nf_conntrack_core.c revision 4e29e9ec7e0707d3925f5dcc29af0d3f04e49833
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 14#include <linux/types.h> 15#include <linux/netfilter.h> 16#include <linux/module.h> 17#include <linux/skbuff.h> 18#include <linux/proc_fs.h> 19#include <linux/vmalloc.h> 20#include <linux/stddef.h> 21#include <linux/slab.h> 22#include <linux/random.h> 23#include <linux/jhash.h> 24#include <linux/err.h> 25#include <linux/percpu.h> 26#include <linux/moduleparam.h> 27#include <linux/notifier.h> 28#include <linux/kernel.h> 29#include <linux/netdevice.h> 30#include <linux/socket.h> 31#include <linux/mm.h> 32 33#include <net/netfilter/nf_conntrack.h> 34#include <net/netfilter/nf_conntrack_l3proto.h> 35#include <net/netfilter/nf_conntrack_l4proto.h> 36#include <net/netfilter/nf_conntrack_expect.h> 37#include <net/netfilter/nf_conntrack_helper.h> 38#include <net/netfilter/nf_conntrack_core.h> 39#include <net/netfilter/nf_conntrack_extend.h> 40 41#define NF_CONNTRACK_VERSION "0.5.0" 42 43DEFINE_SPINLOCK(nf_conntrack_lock); 44EXPORT_SYMBOL_GPL(nf_conntrack_lock); 45 46/* nf_conntrack_standalone needs this */ 47atomic_t nf_conntrack_count = ATOMIC_INIT(0); 48EXPORT_SYMBOL_GPL(nf_conntrack_count); 49 50unsigned int nf_conntrack_htable_size __read_mostly; 51EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); 52 53int nf_conntrack_max __read_mostly; 54EXPORT_SYMBOL_GPL(nf_conntrack_max); 55 56struct hlist_head *nf_conntrack_hash __read_mostly; 57EXPORT_SYMBOL_GPL(nf_conntrack_hash); 58 59struct nf_conn nf_conntrack_untracked __read_mostly; 60EXPORT_SYMBOL_GPL(nf_conntrack_untracked); 61 62unsigned int nf_ct_log_invalid __read_mostly; 63HLIST_HEAD(unconfirmed); 64static int nf_conntrack_vmalloc __read_mostly; 65static struct kmem_cache *nf_conntrack_cachep __read_mostly; 66 67DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); 68EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); 69 70static int nf_conntrack_hash_rnd_initted; 71static unsigned int nf_conntrack_hash_rnd; 72 73static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 74 unsigned int size, unsigned int rnd) 75{ 76 unsigned int n; 77 u_int32_t h; 78 79 /* The direction must be ignored, so we hash everything up to the 80 * destination ports (which is a multiple of 4) and treat the last 81 * three bytes manually. 82 */ 83 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); 84 h = jhash2((u32 *)tuple, n, 85 rnd ^ (((__force __u16)tuple->dst.u.all << 16) | 86 tuple->dst.protonum)); 87 88 return ((u64)h * size) >> 32; 89} 90 91static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) 92{ 93 return __hash_conntrack(tuple, nf_conntrack_htable_size, 94 nf_conntrack_hash_rnd); 95} 96 97int 98nf_ct_get_tuple(const struct sk_buff *skb, 99 unsigned int nhoff, 100 unsigned int dataoff, 101 u_int16_t l3num, 102 u_int8_t protonum, 103 struct nf_conntrack_tuple *tuple, 104 const struct nf_conntrack_l3proto *l3proto, 105 const struct nf_conntrack_l4proto *l4proto) 106{ 107 NF_CT_TUPLE_U_BLANK(tuple); 108 109 tuple->src.l3num = l3num; 110 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 111 return 0; 112 113 tuple->dst.protonum = protonum; 114 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 115 116 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 117} 118EXPORT_SYMBOL_GPL(nf_ct_get_tuple); 119 120int nf_ct_get_tuplepr(const struct sk_buff *skb, 121 unsigned int nhoff, 122 u_int16_t l3num, 123 struct nf_conntrack_tuple *tuple) 124{ 125 struct nf_conntrack_l3proto *l3proto; 126 struct nf_conntrack_l4proto *l4proto; 127 unsigned int protoff; 128 u_int8_t protonum; 129 int ret; 130 131 rcu_read_lock(); 132 133 l3proto = __nf_ct_l3proto_find(l3num); 134 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); 135 if (ret != NF_ACCEPT) { 136 rcu_read_unlock(); 137 return 0; 138 } 139 140 l4proto = __nf_ct_l4proto_find(l3num, protonum); 141 142 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, 143 