1 /* 2 * Copyright (C) 1995-2003 by Darren Reed. 3 * 4 * See the IPFILTER.LICENCE file for details on licencing. 5 * 6 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 7 * Use is subject to license terms. 8 */ 9 10 #pragma ident "%Z%%M% %I% %E% SMI" 11 12 #if defined(KERNEL) || defined(_KERNEL) 13 # undef KERNEL 14 # undef _KERNEL 15 # define KERNEL 1 16 # define _KERNEL 1 17 #endif 18 #include <sys/errno.h> 19 #include <sys/types.h> 20 #include <sys/param.h> 21 #include <sys/file.h> 22 #if defined(__NetBSD__) && (NetBSD >= 199905) && !defined(IPFILTER_LKM) && \ 23 defined(_KERNEL) 24 # include "opt_ipfilter_log.h" 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 #include "netinet/ipf_stack.h" 96 #ifdef IPFILTER_SYNC 97 #include "netinet/ip_sync.h" 98 #endif 99 #ifdef IPFILTER_SCAN 100 #include "netinet/ip_scan.h" 101 #endif 102 #ifdef USE_INET6 103 #include <netinet/icmp6.h> 104 #endif 105 #if (__FreeBSD_version >= 300000) 106 # include <sys/malloc.h> 107 # if defined(_KERNEL) && !defined(IPFILTER_LKM) 108 # include <sys/libkern.h> 109 # include <sys/systm.h> 110 # endif 111 #endif 112 /* END OF INCLUDES */ 113 114 115 #if !defined(lint) 116 static const char sccsid[] = "@(#)ip_state.c 1.8 6/5/96 (C) 1993-2000 Darren Reed"; 117 static const char rcsid[] = "@(#)$Id: ip_state.c,v 2.186.2.36 2005/08/11 19:58:03 darrenr Exp $"; 118 #endif 119 120 #ifdef USE_INET6 121 static ipstate_t *fr_checkicmp6matchingstate __P((fr_info_t *)); 122 #endif 123 static ipstate_t *fr_matchsrcdst __P((fr_info_t *, ipstate_t *, i6addr_t *, 124 i6addr_t *, tcphdr_t *, u_32_t)); 125 static ipstate_t *fr_checkicmpmatchingstate __P((fr_info_t *)); 126 static int fr_state_flush __P((int, int, ipf_stack_t *)); 127 static ips_stat_t *fr_statetstats __P((ipf_stack_t *)); 128 static void fr_delstate __P((ipstate_t *, int, ipf_stack_t *)); 129 static int fr_state_remove __P((caddr_t, ipf_stack_t *)); 130 static void fr_ipsmove __P((ipstate_t *, u_int, ipf_stack_t *)); 131 static int fr_tcpstate __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 132 static int fr_tcpoptions __P((fr_info_t *, tcphdr_t *, tcpdata_t *)); 133 static ipstate_t *fr_stclone __P((fr_info_t *, tcphdr_t *, ipstate_t *)); 134 static void fr_fixinisn __P((fr_info_t *, ipstate_t *)); 135 static void fr_fixoutisn __P((fr_info_t *, ipstate_t *)); 136 static void fr_checknewisn __P((fr_info_t *, ipstate_t *)); 137 static int fr_stateiter __P((ipftoken_t *, ipfgeniter_t *, ipf_stack_t *)); 138 139 int fr_stputent __P((caddr_t, ipf_stack_t *)); 140 int fr_stgetent __P((caddr_t, ipf_stack_t *)); 141 142 #define ONE_DAY IPF_TTLVAL(1 * 86400) /* 1 day */ 143 #define FIVE_DAYS (5 * ONE_DAY) 144 #define DOUBLE_HASH(x, ifs) \ 145 (((x) + ifs->ifs_ips_seed[(x) % ifs->ifs_fr_statesize]) % ifs->ifs_fr_statesize) 146 147 148 149 /* ------------------------------------------------------------------------ */ 150 /* Function: fr_stateinit */ 151 /* Returns: int - 0 == success, -1 == failure */ 152 /* Parameters: Nil */ 153 /* */ 154 /* Initialise all the global variables used within the state code. */ 155 /* This action also includes initiailising locks. */ 156 /* ------------------------------------------------------------------------ */ 157 int fr_stateinit(ifs) 158 ipf_stack_t *ifs; 159 { 160 int i; 161 162 KMALLOCS(ifs->ifs_ips_table, ipstate_t **, 163 ifs->ifs_fr_statesize * sizeof(ipstate_t *)); 164 if (ifs->ifs_ips_table == NULL) 165 return -1; 166 bzero((char *)ifs->ifs_ips_table, 167 ifs->ifs_fr_statesize * sizeof(ipstate_t *)); 168 169 KMALLOCS(ifs->ifs_ips_seed, u_long *, 170 ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_seed)); 171 if (ifs->ifs_ips_seed == NULL) 172 return -2; 173 for (i = 0; i < ifs->ifs_fr_statesize; i++) { 174 /* 175 * XXX - ips_seed[X] should be a random number of sorts. 176 */ 177 #if (__FreeBSD_version >= 400000) 178 ifs->ifs_ips_seed[i] = arc4random(); 179 #else 180 ifs->ifs_ips_seed[i] = ((u_long)ifs->ifs_ips_seed + i) * 181 ifs->ifs_fr_statesize; 182 ifs->ifs_ips_seed[i] ^= 0xa5a55a5a; 183 ifs->ifs_ips_seed[i] *= (u_long)ifs->ifs_ips_seed; 184 ifs->ifs_ips_seed[i] ^= 0x5a5aa5a5; 185 ifs->ifs_ips_seed[i] *= ifs->ifs_fr_statemax; 186 #endif 187 } 188 189 /* fill icmp reply type table */ 190 for (i = 0; i <= ICMP_MAXTYPE; i++) 191 icmpreplytype4[i] = -1; 192 icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; 193 icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; 194 icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; 195 icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; 196 #ifdef USE_INET6 197 /* fill icmp reply type table */ 198 for (i = 0; i <= ICMP6_MAXTYPE; i++) 199 icmpreplytype6[i] = -1; 200 icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY; 201 icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT; 202 icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY; 203 icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT; 204 icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT; 205 #endif 206 207 KMALLOCS(ifs->ifs_ips_stats.iss_bucketlen, u_long *, 208 ifs->ifs_fr_statesize * sizeof(u_long)); 209 if (ifs->ifs_ips_stats.iss_bucketlen == NULL) 210 return -1; 211 bzero((char *)ifs->ifs_ips_stats.iss_bucketlen, 212 ifs->ifs_fr_statesize * sizeof(u_long)); 213 214 if (ifs->ifs_fr_state_maxbucket == 0) { 215 for (i = ifs->ifs_fr_statesize; i > 0; i >>= 1) 216 ifs->ifs_fr_state_maxbucket++; 217 ifs->ifs_fr_state_maxbucket *= 2; 218 } 219 220 fr_sttab_init(ifs->ifs_ips_tqtqb, ifs); 221 ifs->ifs_ips_tqtqb[IPF_TCP_NSTATES - 1].ifq_next = &ifs->ifs_ips_udptq; 222 ifs->ifs_ips_udptq.ifq_ttl = (u_long)ifs->ifs_fr_udptimeout; 223 ifs->ifs_ips_udptq.ifq_ref = 1; 224 ifs->ifs_ips_udptq.ifq_head = NULL; 225 ifs->ifs_ips_udptq.ifq_tail = &ifs->ifs_ips_udptq.ifq_head; 226 MUTEX_INIT(&ifs->ifs_ips_udptq.ifq_lock, "ipftq udp tab"); 227 ifs->ifs_ips_udptq.ifq_next = &ifs->ifs_ips_udpacktq; 228 ifs->ifs_ips_udpacktq.ifq_ttl = (u_long)ifs->ifs_fr_udpacktimeout; 229 ifs->ifs_ips_udpacktq.ifq_ref = 1; 230 ifs->ifs_ips_udpacktq.ifq_head = NULL; 231 ifs->ifs_ips_udpacktq.ifq_tail = &ifs->ifs_ips_udpacktq.ifq_head; 232 MUTEX_INIT(&ifs->ifs_ips_udpacktq.ifq_lock, "ipftq udpack tab"); 233 ifs->ifs_ips_udpacktq.ifq_next = &ifs->ifs_ips_icmptq; 234 ifs->ifs_ips_icmptq.ifq_ttl = (u_long)ifs->ifs_fr_icmptimeout; 235 ifs->ifs_ips_icmptq.ifq_ref = 1; 236 ifs->ifs_ips_icmptq.ifq_head = NULL; 237 ifs->ifs_ips_icmptq.ifq_tail = &ifs->ifs_ips_icmptq.ifq_head; 238 MUTEX_INIT(&ifs->ifs_ips_icmptq.ifq_lock, "ipftq icmp tab"); 239 ifs->ifs_ips_icmptq.ifq_next = &ifs->ifs_ips_icmpacktq; 240 ifs->ifs_ips_icmpacktq.ifq_ttl = (u_long)ifs->ifs_fr_icmpacktimeout; 241 ifs->ifs_ips_icmpacktq.ifq_ref = 1; 242 ifs->ifs_ips_icmpacktq.ifq_head = NULL; 243 ifs->ifs_ips_icmpacktq.ifq_tail = &ifs->ifs_ips_icmpacktq.ifq_head; 244 MUTEX_INIT(&ifs->ifs_ips_icmpacktq.ifq_lock, "ipftq icmpack tab"); 245 ifs->ifs_ips_icmpacktq.ifq_next = &ifs->ifs_ips_iptq; 246 ifs->ifs_ips_iptq.ifq_ttl = (u_long)ifs->ifs_fr_iptimeout; 247 ifs->ifs_ips_iptq.ifq_ref = 1; 248 ifs->ifs_ips_iptq.ifq_head = NULL; 249 ifs->ifs_ips_iptq.ifq_tail = &ifs->ifs_ips_iptq.ifq_head; 250 MUTEX_INIT(&ifs->ifs_ips_iptq.ifq_lock, "ipftq ip tab"); 251 ifs->ifs_ips_iptq.ifq_next = &ifs->ifs_ips_deletetq; 252 /* entry's ttl in deletetq is just 1 tick */ 253 ifs->ifs_ips_deletetq.ifq_ttl = (u_long) 1; 254 ifs->ifs_ips_deletetq.ifq_ref = 1; 255 ifs->ifs_ips_deletetq.ifq_head = NULL; 256 ifs->ifs_ips_deletetq.ifq_tail = &ifs->ifs_ips_deletetq.ifq_head; 257 MUTEX_INIT(&ifs->ifs_ips_deletetq.ifq_lock, "state delete queue"); 258 ifs->ifs_ips_deletetq.ifq_next = NULL; 259 260 RWLOCK_INIT(&ifs->ifs_ipf_state, "ipf IP state rwlock"); 261 MUTEX_INIT(&ifs->ifs_ipf_stinsert, "ipf state insert mutex"); 262 ifs->ifs_fr_state_init = 1; 263 264 ifs->ifs_ips_last_force_flush = ifs->ifs_fr_ticks; 265 return 0; 266 } 267 268 269 /* ------------------------------------------------------------------------ */ 270 /* Function: fr_stateunload */ 271 /* Returns: Nil */ 272 /* Parameters: Nil */ 273 /* */ 274 /* Release and destroy any resources acquired or initialised so that */ 275 /* IPFilter can be unloaded or re-initialised. */ 276 /* ------------------------------------------------------------------------ */ 277 void fr_stateunload(ifs) 278 ipf_stack_t *ifs; 279 { 280 ipftq_t *ifq, *ifqnext; 281 ipstate_t *is; 282 283 while ((is = ifs->ifs_ips_list) != NULL) 284 fr_delstate(is, 0, ifs); 285 286 /* 287 * Proxy timeout queues are not cleaned here because although they 288 * exist on the state list, appr_unload is called after fr_stateunload 289 * and the proxies actually are responsible for them being created. 290 * Should the proxy timeouts have their own list? There's no real 291 * justification as this is the only complicationA 292 */ 293 for (ifq = ifs->ifs_ips_utqe; ifq != NULL; ifq = ifqnext) { 294 ifqnext = ifq->ifq_next; 295 if (((ifq->ifq_flags & IFQF_PROXY) == 0) && 296 (fr_deletetimeoutqueue(ifq) == 0)) 297 fr_freetimeoutqueue(ifq, ifs); 298 } 299 300 ifs->ifs_ips_stats.iss_inuse = 0; 301 ifs->ifs_ips_num = 0; 302 303 if (ifs->ifs_fr_state_init == 1) { 304 fr_sttab_destroy(ifs->ifs_ips_tqtqb); 305 MUTEX_DESTROY(&ifs->ifs_ips_udptq.ifq_lock); 306 MUTEX_DESTROY(&ifs->ifs_ips_icmptq.ifq_lock); 307 MUTEX_DESTROY(&ifs->ifs_ips_udpacktq.ifq_lock); 308 MUTEX_DESTROY(&ifs->ifs_ips_icmpacktq.ifq_lock); 309 MUTEX_DESTROY(&ifs->ifs_ips_iptq.ifq_lock); 310 MUTEX_DESTROY(&ifs->ifs_ips_deletetq.ifq_lock); 311 } 312 313 if (ifs->ifs_ips_table != NULL) { 314 KFREES(ifs->ifs_ips_table, 315 ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_table)); 316 ifs->ifs_ips_table = NULL; 317 } 318 319 if (ifs->ifs_ips_seed != NULL) { 320 KFREES(ifs->ifs_ips_seed, 321 ifs->ifs_fr_statesize * sizeof(*ifs->ifs_ips_seed)); 322 ifs->ifs_ips_seed = NULL; 323 } 324 325 if (ifs->ifs_ips_stats.iss_bucketlen != NULL) { 326 KFREES(ifs->ifs_ips_stats.iss_bucketlen, 327 ifs->ifs_fr_statesize * sizeof(u_long)); 328 ifs->ifs_ips_stats.iss_bucketlen = NULL; 329 } 330 331 if (ifs->ifs_fr_state_maxbucket_reset == 1) 332 ifs->ifs_fr_state_maxbucket = 0; 333 334 if (ifs->ifs_fr_state_init == 1) { 335 ifs->ifs_fr_state_init = 0; 336 RW_DESTROY(&ifs->ifs_ipf_state); 337 MUTEX_DESTROY(&ifs->ifs_ipf_stinsert); 338 } 339 } 340 341 342 /* ------------------------------------------------------------------------ */ 343 /* Function: fr_statetstats */ 344 /* Returns: ips_state_t* - pointer to state stats structure */ 345 /* Parameters: Nil */ 346 /* */ 347 /* Put all the current numbers and pointers into a single struct and return */ 348 /* a pointer to it. */ 349 /* ------------------------------------------------------------------------ */ 350 static ips_stat_t *fr_statetstats(ifs) 351 ipf_stack_t *ifs; 352 { 353 ifs->ifs_ips_stats.iss_active = ifs->ifs_ips_num; 354 ifs->ifs_ips_stats.iss_statesize = ifs->ifs_fr_statesize; 355 ifs->ifs_ips_stats.iss_statemax = ifs->ifs_fr_statemax; 356 ifs->ifs_ips_stats.iss_table = ifs->ifs_ips_table; 357 ifs->ifs_ips_stats.iss_list = ifs->ifs_ips_list; 358 ifs->ifs_ips_stats.iss_ticks = ifs->ifs_fr_ticks; 359 return &ifs->ifs_ips_stats; 360 } 361 362 /* ------------------------------------------------------------------------ */ 363 /* Function: fr_state_remove */ 364 /* Returns: int - 0 == success, != 0 == failure */ 365 /* Parameters: data(I) - pointer to state structure to delete from table */ 366 /* */ 367 /* Search for a state structure that matches the one passed, according to */ 368 /* the IP addresses and other protocol specific information. */ 369 /* ------------------------------------------------------------------------ */ 370 static int fr_state_remove(data, ifs) 371 caddr_t data; 372 ipf_stack_t *ifs; 373 { 374 ipstate_t *sp, st; 375 int error; 376 377 sp = &st; 378 error = fr_inobj(data, &st, IPFOBJ_IPSTATE); 379 if (error) 380 return EFAULT; 381 382 WRITE_ENTER(&ifs->ifs_ipf_state); 383 for (sp = ifs->ifs_ips_list; sp; sp = sp->is_next) 384 if ((sp->is_p == st.is_p) && (sp->is_v == st.is_v) && 385 !bcmp((caddr_t)&sp->is_src, (caddr_t)&st.is_src, 386 sizeof(st.is_src)) && 387 !bcmp((caddr_t)&sp->is_dst, (caddr_t)&st.is_src, 388 sizeof(st.is_dst)) && 389 !bcmp((caddr_t)&sp->is_ps, (caddr_t)&st.is_ps, 390 sizeof(st.is_ps))) { 391 fr_delstate(sp, ISL_REMOVE, ifs); 392 RWLOCK_EXIT(&ifs->ifs_ipf_state); 393 return 0; 394 } 395 RWLOCK_EXIT(&ifs->ifs_ipf_state); 396 return ESRCH; 397 } 398 399 400 /* ------------------------------------------------------------------------ */ 401 /* Function: fr_state_ioctl */ 402 /* Returns: int - 0 == success, != 0 == failure */ 403 /* Parameters: data(I) - pointer to ioctl data */ 404 /* cmd(I) - ioctl command integer */ 405 /* mode(I) - file mode bits used with open */ 406 /* */ 407 /* Processes an ioctl call made to operate on the IP Filter state device. */ 408 /* ------------------------------------------------------------------------ */ 409 int fr_state_ioctl(data, cmd, mode, uid, ctx, ifs) 410 caddr_t data; 411 ioctlcmd_t cmd; 412 int mode, uid; 413 void *ctx; 414 ipf_stack_t *ifs; 415 { 416 int arg, ret, error = 0; 417 418 switch (cmd) 419 { 420 /* 421 * Delete an entry from the state table. 422 */ 423 case SIOCDELST : 424 error = fr_state_remove(data, ifs); 425 break; 426 /* 427 * Flush the state table 428 */ 429 case SIOCIPFFL : 430 BCOPYIN(data, (char *)&arg, sizeof(arg)); 431 if (arg == 0 || arg == 1) { 432 WRITE_ENTER(&ifs->ifs_ipf_state); 433 ret = fr_state_flush(arg, 4, ifs); 434 RWLOCK_EXIT(&ifs->ifs_ipf_state); 435 BCOPYOUT((char *)&ret, data, sizeof(ret)); 436 } else 437 error = EINVAL; 438 break; 439 #ifdef USE_INET6 440 case SIOCIPFL6 : 441 BCOPYIN(data, (char *)&arg, sizeof(arg)); 442 if (arg == 0 || arg == 1) { 443 WRITE_ENTER(&ifs->ifs_ipf_state); 444 ret = fr_state_flush(arg, 6, ifs); 445 RWLOCK_EXIT(&ifs->ifs_ipf_state); 446 BCOPYOUT((char *)&ret, data, sizeof(ret)); 447 } else 448 error = EINVAL; 449 break; 450 #endif 451 #ifdef IPFILTER_LOG 452 /* 453 * Flush the state log. 454 */ 455 case SIOCIPFFB : 456 if (!(mode & FWRITE)) 457 error = EPERM; 458 else { 459 int tmp; 460 461 tmp = ipflog_clear(IPL_LOGSTATE, ifs); 462 BCOPYOUT((char *)&tmp, data, sizeof(tmp)); 463 } 464 break; 465 /* 466 * Turn logging of state information on/off. 467 */ 468 case SIOCSETLG : 469 if (!(mode & FWRITE)) 470 error = EPERM; 471 else { 472 BCOPYIN((char *)data, 473 (char *)&ifs->ifs_ipstate_logging, 474 sizeof(ifs->ifs_ipstate_logging)); 475 } 476 break; 477 /* 478 * Return the current state of logging. 479 */ 480 case SIOCGETLG : 481 BCOPYOUT((char *)&ifs->ifs_ipstate_logging, (char *)data, 482 sizeof(ifs->ifs_ipstate_logging)); 483 break; 484 /* 485 * Return the number of bytes currently waiting to be read. 486 */ 487 case FIONREAD : 488 arg = ifs->ifs_iplused[IPL_LOGSTATE]; /* returned in an int */ 489 BCOPYOUT((char *)&arg, data, sizeof(arg)); 490 break; 491 #endif 492 /* 493 * Get the current state statistics. 494 */ 495 case SIOCGETFS : 496 error = fr_outobj(data, fr_statetstats(ifs), IPFOBJ_STATESTAT); 497 break; 498 /* 499 * Lock/Unlock the state table. (Locking prevents any changes, which 500 * means no packets match). 501 */ 502 case SIOCSTLCK : 503 if (!(mode & FWRITE)) { 504 error = EPERM; 505 } else { 506 fr_lock(data, &ifs->ifs_fr_state_lock); 507 } 508 break; 509 /* 510 * Add an entry to the current state table. 511 */ 512 case SIOCSTPUT : 513 if (!ifs->ifs_fr_state_lock || !(mode &FWRITE)) { 514 error = EACCES; 515 break; 516 } 517 error = fr_stputent(data, ifs); 518 break; 519 /* 520 * Get a state table entry. 