nf_conntrack_core.c revision e2b7606cdb602a4f69c02cfc8bebe9c63b595e24
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/* This rwlock protects the main hash table, protocol/helper/expected 51 registrations, conntrack timers*/ 52#define ASSERT_READ_LOCK(x) 53#define ASSERT_WRITE_LOCK(x) 54 55#include <net/netfilter/nf_conntrack.h> 56#include <net/netfilter/nf_conntrack_l3proto.h> 57#include <net/netfilter/nf_conntrack_protocol.h> 58#include <net/netfilter/nf_conntrack_expect.h> 59#include <net/netfilter/nf_conntrack_helper.h> 60#include <net/netfilter/nf_conntrack_core.h> 61 62#define NF_CONNTRACK_VERSION "0.5.0" 63 64#if 0 65#define DEBUGP printk 66#else 67#define DEBUGP(format, args...) 68#endif 69 70DEFINE_RWLOCK(nf_conntrack_lock); 71 72/* nf_conntrack_standalone needs this */ 73atomic_t nf_conntrack_count = ATOMIC_INIT(0); 74 75void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL; 76unsigned int nf_conntrack_htable_size __read_mostly; 77int nf_conntrack_max __read_mostly; 78struct list_head *nf_conntrack_hash __read_mostly; 79struct nf_conn nf_conntrack_untracked __read_mostly; 80unsigned int nf_ct_log_invalid __read_mostly; 81LIST_HEAD(unconfirmed); 82static int nf_conntrack_vmalloc __read_mostly; 83 84static unsigned int nf_conntrack_next_id; 85 86DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); 87EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); 88 89/* 90 * This scheme offers various size of "struct nf_conn" dependent on 91 * features(helper, nat, ...) 92 */ 93 94#define NF_CT_FEATURES_NAMELEN 256 95static struct { 96 /* name of slab cache. printed in /proc/slabinfo */ 97 char *name; 98 99 /* size of slab cache */ 100 size_t size; 101 102 /* slab cache pointer */ 103 kmem_cache_t *cachep; 104 105 /* allocated slab cache + modules which uses this slab cache */ 106 int use; 107 108} nf_ct_cache[NF_CT_F_NUM]; 109 110/* protect members of nf_ct_cache except of "use" */ 111DEFINE_RWLOCK(nf_ct_cache_lock); 112 113/* This avoids calling kmem_cache_create() with same name simultaneously */ 114static DEFINE_MUTEX(nf_ct_cache_mutex); 115 116static int nf_conntrack_hash_rnd_initted; 117static unsigned int nf_conntrack_hash_rnd; 118 119static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 120 unsigned int size, unsigned int rnd) 121{ 122 unsigned int a, b; 123 a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all), 124 ((tuple->src.l3num) << 16) | tuple->dst.protonum); 125 b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all), 126 (tuple->src.u.all << 16) | tuple->dst.u.all); 127 128 return jhash_2words(a, b, rnd) % size; 129} 130 131static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) 132{ 133 return __hash_conntrack(tuple, nf_conntrack_htable_size, 134 nf_conntrack_hash_rnd); 135} 136 137int nf_conntrack_register_cache(u_int32_t features, const char *name, 138 size_t size) 139{ 140 int ret = 0; 141 char *cache_name; 142 kmem_cache_t *cachep; 143 144 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n", 145 features, name, size); 146 147 if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) { 148 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n", 149 features); 150 return -EINVAL; 151 } 152 153 mutex_lock(&nf_ct_cache_mutex); 154 155 write_lock_bh(&nf_ct_cache_lock); 156 /* e.g: multiple helpers are loaded */ 157 if (nf_ct_cache[features].use > 0) { 158 DEBUGP("nf_conntrack_register_cache: already resisterd.\n"); 159 if ((!strncmp(nf_ct_cache[features].name, name, 160 NF_CT_FEATURES_NAMELEN)) 161 && nf_ct_cache[features].size == size) { 162 DEBUGP("nf_conntrack_register_cache: reusing.\n"); 163 nf_ct_cache[features].use++; 164 ret = 0; 165 } else 166 ret = -EBUSY; 167 168 write_unlock_bh(&nf_ct_cache_lock); 169 mutex_unlock(&nf_ct_cache_mutex); 170 return ret; 171 } 172 write_unlock_bh(&nf_ct_cache_lock); 173 174 /* 175 * The memory space for name of slab cache must be alive until 176 * cache is destroyed. 