l3proto, l4proto); 144 145 rcu_read_unlock(); 146 return ret; 147} 148EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); 149 150int 151nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 152 const struct nf_conntrack_tuple *orig, 153 const struct nf_conntrack_l3proto *l3proto, 154 const struct nf_conntrack_l4proto *l4proto) 155{ 156 NF_CT_TUPLE_U_BLANK(inverse); 157 158 inverse->src.l3num = orig->src.l3num; 159 if (l3proto->invert_tuple(inverse, orig) == 0) 160 return 0; 161 162 inverse->dst.dir = !orig->dst.dir; 163 164 inverse->dst.protonum = orig->dst.protonum; 165 return l4proto->invert_tuple(inverse, orig); 166} 167EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); 168 169static void 170clean_from_lists(struct nf_conn *ct) 171{ 172 pr_debug("clean_from_lists(%p)\n", ct); 173 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 174 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); 175 176 /* Destroy all pending expectations */ 177 nf_ct_remove_expectations(ct); 178} 179 180static void 181destroy_conntrack(struct nf_conntrack *nfct) 182{ 183 struct nf_conn *ct = (struct nf_conn *)nfct; 184 struct nf_conntrack_l4proto *l4proto; 185 186 pr_debug("destroy_conntrack(%p)\n", ct); 187 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 188 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 189 190 nf_conntrack_event(IPCT_DESTROY, ct); 191 set_bit(IPS_DYING_BIT, &ct->status); 192 193 /* To make sure we don't get any weird locking issues here: 194 * destroy_conntrack() MUST NOT be called with a write lock 195 * to nf_conntrack_lock!!! -HW */ 196 rcu_read_lock(); 197 l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num, 198 ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum); 199 if (l4proto && l4proto->destroy) 200 l4proto->destroy(ct); 201 202 nf_ct_ext_destroy(ct); 203 204 rcu_read_unlock(); 205 206 spin_lock_bh(&nf_conntrack_lock); 207 /* Expectations will have been removed in clean_from_lists, 208 * except TFTP can create an expectation on the first packet, 209 * before connection is in the list, so we need to clean here, 210 * too. */ 211 nf_ct_remove_expectations(ct); 212 213 /* We overload first tuple to link into unconfirmed list. */ 214 if (!nf_ct_is_confirmed(ct)) { 215 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); 216 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 217 } 218 219 NF_CT_STAT_INC(delete); 220 spin_unlock_bh(&nf_conntrack_lock); 221 222 if (ct->master) 223 nf_ct_put(ct->master); 224 225 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); 226 nf_conntrack_free(ct); 227} 228 229static void death_by_timeout(unsigned long ul_conntrack) 230{ 231 struct nf_conn *ct = (void *)ul_conntrack; 232 struct nf_conn_help *help = nfct_help(ct); 233 struct nf_conntrack_helper *helper; 234 235 if (help) { 236 rcu_read_lock(); 237 helper = rcu_dereference(help->helper); 238 if (helper && helper->destroy) 239 helper->destroy(ct); 240 rcu_read_unlock(); 241 } 242 243 spin_lock_bh(&nf_conntrack_lock); 244 /* Inside lock so preempt is disabled on module removal path. 245 * Otherwise we can get spurious warnings. */ 246 NF_CT_STAT_INC(delete_list); 247 clean_from_lists(ct); 248 spin_unlock_bh(&nf_conntrack_lock); 249 nf_ct_put(ct); 250} 251 252struct nf_conntrack_tuple_hash * 253__nf_conntrack_find(const struct nf_conntrack_tuple *tuple) 254{ 255 struct nf_conntrack_tuple_hash *h; 256 struct hlist_node *n; 257 unsigned int hash = hash_conntrack(tuple); 258 259 /* Disable BHs the entire time since we normally need to disable them 260 * at least once for the stats anyway. 