nf_conntrack_core.c revision be00c8e48993368663e2714bd1e7c886b7736406
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); 69 70void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL; 71unsigned int nf_conntrack_htable_size __read_mostly; 72int nf_conntrack_max __read_mostly; 73struct list_head *nf_conntrack_hash __read_mostly; 74struct nf_conn nf_conntrack_untracked __read_mostly; 75unsigned int nf_ct_log_invalid __read_mostly; 76LIST_HEAD(unconfirmed); 77static int nf_conntrack_vmalloc __read_mostly; 78 79static unsigned int nf_conntrack_next_id; 80 81DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); 82EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); 83 84/* 85 * This scheme offers various size of "struct nf_conn" dependent on 86 * features(helper, nat, ...) 87 */ 88 89#define NF_CT_FEATURES_NAMELEN 256 90static struct { 91 /* name of slab cache. printed in /proc/slabinfo */ 92 char *name; 93 94 /* size of slab cache */ 95 size_t size; 96 97 /* slab cache pointer */ 98 kmem_cache_t *cachep; 99 100 /* allocated slab cache + modules which uses this slab cache */ 101 int use; 102 103} nf_ct_cache[NF_CT_F_NUM]; 104 105/* protect members of nf_ct_cache except of "use" */ 106DEFINE_RWLOCK(nf_ct_cache_lock); 107 108/* This avoids calling kmem_cache_create() with same name simultaneously */ 109static DEFINE_MUTEX(nf_ct_cache_mutex); 110 111static int nf_conntrack_hash_rnd_initted; 112static unsigned int nf_conntrack_hash_rnd; 113 114static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 115 unsigned int size, unsigned int rnd) 116{ 117 unsigned int a, b; 118 a = jhash((void *)tuple->src.u3.all, sizeof(tuple->src.u3.all), 119 ((tuple->src.l3num) << 16) | tuple->dst.protonum); 120 b = jhash((void *)tuple->dst.u3.all, sizeof(tuple->dst.u3.all), 121 (tuple->src.u.all << 16) | tuple->dst.u.all); 122 123 return jhash_2words(a, b, rnd) % size; 124} 125 126static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) 127{ 128 return __hash_conntrack(tuple, nf_conntrack_htable_size, 129 nf_conntrack_hash_rnd); 130} 131 132int nf_conntrack_register_cache(u_int32_t features, const char *name, 133 size_t size) 134{ 135 int ret = 0; 136 char *cache_name; 137 kmem_cache_t *cachep; 138 139 DEBUGP("nf_conntrack_register_cache: features=0x%x, name=%s, size=%d\n", 140 features, name, size); 141 142 if (features < NF_CT_F_BASIC || features >= NF_CT_F_NUM) { 143 DEBUGP("nf_conntrack_register_cache: invalid features.: 0x%x\n", 144 features); 145 return -EINVAL; 146 } 147 148 mutex_lock(&nf_ct_cache_mutex); 149 150 write_lock_bh(&nf_ct_cache_lock); 151 /* e.g: multiple helpers are loaded */ 152 if (nf_ct_cache[features].use > 0) { 153 DEBUGP("nf_conntrack_register_cache: already resisterd.\n"); 154 if ((!strncmp(nf_ct_cache[features].name, name, 155 NF_CT_FEATURES_NAMELEN)) 156 && nf_ct_cache[features].size == size) { 157 DEBUGP("nf_conntrack_register_cache: reusing.\n"); 158 nf_ct_cache[features].use++; 159 ret = 0; 160 } else 161 ret = -EBUSY; 162 163 write_unlock_bh(&nf_ct_cache_lock); 164 mutex_unlock(&nf_ct_cache_mutex); 165 return ret; 166 } 167 write_unlock_bh(&nf_ct_cache_lock); 168 169 /* 170 * The memory space for name of slab cache must be alive until 171 * cache is destroyed. 172 */ 173 cache_name = kmalloc(sizeof(char)*NF_CT_FEATURES_NAMELEN, GFP_ATOMIC); 174 if (cache_name == NULL) { 175 DEBUGP("nf_conntrack_register_cache: can't alloc cache_name\n"); 176 ret = -ENOMEM; 177 goto out_up_mutex; 178 } 179 180 if (strlcpy(cache_name, name, NF_CT_FEATURES_NAMELEN) 181 >= NF_CT_FEATURES_NAMELEN) { 182 printk("nf_conntrack_register_cache: name too long\n"); 183 ret = -EINVAL; 184 goto out_free_name; 185 } 186 187 cachep = kmem_cache_create(cache_name, size, 0, 0, 188 NULL, NULL); 189 if (!cachep) { 190 printk("nf_conntrack_register_cache: Can't create slab cache " 191 "for the features = 0x%x\n", features); 192 ret = -ENOMEM; 193 goto out_free_name; 194 } 195 196 write_lock_bh(&nf_ct_cache_lock); 197 nf_ct_cache[features].use = 1; 198 nf_ct_cache[features].size = size; 199 nf_ct_cache[features].cachep = cachep; 200 nf_ct_cache[features].name = cache_name; 201 write_unlock_bh(&nf_ct_cache_lock); 202 203 goto out_up_mutex; 204 205out_free_name: 206 kfree(cache_name); 207out_up_mutex: 208 mutex_unlock(&nf_ct_cache_mutex); 209 return ret; 210} 211 212/* FIXME: In the current, only nf_conntrack_cleanup() can call this function. */ 213void nf_conntrack_unregister_cache(u_int32_t features) 214{ 215 kmem_cache_t *cachep; 216 char *name; 217 218 /* 219 * This assures that kmem_cache_create() isn't called before destroying 220 * slab cache. 