521 */ 522 case SIOCSTGET : 523 if (!ifs->ifs_fr_state_lock) { 524 error = EACCES; 525 break; 526 } 527 error = fr_stgetent(data, ifs); 528 break; 529 530 case SIOCGENITER : 531 { 532 ipftoken_t *token; 533 ipfgeniter_t iter; 534 535 error = fr_inobj(data, &iter, IPFOBJ_GENITER); 536 if (error != 0) 537 break; 538 539 token = ipf_findtoken(IPFGENITER_STATE, uid, ctx, ifs); 540 if (token != NULL) 541 error = fr_stateiter(token, &iter, ifs); 542 else 543 error = ESRCH; 544 RWLOCK_EXIT(&ifs->ifs_ipf_tokens); 545 break; 546 } 547 548 case SIOCIPFDELTOK : 549 (void) BCOPYIN(data, (char *)&arg, sizeof(arg)); 550 error = ipf_deltoken(arg, uid, ctx, ifs); 551 break; 552 553 default : 554 error = EINVAL; 555 break; 556 } 557 return error; 558 } 559 560 561 /* ------------------------------------------------------------------------ */ 562 /* Function: fr_stgetent */ 563 /* Returns: int - 0 == success, != 0 == failure */ 564 /* Parameters: data(I) - pointer to state structure to retrieve from table */ 565 /* */ 566 /* Copy out state information from the kernel to a user space process. If */ 567 /* there is a filter rule associated with the state entry, copy that out */ 568 /* as well. The entry to copy out is taken from the value of "ips_next" in */ 569 /* the struct passed in and if not null and not found in the list of current*/ 570 /* state entries, the retrieval fails. */ 571 /* ------------------------------------------------------------------------ */ 572 int fr_stgetent(data, ifs) 573 caddr_t data; 574 ipf_stack_t *ifs; 575 { 576 ipstate_t *is, *isn; 577 ipstate_save_t ips; 578 int error; 579 580 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 581 if (error) 582 return EFAULT; 583 584 isn = ips.ips_next; 585 if (isn == NULL) { 586 isn = ifs->ifs_ips_list; 587 if (isn == NULL) { 588 if (ips.ips_next == NULL) 589 return ENOENT; 590 return 0; 591 } 592 } else { 593 /* 594 * Make sure the pointer we're copying from exists in the 595 * current list of entries. Security precaution to prevent 596 * copying of random kernel data. 597 */ 598 for (is = ifs->ifs_ips_list; is; is = is->is_next) 599 if (is == isn) 600 break; 601 if (!is) 602 return ESRCH; 603 } 604 ips.ips_next = isn->is_next; 605 bcopy((char *)isn, (char *)&ips.ips_is, sizeof(ips.ips_is)); 606 ips.ips_rule = isn->is_rule; 607 if (isn->is_rule != NULL) 608 bcopy((char *)isn->is_rule, (char *)&ips.ips_fr, 609 sizeof(ips.ips_fr)); 610 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 611 if (error) 612 return EFAULT; 613 return 0; 614 } 615 616 617 /* ------------------------------------------------------------------------ */ 618 /* Function: fr_stputent */ 619 /* Returns: int - 0 == success, != 0 == failure */ 620 /* Parameters: data(I) - pointer to state information struct */ 621 /* */ 622 /* This function implements the SIOCSTPUT ioctl: insert a state entry into */ 623 /* the state table. If the state info. includes a pointer to a filter rule */ 624 /* then also add in an orphaned rule (will not show up in any "ipfstat -io" */ 625 /* output. */ 626 /* ------------------------------------------------------------------------ */ 627 int fr_stputent(data, ifs) 628 caddr_t data; 629 ipf_stack_t *ifs; 630 { 631 ipstate_t *is, *isn; 632 ipstate_save_t ips; 633 int error, i; 634 frentry_t *fr; 635 char *name; 636 637 error = fr_inobj(data, &ips, IPFOBJ_STATESAVE); 638 if (error) 639 return EFAULT; 640 641 KMALLOC(isn, ipstate_t *); 642 if (isn == NULL) 643 return ENOMEM; 644 645 bcopy((char *)&ips.ips_is, (char *)isn, sizeof(*isn)); 646 bzero((char *)isn, offsetof(struct ipstate, is_pkts)); 647 isn->is_sti.tqe_pnext = NULL; 648 isn->is_sti.tqe_next = NULL; 649 isn->is_sti.tqe_ifq = NULL; 650 isn->is_sti.tqe_parent = isn; 651 isn->is_ifp[0] = NULL; 652 isn->is_ifp[1] = NULL; 653 isn->is_ifp[2] = NULL; 654 isn->is_ifp[3] = NULL; 655 isn->is_sync = NULL; 656 fr = ips.ips_rule; 657 658 if (fr == NULL) { 659 READ_ENTER(&ifs->ifs_ipf_state); 660 fr_stinsert(isn, 0, ifs); 661 MUTEX_EXIT(&isn->is_lock); 662 RWLOCK_EXIT(&ifs->ifs_ipf_state); 663 return 0; 664 } 665 666 if (isn->is_flags & SI_NEWFR) { 667 KMALLOC(fr, frentry_t *); 668 if (fr == NULL) { 669 KFREE(isn); 670 return ENOMEM; 671 } 672 bcopy((char *)&ips.ips_fr, (char *)fr, sizeof(*fr)); 673 isn->is_rule = fr; 674 ips.ips_is.is_rule = fr; 675 MUTEX_NUKE(&fr->fr_lock); 676 MUTEX_INIT(&fr->fr_lock, "state filter rule lock"); 677 678 /* 679 * Look up all the interface names in the rule. 680 */ 681 for (i = 0; i < 4; i++) { 682 name = fr->fr_ifnames[i]; 683 fr->fr_ifas[i] = fr_resolvenic(name, fr->fr_v, ifs); 684 name = isn->is_ifname[i]; 685 isn->is_ifp[i] = fr_resolvenic(name, isn->is_v, ifs); 686 } 687 688 fr->fr_ref = 0; 689 fr->fr_dsize = 0; 690 fr->fr_data = NULL; 691 692 fr_resolvedest(&fr->fr_tif, fr->fr_v, ifs); 693 fr_resolvedest(&fr->fr_dif, fr->fr_v, ifs); 694 fr_resolvedest(&fr->fr_rif, fr->fr_v, ifs); 695 696 /* 697 * send a copy back to userland of what we ended up 698 * to allow for verification. 699 */ 700 error = fr_outobj(data, &ips, IPFOBJ_STATESAVE); 701 if (error) { 702 KFREE(isn); 703 MUTEX_DESTROY(&fr->fr_lock); 704 KFREE(fr); 705 return EFAULT; 706 } 707 READ_ENTER(&ifs->ifs_ipf_state); 708 fr_stinsert(isn, 0, ifs); 709 MUTEX_EXIT(&isn->is_lock); 710 RWLOCK_EXIT(&ifs->ifs_ipf_state); 711 712 } else { 713 READ_ENTER(&ifs->ifs_ipf_state); 714 for (is = ifs->ifs_ips_list; is; is = is->is_next) 715 if (is->is_rule == fr) { 716 fr_stinsert(isn, 0, ifs); 717 MUTEX_EXIT(&isn->is_lock); 718 break; 719 } 720 721 if (is == NULL) { 722 KFREE(isn); 723 isn = NULL; 724 } 725 RWLOCK_EXIT(&ifs->ifs_ipf_state); 726 727 return (isn == NULL) ? ESRCH : 0; 728 } 729 730 return 0; 731 } 732 733 734 /* ------------------------------------------------------------------------ */ 735 /* Function: fr_stinsert */ 736 /* Returns: Nil */ 737 /* Parameters: is(I) - pointer to state structure */ 738 /* rev(I) - flag indicating forward/reverse direction of packet */ 739 /* */ 740 /* Inserts a state structure into the hash table (for lookups) and the list */ 741 /* of state entries (for enumeration). Resolves all of the interface names */ 742 /* to pointers and adjusts running stats for the hash table as appropriate. */ 743 /* */ 744 /* Locking: it is assumed that some kind of lock on ipf_state is held. */ 745 /* Exits with is_lock initialised and held. */ 746 /* ------------------------------------------------------------------------ */ 747 void fr_stinsert(is, rev, ifs) 748 ipstate_t *is; 749 int rev; 750 ipf_stack_t *ifs; 751 { 752 frentry_t *fr; 753 u_int hv; 754 int i; 755 756 MUTEX_INIT(&is->is_lock, "ipf state entry"); 757 758 fr = is->is_rule; 759 if (fr != NULL) { 760 MUTEX_ENTER(&fr->fr_lock); 761 fr->fr_ref++; 762 fr->fr_statecnt++; 763 MUTEX_EXIT(&fr->fr_lock); 764 } 765 766 /* 767 * Look up all the interface names in the state entry. 768 */ 769 for (i = 0; i < 4; i++) { 770 if (is->is_ifp[i] != NULL) 771 continue; 772 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], is->is_v, ifs); 773 } 774 775 /* 776 * If we could trust is_hv, then the modulous would not be needed, but 777 * when running with IPFILTER_SYNC, this stops bad values. 778 */ 779 hv = is->is_hv % ifs->ifs_fr_statesize; 780 is->is_hv = hv; 781 782 /* 783 * We need to get both of these locks...the first because it is 784 * possible that once the insert is complete another packet might 785 * come along, match the entry and want to update it. 786 */ 787 MUTEX_ENTER(&is->is_lock); 788 MUTEX_ENTER(&ifs->ifs_ipf_stinsert); 789 790 /* 791 * add into list table. 792 */ 793 if (ifs->ifs_ips_list != NULL) 794 ifs->ifs_ips_list->is_pnext = &is->is_next; 795 is->is_pnext = &ifs->ifs_ips_list; 796 is->is_next = ifs->ifs_ips_list; 797 ifs->ifs_ips_list = is; 798 799 if (ifs->ifs_ips_table[hv] != NULL) 800 ifs->ifs_ips_table[hv]->is_phnext = &is->is_hnext; 801 else 802 ifs->ifs_ips_stats.iss_inuse++; 803 is->is_phnext = ifs->ifs_ips_table + hv; 804 is->is_hnext = ifs->ifs_ips_table[hv]; 805 ifs->ifs_ips_table[hv] = is; 806 ifs->ifs_ips_stats.iss_bucketlen[hv]++; 807 ifs->ifs_ips_num++; 808 MUTEX_EXIT(&ifs->ifs_ipf_stinsert); 809 810 fr_setstatequeue(is, rev, ifs); 811 } 812 813 /* ------------------------------------------------------------------------ */ 814 /* Function: fr_match_ipv4addrs */ 815 /* Returns: int - 2 strong match (same addresses, same direction) */ 816 /* 1 weak match (same address, opposite direction) */ 817 /* 0 no match */ 818 /* */ 819 /* Function matches IPv4 addresses. */ 820 /* ------------------------------------------------------------------------ */ 821 static int fr_match_ipv4addrs(is1, is2) 822 ipstate_t *is1; 823 ipstate_t *is2; 824 { 825 int rv; 826 827 if (is1->is_saddr == is2->is_saddr && is1->is_daddr == is2->is_daddr) 828 rv = 2; 829 else if (is1->is_saddr == is2->is_daddr && 830 is1->is_daddr == is2->is_saddr) 831 rv = 1; 832 else 833 rv = 0; 834 835 return (rv); 836 } 837 838 /* ------------------------------------------------------------------------ */ 839 /* Function: fr_match_ipv6addrs */ 840 /* Returns: int - 2 strong match (same addresses, same direction) */ 841 /* 1 weak match (same addresses, opposite direction) */ 842 /* 0 no match */ 843 /* */ 844 /* Function matches IPv6 addresses. */ 845 /* ------------------------------------------------------------------------ */ 846 static int fr_match_ipv6addrs(is1, is2) 847 ipstate_t *is1; 848 ipstate_t *is2; 849 { 850 int rv; 851 852 if (IP6_EQ(&is1->is_src, &is2->is_src) && 853 IP6_EQ(&is1->is_dst, &is2->is_dst)) 854 rv = 2; 855 else if (IP6_EQ(&is1->is_src, &is2->is_dst) && 856 IP6_EQ(&is1->is_dst, &is2->is_src)) { 857 rv = 1; 858 } 859 else 860 rv = 0; 861 862 return (rv); 863 } 864 /* ------------------------------------------------------------------------ */ 865 /* Function: fr_match_addresses */ 866 /* Returns: int - 2 strong match (same addresses, same direction) */ 867 /* 1 weak match (same address, opposite directions) */ 868 /* 0 no match */ 869 /* Parameters: is1, is2 pointers to states we are checking */ 870 /* */ 871 /* Matches addresses, function uses fr_match_ipvXaddrs() to deal with IPv4 */ 872 /* and IPv6 address format. */ 873 /* ------------------------------------------------------------------------ */ 874 static int fr_match_addresses(is1, is2) 875 ipstate_t *is1; 876 ipstate_t *is2; 877 { 878 int rv; 879 880 if (is1->is_v == 4) { 881 rv = fr_match_ipv4addrs(is1, is2); 882 } 883 else { 884 rv = fr_match_ipv6addrs(is1, is2); 885 } 886 887 return (rv); 888 } 889 890 /* ------------------------------------------------------------------------ */ 891 /* Function: fr_match_ppairs */ 892 /* Returns: int - 2 strong match (same ports, same direction) */ 893 /* 1 weak match (same ports, different direction) */ 894 /* 0 no match */ 895 /* Parameters ppairs1, ppairs - src, dst ports we want to match. */ 896 /* */ 897 /* Matches two port_pair_t types (port pairs). Each port pair contains */ 898 /* src, dst port, which belong to session (state entry). */ 899 /* ------------------------------------------------------------------------ */ 900 static int fr_match_ppairs(ppairs1, ppairs2) 901 port_pair_t *ppairs1; 902 port_pair_t *ppairs2; 903 { 904 int rv; 905 906 if (ppairs1->pp_sport == ppairs2->pp_sport && 907 ppairs1->pp_dport == ppairs2->pp_dport) 908 rv = 2; 909 else if (ppairs1->pp_sport == ppairs2->pp_dport && 910 ppairs1->pp_dport == ppairs2->pp_sport) 911 rv = 1; 912 else 913 rv = 0; 914 915 return (rv); 916 } 917 918 /* ------------------------------------------------------------------------ */ 919 /* Function: fr_match_l4_hdr */ 920 /* Returns: int - 0 no match, */ 921 /* 1 weak match (same ports, different directions) */ 922 /* 2 strong match (same ports, same direction) */ 923 /* Parameters is1, is2 - states we want to match */ 924 /* */ 925 /* Function matches L4 header data (source ports for TCP, UDP, CallIds for */ 926 /* GRE protocol). */ 927 /* ------------------------------------------------------------------------ */ 928 static int fr_match_l4_hdr(is1, is2) 929 ipstate_t *is1; 930 ipstate_t *is2; 931 { 932 int rv = 0; 933 port_pair_t pp1; 934 port_pair_t pp2; 935 936 if (is1->is_p != is2->is_p) 937 return (0); 938 939 switch (is1->is_p) { 940 case IPPROTO_TCP: 941 pp1.pp_sport = is1->is_ps.is_ts.ts_sport; 942 pp1.pp_dport = is1->is_ps.is_ts.ts_dport; 943 pp2.pp_sport = is2->is_ps.is_ts.ts_sport; 944 pp2.pp_dport = is2->is_ps.is_ts.ts_dport; 945 rv = fr_match_ppairs(&pp1, &pp2); 946 break; 947 case IPPROTO_UDP: 948 pp1.pp_sport = is1->is_ps.is_us.us_sport; 949 pp1.pp_dport = is1->is_ps.is_us.us_dport; 950 pp2.pp_sport = is2->is_ps.is_us.us_sport; 951 pp2.pp_dport = is2->is_ps.is_us.us_dport; 952 rv = fr_match_ppairs(&pp1, &pp2); 953 break; 954 case IPPROTO_GRE: 955 /* greinfo_t can be also interprted as port pair */ 956 pp1.pp_sport = is1->is_ps.is_ug.gs_call[0]; 957 pp1.pp_dport = is1->is_ps.is_ug.gs_call[1]; 958 pp2.pp_sport = is2->is_ps.is_ug.gs_call[0]; 959 pp2.pp_dport = is2->is_ps.is_ug.gs_call[1]; 960 rv = fr_match_ppairs(&pp1, &pp2); 961 break; 962 case IPPROTO_ICMP: 963 case IPPROTO_ICMPV6: 964 if (bcmp(&is1->is_ps, &is2->is_ps, sizeof(icmpinfo_t))) 965 rv = 1; 966 else 967 rv = 0; 968 break; 969 default: 970 rv = 0; 971 } 972 973 return (rv); 974 } 975 976 /* ------------------------------------------------------------------------ */ 977 /* Function: fr_matchstates */ 978 /* Returns: int - nonzero match, zero no match */ 979 /* Parameters is1, is2 - states we want to match */ 980 /* */ 981 /* The state entries are equal (identical match) if they belong to the same */ 982 /* session. Any time new state entry is being added the fr_addstate() */ 983 /* function creates temporal state entry from the data it gets from IP and */ 984 /* L4 header. The fr_matchstats() must be also aware of packet direction, */ 985 /* which is also stored within the state entry. We should keep in mind the */ 986 /* information about packet direction is spread accross L3 (addresses) and */ 987 /* L4 (ports). There are three possible relationships betwee is1, is2: */ 988 /* - no match (match(is1, is2) == 0)) */ 989 /* - weak match same addresses (ports), but different */ 990 /* directions (1) (fr_match_xxxx(is1, is2) == 1) */ 991 /* - strong match same addresses (ports) and same directions */ 992 /* (2) (fr_match_xxxx(is1, is2) == 2) */ 993 /* */ 994 /* There are functions, which match match addresses (L3 header) in is1, is2 */ 995 /* and functions, which are used to compare ports (L4 header) data. We say */ 996 /* the is1 and is2 are same (identical) if there is a match */ 997 /* (fr_match_l4_hdr(is1, is2) != 0) and matchlevels are same for entries */ 998 /* (fr_match_l3_hdr(is1, is2) == fr_match_l4_hdr(is1, is2)) for is1, is2. */ 999 /* Such requirement deals with case as follows: */ 1000 /* suppose there are two connections between hosts A, B. Connection 1: */ 1001 /* a.a.a.a:12345 <=> b.b.b.b:54321 */ 1002 /* Connection 2: */ 1003 /* a.a.a.a:54321 <=> b.b.b.b:12345 */ 1004 /* since we've introduced match levels into our fr_matchstates(), we are */ 1005 /* able to identify, which packets belong to connection A and which belong */ 1006 /* to connection B. Assume there are two entries is1, is2. is1 has been */ 1007 /* from con. 1 packet, which travelled from A to B: */ 1008 /* a.a.a.a:12345 -> b.b.b.b:54321 */ 1009 /* while s2, has been created from packet which belongs to con. 2 and is */ 1010 /* also coming from A to B: */ 1011 /* a.a.a.a:54321 -> b.b.b.b:12345 */ 1012 /* fr_match_l3_hdr(is1, is2) == 2 -> strong match, while */ 1013 /* fr_match_l4_hdr(is1, is2) == 1 -> weak match. Since match levels are */ 1014 /* different the state entries are not identical -> no match as a final */ 1015 /* result. */ 1016 /* ------------------------------------------------------------------------ */ 1017 static int fr_matchstates(is1, is2) 1018 ipstate_t *is1; 1019 ipstate_t *is2; 1020 { 1021 int rv; 1022 int amatch; 1023 int pmatch; 1024 1025 if (bcmp(&is1->is_pass, &is2->is_pass, 1026 offsetof(struct ipstate, is_ps) - 1027 offsetof(struct ipstate, is_pass)) == 0) { 1028 1029 pmatch = fr_match_l4_hdr(is1, is2); 1030 amatch = fr_match_addresses(is1, is2); 1031 /* 1032 * If addresses match (amatch != 0), then 'match levels' 1033 * must be same for matching entries. If amatch and pmatch 1034 * have different values (different match levels), then 1035 * is1 and is2 belong to different sessions. 