261 */ 262 local_bh_disable(); 263 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { 264 if (nf_ct_tuple_equal(tuple, &h->tuple)) { 265 NF_CT_STAT_INC(found); 266 local_bh_enable(); 267 return h; 268 } 269 NF_CT_STAT_INC(searched); 270 } 271 local_bh_enable(); 272 273 return NULL; 274} 275EXPORT_SYMBOL_GPL(__nf_conntrack_find); 276 277/* Find a connection corresponding to a tuple. */ 278struct nf_conntrack_tuple_hash * 279nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple) 280{ 281 struct nf_conntrack_tuple_hash *h; 282 struct nf_conn *ct; 283 284 rcu_read_lock(); 285 h = __nf_conntrack_find(tuple); 286 if (h) { 287 ct = nf_ct_tuplehash_to_ctrack(h); 288 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) 289 h = NULL; 290 } 291 rcu_read_unlock(); 292 293 return h; 294} 295EXPORT_SYMBOL_GPL(nf_conntrack_find_get); 296 297static void __nf_conntrack_hash_insert(struct nf_conn *ct, 298 unsigned int hash, 299 unsigned int repl_hash) 300{ 301 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, 302 &nf_conntrack_hash[hash]); 303 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, 304 &nf_conntrack_hash[repl_hash]); 305} 306 307void nf_conntrack_hash_insert(struct nf_conn *ct) 308{ 309 unsigned int hash, repl_hash; 310 311 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 312 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 313 314 spin_lock_bh(&nf_conntrack_lock); 315 __nf_conntrack_hash_insert(ct, hash, repl_hash); 316 spin_unlock_bh(&nf_conntrack_lock); 317} 318EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); 319 320/* Confirm a connection given skb; places it in hash table */ 321int 322__nf_conntrack_confirm(struct sk_buff *skb) 323{ 324 unsigned int hash, repl_hash; 325 struct nf_conntrack_tuple_hash *h; 326 struct nf_conn *ct; 327 struct nf_conn_help *help; 328 struct hlist_node *n; 329 enum ip_conntrack_info ctinfo; 330 331 ct = nf_ct_get(skb, &ctinfo); 332 333 /* ipt_REJECT uses nf_conntrack_attach to attach related 334 ICMP/TCP RST packets in other direction. Actual packet 335 which created connection will be IP_CT_NEW or for an 336 expected connection, IP_CT_RELATED. */ 337 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 338 return NF_ACCEPT; 339 340 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 341 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 342 343 /* We're not in hash table, and we refuse to set up related 344 connections for unconfirmed conns. But packet copies and 345 REJECT will give spurious warnings here. */ 346 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 347 348 /* No external references means noone else could have 349 confirmed us. */ 350 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 351 pr_debug("Confirming conntrack %p\n", ct); 352 353 spin_lock_bh(&nf_conntrack_lock); 354 355 /* See if there's one in the list already, including reverse: 356 NAT could have grabbed it without realizing, since we're 357 not in the hash. If there is, we lost race. */ 358 hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) 359 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 360 &h->tuple)) 361 goto out; 362 hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode) 363 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 364 &h->tuple)) 365 goto out; 366 367 /* Remove from unconfirmed list */ 368 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 369 370 __nf_conntrack_hash_insert(ct, hash, repl_hash); 371 /* Timer relative to confirmation time, not original 372 setting time, otherwise we'd get timer wrap in 373 weird delay cases. */ 374 ct->timeout.expires += jiffies; 375 add_timer(&ct->timeout); 376 atomic_inc(&ct->ct_general.use); 377 set_bit(IPS_CONFIRMED_BIT, &ct->status); 378 NF_CT_STAT_INC(insert); 379 spin_unlock_bh(&nf_conntrack_lock); 380 help = nfct_help(ct); 381 if (help && help->helper) 382 nf_conntrack_event_cache(IPCT_HELPER, skb); 383#ifdef CONFIG_NF_NAT_NEEDED 384 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || 385 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) 386 nf_conntrack_event_cache(IPCT_NATINFO, skb); 387#endif 388 nf_conntrack_event_cache(master_ct(ct) ? 