nf_conntrack_core.c revision 5b1158e909ecbe1a052203e0d8df15633f829930
1/* Connection state tracking for netfilter. This is separated from, 2 but required by, the NAT layer; it can also be used by an iptables 3 extension. */ 4 5/* (C) 1999-2001 Paul `Rusty' Russell 6 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org> 7 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org> 8 * 9 * This program is free software; you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License version 2 as 11 * published by the Free Software Foundation. 12 * 13 * 23 Apr 2001: Harald Welte <laforge@gnumonks.org> 14 * - new API and handling of conntrack/nat helpers 15 * - now capable of multiple expectations for one master 16 * 16 Jul 2002: Harald Welte <laforge@gnumonks.org> 17 * - add usage/reference counts to ip_conntrack_expect 18 * - export ip_conntrack[_expect]_{find_get,put} functions 19 * 16 Dec 2003: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp> 20 * - generalize L3 protocol denendent part. 21 * 23 Mar 2004: Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp> 22 * - add support various size of conntrack structures. 23 * 26 Jan 2006: Harald Welte <laforge@netfilter.org> 24 * - restructure nf_conn (introduce nf_conn_help) 25 * - redesign 'features' how they were originally intended 26 * 26 Feb 2006: Pablo Neira Ayuso <pablo@eurodev.net> 27 * - add support for L3 protocol module load on demand. 28 * 29 * Derived from net/ipv4/netfilter/ip_conntrack_core.c 30 */ 31 32#include <linux/types.h> 33#include <linux/netfilter.h> 34#include <linux/module.h> 35#include <linux/skbuff.h> 36#include <linux/proc_fs.h> 37#include <linux/vmalloc.h> 38#include <linux/stddef.h> 39#include <linux/slab.h> 40#include <linux/random.h> 41#include <linux/jhash.h> 42#include <linux/err.h> 43#include <linux/percpu.h> 44#include <linux/moduleparam.h> 45#include <linux/notifier.h> 46#include <linux/kernel.h> 47#include <linux/netdevice.h> 48#include <linux/socket.h> 49 50#include <net/netfilter/nf_conntrack.h> 51#include <net/netfilter/nf_conntrack_l3proto.h> 52#include <net/netfilter/nf_conntrack_l4proto.h> 53#include <net/netfilter/nf_conntrack_expect.h> 54#include <net/netfilter/nf_conntrack_helper.h> 55#include <net/netfilter/nf_conntrack_core.h> 56 57#define NF_CONNTRACK_VERSION "0.5.0" 58 59#if 0 60#define DEBUGP printk 61#else 62#define DEBUGP(format, args...) 63#endif 64 65DEFINE_RWLOCK(nf_conntrack_lock); 66 67/* nf_conntrack_standalone needs this */ 68atomic_t nf_conntrack_count = ATOMIC_INIT(0); 69EXPORT_SYMBOL_GPL(nf_conntrack_count); 70 71void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL; 72unsigned int nf_conntrack_htable_size __read_mostly; 73int nf_conntrack_max __read_mostly; 74EXPORT_SYMBOL_GPL(nf_conntrack_max); 75struct list_head *nf_conntrack_hash __read_mostly; 76struct nf_conn nf_conntrack_untracked __read_mostly; 77unsigned int nf_ct_log_invalid __read_mostly; 78LIST_HEAD(unconfirmed); 79static int nf_conntrack_vmalloc __read_mostly; 80 81static unsigned int nf_conntrack_next_id; 82 83DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); 84EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); 85 86/* 87 * This scheme offers various size of "struct nf_conn" dependent on 88 * features(helper, nat, ...) 89 */ 90 91#define NF_CT_FEATURES_NAMELEN 256 92static struct { 93 /* name of slab cache. printed in /proc/slabinfo */ 94 char *name; 95 96 /* size of slab cache */ 97 size_t size; 98 99 /* slab cache pointer */ 100 kmem_cache_t *cachep; 101 102 /* allocated slab cache + modules which uses this slab cache */ 103 int use; 104 105} nf_ct_cache[NF_CT_F_NUM]; 106 107/* protect members of nf_ct_cache except of "use" */ 108DEFINE_RWLOCK(nf_ct_cache_lock); 109 110/* This avoids calling kmem_cache_create() with same name simultaneously */ 111static DEFINE_MUTEX(nf_ct_cache_mutex); 112 113static int nf_conntrack_hash_rnd_initted; 114static unsigned int nf_conntrack_hash_rnd; 115 116static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 117 unsigned int size, unsigned int rnd) 118{ 119 unsigned int a, b; 120 a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all), 121 ((tuple->src.l3num) << 16) | tuple->dst.protonum); 122 b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all), 123 (tuple->src.u.all << 16) | tuple->dst.u.all); 124 125 return jhash_2words(a, b, rnd) % size; 126} 127 128static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) 129{ 130 return __hash_conntrack(tuple, nf_conntrack_htable_size, 131 nf_conntrack_hash_rnd); 132} 133 134int nf_conntrack_register_cache(u_int32_t features, const char *name, 135 size_t size) 136{ 137 int ret = 0; 138 char *cache_name; 139 kmem_cache_t *cachep; 140 141 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n", 142 features, name, size); 143 144 if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) { 145 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n", 146 features); 147 return -EINVAL; 148 } 149 150 mutex_lock(&nf_ct_cache_mutex); 151 152 write_lock_bh(&nf_ct_cache_lock); 153 /* e.g: multiple helpers are loaded */ 154 if (nf_ct_cache[features].use > 0) { 155 DEBUGP("nf_conntrack_register_cache: already resisterd.