use syn-ack packet as it
1495                  * contains both pieces of required information.
1496                  */
1497                 /*
1498                  * Window scale option is only present in SYN/SYN-ACK packet.
1499                  * Compare with ~TH_FIN to mask out T/TCP setups.
1500                  */
1501                 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL);
1502                 if (flags == (TH_SYN|TH_ACK)) {
1503                         is->is_s0[source] = ntohl(tcp->th_ack);
1504                         is->is_s0[!source] = ntohl(tcp->th_seq) + 1;
1505                         if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
1506                                 if (fr_tcpoptions(fin, tcp, fdata) == -1)
1507                                         fin->fin_flx |= FI_BAD;
1508                         }
1509                         if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
1510                                 fr_checknewisn(fin, is);
1511                 } else if (flags == TH_SYN) {
1512                         is->is_s0[source] = ntohl(tcp->th_seq) + 1;
1513                         if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) {
1514                                 if (fr_tcpoptions(fin, tcp, fdata) == -1)
1515                                         fin->fin_flx |= FI_BAD;
1516                         }
1517
1518                         if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN))
1519                                 fr_checknewisn(fin, is);
1520
1521                 }
1522                 ret = 1;
1523         } else {
1524                 fin->fin_flx |= FI_OOW;
1525         }
1526         MUTEX_EXIT(&is->is_lock);
1527         return ret;
1528 }
1529
1530
1531 /* ------------------------------------------------------------------------ */
1532 /* Function:    fr_checknewisn                                              */
1533 /* Returns:     Nil                                                         */
1534 /* Parameters:  fin(I)   - pointer to packet information                    */
1535 /*              is(I)  - pointer to master state structure                  */
1536 /*                                                                          */
1537 /* Check to see if this TCP connection is expecting and needs a new         */
1538 /* sequence number for a particular direction of the connection.            */
1539 /*                                                                          */
1540 /* NOTE: This does not actually change the sequence numbers, only gets new  */
1541 /* one ready.                                                               */
1542 /* ------------------------------------------------------------------------ */
1543 static void fr_checknewisn(fin, is)
1544 fr_info_t *fin;
1545 ipstate_t *is;
1546 {
1547         u_32_t sumd, old, new;
1548         tcphdr_t *tcp;
1549         int i;
1550
1551         i = fin->fin_rev;
1552         tcp = fin->fin_dp;
1553
1554         if (((i == 0) && !(is->is_flags & IS_ISNSYN)) ||
1555             ((i == 1) && !(is->is_flags & IS_ISNACK))) {
1556                 old = ntohl(tcp->th_seq);
1557                 new = fr_newisn(fin);
1558                 is->is_isninc[i] = new - old;
1559                 CALC_SUMD(old, new, sumd);
1560                 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16);
1561
1562                 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK);
1563         }
1564 }
1565
1566
1567 /* ------------------------------------------------------------------------ */
1568 /* Function:    fr_tcpinwindow                                              */
1569 /* Returns:     int - 1 == packet inside TCP "window", 0 == not inside,     */
1570 /*                    2 == packet seq number matches next expected          */
1571 /* Parameters:  fin(I)   - pointer to packet information                    */
1572 /*              fdata(I) - pointer to tcp state informatio (forward)        */
1573 /*              tdata(I) - pointer to tcp state informatio (reverse)        */
1574 /*              tcp(I)   - pointer to TCP packet header                     */
1575 /*                                                                          */
1576 /* Given a packet has matched addresses and ports, check to see if it is    */
1577 /* within the TCP data window.  In a show of generosity, allow packets that */
1578 /* are within the window space behind the current sequence # as well.       */
1579 /* ------------------------------------------------------------------------ */
1580 int fr_tcpinwindow(fin, fdata, tdata, tcp, flags)
1581 fr_info_t *fin;
1582 tcpdata_t  *fdata, *tdata;
1583 tcphdr_t *tcp;
1584 int flags;
1585 {
1586         tcp_seq seq, ack, end;
1587         int ackskew, tcpflags;
1588         u_32_t win, maxwin;
1589         int dsize, inseq;
1590
1591         /*
1592          * Find difference between last checked packet and this packet.
1593          */
1594         tcpflags = tcp->th_flags;
1595         seq = ntohl(tcp->th_seq);
1596         ack = ntohl(tcp->th_ack);
1597         if (tcpflags & TH_SYN)
1598                 win = ntohs(tcp->th_win);
1599         else
1600                 win = ntohs(tcp->th_win) << fdata->td_winscale;
1601
1602         /*
1603          * A window of 0 produces undesirable behaviour from this function.
1604          */
1605         if (win == 0)
1606                 win = 1;
1607
1608         dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) +
1609                 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0);
1610
1611         /*
1612          * if window scaling is present, the scaling is only allowed
1613          * for windows not in the first SYN packet. In that packet the
1614          * window is 65535 to specify the largest window possible
1615          * for receivers not implementing the window scale option.
1616          * Currently, we do not assume TTCP here. That means that
1617          * if we see a second packet from a host (after the initial
1618          * SYN), we can assume that the receiver of the SYN did
1619          * already send back the SYN/ACK (and thus that we know if
1620          * the receiver also does window scaling)
1621          */
1622         if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) {
1623                 fdata->td_winflags &= ~TCP_WSCALE_FIRST;
1624                 fdata->td_maxwin = win;
1625         }
1626
1627         end = seq + dsize;
1628
1629         if ((fdata->td_end == 0) &&
1630             (!(flags & IS_TCPFSM) ||
1631              ((tcpflags & TH_OPENING) == TH_OPENING))) {
1632                 /*
1633                  * Must be a (outgoing) SYN-ACK in reply to a SYN.
1634                  */
1635                 fdata->td_end = end - 1;
1636                 fdata->td_maxwin = 1;
1637                 fdata->td_maxend = end + win;
1638         }
1639
1640         if (!(tcpflags & TH_ACK)) {  /* Pretend an ack was sent */
1641                 ack = tdata->td_end;
1642         } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) &&
1643                    (ack == 0)) {
1644                 /* gross hack to get around certain broken tcp stacks */
1645                 ack = tdata->td_end;
1646         }
1647
1648         maxwin = tdata->td_maxwin;
1649         ackskew = tdata->td_end - ack;
1650
1651         /*
1652          * Strict sequencing only allows in-order delivery.
1653          */
1654         if (seq != fdata->td_end) {
1655                 if ((flags & IS_STRICT) != 0) {
1656                         return 0;
1657                 }
1658         }
1659
1660         inseq = 0;
1661         if ((SEQ_GE(fdata->td_maxend, end)) &&
1662             (SEQ_GE(seq, fdata->td_end - maxwin)) &&
1663 /* XXX what about big packets */
1664 #define MAXACKWINDOW 66000
1665             (-ackskew <= (MAXACKWINDOW)) &&
1666             ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) {
1667                 inseq = 1;
1668         /*
1669          * Microsoft Windows will send the next packet to the right of the
1670          * window if SACK is in use.
1671          */
1672         } else if ((seq == fdata->td_maxend) && (ackskew == 0) &&
1673             (fdata->td_winflags & TCP_SACK_PERMIT) &&
1674             (tdata->td_winflags & TCP_SACK_PERMIT)) {
1675                 inseq = 1;
1676         /*
1677          * Sometimes a TCP RST will be generated with only the ACK field
1678          * set to non-zero.
1679          */
1680         } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) &&
1681                    (ackskew >= -1) && (ackskew <= 1)) {
1682                 inseq = 1;
1683         } else if (!(flags & IS_TCPFSM)) {
1684                 int i;
1685
1686                 i = (fin->fin_rev << 1) + fin->fin_out;
1687
1688 #if 0
1689                 if (is_pkts[i]0 == 0) {
1690                         /*
1691                          * Picking up a connection in the middle, the "next"
1692                          * packet seen from a direction that is new should be
1693                          * accepted, even if it appears out of sequence.
1694                          */
1695                         inseq = 1;
1696                 } else
1697 #endif
1698                 if (!(fdata->td_winflags &
1699                             (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) {
1700                         /*
1701                          * No TCPFSM and no window scaling, so make some
1702                          * extra guesses.
1703                          */
1704                         if ((seq == fdata->td_maxend) && (ackskew == 0))
1705                                 inseq = 1;
1706                         else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin))
1707                                 inseq = 1;
1708                 }
1709         }
1710
1711         /* TRACE(inseq, fdata, tdata, seq, end, ack, ackskew, win, maxwin) */
1712
1713         if (inseq) {
1714                 /* if ackskew < 0 then this should be due to fragmented
1715                  * packets. There is no way to know the length of the
1716                  * total packet in advance.
1717                  * We do know the total length from the fragment cache though.
1718                  * Note however that there might be more sessions with
1719                  * exactly the same source and destination parameters in the
1720                  * state cache (and source and destination is the only stuff
1721                  * that is saved in the fragment cache). Note further that
1722                  * some TCP connections in the state cache are hashed with
1723                  * sport and dport as well which makes it not worthwhile to
1724                  * look for them.
1725                  * Thus, when ackskew is negative but still seems to belong
1726                  * to this session, we bump up the destinations end value.
1727                  */
1728                 if (ackskew < 0)
1729                         tdata->td_end = ack;
1730
1731                 /* update max window seen */
1732                 if (fdata->td_maxwin < win)
1733                         fdata->td_maxwin = win;
1734                 if (SEQ_GT(end, fdata->td_end))
1735                         fdata->td_end = end;
1736                 if (SEQ_GE(ack + win, tdata->td_maxend))
1737                         tdata->td_maxend = ack + win;
1738                 return 1;
1739         }
1740         return 0;
1741 }
1742
1743
1744 /* ------------------------------------------------------------------------ */
1745 /* Function:    fr_stclone                                                  */
1746 /* Returns:     ipstate_t* - NULL == cloning failed,                        */
1747 /*                           else pointer to new state structure            */
1748 /* Parameters:  fin(I) - pointer to packet information                      */
1749 /*              tcp(I) - pointer to TCP/UDP header                          */
1750 /*              is(I)  - pointer to master state structure                  */
1751 /*                                                                          */
1752 /* Create a "duplcate" state table entry from the master.                   */
1753 /* ------------------------------------------------------------------------ */
1754 static ipstate_t *fr_stclone(fin, tcp, is)
1755 fr_info_t *fin;
1756 tcphdr_t *tcp;
1757 ipstate_t *is;
1758 {
1759         ipstate_t *clone;
1760         u_32_t send;
1761
1762         if (ips_num == fr_statemax) {
1763                 ATOMIC_INCL(ips_stats.iss_max);
1764                 fr_state_doflush = 1;
1765                 return NULL;
1766         }
1767         KMALLOC(clone, ipstate_t *);
1768         if (clone == NULL)
1769                 return NULL;
1770         bcopy((char *)is, (char *)clone, sizeof(*clone));
1771
1772         MUTEX_NUKE(&clone->is_lock);
1773
1774         clone->is_die = ONE_DAY + fr_ticks;
1775         clone->is_state[0] = 0;
1776         clone->is_state[1] = 0;
1777         send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) +
1778                 ((tcp->th_flags & TH_SYN) ? 1 : 0) +
1779                 ((tcp->th_flags & TH_FIN) ? 1 : 0);
1780
1781         if (fin->fin_rev == 1) {
1782                 clone->is_dend = send;
1783                 clone->is_maxdend = send;
1784                 clone->is_send = 0;
1785                 clone->is_maxswin = 1;
1786                 clone->is_maxdwin = ntohs(tcp->th_win);
1787                 if (clone->is_maxdwin == 0)
1788                         clone->is_maxdwin = 1;
1789         } else {
1790                 clone->is_send = send;
1791                 clone->is_maxsend = send;
1792                 clone->is_dend = 0;
1793                 clone->is_maxdwin = 1;
1794                 clone->is_maxswin = ntohs(tcp->th_win);
1795                 if (clone->is_maxswin == 0)
1796                         clone->is_maxswin = 1;
1797         }
1798
1799         clone->is_flags &= ~SI_CLONE;
1800         clone->is_flags |= SI_CLONED;
1801         fr_stinsert(clone, fin->fin_rev);
1802         clone->is_ref = 2;
1803         if (clone->is_p == IPPROTO_TCP) {
1804                 (void) fr_tcp_age(&clone->is_sti, fin, ips_tqtqb,
1805                                   clone->is_flags);
1806         }
1807         MUTEX_EXIT(&clone->is_lock);
1808 #ifdef  IPFILTER_SCAN
1809         (void) ipsc_attachis(is);
1810 #endif
1811 #ifdef  IPFILTER_SYNC
1812         if (is->is_flags & IS_STATESYNC)
1813                 clone->is_sync = ipfsync_new(SMC_STATE, fin, clone);
1814 #endif
1815         return clone;
1816 }
1817
1818
1819 /* ------------------------------------------------------------------------ */
1820 /* Function:    fr_matchsrcdst                                              */
1821 /* Returns:     Nil                                                         */
1822 /* Parameters:  fin(I) - pointer to packet information                      */
1823 /*              is(I)  - pointer to state structure                         */
1824 /*              src(I) - pointer to source address                          */
1825 /*              dst(I) - pointer to destination address                     */
1826 /*              tcp(I) - pointer to TCP/UDP header                          */
1827 /*                                                                          */
1828 /* Match a state table entry against an IP packet.  The logic below is that */
1829 /* ret gets set to one if the match succeeds, else remains 0.  If it is     */
1830 /* still 0 after the test. no match.                                        */
1831 /* ------------------------------------------------------------------------ */
1832 static ipstate_t *fr_matchsrcdst(fin, is, src, dst, tcp, cmask)
1833 fr_info_t *fin;
1834 ipstate_t *is;
1835 i6addr_t *src, *dst;
1836 tcphdr_t *tcp;
1837 u_32_t cmask;
1838 {
1839         int ret = 0, rev, out, flags, flx = 0, idx;
1840         u_short sp, dp;
1841         u_32_t cflx;
1842         void *ifp;
1843
1844         rev = IP6_NEQ(&is->is_dst, dst);
1845         ifp = fin->fin_ifp;
1846         out = fin->fin_out;
1847         flags = is->is_flags;
1848         sp = 0;
1849         dp = 0;
1850
1851         if (tcp != NULL) {
1852                 sp = htons(fin->fin_sport);
1853                 dp = ntohs(fin->fin_dport);
1854         }
1855         if (!rev) {
1856                 if (tcp != NULL) {
1857                         if (!(flags & SI_W_SPORT) && (sp != is->is_sport))
1858                                 rev = 1;
1859                         else if (!(flags & SI_W_DPORT) && (dp != is->is_dport))
1860                                 rev = 1;
1861                 }
1862         }
1863
1864         idx = (out << 1) + rev;
1865
1866         /*
1867          * If the interface for this 'direction' is set, make sure it matches.