389 IPCT_RELATED : IPCT_NEW, skb); 390 return NF_ACCEPT; 391 392out: 393 NF_CT_STAT_INC(insert_failed); 394 spin_unlock_bh(&nf_conntrack_lock); 395 return NF_DROP; 396} 397EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); 398 399/* Returns true if a connection correspondings to the tuple (required 400 for NAT). */ 401int 402nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 403 const struct nf_conn *ignored_conntrack) 404{ 405 struct nf_conntrack_tuple_hash *h; 406 struct hlist_node *n; 407 unsigned int hash = hash_conntrack(tuple); 408 409 /* Disable BHs the entire time since we need to disable them at 410 * least once for the stats anyway. 411 */ 412 rcu_read_lock_bh(); 413 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) { 414 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && 415 nf_ct_tuple_equal(tuple, &h->tuple)) { 416 NF_CT_STAT_INC(found); 417 rcu_read_unlock_bh(); 418 return 1; 419 } 420 NF_CT_STAT_INC(searched); 421 } 422 rcu_read_unlock_bh(); 423 424 return 0; 425} 426EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); 427 428#define NF_CT_EVICTION_RANGE 8 429 430/* There's a small race here where we may free a just-assured 431 connection. Too bad: we're in trouble anyway. */ 432static noinline int early_drop(unsigned int hash) 433{ 434 /* Use oldest entry, which is roughly LRU */ 435 struct nf_conntrack_tuple_hash *h; 436 struct nf_conn *ct = NULL, *tmp; 437 struct hlist_node *n; 438 unsigned int i, cnt = 0; 439 int dropped = 0; 440 441 rcu_read_lock(); 442 for (i = 0; i < nf_conntrack_htable_size; i++) { 443 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], 444 hnode) { 445 tmp = nf_ct_tuplehash_to_ctrack(h); 446 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) 447 ct = tmp; 448 cnt++; 449 } 450 451 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) 452 ct = NULL; 453 if (ct || cnt >= NF_CT_EVICTION_RANGE) 454 break; 455 hash = (hash + 1) % nf_conntrack_htable_size; 456 } 457 rcu_read_unlock(); 458 459 if (!ct) 460 return dropped; 461 462 if (del_timer(&ct->timeout)) { 463 death_by_timeout((unsigned long)ct); 464 dropped = 1; 465 NF_CT_STAT_INC_ATOMIC(early_drop); 466 } 467 nf_ct_put(ct); 468 return dropped; 469} 470 471struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig, 472 const struct nf_conntrack_tuple *repl) 473{ 474 struct nf_conn *ct = NULL; 475 476 if (unlikely(!nf_conntrack_hash_rnd_initted)) { 477 get_random_bytes(&nf_conntrack_hash_rnd, 4); 478 nf_conntrack_hash_rnd_initted = 1; 479 } 480 481 /* We don't want any race condition at early drop stage */ 482 atomic_inc(&nf_conntrack_count); 483 484 if (nf_conntrack_max && 485 unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) { 486 unsigned int hash = hash_conntrack(orig); 487 if (!early_drop(hash)) { 488 atomic_dec(&nf_conntrack_count); 489 if (net_ratelimit()) 490 printk(KERN_WARNING 491 "nf_conntrack: table full, dropping" 492 " packet.\n"); 493 return ERR_PTR(-ENOMEM); 494 } 495 } 496 497 ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC); 498 if (ct == NULL) { 499 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); 500 atomic_dec(&nf_conntrack_count); 501 return ERR_PTR(-ENOMEM); 502 } 503 504 atomic_set(&ct->ct_general.use, 1); 505 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 506 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 507 /* Don't set timer yet: wait for confirmation */ 508 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); 509 INIT_RCU_HEAD(&ct->rcu); 510 511 return ct; 512} 513EXPORT_SYMBOL_GPL(nf_conntrack_alloc); 514 515static void nf_conntrack_free_rcu(struct rcu_head *head) 516{ 517 struct nf_conn *ct = container_of(head, struct nf_conn, rcu); 518 519 nf_ct_ext_free(ct); 520 kmem_cache_free(nf_conntrack_cachep, ct); 521 atomic_dec(&nf_conntrack_count); 522} 523 524void nf_conntrack_free(struct nf_conn *ct) 525{ 526 call_rcu(&ct->rcu, nf_conntrack_free_rcu); 527} 528EXPORT_SYMBOL_GPL(nf_conntrack_free); 529 530/* Allocate a new conntrack: we return -ENOMEM if classification 531 failed due to stress. Otherwise it really is unclassifiable. */ 532static struct nf_conntrack_tuple_hash * 533init_conntrack(const struct nf_conntrack_tuple *tuple, 534 struct nf_conntrack_l3proto *l3proto, 535 struct nf_conntrack_l4proto *l4proto, 536 struct sk_buff *skb, 537 unsigned int dataoff) 538{ 539 struct nf_conn *ct; 540 struct nf_conn_help *help; 541 struct nf_conntrack_tuple repl_tuple; 542 struct nf_conntrack_expect *exp; 543 544 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 545 pr_debug("Can't invert tuple.