\n"); 156 if ((!strncmp(nf_ct_cache[features].name, name, 157 NF_CT_FEATURES_NAMELEN)) 158 && nf_ct_cache[features].size == size) { 159 DEBUGP("nf_conntrack_register_cache: reusing.\n"); 160 nf_ct_cache[features].use++; 161 ret = 0; 162 } else 163 ret = -EBUSY; 164 165 write_unlock_bh(&nf_ct_cache_lock); 166 mutex_unlock(&nf_ct_cache_mutex); 167 return ret; 168 } 169 write_unlock_bh(&nf_ct_cache_lock); 170 171 /* 172 * The memory space for name of slab cache must be alive until 173 * cache is destroyed. 174 */ 175 cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC); 176 if (cache_name == NULL) { 177 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n"); 178 ret = -ENOMEM; 179 goto out_up_mutex; 180 } 181 182 if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN) 183 >= NF_CT_FEATURES_NAMELEN) { 184 printk("nf_conntrack_register_cache: name too long\n"); 185 ret = -EINVAL; 186 goto out_free_name; 187 } 188 189 cachep = kmem_cache_create(cache_name, size, 0, 0, 190 NULL, NULL); 191 if (!cachep) { 192 printk("nf_conntrack_register_cache: Can't create slab cache " 193 "for the features = 0x%x\n", features); 194 ret = -ENOMEM; 195 goto out_free_name; 196 } 197 198 write_lock_bh(&nf_ct_cache_lock); 199 nf_ct_cache[features].use = 1; 200 nf_ct_cache[features].size = size; 201 nf_ct_cache[features].cachep = cachep; 202 nf_ct_cache[features].name = cache_name; 203 write_unlock_bh(&nf_ct_cache_lock); 204 205 goto out_up_mutex; 206 207out_free_name: 208 kfree(cache_name); 209out_up_mutex: 210 mutex_unlock(&nf_ct_cache_mutex); 211 return ret; 212} 213 214/* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */ 215void nf_conntrack_unregister_cache(u_int32_t features) 216{ 217 kmem_cache_t *cachep; 218 char *name; 219 220 /* 221 * This assures that kmem_cache_create() isn't called before destroying 222 * slab cache. 223 */ 224 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features); 225 mutex_lock(&nf_ct_cache_mutex); 226 227 write_lock_bh(&nf_ct_cache_lock); 228 if (--nf_ct_cache[features].use > 0) { 229 write_unlock_bh(&nf_ct_cache_lock); 230 mutex_unlock(&nf_ct_cache_mutex); 231 return; 232 } 233 cachep = nf_ct_cache[features].cachep; 234 name = nf_ct_cache[features].name; 235 nf_ct_cache[features].cachep = NULL; 236 nf_ct_cache[features].name = NULL; 237 nf_ct_cache[features].size = 0; 238 write_unlock_bh(&nf_ct_cache_lock); 239 240 synchronize_net(); 241 242 kmem_cache_destroy(cachep); 243 kfree(name); 244 245 mutex_unlock(&nf_ct_cache_mutex); 246} 247 248int 249nf_ct_get_tuple(const struct sk_buff *skb, 250 unsigned int nhoff, 251 unsigned int dataoff, 252 u_int16_t l3num, 253 u_int8_t protonum, 254 struct nf_conntrack_tuple *tuple, 255 const struct nf_conntrack_l3proto *l3proto, 256 const struct nf_conntrack_l4proto *l4proto) 257{ 258 NF_CT_TUPLE_U_BLANK(tuple); 259 260 tuple->src.l3num = l3num; 261 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 262 return 0; 263 264 tuple->dst.protonum = protonum; 265 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 266 267 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 268} 269 270int 271nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 272 const struct nf_conntrack_tuple *orig, 273 const struct nf_conntrack_l3proto *l3proto, 274 const struct nf_conntrack_l4proto *l4proto) 275{ 276 NF_CT_TUPLE_U_BLANK(inverse); 277 278 inverse->src.l3num = orig->src.l3num; 279 if (l3proto->invert_tuple(inverse, orig) == 0) 280 return 0; 281 282 inverse->dst.dir = !orig->dst.dir; 283 284 inverse->dst.protonum = orig->dst.protonum; 285 return l4proto->invert_tuple(inverse, orig); 286} 287 288static void 289clean_from_lists(struct nf_conn *ct) 290{ 291 DEBUGP("clean_from_lists(%p)\n", ct); 292 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 293 list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list); 294 295 /* Destroy all pending expectations */ 296 nf_ct_remove_expectations(ct); 297} 298 299static void 300destroy_conntrack(struct nf_conntrack *nfct) 301{ 302 struct nf_conn *ct = (struct nf_conn *)nfct; 303 struct nf_conntrack_l3proto *l3proto; 304 struct nf_conntrack_l4proto *l4proto; 305 306 DEBUGP("destroy_conntrack(%p)\n", ct); 307 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 308 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 309 310 nf_conntrack_event(IPCT_DESTROY, ct); 311 set_bit(IPS_DYING_BIT, &ct->status); 312 313 /* To make sure we don't get any weird