1036 */ 1037 rv = (amatch != 0) && (amatch == pmatch); 1038 } 1039 else 1040 rv = 0; 1041 1042 return (rv); 1043 } 1044 1045 /* ------------------------------------------------------------------------ */ 1046 /* Function: fr_addstate */ 1047 /* Returns: ipstate_t* - NULL == failure, else pointer to new state */ 1048 /* Parameters: fin(I) - pointer to packet information */ 1049 /* stsave(O) - pointer to place to save pointer to created */ 1050 /* state structure. */ 1051 /* flags(I) - flags to use when creating the structure */ 1052 /* */ 1053 /* Creates a new IP state structure from the packet information collected. */ 1054 /* Inserts it into the state table and appends to the bottom of the active */ 1055 /* list. If the capacity of the table has reached the maximum allowed then */ 1056 /* the call will fail and a flush is scheduled for the next timeout call. */ 1057 /* ------------------------------------------------------------------------ */ 1058 ipstate_t *fr_addstate(fin, stsave, flags) 1059 fr_info_t *fin; 1060 ipstate_t **stsave; 1061 u_int flags; 1062 { 1063 ipstate_t *is, ips; 1064 struct icmp *ic; 1065 u_int pass, hv; 1066 frentry_t *fr; 1067 tcphdr_t *tcp; 1068 grehdr_t *gre; 1069 void *ifp; 1070 int out; 1071 ipf_stack_t *ifs = fin->fin_ifs; 1072 1073 if (ifs->ifs_fr_state_lock || 1074 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 1075 return NULL; 1076 1077 if ((fin->fin_flx & FI_OOW) && !(fin->fin_tcpf & TH_SYN)) 1078 return NULL; 1079 1080 /* 1081 * If a "keep state" rule has reached the maximum number of references 1082 * to it, then schedule an automatic flush in case we can clear out 1083 * some "dead old wood". Note that because the lock isn't held on 1084 * fr it is possible that we could overflow. The cost of overflowing 1085 * is being ignored here as the number by which it can overflow is 1086 * a product of the number of simultaneous threads that could be 1087 * executing in here, so a limit of 100 won't result in 200, but could 1088 * result in 101 or 102. 1089 */ 1090 fr = fin->fin_fr; 1091 if (fr != NULL) { 1092 if ((ifs->ifs_ips_num == ifs->ifs_fr_statemax) && (fr->fr_statemax == 0)) { 1093 ATOMIC_INCL(ifs->ifs_ips_stats.iss_max); 1094 ifs->ifs_fr_state_doflush = 1; 1095 return NULL; 1096 } 1097 if ((fr->fr_statemax != 0) && 1098 (fr->fr_statecnt >= fr->fr_statemax)) { 1099 ATOMIC_INCL(ifs->ifs_ips_stats.iss_maxref); 1100 ifs->ifs_fr_state_doflush = 1; 1101 return NULL; 1102 } 1103 } 1104 1105 ic = NULL; 1106 tcp = NULL; 1107 out = fin->fin_out; 1108 is = &ips; 1109 bzero((char *)is, sizeof(*is)); 1110 1111 if (fr == NULL) { 1112 pass = ifs->ifs_fr_flags; 1113 is->is_tag = FR_NOLOGTAG; 1114 } 1115 else { 1116 pass = fr->fr_flags; 1117 } 1118 1119 is->is_die = 1 + ifs->ifs_fr_ticks; 1120 /* 1121 * We want to check everything that is a property of this packet, 1122 * but we don't (automatically) care about it's fragment status as 1123 * this may change. 1124 */ 1125 is->is_pass = pass; 1126 is->is_v = fin->fin_v; 1127 is->is_opt[0] = fin->fin_optmsk; 1128 is->is_optmsk[0] = 0xffffffff; 1129 is->is_optmsk[1] = 0xffffffff; 1130 if (is->is_v == 6) { 1131 is->is_opt[0] &= ~0x8; 1132 is->is_optmsk[0] &= ~0x8; 1133 is->is_optmsk[1] &= ~0x8; 1134 } 1135 is->is_sec = fin->fin_secmsk; 1136 is->is_secmsk = 0xffff; 1137 is->is_auth = fin->fin_auth; 1138 is->is_authmsk = 0xffff; 1139 1140 /* 1141 * Copy and calculate... 1142 */ 1143 hv = (is->is_p = fin->fin_fi.fi_p); 1144 is->is_src = fin->fin_fi.fi_src; 1145 hv += is->is_saddr; 1146 is->is_dst = fin->fin_fi.fi_dst; 1147 hv += is->is_daddr; 1148 #ifdef USE_INET6 1149 if (fin->fin_v == 6) { 1150 /* 1151 * For ICMPv6, we check to see if the destination address is 1152 * a multicast address. If it is, do not include it in the 1153 * calculation of the hash because the correct reply will come 1154 * back from a real address, not a multicast address. 1155 */ 1156 if ((is->is_p == IPPROTO_ICMPV6) && 1157 IN6_IS_ADDR_MULTICAST(&is->is_dst.in6)) { 1158 /* 1159 * So you can do keep state with neighbour discovery. 1160 * 1161 * Here we could use the address from the neighbour 1162 * solicit message to put in the state structure and 1163 * we could use that without a wildcard flag too... 1164 */ 1165 is->is_flags |= SI_W_DADDR; 1166 hv -= is->is_daddr; 1167 } else { 1168 hv += is->is_dst.i6[1]; 1169 hv += is->is_dst.i6[2]; 1170 hv += is->is_dst.i6[3]; 1171 } 1172 hv += is->is_src.i6[1]; 1173 hv += is->is_src.i6[2]; 1174 hv += is->is_src.i6[3]; 1175 } 1176 #endif 1177 1178 switch (is->is_p) 1179 { 1180 #ifdef USE_INET6 1181 case IPPROTO_ICMPV6 : 1182 ic = fin->fin_dp; 1183 1184 switch (ic->icmp_type) 1185 { 1186 case ICMP6_ECHO_REQUEST : 1187 is->is_icmp.ici_type = ic->icmp_type; 1188 hv += (is->is_icmp.ici_id = ic->icmp_id); 1189 break; 1190 case ICMP6_MEMBERSHIP_QUERY : 1191 case ND_ROUTER_SOLICIT : 1192 case ND_NEIGHBOR_SOLICIT : 1193 case ICMP6_NI_QUERY : 1194 is->is_icmp.ici_type = ic->icmp_type; 1195 break; 1196 default : 1197 return NULL; 1198 } 1199 ATOMIC_INCL(ifs->ifs_ips_stats.iss_icmp); 1200 break; 1201 #endif 1202 case IPPROTO_ICMP : 1203 ic = fin->fin_dp; 1204 1205 switch (ic->icmp_type) 1206 { 1207 case ICMP_ECHO : 1208 case ICMP_TSTAMP : 1209 case ICMP_IREQ : 1210 case ICMP_MASKREQ : 1211 is->is_icmp.ici_type = ic->icmp_type; 1212 hv += (is->is_icmp.ici_id = ic->icmp_id); 1213 break; 1214 default : 1215 return NULL; 1216 } 1217 ATOMIC_INCL(ifs->ifs_ips_stats.iss_icmp); 1218 break; 1219 1220 case IPPROTO_GRE : 1221 gre = fin->fin_dp; 1222 1223 is->is_gre.gs_flags = gre->gr_flags; 1224 is->is_gre.gs_ptype = gre->gr_ptype; 1225 if (GRE_REV(is->is_gre.gs_flags) == 1) { 1226 is->is_call[0] = fin->fin_data[0]; 1227 is->is_call[1] = fin->fin_data[1]; 1228 } 1229 break; 1230 1231 case IPPROTO_TCP : 1232 tcp = fin->fin_dp; 1233 1234 if (tcp->th_flags & TH_RST) 1235 return NULL; 1236 /* 1237 * The endian of the ports doesn't matter, but the ack and 1238 * sequence numbers do as we do mathematics on them later. 1239 */ 1240 is->is_sport = htons(fin->fin_data[0]); 1241 is->is_dport = htons(fin->fin_data[1]); 1242 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 1243 hv += is->is_sport; 1244 hv += is->is_dport; 1245 } 1246 1247 /* 1248 * If this is a real packet then initialise fields in the 1249 * state information structure from the TCP header information. 1250 */ 1251 1252 is->is_maxdwin = 1; 1253 is->is_maxswin = ntohs(tcp->th_win); 1254 if (is->is_maxswin == 0) 1255 is->is_maxswin = 1; 1256 1257 if ((fin->fin_flx & FI_IGNORE) == 0) { 1258 is->is_send = ntohl(tcp->th_seq) + fin->fin_dlen - 1259 (TCP_OFF(tcp) << 2) + 1260 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 1261 ((tcp->th_flags & TH_FIN) ? 1 : 0); 1262 is->is_maxsend = is->is_send; 1263 1264 /* 1265 * Window scale option is only present in 1266 * SYN/SYN-ACK packet. 1267 */ 1268 if ((tcp->th_flags & ~(TH_FIN|TH_ACK|TH_ECNALL)) == 1269 TH_SYN && 1270 (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) { 1271 if (fr_tcpoptions(fin, tcp, 1272 &is->is_tcp.ts_data[0]) == -1) { 1273 fin->fin_flx |= FI_BAD; 1274 } 1275 } 1276 1277 if ((fin->fin_out != 0) && (pass & FR_NEWISN) != 0) { 1278 fr_checknewisn(fin, is); 1279 fr_fixoutisn(fin, is); 1280 } 1281 1282 if ((tcp->th_flags & TH_OPENING) == TH_SYN) 1283 flags |= IS_TCPFSM; 1284 else { 1285 is->is_maxdwin = is->is_maxswin * 2; 1286 is->is_dend = ntohl(tcp->th_ack); 1287 is->is_maxdend = ntohl(tcp->th_ack); 1288 is->is_maxdwin *= 2; 1289 } 1290 } 1291 1292 /* 1293 * If we're creating state for a starting connection, start the 1294 * timer on it as we'll never see an error if it fails to 1295 * connect. 1296 */ 1297 ATOMIC_INCL(ifs->ifs_ips_stats.iss_tcp); 1298 break; 1299 1300 case IPPROTO_UDP : 1301 tcp = fin->fin_dp; 1302 1303 is->is_sport = htons(fin->fin_data[0]); 1304 is->is_dport = htons(fin->fin_data[1]); 1305 if ((flags & (SI_W_DPORT|SI_W_SPORT)) == 0) { 1306 hv += tcp->th_dport; 1307 hv += tcp->th_sport; 1308 } 1309 ATOMIC_INCL(ifs->ifs_ips_stats.iss_udp); 1310 break; 1311 1312 default : 1313 break; 1314 } 1315 hv = DOUBLE_HASH(hv, ifs); 1316 is->is_hv = hv; 1317 is->is_rule = fr; 1318 is->is_flags = flags & IS_INHERITED; 1319 1320 /* 1321 * Look for identical state. 1322 */ 1323 for (is = ifs->ifs_ips_table[is->is_hv % ifs->ifs_fr_statesize]; 1324 is != NULL; 1325 is = is->is_hnext) { 1326 if (fr_matchstates(&ips, is) == 1) 1327 break; 1328 } 1329 1330 /* 1331 * we've found a matching state -> state already exists, 1332 * we are not going to add a duplicate record. 1333 */ 1334 if (is != NULL) 1335 return NULL; 1336 1337 if (ifs->ifs_ips_stats.iss_bucketlen[hv] >= ifs->ifs_fr_state_maxbucket) { 1338 ATOMIC_INCL(ifs->ifs_ips_stats.iss_bucketfull); 1339 return NULL; 1340 } 1341 KMALLOC(is, ipstate_t *); 1342 if (is == NULL) { 1343 ATOMIC_INCL(ifs->ifs_ips_stats.iss_nomem); 1344 return NULL; 1345 } 1346 bcopy((char *)&ips, (char *)is, sizeof(*is)); 1347 /* 1348 * Do not do the modulous here, it is done in fr_stinsert(). 1349 */ 1350 if (fr != NULL) { 1351 (void) strncpy(is->is_group, fr->fr_group, FR_GROUPLEN); 1352 if (fr->fr_age[0] != 0) { 1353 is->is_tqehead[0] = 1354 fr_addtimeoutqueue(&ifs->ifs_ips_utqe, 1355 fr->fr_age[0], ifs); 1356 is->is_sti.tqe_flags |= TQE_RULEBASED; 1357 } 1358 if (fr->fr_age[1] != 0) { 1359 is->is_tqehead[1] = 1360 fr_addtimeoutqueue(&ifs->ifs_ips_utqe, 1361 fr->fr_age[1], ifs); 1362 is->is_sti.tqe_flags |= TQE_RULEBASED; 1363 } 1364 is->is_tag = fr->fr_logtag; 1365 1366 is->is_ifp[(out << 1) + 1] = fr->fr_ifas[1]; 1367 is->is_ifp[(1 - out) << 1] = fr->fr_ifas[2]; 1368 is->is_ifp[((1 - out) << 1) + 1] = fr->fr_ifas[3]; 1369 1370 if (((ifp = fr->fr_ifas[1]) != NULL) && 1371 (ifp != (void *)-1)) { 1372 COPYIFNAME(ifp, is->is_ifname[(out << 1) + 1], fr->fr_v); 1373 } 1374 if (((ifp = fr->fr_ifas[2]) != NULL) && 1375 (ifp != (void *)-1)) { 1376 COPYIFNAME(ifp, is->is_ifname[(1 - out) << 1], fr->fr_v); 1377 } 1378 if (((ifp = fr->fr_ifas[3]) != NULL) && 1379 (ifp != (void *)-1)) { 1380 COPYIFNAME(ifp, is->is_ifname[((1 - out) << 1) + 1], fr->fr_v); 1381 } 1382 } 1383 1384 is->is_ifp[out << 1] = fin->fin_ifp; 1385 if (fin->fin_ifp != NULL) { 1386 COPYIFNAME(fin->fin_ifp, is->is_ifname[out << 1], fr->fr_v); 1387 } 1388 1389 /* 1390 * It may seem strange to set is_ref to 2, but fr_check() will call 1391 * fr_statederef() after calling fr_addstate() and the idea is to 1392 * have it exist at the end of fr_check() with is_ref == 1. 1393 */ 1394 is->is_ref = 2; 1395 is->is_pkts[0] = 0, is->is_bytes[0] = 0; 1396 is->is_pkts[1] = 0, is->is_bytes[1] = 0; 1397 is->is_pkts[2] = 0, is->is_bytes[2] = 0; 1398 is->is_pkts[3] = 0, is->is_bytes[3] = 0; 1399 if ((fin->fin_flx & FI_IGNORE) == 0) { 1400 is->is_pkts[out] = 1; 1401 is->is_bytes[out] = fin->fin_plen; 1402 is->is_flx[out][0] = fin->fin_flx & FI_CMP; 1403 is->is_flx[out][0] &= ~FI_OOW; 1404 } 1405 1406 if (pass & FR_STSTRICT) 1407 is->is_flags |= IS_STRICT; 1408 1409 if (pass & FR_STATESYNC) 1410 is->is_flags |= IS_STATESYNC; 1411 1412 if (flags & (SI_WILDP|SI_WILDA)) { 1413 ATOMIC_INCL(ifs->ifs_ips_stats.iss_wild); 1414 } 1415 is->is_rulen = fin->fin_rule; 1416 1417 1418 if (pass & FR_LOGFIRST) 1419 is->is_pass &= ~(FR_LOGFIRST|FR_LOG); 1420 1421 READ_ENTER(&ifs->ifs_ipf_state); 1422 is->is_me = stsave; 1423 1424 fr_stinsert(is, fin->fin_rev, ifs); 1425 1426 if (fin->fin_p == IPPROTO_TCP) { 1427 /* 1428 * If we're creating state for a starting connection, start the 1429 * timer on it as we'll never see an error if it fails to 1430 * connect. 1431 */ 1432 (void) fr_tcp_age(&is->is_sti, fin, ifs->ifs_ips_tqtqb, 1433 is->is_flags); 1434 MUTEX_EXIT(&is->is_lock); 1435 #ifdef IPFILTER_SCAN 1436 if ((is->is_flags & SI_CLONE) == 0) 1437 (void) ipsc_attachis(is); 1438 #endif 1439 } else { 1440 MUTEX_EXIT(&is->is_lock); 1441 } 1442 #ifdef IPFILTER_SYNC 1443 if ((is->is_flags & IS_STATESYNC) && ((is->is_flags & SI_CLONE) == 0)) 1444 is->is_sync = ipfsync_new(SMC_STATE, fin, is); 1445 #endif 1446 if (ifs->ifs_ipstate_logging) 1447 ipstate_log(is, ISL_NEW, ifs); 1448 1449 RWLOCK_EXIT(&ifs->ifs_ipf_state); 1450 fin->fin_state = is; 1451 fin->fin_rev = IP6_NEQ(&is->is_dst, &fin->fin_daddr); 1452 fin->fin_flx |= FI_STATE; 1453 if (fin->fin_flx & FI_FRAG) 1454 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 1455 1456 return is; 1457 } 1458 1459 1460 /* ------------------------------------------------------------------------ */ 1461 /* Function: fr_tcpoptions */ 1462 /* Returns: int - 1 == packet matches state entry, 0 == it does not */ 1463 /* Parameters: fin(I) - pointer to packet information */ 1464 /* tcp(I) - pointer to TCP packet header */ 1465 /* td(I) - pointer to TCP data held as part of the state */ 1466 /* */ 1467 /* Look after the TCP header for any options and deal with those that are */ 1468 /* present. Record details about those that we recogise. */ 1469 /* ------------------------------------------------------------------------ */ 1470 static int fr_tcpoptions(fin, tcp, td) 1471 fr_info_t *fin; 1472 tcphdr_t *tcp; 1473 tcpdata_t *td; 1474 { 1475 int off, mlen, ol, i, len, retval; 1476 char buf[64], *s, opt; 1477 mb_t *m = NULL; 1478 1479 len = (TCP_OFF(tcp) << 2); 1480 if (fin->fin_dlen < len) 1481 return 0; 1482 len -= sizeof(*tcp); 1483 1484 off = fin->fin_plen - fin->fin_dlen + sizeof(*tcp) + fin->fin_ipoff; 1485 1486 m = fin->fin_m; 1487 mlen = MSGDSIZE(m) - off; 1488 if (len > mlen) { 1489 len = mlen; 1490 retval = 0; 1491 } else { 1492 retval = 1; 1493 } 1494 1495 COPYDATA(m, off, len, buf); 1496 1497 for (s = buf; len > 0; ) { 1498 opt = *s; 1499 if (opt == TCPOPT_EOL) 1500 break; 1501 else if (opt == TCPOPT_NOP) 1502 ol = 1; 1503 else { 1504 if (len < 2) 1505 break; 1506 ol = (int)*(s + 1); 1507 if (ol < 2 || ol > len) 1508 break; 1509 1510 /* 1511 * Extract the TCP options we are interested in out of 1512 * the header and store them in the the tcpdata struct. 1513 */ 1514 switch (opt) 1515 { 1516 case TCPOPT_WINDOW : 1517 if (ol == TCPOLEN_WINDOW) { 1518 i = (int)*(s + 2); 1519 if (i > TCP_WSCALE_MAX) 1520 i = TCP_WSCALE_MAX; 1521 else if (i < 0) 1522 i = 0; 1523 td->td_winscale = i; 1524 td->td_winflags |= TCP_WSCALE_SEEN| 1525 TCP_WSCALE_FIRST; 1526 } else 1527 retval = -1; 1528 break; 1529 case TCPOPT_MAXSEG : 1530 /* 1531 * So, if we wanted to set the TCP MAXSEG, 1532 * it should be done here... 1533 */ 1534 if (ol == TCPOLEN_MAXSEG) { 1535 i = (int)*(s + 2); 1536 i <<= 8; 1537 i += (int)*(s + 3); 1538 td->td_maxseg = i; 1539 } else 1540 retval = -1; 1541 break; 1542 case TCPOPT_SACK_PERMITTED : 1543 if (ol == TCPOLEN_SACK_PERMITTED) 1544 td->td_winflags |= TCP_SACK_PERMIT; 1545 else 1546 retval = -1; 1547 break; 1548 } 1549 } 1550 len -= ol; 1551 s += ol; 1552 } 1553 return retval; 1554 } 1555 1556 1557 /* ------------------------------------------------------------------------ */ 1558 /* Function: fr_tcpstate */ 1559 /* Returns: int - 1 == packet matches state entry, 0 == it does not */ 1560 /* Parameters: fin(I) - pointer to packet information */ 1561 /* tcp(I) - pointer to TCP packet header */ 1562 /* is(I) - pointer to master state structure */ 1563 /* */ 1564 /* Check to see if a packet with TCP headers fits within the TCP window. */ 1565 /* Change timeout depending on whether new packet is a SYN-ACK returning */ 1566 /* for a SYN or a RST or FIN which indicate time to close up shop. */ 1567 /* ------------------------------------------------------------------------ */ 1568 static int fr_tcpstate(fin, tcp, is) 1569 fr_info_t *fin; 1570 tcphdr_t *tcp; 1571 ipstate_t *is; 1572 { 1573 int source, ret = 0, flags; 1574 tcpdata_t *fdata, *tdata; 1575 ipf_stack_t *ifs = fin->fin_ifs; 1576 1577 source = !fin->fin_rev; 1578 if (((is->is_flags & IS_TCPFSM) != 0) && (source == 1) && 1579 (ntohs(is->is_sport) != fin->fin_data[0])) 1580 source = 0; 1581 fdata = &is->is_tcp.ts_data[!source]; 1582 tdata = &is->is_tcp.ts_data[source]; 1583 1584 MUTEX_ENTER(&is->is_lock); 1585 1586 /* 1587 * If a SYN packet is received for a connection that is in a half 1588 * closed state, then move its state entry to deletetq. In such case 1589 * the SYN packet will be consequently dropped. This allows new state 1590 * entry to be created with a retransmited SYN packet. 