1868          * An interface name that is not set matches any, as does a name of *.
1869          */
1870         if ((is->is_ifp[idx] == ifp) || (is->is_ifp[idx] == NULL &&
1871             (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '-' ||
1872              *is->is_ifname[idx] == '*')))
1873                 ret = 1;
1874
1875         if (ret == 0)
1876                 return NULL;
1877         ret = 0;
1878
1879         /*
1880          * Match addresses and ports.
1881          */
1882         if (rev == 0) {
1883                 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) &&
1884                     (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) {
1885                         if (tcp) {
1886                                 if ((sp == is->is_sport || flags & SI_W_SPORT)&&
1887                                     (dp == is->is_dport || flags & SI_W_DPORT))
1888                                         ret = 1;
1889                         } else {
1890                                 ret = 1;
1891                         }
1892                 }
1893         } else {
1894                 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) &&
1895                     (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) {
1896                         if (tcp) {
1897                                 if ((dp == is->is_sport || flags & SI_W_SPORT)&&
1898                                     (sp == is->is_dport || flags & SI_W_DPORT))
1899                                         ret = 1;
1900                         } else {
1901                                 ret = 1;
1902                         }
1903                 }
1904         }
1905
1906         if (ret == 0)
1907                 return NULL;
1908
1909         /*
1910          * Whether or not this should be here, is questionable, but the aim
1911          * is to get this out of the main line.
1912          */
1913         if (tcp == NULL)
1914                 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED);
1915
1916         /*
1917          * Only one of the source or destination address can be flaged as a
1918          * wildcard.  Fill in the missing address, if set.
1919          * For IPv6, if the address being copied in is multicast, then
1920          * don't reset the wild flag - multicast causes it to be set in the
1921          * first place!
1922          */
1923         if ((flags & (SI_W_SADDR|SI_W_DADDR))) {
1924                 fr_ip_t *fi = &fin->fin_fi;
1925
1926                 if ((flags & SI_W_SADDR) != 0) {
1927                         if (rev == 0) {
1928 #ifdef USE_INET6
1929                                 if (is->is_v == 6 &&
1930                                     IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6))
1931                                         /*EMPTY*/;
1932                                 else
1933 #endif
1934                                 {
1935                                         is->is_src = fi->fi_src;
1936                                         is->is_flags &= ~SI_W_SADDR;
1937                                 }
1938                         } else {
1939 #ifdef USE_INET6
1940                                 if (is->is_v == 6 &&
1941                                     IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6))
1942                                         /*EMPTY*/;
1943                                 else
1944 #endif
1945                                 {
1946                                         is->is_src = fi->fi_dst;
1947                                         is->is_flags &= ~SI_W_SADDR;
1948                                 }
1949                         }
1950                 } else if ((flags & SI_W_DADDR) != 0) {
1951                         if (rev == 0) {
1952 #ifdef USE_INET6
1953                                 if (is->is_v == 6 &&
1954                                     IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6))
1955                                         /*EMPTY*/;
1956                                 else
1957 #endif
1958                                 {
1959                                         is->is_dst = fi->fi_dst;
1960                                         is->is_flags &= ~SI_W_DADDR;
1961                                 }
1962                         } else {
1963 #ifdef USE_INET6
1964                                 if (is->is_v == 6 &&
1965                                     IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6))
1966                                         /*EMPTY*/;
1967                                 else
1968 #endif
1969                                 {
1970                                         is->is_dst = fi->fi_src;
1971                                         is->is_flags &= ~SI_W_DADDR;
1972                                 }
1973                         }
1974                 }
1975                 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) {
1976                         ATOMIC_DECL(ips_stats.iss_wild);
1977                 }
1978         }
1979
1980         flx = fin->fin_flx & cmask;
1981         cflx = is->is_flx[out][rev];
1982
1983         /*
1984          * Match up any flags set from IP options.
1985          */
1986         if ((cflx && (flx != (cflx & cmask))) ||
1987             ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) ||
1988             ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) ||
1989             ((fin->fin_auth & is->is_authmsk) != is->is_auth))
1990                 return NULL;
1991
1992         /*
1993          * Only one of the source or destination port can be flagged as a
1994          * wildcard.  When filling it in, fill in a copy of the matched entry
1995          * if it has the cloning flag set.
1996          */
1997         if ((fin->fin_flx & FI_IGNORE) != 0) {
1998                 fin->fin_rev = rev;
1999                 return is;
2000         }
2001
2002         if ((flags & (SI_W_SPORT|SI_W_DPORT))) {
2003                 if ((flags & SI_CLONE) != 0) {
2004                         ipstate_t *clone;
2005
2006                         clone = fr_stclone(fin, tcp, is);
2007                         if (clone == NULL)
2008                                 return NULL;
2009                         is = clone;
2010                 } else {
2011                         ATOMIC_DECL(ips_stats.iss_wild);
2012                 }
2013
2014                 if ((flags & SI_W_SPORT) != 0) {
2015                         if (rev == 0) {
2016                                 is->is_sport = sp;
2017                                 is->is_send = ntohl(tcp->th_seq);
2018                         } else {
2019                                 is->is_sport = dp;
2020                                 is->is_send = ntohl(tcp->th_ack);
2021                         }
2022                         is->is_maxsend = is->is_send + 1;
2023                 } else if ((flags & SI_W_DPORT) != 0) {
2024                         if (rev == 0) {
2025                                 is->is_dport = dp;
2026                                 is->is_dend = ntohl(tcp->th_ack);
2027                         } else {
2028                                 is->is_dport = sp;
2029                                 is->is_dend = ntohl(tcp->th_seq);
2030                         }
2031                         is->is_maxdend = is->is_dend + 1;
2032                 }
2033                 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT);
2034                 if ((flags & SI_CLONED) && ipstate_logging)
2035                         ipstate_log(is, ISL_CLONE);
2036         }
2037
2038         ret = -1;
2039
2040         if (is->is_flx[out][rev] == 0) {
2041                 is->is_flx[out][rev] = flx;
2042                 is->is_opt[rev] = fin->fin_optmsk;
2043                 if (is->is_v == 6) {
2044                         is->is_opt[rev] &= ~0x8;
2045                         is->is_optmsk[rev] &= ~0x8;
2046                 }
2047         }
2048
2049         /*
2050          * Check if the interface name for this "direction" is set and if not,
2051          * fill it in.
2052          */
2053         if (is->is_ifp[idx] == NULL &&
2054             (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) {
2055                 is->is_ifp[idx] = ifp;
2056                 COPYIFNAME(is->is_v, ifp, is->is_ifname[idx]);
2057         }
2058         fin->fin_rev = rev;
2059         return is;
2060 }
2061
2062
2063 /* ------------------------------------------------------------------------ */
2064 /* Function:    fr_checkicmpmatchingstate                                   */
2065 /* Returns:     Nil                                                         */
2066 /* Parameters:  fin(I) - pointer to packet information                      */
2067 /*                                                                          */
2068 /* If we've got an ICMP error message, using the information stored in the  */
2069 /* ICMP packet, look for a matching state table entry.                      */
2070 /*                                                                          */
2071 /* If we return NULL then no lock on ipf_state is held.                     */
2072 /* If we return non-null then a read-lock on ipf_state is held.             */
2073 /* ------------------------------------------------------------------------ */
2074 static ipstate_t *fr_checkicmpmatchingstate(fin)
2075 fr_info_t *fin;
2076 {
2077         ipstate_t *is, **isp;
2078         u_short sport, dport;
2079         u_char  pr;
2080         int backward, i, oi;
2081         i6addr_t dst, src;
2082         struct icmp *ic;
2083         u_short savelen;
2084         icmphdr_t *icmp;
2085         fr_info_t ofin;
2086         tcphdr_t *tcp;
2087         int type, len;
2088         ip_t *oip;
2089         u_int hv;
2090
2091         /*
2092          * Does it at least have the return (basic) IP header ?
2093          * Is it an actual recognised ICMP error type?
2094          * Only a basic IP header (no options) should be with
2095          * an ICMP error header.
2096          */
2097         if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) ||
2098             (fin->fin_plen < ICMPERR_MINPKTLEN) ||
2099             !(fin->fin_flx & FI_ICMPERR))
2100                 return NULL;
2101         ic = fin->fin_dp;
2102         type = ic->icmp_type;
2103
2104         oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN);
2105         /*
2106          * Check if the at least the old IP header (with options) and
2107          * 8 bytes of payload is present.
2108          */
2109         if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2))
2110                 return NULL;
2111
2112         /*
2113          * Sanity Checks.
2114          */
2115         len = fin->fin_dlen - ICMPERR_ICMPHLEN;
2116         if ((len <= 0) || ((IP_HL(oip) << 2) > len))
2117                 return NULL;
2118
2119         /*
2120          * Is the buffer big enough for all of it ?  It's the size of the IP
2121          * header claimed in the encapsulated part which is of concern.  It
2122          * may be too big to be in this buffer but not so big that it's
2123          * outside the ICMP packet, leading to TCP deref's causing problems.
2124          * This is possible because we don't know how big oip_hl is when we
2125          * do the pullup early in fr_check() and thus can't guarantee it is
2126          * all here now.
2127          */
2128 #ifdef  _KERNEL
2129         {
2130         mb_t *m;
2131
2132         m = fin->fin_m;
2133 # if defined(MENTAT)
2134         if ((char *)oip + len > (char *)m->b_wptr)
2135                 return NULL;
2136 # else
2137         if ((char *)oip + len > (char *)fin->fin_ip + m->m_len)
2138                 return NULL;
2139 # endif
2140         }
2141 #endif
2142         bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
2143
2144         /*
2145          * in the IPv4 case we must zero the i6addr union otherwise
2146          * the IP6_EQ and IP6_NEQ macros produce the wrong results because
2147          * of the 'junk' in the unused part of the union
2148          */
2149         bzero((char *)&src, sizeof(src));
2150         bzero((char *)&dst, sizeof(dst));
2151
2152         /*
2153          * we make an fin entry to be able to feed it to
2154          * matchsrcdst note that not all fields are encessary
2155          * but this is the cleanest way. Note further we fill
2156          * in fin_mp such that if someone uses it we'll get
2157          * a kernel panic. fr_matchsrcdst does not use this.
2158          *
2159          * watch out here, as ip is in host order and oip in network
2160          * order. Any change we make must be undone afterwards, like
2161          * oip->ip_off - it is still in network byte order so fix it.
2162          */
2163         savelen = oip->ip_len;
2164         oip->ip_len = len;
2165         oip->ip_off = ntohs(oip->ip_off);
2166
2167         ofin.fin_flx = FI_NOCKSUM;
2168         ofin.fin_v = 4;
2169         ofin.fin_ip = oip;
2170         ofin.fin_m = NULL;      /* if dereferenced, panic XXX */
2171         ofin.fin_mp = NULL;     /* if dereferenced, panic XXX */
2172         (void) fr_makefrip(IP_HL(oip) << 2, oip, &ofin);
2173         ofin.fin_ifp = fin->fin_ifp;
2174         ofin.fin_out = !fin->fin_out;
2175         /*
2176          * Reset the short and bad flag here because in fr_matchsrcdst()
2177          * the flags for the current packet (fin_flx) are compared against
2178          * those for the existing session.
2179          */
2180         ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
2181
2182         /*
2183          * Put old values of ip_len and ip_off back as we don't know
2184          * if we have to forward the packet (or process it again.