\n"); 546 return NULL; 547 } 548 549 ct = nf_conntrack_alloc(tuple, &repl_tuple); 550 if (ct == NULL || IS_ERR(ct)) { 551 pr_debug("Can't allocate conntrack.\n"); 552 return (struct nf_conntrack_tuple_hash *)ct; 553 } 554 555 if (!l4proto->new(ct, skb, dataoff)) { 556 nf_conntrack_free(ct); 557 pr_debug("init conntrack: can't track with proto module\n"); 558 return NULL; 559 } 560 561 spin_lock_bh(&nf_conntrack_lock); 562 exp = nf_ct_find_expectation(tuple); 563 if (exp) { 564 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", 565 ct, exp); 566 /* Welcome, Mr. Bond. We've been expecting you... */ 567 __set_bit(IPS_EXPECTED_BIT, &ct->status); 568 ct->master = exp->master; 569 if (exp->helper) { 570 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 571 if (help) 572 rcu_assign_pointer(help->helper, exp->helper); 573 } 574 575#ifdef CONFIG_NF_CONNTRACK_MARK 576 ct->mark = exp->master->mark; 577#endif 578#ifdef CONFIG_NF_CONNTRACK_SECMARK 579 ct->secmark = exp->master->secmark; 580#endif 581 nf_conntrack_get(&ct->master->ct_general); 582 NF_CT_STAT_INC(expect_new); 583 } else { 584 struct nf_conntrack_helper *helper; 585 586 helper = __nf_ct_helper_find(&repl_tuple); 587 if (helper) { 588 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 589 if (help) 590 rcu_assign_pointer(help->helper, helper); 591 } 592 NF_CT_STAT_INC(new); 593 } 594 595 /* Overload tuple linked list to put us in unconfirmed list. */ 596 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed); 597 598 spin_unlock_bh(&nf_conntrack_lock); 599 600 if (exp) { 601 if (exp->expectfn) 602 exp->expectfn(ct, exp); 603 nf_ct_expect_put(exp); 604 } 605 606 return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; 607} 608 609/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 610static inline struct nf_conn * 611resolve_normal_ct(struct sk_buff *skb, 612 unsigned int dataoff, 613 u_int16_t l3num, 614 u_int8_t protonum, 615 struct nf_conntrack_l3proto *l3proto, 616 struct nf_conntrack_l4proto *l4proto, 617 int *set_reply, 618 enum ip_conntrack_info *ctinfo) 619{ 620 struct nf_conntrack_tuple tuple; 621 struct nf_conntrack_tuple_hash *h; 622 struct nf_conn *ct; 623 624 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), 625 dataoff, l3num, protonum, &tuple, l3proto, 626 l4proto)) { 627 pr_debug("resolve_normal_ct: Can't get tuple\n"); 628 return NULL; 629 } 630 631 /* look for tuple match */ 632 h = nf_conntrack_find_get(&tuple); 633 if (!h) { 634 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff); 635 if (!h) 636 return NULL; 637 if (IS_ERR(h)) 638 return (void *)h; 639 } 640 ct = nf_ct_tuplehash_to_ctrack(h); 641 642 /* It exists; we have (non-exclusive) reference. */ 643 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 644 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 645 /* Please set reply bit if this packet OK */ 646 *set_reply = 1; 647 } else { 648 /* Once we've had two way comms, always ESTABLISHED. */ 649 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 650 pr_debug("nf_conntrack_in: normal packet for %p\n", ct); 651 *ctinfo = IP_CT_ESTABLISHED; 652 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 653 pr_debug("nf_conntrack_in: related packet for %p\n", 654 ct); 655 *ctinfo = IP_CT_RELATED; 656 } else { 657 pr_debug("nf_conntrack_in: new packet for %p\n", ct); 658 *ctinfo = IP_CT_NEW; 659 } 660 *set_reply = 0; 661 } 662 skb->nfct = &ct->ct_general; 663 skb->nfctinfo = *ctinfo; 664 return ct; 665} 666 667unsigned int 668nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb) 