177 */ 178 cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC); 179 if (cache_name == NULL) { 180 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n"); 181 ret = -ENOMEM; 182 goto out_up_mutex; 183 } 184 185 if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN) 186 >= NF_CT_FEATURES_NAMELEN) { 187 printk("nf_conntrack_register_cache: name too long\n"); 188 ret = -EINVAL; 189 goto out_free_name; 190 } 191 192 cachep = kmem_cache_create(cache_name, size, 0, 0, 193 NULL, NULL); 194 if (!cachep) { 195 printk("nf_conntrack_register_cache: Can't create slab cache " 196 "for the features = 0x%x\n", features); 197 ret = -ENOMEM; 198 goto out_free_name; 199 } 200 201 write_lock_bh(&nf_ct_cache_lock); 202 nf_ct_cache[features].use = 1; 203 nf_ct_cache[features].size = size; 204 nf_ct_cache[features].cachep = cachep; 205 nf_ct_cache[features].name = cache_name; 206 write_unlock_bh(&nf_ct_cache_lock); 207 208 goto out_up_mutex; 209 210out_free_name: 211 kfree(cache_name); 212out_up_mutex: 213 mutex_unlock(&nf_ct_cache_mutex); 214 return ret; 215} 216 217/* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */ 218void nf_conntrack_unregister_cache(u_int32_t features) 219{ 220 kmem_cache_t *cachep; 221 char *name; 222 223 /* 224 * This assures that kmem_cache_create() isn't called before destroying 225 * slab cache. 226 */ 227 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features); 228 mutex_lock(&nf_ct_cache_mutex); 229 230 write_lock_bh(&nf_ct_cache_lock); 231 if (--nf_ct_cache[features].use > 0) { 232 write_unlock_bh(&nf_ct_cache_lock); 233 mutex_unlock(&nf_ct_cache_mutex); 234 return; 235 } 236 cachep = nf_ct_cache[features].cachep; 237 name = nf_ct_cache[features].name; 238 nf_ct_cache[features].cachep = NULL; 239 nf_ct_cache[features].name = NULL; 240 nf_ct_cache[features].size = 0; 241 write_unlock_bh(&nf_ct_cache_lock); 242 243 synchronize_net(); 244 245 kmem_cache_destroy(cachep); 246 kfree(name); 247 248 mutex_unlock(&nf_ct_cache_mutex); 249} 250 251int 252nf_ct_get_tuple(const struct sk_buff *skb, 253 unsigned int nhoff, 254 unsigned int dataoff, 255 u_int16_t l3num, 256 u_int8_t protonum, 257 struct nf_conntrack_tuple *tuple, 258 const struct nf_conntrack_l3proto *l3proto, 259 const struct nf_conntrack_protocol *protocol) 260{ 261 NF_CT_TUPLE_U_BLANK(tuple); 262 263 tuple->src.l3num = l3num; 264 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 265 return 0; 266 267 tuple->dst.protonum = protonum; 268 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 269 270 return protocol->pkt_to_tuple(skb, dataoff, tuple); 271} 272 273int 274nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 275 const struct nf_conntrack_tuple *orig, 276 const struct nf_conntrack_l3proto *l3proto, 277 const struct nf_conntrack_protocol *protocol) 278{ 279 NF_CT_TUPLE_U_BLANK(inverse); 280 281 inverse->src.l3num = orig->src.l3num; 282 if (l3proto->invert_tuple(inverse, orig) == 0) 283 return 0; 284 285 inverse->dst.dir = !orig->dst.dir; 286 287 inverse->dst.protonum = orig->dst.protonum; 288 return protocol->invert_tuple(inverse, orig); 289} 290 291static void 292clean_from_lists(struct nf_conn *ct) 293{ 294 DEBUGP("clean_from_lists(%p)\n", ct); 295 ASSERT_WRITE_LOCK(&nf_conntrack_lock); 296 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 297 list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list); 298 299 /* Destroy all pending expectations */ 300 nf_ct_remove_expectations(ct); 301} 302 303static void 304destroy_conntrack(struct nf_conntrack *nfct) 305{ 306 struct nf_conn *ct = (struct nf_conn *)nfct; 307 struct nf_conntrack_l3proto *l3proto; 308 struct nf_conntrack_protocol *proto; 309 310 DEBUGP("destroy_conntrack(%p)\n", ct); 311 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 312 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 313 314 nf_conntrack_event(IPCT_DESTROY, ct); 315 set_bit(IPS_DYING_BIT, &ct->status); 316 317 /* To make sure we don't get any weird locking issues here: 318 * destroy_conntrack() MUST NOT be called with a write lock 319 * to nf_conntrack_lock!!! -HW */ 320 l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num); 321 if (l3proto && l3proto->destroy) 322 l3proto->destroy(ct); 323 324 proto = __nf_ct_proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); 325 if (proto && proto->destroy) 326 proto->destroy(ct); 327 328 if (nf_conntrack_destroyed) 329 nf_conntrack_destroyed(ct); 330 331 write_lock_bh(&nf_conntrack_lock); 332 /* Expectations will have been removed in clean_from_lists, 333 * except TFTP can create an expectation on the first packet, 334 * before connection is in the list, so we need to clean here, 335 * too. */ 336 nf_ct_remove_expectations(ct); 337 338 /* We overload first tuple to link into unconfirmed list. */ 339 if (!nf_ct_is_confirmed(ct)) { 340 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); 341 