2185          */
2186         oip->ip_len = savelen;
2187         oip->ip_off = htons(oip->ip_off);
2188
2189         switch (oip->ip_p)
2190         {
2191         case IPPROTO_ICMP :
2192                 /*
2193                  * an ICMP error can only be generated as a result of an
2194                  * ICMP query, not as the response on an ICMP error
2195                  *
2196                  * XXX theoretically ICMP_ECHOREP and the other reply's are
2197                  * ICMP query's as well, but adding them here seems strange XXX
2198                  */
2199                 if ((ofin.fin_flx & FI_ICMPERR) != 0)
2200                         return NULL;
2201
2202                 /*
2203                  * perform a lookup of the ICMP packet in the state table
2204                  */
2205                 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2206                 hv = (pr = oip->ip_p);
2207                 src.in4 = oip->ip_src;
2208                 hv += src.in4.s_addr;
2209                 dst.in4 = oip->ip_dst;
2210                 hv += dst.in4.s_addr;
2211                 hv += icmp->icmp_id;
2212                 hv = DOUBLE_HASH(hv);
2213
2214                 READ_ENTER(&ipf_state);
2215                 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) {
2216                         isp = &is->is_hnext;
2217                         if ((is->is_p != pr) || (is->is_v != 4))
2218                                 continue;
2219                         if (is->is_pass & FR_NOICMPERR)
2220                                 continue;
2221                         is = fr_matchsrcdst(&ofin, is, &src, &dst,
2222                                             NULL, FI_ICMPCMP);
2223                         if (is != NULL) {
2224                                 /*
2225                                  * i  : the index of this packet (the icmp
2226                                  *      unreachable)
2227                                  * oi : the index of the original packet found
2228                                  *      in the icmp header (i.e. the packet
2229                                  *      causing this icmp)
2230                                  * backward : original packet was backward
2231                                  *      compared to the state
2232                                  */
2233                                 backward = IP6_NEQ(&is->is_src, &src);
2234                                 fin->fin_rev = !backward;
2235                                 i = (!backward << 1) + fin->fin_out;
2236                                 oi = (backward << 1) + ofin.fin_out;
2237                                 if (is->is_icmppkts[i] > is->is_pkts[oi])
2238                                         continue;
2239                                 ips_stats.iss_hits++;
2240                                 is->is_icmppkts[i]++;
2241                                 return is;
2242                         }
2243                 }
2244                 RWLOCK_EXIT(&ipf_state);
2245                 return NULL;
2246         case IPPROTO_TCP :
2247         case IPPROTO_UDP :
2248                 break;
2249         default :
2250                 return NULL;
2251         }
2252
2253         tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2));
2254         dport = tcp->th_dport;
2255         sport = tcp->th_sport;
2256
2257         hv = (pr = oip->ip_p);
2258         src.in4 = oip->ip_src;
2259         hv += src.in4.s_addr;
2260         dst.in4 = oip->ip_dst;
2261         hv += dst.in4.s_addr;
2262         hv += dport;
2263         hv += sport;
2264         hv = DOUBLE_HASH(hv);
2265
2266         READ_ENTER(&ipf_state);
2267         for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) {
2268                 isp = &is->is_hnext;
2269                 /*
2270                  * Only allow this icmp though if the
2271                  * encapsulated packet was allowed through the
2272                  * other way around. Note that the minimal amount
2273                  * of info present does not allow for checking against
2274                  * tcp internals such as seq and ack numbers.   Only the
2275                  * ports are known to be present and can be even if the
2276                  * short flag is set.
2277                  */
2278                 if ((is->is_p == pr) && (is->is_v == 4) &&
2279                     (is = fr_matchsrcdst(&ofin, is, &src, &dst,
2280                                          tcp, FI_ICMPCMP))) {
2281                         /*
2282                          * i  : the index of this packet (the icmp unreachable)
2283                          * oi : the index of the original packet found in the
2284                          *      icmp header (i.e. the packet causing this icmp)
2285                          * backward : original packet was backward compared to
2286                          *            the state
2287                          */
2288                         backward = IP6_NEQ(&is->is_src, &src);
2289                         fin->fin_rev = !backward;
2290                         i = (!backward << 1) + fin->fin_out;
2291                         oi = (backward << 1) + ofin.fin_out;
2292
2293                         if (((is->is_pass & FR_NOICMPERR) != 0) ||
2294                             (is->is_icmppkts[i] > is->is_pkts[oi]))
2295                                 break;
2296                         ips_stats.iss_hits++;
2297                         is->is_icmppkts[i]++;
2298                         /*
2299                          * we deliberately do not touch the timeouts
2300                          * for the accompanying state table entry.
2301                          * It remains to be seen if that is correct. XXX
2302                          */
2303                         return is;
2304                 }
2305         }
2306         RWLOCK_EXIT(&ipf_state);
2307         return NULL;
2308 }
2309
2310
2311 /* ------------------------------------------------------------------------ */
2312 /* Function:    fr_ipsmove                                                  */
2313 /* Returns:     Nil                                                         */
2314 /* Parameters:  is(I) - pointer to state table entry                        */
2315 /*              hv(I) - new hash value for state table entry                */
2316 /* Write Locks: ipf_state                                                   */
2317 /*                                                                          */
2318 /* Move a state entry from one position in the hash table to another.       */
2319 /* ------------------------------------------------------------------------ */
2320 static void fr_ipsmove(is, hv)
2321 ipstate_t *is;
2322 u_int hv;
2323 {
2324         ipstate_t **isp;
2325         u_int hvm;
2326
2327         hvm = is->is_hv;
2328         /*
2329          * Remove the hash from the old location...
2330          */
2331         isp = is->is_phnext;
2332         if (is->is_hnext)
2333                 is->is_hnext->is_phnext = isp;
2334         *isp = is->is_hnext;
2335         if (ips_table[hvm] == NULL)
2336                 ips_stats.iss_inuse--;
2337         ips_stats.iss_bucketlen[hvm]--;
2338
2339         /*
2340          * ...and put the hash in the new one.
2341          */
2342         hvm = DOUBLE_HASH(hv);
2343         is->is_hv = hvm;
2344         isp = &ips_table[hvm];
2345         if (*isp)
2346                 (*isp)->is_phnext = &is->is_hnext;
2347         else
2348                 ips_stats.iss_inuse++;
2349         ips_stats.iss_bucketlen[hvm]++;
2350         is->is_phnext = isp;
2351         is->is_hnext = *isp;
2352         *isp = is;
2353 }
2354
2355
2356 /* ------------------------------------------------------------------------ */
2357 /* Function:    fr_stlookup                                                 */
2358 /* Returns:     ipstate_t* - NULL == no matching state found,               */
2359 /*                           else pointer to state information is returned  */
2360 /* Parameters:  fin(I) - pointer to packet information                      */
2361 /*              tcp(I) - pointer to TCP/UDP header.                         */
2362 /*                                                                          */
2363 /* Search the state table for a matching entry to the packet described by   */
2364 /* the contents of *fin.                                                    */
2365 /*                                                                          */
2366 /* If we return NULL then no lock on ipf_state is held.                     */
2367 /* If we return non-null then a read-lock on ipf_state is held.             */
2368 /* ------------------------------------------------------------------------ */
2369 ipstate_t *fr_stlookup(fin, tcp, ifqp)
2370 fr_info_t *fin;
2371 tcphdr_t *tcp;
2372 ipftq_t **ifqp;
2373 {
2374         u_int hv, hvm, pr, v, tryagain;
2375         ipstate_t *is, **isp;
2376         u_short dport, sport;
2377         i6addr_t src, dst;
2378         struct icmp *ic;
2379         ipftq_t *ifq;
2380         int oow;
2381
2382         is = NULL;
2383         ifq = NULL;
2384         tcp = fin->fin_dp;
2385         ic = (struct icmp *)tcp;
2386         hv = (pr = fin->fin_fi.fi_p);
2387         src = fin->fin_fi.fi_src;
2388         dst = fin->fin_fi.fi_dst;
2389         hv += src.in4.s_addr;
2390         hv += dst.in4.s_addr;
2391
2392         v = fin->fin_fi.fi_v;
2393 #ifdef  USE_INET6
2394         if (v == 6) {
2395                 hv  += fin->fin_fi.fi_src.i6[1];
2396                 hv  += fin->fin_fi.fi_src.i6[2];
2397                 hv  += fin->fin_fi.fi_src.i6[3];
2398
2399                 if ((fin->fin_p == IPPROTO_ICMPV6) &&
2400                     IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) {
2401                         hv -= dst.in4.s_addr;
2402                 } else {
2403                         hv += fin->fin_fi.fi_dst.i6[1];
2404                         hv += fin->fin_fi.fi_dst.i6[2];
2405                         hv += fin->fin_fi.fi_dst.i6[3];
2406                 }
2407         }
2408 #endif
2409         if ((v == 4) &&
2410             (fin->fin_flx & (FI_MULTICAST|FI_BROADCAST|FI_MBCAST))) {
2411                 if (fin->fin_out == 0) {
2412                         hv -= src.in4.s_addr;
2413                 } else {
2414                         hv -= dst.in4.s_addr;
2415                 }
2416         }
2417
2418         /*
2419          * Search the hash table for matching packet header info.
2420          */
2421         switch (pr)
2422         {
2423 #ifdef  USE_INET6
2424         case IPPROTO_ICMPV6 :
2425                 tryagain = 0;
2426                 if (v == 6) {
2427                         if ((ic->icmp_type == ICMP6_ECHO_REQUEST) ||
2428                             (ic->icmp_type == ICMP6_ECHO_REPLY)) {
2429                                 hv += ic->icmp_id;
2430                         }
2431                 }
2432                 READ_ENTER(&ipf_state);
2433 icmp6again:
2434                 hvm = DOUBLE_HASH(hv);
2435                 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) {
2436                         isp = &is->is_hnext;
2437                         /*
2438                          * If a connection is about to be deleted, no packets
2439                          * are allowed to match it.
2440                          */
2441                         if (is->is_sti.tqe_ifq == &ips_deletetq)
2442                                 continue;
2443
2444                         if ((is->is_p != pr) || (is->is_v != v))
2445                                 continue;
2446                         is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
2447                         if (is != NULL &&
2448                             fr_matchicmpqueryreply(v, &is->is_icmp,
2449                                                    ic, fin->fin_rev)) {
2450                                 if (fin->fin_rev)
2451                                         ifq = &ips_icmpacktq;
2452                                 else
2453                                         ifq = &ips_icmptq;
2454                                 break;
2455                         }
2456                 }
2457
2458                 if (is != NULL) {
2459                         if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) {
2460                                 hv += fin->fin_fi.fi_src.i6[0];
2461                                 hv += fin->fin_fi.fi_src.i6[1];
2462                                 hv += fin->fin_fi.fi_src.i6[2];
2463                                 hv += fin->fin_fi.fi_src.i6[3];
2464                                 fr_ipsmove(is, hv);
2465                                 MUTEX_DOWNGRADE(&ipf_state);
2466                         }
2467                         break;
2468                 }
2469                 RWLOCK_EXIT(&ipf_state);
2470
2471                 /*
2472                  * No matching icmp state entry. Perhaps this is a
2473                  * response to another state entry.
2474                  *
2475                  * XXX With some ICMP6 packets, the "other" address is already
2476                  * in the packet, after the ICMP6 header, and this could be
2477                  * used in place of the multicast address.  However, taking
2478                  * advantage of this requires some significant code changes
2479                  * to handle the specific types where that is the case.
2480                  */
2481                 if ((ips_stats.iss_wild != 0) && (v == 6) && (tryagain == 0) &&
2482                     !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) {
2483                         hv -= fin->fin_fi.fi_src.i6[0];
2484                         hv -= fin->fin_fi.fi_src.i6[1];
2485                         hv -= fin->fin_fi.fi_src.i6[2];
2486                         hv -= fin->fin_fi.fi_src.i6[3];
2487                         tryagain = 1;
2488                         WRITE_ENTER(&ipf_state);
2489                         goto icmp6again;
2490                 }
2491
2492                 is = fr_checkicmp6matchingstate(fin);
2493                 if (is != NULL)
2494                         return is;
2495                 break;
2496 #endif
2497
2498         case IPPROTO_ICMP :
2499                 if (v == 4) {
2500                         hv += ic->icmp_id;
2501                 }
2502                 hv = DOUBLE_HASH(hv);
2503                 READ_ENTER(&ipf_state);
2504                 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) {
2505                         isp = &is->is_hnext;
2506                         if ((is->is_p != pr) || (is->is_v != v))
2507                                 continue;
2508                         is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
2509                         if ((is != NULL) &&
2510                             (ic->icmp_id == is->is_icmp.ici_id) &&
2511                             fr_matchicmpqueryreply(v, &is->is_icmp,
2512                                                    ic, fin->fin_rev)) {
2513                                 if (fin->fin_rev)
2514                                         ifq = &ips_icmpacktq;
2515                                 else
2516                                         ifq = &ips_icmptq;
2517                                 break;
2518                         }
2519                 }
2520                 if (is == NULL) {
2521                         RWLOCK_EXIT(&ipf_state);
2522                 }
2523                 break;
2524
2525         case IPPROTO_TCP :
2526         case IPPROTO_UDP :
2527                 ifqp = NULL;
2528                 sport = htons(fin->fin_data[0]);
2529                 hv += sport;
2530                 dport = htons(fin->fin_data[1]);
2531                 hv += dport;
2532                 oow = 0;
2533                 tryagain = 0;
2534                 READ_ENTER(&ipf_state);
2535 retry_tcpudp:
2536                 hvm = DOUBLE_HASH(hv);
2537                 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) {
2538                         isp = &is->is_hnext;
2539                         if ((is->is_p != pr) || (is->is_v != v))
2540                                 continue;
2541                         fin->fin_flx &= ~FI_OOW;
2542                         is = fr_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP);
2543                         if (is != NULL) {
2544                                 if (pr == IPPROTO_TCP) {
2545                                         if (!fr_tcpstate(fin, tcp, is)) {
2546                                                 oow |= fin->fin_flx & FI_OOW;
2547                                                 continue;
2548                                         }
2549                                 }
2550                                 break;
2551                         }
2552                 }
2553                 if (is != NULL) {
2554                         if (tryagain &&
2555                             !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) {
2556                                 hv += dport;
2557                                 hv += sport;
2558                                 fr_ipsmove(is, hv);
2559                                 MUTEX_DOWNGRADE(&ipf_state);
2560                         }
2561                         break;
2562                 }
2563                 RWLOCK_EXIT(&ipf_state);
2564
2565                 if (ips_stats.iss_wild) {
2566                         if (tryagain == 0) {
2567                                 hv -= dport;
2568                                 hv -= sport;
2569                         } else if (tryagain == 1) {
2570                                 hv = fin->fin_fi.fi_p;
2571                                 /*
2572                                  * If we try to pretend this is a reply to a
2573                                  * multicast/broadcast packet then we need to
2574                                  * exclude part of the address from the hash
2575                                  * calculation.