locking issues here: 314 * destroy_conntrack() MUST NOT be called with a write lock 315 * to nf_conntrack_lock!!! -HW */ 316 l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num); 317 if (l3proto && l3proto->destroy) 318 l3proto->destroy(ct); 319 320 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); 321 if (l4proto && l4proto->destroy) 322 l4proto->destroy(ct); 323 324 if (nf_conntrack_destroyed) 325 nf_conntrack_destroyed(ct); 326 327 write_lock_bh(&nf_conntrack_lock); 328 /* Expectations will have been removed in clean_from_lists, 329 * except TFTP can create an expectation on the first packet, 330 * before connection is in the list, so we need to clean here, 331 * too. */ 332 nf_ct_remove_expectations(ct); 333 334 /* We overload first tuple to link into unconfirmed list. */ 335 if (!nf_ct_is_confirmed(ct)) { 336 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); 337 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 338 } 339 340 NF_CT_STAT_INC(delete); 341 write_unlock_bh(&nf_conntrack_lock); 342 343 if (ct->master) 344 nf_ct_put(ct->master); 345 346 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); 347 nf_conntrack_free(ct); 348} 349 350static void death_by_timeout(unsigned long ul_conntrack) 351{ 352 struct nf_conn *ct = (void *)ul_conntrack; 353 354 write_lock_bh(&nf_conntrack_lock); 355 /* Inside lock so preempt is disabled on module removal path. 356 * Otherwise we can get spurious warnings. */ 357 NF_CT_STAT_INC(delete_list); 358 clean_from_lists(ct); 359 write_unlock_bh(&nf_conntrack_lock); 360 nf_ct_put(ct); 361} 362 363struct nf_conntrack_tuple_hash * 364__nf_conntrack_find(const struct nf_conntrack_tuple *tuple, 365 const struct nf_conn *ignored_conntrack) 366{ 367 struct nf_conntrack_tuple_hash *h; 368 unsigned int hash = hash_conntrack(tuple); 369 370 list_for_each_entry(h, &nf_conntrack_hash[hash], list) { 371 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && 372 nf_ct_tuple_equal(tuple, &h->tuple)) { 373 NF_CT_STAT_INC(found); 374 return h; 375 } 376 NF_CT_STAT_INC(searched); 377 } 378 379 return NULL; 380} 381 382/* Find a connection corresponding to a tuple. */ 383struct nf_conntrack_tuple_hash * 384nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple, 385 const struct nf_conn *ignored_conntrack) 386{ 387 struct nf_conntrack_tuple_hash *h; 388 389 read_lock_bh(&nf_conntrack_lock); 390 h = __nf_conntrack_find(tuple, ignored_conntrack); 391 if (h) 392 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); 393 read_unlock_bh(&nf_conntrack_lock); 394 395 return h; 396} 397 398static void __nf_conntrack_hash_insert(struct nf_conn *ct, 399 unsigned int hash, 400 unsigned int repl_hash) 401{ 402 ct->id = ++nf_conntrack_next_id; 403 list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list, 404 &nf_conntrack_hash[hash]); 405 list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list, 406 &nf_conntrack_hash[repl_hash]); 407} 408 409void nf_conntrack_hash_insert(struct nf_conn *ct) 410{ 411 unsigned int hash, repl_hash; 412 413 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 414 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 415 416 write_lock_bh(&nf_conntrack_lock); 417 __nf_conntrack_hash_insert(ct, hash, repl_hash); 418 write_unlock_bh(&nf_conntrack_lock); 419} 420 421/* Confirm a connection given skb; places it in hash table */ 422int 423__nf_conntrack_confirm(struct sk_buff **pskb) 424{ 425 unsigned int hash, repl_hash; 426 struct nf_conntrack_tuple_hash *h; 427 struct nf_conn *ct; 428 struct nf_conn_help *help; 429 enum ip_conntrack_info ctinfo; 430 431 ct = nf_ct_get(*pskb, &ctinfo); 432 433 /* ipt_REJECT uses nf_conntrack_attach to attach related 434 ICMP/TCP RST packets in other direction. Actual packet 435 which created connection will be IP_CT_NEW or for an 436 expected connection, IP_CT_RELATED. */ 437 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 438 return NF_ACCEPT; 439 440 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 441 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 442 443 /* We're not in hash table, and we refuse to set up related 444 connections for unconfirmed conns. But packet copies and 445 REJECT will give spurious warnings here. */ 446 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 447 448 /* No external references means noone else could have 449 confirmed us. */ 450 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 451 DEBUGP("Confirming conntrack %p\n", ct); 452 453 write_lock_bh(&nf_conntrack_lock); 454 455 /* See if there's one in the list already, including reverse: 456 NAT could have grabbed it without realizing, since we're 457 not in the hash. If there is, we lost race. */ 458 list_for_each_entry(h, &nf_conntrack_hash[hash], list) 459 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 