1591 */ 1592 if ((tcp->th_flags & TH_OPENING) == TH_SYN) { 1593 if (((is->is_state[source] > IPF_TCPS_ESTABLISHED) || 1594 (is->is_state[source] == IPF_TCPS_CLOSED)) && 1595 ((is->is_state[!source] > IPF_TCPS_ESTABLISHED) || 1596 (is->is_state[!source] == IPF_TCPS_CLOSED))) { 1597 /* 1598 * Do not update is->is_sti.tqe_die in case state entry 1599 * is already present in deletetq. It prevents state 1600 * entry ttl update by retransmitted SYN packets, which 1601 * may arrive before timer tick kicks off. The SYN 1602 * packet will be dropped again. 1603 */ 1604 if (is->is_sti.tqe_ifq != &ifs->ifs_ips_deletetq) 1605 fr_movequeue(&is->is_sti, is->is_sti.tqe_ifq, 1606 &fin->fin_ifs->ifs_ips_deletetq, 1607 fin->fin_ifs); 1608 1609 MUTEX_EXIT(&is->is_lock); 1610 return 0; 1611 } 1612 } 1613 1614 if (fr_tcpinwindow(fin, fdata, tdata, tcp, is->is_flags)) { 1615 #ifdef IPFILTER_SCAN 1616 if (is->is_flags & (IS_SC_CLIENT|IS_SC_SERVER)) { 1617 ipsc_packet(fin, is); 1618 if (FR_ISBLOCK(is->is_pass)) { 1619 MUTEX_EXIT(&is->is_lock); 1620 return 1; 1621 } 1622 } 1623 #endif 1624 1625 /* 1626 * Nearing end of connection, start timeout. 1627 */ 1628 ret = fr_tcp_age(&is->is_sti, fin, ifs->ifs_ips_tqtqb, 1629 is->is_flags); 1630 if (ret == 0) { 1631 MUTEX_EXIT(&is->is_lock); 1632 return 0; 1633 } 1634 1635 /* 1636 * set s0's as appropriate. Use syn-ack packet as it 1637 * contains both pieces of required information. 1638 */ 1639 /* 1640 * Window scale option is only present in SYN/SYN-ACK packet. 1641 * Compare with ~TH_FIN to mask out T/TCP setups. 1642 */ 1643 flags = tcp->th_flags & ~(TH_FIN|TH_ECNALL); 1644 if (flags == (TH_SYN|TH_ACK)) { 1645 is->is_s0[source] = ntohl(tcp->th_ack); 1646 is->is_s0[!source] = ntohl(tcp->th_seq) + 1; 1647 if (TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2)) { 1648 (void) fr_tcpoptions(fin, tcp, fdata); 1649 } 1650 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1651 fr_checknewisn(fin, is); 1652 } else if (flags == TH_SYN) { 1653 is->is_s0[source] = ntohl(tcp->th_seq) + 1; 1654 if ((TCP_OFF(tcp) > (sizeof(tcphdr_t) >> 2))) 1655 (void) fr_tcpoptions(fin, tcp, tdata); 1656 1657 if ((fin->fin_out != 0) && (is->is_pass & FR_NEWISN)) 1658 fr_checknewisn(fin, is); 1659 1660 } 1661 ret = 1; 1662 } else 1663 fin->fin_flx |= FI_OOW; 1664 MUTEX_EXIT(&is->is_lock); 1665 return ret; 1666 } 1667 1668 1669 /* ------------------------------------------------------------------------ */ 1670 /* Function: fr_checknewisn */ 1671 /* Returns: Nil */ 1672 /* Parameters: fin(I) - pointer to packet information */ 1673 /* is(I) - pointer to master state structure */ 1674 /* */ 1675 /* Check to see if this TCP connection is expecting and needs a new */ 1676 /* sequence number for a particular direction of the connection. */ 1677 /* */ 1678 /* NOTE: This does not actually change the sequence numbers, only gets new */ 1679 /* one ready. */ 1680 /* ------------------------------------------------------------------------ */ 1681 static void fr_checknewisn(fin, is) 1682 fr_info_t *fin; 1683 ipstate_t *is; 1684 { 1685 u_32_t sumd, old, new; 1686 tcphdr_t *tcp; 1687 int i; 1688 1689 i = fin->fin_rev; 1690 tcp = fin->fin_dp; 1691 1692 if (((i == 0) && !(is->is_flags & IS_ISNSYN)) || 1693 ((i == 1) && !(is->is_flags & IS_ISNACK))) { 1694 old = ntohl(tcp->th_seq); 1695 new = fr_newisn(fin); 1696 is->is_isninc[i] = new - old; 1697 CALC_SUMD(old, new, sumd); 1698 is->is_sumd[i] = (sumd & 0xffff) + (sumd >> 16); 1699 1700 is->is_flags |= ((i == 0) ? IS_ISNSYN : IS_ISNACK); 1701 } 1702 } 1703 1704 1705 /* ------------------------------------------------------------------------ */ 1706 /* Function: fr_tcpinwindow */ 1707 /* Returns: int - 1 == packet inside TCP "window", 0 == not inside. */ 1708 /* Parameters: fin(I) - pointer to packet information */ 1709 /* fdata(I) - pointer to tcp state informatio (forward) */ 1710 /* tdata(I) - pointer to tcp state informatio (reverse) */ 1711 /* tcp(I) - pointer to TCP packet header */ 1712 /* */ 1713 /* Given a packet has matched addresses and ports, check to see if it is */ 1714 /* within the TCP data window. In a show of generosity, allow packets that */ 1715 /* are within the window space behind the current sequence # as well. */ 1716 /* ------------------------------------------------------------------------ */ 1717 int fr_tcpinwindow(fin, fdata, tdata, tcp, flags) 1718 fr_info_t *fin; 1719 tcpdata_t *fdata, *tdata; 1720 tcphdr_t *tcp; 1721 int flags; 1722 { 1723 tcp_seq seq, ack, end; 1724 int ackskew, tcpflags; 1725 u_32_t win, maxwin; 1726 int dsize, inseq; 1727 1728 /* 1729 * Find difference between last checked packet and this packet. 1730 */ 1731 tcpflags = tcp->th_flags; 1732 seq = ntohl(tcp->th_seq); 1733 ack = ntohl(tcp->th_ack); 1734 1735 if (tcpflags & TH_SYN) 1736 win = ntohs(tcp->th_win); 1737 else 1738 win = ntohs(tcp->th_win) << fdata->td_winscale; 1739 1740 /* 1741 * win 0 means the receiving endpoint has closed the window, because it 1742 * has not enough memory to receive data from sender. In such case we 1743 * are pretending window size to be 1 to let TCP probe data through. 1744 * TCP probe data can be either 0 or 1 octet of data, the RFC does not 1745 * state this accurately, so we have to allow 1 octet (win = 1) even if 1746 * the window is closed (win == 0). 1747 */ 1748 if (win == 0) 1749 win = 1; 1750 1751 dsize = fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1752 ((tcpflags & TH_SYN) ? 1 : 0) + ((tcpflags & TH_FIN) ? 1 : 0); 1753 1754 /* 1755 * if window scaling is present, the scaling is only allowed 1756 * for windows not in the first SYN packet. In that packet the 1757 * window is 65535 to specify the largest window possible 1758 * for receivers not implementing the window scale option. 1759 * Currently, we do not assume TTCP here. That means that 1760 * if we see a second packet from a host (after the initial 1761 * SYN), we can assume that the receiver of the SYN did 1762 * already send back the SYN/ACK (and thus that we know if 1763 * the receiver also does window scaling) 1764 */ 1765 if (!(tcpflags & TH_SYN) && (fdata->td_winflags & TCP_WSCALE_FIRST)) { 1766 fdata->td_maxwin = win; 1767 } 1768 1769 end = seq + dsize; 1770 1771 if ((fdata->td_end == 0) && 1772 (!(flags & IS_TCPFSM) || 1773 ((tcpflags & TH_OPENING) == TH_OPENING))) { 1774 /* 1775 * Must be a (outgoing) SYN-ACK in reply to a SYN. 1776 */ 1777 fdata->td_end = end - 1; 1778 fdata->td_maxwin = 1; 1779 fdata->td_maxend = end + win; 1780 } 1781 1782 if (!(tcpflags & TH_ACK)) { /* Pretend an ack was sent */ 1783 ack = tdata->td_end; 1784 } else if (((tcpflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) && 1785 (ack == 0)) { 1786 /* gross hack to get around certain broken tcp stacks */ 1787 ack = tdata->td_end; 1788 } 1789 1790 maxwin = tdata->td_maxwin; 1791 ackskew = tdata->td_end - ack; 1792 1793 /* 1794 * Strict sequencing only allows in-order delivery. 1795 */ 1796 if ((flags & IS_STRICT) != 0) { 1797 if (seq != fdata->td_end) { 1798 return 0; 1799 } 1800 } 1801 1802 #define SEQ_GE(a,b) ((int)((a) - (b)) >= 0) 1803 #define SEQ_GT(a,b) ((int)((a) - (b)) > 0) 1804 inseq = 0; 1805 if ( 1806 #if defined(_KERNEL) 1807 (SEQ_GE(fdata->td_maxend, end)) && 1808 (SEQ_GE(seq, fdata->td_end - maxwin)) && 1809 #endif 1810 /* XXX what about big packets */ 1811 #define MAXACKWINDOW 66000 1812 (-ackskew <= (MAXACKWINDOW << fdata->td_winscale)) && 1813 ( ackskew <= (MAXACKWINDOW << fdata->td_winscale))) { 1814 inseq = 1; 1815 /* 1816 * Microsoft Windows will send the next packet to the right of the 1817 * window if SACK is in use. 1818 */ 1819 } else if ((seq == fdata->td_maxend) && (ackskew == 0) && 1820 (fdata->td_winflags & TCP_SACK_PERMIT) && 1821 (tdata->td_winflags & TCP_SACK_PERMIT)) { 1822 inseq = 1; 1823 /* 1824 * RST ACK with SEQ equal to 0 is sent by some OSes (i.e. Solaris) as a 1825 * response to initial SYN packet, when there is no application 1826 * listeing to on a port, where the SYN packet has came to. 1827 */ 1828 } else if ((seq == 0) && (tcpflags == (TH_RST|TH_ACK)) && 1829 (ackskew >= -1) && (ackskew <= 1)) { 1830 inseq = 1; 1831 } else if (!(flags & IS_TCPFSM)) { 1832 1833 if (!(fdata->td_winflags & 1834 (TCP_WSCALE_SEEN|TCP_WSCALE_FIRST))) { 1835 /* 1836 * No TCPFSM and no window scaling, so make some 1837 * extra guesses. 1838 */ 1839 if ((seq == fdata->td_maxend) && (ackskew == 0)) 1840 inseq = 1; 1841 else if (SEQ_GE(seq + maxwin, fdata->td_end - maxwin)) 1842 inseq = 1; 1843 } 1844 } 1845 1846 if (inseq) { 1847 /* if ackskew < 0 then this should be due to fragmented 1848 * packets. There is no way to know the length of the 1849 * total packet in advance. 1850 * We do know the total length from the fragment cache though. 1851 * Note however that there might be more sessions with 1852 * exactly the same source and destination parameters in the 1853 * state cache (and source and destination is the only stuff 1854 * that is saved in the fragment cache). Note further that 1855 * some TCP connections in the state cache are hashed with 1856 * sport and dport as well which makes it not worthwhile to 1857 * look for them. 1858 * Thus, when ackskew is negative but still seems to belong 1859 * to this session, we bump up the destinations end value. 1860 */ 1861 if (ackskew < 0) 1862 tdata->td_end = ack; 1863 1864 /* update max window seen */ 1865 if (fdata->td_maxwin < win) 1866 fdata->td_maxwin = win; 1867 if (SEQ_GT(end, fdata->td_end)) 1868 fdata->td_end = end; 1869 if (SEQ_GE(ack + win, tdata->td_maxend)) 1870 tdata->td_maxend = ack + win; 1871 return 1; 1872 } 1873 fin->fin_flx |= FI_OOW; 1874 return 0; 1875 } 1876 1877 1878 /* ------------------------------------------------------------------------ */ 1879 /* Function: fr_stclone */ 1880 /* Returns: ipstate_t* - NULL == cloning failed, */ 1881 /* else pointer to new state structure */ 1882 /* Parameters: fin(I) - pointer to packet information */ 1883 /* tcp(I) - pointer to TCP/UDP header */ 1884 /* is(I) - pointer to master state structure */ 1885 /* */ 1886 /* Create a "duplcate" state table entry from the master. */ 1887 /* ------------------------------------------------------------------------ */ 1888 static ipstate_t *fr_stclone(fin, tcp, is) 1889 fr_info_t *fin; 1890 tcphdr_t *tcp; 1891 ipstate_t *is; 1892 { 1893 ipstate_t *clone; 1894 u_32_t send; 1895 ipf_stack_t *ifs = fin->fin_ifs; 1896 1897 if (ifs->ifs_ips_num == ifs->ifs_fr_statemax) { 1898 ATOMIC_INCL(ifs->ifs_ips_stats.iss_max); 1899 ifs->ifs_fr_state_doflush = 1; 1900 return NULL; 1901 } 1902 KMALLOC(clone, ipstate_t *); 1903 if (clone == NULL) 1904 return NULL; 1905 bcopy((char *)is, (char *)clone, sizeof(*clone)); 1906 1907 MUTEX_NUKE(&clone->is_lock); 1908 1909 clone->is_die = ONE_DAY + ifs->ifs_fr_ticks; 1910 clone->is_state[0] = 0; 1911 clone->is_state[1] = 0; 1912 send = ntohl(tcp->th_seq) + fin->fin_dlen - (TCP_OFF(tcp) << 2) + 1913 ((tcp->th_flags & TH_SYN) ? 1 : 0) + 1914 ((tcp->th_flags & TH_FIN) ? 1 : 0); 1915 1916 if (fin->fin_rev == 1) { 1917 clone->is_dend = send; 1918 clone->is_maxdend = send; 1919 clone->is_send = 0; 1920 clone->is_maxswin = 1; 1921 clone->is_maxdwin = ntohs(tcp->th_win); 1922 if (clone->is_maxdwin == 0) 1923 clone->is_maxdwin = 1; 1924 } else { 1925 clone->is_send = send; 1926 clone->is_maxsend = send; 1927 clone->is_dend = 0; 1928 clone->is_maxdwin = 1; 1929 clone->is_maxswin = ntohs(tcp->th_win); 1930 if (clone->is_maxswin == 0) 1931 clone->is_maxswin = 1; 1932 } 1933 1934 clone->is_flags &= ~SI_CLONE; 1935 clone->is_flags |= SI_CLONED; 1936 fr_stinsert(clone, fin->fin_rev, ifs); 1937 clone->is_ref = 2; 1938 if (clone->is_p == IPPROTO_TCP) { 1939 (void) fr_tcp_age(&clone->is_sti, fin, ifs->ifs_ips_tqtqb, 1940 clone->is_flags); 1941 } 1942 MUTEX_EXIT(&clone->is_lock); 1943 #ifdef IPFILTER_SCAN 1944 (void) ipsc_attachis(is); 1945 #endif 1946 #ifdef IPFILTER_SYNC 1947 if (is->is_flags & IS_STATESYNC) 1948 clone->is_sync = ipfsync_new(SMC_STATE, fin, clone); 1949 #endif 1950 return clone; 1951 } 1952 1953 1954 /* ------------------------------------------------------------------------ */ 1955 /* Function: fr_matchsrcdst */ 1956 /* Returns: Nil */ 1957 /* Parameters: fin(I) - pointer to packet information */ 1958 /* is(I) - pointer to state structure */ 1959 /* src(I) - pointer to source address */ 1960 /* dst(I) - pointer to destination address */ 1961 /* tcp(I) - pointer to TCP/UDP header */ 1962 /* */ 1963 /* Match a state table entry against an IP packet. The logic below is that */ 1964 /* ret gets set to one if the match succeeds, else remains 0. If it is */ 1965 /* still 0 after the test. no match. */ 1966 /* ------------------------------------------------------------------------ */ 1967 static ipstate_t *fr_matchsrcdst(fin, is, src, dst, tcp, cmask) 1968 fr_info_t *fin; 1969 ipstate_t *is; 1970 i6addr_t *src, *dst; 1971 tcphdr_t *tcp; 1972 u_32_t cmask; 1973 { 1974 int ret = 0, rev, out, flags, flx = 0, idx; 1975 u_short sp, dp; 1976 u_32_t cflx; 1977 void *ifp; 1978 ipf_stack_t *ifs = fin->fin_ifs; 1979 1980 rev = IP6_NEQ(&is->is_dst, dst); 1981 ifp = fin->fin_ifp; 1982 out = fin->fin_out; 1983 flags = is->is_flags; 1984 sp = 0; 1985 dp = 0; 1986 1987 if (tcp != NULL) { 1988 sp = htons(fin->fin_sport); 1989 dp = ntohs(fin->fin_dport); 1990 } 1991 if (!rev) { 1992 if (tcp != NULL) { 1993 if (!(flags & SI_W_SPORT) && (sp != is->is_sport)) 1994 rev = 1; 1995 else if (!(flags & SI_W_DPORT) && (dp != is->is_dport)) 1996 rev = 1; 1997 } 1998 } 1999 2000 idx = (out << 1) + rev; 2001 2002 /* 2003 * If the interface for this 'direction' is set, make sure it matches. 2004 * An interface name that is not set matches any, as does a name of *. 2005 */ 2006 if ((is->is_ifp[idx] == NULL && 2007 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) || 2008 is->is_ifp[idx] == ifp) 2009 ret = 1; 2010 2011 if (ret == 0) 2012 return NULL; 2013 ret = 0; 2014 2015 /* 2016 * Match addresses and ports. 2017 */ 2018 if (rev == 0) { 2019 if ((IP6_EQ(&is->is_dst, dst) || (flags & SI_W_DADDR)) && 2020 (IP6_EQ(&is->is_src, src) || (flags & SI_W_SADDR))) { 2021 if (tcp) { 2022 if ((sp == is->is_sport || flags & SI_W_SPORT)&& 2023 (dp == is->is_dport || flags & SI_W_DPORT)) 2024 ret = 1; 2025 } else { 2026 ret = 1; 2027 } 2028 } 2029 } else { 2030 if ((IP6_EQ(&is->is_dst, src) || (flags & SI_W_DADDR)) && 2031 (IP6_EQ(&is->is_src, dst) || (flags & SI_W_SADDR))) { 2032 if (tcp) { 2033 if ((dp == is->is_sport || flags & SI_W_SPORT)&& 2034 (sp == is->is_dport || flags & SI_W_DPORT)) 2035 ret = 1; 2036 } else { 2037 ret = 1; 2038 } 2039 } 2040 } 2041 2042 if (ret == 0) 2043 return NULL; 2044 2045 /* 2046 * Whether or not this should be here, is questionable, but the aim 2047 * is to get this out of the main line. 2048 */ 2049 if (tcp == NULL) 2050 flags = is->is_flags & ~(SI_WILDP|SI_NEWFR|SI_CLONE|SI_CLONED); 2051 2052 /* 2053 * Only one of the source or destination address can be flaged as a 2054 * wildcard. Fill in the missing address, if set. 2055 * For IPv6, if the address being copied in is multicast, then 2056 * don't reset the wild flag - multicast causes it to be set in the 2057 * first place! 