2576                                  */
2577                                 if (fin->fin_out == 0) {
2578                                         hv += src.in4.s_addr;
2579                                 } else {
2580                                         hv += dst.in4.s_addr;
2581                                 }
2582                                 hv += dport;
2583                                 hv += sport;
2584                         }
2585                         tryagain++;
2586                         if (tryagain <= 2) {
2587                                 WRITE_ENTER(&ipf_state);
2588                                 goto retry_tcpudp;
2589                         }
2590                 }
2591                 fin->fin_flx |= oow;
2592                 break;
2593
2594 #if 0
2595         case IPPROTO_GRE :
2596                 gre = fin->fin_dp;
2597                 if (GRE_REV(gre->gr_flags) == 1) {
2598                         hv += gre->gr_call;
2599                 }
2600                 /* FALLTHROUGH */
2601 #endif
2602         default :
2603                 ifqp = NULL;
2604                 hvm = DOUBLE_HASH(hv);
2605                 READ_ENTER(&ipf_state);
2606                 for (isp = &ips_table[hvm]; ((is = *isp) != NULL); ) {
2607                         isp = &is->is_hnext;
2608                         if ((is->is_p != pr) || (is->is_v != v))
2609                                 continue;
2610                         is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP);
2611                         if (is != NULL) {
2612                                 ifq = &ips_iptq;
2613                                 break;
2614                         }
2615                 }
2616                 if (is == NULL) {
2617                         RWLOCK_EXIT(&ipf_state);
2618                 }
2619                 break;
2620         }
2621
2622         if (is != NULL) {
2623                 if (((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) &&
2624                     (is->is_tqehead[fin->fin_rev] != NULL))
2625                         ifq = is->is_tqehead[fin->fin_rev];
2626                 if (ifq != NULL && ifqp != NULL)
2627                         *ifqp = ifq;
2628         }
2629         return is;
2630 }
2631
2632
2633 /* ------------------------------------------------------------------------ */
2634 /* Function:    fr_updatestate                                              */
2635 /* Returns:     Nil                                                         */
2636 /* Parameters:  fin(I) - pointer to packet information                      */
2637 /*              is(I)  - pointer to state table entry                       */
2638 /* Read Locks:  ipf_state                                                   */
2639 /*                                                                          */
2640 /* Updates packet and byte counters for a newly received packet.  Seeds the */
2641 /* fragment cache with a new entry as required.                             */
2642 /* ------------------------------------------------------------------------ */
2643 void fr_updatestate(fin, is, ifq)
2644 fr_info_t *fin;
2645 ipstate_t *is;
2646 ipftq_t *ifq;
2647 {
2648         ipftqent_t *tqe;
2649         int i, pass;
2650
2651         i = (fin->fin_rev << 1) + fin->fin_out;
2652
2653         /*
2654          * For TCP packets, ifq == NULL.  For all others, check if this new
2655          * queue is different to the last one it was on and move it if so.
2656          */
2657         tqe = &is->is_sti;
2658         MUTEX_ENTER(&is->is_lock);
2659         if ((tqe->tqe_flags & TQE_RULEBASED) != 0)
2660                 ifq = is->is_tqehead[fin->fin_rev];
2661
2662         if (ifq != NULL)
2663                 fr_movequeue(tqe, tqe->tqe_ifq, ifq);
2664
2665         is->is_pkts[i]++;
2666         is->is_bytes[i] += fin->fin_plen;
2667         MUTEX_EXIT(&is->is_lock);
2668
2669 #ifdef  IPFILTER_SYNC
2670         if (is->is_flags & IS_STATESYNC)
2671                 ipfsync_update(SMC_STATE, fin, is->is_sync);
2672 #endif
2673
2674         ATOMIC_INCL(ips_stats.iss_hits);
2675
2676         fin->fin_fr = is->is_rule;
2677
2678         /*
2679          * If this packet is a fragment and the rule says to track fragments,
2680          * then create a new fragment cache entry.
2681          */
2682         pass = is->is_pass;
2683         if ((fin->fin_flx & FI_FRAG) && FR_ISPASS(pass))
2684                 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE);
2685 }
2686
2687
2688 /* ------------------------------------------------------------------------ */
2689 /* Function:    fr_checkstate                                               */
2690 /* Returns:     frentry_t* - NULL == search failed,                         */
2691 /*                           else pointer to rule for matching state        */
2692 /* Parameters:  ifp(I)   - pointer to interface                             */
2693 /*              passp(I) - pointer to filtering result flags                */
2694 /*                                                                          */
2695 /* Check if a packet is associated with an entry in the state table.        */
2696 /* ------------------------------------------------------------------------ */
2697 frentry_t *fr_checkstate(fin, passp)
2698 fr_info_t *fin;
2699 u_32_t *passp;
2700 {
2701         ipstate_t *is;
2702         frentry_t *fr;
2703         tcphdr_t *tcp;
2704         ipftq_t *ifq;
2705         u_int pass;
2706
2707         if (fr_state_lock || (ips_list == NULL) ||
2708             (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD)))
2709                 return NULL;
2710
2711         is = NULL;
2712         if ((fin->fin_flx & FI_TCPUDP) ||
2713             (fin->fin_fi.fi_p == IPPROTO_ICMP)
2714 #ifdef  USE_INET6
2715             || (fin->fin_fi.fi_p == IPPROTO_ICMPV6)
2716 #endif
2717             )
2718                 tcp = fin->fin_dp;
2719         else
2720                 tcp = NULL;
2721
2722         /*
2723          * Search the hash table for matching packet header info.
2724          */
2725         ifq = NULL;
2726         is = fin->fin_state;
2727         if (is == NULL)
2728                 is = fr_stlookup(fin, tcp, &ifq);
2729         switch (fin->fin_p)
2730         {
2731 #ifdef  USE_INET6
2732         case IPPROTO_ICMPV6 :
2733                 if (is != NULL)
2734                         break;
2735                 if (fin->fin_v == 6) {
2736                         is = fr_checkicmp6matchingstate(fin);
2737                         if (is != NULL)
2738                                 goto matched;
2739                 }
2740                 break;
2741 #endif
2742         case IPPROTO_ICMP :
2743                 if (is != NULL)
2744                         break;
2745                 /*
2746                  * No matching icmp state entry. Perhaps this is a
2747                  * response to another state entry.
2748                  */
2749                 is = fr_checkicmpmatchingstate(fin);
2750                 if (is != NULL)
2751                         goto matched;
2752                 break;
2753         case IPPROTO_TCP :
2754                 if (is == NULL)
2755                         break;
2756
2757                 if (is->is_pass & FR_NEWISN) {
2758                         if (fin->fin_out == 0)
2759                                 fr_fixinisn(fin, is);
2760                         else if (fin->fin_out == 1)
2761                                 fr_fixoutisn(fin, is);
2762                 }
2763                 break;
2764         default :
2765                 if (fin->fin_rev)
2766                         ifq = &ips_udpacktq;
2767                 else
2768                         ifq = &ips_udptq;
2769                 break;
2770         }
2771         if (is == NULL) {
2772                 ATOMIC_INCL(ips_stats.iss_miss);
2773                 return NULL;
2774         }
2775
2776 matched:
2777         fr = is->is_rule;
2778         if (fr != NULL) {
2779                 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) {
2780                         if (fin->fin_nattag == NULL)
2781                                 return NULL;
2782                         if (fr_matchtag(&fr->fr_nattag, fin->fin_nattag) != 0)
2783                                 return NULL;
2784                 }
2785                 (void) strncpy(fin->fin_group, fr->fr_group, FR_GROUPLEN);
2786                 fin->fin_icode = fr->fr_icode;
2787         }
2788
2789         fin->fin_rule = is->is_rulen;
2790         pass = is->is_pass;
2791         fr_updatestate(fin, is, ifq);
2792
2793         fin->fin_state = is;
2794         is->is_touched = fr_ticks;
2795         MUTEX_ENTER(&is->is_lock);
2796         is->is_ref++;
2797         MUTEX_EXIT(&is->is_lock);
2798         RWLOCK_EXIT(&ipf_state);
2799         fin->fin_flx |= FI_STATE;
2800         if ((pass & FR_LOGFIRST) != 0)
2801                 pass &= ~(FR_LOGFIRST|FR_LOG);
2802         *passp = pass;
2803         return fr;
2804 }
2805
2806
2807 /* ------------------------------------------------------------------------ */
2808 /* Function:    fr_fixoutisn                                                */
2809 /* Returns:     Nil                                                         */
2810 /* Parameters:  fin(I)   - pointer to packet information                    */
2811 /*              is(I)  - pointer to master state structure                  */
2812 /*                                                                          */
2813 /* Called only for outbound packets, adjusts the sequence number and the    */
2814 /* TCP checksum to match that change.                                       */
2815 /* ------------------------------------------------------------------------ */
2816 static void fr_fixoutisn(fin, is)
2817 fr_info_t *fin;
2818 ipstate_t *is;
2819 {
2820         tcphdr_t *tcp;
2821         int rev;
2822         u_32_t seq;
2823
2824         tcp = fin->fin_dp;
2825         rev = fin->fin_rev;
2826         if ((is->is_flags & IS_ISNSYN) != 0) {
2827                 if (rev == 0) {
2828                         seq = ntohl(tcp->th_seq);
2829                         seq += is->is_isninc[0];
2830                         tcp->th_seq = htonl(seq);
2831                         fix_outcksum(fin, &tcp->th_sum, is->is_sumd[0]);
2832                 }
2833         }
2834         if ((is->is_flags & IS_ISNACK) != 0) {
2835                 if (rev == 1) {
2836                         seq = ntohl(tcp->th_seq);
2837                         seq += is->is_isninc[1];
2838                         tcp->th_seq = htonl(seq);
2839                         fix_outcksum(fin, &tcp->th_sum, is->is_sumd[1]);
2840                 }
2841         }
2842 }
2843
2844
2845 /* ------------------------------------------------------------------------ */
2846 /* Function:    fr_fixinisn                                                 */
2847 /* Returns:     Nil                                                         */
2848 /* Parameters:  fin(I)   - pointer to packet information                    */
2849 /*              is(I)  - pointer to master state structure                  */
2850 /*                                                                          */
2851 /* Called only for inbound packets, adjusts the acknowledge number and the  */
2852 /* TCP checksum to match that change.                                       */
2853 /* ------------------------------------------------------------------------ */
2854 static void fr_fixinisn(fin, is)
2855 fr_info_t *fin;
2856 ipstate_t *is;
2857 {
2858         tcphdr_t *tcp;
2859         int rev;
2860         u_32_t ack;
2861
2862         tcp = fin->fin_dp;
2863         rev = fin->fin_rev;
2864         if ((is->is_flags & IS_ISNSYN) != 0) {
2865                 if (rev == 1) {
2866                         ack = ntohl(tcp->th_ack);
2867                         ack -= is->is_isninc[0];
2868                         tcp->th_ack = htonl(ack);
2869                         fix_incksum(fin, &tcp->th_sum, is->is_sumd[0]);
2870                 }
2871         }
2872         if ((is->is_flags & IS_ISNACK) != 0) {
2873                 if (rev == 0) {
2874                         ack = ntohl(tcp->th_ack);
2875                         ack -= is->is_isninc[1];
2876                         tcp->th_ack = htonl(ack);
2877                         fix_incksum(fin, &tcp->th_sum, is->is_sumd[1]);
2878                 }
2879         }
2880 }
2881
2882
2883 /* ------------------------------------------------------------------------ */
2884 /* Function:    fr_statesync                                                */
2885 /* Returns:     Nil                                                         */
2886 /* Parameters:  ifp(I) - pointer to interface                               */
2887 /*                                                                          */
2888 /* Walk through all state entries and if an interface pointer match is      */
2889 /* found then look it up again, based on its name in case the pointer has   */
2890 /* changed since last time.                                                 */
2891 /*                                                                          */
2892 /* If ifp is passed in as being non-null then we are only doing updates for */
2893 /* existing, matching, uses of it.                                          */
2894 /* ------------------------------------------------------------------------ */
2895 void fr_statesync(ifp)
2896 void *ifp;
2897 {
2898         ipstate_t *is;
2899         int i;
2900
2901         if (fr_running <= 0)
2902                 return;
2903
2904         WRITE_ENTER(&ipf_state);
2905
2906         if (fr_running <= 0) {
2907                 RWLOCK_EXIT(&ipf_state);
2908                 return;
2909         }
2910
2911         for (is = ips_list; is; is = is->is_next) {
2912                 /*
2913                  * Look up all the interface names in the state entry.