460 &h->tuple)) 461 goto out; 462 list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list) 463 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 464 &h->tuple)) 465 goto out; 466 467 /* Remove from unconfirmed list */ 468 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 469 470 __nf_conntrack_hash_insert(ct, hash, repl_hash); 471 /* Timer relative to confirmation time, not original 472 setting time, otherwise we'd get timer wrap in 473 weird delay cases. */ 474 ct->timeout.expires += jiffies; 475 add_timer(&ct->timeout); 476 atomic_inc(&ct->ct_general.use); 477 set_bit(IPS_CONFIRMED_BIT, &ct->status); 478 NF_CT_STAT_INC(insert); 479 write_unlock_bh(&nf_conntrack_lock); 480 help = nfct_help(ct); 481 if (help && help->helper) 482 nf_conntrack_event_cache(IPCT_HELPER, *pskb); 483#ifdef CONFIG_NF_NAT_NEEDED 484 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || 485 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) 486 nf_conntrack_event_cache(IPCT_NATINFO, *pskb); 487#endif 488 nf_conntrack_event_cache(master_ct(ct) ? 489 IPCT_RELATED : IPCT_NEW, *pskb); 490 return NF_ACCEPT; 491 492out: 493 NF_CT_STAT_INC(insert_failed); 494 write_unlock_bh(&nf_conntrack_lock); 495 return NF_DROP; 496} 497 498/* Returns true if a connection correspondings to the tuple (required 499 for NAT). */ 500int 501nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 502 const struct nf_conn *ignored_conntrack) 503{ 504 struct nf_conntrack_tuple_hash *h; 505 506 read_lock_bh(&nf_conntrack_lock); 507 h = __nf_conntrack_find(tuple, ignored_conntrack); 508 read_unlock_bh(&nf_conntrack_lock); 509 510 return h != NULL; 511} 512 513/* There's a small race here where we may free a just-assured 514 connection. Too bad: we're in trouble anyway. */ 515static int early_drop(struct list_head *chain) 516{ 517 /* Traverse backwards: gives us oldest, which is roughly LRU */ 518 struct nf_conntrack_tuple_hash *h; 519 struct nf_conn *ct = NULL, *tmp; 520 int dropped = 0; 521 522 read_lock_bh(&nf_conntrack_lock); 523 list_for_each_entry_reverse(h, chain, list) { 524 tmp = nf_ct_tuplehash_to_ctrack(h); 525 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) { 526 ct = tmp; 527 atomic_inc(&ct->ct_general.use); 528 break; 529 } 530 } 531 read_unlock_bh(&nf_conntrack_lock); 532 533 if (!ct) 534 return dropped; 535 536 if (del_timer(&ct->timeout)) { 537 death_by_timeout((unsigned long)ct); 538 dropped = 1; 539 NF_CT_STAT_INC(early_drop); 540 } 541 nf_ct_put(ct); 542 return dropped; 543} 544 545static struct nf_conn * 546__nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 547 const struct nf_conntrack_tuple *repl, 548 const struct nf_conntrack_l3proto *l3proto, 549 u_int32_t features) 550{ 551 struct nf_conn *conntrack = NULL; 552 struct nf_conntrack_helper *helper; 553 554 if (unlikely(!nf_conntrack_hash_rnd_initted)) { 555 get_random_bytes(&nf_conntrack_hash_rnd, 4); 556 nf_conntrack_hash_rnd_initted = 1; 557 } 558 559 /* We don't want any race condition at early drop stage */ 560 atomic_inc(&nf_conntrack_count); 561 562 if (nf_conntrack_max 563 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { 564 unsigned int hash = hash_conntrack(orig); 565 /* Try dropping from this hash chain. */ 566 if (!early_drop(&nf_conntrack_hash[hash])) { 567 atomic_dec(&nf_conntrack_count); 568 if (net_ratelimit()) 569 printk(KERN_WARNING 570 "nf_conntrack: table full, dropping" 571 " packet.\n"); 572 return ERR_PTR(-ENOMEM); 573 } 574 } 575 576 /* find features needed by this conntrack. */ 577 features |= l3proto->get_features(orig); 578 579 /* FIXME: protect helper list per RCU */ 580 read_lock_bh(&nf_conntrack_lock); 581 helper = __nf_ct_helper_find(repl); 582 /* NAT might want to assign a helper later */ 583 if (helper || features & NF_CT_F_NAT) 584 features |= NF_CT_F_HELP; 585 read_unlock_bh(&nf_conntrack_lock); 586 587 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features); 588 589 read_lock_bh(&nf_ct_cache_lock); 590 591 if (unlikely(!nf_ct_cache[features].use)) { 592 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n", 593 features); 594 goto out; 595 } 596 597 conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC); 598 if (conntrack == NULL) { 599 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n"); 600 goto out; 601 } 602 603 memset(conntrack, 0, nf_ct_cache[features].size); 604 conntrack->features = features; 605 atomic_set(&conntrack->ct_general.use, 1); 606 conntrack->ct_general.destroy = destroy_conntrack; 607 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 608 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 609 /* Don't set timer yet: wait for confirmation */ 610 init_timer(&conntrack->timeout); 611 conntrack->timeout.data = (unsigned long)conntrack; 612 conntrack->timeout.function = death_by_timeout; 613 