221 */ 222 DEBUGP("nf_conntrack_unregister_cache: 0x%04x\n", features); 223 mutex_lock(&nf_ct_cache_mutex); 224 225 write_lock_bh(&nf_ct_cache_lock); 226 if (--nf_ct_cache[features].use > 0) { 227 write_unlock_bh(&nf_ct_cache_lock); 228 mutex_unlock(&nf_ct_cache_mutex); 229 return; 230 } 231 cachep = nf_ct_cache[features].cachep; 232 name = nf_ct_cache[features].name; 233 nf_ct_cache[features].cachep = NULL; 234 nf_ct_cache[features].name = NULL; 235 nf_ct_cache[features].size = 0; 236 write_unlock_bh(&nf_ct_cache_lock); 237 238 synchronize_net(); 239 240 kmem_cache_destroy(cachep); 241 kfree(name); 242 243 mutex_unlock(&nf_ct_cache_mutex); 244} 245 246int 247nf_ct_get_tuple(const struct sk_buff *skb, 248 unsigned int nhoff, 249 unsigned int dataoff, 250 u_int16_t l3num, 251 u_int8_t protonum, 252 struct nf_conntrack_tuple *tuple, 253 const struct nf_conntrack_l3proto *l3proto, 254 const struct nf_conntrack_l4proto *l4proto) 255{ 256 NF_CT_TUPLE_U_BLANK(tuple); 257 258 tuple->src.l3num = l3num; 259 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 260 return 0; 261 262 tuple->dst.protonum = protonum; 263 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 264 265 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 266} 267 268int 269nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 270 const struct nf_conntrack_tuple *orig, 271 const struct nf_conntrack_l3proto *l3proto, 272 const struct nf_conntrack_l4proto *l4proto) 273{ 274 NF_CT_TUPLE_U_BLANK(inverse); 275 276 inverse->src.l3num = orig->src.l3num; 277 if (l3proto->invert_tuple(inverse, orig) == 0) 278 return 0; 279 280 inverse->dst.dir = !orig->dst.dir; 281 282 inverse->dst.protonum = orig->dst.protonum; 283 return l4proto->invert_tuple(inverse, orig); 284} 285 286static void 287clean_from_lists(struct nf_conn *ct) 288{ 289 DEBUGP("clean_from_lists(%p)\n", ct); 290 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 291 list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list); 292 293 /* Destroy all pending expectations */ 294 nf_ct_remove_expectations(ct); 295} 296 297static void 298destroy_conntrack(struct nf_conntrack *nfct) 299{ 300 struct nf_conn *ct = (struct nf_conn *)nfct; 301 struct nf_conntrack_l3proto *l3proto; 302 struct nf_conntrack_l4proto *l4proto; 303 304 DEBUGP("destroy_conntrack(%p)\n", ct); 305 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 306 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 307 308 nf_conntrack_event(IPCT_DESTROY, ct); 309 set_bit(IPS_DYING_BIT, &ct->status); 310 311 /* To make sure we don't get any weird locking issues here: 312 * destroy_conntrack() MUST NOT be called with a write lock 313 * to nf_conntrack_lock!!! -HW */ 314 l3proto = __nf_ct_l3proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num); 315 if (l3proto && l3proto->destroy) 316 l3proto->destroy(ct); 317 318 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); 319 if (l4proto && l4proto->destroy) 320 l4proto->destroy(ct); 321 322 if (nf_conntrack_destroyed) 323 nf_conntrack_destroyed(ct); 324 325 write_lock_bh(&nf_conntrack_lock); 326 /* Expectations will have been removed in clean_from_lists, 327 * except TFTP can create an expectation on the first packet, 328 * before connection is in the list, so we need to clean here, 329 * too. */ 330 nf_ct_remove_expectations(ct); 331 332 /* We overload first tuple to link into unconfirmed list. */ 333 if (!nf_ct_is_confirmed(ct)) { 334 BUG_ON(list_empty(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list)); 335 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 336 } 337 338 NF_CT_STAT_INC(delete); 339 write_unlock_bh(&nf_conntrack_lock); 340 341 if (ct->master) 342 nf_ct_put(ct->master); 343 344 DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct); 345 nf_conntrack_free(ct); 346} 347 348static void death_by_timeout(unsigned long ul_conntrack) 349{ 350 struct nf_conn *ct = (void *)ul_conntrack; 351 352 write_lock_bh(&nf_conntrack_lock); 353 /* Inside lock so preempt is disabled on module removal path. 354 * Otherwise we can get spurious warnings. */ 355 NF_CT_STAT_INC(delete_list); 