2914                  */
2915                 for (i = 0; i < 4; i++) {
2916                         if (ifp == NULL || ifp == is->is_ifp[i])
2917                                 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i],
2918                                                               is->is_v);
2919                 }
2920         }
2921         RWLOCK_EXIT(&ipf_state);
2922 }
2923
2924
2925 /* ------------------------------------------------------------------------ */
2926 /* Function:    fr_delstate                                                 */
2927 /* Returns:     int - 0 = entry deleted, else reference count on struct     */
2928 /* Parameters:  is(I)  - pointer to state structure to delete               */
2929 /*              why(I) - if not 0, log reason why it was deleted            */
2930 /* Write Locks: ipf_state                                                   */
2931 /*                                                                          */
2932 /* Deletes a state entry from the enumerated list as well as the hash table */
2933 /* and timeout queue lists.  Make adjustments to hash table statistics and  */
2934 /* global counters as required.                                             */
2935 /* ------------------------------------------------------------------------ */
2936 static int fr_delstate(is, why)
2937 ipstate_t *is;
2938 int why;
2939 {
2940
2941         /*
2942          * Since we want to delete this, remove it from the state table,
2943          * where it can be found & used, first.
2944          */
2945         if (is->is_phnext != NULL) {
2946                 *is->is_phnext = is->is_hnext;
2947                 if (is->is_hnext != NULL)
2948                         is->is_hnext->is_phnext = is->is_phnext;
2949                 if (ips_table[is->is_hv] == NULL)
2950                         ips_stats.iss_inuse--;
2951                 ips_stats.iss_bucketlen[is->is_hv]--;
2952
2953                 is->is_phnext = NULL;
2954                 is->is_hnext = NULL;
2955         }
2956
2957         /*
2958          * Because ips_stats.iss_wild is a count of entries in the state
2959          * table that have wildcard flags set, only decerement it once
2960          * and do it here.
2961          */
2962         if (is->is_flags & (SI_WILDP|SI_WILDA)) {
2963                 if (!(is->is_flags & SI_CLONED)) {
2964                         ATOMIC_DECL(ips_stats.iss_wild);
2965                 }
2966                 is->is_flags &= ~(SI_WILDP|SI_WILDA);
2967         }
2968
2969         /*
2970          * Next, remove it from the timeout queue it is in.
2971          */
2972         if (is->is_sti.tqe_ifq != NULL)
2973                 fr_deletequeueentry(&is->is_sti);
2974
2975         if (is->is_me != NULL) {
2976                 *is->is_me = NULL;
2977                 is->is_me = NULL;
2978         }
2979
2980         /*
2981          * If it is still in use by something else, do not go any further,
2982          * but note that at this point it is now an orphan.  How can this
2983          * be?  fr_state_flush() calls fr_delete() directly because it wants
2984          * to empty the table out and if something has a hold on a state
2985          * entry (such as ipfstat), it'll do the deref path that'll bring
2986          * us back here to do the real delete & free.
2987          */
2988         MUTEX_ENTER(&is->is_lock);
2989         if (is->is_ref > 1) {
2990                 is->is_ref--;
2991                 MUTEX_EXIT(&is->is_lock);
2992                 return is->is_ref;
2993         }
2994         MUTEX_EXIT(&is->is_lock);
2995
2996         is->is_ref = 0;
2997
2998         if (is->is_tqehead[0] != NULL) {
2999                 if (fr_deletetimeoutqueue(is->is_tqehead[0]) == 0)
3000                         fr_freetimeoutqueue(is->is_tqehead[0]);
3001         }
3002         if (is->is_tqehead[1] != NULL) {
3003                 if (fr_deletetimeoutqueue(is->is_tqehead[1]) == 0)
3004                         fr_freetimeoutqueue(is->is_tqehead[1]);
3005         }
3006
3007 #ifdef  IPFILTER_SYNC
3008         if (is->is_sync)
3009                 ipfsync_del(is->is_sync);
3010 #endif
3011 #ifdef  IPFILTER_SCAN
3012         (void) ipsc_detachis(is);
3013 #endif
3014
3015         /*
3016          * Now remove it from the linked list of known states
3017          */
3018         if (is->is_pnext != NULL) {
3019                 *is->is_pnext = is->is_next;
3020
3021                 if (is->is_next != NULL)
3022                         is->is_next->is_pnext = is->is_pnext;
3023
3024                 is->is_pnext = NULL;
3025                 is->is_next = NULL;
3026         }
3027
3028         if (ipstate_logging != 0 && why != 0)
3029                 ipstate_log(is, why);
3030
3031         if (is->is_p == IPPROTO_TCP)
3032                 ips_stats.iss_fin++;
3033         else
3034                 ips_stats.iss_expire++;
3035
3036         if (is->is_rule != NULL) {
3037                 is->is_rule->fr_statecnt--;
3038                 (void) fr_derefrule(&is->is_rule);
3039         }
3040
3041 #if defined(NEED_LOCAL_RAND) && defined(_KERNEL)
3042         ipf_rand_push(is, sizeof(*is));
3043 #endif
3044
3045         MUTEX_DESTROY(&is->is_lock);
3046         KFREE(is);
3047         ips_num--;
3048
3049         return 0;
3050 }
3051
3052
3053 /* ------------------------------------------------------------------------ */
3054 /* Function:    fr_timeoutstate                                             */
3055 /* Returns:     Nil                                                         */
3056 /* Parameters:  Nil                                                         */
3057 /*                                                                          */
3058 /* Slowly expire held state for thingslike UDP and ICMP.  The algorithm     */
3059 /* used here is to keep the queue sorted with the oldest things at the top  */
3060 /* and the youngest at the bottom.  So if the top one doesn't need to be    */
3061 /* expired then neither will any under it.                                  */
3062 /* ------------------------------------------------------------------------ */
3063 void fr_timeoutstate()
3064 {
3065         ipftq_t *ifq, *ifqnext;
3066         ipftqent_t *tqe, *tqn;
3067         ipstate_t *is;
3068         SPL_INT(s);
3069
3070         SPL_NET(s);
3071         WRITE_ENTER(&ipf_state);
3072         for (ifq = ips_tqtqb; ifq != NULL; ifq = ifq->ifq_next)
3073                 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3074                         if (tqe->tqe_die > fr_ticks)
3075                                 break;
3076                         tqn = tqe->tqe_next;
3077                         is = tqe->tqe_parent;
3078                         fr_delstate(is, ISL_EXPIRE);
3079                 }
3080
3081         for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) {
3082                 ifqnext = ifq->ifq_next;
3083
3084                 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3085                         if (tqe->tqe_die > fr_ticks)
3086                                 break;
3087                         tqn = tqe->tqe_next;
3088                         is = tqe->tqe_parent;
3089                         fr_delstate(is, ISL_EXPIRE);
3090                 }
3091         }
3092
3093         for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) {
3094                 ifqnext = ifq->ifq_next;
3095
3096                 if (((ifq->ifq_flags & IFQF_DELETE) != 0) &&
3097                     (ifq->ifq_ref == 0)) {
3098                         fr_freetimeoutqueue(ifq);
3099                 }
3100         }
3101
3102         if (fr_state_doflush) {
3103                 (void) fr_state_flush(2, 0);
3104                 fr_state_doflush = 0;
3105         }
3106
3107         RWLOCK_EXIT(&ipf_state);
3108         SPL_X(s);
3109 }
3110
3111
3112 /* ------------------------------------------------------------------------ */
3113 /* Function:    fr_state_flush                                              */
3114 /* Returns:     int - 0 == success, -1 == failure                           */
3115 /* Parameters:  Nil                                                         */
3116 /* Write Locks: ipf_state                                                   */
3117 /*                                                                          */
3118 /* Flush state tables.  Three actions currently defined:                    */
3119 /* which == 0 : flush all state table entries                               */
3120 /* which == 1 : flush TCP connections which have started to close but are   */
3121 /*            stuck for some reason.                                        */
3122 /* which == 2 : flush TCP connections which have been idle for a long time, */
3123 /*            starting at > 4 days idle and working back in successive half-*/
3124 /*            days to at most 12 hours old.  If this fails to free enough   */
3125 /*            slots then work backwards in half hour slots to 30 minutes.   */
3126 /*            If that too fails, then work backwards in 30 second intervals */
3127 /*            for the last 30 minutes to at worst 30 seconds idle.          */
3128 /* ------------------------------------------------------------------------ */
3129 static int fr_state_flush(which, proto)
3130 int which, proto;
3131 {
3132         ipftq_t *ifq, *ifqnext;
3133         ipftqent_t *tqe, *tqn;
3134         ipstate_t *is, **isp;
3135         int removed;
3136         SPL_INT(s);
3137
3138         removed = 0;
3139
3140         SPL_NET(s);
3141
3142         switch (which)
3143         {
3144         case 0 :
3145                 /*
3146                  * Style 0 flush removes everything...
3147                  */
3148                 for (isp = &ips_list; ((is = *isp) != NULL); ) {
3149                         if ((proto != 0) && (is->is_v != proto)) {
3150                                 isp = &is->is_next;
3151                                 continue;
3152                         }
3153                         if (fr_delstate(is, ISL_FLUSH) == 0)
3154                                 removed++;
3155                         else
3156                                 isp = &is->is_next;
3157                 }
3158                 break;
3159
3160         case 1 :
3161                 /*
3162                  * Since we're only interested in things that are closing,
3163                  * we can start with the appropriate timeout queue.
3164                  */
3165                 for (ifq = ips_tqtqb + IPF_TCPS_CLOSE_WAIT; ifq != NULL;
3166                      ifq = ifq->ifq_next) {
3167
3168                         for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3169                                 tqn = tqe->tqe_next;
3170                                 is = tqe->tqe_parent;
3171                                 if (is->is_p != IPPROTO_TCP)
3172                                         break;
3173                                 if (fr_delstate(is, ISL_EXPIRE) == 0)
3174                                         removed++;
3175                         }
3176                 }
3177
3178                 /*
3179                  * Also need to look through the user defined queues.
3180                  */
3181                 for (ifq = ips_utqe; ifq != NULL; ifq = ifqnext) {
3182                         ifqnext = ifq->ifq_next;
3183                         for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) {
3184                                 tqn = tqe->tqe_next;
3185                                 is = tqe->tqe_parent;
3186                                 if (is->is_p != IPPROTO_TCP)
3187                                         continue;
3188
3189                                 if ((is->is_state[0] > IPF_TCPS_ESTABLISHED) &&
3190                                     (is->is_state[1] > IPF_TCPS_ESTABLISHED)) {
3191                                         if (fr_delstate(is, ISL_EXPIRE) == 0)
3192                                                 removed++;
3193                                 }
3194                         }
3195                 }
3196                 break;
3197
3198         case 2 :
3199                 break;
3200
3201                 /*
3202                  * Args 5-11 correspond to flushing those particular states
3203                  * for TCP connections.
3204                  */
3205         case IPF_TCPS_CLOSE_WAIT :
3206         case IPF_TCPS_FIN_WAIT_1 :
3207         case IPF_TCPS_CLOSING :
3208         case IPF_TCPS_LAST_ACK :
3209         case IPF_TCPS_FIN_WAIT_2 :
3210         case IPF_TCPS_TIME_WAIT :
3211         case IPF_TCPS_CLOSED :
3212                 tqn = ips_tqtqb[which].ifq_head;
3213                 while (tqn != NULL) {
3214                         tqe = tqn;
3215                         tqn = tqe->tqe_next;
3216                         is = tqe->tqe_parent;
3217                         if (fr_delstate(is, ISL_FLUSH) == 0)
3218                                 removed++;
3219                 }
3220                 break;
3221
3222         default :
3223                 if (which < 30)
3224                         break;
3225
3226                 /*
3227                  * Take a large arbitrary number to mean the number of seconds
3228                  * for which which consider to be the maximum value we'll allow
3229                  * the expiration to be.