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 342 } 343 344 NF_CT_STAT_INC(delete); 345 write_unlock_bh(&nf_conntrack_lock); 346 347 if (ct->master) 348 nf_ct_put(ct->master); 349 350 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); 351 nf_conntrack_free(ct); 352} 353 354static void death_by_timeout(unsigned long ul_conntrack) 355{ 356 struct nf_conn *ct = (void *)ul_conntrack; 357 358 write_lock_bh(&nf_conntrack_lock); 359 /* Inside lock so preempt is disabled on module removal path. 360 * Otherwise we can get spurious warnings. */ 361 NF_CT_STAT_INC(delete_list); 362 clean_from_lists(ct); 363 write_unlock_bh(&nf_conntrack_lock); 364 nf_ct_put(ct); 365} 366 367struct nf_conntrack_tuple_hash * 368__nf_conntrack_find(const struct nf_conntrack_tuple *tuple, 369 const struct nf_conn *ignored_conntrack) 370{ 371 struct nf_conntrack_tuple_hash *h; 372 unsigned int hash = hash_conntrack(tuple); 373 374 ASSERT_READ_LOCK(&nf_conntrack_lock); 375 list_for_each_entry(h, &nf_conntrack_hash[hash], list) { 376 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && 377 nf_ct_tuple_equal(tuple, &h->tuple)) { 378 NF_CT_STAT_INC(found); 379 return h; 380 } 381 NF_CT_STAT_INC(searched); 382 } 383 384 return NULL; 385} 386 387/* Find a connection corresponding to a tuple. */ 388struct nf_conntrack_tuple_hash * 389nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple, 390 const struct nf_conn *ignored_conntrack) 391{ 392 struct nf_conntrack_tuple_hash *h; 393 394 read_lock_bh(&nf_conntrack_lock); 395 h = __nf_conntrack_find(tuple, ignored_conntrack); 396 if (h) 397 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); 398 read_unlock_bh(&nf_conntrack_lock); 399 400 return h; 401} 402 403static void __nf_conntrack_hash_insert(struct nf_conn *ct, 404 unsigned int hash, 405 unsigned int repl_hash) 406{ 407 ct->id = ++nf_conntrack_next_id; 408 list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list, 409 &nf_conntrack_hash[hash]); 410 list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list, 411 &nf_conntrack_hash[repl_hash]); 412} 413 414void nf_conntrack_hash_insert(struct nf_conn *ct) 415{ 416 unsigned int hash, repl_hash; 417 418 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 419 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 420 421 write_lock_bh(&nf_conntrack_lock); 422 __nf_conntrack_hash_insert(ct, hash, repl_hash); 423 write_unlock_bh(&nf_conntrack_lock); 424} 425 426/* Confirm a connection given skb; places it in hash table */ 427int 428__nf_conntrack_confirm(struct sk_buff **pskb) 429{ 430 unsigned int hash, repl_hash; 431 struct nf_conntrack_tuple_hash *h; 432 struct nf_conn *ct; 433 struct nf_conn_help *help; 434 enum ip_conntrack_info ctinfo; 435 436 ct = nf_ct_get(*pskb, &ctinfo); 437 438 /* ipt_REJECT uses nf_conntrack_attach to attach related 439 ICMP/TCP RST packets in other direction. Actual packet 440 which created connection will be IP_CT_NEW or for an 441 expected connection, IP_CT_RELATED. */ 442 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 443 return NF_ACCEPT; 444 445 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 446 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 447 448 /* We're not in hash table, and we refuse to set up related 449 connections for unconfirmed conns. But packet copies and 450 REJECT will give spurious warnings here. */ 451 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 452 453 /* No external references means noone else could have 454 confirmed us. */ 455 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 456 DEBUGP("Confirming conntrack %p\n", ct); 457 458 write_lock_bh(&nf_conntrack_lock); 459 460 /* See if there's one in the list already, including reverse: 461 NAT could have grabbed it without realizing, since we're 462 not in the hash. If there is, we lost race. */ 463 list_for_each_entry(h, &nf_conntrack_hash[hash], list) 464 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 465 &h->tuple)) 466 goto out; 467 list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list) 468 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 469 &h->tuple)) 470 goto out; 471 472 /* Remove from unconfirmed list */ 473 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 474 475 __nf_conntrack_hash_insert(ct, hash, repl_hash); 476 /* Timer relative to confirmation time, not original 477 setting time, otherwise we'd get timer wrap in 478 weird delay cases. */ 479 ct->timeout.expires += jiffies; 480 add_timer(&ct->timeout); 481 atomic_inc(&ct->ct_general.use); 482 set_bit(IPS_CONFIRMED_BIT, &ct->status); 483 NF_CT_STAT_INC(insert); 484 write_unlock_bh(&nf_conntrack_lock); 485 help = nfct_help(ct); 486 if (help && help->helper) 487 nf_conntrack_event_cache(IPCT_HELPER, *pskb); 488#ifdef CONFIG_NF_NAT_NEEDED 489 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || 490 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) 491 nf_conntrack_event_cache(IPCT_NATINFO, *pskb); 492#endif 493 nf_conntrack_event_cache(master_ct(ct) ? 