669{ 670 struct nf_conn *ct; 671 enum ip_conntrack_info ctinfo; 672 struct nf_conntrack_l3proto *l3proto; 673 struct nf_conntrack_l4proto *l4proto; 674 unsigned int dataoff; 675 u_int8_t protonum; 676 int set_reply = 0; 677 int ret; 678 679 /* Previously seen (loopback or untracked)? Ignore. */ 680 if (skb->nfct) { 681 NF_CT_STAT_INC_ATOMIC(ignore); 682 return NF_ACCEPT; 683 } 684 685 /* rcu_read_lock()ed by nf_hook_slow */ 686 l3proto = __nf_ct_l3proto_find((u_int16_t)pf); 687 ret = l3proto->get_l4proto(skb, skb_network_offset(skb), 688 &dataoff, &protonum); 689 if (ret <= 0) { 690 pr_debug("not prepared to track yet or error occured\n"); 691 NF_CT_STAT_INC_ATOMIC(error); 692 NF_CT_STAT_INC_ATOMIC(invalid); 693 return -ret; 694 } 695 696 l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum); 697 698 /* It may be an special packet, error, unclean... 699 * inverse of the return code tells to the netfilter 700 * core what to do with the packet. */ 701 if (l4proto->error != NULL && 702 (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) { 703 NF_CT_STAT_INC_ATOMIC(error); 704 NF_CT_STAT_INC_ATOMIC(invalid); 705 return -ret; 706 } 707 708 ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto, 709 &set_reply, &ctinfo); 710 if (!ct) { 711 /* Not valid part of a connection */ 712 NF_CT_STAT_INC_ATOMIC(invalid); 713 return NF_ACCEPT; 714 } 715 716 if (IS_ERR(ct)) { 717 /* Too stressed to deal. */ 718 NF_CT_STAT_INC_ATOMIC(drop); 719 return NF_DROP; 720 } 721 722 NF_CT_ASSERT(skb->nfct); 723 724 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); 725 if (ret < 0) { 726 /* Invalid: inverse of the return code tells 727 * the netfilter core what to do */ 728 pr_debug("nf_conntrack_in: Can't track with proto module\n"); 729 nf_conntrack_put(skb->nfct); 730 skb->nfct = NULL; 731 NF_CT_STAT_INC_ATOMIC(invalid); 732 return -ret; 733 } 734 735 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 736 nf_conntrack_event_cache(IPCT_STATUS, skb); 737 738 return ret; 739} 740EXPORT_SYMBOL_GPL(nf_conntrack_in); 741 742int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 743 const struct nf_conntrack_tuple *orig) 744{ 745 int ret; 746 747 rcu_read_lock(); 748 ret = nf_ct_invert_tuple(inverse, orig, 749 __nf_ct_l3proto_find(orig->src.l3num), 750 __nf_ct_l4proto_find(orig->src.l3num, 751 orig->dst.protonum)); 752 rcu_read_unlock(); 753 return ret; 754} 755EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); 756 757/* Alter reply tuple (maybe alter helper). This is for NAT, and is 758 implicitly racy: see __nf_conntrack_confirm */ 759void nf_conntrack_alter_reply(struct nf_conn *ct, 760 const struct nf_conntrack_tuple *newreply) 761{ 762 struct nf_conn_help *help = nfct_help(ct); 763 struct nf_conntrack_helper *helper; 764 765 /* Should be unconfirmed, so not in hash table yet */ 766 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 767 768 pr_debug("Altering reply tuple of %p to ", ct); 769 NF_CT_DUMP_TUPLE(newreply); 770 771 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 772 if (ct->master || (help && help->expecting != 0)) 773 return; 774 775 rcu_read_lock(); 776 helper = __nf_ct_helper_find(newreply); 777 if (helper == NULL) { 778 if (help) 779 rcu_assign_pointer(help->helper, NULL); 780 goto out; 781 } 782 783 if (help == NULL) { 784 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 785 if (help == NULL) 786 goto out; 787 } else { 788 memset(&help->help, 0, sizeof(help->help)); 789 } 790 791 rcu_assign_pointer(help->helper, helper); 792out: 793 rcu_read_unlock(); 794} 795EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); 796 797/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 798void __nf_ct_refresh_acct(struct nf_conn *ct, 799 enum ip_conntrack_info ctinfo, 800 const struct sk_buff *skb, 801 unsigned long extra_jiffies, 802 int do_acct) 803{ 804 int event = 0; 805 806 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 807 NF_CT_ASSERT(skb); 808 809 