356 clean_from_lists(ct); 357 write_unlock_bh(&nf_conntrack_lock); 358 nf_ct_put(ct); 359} 360 361struct nf_conntrack_tuple_hash * 362__nf_conntrack_find(const struct nf_conntrack_tuple *tuple, 363 const struct nf_conn *ignored_conntrack) 364{ 365 struct nf_conntrack_tuple_hash *h; 366 unsigned int hash = hash_conntrack(tuple); 367 368 list_for_each_entry(h, &nf_conntrack_hash[hash], list) { 369 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && 370 nf_ct_tuple_equal(tuple, &h->tuple)) { 371 NF_CT_STAT_INC(found); 372 return h; 373 } 374 NF_CT_STAT_INC(searched); 375 } 376 377 return NULL; 378} 379 380/* Find a connection corresponding to a tuple. */ 381struct nf_conntrack_tuple_hash * 382nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple, 383 const struct nf_conn *ignored_conntrack) 384{ 385 struct nf_conntrack_tuple_hash *h; 386 387 read_lock_bh(&nf_conntrack_lock); 388 h = __nf_conntrack_find(tuple, ignored_conntrack); 389 if (h) 390 atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use); 391 read_unlock_bh(&nf_conntrack_lock); 392 393 return h; 394} 395 396static void __nf_conntrack_hash_insert(struct nf_conn *ct, 397 unsigned int hash, 398 unsigned int repl_hash) 399{ 400 ct->id = ++nf_conntrack_next_id; 401 list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list, 402 &nf_conntrack_hash[hash]); 403 list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list, 404 &nf_conntrack_hash[repl_hash]); 405} 406 407void nf_conntrack_hash_insert(struct nf_conn *ct) 408{ 409 unsigned int hash, repl_hash; 410 411 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 412 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 413 414 write_lock_bh(&nf_conntrack_lock); 415 __nf_conntrack_hash_insert(ct, hash, repl_hash); 416 write_unlock_bh(&nf_conntrack_lock); 417} 418 419/* Confirm a connection given skb; places it in hash table */ 420int 421__nf_conntrack_confirm(struct sk_buff **pskb) 422{ 423 unsigned int hash, repl_hash; 424 struct nf_conntrack_tuple_hash *h; 425 struct nf_conn *ct; 426 struct nf_conn_help *help; 427 enum ip_conntrack_info ctinfo; 428 429 ct = nf_ct_get(*pskb, &ctinfo); 430 431 /* ipt_REJECT uses nf_conntrack_attach to attach related 432 ICMP/TCP RST packets in other direction. Actual packet 433 which created connection will be IP_CT_NEW or for an 434 expected connection, IP_CT_RELATED. */ 435 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 436 return NF_ACCEPT; 437 438 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 439 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 440 441 /* We're not in hash table, and we refuse to set up related 442 connections for unconfirmed conns. But packet copies and 443 REJECT will give spurious warnings here. */ 444 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 445 446 /* No external references means noone else could have 447 confirmed us. */ 448 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 449 DEBUGP("Confirming conntrack %p\n", ct); 450 451 write_lock_bh(&nf_conntrack_lock); 452 453 /* See if there's one in the list already, including reverse: 454 NAT could have grabbed it without realizing, since we're 455 not in the hash. If there is, we lost race. */ 456 list_for_each_entry(h, &nf_conntrack_hash[hash], list) 457 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 458 &h->tuple)) 459 goto out; 460 list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list) 461 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 462 &h->tuple)) 463 goto out; 464 465 /* Remove from unconfirmed list */ 466 list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list); 467 468 __nf_conntrack_hash_insert(ct, hash, repl_hash); 469 /* Timer relative to confirmation time, not original 470 setting time, otherwise we'd get timer wrap in 471 weird delay cases. */ 472 ct->timeout.expires += jiffies; 473 add_timer(&ct->timeout); 474 atomic_inc(&ct->ct_general.use); 475 set_bit(IPS_CONFIRMED_BIT, &ct->status); 476 NF_CT_STAT_INC(insert); 477 write_unlock_bh(&nf_conntrack_lock); 478 help = nfct_help(ct); 479 if (help && help->helper) 480 nf_conntrack_event_cache(IPCT_HELPER, *pskb); 481#ifdef CONFIG_NF_NAT_NEEDED 482 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || 483 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) 484 nf_conntrack_event_cache(IPCT_NATINFO, *pskb); 485#endif 486 nf_conntrack_event_cache(master_ct(ct) ? 