read_unlock_bh(&nf_ct_cache_lock); 614 615 return conntrack; 616out: 617 read_unlock_bh(&nf_ct_cache_lock); 618 atomic_dec(&nf_conntrack_count); 619 return conntrack; 620} 621 622struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 623 const struct nf_conntrack_tuple *repl) 624{ 625 struct nf_conntrack_l3proto *l3proto; 626 627 l3proto = __nf_ct_l3proto_find(orig->src.l3num); 628 return __nf_conntrack_alloc(orig, repl, l3proto, 0); 629} 630 631void nf_conntrack_free(struct nf_conn *conntrack) 632{ 633 u_int32_t features = conntrack->features; 634 NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM); 635 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features, 636 conntrack); 637 kmem_cache_free(nf_ct_cache[features].cachep, conntrack); 638 atomic_dec(&nf_conntrack_count); 639} 640 641/* Allocate a new conntrack: we return -ENOMEM if classification 642 failed due to stress. Otherwise it really is unclassifiable. */ 643static struct nf_conntrack_tuple_hash * 644init_conntrack(const struct nf_conntrack_tuple *tuple, 645 struct nf_conntrack_l3proto *l3proto, 646 struct nf_conntrack_l4proto *l4proto, 647 struct sk_buff *skb, 648 unsigned int dataoff) 649{ 650 struct nf_conn *conntrack; 651 struct nf_conntrack_tuple repl_tuple; 652 struct nf_conntrack_expect *exp; 653 u_int32_t features = 0; 654 655 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 656 DEBUGP("Can't invert tuple.\n"); 657 return NULL; 658 } 659 660 read_lock_bh(&nf_conntrack_lock); 661 exp = __nf_conntrack_expect_find(tuple); 662 if (exp && exp->helper) 663 features = NF_CT_F_HELP; 664 read_unlock_bh(&nf_conntrack_lock); 665 666 conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto, features); 667 if (conntrack == NULL || IS_ERR(conntrack)) { 668 DEBUGP("Can't allocate conntrack.\n"); 669 return (struct nf_conntrack_tuple_hash *)conntrack; 670 } 671 672 if (!l4proto->new(conntrack, skb, dataoff)) { 673 nf_conntrack_free(conntrack); 674 DEBUGP("init conntrack: can't track with proto module\n"); 675 return NULL; 676 } 677 678 write_lock_bh(&nf_conntrack_lock); 679 exp = find_expectation(tuple); 680 681 if (exp) { 682 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", 683 conntrack, exp); 684 /* Welcome, Mr. Bond. We've been expecting you... */ 685 __set_bit(IPS_EXPECTED_BIT, &conntrack->status); 686 conntrack->master = exp->master; 687 if (exp->helper) 688 nfct_help(conntrack)->helper = exp->helper; 689#ifdef CONFIG_NF_CONNTRACK_MARK 690 conntrack->mark = exp->master->mark; 691#endif 692#ifdef CONFIG_NF_CONNTRACK_SECMARK 693 conntrack->secmark = exp->master->secmark; 694#endif 695 nf_conntrack_get(&conntrack->master->ct_general); 696 NF_CT_STAT_INC(expect_new); 697 } else { 698 struct nf_conn_help *help = nfct_help(conntrack); 699 700 if (help) 701 help->helper = __nf_ct_helper_find(&repl_tuple); 702 NF_CT_STAT_INC(new); 703 } 704 705 /* Overload tuple linked list to put us in unconfirmed list. */ 706 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); 707 708 write_unlock_bh(&nf_conntrack_lock); 709 710 if (exp) { 711 if (exp->expectfn) 712 exp->expectfn(conntrack, exp); 713 nf_conntrack_expect_put(exp); 714 } 715 716 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; 717} 718 719/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 720static inline struct nf_conn * 721resolve_normal_ct(struct sk_buff *skb, 722 unsigned int dataoff, 723 u_int16_t l3num, 724 u_int8_t protonum, 725 struct nf_conntrack_l3proto *l3proto, 726 struct nf_conntrack_l4proto *l4proto, 727 int *set_reply, 728 enum ip_conntrack_info *ctinfo) 729{ 730 struct nf_conntrack_tuple tuple; 731 struct nf_conntrack_tuple_hash *h; 732 struct nf_conn *ct; 733 734 if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data), 735 dataoff, l3num, protonum, &tuple, l3proto, 736 l4proto)) { 737 DEBUGP("resolve_normal_ct: Can't get tuple\n"); 738 return NULL; 739 } 740 741 /* look for tuple match */ 742 h = nf_conntrack_find_get(&tuple, NULL); 743 if (!h) { 744 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); 745 if (!h) 746 return NULL; 747 if (IS_ERR(h)) 748 return (void *)h; 749 } 750 ct = nf_ct_tuplehash_to_ctrack(h); 751 752 /* It exists; we have (non-exclusive) reference. */ 753 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 754 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 755 /* Please set reply bit if this packet OK */ 756 *set_reply = 1; 757 } else { 758 /* Once we've had two way comms, always ESTABLISHED. */ 759 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 760 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct); 761 *ctinfo = IP_CT_ESTABLISHED; 762 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 763 DEBUGP("nf_conntrack_in: related packet for %p\n", ct); 764 *ctinfo = IP_CT_RELATED; 765 } else { 766 DEBUGP("nf_conntrack_in: new packet for %p\n", ct); 767 *ctinfo = IP_CT_NEW; 768 } 769 *set_reply = 0; 770 } 771 skb->nfct = &ct->ct_general; 772 skb->nfctinfo = *ctinfo; 773 return ct; 774} 775 776unsigned int 777nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) 778{ 779 struct nf_conn *ct; 780 enum ip_conntrack_info ctinfo; 781 struct nf_conntrack_l3proto *l3proto; 782 struct nf_conntrack_l4proto *l4proto; 783 unsigned int dataoff; 784 u_int8_t protonum; 785 int set_reply = 0; 786 int ret; 787 788 /* Previously seen (loopback or untracked)? Ignore. */ 789 if ((*pskb)->nfct) { 790 NF_CT_STAT_INC(ignore); 791 return NF_ACCEPT; 792 } 793 794 l3proto = __nf_ct_l3proto_find((u_int16_t)pf); 795 if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) { 796 DEBUGP("not prepared to track yet or error occured\n"); 797 return -ret; 798 } 799 800 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); 801 802 /* It may be an special packet, error, unclean... 803 * inverse of the return code tells to the netfilter 804 * core what to do with the packet. */ 805 if (l4proto->error != NULL && 806 (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { 807 NF_CT_STAT_INC(error); 808 NF_CT_STAT_INC(invalid); 809 return -ret; 810 } 811 812 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto, 813 &set_reply, &ctinfo); 814 if (!ct) { 815 /* Not valid part of a connection */ 816 NF_CT_STAT_INC(invalid); 817 return NF_ACCEPT; 818 } 819 820 if (IS_ERR(ct)) { 821 /* Too stressed to deal. */ 822 NF_CT_STAT_INC(drop); 823 return NF_DROP; 824 } 825 826 NF_CT_ASSERT((*pskb)->nfct); 827 828 ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); 829 if (ret < 0) { 830 /* Invalid: inverse of the return code tells 831 * the netfilter core what to do */ 832 DEBUGP("nf_conntrack_in: Can't track with proto module\n"); 833 nf_conntrack_put((*pskb)->nfct); 834 (*pskb)->nfct = NULL; 835 NF_CT_STAT_INC(invalid); 836 return -ret; 837 } 838 839 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 840 nf_conntrack_event_cache(IPCT_STATUS, *pskb); 841 842 return ret; 843} 844 845int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 846 const struct nf_conntrack_tuple *orig) 847{ 848 return nf_ct_invert_tuple(inverse, orig, 849 __nf_ct_l3proto_find(orig->src.l3num), 850 __nf_ct_l4proto_find(orig->src.l3num, 851 orig->dst.protonum)); 852} 853 854/* Alter reply tuple (maybe alter helper). This is for NAT, and is 855 implicitly racy: see __nf_conntrack_confirm */ 856void nf_conntrack_alter_reply(struct nf_conn *ct, 857 const struct nf_conntrack_tuple *newreply) 858{ 859 struct nf_conn_help *help = nfct_help(ct); 860 861 write_lock_bh(&nf_conntrack_lock); 862 /* Should be unconfirmed, so not in hash table yet */ 863 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 864 865 DEBUGP("Altering reply tuple of %p to ", ct); 866 NF_CT_DUMP_TUPLE(newreply); 867 868 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 869 if (!ct->master && help && help->expecting == 0) 870 help->helper = __nf_ct_helper_find(newreply); 871 write_unlock_bh(&nf_conntrack_lock); 872} 873 874/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 875void __nf_ct_refresh_acct(struct nf_conn *ct, 876 enum ip_conntrack_info ctinfo, 877 const struct sk_buff *skb, 878 unsigned long extra_jiffies, 879 int do_acct) 880{ 881 int event = 0; 882 883 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 884 NF_CT_ASSERT(skb); 885 886 write_lock_bh(&nf_conntrack_lock); 887 888 /* Only update if this is not a fixed timeout */ 889 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { 890 write_unlock_bh(&nf_conntrack_lock); 891 return; 892 } 893 894 /* If not in hash table, timer will not be active yet */ 895 if (!nf_ct_is_confirmed(ct)) { 896 ct->timeout.expires = extra_jiffies; 897 event = IPCT_REFRESH; 898 } else { 899 unsigned long newtime = jiffies + extra_jiffies; 900 901 /* Only update the timeout if the new timeout is at least 902 HZ jiffies from the old timeout. Need del_timer for race 903 avoidance (may already be dying). */ 904 if (newtime - ct->timeout.expires >= HZ 905 && del_timer(&ct->timeout)) { 906 ct->timeout.expires = newtime; 907 add_timer(&ct->timeout); 908 event = IPCT_REFRESH; 909 } 910 } 911 912#ifdef CONFIG_NF_CT_ACCT 913 if (do_acct) { 914 ct->counters[CTINFO2DIR(ctinfo)].packets++; 915 ct->counters[CTINFO2DIR(ctinfo)].bytes += 916 skb->len - (unsigned int)(skb->nh.raw - skb->data); 917 918 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) 919 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) 920 event |= IPCT_COUNTER_FILLING; 921 } 922#endif 923 924 write_unlock_bh(&nf_conntrack_lock); 925 926 /* must be unlocked when calling event cache */ 927 if (event) 928 nf_conntrack_event_cache(event, skb); 929} 930 931#if defined(CONFIG_NF_CT_NETLINK) || \ 932 defined(CONFIG_NF_CT_NETLINK_MODULE) 