spin_lock_bh(&nf_conntrack_lock); 810 811 /* Only update if this is not a fixed timeout */ 812 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) 813 goto acct; 814 815 /* If not in hash table, timer will not be active yet */ 816 if (!nf_ct_is_confirmed(ct)) { 817 ct->timeout.expires = extra_jiffies; 818 event = IPCT_REFRESH; 819 } else { 820 unsigned long newtime = jiffies + extra_jiffies; 821 822 /* Only update the timeout if the new timeout is at least 823 HZ jiffies from the old timeout. Need del_timer for race 824 avoidance (may already be dying). */ 825 if (newtime - ct->timeout.expires >= HZ 826 && del_timer(&ct->timeout)) { 827 ct->timeout.expires = newtime; 828 add_timer(&ct->timeout); 829 event = IPCT_REFRESH; 830 } 831 } 832 833acct: 834#ifdef CONFIG_NF_CT_ACCT 835 if (do_acct) { 836 ct->counters[CTINFO2DIR(ctinfo)].packets++; 837 ct->counters[CTINFO2DIR(ctinfo)].bytes += 838 skb->len - skb_network_offset(skb); 839 840 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000) 841 || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000)) 842 event |= IPCT_COUNTER_FILLING; 843 } 844#endif 845 846 spin_unlock_bh(&nf_conntrack_lock); 847 848 /* must be unlocked when calling event cache */ 849 if (event) 850 nf_conntrack_event_cache(event, skb); 851} 852EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); 853 854#if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) 855 856#include <linux/netfilter/nfnetlink.h> 857#include <linux/netfilter/nfnetlink_conntrack.h> 858#include <linux/mutex.h> 859 860/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 861 * in ip_conntrack_core, since we don't want the protocols to autoload 862 * or depend on ctnetlink */ 863int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, 864 const struct nf_conntrack_tuple *tuple) 865{ 866 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); 867 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); 868 return 0; 869 870nla_put_failure: 871 return -1; 872} 873EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); 874 875const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { 876 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, 877 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, 878}; 879EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); 880 881int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], 882 struct nf_conntrack_tuple *t) 883{ 884 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) 885 return -EINVAL; 886 887 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); 888 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); 889 890 return 0; 891} 892EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); 893#endif 894 895/* Used by ipt_REJECT and ip6t_REJECT. */ 896static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 897{ 898 struct nf_conn *ct; 899 enum ip_conntrack_info ctinfo; 900 901 /* This ICMP is in reverse direction to the packet which caused it */ 902 ct = nf_ct_get(skb, &ctinfo); 903 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 904 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 905 else 906 ctinfo = IP_CT_RELATED; 907 908 /* Attach to new skbuff, and increment count */ 909 nskb->nfct = &ct->ct_general; 910 nskb->nfctinfo = ctinfo; 911 nf_conntrack_get(nskb->nfct); 912} 913 914/* Bring out ya dead! */ 915static struct nf_conn * 916get_next_corpse(int (*iter)(struct nf_conn *i, void *data), 917 void *data, unsigned int *bucket) 918{ 919 struct nf_conntrack_tuple_hash *h; 920 struct nf_conn *ct; 921 struct hlist_node *n; 922 923 spin_lock_bh(&nf_conntrack_lock); 924 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 925 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) { 926 ct = nf_ct_tuplehash_to_ctrack(h); 927 if (iter(ct, data)) 928 goto found; 929 } 930 } 931 hlist_for_each_entry(h, n, &unconfirmed, hnode) { 932 ct = nf_ct_tuplehash_to_ctrack(h); 933 if (iter(ct, data)) 934 set_bit(IPS_DYING_BIT, &ct->status); 