2058 */ 2059 if ((flags & (SI_W_SADDR|SI_W_DADDR))) { 2060 fr_ip_t *fi = &fin->fin_fi; 2061 2062 if ((flags & SI_W_SADDR) != 0) { 2063 if (rev == 0) { 2064 #ifdef USE_INET6 2065 if (is->is_v == 6 && 2066 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 2067 /*EMPTY*/; 2068 else 2069 #endif 2070 { 2071 is->is_src = fi->fi_src; 2072 is->is_flags &= ~SI_W_SADDR; 2073 } 2074 } else { 2075 #ifdef USE_INET6 2076 if (is->is_v == 6 && 2077 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 2078 /*EMPTY*/; 2079 else 2080 #endif 2081 { 2082 is->is_src = fi->fi_dst; 2083 is->is_flags &= ~SI_W_SADDR; 2084 } 2085 } 2086 } else if ((flags & SI_W_DADDR) != 0) { 2087 if (rev == 0) { 2088 #ifdef USE_INET6 2089 if (is->is_v == 6 && 2090 IN6_IS_ADDR_MULTICAST(&fi->fi_dst.in6)) 2091 /*EMPTY*/; 2092 else 2093 #endif 2094 { 2095 is->is_dst = fi->fi_dst; 2096 is->is_flags &= ~SI_W_DADDR; 2097 } 2098 } else { 2099 #ifdef USE_INET6 2100 if (is->is_v == 6 && 2101 IN6_IS_ADDR_MULTICAST(&fi->fi_src.in6)) 2102 /*EMPTY*/; 2103 else 2104 #endif 2105 { 2106 is->is_dst = fi->fi_src; 2107 is->is_flags &= ~SI_W_DADDR; 2108 } 2109 } 2110 } 2111 if ((is->is_flags & (SI_WILDA|SI_WILDP)) == 0) { 2112 ATOMIC_DECL(ifs->ifs_ips_stats.iss_wild); 2113 } 2114 } 2115 2116 flx = fin->fin_flx & cmask; 2117 cflx = is->is_flx[out][rev]; 2118 2119 /* 2120 * Match up any flags set from IP options. 2121 */ 2122 if ((cflx && (flx != (cflx & cmask))) || 2123 ((fin->fin_optmsk & is->is_optmsk[rev]) != is->is_opt[rev]) || 2124 ((fin->fin_secmsk & is->is_secmsk) != is->is_sec) || 2125 ((fin->fin_auth & is->is_authmsk) != is->is_auth)) 2126 return NULL; 2127 2128 /* 2129 * Only one of the source or destination port can be flagged as a 2130 * wildcard. When filling it in, fill in a copy of the matched entry 2131 * if it has the cloning flag set. 2132 */ 2133 if ((fin->fin_flx & FI_IGNORE) != 0) { 2134 fin->fin_rev = rev; 2135 return is; 2136 } 2137 2138 if ((flags & (SI_W_SPORT|SI_W_DPORT))) { 2139 if ((flags & SI_CLONE) != 0) { 2140 ipstate_t *clone; 2141 2142 clone = fr_stclone(fin, tcp, is); 2143 if (clone == NULL) 2144 return NULL; 2145 is = clone; 2146 } else { 2147 ATOMIC_DECL(ifs->ifs_ips_stats.iss_wild); 2148 } 2149 2150 if ((flags & SI_W_SPORT) != 0) { 2151 if (rev == 0) { 2152 is->is_sport = sp; 2153 is->is_send = ntohl(tcp->th_seq); 2154 } else { 2155 is->is_sport = dp; 2156 is->is_send = ntohl(tcp->th_ack); 2157 } 2158 is->is_maxsend = is->is_send + 1; 2159 } else if ((flags & SI_W_DPORT) != 0) { 2160 if (rev == 0) { 2161 is->is_dport = dp; 2162 is->is_dend = ntohl(tcp->th_ack); 2163 } else { 2164 is->is_dport = sp; 2165 is->is_dend = ntohl(tcp->th_seq); 2166 } 2167 is->is_maxdend = is->is_dend + 1; 2168 } 2169 is->is_flags &= ~(SI_W_SPORT|SI_W_DPORT); 2170 if ((flags & SI_CLONED) && ifs->ifs_ipstate_logging) 2171 ipstate_log(is, ISL_CLONE, ifs); 2172 } 2173 2174 ret = -1; 2175 2176 if (is->is_flx[out][rev] == 0) { 2177 is->is_flx[out][rev] = flx; 2178 is->is_opt[rev] = fin->fin_optmsk; 2179 if (is->is_v == 6) { 2180 is->is_opt[rev] &= ~0x8; 2181 is->is_optmsk[rev] &= ~0x8; 2182 } 2183 } 2184 2185 /* 2186 * Check if the interface name for this "direction" is set and if not, 2187 * fill it in. 2188 */ 2189 if (is->is_ifp[idx] == NULL && 2190 (*is->is_ifname[idx] == '\0' || *is->is_ifname[idx] == '*')) { 2191 is->is_ifp[idx] = ifp; 2192 COPYIFNAME(ifp, is->is_ifname[idx], fin->fin_v); 2193 } 2194 fin->fin_rev = rev; 2195 return is; 2196 } 2197 2198 2199 /* ------------------------------------------------------------------------ */ 2200 /* Function: fr_checkicmpmatchingstate */ 2201 /* Returns: Nil */ 2202 /* Parameters: fin(I) - pointer to packet information */ 2203 /* */ 2204 /* If we've got an ICMP error message, using the information stored in the */ 2205 /* ICMP packet, look for a matching state table entry. */ 2206 /* */ 2207 /* If we return NULL then no lock on ipf_state is held. */ 2208 /* If we return non-null then a read-lock on ipf_state is held. */ 2209 /* ------------------------------------------------------------------------ */ 2210 static ipstate_t *fr_checkicmpmatchingstate(fin) 2211 fr_info_t *fin; 2212 { 2213 ipstate_t *is, **isp; 2214 u_short sport, dport; 2215 u_char pr; 2216 int backward, i, oi; 2217 i6addr_t dst, src; 2218 struct icmp *ic; 2219 u_short savelen; 2220 icmphdr_t *icmp; 2221 fr_info_t ofin; 2222 tcphdr_t *tcp; 2223 int len; 2224 ip_t *oip; 2225 u_int hv; 2226 ipf_stack_t *ifs = fin->fin_ifs; 2227 2228 /* 2229 * Does it at least have the return (basic) IP header ? 2230 * Is it an actual recognised ICMP error type? 2231 * Only a basic IP header (no options) should be with 2232 * an ICMP error header. 2233 */ 2234 if ((fin->fin_v != 4) || (fin->fin_hlen != sizeof(ip_t)) || 2235 (fin->fin_plen < ICMPERR_MINPKTLEN) || 2236 !(fin->fin_flx & FI_ICMPERR)) 2237 return NULL; 2238 ic = fin->fin_dp; 2239 2240 oip = (ip_t *)((char *)ic + ICMPERR_ICMPHLEN); 2241 /* 2242 * Check if the at least the old IP header (with options) and 2243 * 8 bytes of payload is present. 2244 */ 2245 if (fin->fin_plen < ICMPERR_MAXPKTLEN + ((IP_HL(oip) - 5) << 2)) 2246 return NULL; 2247 2248 /* 2249 * Sanity Checks. 2250 */ 2251 len = fin->fin_dlen - ICMPERR_ICMPHLEN; 2252 if ((len <= 0) || ((IP_HL(oip) << 2) > len)) 2253 return NULL; 2254 2255 /* 2256 * Is the buffer big enough for all of it ? It's the size of the IP 2257 * header claimed in the encapsulated part which is of concern. It 2258 * may be too big to be in this buffer but not so big that it's 2259 * outside the ICMP packet, leading to TCP deref's causing problems. 2260 * This is possible because we don't know how big oip_hl is when we 2261 * do the pullup early in fr_check() and thus can't guarantee it is 2262 * all here now. 2263 */ 2264 #ifdef _KERNEL 2265 { 2266 mb_t *m; 2267 2268 m = fin->fin_m; 2269 # if defined(MENTAT) 2270 if ((char *)oip + len > (char *)m->b_wptr) 2271 return NULL; 2272 # else 2273 if ((char *)oip + len > (char *)fin->fin_ip + m->m_len) 2274 return NULL; 2275 # endif 2276 } 2277 #endif 2278 bcopy((char *)fin, (char *)&ofin, sizeof(*fin)); 2279 2280 /* 2281 * in the IPv4 case we must zero the i6addr union otherwise 2282 * the IP6_EQ and IP6_NEQ macros produce the wrong results because 2283 * of the 'junk' in the unused part of the union 2284 */ 2285 bzero((char *)&src, sizeof(src)); 2286 bzero((char *)&dst, sizeof(dst)); 2287 2288 /* 2289 * we make an fin entry to be able to feed it to 2290 * matchsrcdst note that not all fields are encessary 2291 * but this is the cleanest way. Note further we fill 2292 * in fin_mp such that if someone uses it we'll get 2293 * a kernel panic. fr_matchsrcdst does not use this. 2294 * 2295 * watch out here, as ip is in host order and oip in network 2296 * order. Any change we make must be undone afterwards, like 2297 * oip->ip_off - it is still in network byte order so fix it. 2298 */ 2299 savelen = oip->ip_len; 2300 oip->ip_len = len; 2301 oip->ip_off = ntohs(oip->ip_off); 2302 2303 ofin.fin_flx = FI_NOCKSUM; 2304 ofin.fin_v = 4; 2305 ofin.fin_ip = oip; 2306 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 2307 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 2308 ofin.fin_plen = fin->fin_dlen - ICMPERR_ICMPHLEN; 2309 (void) fr_makefrip(IP_HL(oip) << 2, oip, &ofin); 2310 ofin.fin_ifp = fin->fin_ifp; 2311 ofin.fin_out = !fin->fin_out; 2312 /* 2313 * Reset the short and bad flag here because in fr_matchsrcdst() 2314 * the flags for the current packet (fin_flx) are compared against 2315 * those for the existing session. 2316 */ 2317 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 2318 2319 /* 2320 * Put old values of ip_len and ip_off back as we don't know 2321 * if we have to forward the packet (or process it again. 2322 */ 2323 oip->ip_len = savelen; 2324 oip->ip_off = htons(oip->ip_off); 2325 2326 switch (oip->ip_p) 2327 { 2328 case IPPROTO_ICMP : 2329 /* 2330 * an ICMP error can only be generated as a result of an 2331 * ICMP query, not as the response on an ICMP error 2332 * 2333 * XXX theoretically ICMP_ECHOREP and the other reply's are 2334 * ICMP query's as well, but adding them here seems strange XXX 2335 */ 2336 if ((ofin.fin_flx & FI_ICMPERR) != 0) 2337 return NULL; 2338 2339 /* 2340 * perform a lookup of the ICMP packet in the state table 2341 */ 2342 icmp = (icmphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2343 hv = (pr = oip->ip_p); 2344 src.in4 = oip->ip_src; 2345 hv += src.in4.s_addr; 2346 dst.in4 = oip->ip_dst; 2347 hv += dst.in4.s_addr; 2348 hv += icmp->icmp_id; 2349 hv = DOUBLE_HASH(hv, ifs); 2350 2351 READ_ENTER(&ifs->ifs_ipf_state); 2352 for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) { 2353 isp = &is->is_hnext; 2354 if ((is->is_p != pr) || (is->is_v != 4)) 2355 continue; 2356 if (is->is_pass & FR_NOICMPERR) 2357 continue; 2358 is = fr_matchsrcdst(&ofin, is, &src, &dst, 2359 NULL, FI_ICMPCMP); 2360 if (is != NULL) { 2361 if ((is->is_pass & FR_NOICMPERR) != 0) { 2362 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2363 return NULL; 2364 } 2365 /* 2366 * i : the index of this packet (the icmp 2367 * unreachable) 2368 * oi : the index of the original packet found 2369 * in the icmp header (i.e. the packet 2370 * causing this icmp) 2371 * backward : original packet was backward 2372 * compared to the state 2373 */ 2374 backward = IP6_NEQ(&is->is_src, &src); 2375 fin->fin_rev = !backward; 2376 i = (!backward << 1) + fin->fin_out; 2377 oi = (backward << 1) + ofin.fin_out; 2378 if (is->is_icmppkts[i] > is->is_pkts[oi]) 2379 continue; 2380 ifs->ifs_ips_stats.iss_hits++; 2381 is->is_icmppkts[i]++; 2382 return is; 2383 } 2384 } 2385 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2386 return NULL; 2387 case IPPROTO_TCP : 2388 case IPPROTO_UDP : 2389 break; 2390 default : 2391 return NULL; 2392 } 2393 2394 tcp = (tcphdr_t *)((char *)oip + (IP_HL(oip) << 2)); 2395 dport = tcp->th_dport; 2396 sport = tcp->th_sport; 2397 2398 hv = (pr = oip->ip_p); 2399 src.in4 = oip->ip_src; 2400 hv += src.in4.s_addr; 2401 dst.in4 = oip->ip_dst; 2402 hv += dst.in4.s_addr; 2403 hv += dport; 2404 hv += sport; 2405 hv = DOUBLE_HASH(hv, ifs); 2406 2407 READ_ENTER(&ifs->ifs_ipf_state); 2408 for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) { 2409 isp = &is->is_hnext; 2410 /* 2411 * Only allow this icmp though if the 2412 * encapsulated packet was allowed through the 2413 * other way around. Note that the minimal amount 2414 * of info present does not allow for checking against 2415 * tcp internals such as seq and ack numbers. Only the 2416 * ports are known to be present and can be even if the 2417 * short flag is set. 2418 */ 2419 if ((is->is_p == pr) && (is->is_v == 4) && 2420 (is = fr_matchsrcdst(&ofin, is, &src, &dst, 2421 tcp, FI_ICMPCMP))) { 2422 /* 2423 * i : the index of this packet (the icmp unreachable) 2424 * oi : the index of the original packet found in the 2425 * icmp header (i.e. the packet causing this icmp) 2426 * backward : original packet was backward compared to 2427 * the state 2428 */ 2429 backward = IP6_NEQ(&is->is_src, &src); 2430 fin->fin_rev = !backward; 2431 i = (!backward << 1) + fin->fin_out; 2432 oi = (backward << 1) + ofin.fin_out; 2433 2434 if (((is->is_pass & FR_NOICMPERR) != 0) || 2435 (is->is_icmppkts[i] > is->is_pkts[oi])) 2436 break; 2437 ifs->ifs_ips_stats.iss_hits++; 2438 is->is_icmppkts[i]++; 2439 /* 2440 * we deliberately do not touch the timeouts 2441 * for the accompanying state table entry. 2442 * It remains to be seen if that is correct. XXX 2443 */ 2444 return is; 2445 } 2446 } 2447 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2448 return NULL; 2449 } 2450 2451 2452 /* ------------------------------------------------------------------------ */ 2453 /* Function: fr_ipsmove */ 2454 /* Returns: Nil */ 2455 /* Parameters: is(I) - pointer to state table entry */ 2456 /* hv(I) - new hash value for state table entry */ 2457 /* Write Locks: ipf_state */ 2458 /* */ 2459 /* Move a state entry from one position in the hash table to another. */ 2460 /* ------------------------------------------------------------------------ */ 2461 static void fr_ipsmove(is, hv, ifs) 2462 ipstate_t *is; 2463 u_int hv; 2464 ipf_stack_t *ifs; 2465 { 2466 ipstate_t **isp; 2467 u_int hvm; 2468 2469 ASSERT(rw_read_locked(&ifs->ifs_ipf_state.ipf_lk) == 0); 2470 2471 hvm = is->is_hv; 2472 /* 2473 * Remove the hash from the old location... 2474 */ 2475 isp = is->is_phnext; 2476 if (is->is_hnext) 2477 is->is_hnext->is_phnext = isp; 2478 *isp = is->is_hnext; 2479 if (ifs->ifs_ips_table[hvm] == NULL) 2480 ifs->ifs_ips_stats.iss_inuse--; 2481 ifs->ifs_ips_stats.iss_bucketlen[hvm]--; 2482 2483 /* 2484 * ...and put the hash in the new one. 2485 */ 2486 hvm = DOUBLE_HASH(hv, ifs); 2487 is->is_hv = hvm; 2488 isp = &ifs->ifs_ips_table[hvm]; 2489 if (*isp) 2490 (*isp)->is_phnext = &is->is_hnext; 2491 else 2492 ifs->ifs_ips_stats.iss_inuse++; 2493 ifs->ifs_ips_stats.iss_bucketlen[hvm]++; 2494 is->is_phnext = isp; 2495 is->is_hnext = *isp; 2496 *isp = is; 2497 } 2498 2499 2500 /* ------------------------------------------------------------------------ */ 2501 /* Function: fr_stlookup */ 2502 /* Returns: ipstate_t* - NULL == no matching state found, */ 2503 /* else pointer to state information is returned */ 2504 /* Parameters: fin(I) - pointer to packet information */ 2505 /* tcp(I) - pointer to TCP/UDP header. */ 2506 /* */ 2507 /* Search the state table for a matching entry to the packet described by */ 2508 /* the contents of *fin. */ 2509 /* */ 2510 /* If we return NULL then no lock on ipf_state is held. */ 2511 /* If we return non-null then a read-lock on ipf_state is held. */ 2512 /* ------------------------------------------------------------------------ */ 2513 ipstate_t *fr_stlookup(fin, tcp, ifqp) 2514 fr_info_t *fin; 2515 tcphdr_t *tcp; 2516 ipftq_t **ifqp; 2517 { 2518 u_int hv, hvm, pr, v, tryagain; 2519 ipstate_t *is, **isp; 2520 u_short dport, sport; 2521 i6addr_t src, dst; 2522 struct icmp *ic; 2523 ipftq_t *ifq; 2524 int oow; 2525 ipf_stack_t *ifs = fin->fin_ifs; 2526 2527 is = NULL; 2528 ifq = NULL; 2529 tcp = fin->fin_dp; 2530 ic = (struct icmp *)tcp; 2531 hv = (pr = fin->fin_fi.fi_p); 2532 src = fin->fin_fi.fi_src; 2533 dst = fin->fin_fi.fi_dst; 2534 hv += src.in4.s_addr; 2535 hv += dst.in4.s_addr; 2536 2537 v = fin->fin_fi.fi_v; 2538 #ifdef USE_INET6 2539 if (v == 6) { 2540 hv += fin->fin_fi.fi_src.i6[1]; 2541 hv += fin->fin_fi.fi_src.i6[2]; 2542 hv += fin->fin_fi.fi_src.i6[3]; 2543 2544 if ((fin->fin_p == IPPROTO_ICMPV6) && 2545 IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_dst.in6)) { 2546 hv -= dst.in4.s_addr; 2547 } else { 2548 hv += fin->fin_fi.fi_dst.i6[1]; 2549 hv += fin->fin_fi.fi_dst.i6[2]; 2550 hv += fin->fin_fi.fi_dst.i6[3]; 2551 } 2552 } 2553 #endif 2554 2555 /* 2556 * Search the hash table for matching packet header info. 2557 */ 2558 switch (pr) 2559 { 2560 #ifdef USE_INET6 2561 case IPPROTO_ICMPV6 : 2562 tryagain = 0; 2563 if (v == 6) { 2564 if ((ic->icmp_type == ICMP6_ECHO_REQUEST) || 2565 (ic->icmp_type == ICMP6_ECHO_REPLY)) { 2566 hv += ic->icmp_id; 2567 } 2568 } 2569 READ_ENTER(&ifs->ifs_ipf_state); 2570 icmp6again: 2571 hvm = DOUBLE_HASH(hv, ifs); 2572 for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) { 2573 isp = &is->is_hnext; 2574 if ((is->is_p != pr) || (is->is_v != v)) 2575 continue; 2576 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2577 if (is != NULL && 2578 fr_matchicmpqueryreply(v, &is->is_icmp, 2579 ic, fin->fin_rev)) { 2580 if (fin->fin_rev) 2581 ifq = &ifs->ifs_ips_icmpacktq; 2582 else 2583 ifq = &ifs->ifs_ips_icmptq; 2584 break; 2585 } 2586 } 2587 2588 if (is != NULL) { 2589 if ((tryagain != 0) && !(is->is_flags & SI_W_DADDR)) { 2590 hv += fin->fin_fi.fi_src.i6[0]; 2591 hv += fin->fin_fi.fi_src.i6[1]; 2592 hv += fin->fin_fi.fi_src.i6[2]; 2593 hv += fin->fin_fi.fi_src.i6[3]; 2594 fr_ipsmove(is, hv, ifs); 2595 MUTEX_DOWNGRADE(&ifs->ifs_ipf_state); 2596 } 2597 break; 2598 } 2599 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2600 2601 /* 2602 * No matching icmp state entry. Perhaps this is a 2603 * response to another state entry. 