494 IPCT_RELATED : IPCT_NEW, *pskb); 495 return NF_ACCEPT; 496 497out: 498 NF_CT_STAT_INC(insert_failed); 499 write_unlock_bh(&nf_conntrack_lock); 500 return NF_DROP; 501} 502 503/* Returns true if a connection correspondings to the tuple (required 504 for NAT). */ 505int 506nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 507 const struct nf_conn *ignored_conntrack) 508{ 509 struct nf_conntrack_tuple_hash *h; 510 511 read_lock_bh(&nf_conntrack_lock); 512 h = __nf_conntrack_find(tuple, ignored_conntrack); 513 read_unlock_bh(&nf_conntrack_lock); 514 515 return h != NULL; 516} 517 518/* There's a small race here where we may free a just-assured 519 connection. Too bad: we're in trouble anyway. */ 520static int early_drop(struct list_head *chain) 521{ 522 /* Traverse backwards: gives us oldest, which is roughly LRU */ 523 struct nf_conntrack_tuple_hash *h; 524 struct nf_conn *ct = NULL, *tmp; 525 int dropped = 0; 526 527 read_lock_bh(&nf_conntrack_lock); 528 list_for_each_entry_reverse(h, chain, list) { 529 tmp = nf_ct_tuplehash_to_ctrack(h); 530 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) { 531 ct = tmp; 532 atomic_inc(&ct->ct_general.use); 533 break; 534 } 535 } 536 read_unlock_bh(&nf_conntrack_lock); 537 538 if (!ct) 539 return dropped; 540 541 if (del_timer(&ct->timeout)) { 542 death_by_timeout((unsigned long)ct); 543 dropped = 1; 544 NF_CT_STAT_INC(early_drop); 545 } 546 nf_ct_put(ct); 547 return dropped; 548} 549 550static struct nf_conn * 551__nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 552 const struct nf_conntrack_tuple *repl, 553 const struct nf_conntrack_l3proto *l3proto) 554{ 555 struct nf_conn *conntrack = NULL; 556 u_int32_t features = 0; 557 struct nf_conntrack_helper *helper; 558 559 if (unlikely(!nf_conntrack_hash_rnd_initted)) { 560 get_random_bytes(&nf_conntrack_hash_rnd, 4); 561 nf_conntrack_hash_rnd_initted = 1; 562 } 563 564 /* We don't want any race condition at early drop stage */ 565 atomic_inc(&nf_conntrack_count); 566 567 if (nf_conntrack_max 568 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { 569 unsigned int hash = hash_conntrack(orig); 570 /* Try dropping from this hash chain. */ 571 if (!early_drop(&nf_conntrack_hash[hash])) { 572 atomic_dec(&nf_conntrack_count); 573 if (net_ratelimit()) 574 printk(KERN_WARNING 575 "nf_conntrack: table full, dropping" 576 " packet.\n"); 577 return ERR_PTR(-ENOMEM); 578 } 579 } 580 581 /* find features needed by this conntrack. */ 582 features = l3proto->get_features(orig); 583 584 /* FIXME: protect helper list per RCU */ 585 read_lock_bh(&nf_conntrack_lock); 586 helper = __nf_ct_helper_find(repl); 587 if (helper) 588 features |= NF_CT_F_HELP; 589 read_unlock_bh(&nf_conntrack_lock); 590 591 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features); 592 593 read_lock_bh(&nf_ct_cache_lock); 594 595 if (unlikely(!nf_ct_cache[features].use)) { 596 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n", 597 features); 598 goto out; 599 } 600 601 conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC); 602 if (conntrack == NULL) { 603 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n"); 604 goto out; 605 } 606 607 memset(conntrack, 0, nf_ct_cache[features].size); 608 conntrack->features = features; 609 atomic_set(&conntrack->ct_general.use, 1); 610 conntrack->ct_general.destroy = destroy_conntrack; 611 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 612 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 613 /* Don't set timer yet: wait for confirmation */ 614 init_timer(&conntrack->timeout); 615 conntrack->timeout.data = (unsigned long)conntrack; 616 conntrack->timeout.function = death_by_timeout; 617 read_unlock_bh(&nf_ct_cache_lock); 618 619 return conntrack; 620out: 621 read_unlock_bh(&nf_ct_cache_lock); 622 atomic_dec(&nf_conntrack_count); 623 