933 934#include <linux/netfilter/nfnetlink.h> 935#include <linux/netfilter/nfnetlink_conntrack.h> 936#include <linux/mutex.h> 937 938 939/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 940 * in ip_conntrack_core, since we don't want the protocols to autoload 941 * or depend on ctnetlink */ 942int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb, 943 const struct nf_conntrack_tuple *tuple) 944{ 945 NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), 946 &tuple->src.u.tcp.port); 947 NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), 948 &tuple->dst.u.tcp.port); 949 return 0; 950 951nfattr_failure: 952 return -1; 953} 954 955static const size_t cta_min_proto[CTA_PROTO_MAX] = { 956 [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t), 957 [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t) 958}; 959 960int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[], 961 struct nf_conntrack_tuple *t) 962{ 963 if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) 964 return -EINVAL; 965 966 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) 967 return -EINVAL; 968 969 t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); 970 t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); 971 972 return 0; 973} 974#endif 975 976/* Used by ipt_REJECT and ip6t_REJECT. */ 977void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 978{ 979 struct nf_conn *ct; 980 enum ip_conntrack_info ctinfo; 981 982 /* This ICMP is in reverse direction to the packet which caused it */ 983 ct = nf_ct_get(skb, &ctinfo); 984 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 985 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 986 else 987 ctinfo = IP_CT_RELATED; 988 989 /* Attach to new skbuff, and increment count */ 990 nskb->nfct = &ct->ct_general; 991 nskb->nfctinfo = ctinfo; 992 nf_conntrack_get(nskb->nfct); 993} 994 995static inline int 996do_iter(const struct nf_conntrack_tuple_hash *i, 997 int (*iter)(struct nf_conn *i, void *data), 998 void *data) 999{ 1000 return iter(nf_ct_tuplehash_to_ctrack(i), data); 1001} 1002 1003/* Bring out ya dead! */ 1004static struct nf_conn * 1005get_next_corpse(int (*iter)(struct nf_conn *i, void *data), 1006 void *data, unsigned int *bucket) 1007{ 1008 struct nf_conntrack_tuple_hash *h; 1009 struct nf_conn *ct; 1010 1011 write_lock_bh(&nf_conntrack_lock); 1012 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 1013 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) { 1014 ct = nf_ct_tuplehash_to_ctrack(h); 1015 if (iter(ct, data)) 1016 goto found; 1017 } 1018 } 1019 list_for_each_entry(h, &unconfirmed, list) { 1020 ct = nf_ct_tuplehash_to_ctrack(h); 1021 if (iter(ct, data)) 1022 goto found; 1023 } 1024 write_unlock_bh(&nf_conntrack_lock); 1025 return NULL; 1026found: 1027 atomic_inc(&ct->ct_general.use); 1028 write_unlock_bh(&nf_conntrack_lock); 1029 return ct; 1030} 1031 1032void 1033nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) 1034{ 1035 struct nf_conn *ct; 1036 unsigned int bucket = 0; 1037 1038 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { 1039 /* Time to push up daises... */ 1040 if (del_timer(&ct->timeout)) 1041 death_by_timeout((unsigned long)ct); 1042 /* ... else the timer will get him soon. */ 1043 1044 nf_ct_put(ct); 1045 } 1046} 1047 1048static int kill_all(struct nf_conn *i, void *data) 1049{ 1050 return 1; 1051} 1052 1053static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size) 1054{ 1055 if (vmalloced) 1056 vfree(hash); 1057 else 1058 free_pages((unsigned long)hash, 1059 get_order(sizeof(struct list_head) * size)); 1060} 1061 1062void nf_conntrack_flush() 1063{ 1064 nf_ct_iterate_cleanup(kill_all, NULL); 1065} 1066 1067/* Mishearing the voices in his head, our hero wonders how he's 1068 supposed to kill the mall. */ 1069void nf_conntrack_cleanup(void) 1070{ 1071 int i; 1072 1073 ip_ct_attach = NULL; 1074 1075 /* This makes sure all current packets have passed through 1076 netfilter framework. Roll on, two-stage module 1077 delete... */ 1078 synchronize_net(); 1079 1080 nf_ct_event_cache_flush(); 1081 i_see_dead_people: 1082 nf_conntrack_flush(); 1083 if (atomic_read(&nf_conntrack_count) != 0) { 1084 schedule(); 1085 goto i_see_dead_people; 1086 } 1087 /* wait until all references to nf_conntrack_untracked are dropped */ 1088 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) 1089 schedule(); 1090 1091 for (i = 0; i < NF_CT_F_NUM; i++) { 1092 if (nf_ct_cache[i].use == 0) 1093 continue; 1094 1095 NF_CT_ASSERT(nf_ct_cache[i].use == 1); 1096 nf_ct_cache[i].use = 1; 1097 nf_conntrack_unregister_cache(i); 1098 } 1099 kmem_cache_destroy(nf_conntrack_expect_cachep); 1100 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1101 nf_conntrack_htable_size); 1102 1103 nf_conntrack_l4proto_unregister(&nf_conntrack_l4proto_generic); 