487 IPCT_RELATED : IPCT_NEW, *pskb); 488 return NF_ACCEPT; 489 490out: 491 NF_CT_STAT_INC(insert_failed); 492 write_unlock_bh(&nf_conntrack_lock); 493 return NF_DROP; 494} 495 496/* Returns true if a connection correspondings to the tuple (required 497 for NAT). */ 498int 499nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 500 const struct nf_conn *ignored_conntrack) 501{ 502 struct nf_conntrack_tuple_hash *h; 503 504 read_lock_bh(&nf_conntrack_lock); 505 h = __nf_conntrack_find(tuple, ignored_conntrack); 506 read_unlock_bh(&nf_conntrack_lock); 507 508 return h != NULL; 509} 510 511/* There's a small race here where we may free a just-assured 512 connection. Too bad: we're in trouble anyway. */ 513static int early_drop(struct list_head *chain) 514{ 515 /* Traverse backwards: gives us oldest, which is roughly LRU */ 516 struct nf_conntrack_tuple_hash *h; 517 struct nf_conn *ct = NULL, *tmp; 518 int dropped = 0; 519 520 read_lock_bh(&nf_conntrack_lock); 521 list_for_each_entry_reverse(h, chain, list) { 522 tmp = nf_ct_tuplehash_to_ctrack(h); 523 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) { 524 ct = tmp; 525 atomic_inc(&ct->ct_general.use); 526 break; 527 } 528 } 529 read_unlock_bh(&nf_conntrack_lock); 530 531 if (!ct) 532 return dropped; 533 534 if (del_timer(&ct->timeout)) { 535 death_by_timeout((unsigned long)ct); 536 dropped = 1; 537 NF_CT_STAT_INC(early_drop); 538 } 539 nf_ct_put(ct); 540 return dropped; 541} 542 543static struct nf_conn * 544__nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 545 const struct nf_conntrack_tuple *repl, 546 const struct nf_conntrack_l3proto *l3proto) 547{ 548 struct nf_conn *conntrack = NULL; 549 u_int32_t features = 0; 550 struct nf_conntrack_helper *helper; 551 552 if (unlikely(!nf_conntrack_hash_rnd_initted)) { 553 get_random_bytes(&nf_conntrack_hash_rnd, 4); 554 nf_conntrack_hash_rnd_initted = 1; 555 } 556 557 /* We don't want any race condition at early drop stage */ 558 atomic_inc(&nf_conntrack_count); 559 560 if (nf_conntrack_max 561 && atomic_read(&nf_conntrack_count) > nf_conntrack_max) { 562 unsigned int hash = hash_conntrack(orig); 563 /* Try dropping from this hash chain. */ 564 if (!early_drop(&nf_conntrack_hash[hash])) { 565 atomic_dec(&nf_conntrack_count); 566 if (net_ratelimit()) 567 printk(KERN_WARNING 568 "nf_conntrack: table full, dropping" 569 " packet.\n"); 570 return ERR_PTR(-ENOMEM); 571 } 572 } 573 574 /* find features needed by this conntrack. */ 575 features = l3proto->get_features(orig); 576 577 /* FIXME: protect helper list per RCU */ 578 read_lock_bh(&nf_conntrack_lock); 579 helper = __nf_ct_helper_find(repl); 580 if (helper) 581 features |= NF_CT_F_HELP; 582 read_unlock_bh(&nf_conntrack_lock); 583 584 DEBUGP("nf_conntrack_alloc: features=0x%x\n", features); 585 586 read_lock_bh(&nf_ct_cache_lock); 587 588 if (unlikely(!nf_ct_cache[features].use)) { 589 DEBUGP("nf_conntrack_alloc: not supported features = 0x%x\n", 590 features); 591 goto out; 592 } 593 594 conntrack = kmem_cache_alloc(nf_ct_cache[features].cachep, GFP_ATOMIC); 595 if (conntrack == NULL) { 596 DEBUGP("nf_conntrack_alloc: Can't alloc conntrack from cache\n"); 597 goto out; 598 } 599 600 memset(conntrack, 0, nf_ct_cache[features].size); 601 conntrack->features = features; 602 atomic_set(&conntrack->ct_general.use, 1); 603 conntrack->ct_general.destroy = destroy_conntrack; 604 conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 605 conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 606 /* Don't set timer yet: wait for confirmation */ 607 init_timer(&conntrack->timeout); 608 conntrack->timeout.data = (unsigned long)conntrack; 609 conntrack->timeout.function = death_by_timeout; 610 read_unlock_bh(&nf_ct_cache_lock); 611 612 return conntrack; 613out: 614 read_unlock_bh(&nf_ct_cache_lock); 615 atomic_dec(&nf_conntrack_count); 616 return conntrack; 617} 618 619struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 620 const struct nf_conntrack_tuple *repl) 621{ 622 struct nf_conntrack_l3proto *l3proto; 623 624 l3proto = __nf_ct_l3proto_find(orig->src.l3num); 625 return __nf_conntrack_alloc(orig, repl, l3proto); 626} 627 628void nf_conntrack_free(struct nf_conn *conntrack) 629{ 630 u_int32_t features = conntrack->features; 631 NF_CT_ASSERT(features >= NF_CT_F_BASIC && features < NF_CT_F_NUM); 