3230                  */
3231                 which = IPF_TTLVAL(which);
3232                 for (isp = &ips_list; ((is = *isp) != NULL); ) {
3233                         if ((proto == 0) || (is->is_v == proto)) {
3234                                 if (fr_ticks - is->is_touched > which) {
3235                                         if (fr_delstate(is, ISL_FLUSH) == 0) {
3236                                                 removed++;
3237                                                 continue;
3238                                         }
3239                                 }
3240                         }
3241                         isp = &is->is_next;
3242                 }
3243                 break;
3244         }
3245
3246         if (which != 2) {
3247                 SPL_X(s);
3248                 return removed;
3249         }
3250
3251         /*
3252          * Asked to remove inactive entries because the table is full.
3253          */
3254         if (fr_ticks - ips_last_force_flush > IPF_TTLVAL(5)) {
3255                 ips_last_force_flush = fr_ticks;
3256                 removed = ipf_queueflush(fr_state_flush_entry, ips_tqtqb,
3257                                          ips_utqe);
3258         }
3259
3260         SPL_X(s);
3261         return removed;
3262 }
3263
3264
3265 /* ------------------------------------------------------------------------ */
3266 /* Function:    fr_state_flush_entry                                        */
3267 /* Returns:     int - 0 = entry deleted, else not deleted                   */
3268 /* Parameters:  entry(I)  - pointer to state structure to delete            */
3269 /* Write Locks: ipf_state                                                   */
3270 /*                                                                          */
3271 /* This function is a stepping stone between ipf_queueflush() and           */
3272 /* fr_delstate().  It is used so we can provide a uniform interface via the */
3273 /* ipf_queueflush() function.                                               */
3274 /* ------------------------------------------------------------------------ */
3275 static int fr_state_flush_entry(entry)
3276 void *entry;
3277 {
3278         return fr_delstate(entry, ISL_FLUSH);
3279 }
3280
3281
3282 /* ------------------------------------------------------------------------ */
3283 /* Function:    fr_tcp_age                                                  */
3284 /* Returns:     int - 1 == state transition made, 0 == no change (rejected) */
3285 /* Parameters:  tq(I)    - pointer to timeout queue information             */
3286 /*              fin(I)   - pointer to packet information                    */
3287 /*              tqtab(I) - TCP timeout queue table this is in               */
3288 /*              flags(I) - flags from state/NAT entry                       */
3289 /*                                                                          */
3290 /* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29:          */
3291 /*                                                                          */
3292 /* - (try to) base state transitions on real evidence only,                 */
3293 /*   i.e. packets that are sent and have been received by ipfilter;         */
3294 /*   diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used.       */
3295 /*                                                                          */
3296 /* - deal with half-closed connections correctly;                           */
3297 /*                                                                          */
3298 /* - store the state of the source in state[0] such that ipfstat            */
3299 /*   displays the state as source/dest instead of dest/source; the calls    */
3300 /*   to fr_tcp_age have been changed accordingly.                           */
3301 /*                                                                          */
3302 /* Internal Parameters:                                                     */
3303 /*                                                                          */
3304 /*    state[0] = state of source (host that initiated connection)           */
3305 /*    state[1] = state of dest   (host that accepted the connection)        */
3306 /*                                                                          */
3307 /*    dir == 0 : a packet from source to dest                               */
3308 /*    dir == 1 : a packet from dest to source                               */
3309 /*                                                                          */
3310 /* A typical procession for a connection is as follows:                     */
3311 /*                                                                          */
3312 /* +--------------+-------------------+                                     */
3313 /* | Side '0'     | Side '1'          |                                     */
3314 /* +--------------+-------------------+                                     */
3315 /* | 0 -> 1 (SYN) |                   |                                     */
3316 /* |              | 0 -> 2 (SYN-ACK)  |                                     */
3317 /* | 1 -> 3 (ACK) |                   |                                     */
3318 /* |              | 2 -> 4 (ACK-PUSH) |                                     */
3319 /* | 3 -> 4 (ACK) |                   |                                     */
3320 /* |   ...        |   ...             |                                     */
3321 /* |              | 4 -> 6 (FIN-ACK)  |                                     */
3322 /* | 4 -> 5 (ACK) |                   |                                     */
3323 /* |              | 6 -> 6 (ACK-PUSH) |                                     */
3324 /* | 5 -> 5 (ACK) |                   |                                     */
3325 /* | 5 -> 8 (FIN) |                   |                                     */
3326 /* |              | 6 -> 10 (ACK)     |                                     */
3327 /* +--------------+-------------------+                                     */
3328 /*                                                                          */
3329 /* Locking: it is assumed that the parent of the tqe structure is locked.   */
3330 /* ------------------------------------------------------------------------ */
3331 int fr_tcp_age(tqe, fin, tqtab, flags)
3332 ipftqent_t *tqe;
3333 fr_info_t *fin;
3334 ipftq_t *tqtab;
3335 int flags;
3336 {
3337         int dlen, ostate, nstate, rval, dir;
3338         u_char tcpflags;
3339         tcphdr_t *tcp;
3340
3341         tcp = fin->fin_dp;
3342
3343         rval = 0;
3344         dir = fin->fin_rev;
3345         tcpflags = tcp->th_flags;
3346         dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2);
3347
3348         if (tcpflags & TH_RST) {
3349                 if (!(tcpflags & TH_PUSH) && !dlen)
3350                         nstate = IPF_TCPS_CLOSED;
3351                 else
3352                         nstate = IPF_TCPS_CLOSE_WAIT;
3353                 rval = 1;
3354         } else {
3355                 ostate = tqe->tqe_state[1 - dir];
3356                 nstate = tqe->tqe_state[dir];
3357
3358                 switch (nstate)
3359                 {
3360                 case IPF_TCPS_LISTEN: /* 0 */
3361                         if ((tcpflags & TH_OPENING) == TH_OPENING) {
3362                                 /*
3363                                  * 'dir' received an S and sends SA in
3364                                  * response, LISTEN -> SYN_RECEIVED
3365                                  */
3366                                 nstate = IPF_TCPS_SYN_RECEIVED;
3367                                 rval = 1;
3368                         } else if ((tcpflags & TH_OPENING) == TH_SYN) {
3369                                 /* 'dir' sent S, LISTEN -> SYN_SENT */
3370                                 nstate = IPF_TCPS_SYN_SENT;
3371                                 rval = 1;
3372                         }
3373                         /*
3374                          * the next piece of code makes it possible to get
3375                          * already established connections into the state table
3376                          * after a restart or reload of the filter rules; this
3377                          * does not work when a strict 'flags S keep state' is
3378                          * used for tcp connections of course
3379                          */
3380                         if (((flags & IS_TCPFSM) == 0) &&
3381                             ((tcpflags & TH_ACKMASK) == TH_ACK)) {
3382                                 /*
3383                                  * we saw an A, guess 'dir' is in ESTABLISHED
3384                                  * mode
3385                                  */
3386                                 switch (ostate)
3387                                 {
3388                                 case IPF_TCPS_LISTEN :
3389                                 case IPF_TCPS_SYN_RECEIVED :
3390                                         nstate = IPF_TCPS_HALF_ESTAB;
3391                                         rval = 1;
3392                                         break;
3393                                 case IPF_TCPS_HALF_ESTAB :
3394                                 case IPF_TCPS_ESTABLISHED :
3395                                         nstate = IPF_TCPS_ESTABLISHED;
3396                                         rval = 1;
3397                                         break;
3398                                 default :
3399                                         break;
3400                                 }
3401                         }
3402                         /*
3403                          * TODO: besides regular ACK packets we can have other
3404                          * packets as well; it is yet to be determined how we
3405                          * should initialize the states in those cases
3406                          */
3407                         break;
3408
3409                 case IPF_TCPS_SYN_SENT: /* 1 */
3410                         if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) {
3411                                 /*
3412                                  * A retransmitted SYN packet.  We do not reset
3413                                  * the timeout here to fr_tcptimeout because a
3414                                  * connection connect timeout does not renew
3415                                  * after every packet that is sent.  We need to
3416                                  * set rval so as to indicate the packet has
3417                                  * passed the check for its flags being valid
3418                                  * in the TCP FSM.  Setting rval to 2 has the
3419                                  * result of not resetting the timeout.
3420                                  */
3421                                 rval = 2;
3422                         } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) ==
3423                                    TH_ACK) {
3424                                 /*
3425                                  * we see an A from 'dir' which is in SYN_SENT
3426                                  * state: 'dir' sent an A in response to an SA
3427                                  * which it received, SYN_SENT -> ESTABLISHED
3428                                  */
3429                                 nstate = IPF_TCPS_ESTABLISHED;
3430                                 rval = 1;
3431                         } else if (tcpflags & TH_FIN) {
3432                                 /*
3433                                  * we see an F from 'dir' which is in SYN_SENT
3434                                  * state and wants to close its side of the
3435                                  * connection; SYN_SENT -> FIN_WAIT_1
3436                                  */
3437                                 nstate = IPF_TCPS_FIN_WAIT_1;
3438                                 rval = 1;
3439                         } else if ((tcpflags & TH_OPENING) == TH_OPENING) {
3440                                 /*
3441                                  * we see an SA from 'dir' which is already in
3442                                  * SYN_SENT state, this means we have a
3443                                  * simultaneous open; SYN_SENT -> SYN_RECEIVED
3444                                  */
3445                                 nstate = IPF_TCPS_SYN_RECEIVED;
3446                                 rval = 1;
3447                         }
3448                         break;
3449
3450                 case IPF_TCPS_SYN_RECEIVED: /* 2 */
3451                         if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) {
3452                                 /*
3453                                  * we see an A from 'dir' which was in
3454                                  * SYN_RECEIVED state so it must now be in
3455                                  * established state, SYN_RECEIVED ->
3456                                  * ESTABLISHED
3457                                  */
3458                                 nstate = IPF_TCPS_ESTABLISHED;
3459                                 rval = 1;
3460                         } else if ((tcpflags & ~(TH_ECN|TH_CWR)) ==
3461                                    TH_OPENING) {
3462                                 /*
3463                                  * We see an SA from 'dir' which is already in
3464                                  * SYN_RECEIVED state.
3465                                  */
3466                                 rval = 2;
3467                         } else if (tcpflags & TH_FIN) {
3468                                 /*
3469                                  * we see an F from 'dir' which is in
3470                                  * SYN_RECEIVED state and wants to close its
3471                                  * side of the connection; SYN_RECEIVED ->
3472                                  * FIN_WAIT_1
3473                                  */
3474                                 nstate = IPF_TCPS_FIN_WAIT_1;
3475                                 rval = 1;
3476                         }
3477                         break;
3478
3479                 case IPF_TCPS_HALF_ESTAB: /* 3 */
3480                         if (tcpflags & TH_FIN) {
3481                                 nstate = IPF_TCPS_FIN_WAIT_1;
3482                                 rval = 1;
3483                         } else if ((tcpflags & TH_ACKMASK) == TH_ACK) {
3484                                 /*
3485                                  * If we've picked up a connection in mid
3486                                  * flight, we could be looking at a follow on
3487                                  * packet from the same direction as the one
3488                                  * that created this state.  Recognise it but
3489                                  * do not advance the entire connection's
3490                                  * state.
3491                                  */
3492                                 switch (ostate)
3493                                 {
3494                                 case IPF_TCPS_LISTEN :
3495                                 case IPF_TCPS_SYN_SENT :
3496                                 case IPF_TCPS_SYN_RECEIVED :
3497                                         rval = 1;
3498                                         break;
3499                                 case IPF_TCPS_HALF_ESTAB :
3500                                 case IPF_TCPS_ESTABLISHED :
3501                                         nstate = IPF_TCPS_ESTABLISHED;
3502                                         rval = 1;
3503                                         break;
3504                                 default :
3505                                         break;
3506                                 }
3507                         }
3508                         break;
3509
3510                 case IPF_TCPS_ESTABLISHED: /* 4 */
3511                         rval = 1;
3512                         if (tcpflags & TH_FIN) {
3513                                 /*
3514                                  * 'dir' closed its side of the connection;
3515                                  * this gives us a half-closed connection;
3516                                  * ESTABLISHED -> FIN_WAIT_1
3517                                  */
3518                                 if (ostate == IPF_TCPS_FIN_WAIT_1) {
3519                                         nstate = IPF_TCPS_CLOSING;
3520                                 } else {
3521                                         nstate = IPF_TCPS_FIN_WAIT_1;
3522                                 }
3523                         } else if (tcpflags & TH_ACK) {
3524                                 /*
3525                                  * an ACK, should we exclude other flags here?