2604 * 2605 * XXX With some ICMP6 packets, the "other" address is already 2606 * in the packet, after the ICMP6 header, and this could be 2607 * used in place of the multicast address. However, taking 2608 * advantage of this requires some significant code changes 2609 * to handle the specific types where that is the case. 2610 */ 2611 if ((ifs->ifs_ips_stats.iss_wild != 0) && (v == 6) && (tryagain == 0) && 2612 !IN6_IS_ADDR_MULTICAST(&fin->fin_fi.fi_src.in6)) { 2613 hv -= fin->fin_fi.fi_src.i6[0]; 2614 hv -= fin->fin_fi.fi_src.i6[1]; 2615 hv -= fin->fin_fi.fi_src.i6[2]; 2616 hv -= fin->fin_fi.fi_src.i6[3]; 2617 tryagain = 1; 2618 WRITE_ENTER(&ifs->ifs_ipf_state); 2619 goto icmp6again; 2620 } 2621 2622 is = fr_checkicmp6matchingstate(fin); 2623 if (is != NULL) 2624 return is; 2625 break; 2626 #endif 2627 2628 case IPPROTO_ICMP : 2629 if (v == 4) { 2630 hv += ic->icmp_id; 2631 } 2632 hv = DOUBLE_HASH(hv, ifs); 2633 READ_ENTER(&ifs->ifs_ipf_state); 2634 for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) { 2635 isp = &is->is_hnext; 2636 if ((is->is_p != pr) || (is->is_v != v)) 2637 continue; 2638 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2639 if (is != NULL && 2640 fr_matchicmpqueryreply(v, &is->is_icmp, 2641 ic, fin->fin_rev)) { 2642 if (fin->fin_rev) 2643 ifq = &ifs->ifs_ips_icmpacktq; 2644 else 2645 ifq = &ifs->ifs_ips_icmptq; 2646 break; 2647 } 2648 } 2649 if (is == NULL) { 2650 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2651 } 2652 break; 2653 2654 case IPPROTO_TCP : 2655 case IPPROTO_UDP : 2656 ifqp = NULL; 2657 sport = htons(fin->fin_data[0]); 2658 hv += sport; 2659 dport = htons(fin->fin_data[1]); 2660 hv += dport; 2661 oow = 0; 2662 tryagain = 0; 2663 READ_ENTER(&ifs->ifs_ipf_state); 2664 retry_tcpudp: 2665 hvm = DOUBLE_HASH(hv, ifs); 2666 for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) { 2667 isp = &is->is_hnext; 2668 if ((is->is_p != pr) || (is->is_v != v)) 2669 continue; 2670 fin->fin_flx &= ~FI_OOW; 2671 is = fr_matchsrcdst(fin, is, &src, &dst, tcp, FI_CMP); 2672 if (is != NULL) { 2673 if (pr == IPPROTO_TCP) { 2674 if (!fr_tcpstate(fin, tcp, is)) { 2675 oow |= fin->fin_flx & FI_OOW; 2676 continue; 2677 } 2678 } 2679 break; 2680 } 2681 } 2682 if (is != NULL) { 2683 if (tryagain && 2684 !(is->is_flags & (SI_CLONE|SI_WILDP|SI_WILDA))) { 2685 hv += dport; 2686 hv += sport; 2687 fr_ipsmove(is, hv, ifs); 2688 MUTEX_DOWNGRADE(&ifs->ifs_ipf_state); 2689 } 2690 break; 2691 } 2692 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2693 2694 if (!tryagain && ifs->ifs_ips_stats.iss_wild) { 2695 hv -= dport; 2696 hv -= sport; 2697 tryagain = 1; 2698 WRITE_ENTER(&ifs->ifs_ipf_state); 2699 goto retry_tcpudp; 2700 } 2701 fin->fin_flx |= oow; 2702 break; 2703 2704 #if 0 2705 case IPPROTO_GRE : 2706 gre = fin->fin_dp; 2707 if (GRE_REV(gre->gr_flags) == 1) { 2708 hv += gre->gr_call; 2709 } 2710 /* FALLTHROUGH */ 2711 #endif 2712 default : 2713 ifqp = NULL; 2714 hvm = DOUBLE_HASH(hv, ifs); 2715 READ_ENTER(&ifs->ifs_ipf_state); 2716 for (isp = &ifs->ifs_ips_table[hvm]; ((is = *isp) != NULL); ) { 2717 isp = &is->is_hnext; 2718 if ((is->is_p != pr) || (is->is_v != v)) 2719 continue; 2720 is = fr_matchsrcdst(fin, is, &src, &dst, NULL, FI_CMP); 2721 if (is != NULL) { 2722 ifq = &ifs->ifs_ips_iptq; 2723 break; 2724 } 2725 } 2726 if (is == NULL) { 2727 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2728 } 2729 break; 2730 } 2731 2732 if ((is != NULL) && ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) && 2733 (is->is_tqehead[fin->fin_rev] != NULL)) 2734 ifq = is->is_tqehead[fin->fin_rev]; 2735 if (ifq != NULL && ifqp != NULL) 2736 *ifqp = ifq; 2737 return is; 2738 } 2739 2740 2741 /* ------------------------------------------------------------------------ */ 2742 /* Function: fr_updatestate */ 2743 /* Returns: Nil */ 2744 /* Parameters: fin(I) - pointer to packet information */ 2745 /* is(I) - pointer to state table entry */ 2746 /* Read Locks: ipf_state */ 2747 /* */ 2748 /* Updates packet and byte counters for a newly received packet. Seeds the */ 2749 /* fragment cache with a new entry as required. */ 2750 /* ------------------------------------------------------------------------ */ 2751 void fr_updatestate(fin, is, ifq) 2752 fr_info_t *fin; 2753 ipstate_t *is; 2754 ipftq_t *ifq; 2755 { 2756 ipftqent_t *tqe; 2757 int i, pass; 2758 ipf_stack_t *ifs = fin->fin_ifs; 2759 2760 i = (fin->fin_rev << 1) + fin->fin_out; 2761 2762 /* 2763 * For TCP packets, ifq == NULL. For all others, check if this new 2764 * queue is different to the last one it was on and move it if so. 2765 */ 2766 tqe = &is->is_sti; 2767 MUTEX_ENTER(&is->is_lock); 2768 if ((tqe->tqe_flags & TQE_RULEBASED) != 0) 2769 ifq = is->is_tqehead[fin->fin_rev]; 2770 2771 if (ifq != NULL) 2772 fr_movequeue(tqe, tqe->tqe_ifq, ifq, ifs); 2773 2774 is->is_pkts[i]++; 2775 is->is_bytes[i] += fin->fin_plen; 2776 MUTEX_EXIT(&is->is_lock); 2777 2778 #ifdef IPFILTER_SYNC 2779 if (is->is_flags & IS_STATESYNC) 2780 ipfsync_update(SMC_STATE, fin, is->is_sync); 2781 #endif 2782 2783 ATOMIC_INCL(ifs->ifs_ips_stats.iss_hits); 2784 2785 fin->fin_fr = is->is_rule; 2786 2787 /* 2788 * If this packet is a fragment and the rule says to track fragments, 2789 * then create a new fragment cache entry. 2790 */ 2791 pass = is->is_pass; 2792 if ((fin->fin_flx & FI_FRAG) && FR_ISPASS(pass)) 2793 (void) fr_newfrag(fin, pass ^ FR_KEEPSTATE); 2794 } 2795 2796 2797 /* ------------------------------------------------------------------------ */ 2798 /* Function: fr_checkstate */ 2799 /* Returns: frentry_t* - NULL == search failed, */ 2800 /* else pointer to rule for matching state */ 2801 /* Parameters: ifp(I) - pointer to interface */ 2802 /* passp(I) - pointer to filtering result flags */ 2803 /* */ 2804 /* Check if a packet is associated with an entry in the state table. */ 2805 /* ------------------------------------------------------------------------ */ 2806 frentry_t *fr_checkstate(fin, passp) 2807 fr_info_t *fin; 2808 u_32_t *passp; 2809 { 2810 ipstate_t *is; 2811 frentry_t *fr; 2812 tcphdr_t *tcp; 2813 ipftq_t *ifq; 2814 u_int pass; 2815 ipf_stack_t *ifs = fin->fin_ifs; 2816 2817 if (ifs->ifs_fr_state_lock || (ifs->ifs_ips_list == NULL) || 2818 (fin->fin_flx & (FI_SHORT|FI_STATE|FI_FRAGBODY|FI_BAD))) 2819 return NULL; 2820 2821 is = NULL; 2822 if ((fin->fin_flx & FI_TCPUDP) || 2823 (fin->fin_fi.fi_p == IPPROTO_ICMP) 2824 #ifdef USE_INET6 2825 || (fin->fin_fi.fi_p == IPPROTO_ICMPV6) 2826 #endif 2827 ) 2828 tcp = fin->fin_dp; 2829 else 2830 tcp = NULL; 2831 2832 /* 2833 * Search the hash table for matching packet header info. 2834 */ 2835 ifq = NULL; 2836 is = fin->fin_state; 2837 if (is == NULL) 2838 is = fr_stlookup(fin, tcp, &ifq); 2839 switch (fin->fin_p) 2840 { 2841 #ifdef USE_INET6 2842 case IPPROTO_ICMPV6 : 2843 if (is != NULL) 2844 break; 2845 if (fin->fin_v == 6) { 2846 is = fr_checkicmp6matchingstate(fin); 2847 if (is != NULL) 2848 goto matched; 2849 } 2850 break; 2851 #endif 2852 case IPPROTO_ICMP : 2853 if (is != NULL) 2854 break; 2855 /* 2856 * No matching icmp state entry. Perhaps this is a 2857 * response to another state entry. 2858 */ 2859 is = fr_checkicmpmatchingstate(fin); 2860 if (is != NULL) 2861 goto matched; 2862 break; 2863 case IPPROTO_TCP : 2864 if (is == NULL) 2865 break; 2866 2867 if (is->is_pass & FR_NEWISN) { 2868 if (fin->fin_out == 0) 2869 fr_fixinisn(fin, is); 2870 else if (fin->fin_out == 1) 2871 fr_fixoutisn(fin, is); 2872 } 2873 break; 2874 default : 2875 if (fin->fin_rev) 2876 ifq = &ifs->ifs_ips_udpacktq; 2877 else 2878 ifq = &ifs->ifs_ips_udptq; 2879 break; 2880 } 2881 if (is == NULL) { 2882 ATOMIC_INCL(ifs->ifs_ips_stats.iss_miss); 2883 return NULL; 2884 } 2885 2886 matched: 2887 fr = is->is_rule; 2888 if (fr != NULL) { 2889 if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) { 2890 if (fin->fin_nattag == NULL) 2891 return NULL; 2892 if (fr_matchtag(&fr->fr_nattag, fin->fin_nattag) != 0) 2893 return NULL; 2894 } 2895 (void) strncpy(fin->fin_group, fr->fr_group, FR_GROUPLEN); 2896 fin->fin_icode = fr->fr_icode; 2897 } 2898 2899 fin->fin_rule = is->is_rulen; 2900 pass = is->is_pass; 2901 fr_updatestate(fin, is, ifq); 2902 if (fin->fin_out == 1) 2903 fin->fin_nat = is->is_nat[fin->fin_rev]; 2904 2905 fin->fin_state = is; 2906 is->is_touched = ifs->ifs_fr_ticks; 2907 MUTEX_ENTER(&is->is_lock); 2908 is->is_ref++; 2909 MUTEX_EXIT(&is->is_lock); 2910 RWLOCK_EXIT(&ifs->ifs_ipf_state); 2911 fin->fin_flx |= FI_STATE; 2912 if ((pass & FR_LOGFIRST) != 0) 2913 pass &= ~(FR_LOGFIRST|FR_LOG); 2914 *passp = pass; 2915 return fr; 2916 } 2917 2918 2919 /* ------------------------------------------------------------------------ */ 2920 /* Function: fr_fixoutisn */ 2921 /* Returns: Nil */ 2922 /* Parameters: fin(I) - pointer to packet information */ 2923 /* is(I) - pointer to master state structure */ 2924 /* */ 2925 /* Called only for outbound packets, adjusts the sequence number and the */ 2926 /* TCP checksum to match that change. */ 2927 /* ------------------------------------------------------------------------ */ 2928 static void fr_fixoutisn(fin, is) 2929 fr_info_t *fin; 2930 ipstate_t *is; 2931 { 2932 tcphdr_t *tcp; 2933 int rev; 2934 u_32_t seq; 2935 2936 tcp = fin->fin_dp; 2937 rev = fin->fin_rev; 2938 if ((is->is_flags & IS_ISNSYN) != 0) { 2939 if (rev == 0) { 2940 seq = ntohl(tcp->th_seq); 2941 seq += is->is_isninc[0]; 2942 tcp->th_seq = htonl(seq); 2943 fix_outcksum(&tcp->th_sum, is->is_sumd[0]); 2944 } 2945 } 2946 if ((is->is_flags & IS_ISNACK) != 0) { 2947 if (rev == 1) { 2948 seq = ntohl(tcp->th_seq); 2949 seq += is->is_isninc[1]; 2950 tcp->th_seq = htonl(seq); 2951 fix_outcksum(&tcp->th_sum, is->is_sumd[1]); 2952 } 2953 } 2954 } 2955 2956 2957 /* ------------------------------------------------------------------------ */ 2958 /* Function: fr_fixinisn */ 2959 /* Returns: Nil */ 2960 /* Parameters: fin(I) - pointer to packet information */ 2961 /* is(I) - pointer to master state structure */ 2962 /* */ 2963 /* Called only for inbound packets, adjusts the acknowledge number and the */ 2964 /* TCP checksum to match that change. */ 2965 /* ------------------------------------------------------------------------ */ 2966 static void fr_fixinisn(fin, is) 2967 fr_info_t *fin; 2968 ipstate_t *is; 2969 { 2970 tcphdr_t *tcp; 2971 int rev; 2972 u_32_t ack; 2973 2974 tcp = fin->fin_dp; 2975 rev = fin->fin_rev; 2976 if ((is->is_flags & IS_ISNSYN) != 0) { 2977 if (rev == 1) { 2978 ack = ntohl(tcp->th_ack); 2979 ack -= is->is_isninc[0]; 2980 tcp->th_ack = htonl(ack); 2981 fix_incksum(&tcp->th_sum, is->is_sumd[0]); 2982 } 2983 } 2984 if ((is->is_flags & IS_ISNACK) != 0) { 2985 if (rev == 0) { 2986 ack = ntohl(tcp->th_ack); 2987 ack -= is->is_isninc[1]; 2988 tcp->th_ack = htonl(ack); 2989 fix_incksum(&tcp->th_sum, is->is_sumd[1]); 2990 } 2991 } 2992 } 2993 2994 2995 /* ------------------------------------------------------------------------ */ 2996 /* Function: fr_statesync */ 2997 /* Returns: Nil */ 2998 /* Parameters: action(I) - type of synchronisation to do */ 2999 /* v(I) - IP version being sync'd (v4 or v6) */ 3000 /* ifp(I) - interface identifier associated with action */ 3001 /* name(I) - name associated with ifp parameter */ 3002 /* */ 3003 /* Walk through all state entries and if an interface pointer match is */ 3004 /* found then look it up again, based on its name in case the pointer has */ 3005 /* changed since last time. */ 3006 /* */ 3007 /* If ifp is passed in as being non-null then we are only doing updates for */ 3008 /* existing, matching, uses of it. */ 3009 /* ------------------------------------------------------------------------ */ 3010 void fr_statesync(action, v, ifp, name, ifs) 3011 int action, v; 3012 void *ifp; 3013 char *name; 3014 ipf_stack_t *ifs; 3015 { 3016 ipstate_t *is; 3017 int i; 3018 3019 if (ifs->ifs_fr_running <= 0) 3020 return; 3021 3022 WRITE_ENTER(&ifs->ifs_ipf_state); 3023 3024 if (ifs->ifs_fr_running <= 0) { 3025 RWLOCK_EXIT(&ifs->ifs_ipf_state); 3026 return; 3027 } 3028 3029 switch (action) 3030 { 3031 case IPFSYNC_RESYNC : 3032 for (is = ifs->ifs_ips_list; is; is = is->is_next) { 3033 if (v != 0 && is->is_v != v) 3034 continue; 3035 /* 3036 * Look up all the interface names in the state entry. 3037 */ 3038 for (i = 0; i < 4; i++) { 3039 is->is_ifp[i] = fr_resolvenic(is->is_ifname[i], 3040 is->is_v, ifs); 3041 } 3042 } 3043 break; 3044 case IPFSYNC_NEWIFP : 3045 for (is = ifs->ifs_ips_list; is; is = is->is_next) { 3046 if (v != 0 && is->is_v != v) 3047 continue; 3048 /* 3049 * Look up all the interface names in the state entry. 3050 */ 3051 for (i = 0; i < 4; i++) { 3052 if (!strncmp(is->is_ifname[i], name, 3053 sizeof(is->is_ifname[i]))) 3054 is->is_ifp[i] = ifp; 3055 } 3056 } 3057 break; 3058 case IPFSYNC_OLDIFP : 3059 for (is = ifs->ifs_ips_list; is; is = is->is_next) { 3060 if (v != 0 && is->is_v != v) 3061 continue; 3062 /* 3063 * Look up all the interface names in the state entry. 3064 */ 3065 for (i = 0; i < 4; i++) { 3066 if (is->is_ifp[i] == ifp) 3067 is->is_ifp[i] = (void *)-1; 3068 } 3069 } 3070 break; 3071 } 3072 RWLOCK_EXIT(&ifs->ifs_ipf_state); 3073 } 3074 3075 3076 /* ------------------------------------------------------------------------ */ 3077 /* Function: fr_delstate */ 3078 /* Returns: Nil */ 3079 /* Parameters: is(I) - pointer to state structure to delete */ 3080 /* why(I) - if not 0, log reason why it was deleted */ 3081 /* Write Locks: ipf_state/ipf_global */ 3082 /* */ 3083 /* Deletes a state entry from the enumerated list as well as the hash table */ 3084 /* and timeout queue lists. Make adjustments to hash table statistics and */ 3085 /* global counters as required. */ 3086 /* ------------------------------------------------------------------------ */ 3087 static void fr_delstate(is, why, ifs) 3088 ipstate_t *is; 3089 int why; 3090 ipf_stack_t *ifs; 3091 { 3092 3093 ASSERT(rw_write_held(&ifs->ifs_ipf_global.ipf_lk) == 0 || 3094 rw_write_held(&ifs->ifs_ipf_state.ipf_lk) == 0); 3095 3096 /* 3097 * Since we want to delete this, remove it from the state table, 3098 * where it can be found & used, first. 3099 */ 3100 if (is->is_pnext != NULL) { 3101 *is->is_pnext = is->is_next; 3102 3103 if (is->is_next != NULL) 3104 is->is_next->is_pnext = is->is_pnext; 3105 3106 is->is_pnext = NULL; 3107 is->is_next = NULL; 3108 } 3109 3110 if (is->is_phnext != NULL) { 3111 *is->is_phnext = is->is_hnext; 3112 if (is->is_hnext != NULL) 3113 is->is_hnext->is_phnext = is->is_phnext; 3114 if (ifs->ifs_ips_table[is->is_hv] == NULL) 3115 ifs->ifs_ips_stats.iss_inuse--; 3116 ifs->ifs_ips_stats.iss_bucketlen[is->is_hv]--; 3117 3118 is->is_phnext = NULL; 3119 is->is_hnext = NULL; 3120 } 3121 3122 /* 3123 * Because ifs->ifs_ips_stats.iss_wild is a count of entries in the state 3124 * table that have wildcard flags set, only decerement it once 3125 * and do it here. 3126 */ 3127 if (is->is_flags & (SI_WILDP|SI_WILDA)) { 3128 if (!(is->is_flags & SI_CLONED)) { 3129 ATOMIC_DECL(ifs->ifs_ips_stats.iss_wild); 3130 } 3131 is->is_flags &= ~(SI_WILDP|SI_WILDA); 3132 } 3133 3134 /* 3135 * Next, remove it from the timeout queue it is in. 3136 */ 3137 fr_deletequeueentry(&is->is_sti); 3138 3139 is->is_me = NULL; 3140 3141 /* 3142 * If it is still in use by something else, do not go any further, 3143 * but note that at this point it is now an orphan. 