632 DEBUGP("nf_conntrack_free: features = 0x%x, conntrack=%p\n", features, 633 conntrack); 634 kmem_cache_free(nf_ct_cache[features].cachep, conntrack); 635 atomic_dec(&nf_conntrack_count); 636} 637 638/* Allocate a new conntrack: we return -ENOMEM if classification 639 failed due to stress. Otherwise it really is unclassifiable. */ 640static struct nf_conntrack_tuple_hash * 641init_conntrack(const struct nf_conntrack_tuple *tuple, 642 struct nf_conntrack_l3proto *l3proto, 643 struct nf_conntrack_l4proto *l4proto, 644 struct sk_buff *skb, 645 unsigned int dataoff) 646{ 647 struct nf_conn *conntrack; 648 struct nf_conntrack_tuple repl_tuple; 649 struct nf_conntrack_expect *exp; 650 651 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 652 DEBUGP("Can't invert tuple.\n"); 653 return NULL; 654 } 655 656 conntrack = __nf_conntrack_alloc(tuple, &repl_tuple, l3proto); 657 if (conntrack == NULL || IS_ERR(conntrack)) { 658 DEBUGP("Can't allocate conntrack.\n"); 659 return (struct nf_conntrack_tuple_hash *)conntrack; 660 } 661 662 if (!l4proto->new(conntrack, skb, dataoff)) { 663 nf_conntrack_free(conntrack); 664 DEBUGP("init conntrack: can't track with proto module\n"); 665 return NULL; 666 } 667 668 write_lock_bh(&nf_conntrack_lock); 669 exp = find_expectation(tuple); 670 671 if (exp) { 672 DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n", 673 conntrack, exp); 674 /* Welcome, Mr. Bond. We've been expecting you... */ 675 __set_bit(IPS_EXPECTED_BIT, &conntrack->status); 676 conntrack->master = exp->master; 677#ifdef CONFIG_NF_CONNTRACK_MARK 678 conntrack->mark = exp->master->mark; 679#endif 680#ifdef CONFIG_NF_CONNTRACK_SECMARK 681 conntrack->secmark = exp->master->secmark; 682#endif 683 nf_conntrack_get(&conntrack->master->ct_general); 684 NF_CT_STAT_INC(expect_new); 685 } else { 686 struct nf_conn_help *help = nfct_help(conntrack); 687 688 if (help) 689 help->helper = __nf_ct_helper_find(&repl_tuple); 690 NF_CT_STAT_INC(new); 691 } 692 693 /* Overload tuple linked list to put us in unconfirmed list. */ 694 list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed); 695 696 write_unlock_bh(&nf_conntrack_lock); 697 698 if (exp) { 699 if (exp->expectfn) 700 exp->expectfn(conntrack, exp); 701 nf_conntrack_expect_put(exp); 702 } 703 704 return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL]; 705} 706 707/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 708static inline struct nf_conn * 709resolve_normal_ct(struct sk_buff *skb, 710 unsigned int dataoff, 711 u_int16_t l3num, 712 u_int8_t protonum, 713 struct nf_conntrack_l3proto *l3proto, 714 struct nf_conntrack_l4proto *l4proto, 715 int *set_reply, 716 enum ip_conntrack_info *ctinfo) 717{ 718 struct nf_conntrack_tuple tuple; 719 struct nf_conntrack_tuple_hash *h; 720 struct nf_conn *ct; 721 722 if (!nf_ct_get_tuple(skb, (unsigned int)(skb->nh.raw - skb->data), 723 dataoff, l3num, protonum, &tuple, l3proto, 724 l4proto)) { 725 DEBUGP("resolve_normal_ct: Can't get tuple\n"); 726 return NULL; 727 } 728 729 /* look for tuple match */ 730 h = nf_conntrack_find_get(&tuple, NULL); 731 if (!h) { 732 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); 733 if (!h) 734 return NULL; 735 if (IS_ERR(h)) 736 return (void *)h; 737 } 738 ct = nf_ct_tuplehash_to_ctrack(h); 739 740 /* It exists; we have (non-exclusive) reference. */ 741 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 742 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 743 /* Please set reply bit if this packet OK */ 744 *set_reply = 1; 745 } else { 746 /* Once we've had two way comms, always ESTABLISHED. */ 747 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 748 DEBUGP("nf_conntrack_in: normal packet for %p\n", ct); 749 *ctinfo = IP_CT_ESTABLISHED; 750 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 751 DEBUGP("nf_conntrack_in: related packet for %p\n", ct); 752 *ctinfo = IP_CT_RELATED; 753 } else { 754 DEBUGP("nf_conntrack_in: new packet for %p\n", ct); 755 *ctinfo = IP_CT_NEW; 756 } 757 *set_reply = 0; 758 } 759 skb->nfct = &ct->ct_general; 760 skb->nfctinfo = *ctinfo; 761 return ct; 762} 763 764unsigned int 765nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb) 766{ 767 struct nf_conn *ct; 768 enum ip_conntrack_info ctinfo; 769 struct nf_conntrack_l3proto *l3proto; 770 struct nf_conntrack_l4proto *l4proto; 771 unsigned int dataoff; 772 u_int8_t protonum; 773 int set_reply = 0; 774 int ret; 775 776 /* Previously seen (loopback or untracked)? Ignore. */ 777 if ((*pskb)->nfct) { 778 NF_CT_STAT_INC(ignore); 779 return NF_ACCEPT; 780 } 781 782 l3proto = __nf_ct_l3proto_find((u_int16_t)pf); 783 if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) { 784 DEBUGP("not prepared to track yet or error occured\n"); 785 return -ret; 786 } 787 788 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); 789 790 /* It may be an special packet, error, unclean... 