3526                                  */
3527                                 if (ostate == IPF_TCPS_FIN_WAIT_1) {
3528                                         /*
3529                                          * We know the other side did an active
3530                                          * close, so we are ACKing the recvd
3531                                          * FIN packet (does the window matching
3532                                          * code guarantee this?) and go into
3533                                          * CLOSE_WAIT state; this gives us a
3534                                          * half-closed connection
3535                                          */
3536                                         nstate = IPF_TCPS_CLOSE_WAIT;
3537                                 } else if (ostate < IPF_TCPS_CLOSE_WAIT) {
3538                                         /*
3539                                          * still a fully established
3540                                          * connection reset timeout
3541                                          */
3542                                         nstate = IPF_TCPS_ESTABLISHED;
3543                                 }
3544                         }
3545                         break;
3546
3547                 case IPF_TCPS_CLOSE_WAIT: /* 5 */
3548                         rval = 1;
3549                         if (tcpflags & TH_FIN) {
3550                                 /*
3551                                  * application closed and 'dir' sent a FIN,
3552                                  * we're now going into LAST_ACK state
3553                                  */
3554                                 nstate = IPF_TCPS_LAST_ACK;
3555                         } else {
3556                                 /*
3557                                  * we remain in CLOSE_WAIT because the other
3558                                  * side has closed already and we did not
3559                                  * close our side yet; reset timeout
3560                                  */
3561                                 nstate = IPF_TCPS_CLOSE_WAIT;
3562                         }
3563                         break;
3564
3565                 case IPF_TCPS_FIN_WAIT_1: /* 6 */
3566                         rval = 1;
3567                         if ((tcpflags & TH_ACK) &&
3568                             ostate > IPF_TCPS_CLOSE_WAIT) {
3569                                 /*
3570                                  * if the other side is not active anymore
3571                                  * it has sent us a FIN packet that we are
3572                                  * ack'ing now with an ACK; this means both
3573                                  * sides have now closed the connection and
3574                                  * we go into TIME_WAIT
3575                                  */
3576                                 /*
3577                                  * XXX: how do we know we really are ACKing
3578                                  * the FIN packet here? does the window code
3579                                  * guarantee that?
3580                                  */
3581                                 nstate = IPF_TCPS_TIME_WAIT;
3582                         } else {
3583                                 /*
3584                                  * we closed our side of the connection
3585                                  * already but the other side is still active
3586                                  * (ESTABLISHED/CLOSE_WAIT); continue with
3587                                  * this half-closed connection
3588                                  */
3589                                 nstate = IPF_TCPS_FIN_WAIT_1;
3590                         }
3591                         break;
3592
3593                 case IPF_TCPS_CLOSING: /* 7 */
3594                         if ((tcpflags & (TH_FIN|TH_ACK)) == TH_ACK) {
3595                                 nstate = IPF_TCPS_TIME_WAIT;
3596                         }
3597                         rval = 2;
3598                         break;
3599
3600                 case IPF_TCPS_LAST_ACK: /* 8 */
3601                         if (tcpflags & TH_ACK) {
3602                                 if ((tcpflags & TH_PUSH) || dlen)
3603                                         /*
3604                                          * there is still data to be delivered,
3605                                          * reset timeout
3606                                          */
3607                                         rval = 1;
3608                                 else
3609                                         rval = 2;
3610                         }
3611                         /*
3612                          * we cannot detect when we go out of LAST_ACK state to
3613                          * CLOSED because that is based on the reception of ACK
3614                          * packets; ipfilter can only detect that a packet
3615                          * has been sent by a host
3616                          */
3617                         break;
3618
3619                 case IPF_TCPS_FIN_WAIT_2: /* 9 */
3620                         /* NOT USED */
3621                         break;
3622
3623                 case IPF_TCPS_TIME_WAIT: /* 10 */
3624                         /* we're in 2MSL timeout now */
3625                         if (ostate == IPF_TCPS_LAST_ACK) {
3626                                 nstate = IPF_TCPS_CLOSED;
3627                         }
3628                         rval = 1;
3629                         break;
3630
3631                 case IPF_TCPS_CLOSED: /* 11 */
3632                         rval = 2;
3633                         break;
3634
3635                 default :
3636 #if defined(_KERNEL)
3637 # if SOLARIS
3638                         cmn_err(CE_NOTE,
3639                                 "tcp %lx flags %x si %lx nstate %d ostate %d\n",
3640                                 (u_long)tcp, tcpflags, (u_long)tqe,
3641                                 nstate, ostate);
3642 # else
3643                         printf("tcp %lx flags %x si %lx nstate %d ostate %d\n",
3644                                 (u_long)tcp, tcpflags, (u_long)tqe,
3645                                 nstate, ostate);
3646 # endif
3647 #else
3648                         abort();
3649 #endif
3650                         break;
3651                 }
3652         }
3653
3654         /*
3655          * If rval == 2 then do not update the queue position, but treat the
3656          * packet as being ok.
3657          */
3658         if (rval == 2)
3659                 rval = 1;
3660         else if (rval == 1) {
3661                 tqe->tqe_state[dir] = nstate;
3662                 if ((tqe->tqe_flags & TQE_RULEBASED) == 0)
3663                         fr_movequeue(tqe, tqe->tqe_ifq, tqtab + nstate);
3664         }
3665
3666         return rval;
3667 }
3668
3669
3670 /* ------------------------------------------------------------------------ */
3671 /* Function:    ipstate_log                                                 */
3672 /* Returns:     Nil                                                         */
3673 /* Parameters:  is(I)   - pointer to state structure                        */
3674 /*              type(I) - type of log entry to create                       */
3675 /*                                                                          */
3676 /* Creates a state table log entry using the state structure and type info. */
3677 /* passed in.  Log packet/byte counts, source/destination address and other */
3678 /* protocol specific information.                                           */
3679 /* ------------------------------------------------------------------------ */
3680 void ipstate_log(is, type)
3681 struct ipstate *is;
3682 u_int type;
3683 {
3684 #ifdef  IPFILTER_LOG
3685         struct  ipslog  ipsl;
3686         size_t sizes[1];
3687         void *items[1];
3688         int types[1];
3689
3690         /*
3691          * Copy information out of the ipstate_t structure and into the
3692          * structure used for logging.
3693          */
3694         ipsl.isl_type = type;
3695         ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0];
3696         ipsl.isl_bytes[0] = is->is_bytes[0];
3697         ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1];
3698         ipsl.isl_bytes[1] = is->is_bytes[1];
3699         ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2];
3700         ipsl.isl_bytes[2] = is->is_bytes[2];
3701         ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3];
3702         ipsl.isl_bytes[3] = is->is_bytes[3];
3703         ipsl.isl_src = is->is_src;
3704         ipsl.isl_dst = is->is_dst;
3705         ipsl.isl_p = is->is_p;
3706         ipsl.isl_v = is->is_v;
3707         ipsl.isl_flags = is->is_flags;
3708         ipsl.isl_tag = is->is_tag;
3709         ipsl.isl_rulen = is->is_rulen;
3710         (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN);
3711
3712         if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) {
3713                 ipsl.isl_sport = is->is_sport;
3714                 ipsl.isl_dport = is->is_dport;
3715                 if (ipsl.isl_p == IPPROTO_TCP) {
3716                         ipsl.isl_state[0] = is->is_state[0];
3717                         ipsl.isl_state[1] = is->is_state[1];
3718                 }
3719         } else if (ipsl.isl_p == IPPROTO_ICMP) {
3720                 ipsl.isl_itype = is->is_icmp.ici_type;
3721         } else if (ipsl.isl_p == IPPROTO_ICMPV6) {
3722                 ipsl.isl_itype = is->is_icmp.ici_type;
3723         } else {
3724                 ipsl.isl_ps.isl_filler[0] = 0;
3725                 ipsl.isl_ps.isl_filler[1] = 0;
3726         }
3727
3728         items[0] = &ipsl;
3729         sizes[0] = sizeof(ipsl);
3730         types[0] = 0;
3731
3732         if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1)) {
3733                 ATOMIC_INCL(ips_stats.iss_logged);
3734         } else {
3735                 ATOMIC_INCL(ips_stats.iss_logfail);
3736         }
3737 #endif
3738 }
3739
3740
3741 #ifdef  USE_INET6
3742 /* ------------------------------------------------------------------------ */
3743 /* Function:    fr_checkicmp6matchingstate                                  */
3744 /* Returns:     ipstate_t* - NULL == no match found,                        */
3745 /*                           else  pointer to matching state entry          */
3746 /* Parameters:  fin(I) - pointer to packet information                      */
3747 /* Locks:       NULL == no locks, else Read Lock on ipf_state               */
3748 /*                                                                          */
3749 /* If we've got an ICMPv6 error message, using the information stored in    */
3750 /* the ICMPv6 packet, look for a matching state table entry.                */
3751 /* ------------------------------------------------------------------------ */
3752 static ipstate_t *fr_checkicmp6matchingstate(fin)
3753 fr_info_t *fin;
3754 {
3755         struct icmp6_hdr *ic6, *oic;
3756         int type, backward, i;
3757         ipstate_t *is, **isp;
3758         u_short sport, dport;
3759         i6addr_t dst, src;
3760         u_short savelen;
3761         icmpinfo_t *ic;
3762         fr_info_t ofin;
3763         tcphdr_t *tcp;
3764         ip6_t *oip6;
3765         u_char  pr;
3766         u_int hv;
3767
3768         /*
3769          * Does it at least have the return (basic) IP header ?
3770          * Is it an actual recognised ICMP error type?
3771          * Only a basic IP header (no options) should be with
3772          * an ICMP error header.
3773          */
3774         if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) ||
3775             !(fin->fin_flx & FI_ICMPERR))
3776                 return NULL;
3777
3778         ic6 = fin->fin_dp;
3779         type = ic6->icmp6_type;
3780
3781         oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN);
3782         if (fin->fin_plen < sizeof(*oip6))
3783                 return NULL;
3784
3785         bcopy((char *)fin, (char *)&ofin, sizeof(*fin));
3786         ofin.fin_v = 6;
3787         ofin.fin_ifp = fin->fin_ifp;
3788         ofin.fin_out = !fin->fin_out;
3789         ofin.fin_m = NULL;      /* if dereferenced, panic XXX */
3790         ofin.fin_mp = NULL;     /* if dereferenced, panic XXX */
3791
3792         /*
3793          * We make a fin entry to be able to feed it to
3794          * matchsrcdst. Note that not all fields are necessary
3795          * but this is the cleanest way. Note further we fill
3796          * in fin_mp such that if someone uses it we'll get
3797          * a kernel panic. fr_matchsrcdst does not use this.
3798          *
3799          * watch out here, as ip is in host order and oip6 in network
3800          * order. Any change we make must be undone afterwards.
3801          */
3802         savelen = oip6->ip6_plen;
3803         oip6->ip6_plen = fin->fin_dlen - ICMPERR_ICMPHLEN;
3804         ofin.fin_flx = FI_NOCKSUM;
3805         ofin.fin_ip = (ip_t *)oip6;
3806         (void) fr_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin);
3807         ofin.fin_flx &= ~(FI_BAD|FI_SHORT);
3808         oip6->ip6_plen = savelen;
3809
3810         if (oip6->ip6_nxt == IPPROTO_ICMPV6) {
3811                 oic = (struct icmp6_hdr *)(oip6 + 1);
3812                 /*
3813                  * an ICMP error can only be generated as a result of an
3814                  * ICMP query, not as the response on an ICMP error
3815                  *
3816                  * XXX theoretically ICMP_ECHOREP and the other reply's are
3817                  * ICMP query's as well, but adding them here seems strange XXX
3818                  */
3819                  if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK))
3820                         return NULL;
3821
3822                 /*
3823                  * perform a lookup of the ICMP packet in the state table
3824                  */
3825                 hv = (pr = oip6->ip6_nxt);
3826                 src.in6 = oip6->ip6_src;
3827                 hv += src.in4.s_addr;
3828                 dst.in6 = oip6->ip6_dst;
3829                 hv += dst.in4.s_addr;
3830                 hv += oic->icmp6_id;
3831                 hv += oic->icmp6_seq;
3832                 hv = DOUBLE_HASH(hv);
3833
3834                 READ_ENTER(&ipf_state);
3835                 for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) {
3836                         ic = &is->is_icmp;
3837                         isp = &is->is_hnext;
3838                         if ((is->is_p == pr) &&
3839                             !(is->is_pass & FR_NOICMPERR) &&
3840                             (oic->icmp6_id == ic->ici_id) &&
3841                             (oic->icmp6_seq == ic->ici_seq) &&
3842                             (is = fr_matchsrcdst(&ofin, is, &src,
3843                                                  &dst, NULL, FI_ICMPCMP))) {
3844                                 /*
3845                                  * in the state table ICMP query's are stored
3846                                  * with the type of the corresponding ICMP
3847                                  * response. Correct here
3848                                  */
3849                                 if (((ic->ici_type == ICMP6_ECHO_REPLY) &&
3850                                      (oic->icmp6_type == ICMP6_ECHO_REQUEST)) ||
3851                                      (ic->ici_type - 1 == oic->icmp6_type )) {
3852                                         ips_stats.iss_hits++;
3853                                         backward = IP6_NEQ(&is->is_dst, &src);
3854                                         fin->fin_rev = !backward;
3855                                         i = (backward << 1) + fin->fin_out;
3856                                         is->is_icmppkts[i]++;
3857                                         return is;
3858                                 }
3859                         }
3860                 }
3861                 RWLOCK_EXIT(&ipf_state);
3862                 return NULL;
3863         }
3864
3865         hv = (pr = oip6->ip6_nxt);
3866         src.in6 = oip6->ip6_src;
3867         hv += src.i6[0];
3868         hv += src.i6[1];
3869         hv += src.i6[2];
3870         hv += src.i6[3];
3871         dst.in6 = oip6->ip6_dst;
3872         hv += dst.i6[0];
3873         hv += dst.i6[1];
3874         hv += dst.i6[2];
3875         hv += dst.i6[3];
3876
3877         if ((oip6->ip6_nxt == IPPROTO_TCP) || (oip6->ip6_nxt == IPPROTO_UDP)) {
3878                 tcp = (tcphdr_t *)(oip6 + 1);
3879                 dport = tcp->th_dport;
3880                 sport = tcp->th_sport;
3881                 hv += dport;
3882                 hv += sport;
3883         } else
3884                 tcp = NULL;
3885         hv = DOUBLE_HASH(hv);
3886
3887         READ_ENTER(&ipf_state);
3888         for (isp = &ips_table[hv]; ((is = *isp) != NULL); ) {
3889                 isp = &is->is_hnext;
3890                 /*
3891                  * Only allow this icmp though if the
3892                  * encapsulated packet was allowed through the
3893                  * other way around. Note that the minimal amount
3894                  * of info present does not allow for checking against
3895                  * tcp internals such as seq and ack numbers.