3144 */ 3145 MUTEX_ENTER(&is->is_lock); 3146 if (is->is_ref > 1) { 3147 is->is_ref--; 3148 MUTEX_EXIT(&is->is_lock); 3149 return; 3150 } 3151 MUTEX_EXIT(&is->is_lock); 3152 3153 is->is_ref = 0; 3154 3155 if (is->is_tqehead[0] != NULL) 3156 (void) fr_deletetimeoutqueue(is->is_tqehead[0]); 3157 3158 if (is->is_tqehead[1] != NULL) 3159 (void) fr_deletetimeoutqueue(is->is_tqehead[1]); 3160 3161 #ifdef IPFILTER_SYNC 3162 if (is->is_sync) 3163 ipfsync_del(is->is_sync); 3164 #endif 3165 #ifdef IPFILTER_SCAN 3166 (void) ipsc_detachis(is); 3167 #endif 3168 3169 if (ifs->ifs_ipstate_logging != 0 && why != 0) 3170 ipstate_log(is, why, ifs); 3171 3172 if (is->is_rule != NULL) { 3173 is->is_rule->fr_statecnt--; 3174 (void)fr_derefrule(&is->is_rule, ifs); 3175 } 3176 3177 MUTEX_DESTROY(&is->is_lock); 3178 KFREE(is); 3179 ifs->ifs_ips_num--; 3180 } 3181 3182 3183 /* ------------------------------------------------------------------------ */ 3184 /* Function: fr_timeoutstate */ 3185 /* Returns: Nil */ 3186 /* Parameters: Nil */ 3187 /* */ 3188 /* Slowly expire held state for thingslike UDP and ICMP. The algorithm */ 3189 /* used here is to keep the queue sorted with the oldest things at the top */ 3190 /* and the youngest at the bottom. So if the top one doesn't need to be */ 3191 /* expired then neither will any under it. */ 3192 /* ------------------------------------------------------------------------ */ 3193 void fr_timeoutstate(ifs) 3194 ipf_stack_t *ifs; 3195 { 3196 ipftq_t *ifq, *ifqnext; 3197 ipftqent_t *tqe, *tqn; 3198 ipstate_t *is; 3199 SPL_INT(s); 3200 3201 SPL_NET(s); 3202 WRITE_ENTER(&ifs->ifs_ipf_state); 3203 for (ifq = ifs->ifs_ips_tqtqb; ifq != NULL; ifq = ifq->ifq_next) 3204 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3205 if (tqe->tqe_die > ifs->ifs_fr_ticks) 3206 break; 3207 tqn = tqe->tqe_next; 3208 is = tqe->tqe_parent; 3209 fr_delstate(is, ISL_EXPIRE, ifs); 3210 } 3211 3212 for (ifq = ifs->ifs_ips_utqe; ifq != NULL; ifq = ifq->ifq_next) { 3213 for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { 3214 if (tqe->tqe_die > ifs->ifs_fr_ticks) 3215 break; 3216 tqn = tqe->tqe_next; 3217 is = tqe->tqe_parent; 3218 fr_delstate(is, ISL_EXPIRE, ifs); 3219 } 3220 } 3221 3222 for (ifq = ifs->ifs_ips_utqe; ifq != NULL; ifq = ifqnext) { 3223 ifqnext = ifq->ifq_next; 3224 3225 if (((ifq->ifq_flags & IFQF_DELETE) != 0) && 3226 (ifq->ifq_ref == 0)) { 3227 fr_freetimeoutqueue(ifq, ifs); 3228 } 3229 } 3230 3231 if (ifs->ifs_fr_state_doflush) { 3232 (void) fr_state_flush(2, 0, ifs); 3233 ifs->ifs_fr_state_doflush = 0; 3234 } 3235 RWLOCK_EXIT(&ifs->ifs_ipf_state); 3236 SPL_X(s); 3237 } 3238 3239 3240 /* ------------------------------------------------------------------------ */ 3241 /* Function: fr_state_flush */ 3242 /* Returns: int - 0 == success, -1 == failure */ 3243 /* Parameters: Nil */ 3244 /* Write Locks: ipf_state */ 3245 /* */ 3246 /* Flush state tables. Three actions currently defined: */ 3247 /* which == 0 : flush all state table entries */ 3248 /* which == 1 : flush TCP connections which have started to close but are */ 3249 /* stuck for some reason. */ 3250 /* which == 2 : flush TCP connections which have been idle for a long time, */ 3251 /* starting at > 4 days idle and working back in successive half-*/ 3252 /* days to at most 12 hours old. If this fails to free enough */ 3253 /* slots then work backwards in half hour slots to 30 minutes. */ 3254 /* If that too fails, then work backwards in 30 second intervals */ 3255 /* for the last 30 minutes to at worst 30 seconds idle. */ 3256 /* ------------------------------------------------------------------------ */ 3257 static int fr_state_flush(which, proto, ifs) 3258 int which, proto; 3259 ipf_stack_t *ifs; 3260 { 3261 ipftq_t *ifq, *ifqnext; 3262 ipftqent_t *tqe, *tqn; 3263 ipstate_t *is, **isp; 3264 int delete, removed; 3265 long try, maxtick; 3266 u_long interval; 3267 SPL_INT(s); 3268 3269 removed = 0; 3270 3271 SPL_NET(s); 3272 for (isp = &ifs->ifs_ips_list; ((is = *isp) != NULL); ) { 3273 delete = 0; 3274 3275 if ((proto != 0) && (is->is_v != proto)) { 3276 isp = &is->is_next; 3277 continue; 3278 } 3279 3280 switch (which) 3281 { 3282 case 0 : 3283 delete = 1; 3284 break; 3285 case 1 : 3286 case 2 : 3287 if (is->is_p != IPPROTO_TCP) 3288 break; 3289 if ((is->is_state[0] != IPF_TCPS_ESTABLISHED) || 3290 (is->is_state[1] != IPF_TCPS_ESTABLISHED)) 3291 delete = 1; 3292 break; 3293 } 3294 3295 if (delete) { 3296 if (is->is_p == IPPROTO_TCP) 3297 ifs->ifs_ips_stats.iss_fin++; 3298 else 3299 ifs->ifs_ips_stats.iss_expire++; 3300 fr_delstate(is, ISL_FLUSH, ifs); 3301 removed++; 3302 } else 3303 isp = &is->is_next; 3304 } 3305 3306 if (which != 2) { 3307 SPL_X(s); 3308 return removed; 3309 } 3310 3311 /* 3312 * Asked to remove inactive entries because the table is full, try 3313 * again, 3 times, if first attempt failed with a different criteria 3314 * each time. The order tried in must be in decreasing age. 3315 * Another alternative is to implement random drop and drop N entries 3316 * at random until N have been freed up. 3317 */ 3318 if (ifs->ifs_fr_ticks - ifs->ifs_ips_last_force_flush < IPF_TTLVAL(5)) 3319 goto force_flush_skipped; 3320 ifs->ifs_ips_last_force_flush = ifs->ifs_fr_ticks; 3321 3322 if (ifs->ifs_fr_ticks > IPF_TTLVAL(43200)) 3323 interval = IPF_TTLVAL(43200); 3324 else if (ifs->ifs_fr_ticks > IPF_TTLVAL(1800)) 3325 interval = IPF_TTLVAL(1800); 3326 else if (ifs->ifs_fr_ticks > IPF_TTLVAL(30)) 3327 interval = IPF_TTLVAL(30); 3328 else 3329 interval = IPF_TTLVAL(10); 3330 try = ifs->ifs_fr_ticks - (ifs->ifs_fr_ticks - interval); 3331 if (try < 0) 3332 goto force_flush_skipped; 3333 3334 while (removed == 0) { 3335 maxtick = ifs->ifs_fr_ticks - interval; 3336 if (maxtick < 0) 3337 break; 3338 3339 while (try < maxtick) { 3340 for (ifq = ifs->ifs_ips_tqtqb; ifq != NULL; 3341 ifq = ifq->ifq_next) { 3342 for (tqn = ifq->ifq_head; 3343 ((tqe = tqn) != NULL); ) { 3344 if (tqe->tqe_die > try) 3345 break; 3346 tqn = tqe->tqe_next; 3347 is = tqe->tqe_parent; 3348 fr_delstate(is, ISL_EXPIRE, ifs); 3349 removed++; 3350 } 3351 } 3352 3353 for (ifq = ifs->ifs_ips_utqe; ifq != NULL; ifq = ifqnext) { 3354 ifqnext = ifq->ifq_next; 3355 3356 for (tqn = ifq->ifq_head; 3357 ((tqe = tqn) != NULL); ) { 3358 if (tqe->tqe_die > try) 3359 break; 3360 tqn = tqe->tqe_next; 3361 is = tqe->tqe_parent; 3362 fr_delstate(is, ISL_EXPIRE, ifs); 3363 removed++; 3364 } 3365 } 3366 if (try + interval > maxtick) 3367 break; 3368 try += interval; 3369 } 3370 3371 if (removed == 0) { 3372 if (interval == IPF_TTLVAL(43200)) { 3373 interval = IPF_TTLVAL(1800); 3374 } else if (interval == IPF_TTLVAL(1800)) { 3375 interval = IPF_TTLVAL(30); 3376 } else if (interval == IPF_TTLVAL(30)) { 3377 interval = IPF_TTLVAL(10); 3378 } else { 3379 break; 3380 } 3381 } 3382 } 3383 force_flush_skipped: 3384 SPL_X(s); 3385 return removed; 3386 } 3387 3388 3389 3390 /* ------------------------------------------------------------------------ */ 3391 /* Function: fr_tcp_age */ 3392 /* Returns: int - 1 == state transition made, 0 == no change (rejected) */ 3393 /* Parameters: tq(I) - pointer to timeout queue information */ 3394 /* fin(I) - pointer to packet information */ 3395 /* tqtab(I) - TCP timeout queue table this is in */ 3396 /* flags(I) - flags from state/NAT entry */ 3397 /* */ 3398 /* Rewritten by Arjan de Vet <Arjan.deVet@adv.iae.nl>, 2000-07-29: */ 3399 /* */ 3400 /* - (try to) base state transitions on real evidence only, */ 3401 /* i.e. packets that are sent and have been received by ipfilter; */ 3402 /* diagram 18.12 of TCP/IP volume 1 by W. Richard Stevens was used. */ 3403 /* */ 3404 /* - deal with half-closed connections correctly; */ 3405 /* */ 3406 /* - store the state of the source in state[0] such that ipfstat */ 3407 /* displays the state as source/dest instead of dest/source; the calls */ 3408 /* to fr_tcp_age have been changed accordingly. */ 3409 /* */ 3410 /* Internal Parameters: */ 3411 /* */ 3412 /* state[0] = state of source (host that initiated connection) */ 3413 /* state[1] = state of dest (host that accepted the connection) */ 3414 /* */ 3415 /* dir == 0 : a packet from source to dest */ 3416 /* dir == 1 : a packet from dest to source */ 3417 /* */ 3418 /* Locking: it is assumed that the parent of the tqe structure is locked. */ 3419 /* ------------------------------------------------------------------------ */ 3420 int fr_tcp_age(tqe, fin, tqtab, flags) 3421 ipftqent_t *tqe; 3422 fr_info_t *fin; 3423 ipftq_t *tqtab; 3424 int flags; 3425 { 3426 int dlen, ostate, nstate, rval, dir; 3427 u_char tcpflags; 3428 tcphdr_t *tcp; 3429 ipf_stack_t *ifs = fin->fin_ifs; 3430 3431 tcp = fin->fin_dp; 3432 3433 rval = 0; 3434 dir = fin->fin_rev; 3435 tcpflags = tcp->th_flags; 3436 dlen = fin->fin_dlen - (TCP_OFF(tcp) << 2); 3437 3438 if (tcpflags & TH_RST) { 3439 if (!(tcpflags & TH_PUSH) && !dlen) 3440 nstate = IPF_TCPS_CLOSED; 3441 else 3442 nstate = IPF_TCPS_CLOSE_WAIT; 3443 rval = 1; 3444 } else { 3445 ostate = tqe->tqe_state[1 - dir]; 3446 nstate = tqe->tqe_state[dir]; 3447 3448 switch (nstate) 3449 { 3450 case IPF_TCPS_CLOSED: /* 0 */ 3451 if ((tcpflags & TH_OPENING) == TH_OPENING) { 3452 /* 3453 * 'dir' received an S and sends SA in 3454 * response, CLOSED -> SYN_RECEIVED 3455 */ 3456 nstate = IPF_TCPS_SYN_RECEIVED; 3457 rval = 1; 3458 } else if ((tcpflags & TH_OPENING) == TH_SYN) { 3459 /* 'dir' sent S, CLOSED -> SYN_SENT */ 3460 nstate = IPF_TCPS_SYN_SENT; 3461 rval = 1; 3462 } 3463 /* 3464 * the next piece of code makes it possible to get 3465 * already established connections into the state table 3466 * after a restart or reload of the filter rules; this 3467 * does not work when a strict 'flags S keep state' is 3468 * used for tcp connections of course 3469 */ 3470 if (((flags & IS_TCPFSM) == 0) && 3471 ((tcpflags & TH_ACKMASK) == TH_ACK)) { 3472 /* 3473 * we saw an A, guess 'dir' is in ESTABLISHED 3474 * mode 3475 */ 3476 switch (ostate) 3477 { 3478 case IPF_TCPS_CLOSED : 3479 case IPF_TCPS_SYN_RECEIVED : 3480 nstate = IPF_TCPS_HALF_ESTAB; 3481 rval = 1; 3482 break; 3483 case IPF_TCPS_HALF_ESTAB : 3484 case IPF_TCPS_ESTABLISHED : 3485 nstate = IPF_TCPS_ESTABLISHED; 3486 rval = 1; 3487 break; 3488 default : 3489 break; 3490 } 3491 } 3492 /* 3493 * TODO: besides regular ACK packets we can have other 3494 * packets as well; it is yet to be determined how we 3495 * should initialize the states in those cases 3496 */ 3497 break; 3498 3499 case IPF_TCPS_LISTEN: /* 1 */ 3500 /* NOT USED */ 3501 break; 3502 3503 case IPF_TCPS_SYN_SENT: /* 2 */ 3504 if ((tcpflags & ~(TH_ECN|TH_CWR)) == TH_SYN) { 3505 /* 3506 * A retransmitted SYN packet. We do not reset 3507 * the timeout here to fr_tcptimeout because a 3508 * connection connect timeout does not renew 3509 * after every packet that is sent. We need to 3510 * set rval so as to indicate the packet has 3511 * passed the check for its flags being valid 3512 * in the TCP FSM. Setting rval to 2 has the 3513 * result of not resetting the timeout. 3514 */ 3515 rval = 2; 3516 } else if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == 3517 TH_ACK) { 3518 /* 3519 * we see an A from 'dir' which is in SYN_SENT 3520 * state: 'dir' sent an A in response to an SA 3521 * which it received, SYN_SENT -> ESTABLISHED 3522 */ 3523 nstate = IPF_TCPS_ESTABLISHED; 3524 rval = 1; 3525 } else if (tcpflags & TH_FIN) { 3526 /* 3527 * we see an F from 'dir' which is in SYN_SENT 3528 * state and wants to close its side of the 3529 * connection; SYN_SENT -> FIN_WAIT_1 3530 */ 3531 nstate = IPF_TCPS_FIN_WAIT_1; 3532 rval = 1; 3533 } else if ((tcpflags & TH_OPENING) == TH_OPENING) { 3534 /* 3535 * we see an SA from 'dir' which is already in 3536 * SYN_SENT state, this means we have a 3537 * simultaneous open; SYN_SENT -> SYN_RECEIVED 3538 */ 3539 nstate = IPF_TCPS_SYN_RECEIVED; 3540 rval = 1; 3541 } 3542 break; 3543 3544 case IPF_TCPS_SYN_RECEIVED: /* 3 */ 3545 if ((tcpflags & (TH_SYN|TH_FIN|TH_ACK)) == TH_ACK) { 3546 /* 3547 * we see an A from 'dir' which was in 3548 * SYN_RECEIVED state so it must now be in 3549 * established state, SYN_RECEIVED -> 3550 * ESTABLISHED 3551 */ 3552 nstate = IPF_TCPS_ESTABLISHED; 3553 rval = 1; 3554 } else if ((tcpflags & ~(TH_ECN|TH_CWR)) == 3555 TH_OPENING) { 3556 /* 3557 * We see an SA from 'dir' which is already in 3558 * SYN_RECEIVED state. 3559 */ 3560 rval = 2; 3561 } else if (tcpflags & TH_FIN) { 3562 /* 3563 * we see an F from 'dir' which is in 3564 * SYN_RECEIVED state and wants to close its 3565 * side of the connection; SYN_RECEIVED -> 3566 * FIN_WAIT_1 3567 */ 3568 nstate = IPF_TCPS_FIN_WAIT_1; 3569 rval = 1; 3570 } 3571 break; 3572 3573 case IPF_TCPS_HALF_ESTAB: /* 4 */ 3574 if (ostate >= IPF_TCPS_HALF_ESTAB) { 3575 if ((tcpflags & TH_ACKMASK) == TH_ACK) { 3576 nstate = IPF_TCPS_ESTABLISHED; 3577 } 3578 } 3579 rval = 1; 3580 3581 break; 3582 3583 case IPF_TCPS_ESTABLISHED: /* 5 */ 3584 rval = 1; 3585 if (tcpflags & TH_FIN) { 3586 /* 3587 * 'dir' closed its side of the connection; 3588 * this gives us a half-closed connection; 3589 * ESTABLISHED -> FIN_WAIT_1 3590 */ 3591 nstate = IPF_TCPS_FIN_WAIT_1; 3592 } else if (tcpflags & TH_ACK) { 3593 /* 3594 * an ACK, should we exclude other flags here? 3595 */ 3596 if (ostate == IPF_TCPS_FIN_WAIT_1) { 3597 /* 3598 * We know the other side did an active 3599 * close, so we are ACKing the recvd 3600 * FIN packet (does the window matching 3601 * code guarantee this?) and go into 3602 * CLOSE_WAIT state; this gives us a 3603 * half-closed connection 3604 */ 3605 nstate = IPF_TCPS_CLOSE_WAIT; 3606 } else if (ostate < IPF_TCPS_CLOSE_WAIT) { 3607 /* 3608 * still a fully established 3609 * connection reset timeout 3610 */ 3611 nstate = IPF_TCPS_ESTABLISHED; 3612 } 3613 } 3614 break; 3615 3616 case IPF_TCPS_CLOSE_WAIT: /* 6 */ 3617 rval = 1; 3618 if (tcpflags & TH_FIN) { 3619 /* 3620 * application closed and 'dir' sent a FIN, 3621 * we're now going into LAST_ACK state 3622 */ 3623 nstate = IPF_TCPS_LAST_ACK; 3624 } else { 3625 /* 3626 * we remain in CLOSE_WAIT because the other 3627 * side has closed already and we did not 3628 * close our side yet; reset timeout 3629 */ 3630 nstate = IPF_TCPS_CLOSE_WAIT; 3631 } 3632 break; 3633 3634 case IPF_TCPS_FIN_WAIT_1: /* 7 */ 3635 rval = 1; 3636 if ((tcpflags & TH_ACK) && 3637 ostate > IPF_TCPS_CLOSE_WAIT) { 3638 /* 3639 * if the other side is not active anymore 3640 * it has sent us a FIN packet that we are 3641 * ack'ing now with an ACK; this means both 3642 * sides have now closed the connection and 3643 * we go into TIME_WAIT 3644 */ 3645 /* 3646 * XXX: how do we know we really are ACKing 3647 * the FIN packet here? does the window code 3648 * guarantee that? 3649 */ 3650 nstate = IPF_TCPS_TIME_WAIT; 3651 } else { 3652 /* 3653 * we closed our side of the connection 3654 * already but the other side is still active 3655 * (ESTABLISHED/CLOSE_WAIT); continue with 3656 * this half-closed connection 3657 */ 3658 nstate = IPF_TCPS_FIN_WAIT_1; 3659 } 3660 break; 3661 3662 case IPF_TCPS_CLOSING: /* 8 */ 3663 /* NOT USED */ 3664 break; 3665 3666 case IPF_TCPS_LAST_ACK: /* 9 */ 3667 /* 3668 * We want to reset timer here to keep state in table. 3669 * If we would allow the state to time out here, while 3670 * there would still be packets being retransmitted, we 3671 * would cut off line between the two peers preventing 3672 * them to close connection properly. 3673 */ 3674 rval = 1; 3675 break; 3676 3677 case IPF_TCPS_FIN_WAIT_2: /* 10 */ 3678 rval = 1; 3679 if ((tcpflags & TH_OPENING) == TH_OPENING) 3680 nstate = IPF_TCPS_SYN_RECEIVED; 3681 else if (tcpflags & TH_SYN) 3682 nstate = IPF_TCPS_SYN_SENT; 3683 break; 3684 3685 case IPF_TCPS_TIME_WAIT: /* 11 */ 3686 /* we're in 2MSL timeout now */ 3687 rval = 1; 3688 break; 3689 3690 default : 3691 #if defined(_KERNEL) 3692 # if SOLARIS 3693 cmn_err(CE_NOTE, 3694 "tcp %lx flags %x si %lx nstate %d ostate %d\n", 3695 (u_long)tcp, tcpflags, (u_long)tqe, 3696 nstate, ostate); 3697 # else 3698 printf("tcp %lx flags %x si %lx nstate %d ostate %d\n", 3699 (u_long)tcp, tcpflags, (u_long)tqe, 3700 nstate, ostate); 3701 # endif 3702 #else 3703 abort(); 3704 #endif 3705 break; 3706 } 3707 } 3708 3709 /* 3710 * If rval == 2 then do not update the queue position, but treat the 3711 * packet as being ok. 3712 */ 3713 if (rval == 2) 3714 rval = 1; 3715 else if (rval == 1) { 3716 tqe->tqe_state[dir] = nstate; 3717 if ((tqe->tqe_flags & TQE_RULEBASED) == 0) 3718 fr_movequeue(tqe, tqe->tqe_ifq, tqtab + nstate, ifs); 3719 } 3720 3721 return rval; 3722 } 3723 3724 3725 /* ------------------------------------------------------------------------ */ 3726 /* Function: ipstate_log */ 3727 /* Returns: Nil */ 3728 /* Parameters: is(I) - pointer to state structure */ 3729 /* type(I) - type of log entry to create */ 3730 /* */ 3731 /* Creates a state table log entry using the state structure and type info. */ 3732 /* passed in. Log packet/byte counts, source/destination address and other */ 3733 /* protocol specific information. */ 3734 /* ------------------------------------------------------------------------ */ 3735 void ipstate_log(is, type, ifs) 3736 struct ipstate *is; 3737 u_int type; 3738 ipf_stack_t *ifs; 3739 { 3740 #ifdef IPFILTER_LOG 3741 struct ipslog ipsl; 3742 size_t sizes[1]; 3743 void *items[1]; 3744 int types[1]; 3745 3746 /* 3747 * Copy information out of the ipstate_t structure and into the 3748 * structure used for logging. 