1104 1105 /* free l3proto protocol tables */ 1106 for (i = 0; i < PF_MAX; i++) 1107 if (nf_ct_protos[i]) { 1108 kfree(nf_ct_protos[i]); 1109 nf_ct_protos[i] = NULL; 1110 } 1111} 1112 1113static struct list_head *alloc_hashtable(int size, int *vmalloced) 1114{ 1115 struct list_head *hash; 1116 unsigned int i; 1117 1118 *vmalloced = 0; 1119 hash = (void*)__get_free_pages(GFP_KERNEL, 1120 get_order(sizeof(struct list_head) 1121 * size)); 1122 if (!hash) { 1123 *vmalloced = 1; 1124 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1125 hash = vmalloc(sizeof(struct list_head) * size); 1126 } 1127 1128 if (hash) 1129 for (i = 0; i < size; i++) 1130 INIT_LIST_HEAD(&hash[i]); 1131 1132 return hash; 1133} 1134 1135int set_hashsize(const char *val, struct kernel_param *kp) 1136{ 1137 int i, bucket, hashsize, vmalloced; 1138 int old_vmalloced, old_size; 1139 int rnd; 1140 struct list_head *hash, *old_hash; 1141 struct nf_conntrack_tuple_hash *h; 1142 1143 /* On boot, we can set this without any fancy locking. */ 1144 if (!nf_conntrack_htable_size) 1145 return param_set_uint(val, kp); 1146 1147 hashsize = simple_strtol(val, NULL, 0); 1148 if (!hashsize) 1149 return -EINVAL; 1150 1151 hash = alloc_hashtable(hashsize, &vmalloced); 1152 if (!hash) 1153 return -ENOMEM; 1154 1155 /* We have to rehahs for the new table anyway, so we also can 1156 * use a newrandom seed */ 1157 get_random_bytes(&rnd, 4); 1158 1159 write_lock_bh(&nf_conntrack_lock); 1160 for (i = 0; i < nf_conntrack_htable_size; i++) { 1161 while (!list_empty(&nf_conntrack_hash[i])) { 1162 h = list_entry(nf_conntrack_hash[i].next, 1163 struct nf_conntrack_tuple_hash, list); 1164 list_del(&h->list); 1165 bucket = __hash_conntrack(&h->tuple, hashsize, rnd); 1166 list_add_tail(&h->list, &hash[bucket]); 1167 } 1168 } 1169 old_size = nf_conntrack_htable_size; 1170 old_vmalloced = nf_conntrack_vmalloc; 1171 old_hash = nf_conntrack_hash; 1172 1173 nf_conntrack_htable_size = hashsize; 1174 nf_conntrack_vmalloc = vmalloced; 1175 nf_conntrack_hash = hash; 1176 nf_conntrack_hash_rnd = rnd; 1177 write_unlock_bh(&nf_conntrack_lock); 1178 1179 free_conntrack_hash(old_hash, old_vmalloced, old_size); 1180 return 0; 1181} 1182 1183module_param_call(hashsize, set_hashsize, param_get_uint, 1184 &nf_conntrack_htable_size, 0600); 1185 1186int __init nf_conntrack_init(void) 1187{ 1188 unsigned int i; 1189 int ret; 1190 1191 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1192 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ 1193 if (!nf_conntrack_htable_size) { 1194 nf_conntrack_htable_size 1195 = (((num_physpages << PAGE_SHIFT) / 16384) 1196 / sizeof(struct list_head)); 1197 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1198 nf_conntrack_htable_size = 8192; 1199 if (nf_conntrack_htable_size < 16) 1200 nf_conntrack_htable_size = 16; 1201 } 1202 nf_conntrack_max = 8 * nf_conntrack_htable_size; 1203 1204 printk("nf_conntrack version %s (%u buckets, %d max)\n", 1205 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1206 nf_conntrack_max); 1207 1208 nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size, 1209 &nf_conntrack_vmalloc); 1210 if (!nf_conntrack_hash) { 1211 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1212 goto err_out; 1213 } 1214 1215 ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic", 1216 sizeof(struct nf_conn)); 1217 if (ret < 0) { 1218 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1219 goto err_free_hash; 1220 } 1221 1222 nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect", 1223 sizeof(struct nf_conntrack_expect), 1224 0, 0, NULL, NULL); 1225 if (!nf_conntrack_expect_cachep) { 1226 printk(KERN_ERR "Unable to create nf_expect slab cache\n"); 1227 goto err_free_conntrack_slab; 1228 } 1229 1230 ret = nf_conntrack_l4proto_register(&nf_conntrack_l4proto_generic); 1231 if (ret < 0) 1232 goto out_free_expect_slab; 1233 1234 /* Don't NEED lock here, but good form anyway. */ 1235 write_lock_bh(&nf_conntrack_lock); 1236 for (i = 0; i < AF_MAX; i++) 1237 nf_ct_l3protos[i] = &nf_conntrack_l3proto_generic; 1238 write_unlock_bh(&nf_conntrack_lock); 1239 1240 /* For use by REJECT target */ 1241 ip_ct_attach = __nf_conntrack_attach; 1242 1243 /* Set up fake conntrack: 1244 - to never be deleted, not in any hashes */ 1245 atomic_set(&nf_conntrack_untracked.ct_general.use, 1); 1246 /* - and look it like as a confirmed connection */ 1247 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); 1248 1249 return ret; 1250 1251out_free_expect_slab: 1252 kmem_cache_destroy(nf_conntrack_expect_cachep); 1253err_free_conntrack_slab: 1254 nf_conntrack_unregister_cache(NF_CT_F_BASIC); 1255err_free_hash: 1256 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1257 nf_conntrack_htable_size); 1258err_out: 1259 return -ENOMEM; 1260} 1261