3896                  */
3897                 if ((is->is_p != pr) || (is->is_v != 6) ||
3898                     (is->is_pass & FR_NOICMPERR))
3899                         continue;
3900                 is = fr_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP);
3901                 if (is != NULL) {
3902                         ips_stats.iss_hits++;
3903                         backward = IP6_NEQ(&is->is_dst, &src);
3904                         fin->fin_rev = !backward;
3905                         i = (backward << 1) + fin->fin_out;
3906                         is->is_icmppkts[i]++;
3907                         /*
3908                          * we deliberately do not touch the timeouts
3909                          * for the accompanying state table entry.
3910                          * It remains to be seen if that is correct. XXX
3911                          */
3912                         return is;
3913                 }
3914         }
3915         RWLOCK_EXIT(&ipf_state);
3916         return NULL;
3917 }
3918 #endif
3919
3920
3921 /* ------------------------------------------------------------------------ */
3922 /* Function:    fr_sttab_init                                               */
3923 /* Returns:     Nil                                                         */
3924 /* Parameters:  tqp(I) - pointer to an array of timeout queues for TCP      */
3925 /*                                                                          */
3926 /* Initialise the array of timeout queues for TCP.                          */
3927 /* ------------------------------------------------------------------------ */
3928 void fr_sttab_init(tqp)
3929 ipftq_t *tqp;
3930 {
3931         int i;
3932
3933         for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) {
3934                 tqp[i].ifq_ttl = 0;
3935                 tqp[i].ifq_ref = 1;
3936                 tqp[i].ifq_head = NULL;
3937                 tqp[i].ifq_tail = &tqp[i].ifq_head;
3938                 tqp[i].ifq_next = tqp + i + 1;
3939                 MUTEX_INIT(&tqp[i].ifq_lock, "ipftq tcp tab");
3940         }
3941         tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL;
3942         tqp[IPF_TCPS_CLOSED].ifq_ttl = fr_tcpclosed;
3943         tqp[IPF_TCPS_LISTEN].ifq_ttl = fr_tcptimeout;
3944         tqp[IPF_TCPS_SYN_SENT].ifq_ttl = fr_tcptimeout;
3945         tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = fr_tcptimeout;
3946         tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = fr_tcpidletimeout;
3947         tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = fr_tcphalfclosed;
3948         tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = fr_tcphalfclosed;
3949         tqp[IPF_TCPS_CLOSING].ifq_ttl = fr_tcptimeout;
3950         tqp[IPF_TCPS_LAST_ACK].ifq_ttl = fr_tcplastack;
3951         tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = fr_tcpclosewait;
3952         tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = fr_tcptimewait;
3953         tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = fr_tcptimeout;
3954 }
3955
3956
3957 /* ------------------------------------------------------------------------ */
3958 /* Function:    fr_sttab_destroy                                            */
3959 /* Returns:     Nil                                                         */
3960 /* Parameters:  tqp(I) - pointer to an array of timeout queues for TCP      */
3961 /*                                                                          */
3962 /* Do whatever is necessary to "destroy" each of the entries in the array   */
3963 /* of timeout queues for TCP.                                               */
3964 /* ------------------------------------------------------------------------ */
3965 void fr_sttab_destroy(tqp)
3966 ipftq_t *tqp;
3967 {
3968         int i;
3969
3970         for (i = IPF_TCP_NSTATES - 1; i >= 0; i--)
3971                 MUTEX_DESTROY(&tqp[i].ifq_lock);
3972 }
3973
3974
3975 /* ------------------------------------------------------------------------ */
3976 /* Function:    fr_statederef                                               */
3977 /* Returns:     Nil                                                         */
3978 /* Parameters:  isp(I) - pointer to pointer to state table entry            */
3979 /*                                                                          */
3980 /* Decrement the reference counter for this state table entry and free it   */
3981 /* if there are no more things using it.                                    */
3982 /*                                                                          */
3983 /* This function is only called when cleaning up after increasing is_ref by */
3984 /* one earlier in the 'code path' so if is_ref is 1 when entering, we do    */
3985 /* have an orphan, otherwise not.  However there is a possible race between */
3986 /* the entry being deleted via flushing with an ioctl call (that calls the  */
3987 /* delete function directly) and the tail end of packet processing so we    */
3988 /* need to grab is_lock before doing the check to synchronise the two code  */
3989 /* paths.                                                                   */
3990 /*                                                                          */
3991 /* When operating in userland (ipftest), we have no timers to clear a state */
3992 /* entry.  Therefore, we make a few simple tests before deleting an entry   */
3993 /* outright.  We compare states on each side looking for a combination of   */
3994 /* TIME_WAIT (should really be FIN_WAIT_2?) and LAST_ACK.  Then we factor   */
3995 /* in packet direction with the interface list to make sure we don't        */
3996 /* prematurely delete an entry on a final inbound packet that's we're also  */
3997 /* supposed to route elsewhere.                                             */
3998 /*                                                                          */
3999 /* Internal parameters:                                                     */
4000 /*    state[0] = state of source (host that initiated connection)           */
4001 /*    state[1] = state of dest   (host that accepted the connection)        */
4002 /*                                                                          */
4003 /*    dir == 0 : a packet from source to dest                               */
4004 /*    dir == 1 : a packet from dest to source                               */
4005 /* ------------------------------------------------------------------------ */
4006 void fr_statederef(isp)
4007 ipstate_t **isp;
4008 {
4009         ipstate_t *is;
4010
4011         is = *isp;
4012         *isp = NULL;
4013
4014         MUTEX_ENTER(&is->is_lock);
4015         if (is->is_ref > 1) {
4016                 is->is_ref--;
4017                 MUTEX_EXIT(&is->is_lock);
4018 #ifndef _KERNEL
4019                 if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) ||
4020                    (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) {
4021                         fr_delstate(is, ISL_ORPHAN);
4022                 }
4023 #endif
4024                 return;
4025         }
4026         MUTEX_EXIT(&is->is_lock);
4027
4028         WRITE_ENTER(&ipf_state);
4029         fr_delstate(is, ISL_EXPIRE);
4030         RWLOCK_EXIT(&ipf_state);
4031 }
4032
4033
4034 /* ------------------------------------------------------------------------ */
4035 /* Function:    fr_setstatequeue                                            */
4036 /* Returns:     Nil                                                         */
4037 /* Parameters:  is(I) - pointer to state structure                          */
4038 /*              rev(I) - forward(0) or reverse(1) direction                 */
4039 /* Locks:       ipf_state (read or write)                                   */
4040 /*                                                                          */
4041 /* Put the state entry on its default queue entry, using rev as a helped in */
4042 /* determining which queue it should be placed on.                          */
4043 /* ------------------------------------------------------------------------ */
4044 void fr_setstatequeue(is, rev)
4045 ipstate_t *is;
4046 int rev;
4047 {
4048         ipftq_t *oifq, *nifq;
4049
4050
4051         if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0)
4052                 nifq = is->is_tqehead[rev];
4053         else
4054                 nifq = NULL;
4055
4056         if (nifq == NULL) {
4057                 switch (is->is_p)
4058                 {
4059 #ifdef USE_INET6
4060                 case IPPROTO_ICMPV6 :
4061                         if (rev == 1)
4062                                 nifq = &ips_icmpacktq;
4063                         else
4064                                 nifq = &ips_icmptq;
4065                         break;
4066 #endif
4067                 case IPPROTO_ICMP :
4068                         if (rev == 1)
4069                                 nifq = &ips_icmpacktq;
4070                         else
4071                                 nifq = &ips_icmptq;
4072                         break;
4073                 case IPPROTO_TCP :
4074                         nifq = ips_tqtqb + is->is_state[rev];
4075                         break;
4076
4077                 case IPPROTO_UDP :
4078                         if (rev == 1)
4079                                 nifq = &ips_udpacktq;
4080                         else
4081                                 nifq = &ips_udptq;
4082                         break;
4083
4084                 default :
4085                         nifq = &ips_iptq;
4086                         break;
4087                 }
4088         }
4089
4090         oifq = is->is_sti.tqe_ifq;
4091         /*
4092          * If it's currently on a timeout queue, move it from one queue to
4093          * another, else put it on the end of the newly determined queue.
4094          */
4095         if (oifq != NULL)
4096                 fr_movequeue(&is->is_sti, oifq, nifq);
4097         else
4098                 fr_queueappend(&is->is_sti, nifq, is);
4099         return;
4100 }
4101
4102
4103 /* ------------------------------------------------------------------------ */
4104 /* Function:    fr_stateiter                                                */
4105 /* Returns:     int - 0 == success, else error                              */
4106 /* Parameters:  token(I) - pointer to ipftoken structure                    */
4107 /*              itp(I)   - pointer to ipfgeniter structure                  */
4108 /*                                                                          */
4109 /* This function handles the SIOCGENITER ioctl for the state tables and     */
4110 /* walks through the list of entries in the state table list (ips_list.)    */
4111 /* ------------------------------------------------------------------------ */
4112 static int fr_stateiter(token, itp)
4113 ipftoken_t *token;
4114 ipfgeniter_t *itp;
4115 {
4116         ipstate_t *is, *next, zero;
4117         int error, count;
4118         char *dst;
4119
4120         if (itp->igi_data == NULL)
4121                 return EFAULT;
4122
4123         if (itp->igi_nitems < 1)
4124                 return ENOSPC;
4125
4126         if (itp->igi_type != IPFGENITER_STATE)
4127                 return EINVAL;
4128
4129         is = token->ipt_data;
4130         if (is == (void *)-1) {
4131                 ipf_freetoken(token);
4132                 return ESRCH;
4133         }
4134
4135         error = 0;
4136         dst = itp->igi_data;
4137
4138         READ_ENTER(&ipf_state);
4139         if (is == NULL) {
4140                 next = ips_list;
4141         } else {
4142                 next = is->is_next;
4143         }
4144
4145         count = itp->igi_nitems;
4146         for (;;) {
4147                 if (next != NULL) {
4148                         /*
4149                          * If we find a state entry to use, bump its
4150                          * reference count so that it can be used for
4151                          * is_next when we come back.
4152                          */
4153                         if (count == 1) {
4154                                 MUTEX_ENTER(&next->is_lock);
4155                                 next->is_ref++;
4156                                 MUTEX_EXIT(&next->is_lock);
4157                                 token->ipt_data = next;
4158                         }
4159                 } else {
4160                         bzero(&zero, sizeof(zero));
4161                         next = &zero;
4162                         count = 1;
4163                         token->ipt_data = NULL;
4164                 }
4165                 RWLOCK_EXIT(&ipf_state);
4166
4167                 /*
4168                  * This should arguably be via fr_outobj() so that the state
4169                  * structure can (if required) be massaged going out.
4170                  */
4171                 error = COPYOUT(next, dst, sizeof(*next));
4172                 if (error != 0)
4173                         error = EFAULT;
4174                 if ((count == 1) || (error != 0))
4175                         break;
4176
4177                 dst += sizeof(*next);
4178                 count--;
4179
4180                 READ_ENTER(&ipf_state);
4181                 next = next->is_next;
4182         }
4183
4184         if (is != NULL) {
4185                 fr_statederef(&is);
4186         }
4187
4188         return error;
4189 }
4190
4191
4192 /* ------------------------------------------------------------------------ */
4193 /* Function:    fr_stgettable                                               */
4194 /* Returns:     int     - 0 = success, else error                           */
4195 /* Parameters:  data(I) - pointer to ioctl data                             */
4196 /*                                                                          */
4197 /* This function handles ioctl requests for tables of state information.    */
4198 /* At present the only table it deals with is the hash bucket statistics.   */
4199 /* ------------------------------------------------------------------------ */
4200 static int fr_stgettable(data)
4201 char *data;
4202 {
4203         ipftable_t table;
4204         int error;
4205
4206         error = fr_inobj(data, &table, IPFOBJ_GTABLE);
4207         if (error != 0)
4208                 return error;
4209
4210         if (table.ita_type != IPFTABLE_BUCKETS)
4211                 return EINVAL;
4212
4213         error = COPYOUT(ips_stats.iss_bucketlen, table.ita_table,
4214                         fr_statesize * sizeof(u_long));
4215         if (error != 0)
4216                 error = EFAULT;
4217         return error;
4218 }