791 * inverse of the return code tells to the netfilter 792 * core what to do with the packet. */ 793 if (l4proto->error != NULL && 794 (ret = l4proto->error(*pskb, dataoff, &ctinfo, pf, hooknum)) <= 0) { 795 NF_CT_STAT_INC(error); 796 NF_CT_STAT_INC(invalid); 797 return -ret; 798 } 799 800 ct = resolve_normal_ct(*pskb, dataoff, pf, protonum, l3proto, l4proto, 801 &set_reply, &ctinfo); 802 if (!ct) { 803 /* Not valid part of a connection */ 804 NF_CT_STAT_INC(invalid); 805 return NF_ACCEPT; 806 } 807 808 if (IS_ERR(ct)) { 809 /* Too stressed to deal. */ 810 NF_CT_STAT_INC(drop); 811 return NF_DROP; 812 } 813 814 NF_CT_ASSERT((*pskb)->nfct); 815 816 ret = l4proto->packet(ct, *pskb, dataoff, ctinfo, pf, hooknum); 817 if (ret < 0) { 818 /* Invalid: inverse of the return code tells 819 * the netfilter core what to do */ 820 DEBUGP("nf_conntrack_in: Can't track with proto module\n"); 821 nf_conntrack_put((*pskb)->nfct); 822 (*pskb)->nfct = NULL; 823 NF_CT_STAT_INC(invalid); 824 return -ret; 825 } 826 827 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 828 nf_conntrack_event_cache(IPCT_STATUS, *pskb); 829 830 return ret; 831} 832 833int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 834 const struct nf_conntrack_tuple *orig) 835{ 836 return nf_ct_invert_tuple(inverse, orig, 837 __nf_ct_l3proto_find(orig->src.l3num), 838 __nf_ct_l4proto_find(orig->src.l3num, 839 orig->dst.protonum)); 840} 841 842/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 843void __nf_ct_refresh_acct(struct nf_conn *ct, 844 enum ip_conntrack_info ctinfo, 845 const struct sk_buff *skb, 846 unsigned long extra_jiffies, 847 int do_acct) 848{ 849 int event = 0; 850 851 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 852 NF_CT_ASSERT(skb); 853 854 write_lock_bh(&nf_conntrack_lock); 855 856 /* Only update if this is not a fixed timeout */ 857 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) { 858 write_unlock_bh(&nf_conntrack_lock); 859 return; 860 } 861 862 /* If not in hash table, timer will not be active yet */ 863 if (!nf_ct_is_confirmed(ct)) { 864 ct->timeout.expires = extra_jiffies; 865 event = IPCT_REFRESH; 866 } else { 867 unsigned long newtime = jiffies + extra_jiffies; 868 869 /* Only update the timeout if the new timeout is at least 870 HZ jiffies from the old timeout. Need del_timer for race 871 avoidance (may already be dying). */ 872 if (newtime - ct->timeout.expires >= HZ 873 && del_timer(&ct->timeout)) { 874 ct->timeout.expires = newtime; 875 add_timer(&ct->timeout); 876 event = IPCT_REFRESH; 877 } 878 } 879 880#ifdef CONFIG_NF_CT_ACCT 881 if (do_acct) { 882 ct->counters[CTINFO2DIR(ctinfo)].packets++; 883 ct->counters[CTINFO2DIR(ctinfo)].bytes += 884 skb->len - (unsigned int)(skb->nh.raw - skb->data); 885 886 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) 887 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) 888 event |= IPCT_COUNTER_FILLING; 889 } 890#endif 891 892 write_unlock_bh(&nf_conntrack_lock); 893 894 /* must be unlocked when calling event cache */ 895 if (event) 896 nf_conntrack_event_cache(event, skb); 897} 898 899#if defined(CONFIG_NF_CT_NETLINK) || \ 900 defined(CONFIG_NF_CT_NETLINK_MODULE) 901 902#include <linux/netfilter/nfnetlink.h> 903#include <linux/netfilter/nfnetlink_conntrack.h> 904#include <linux/mutex.h> 905 906 907/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 908 * in ip_conntrack_core, since we don't want the protocols to autoload 909 * or depend on ctnetlink */ 910int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb, 911 const struct nf_conntrack_tuple *tuple) 912{ 913 NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t), 914 &tuple->src.u.tcp.port); 915 NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t), 916 &tuple->dst.u.tcp.port); 917 return 0; 918 919nfattr_failure: 920 return -1; 921} 922 923static