3749 */ 3750 ipsl.isl_type = type; 3751 ipsl.isl_pkts[0] = is->is_pkts[0] + is->is_icmppkts[0]; 3752 ipsl.isl_bytes[0] = is->is_bytes[0]; 3753 ipsl.isl_pkts[1] = is->is_pkts[1] + is->is_icmppkts[1]; 3754 ipsl.isl_bytes[1] = is->is_bytes[1]; 3755 ipsl.isl_pkts[2] = is->is_pkts[2] + is->is_icmppkts[2]; 3756 ipsl.isl_bytes[2] = is->is_bytes[2]; 3757 ipsl.isl_pkts[3] = is->is_pkts[3] + is->is_icmppkts[3]; 3758 ipsl.isl_bytes[3] = is->is_bytes[3]; 3759 ipsl.isl_src = is->is_src; 3760 ipsl.isl_dst = is->is_dst; 3761 ipsl.isl_p = is->is_p; 3762 ipsl.isl_v = is->is_v; 3763 ipsl.isl_flags = is->is_flags; 3764 ipsl.isl_tag = is->is_tag; 3765 ipsl.isl_rulen = is->is_rulen; 3766 (void) strncpy(ipsl.isl_group, is->is_group, FR_GROUPLEN); 3767 3768 if (ipsl.isl_p == IPPROTO_TCP || ipsl.isl_p == IPPROTO_UDP) { 3769 ipsl.isl_sport = is->is_sport; 3770 ipsl.isl_dport = is->is_dport; 3771 if (ipsl.isl_p == IPPROTO_TCP) { 3772 ipsl.isl_state[0] = is->is_state[0]; 3773 ipsl.isl_state[1] = is->is_state[1]; 3774 } 3775 } else if (ipsl.isl_p == IPPROTO_ICMP) { 3776 ipsl.isl_itype = is->is_icmp.ici_type; 3777 } else if (ipsl.isl_p == IPPROTO_ICMPV6) { 3778 ipsl.isl_itype = is->is_icmp.ici_type; 3779 } else { 3780 ipsl.isl_ps.isl_filler[0] = 0; 3781 ipsl.isl_ps.isl_filler[1] = 0; 3782 } 3783 3784 items[0] = &ipsl; 3785 sizes[0] = sizeof(ipsl); 3786 types[0] = 0; 3787 3788 if (ipllog(IPL_LOGSTATE, NULL, items, sizes, types, 1, ifs)) { 3789 ATOMIC_INCL(ifs->ifs_ips_stats.iss_logged); 3790 } else { 3791 ATOMIC_INCL(ifs->ifs_ips_stats.iss_logfail); 3792 } 3793 #endif 3794 } 3795 3796 3797 #ifdef USE_INET6 3798 /* ------------------------------------------------------------------------ */ 3799 /* Function: fr_checkicmp6matchingstate */ 3800 /* Returns: ipstate_t* - NULL == no match found, */ 3801 /* else pointer to matching state entry */ 3802 /* Parameters: fin(I) - pointer to packet information */ 3803 /* Locks: NULL == no locks, else Read Lock on ipf_state */ 3804 /* */ 3805 /* If we've got an ICMPv6 error message, using the information stored in */ 3806 /* the ICMPv6 packet, look for a matching state table entry. */ 3807 /* ------------------------------------------------------------------------ */ 3808 static ipstate_t *fr_checkicmp6matchingstate(fin) 3809 fr_info_t *fin; 3810 { 3811 struct icmp6_hdr *ic6, *oic; 3812 int backward, i; 3813 ipstate_t *is, **isp; 3814 u_short sport, dport; 3815 i6addr_t dst, src; 3816 u_short savelen; 3817 icmpinfo_t *ic; 3818 fr_info_t ofin; 3819 tcphdr_t *tcp; 3820 ip6_t *oip6; 3821 u_char pr; 3822 u_int hv; 3823 ipf_stack_t *ifs = fin->fin_ifs; 3824 3825 /* 3826 * Does it at least have the return (basic) IP header ? 3827 * Is it an actual recognised ICMP error type? 3828 * Only a basic IP header (no options) should be with 3829 * an ICMP error header. 3830 */ 3831 if ((fin->fin_v != 6) || (fin->fin_plen < ICMP6ERR_MINPKTLEN) || 3832 !(fin->fin_flx & FI_ICMPERR)) 3833 return NULL; 3834 3835 ic6 = fin->fin_dp; 3836 3837 oip6 = (ip6_t *)((char *)ic6 + ICMPERR_ICMPHLEN); 3838 if (fin->fin_plen < sizeof(*oip6)) 3839 return NULL; 3840 3841 bcopy((char *)fin, (char *)&ofin, sizeof(*fin)); 3842 ofin.fin_v = 6; 3843 ofin.fin_ifp = fin->fin_ifp; 3844 ofin.fin_out = !fin->fin_out; 3845 ofin.fin_m = NULL; /* if dereferenced, panic XXX */ 3846 ofin.fin_mp = NULL; /* if dereferenced, panic XXX */ 3847 3848 /* 3849 * We make a fin entry to be able to feed it to 3850 * matchsrcdst. Note that not all fields are necessary 3851 * but this is the cleanest way. Note further we fill 3852 * in fin_mp such that if someone uses it we'll get 3853 * a kernel panic. fr_matchsrcdst does not use this. 3854 * 3855 * watch out here, as ip is in host order and oip6 in network 3856 * order. Any change we make must be undone afterwards. 3857 */ 3858 savelen = oip6->ip6_plen; 3859 oip6->ip6_plen = fin->fin_dlen - ICMPERR_ICMPHLEN; 3860 ofin.fin_flx = FI_NOCKSUM; 3861 ofin.fin_ip = (ip_t *)oip6; 3862 ofin.fin_plen = oip6->ip6_plen; 3863 (void) fr_makefrip(sizeof(*oip6), (ip_t *)oip6, &ofin); 3864 ofin.fin_flx &= ~(FI_BAD|FI_SHORT); 3865 oip6->ip6_plen = savelen; 3866 3867 if (oip6->ip6_nxt == IPPROTO_ICMPV6) { 3868 oic = (struct icmp6_hdr *)(oip6 + 1); 3869 /* 3870 * an ICMP error can only be generated as a result of an 3871 * ICMP query, not as the response on an ICMP error 3872 * 3873 * XXX theoretically ICMP_ECHOREP and the other reply's are 3874 * ICMP query's as well, but adding them here seems strange XXX 3875 */ 3876 if (!(oic->icmp6_type & ICMP6_INFOMSG_MASK)) 3877 return NULL; 3878 3879 /* 3880 * perform a lookup of the ICMP packet in the state table 3881 */ 3882 hv = (pr = oip6->ip6_nxt); 3883 src.in6 = oip6->ip6_src; 3884 hv += src.in4.s_addr; 3885 dst.in6 = oip6->ip6_dst; 3886 hv += dst.in4.s_addr; 3887 hv += oic->icmp6_id; 3888 hv += oic->icmp6_seq; 3889 hv = DOUBLE_HASH(hv, ifs); 3890 3891 READ_ENTER(&ifs->ifs_ipf_state); 3892 for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) { 3893 ic = &is->is_icmp; 3894 isp = &is->is_hnext; 3895 if ((is->is_p == pr) && 3896 !(is->is_pass & FR_NOICMPERR) && 3897 (oic->icmp6_id == ic->ici_id) && 3898 (oic->icmp6_seq == ic->ici_seq) && 3899 (is = fr_matchsrcdst(&ofin, is, &src, 3900 &dst, NULL, FI_ICMPCMP))) { 3901 /* 3902 * in the state table ICMP query's are stored 3903 * with the type of the corresponding ICMP 3904 * response. Correct here 3905 */ 3906 if (((ic->ici_type == ICMP6_ECHO_REPLY) && 3907 (oic->icmp6_type == ICMP6_ECHO_REQUEST)) || 3908 (ic->ici_type - 1 == oic->icmp6_type )) { 3909 ifs->ifs_ips_stats.iss_hits++; 3910 backward = IP6_NEQ(&is->is_dst, &src); 3911 fin->fin_rev = !backward; 3912 i = (backward << 1) + fin->fin_out; 3913 is->is_icmppkts[i]++; 3914 return is; 3915 } 3916 } 3917 } 3918 RWLOCK_EXIT(&ifs->ifs_ipf_state); 3919 return NULL; 3920 } 3921 3922 hv = (pr = oip6->ip6_nxt); 3923 src.in6 = oip6->ip6_src; 3924 hv += src.i6[0]; 3925 hv += src.i6[1]; 3926 hv += src.i6[2]; 3927 hv += src.i6[3]; 3928 dst.in6 = oip6->ip6_dst; 3929 hv += dst.i6[0]; 3930 hv += dst.i6[1]; 3931 hv += dst.i6[2]; 3932 hv += dst.i6[3]; 3933 3934 if ((oip6->ip6_nxt == IPPROTO_TCP) || (oip6->ip6_nxt == IPPROTO_UDP)) { 3935 tcp = (tcphdr_t *)(oip6 + 1); 3936 dport = tcp->th_dport; 3937 sport = tcp->th_sport; 3938 hv += dport; 3939 hv += sport; 3940 } else 3941 tcp = NULL; 3942 hv = DOUBLE_HASH(hv, ifs); 3943 3944 READ_ENTER(&ifs->ifs_ipf_state); 3945 for (isp = &ifs->ifs_ips_table[hv]; ((is = *isp) != NULL); ) { 3946 isp = &is->is_hnext; 3947 /* 3948 * Only allow this icmp though if the 3949 * encapsulated packet was allowed through the 3950 * other way around. Note that the minimal amount 3951 * of info present does not allow for checking against 3952 * tcp internals such as seq and ack numbers. 3953 */ 3954 if ((is->is_p != pr) || (is->is_v != 6) || 3955 (is->is_pass & FR_NOICMPERR)) 3956 continue; 3957 is = fr_matchsrcdst(&ofin, is, &src, &dst, tcp, FI_ICMPCMP); 3958 if (is != NULL) { 3959 ifs->ifs_ips_stats.iss_hits++; 3960 backward = IP6_NEQ(&is->is_dst, &src); 3961 fin->fin_rev = !backward; 3962 i = (backward << 1) + fin->fin_out; 3963 is->is_icmppkts[i]++; 3964 /* 3965 * we deliberately do not touch the timeouts 3966 * for the accompanying state table entry. 3967 * It remains to be seen if that is correct. XXX 3968 */ 3969 return is; 3970 } 3971 } 3972 RWLOCK_EXIT(&ifs->ifs_ipf_state); 3973 return NULL; 3974 } 3975 #endif 3976 3977 3978 /* ------------------------------------------------------------------------ */ 3979 /* Function: fr_sttab_init */ 3980 /* Returns: Nil */ 3981 /* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 3982 /* */ 3983 /* Initialise the array of timeout queues for TCP. */ 3984 /* ------------------------------------------------------------------------ */ 3985 void fr_sttab_init(tqp, ifs) 3986 ipftq_t *tqp; 3987 ipf_stack_t *ifs; 3988 { 3989 int i; 3990 3991 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) { 3992 tqp[i].ifq_ttl = 0; 3993 tqp[i].ifq_ref = 1; 3994 tqp[i].ifq_head = NULL; 3995 tqp[i].ifq_tail = &tqp[i].ifq_head; 3996 tqp[i].ifq_next = tqp + i + 1; 3997 MUTEX_INIT(&tqp[i].ifq_lock, "ipftq tcp tab"); 3998 } 3999 tqp[IPF_TCP_NSTATES - 1].ifq_next = NULL; 4000 tqp[IPF_TCPS_CLOSED].ifq_ttl = ifs->ifs_fr_tcpclosed; 4001 tqp[IPF_TCPS_LISTEN].ifq_ttl = ifs->ifs_fr_tcptimeout; 4002 tqp[IPF_TCPS_SYN_SENT].ifq_ttl = ifs->ifs_fr_tcptimeout; 4003 tqp[IPF_TCPS_SYN_RECEIVED].ifq_ttl = ifs->ifs_fr_tcptimeout; 4004 tqp[IPF_TCPS_ESTABLISHED].ifq_ttl = ifs->ifs_fr_tcpidletimeout; 4005 tqp[IPF_TCPS_CLOSE_WAIT].ifq_ttl = ifs->ifs_fr_tcphalfclosed; 4006 tqp[IPF_TCPS_FIN_WAIT_1].ifq_ttl = ifs->ifs_fr_tcphalfclosed; 4007 tqp[IPF_TCPS_CLOSING].ifq_ttl = ifs->ifs_fr_tcptimeout; 4008 tqp[IPF_TCPS_LAST_ACK].ifq_ttl = ifs->ifs_fr_tcplastack; 4009 tqp[IPF_TCPS_FIN_WAIT_2].ifq_ttl = ifs->ifs_fr_tcpclosewait; 4010 tqp[IPF_TCPS_TIME_WAIT].ifq_ttl = ifs->ifs_fr_tcptimeout; 4011 tqp[IPF_TCPS_HALF_ESTAB].ifq_ttl = ifs->ifs_fr_tcptimeout; 4012 } 4013 4014 4015 /* ------------------------------------------------------------------------ */ 4016 /* Function: fr_sttab_destroy */ 4017 /* Returns: Nil */ 4018 /* Parameters: tqp(I) - pointer to an array of timeout queues for TCP */ 4019 /* */ 4020 /* Do whatever is necessary to "destroy" each of the entries in the array */ 4021 /* of timeout queues for TCP. */ 4022 /* ------------------------------------------------------------------------ */ 4023 void fr_sttab_destroy(tqp) 4024 ipftq_t *tqp; 4025 { 4026 int i; 4027 4028 for (i = IPF_TCP_NSTATES - 1; i >= 0; i--) 4029 MUTEX_DESTROY(&tqp[i].ifq_lock); 4030 } 4031 4032 4033 /* ------------------------------------------------------------------------ */ 4034 /* Function: fr_statederef */ 4035 /* Returns: Nil */ 4036 /* Parameters: isp(I) - pointer to pointer to state table entry */ 4037 /* ifs - ipf stack instance */ 4038 /* */ 4039 /* Decrement the reference counter for this state table entry and free it */ 4040 /* if there are no more things using it. */ 4041 /* */ 4042 /* Internal parameters: */ 4043 /* state[0] = state of source (host that initiated connection) */ 4044 /* state[1] = state of dest (host that accepted the connection) */ 4045 /* ------------------------------------------------------------------------ */ 4046 void fr_statederef(isp, ifs) 4047 ipstate_t **isp; 4048 ipf_stack_t *ifs; 4049 { 4050 ipstate_t *is; 4051 4052 is = *isp; 4053 *isp = NULL; 4054 4055 MUTEX_ENTER(&is->is_lock); 4056 if (is->is_ref > 1) { 4057 is->is_ref--; 4058 MUTEX_EXIT(&is->is_lock); 4059 #ifndef _KERNEL 4060 if ((is->is_sti.tqe_state[0] > IPF_TCPS_ESTABLISHED) || 4061 (is->is_sti.tqe_state[1] > IPF_TCPS_ESTABLISHED)) { 4062 fr_delstate(is, ISL_ORPHAN, ifs); 4063 } 4064 #endif 4065 return; 4066 } 4067 MUTEX_EXIT(&is->is_lock); 4068 4069 WRITE_ENTER(&ifs->ifs_ipf_state); 4070 fr_delstate(is, ISL_EXPIRE, ifs); 4071 RWLOCK_EXIT(&ifs->ifs_ipf_state); 4072 } 4073 4074 4075 /* ------------------------------------------------------------------------ */ 4076 /* Function: fr_setstatequeue */ 4077 /* Returns: Nil */ 4078 /* Parameters: is(I) - pointer to state structure */ 4079 /* rev(I) - forward(0) or reverse(1) direction */ 4080 /* Locks: ipf_state (read or write) */ 4081 /* */ 4082 /* Put the state entry on its default queue entry, using rev as a helped in */ 4083 /* determining which queue it should be placed on. */ 4084 /* ------------------------------------------------------------------------ */ 4085 void fr_setstatequeue(is, rev, ifs) 4086 ipstate_t *is; 4087 int rev; 4088 ipf_stack_t *ifs; 4089 { 4090 ipftq_t *oifq, *nifq; 4091 4092 4093 if ((is->is_sti.tqe_flags & TQE_RULEBASED) != 0) 4094 nifq = is->is_tqehead[rev]; 4095 else 4096 nifq = NULL; 4097 4098 if (nifq == NULL) { 4099 switch (is->is_p) 4100 { 4101 #ifdef USE_INET6 4102 case IPPROTO_ICMPV6 : 4103 if (rev == 1) 4104 nifq = &ifs->ifs_ips_icmpacktq; 4105 else 4106 nifq = &ifs->ifs_ips_icmptq; 4107 break; 4108 #endif 4109 case IPPROTO_ICMP : 4110 if (rev == 1) 4111 nifq = &ifs->ifs_ips_icmpacktq; 4112 else 4113 nifq = &ifs->ifs_ips_icmptq; 4114 break; 4115 case IPPROTO_TCP : 4116 nifq = ifs->ifs_ips_tqtqb + is->is_state[rev]; 4117 break; 4118 4119 case IPPROTO_UDP : 4120 if (rev == 1) 4121 nifq = &ifs->ifs_ips_udpacktq; 4122 else 4123 nifq = &ifs->ifs_ips_udptq; 4124 break; 4125 4126 default : 4127 nifq = &ifs->ifs_ips_iptq; 4128 break; 4129 } 4130 } 4131 4132 oifq = is->is_sti.tqe_ifq; 4133 /* 4134 * If it's currently on a timeout queue, move it from one queue to 4135 * another, else put it on the end of the newly determined queue. 4136 */ 4137 if (oifq != NULL) 4138 fr_movequeue(&is->is_sti, oifq, nifq, ifs); 4139 else 4140 fr_queueappend(&is->is_sti, nifq, is, ifs); 4141 return; 4142 } 4143 4144 4145 /* ------------------------------------------------------------------------ */ 4146 /* Function: fr_stateiter */ 4147 /* Returns: int - 0 == success, else error */ 4148 /* Parameters: token(I) - pointer to ipftoken structure */ 4149 /* itp(I) - pointer to ipfgeniter structure */ 4150 /* */ 4151 /* This function handles the SIOCGENITER ioctl for the state tables and */ 4152 /* walks through the list of entries in the state table list (ips_list.) */ 4153 /* ------------------------------------------------------------------------ */ 4154 static int fr_stateiter(token, itp, ifs) 4155 ipftoken_t *token; 4156 ipfgeniter_t *itp; 4157 ipf_stack_t *ifs; 4158 { 4159 ipstate_t *is, *next, zero; 4160 int error, count; 4161 char *dst; 4162 4163 if (itp->igi_data == NULL) 4164 return EFAULT; 4165 4166 if (itp->igi_nitems == 0) 4167 return EINVAL; 4168 4169 if (itp->igi_type != IPFGENITER_STATE) 4170 return EINVAL; 4171 4172 error = 0; 4173 4174 READ_ENTER(&ifs->ifs_ipf_state); 4175 4176 /* 4177 * Get "previous" entry from the token and find the next entry. 4178 */ 4179 is = token->ipt_data; 4180 if (is == NULL) { 4181 next = ifs->ifs_ips_list; 4182 } else { 4183 next = is->is_next; 4184 } 4185 4186 dst = itp->igi_data; 4187 for (count = itp->igi_nitems; count > 0; count--) { 4188 /* 4189 * If we found an entry, add a reference to it and update the token. 4190 * Otherwise, zero out data to be returned and NULL out token. 4191 */ 4192 if (next != NULL) { 4193 MUTEX_ENTER(&next->is_lock); 4194 next->is_ref++; 4195 MUTEX_EXIT(&next->is_lock); 4196 token->ipt_data = next; 4197 } else { 4198 bzero(&zero, sizeof(zero)); 4199 next = &zero; 4200 token->ipt_data = NULL; 4201 } 4202 4203 /* 4204 * Safe to release lock now the we have a reference. 4205 */ 4206 RWLOCK_EXIT(&ifs->ifs_ipf_state); 4207 4208 /* 4209 * Copy out data and clean up references and tokens. 4210 */ 4211 error = COPYOUT(next, dst, sizeof(*next)); 4212 if (error != 0) 4213 error = EFAULT; 4214 if (token->ipt_data == NULL) { 4215 ipf_freetoken(token, ifs); 4216 break; 4217 } else { 4218 if (is != NULL) 4219 fr_statederef(&is, ifs); 4220 if (next->is_next == NULL) { 4221 ipf_freetoken(token, ifs); 4222 break; 4223 } 4224 } 4225 4226 if ((count == 1) || (error != 0)) 4227 break; 4228 4229 READ_ENTER(&ifs->ifs_ipf_state); 4230 dst += sizeof(*next); 4231 is = next; 4232 next = is->is_next; 4233 } 4234 4235 return error; 4236 } 4237