return conntrack; 624} 625 626struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 627 const struct nf_conntrack_tuple *repl) 628{ 629 struct nf_conntrack_l3proto *l3proto; 630 631 l3proto = __nf_ct_l3proto_find(orig->src.l3num); 632 return __nf_conntrack_alloc(orig, repl, l3proto); 633} 634 635void nf_conntrack_free(struct nf_conn *conntrack) 636{ 637 u_int32_t features = conntrack->features; 638 NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM); 639 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features, 640 conntrack); 641 kmem_cache_free(nf_ct_cache[features].cachep, conntrack); 642 atomic_dec(&nf_conntrack_count); 643} 644 645/* Allocate a new conntrack: we return -ENOMEM if classification 646 failed due to stress. Otherwise it really is unclassifiable. */ 647static struct nf_conntrack_tuple_hash * 648init_conntrack(const struct nf_conntrack_tuple *tuple, 649 struct nf_conntrack_l3proto *l3proto, 650 struct nf_conntrack_protocol *protocol, 651 struct sk_buff *skb, 652 unsigned int dataoff) 653{ 654 struct nf_conn *conntrack; 655 struct nf_conntrack_tuple repl_tuple; 656 struct nf_conntrack_expect *exp; 657 658 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, protocol)) { 659 DEBUGP("Can't invert tuple.\n"); 660 return NULL; 661 } 662 663 conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto); 664 if (conntrack == NULL || IS_ERR(conntrack)) { 665 DEBUGP("Can't allocate conntrack.\n"); 666 return (struct nf_conntrack_tuple_hash *)conntrack; 667 } 668 669 if (!protocol->new(conntrack, skb, dataoff)) { 670 nf_conntrack_free(conntrack); 671 DEBUGP("init conntrack: can't track with proto module\n"); 672 return NULL; 673 } 674 675 write_lock_bh(&nf_conntrack_lock); 676 exp = find_expectation(tuple); 677 678 if (exp) { 679 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", 680 conntrack, exp); 681 /* Welcome, Mr. Bond. We've been expecting you... */ 682 __set_bit(IPS_EXPECTED_BIT, &conntrack->status); 683 conntrack->master = exp->master; 684#ifdef CONFIG_NF_CONNTRACK_MARK 685 conntrack->mark = exp->master->mark; 686#endif 687#ifdef CONFIG_NF_CONNTRACK_SECMARK 688 conntrack->secmark = exp->master->secmark; 689#endif 690 nf_conntrack_get(&conntrack->master->ct_general); 691 NF_CT_STAT_INC(expect_new); 692 } else { 693 struct nf_conn_help *help = nfct_help(conntrack); 694 695 if (help) 696 help->helper = __nf_ct_helper_find(&repl_tuple); 697 NF_CT_STAT_INC(new); 698 } 699 700 /* Overload tuple linked list to put us in unconfirmed list. */ 701 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); 702 703 write_unlock_bh(&nf_conntrack_lock); 704 705 if (exp) { 706 if (exp->expectfn) 707 exp->expectfn(conntrack, exp); 708 nf_conntrack_expect_put(exp); 709 } 710 711 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; 712} 713 714/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 715static inline struct nf_conn * 716resolve_normal_ct(struct sk_buff *skb, 717 unsigned int dataoff, 718 u_int16_t l3num, 719 u_int8_t protonum, 720 struct nf_conntrack_l3proto *l3proto, 721 struct nf_conntrack_protocol *proto, 722 int *set_reply, 723 enum ip_conntrack_info *ctinfo) 724{ 725 struct nf_conntrack_tuple tuple; 726 struct nf_conntrack_tuple_hash *h; 727 struct nf_conn *ct; 728 729 if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data), 730 dataoff, l3num, protonum, &tuple, l3proto, 731 proto)) { 732 DEBUGP("resolve_normal_ct: Can't get tuple\n"); 733 return NULL; 734 } 735 736 /* look for tuple match */ 737 h = nf_conntrack_find_get(&tuple, NULL); 738 if (!h) { 739 h = init_conntrack(&tuple, l3proto, proto, skb, dataoff); 740 if (!h) 741 return NULL; 742 if (IS_ERR(h)) 743 return (void *)h; 744 } 745 ct = nf_ct_tuplehash_to_ctrack(h); 746 747 /* It exists; we have (non-exclusive) reference. */ 748 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 749 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 750 /* Please set reply bit if this packet OK */ 751 *set_reply = 1; 752 } else { 753 /* Once we've had two way comms, always ESTABLISHED. */ 754 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 755 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct); 756 *ctinfo = IP_CT_ESTABLISHED; 757 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 758 DEBUGP("nf_conntrack_in: related packet for %p\n", ct); 759 *ctinfo = IP_CT_RELATED; 760 } else { 761 DEBUGP("nf_conntrack_in: new packet for %p\n", ct); 762 *ctinfo = IP_CT_NEW; 763 } 764 *set_reply = 0; 765 } 766 skb->nfct = &ct->ct_general; 767 skb->nfctinfo = *ctinfo; 768 return ct; 769} 770 771unsigned int 772nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) 773{ 774 struct nf_conn *ct; 775 enum ip_conntrack_info ctinfo; 776 struct nf_conntrack_l3proto *l3proto; 777 struct nf_conntrack_protocol *proto; 778 unsigned int dataoff; 779 u_int8_t protonum; 780 int set_reply = 0; 781 int ret; 782 783 /* Previously seen (loopback or untracked)? Ignore. */ 784 if ((*pskb)->nfct) { 785 NF_CT_STAT_INC(ignore); 786 return NF_ACCEPT; 787 } 788 789 l3proto = __nf_ct_l3proto_find((u_int16_t)pf); 790 if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) { 791 DEBUGP("not prepared to track yet or error occured\n"); 792 return -ret; 793 } 794 795 proto = __nf_ct_proto_find((u_int16_t)pf, protonum); 796 797 /* It may be an special packet, error, unclean... 798 * inverse of the return code tells to the netfilter 799 * core what to do with the packet. */ 800 if (proto->error != NULL && 801 (ret = proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { 802 NF_CT_STAT_INC(error); 803 NF_CT_STAT_INC(invalid); 804 return -ret; 805 } 806 807 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, proto, 808 &set_reply, &ctinfo); 809 if (!ct) { 810 /* Not valid part of a connection */ 811 NF_CT_STAT_INC(invalid); 812 return NF_ACCEPT; 813 } 814 815 if (IS_ERR(ct)) { 816 /* Too stressed to deal. */ 817 NF_CT_STAT_INC(drop); 818 return NF_DROP; 819 } 820 821 NF_CT_ASSERT((*pskb)->nfct); 822 823 ret = proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); 824 if (ret < 0) { 825 /* Invalid: inverse of the return code tells 826 * the netfilter core what to do */ 827 DEBUGP("nf_conntrack_in: Can't track with proto module\n"); 828 nf_conntrack_put((*pskb)->nfct); 829 (*pskb)->nfct = NULL; 830 NF_CT_STAT_INC(invalid); 831 return -ret; 832 } 833 834 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 835 nf_conntrack_event_cache(IPCT_STATUS, *pskb); 836 837 return ret; 838} 839 840int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 841 const struct nf_conntrack_tuple *orig) 842{ 843 return nf_ct_invert_tuple(inverse, orig, 844 __nf_ct_l3proto_find(orig->src.l3num), 845 __nf_ct_proto_find(orig->src.l3num, 846 orig->dst.protonum)); 847} 848 849/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 850void __nf_ct_refresh_acct(struct nf_conn *ct, 851 enum ip_conntrack_info ctinfo, 852 const struct sk_buff *skb, 853 unsigned long extra_jiffies, 854 int do_acct) 855{ 856 int event = 0; 857 858 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 859 NF_CT_ASSERT(skb); 860 861 write_lock_bh(&nf_conntrack_lock); 862 863 /* Only update if this is not a fixed timeout */ 864 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { 865 write_unlock_bh(&nf_conntrack_lock); 866 return; 867 } 868 869 /* If not in hash table, timer will not be active yet */ 870 if (!nf_ct_is_confirmed(ct)) { 871 ct->timeout.expires = extra_jiffies; 872 event = IPCT_REFRESH; 873 } else { 874 /* Need del_timer for race avoidance (may already be dying). */ 875 if (del_timer(&ct->timeout)) { 876 ct->timeout.expires = jiffies + extra_jiffies; 877 add_timer(&ct->timeout); 878 event = IPCT_REFRESH; 879 } 880 } 881 882#ifdef CONFIG_NF_CT_ACCT 883 if (do_acct) { 884 ct->counters[CTINFO2DIR(ctinfo)].packets++; 885 ct->counters[CTINFO2DIR(ctinfo)].bytes += 886 skb->len - (unsigned int)(skb->nh.raw - skb->data); 887 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) 888 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) 889 event |= IPCT_COUNTER_FILLING; 890 } 891#endif 892 893 write_unlock_bh(&nf_conntrack_lock); 894 895 /* must be unlocked when calling event cache */ 896 if (event) 897 nf_conntrack_event_cache(event, skb); 898} 899 900#if defined(CONFIG_NF_CT_NETLINK) || \ 901 defined(CONFIG_NF_CT_NETLINK_MODULE) 902 903#include <linux/netfilter/nfnetlink.h> 904#include <linux/netfilter/nfnetlink_conntrack.h> 905#include <linux/mutex.h> 906 907 908/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 909 * in ip_conntrack_core, since we don't want the protocols to autoload 910 * or depend on ctnetlink */ 911int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb, 912 const struct nf_conntrack_tuple *tuple) 913{ 914 NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), 915 &tuple->src.u.tcp.port); 916 NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), 917 &tuple->dst.u.tcp.port); 918 return 0; 919 920nfattr_failure: 921 return -1; 922} 923 924static const size_t cta_min_proto[CTA_PROTO_MAX] = { 925 [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t), 926 [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t) 927}; 928 929int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[], 930 struct nf_conntrack_tuple *t) 931{ 932 if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) 933 return -EINVAL; 934 935 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) 936 return -EINVAL; 937 938 t->src.u.tcp.port = 939 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); 940 t->dst.u.tcp.port = 941 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); 942 943 return 0; 944} 945#endif 946 947/* Used by ipt_REJECT and ip6t_REJECT. */ 948void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 949{ 950 struct nf_conn *ct; 951 enum ip_conntrack_info ctinfo; 952 953 /* This ICMP is in reverse direction to the packet which caused it */ 954 ct = nf_ct_get(skb, &ctinfo); 955 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 956 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 957 else 958 ctinfo = IP_CT_RELATED; 959 960 /* Attach to new skbuff, and increment count */ 961 nskb->nfct = &ct->ct_general; 962 nskb->nfctinfo = ctinfo; 963 nf_conntrack_get(nskb->nfct); 964} 965 966static inline int 967do_iter(const struct nf_conntrack_tuple_hash *i, 968 int (*iter)(struct nf_conn *i, void *data), 969 void *data) 970{ 971 return iter(nf_ct_tuplehash_to_ctrack(i), data); 972} 973 974/* Bring out ya dead! */ 975static struct nf_conn * 976get_next_corpse(int (*iter)(struct nf_conn *i, void *data), 977 void *data, unsigned int *bucket) 978{ 979 struct nf_conntrack_tuple_hash *h; 980 struct nf_conn *ct; 981 982 write_lock_bh(&nf_conntrack_lock); 983 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 984 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) { 985 ct = nf_ct_tuplehash_to_ctrack(h); 986 if (iter(ct, data)) 987 goto found; 988 } 989 } 990 list_for_each_entry(h, &unconfirmed, list) { 991 ct = nf_ct_tuplehash_to_ctrack(h); 992 if (iter(ct, data)) 993 goto found; 994 } 995 write_unlock_bh(&nf_conntrack_lock); 996 return NULL; 997found: 998 atomic_inc(&ct->ct_general.use); 999 write_unlock_bh(&nf_conntrack_lock); 1000 return ct; 1001} 1002 1003void 1004nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) 1005{ 1006 struct nf_conn *ct; 1007 unsigned int bucket = 0; 1008 1009 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { 1010 /* Time to push up daises... */ 1011 if (del_timer(&ct->timeout)) 1012 death_by_timeout((unsigned long)ct); 1013 /* ... else the timer will get him soon. */ 1014 1015 nf_ct_put(ct); 1016 } 1017} 1018 1019static int kill_all(struct nf_conn *i, void *data) 1020{ 1021 return 1; 1022} 1023 1024static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size) 1025{ 1026 if (vmalloced) 1027 vfree(hash); 1028 else 1029 free_pages((unsigned long)hash, 1030 get_order(sizeof(struct list_head) * size)); 1031} 1032 1033void nf_conntrack_flush() 1034{ 1035 nf_ct_iterate_cleanup(kill_all, NULL); 1036} 1037 1038/* Mishearing the voices in his head, our hero wonders how he's 1039 supposed to kill the mall. */ 1040void nf_conntrack_cleanup(void) 1041{ 1042 int i; 1043 1044 ip_ct_attach = NULL; 1045 1046 /* This makes sure all current packets have passed through 1047 netfilter framework. Roll on, two-stage module 1048 delete... */ 1049 synchronize_net(); 1050 1051 nf_ct_event_cache_flush(); 1052 i_see_dead_people: 1053 nf_conntrack_flush(); 1054 if (atomic_read(&nf_conntrack_count) != 0) { 1055 schedule(); 1056 goto i_see_dead_people; 1057 } 1058 /* wait until all references to nf_conntrack_untracked are dropped */ 1059 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) 1060 schedule(); 1061 1062 for (i = 0; i < NF_CT_F_NUM; i++) { 1063 if (nf_ct_cache[i].use == 0) 1064 continue; 1065 1066 NF_CT_ASSERT(nf_ct_cache[i].use == 1); 1067 nf_ct_cache[i].use = 1; 1068 nf_conntrack_unregister_cache(i); 1069 } 1070 kmem_cache_destroy(nf_conntrack_expect_cachep); 1071 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1072 nf_conntrack_htable_size); 1073 1074 /* free l3proto protocol tables */ 1075 for (i = 0; i < PF_MAX; i++) 1076 if (nf_ct_protos[i]) { 1077 kfree(nf_ct_protos[i]); 1078 nf_ct_protos[i] = NULL; 1079 } 1080} 1081 1082static struct list_head *alloc_hashtable(int size, int *vmalloced) 1083{ 1084 struct list_head *hash; 1085 unsigned int i; 1086 1087 *vmalloced = 0; 1088 hash = (void*)__get_free_pages(GFP_KERNEL, 1089 get_order(sizeof(struct list_head) 1090 * size)); 1091 if (!hash) { 1092 *vmalloced = 1; 1093 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1094 hash = vmalloc(sizeof(struct list_head) * size); 1095 } 1096 1097 if (hash) 1098 for (i = 0; i < size; i++) 1099 INIT_LIST_HEAD(&hash[i]); 1100 1101 return hash; 1102} 1103 1104int set_hashsize(const char *val, struct kernel_param *kp) 1105{ 1106 int i, bucket, hashsize, vmalloced; 1107 int old_vmalloced, old_size; 1108 int rnd; 1109 struct list_head *hash, *old_hash; 1110 struct nf_conntrack_tuple_hash *h; 1111 1112 /* On boot, we can set this without any fancy locking. */ 1113 if (!nf_conntrack_htable_size) 1114 return param_set_uint(val, kp); 1115 1116 hashsize = simple_strtol(val, NULL, 0); 1117 if (!hashsize) 1118 return -EINVAL; 1119 1120 hash = alloc_hashtable(hashsize, &vmalloced); 1121 if (!hash) 1122 return -ENOMEM; 1123 1124 /* We have to rehahs for the new table anyway, so we also can 1125 * use a newrandom seed */ 1126 get_random_bytes(&rnd, 4); 1127 1128 write_lock_bh(&nf_conntrack_lock); 1129 for (i = 0; i < nf_conntrack_htable_size; i++) { 1130 while (!list_empty(&nf_conntrack_hash[i])) { 1131 h = list_entry(nf_conntrack_hash[i].next, 1132 struct nf_conntrack_tuple_hash, list); 1133 list_del(&h->list); 1134 bucket = __hash_conntrack(&h->tuple, hashsize, rnd); 1135 list_add_tail(&h->list, &hash[bucket]); 1136 } 1137 } 1138 old_size = nf_conntrack_htable_size; 1139 old_vmalloced = nf_conntrack_vmalloc; 1140 old_hash = nf_conntrack_hash; 1141 1142 nf_conntrack_htable_size = hashsize; 1143 nf_conntrack_vmalloc = vmalloced; 1144 nf_conntrack_hash = hash; 1145 nf_conntrack_hash_rnd = rnd; 1146 write_unlock_bh(&nf_conntrack_lock); 1147 1148 free_conntrack_hash(old_hash, old_vmalloced, old_size); 1149 return 0; 1150} 1151 1152module_param_call(hashsize, set_hashsize, param_get_uint, 1153 &nf_conntrack_htable_size, 0600); 1154 1155int __init nf_conntrack_init(void) 1156{ 1157 unsigned int i; 1158 int ret; 1159 1160 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1161 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ 1162 if (!nf_conntrack_htable_size) { 1163 nf_conntrack_htable_size 1164 = (((num_physpages << PAGE_SHIFT) / 16384) 1165 / sizeof(struct list_head)); 1166 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1167 nf_conntrack_htable_size = 8192; 1168 if (nf_conntrack_htable_size < 16) 1169 nf_conntrack_htable_size = 16; 1170 } 1171 nf_conntrack_max = 8 * nf_conntrack_htable_size; 1172 1173 printk("nf_conntrack version %s (%u buckets, %d max)\n", 1174 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1175 nf_conntrack_max); 1176 1177 nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size, 1178 &nf_conntrack_vmalloc); 1179 if (!nf_conntrack_hash) { 1180 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1181 goto err_out; 1182 } 1183 1184 ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic", 1185 sizeof(struct nf_conn)); 1186 if (ret < 0) { 1187 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1188 goto err_free_hash; 1189 } 1190 1191 nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect", 1192 sizeof(struct nf_conntrack_expect), 1193 0, 0, NULL, NULL); 1194 if (!nf_conntrack_expect_cachep) { 1195 printk(KERN_ERR "Unable to create nf_expect slab cache\n"); 1196 goto err_free_conntrack_slab; 1197 } 1198 1199 /* Don't NEED lock here, but good form anyway. */ 1200 write_lock_bh(&nf_conntrack_lock); 1201 for (i = 0; i < PF_MAX; i++) 1202 nf_ct_l3protos[i] = &nf_conntrack_generic_l3proto; 1203 write_unlock_bh(&nf_conntrack_lock); 1204 1205 /* For use by REJECT target */ 1206 ip_ct_attach = __nf_conntrack_attach; 1207 1208 /* Set up fake conntrack: 1209 - to never be deleted, not in any hashes */ 1210 atomic_set(&nf_conntrack_untracked.ct_general.use, 1); 1211 /* - and look it like as a confirmed connection */ 1212 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); 1213 1214 return ret; 1215 1216err_free_conntrack_slab: 1217 nf_conntrack_unregister_cache(NF_CT_F_BASIC); 1218err_free_hash: 1219 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1220 nf_conntrack_htable_size); 1221err_out: 1222 return -ENOMEM; 1223} 1224