const size_t cta_min_proto[CTA_PROTO_MAX] = { 924 [CTA_PROTO_SRC_PORT-1] = sizeof(u_int16_t), 925 [CTA_PROTO_DST_PORT-1] = sizeof(u_int16_t) 926}; 927 928int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[], 929 struct nf_conntrack_tuple *t) 930{ 931 if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1]) 932 return -EINVAL; 933 934 if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto)) 935 return -EINVAL; 936 937 t->src.u.tcp.port = 938 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]); 939 t->dst.u.tcp.port = 940 *(u_int16_t *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]); 941 942 return 0; 943} 944#endif 945 946/* Used by ipt_REJECT and ip6t_REJECT. */ 947void __nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 948{ 949 struct nf_conn *ct; 950 enum ip_conntrack_info ctinfo; 951 952 /* This ICMP is in reverse direction to the packet which caused it */ 953 ct = nf_ct_get(skb, &ctinfo); 954 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 955 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 956 else 957 ctinfo = IP_CT_RELATED; 958 959 /* Attach to new skbuff, and increment count */ 960 nskb->nfct = &ct->ct_general; 961 nskb->nfctinfo = ctinfo; 962 nf_conntrack_get(nskb->nfct); 963} 964 965static inline int 966do_iter(const struct nf_conntrack_tuple_hash *i, 967 int (*iter)(struct nf_conn *i, void *data), 968 void *data) 969{ 970 return iter(nf_ct_tuplehash_to_ctrack(i), data); 971} 972 973/* Bring out ya dead! */ 974static struct nf_conn * 975get_next_corpse(int (*iter)(struct nf_conn *i, void *data), 976 void *data, unsigned int *bucket) 977{ 978 struct nf_conntrack_tuple_hash *h; 979 struct nf_conn *ct; 980 981 write_lock_bh(&nf_conntrack_lock); 982 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 983 list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) { 984 ct = nf_ct_tuplehash_to_ctrack(h); 985 if (iter(ct, data)) 986 goto found; 987 } 988 } 989 list_for_each_entry(h, &unconfirmed, list) { 990 ct = nf_ct_tuplehash_to_ctrack(h); 991 if (iter(ct, data)) 992 goto found; 993 } 994 write_unlock_bh(&nf_conntrack_lock); 995 return NULL; 996found: 997 atomic_inc(&ct->ct_general.use); 998 write_unlock_bh(&nf_conntrack_lock); 999 return ct; 1000} 1001 1002void 1003nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) 1004{ 1005 struct nf_conn *ct; 1006 unsigned int bucket = 0; 1007 1008 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { 1009 /* Time to push up daises... */ 1010 if (del_timer(&ct->timeout)) 1011 death_by_timeout((unsigned long)ct); 1012 /* ... else the timer will get him soon. */ 1013 1014 nf_ct_put(ct); 1015 } 1016} 1017 1018static int kill_all(struct nf_conn *i, void *data) 1019{ 1020 return 1; 1021} 1022 1023static void free_conntrack_hash(struct list_head *hash, int vmalloced, int size) 1024{ 1025 if (vmalloced) 1026 vfree(hash); 1027 else 1028 free_pages((unsigned long)hash, 1029 get_order(sizeof(struct list_head) * size)); 1030} 1031 1032void nf_conntrack_flush() 1033{ 1034 nf_ct_iterate_cleanup(kill_all, NULL); 1035} 1036 1037/* Mishearing the voices in his head, our hero wonders how he's 1038 supposed to kill the mall. */ 1039void nf_conntrack_cleanup(void) 1040{ 1041 int i; 1042 1043 ip_ct_attach = NULL; 1044 1045 /* This makes sure all current packets have passed through 1046 netfilter framework. Roll on, two-stage module 1047 delete... */ 1048 synchronize_net(); 1049 1050 nf_ct_event_cache_flush(); 1051 i_see_dead_people: 1052 nf_conntrack_flush(); 1053 if (atomic_read(&nf_conntrack_count) != 0) { 1054 schedule(); 1055 goto i_see_dead_people; 1056 } 1057 /* wait until all references to nf_conntrack_untracked are dropped */ 1058 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) 1059 schedule(); 1060 1061 for (i = 0; i < NF_CT_F_NUM; i++) { 1062 if (nf_ct_cache[i].use == 0) 1063 continue; 1064 1065 NF_CT_ASSERT(nf_ct_cache[i].use == 1); 1066 nf_ct_cache[i].use = 1; 1067 nf_conntrack_unregister_cache(i); 1068 } 1069 kmem_cache_destroy(nf_conntrack_expect_cachep); 1070 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1071 nf_conntrack_htable_size); 1072 1073 /* free l3proto protocol tables */ 1074 for (i = 0; i < PF_MAX; i++) 1075 if (nf_ct_protos[i]) { 1076 kfree(nf_ct_protos[i]); 1077 nf_ct_protos[i] = NULL; 1078 } 1079} 1080 1081static struct list_head *alloc_hashtable(int size, int *vmalloced) 1082{ 1083 struct list_head *hash; 1084 unsigned int i; 1085 1086 *vmalloced = 0; 1087 hash = (void*)__get_free_pages(GFP_KERNEL, 1088 get_order(sizeof(struct list_head) 1089 * size)); 1090 if (!hash) { 1091 *vmalloced = 1; 1092 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1093 hash = vmalloc(sizeof(struct list_head) * size); 1094 } 1095 1096 if (hash) 1097 for (i = 0; i < size; i++) 1098 INIT_LIST_HEAD(&hash[i]); 1099 1100 return hash; 1101} 1102 1103int set_hashsize(const char *val, struct kernel_param *kp) 1104{ 1105 int i, bucket, hashsize, vmalloced; 1106 int old_vmalloced, old_size; 1107 int rnd; 1108 struct list_head *hash, *old_hash; 1109 struct nf_conntrack_tuple_hash *h; 1110 1111 /* On boot, we can set this without any fancy locking. */ 1112 if (!nf_conntrack_htable_size) 1113 return param_set_uint(val, kp); 1114 1115 hashsize = simple_strtol(val, NULL, 0); 1116 if (!hashsize) 1117 return -EINVAL; 1118 1119 hash = alloc_hashtable(hashsize, &vmalloced); 1120 if (!hash) 1121 return -ENOMEM; 1122 1123 /* We have to rehahs for the new table anyway, so we also can 1124 * use a newrandom seed */ 1125 get_random_bytes(&rnd, 4); 1126 1127 write_lock_bh(&nf_conntrack_lock); 1128 for (i = 0; i < nf_conntrack_htable_size; i++) { 1129 while (!list_empty(&nf_conntrack_hash[i])) { 1130 h = list_entry(nf_conntrack_hash[i].next, 1131 struct nf_conntrack_tuple_hash, list); 1132 list_del(&h->list); 1133 bucket = __hash_conntrack(&h->tuple, hashsize, rnd); 1134 list_add_tail(&h->list, &hash[bucket]); 1135 } 1136 } 1137 old_size = nf_conntrack_htable_size; 1138 old_vmalloced = nf_conntrack_vmalloc; 1139 old_hash = nf_conntrack_hash; 1140 1141 nf_conntrack_htable_size = hashsize; 1142 nf_conntrack_vmalloc = vmalloced; 1143 nf_conntrack_hash = hash; 1144 nf_conntrack_hash_rnd = rnd; 1145 write_unlock_bh(&nf_conntrack_lock); 1146 1147 free_conntrack_hash(old_hash, old_vmalloced, old_size); 1148 return 0; 1149} 1150 1151module_param_call(hashsize, set_hashsize, param_get_uint, 1152 &nf_conntrack_htable_size, 0600); 1153 1154int __init nf_conntrack_init(void) 1155{ 1156 unsigned int i; 1157 int ret; 1158 1159 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1160 * machine has 256 buckets. >= 1GB machines have 8192 buckets. */ 1161 if (!nf_conntrack_htable_size) { 1162 nf_conntrack_htable_size 1163 = (((num_physpages << PAGE_SHIFT) / 16384) 1164 / sizeof(struct list_head)); 1165 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1166 nf_conntrack_htable_size = 8192; 1167 if (nf_conntrack_htable_size < 16) 1168 nf_conntrack_htable_size = 16; 1169 } 1170 nf_conntrack_max = 8 * nf_conntrack_htable_size; 1171 1172 printk("nf_conntrack version %s (%u buckets, %d max)\n", 1173 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1174 nf_conntrack_max); 1175 1176 nf_conntrack_hash = alloc_hashtable(nf_conntrack_htable_size, 1177 &nf_conntrack_vmalloc); 1178 if (!nf_conntrack_hash) { 1179 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1180 goto err_out; 1181 } 1182 1183 ret = nf_conntrack_register_cache(NF_CT_F_BASIC, "nf_conntrack:basic", 1184 sizeof(struct nf_conn)); 1185 if (ret < 0) { 1186 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1187 goto err_free_hash; 1188 } 1189 1190 nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect", 1191 sizeof(struct nf_conntrack_expect), 1192 0, 0, NULL, NULL); 1193 if (!nf_conntrack_expect_cachep) { 1194 printk(KERN_ERR "Unable to create nf_expect slab cache\n"); 1195 goto err_free_conntrack_slab; 1196 } 1197 1198 /* Don't NEED lock here, but good form anyway. */ 1199 write_lock_bh(&nf_conntrack_lock); 1200 for (i = 0; i < AF_MAX; i++) 1201 nf_ct_l3protos[i] = &nf_conntrack_l3proto_generic; 1202 write_unlock_bh(&nf_conntrack_lock); 1203 1204 /* For use by REJECT target */ 1205 ip_ct_attach = __nf_conntrack_attach; 1206 1207 /* Set up fake conntrack: 1208 - to never be deleted, not in any hashes */ 1209 atomic_set(&nf_conntrack_untracked.ct_general.use, 1); 1210 /* - and look it like as a confirmed connection */ 1211 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); 1212 1213 return ret; 1214 1215err_free_conntrack_slab: 1216 nf_conntrack_unregister_cache(NF_CT_F_BASIC); 1217err_free_hash: 1218 free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc, 1219 nf_conntrack_htable_size); 1220err_out: 1221 return -ENOMEM; 1222} 1223