935 } 936 spin_unlock_bh(&nf_conntrack_lock); 937 return NULL; 938found: 939 atomic_inc(&ct->ct_general.use); 940 spin_unlock_bh(&nf_conntrack_lock); 941 return ct; 942} 943 944void 945nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data) 946{ 947 struct nf_conn *ct; 948 unsigned int bucket = 0; 949 950 while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) { 951 /* Time to push up daises... */ 952 if (del_timer(&ct->timeout)) 953 death_by_timeout((unsigned long)ct); 954 /* ... else the timer will get him soon. */ 955 956 nf_ct_put(ct); 957 } 958} 959EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); 960 961static int kill_all(struct nf_conn *i, void *data) 962{ 963 return 1; 964} 965 966void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) 967{ 968 if (vmalloced) 969 vfree(hash); 970 else 971 free_pages((unsigned long)hash, 972 get_order(sizeof(struct hlist_head) * size)); 973} 974EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); 975 976void nf_conntrack_flush(void) 977{ 978 nf_ct_iterate_cleanup(kill_all, NULL); 979} 980EXPORT_SYMBOL_GPL(nf_conntrack_flush); 981 982/* Mishearing the voices in his head, our hero wonders how he's 983 supposed to kill the mall. */ 984void nf_conntrack_cleanup(void) 985{ 986 rcu_assign_pointer(ip_ct_attach, NULL); 987 988 /* This makes sure all current packets have passed through 989 netfilter framework. Roll on, two-stage module 990 delete... */ 991 synchronize_net(); 992 993 nf_ct_event_cache_flush(); 994 i_see_dead_people: 995 nf_conntrack_flush(); 996 if (atomic_read(&nf_conntrack_count) != 0) { 997 schedule(); 998 goto i_see_dead_people; 999 } 1000 /* wait until all references to nf_conntrack_untracked are dropped */ 1001 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) 1002 schedule(); 1003 1004 rcu_assign_pointer(nf_ct_destroy, NULL); 1005 1006 kmem_cache_destroy(nf_conntrack_cachep); 1007 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, 1008 nf_conntrack_htable_size); 1009 1010 nf_conntrack_proto_fini(); 1011 nf_conntrack_helper_fini(); 1012 nf_conntrack_expect_fini(); 1013} 1014 1015struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) 1016{ 1017 struct hlist_head *hash; 1018 unsigned int size, i; 1019 1020 *vmalloced = 0; 1021 1022 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); 1023 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, 1024 get_order(sizeof(struct hlist_head) 1025 * size)); 1026 if (!hash) { 1027 *vmalloced = 1; 1028 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1029 hash = vmalloc(sizeof(struct hlist_head) * size); 1030 } 1031 1032 if (hash) 1033 for (i = 0; i < size; i++) 1034 INIT_HLIST_HEAD(&hash[i]); 1035 1036 return hash; 1037} 1038EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); 1039 1040int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) 1041{ 1042 int i, bucket, vmalloced, old_vmalloced; 1043 unsigned int hashsize, old_size; 1044 int rnd; 1045 struct hlist_head *hash, *old_hash; 1046 struct nf_conntrack_tuple_hash *h; 1047 1048 /* On boot, we can set this without any fancy locking. */ 1049 if (!nf_conntrack_htable_size) 1050 return param_set_uint(val, kp); 1051 1052 hashsize = simple_strtoul(val, NULL, 0); 1053 if (!hashsize) 1054 return -EINVAL; 1055 1056 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); 1057 if (!hash) 1058 return -ENOMEM; 1059 1060 /* We have to rehahs for the new table anyway, so we also can 1061 * use a newrandom seed */ 1062 get_random_bytes(&rnd, 4); 1063 1064 /* Lookups in the old hash might happen in parallel, which means we 1065 * might get false negatives during connection lookup. New connections 1066 * created because of a false negative won't make it into the hash 1067 * though since that required taking the lock. 1068 */ 1069 spin_lock_bh(&nf_conntrack_lock); 1070 for (i = 0; i < nf_conntrack_htable_size; i++) { 1071 while (!hlist_empty(&nf_conntrack_hash[i])) { 1072 h = hlist_entry(nf_conntrack_hash[i].first, 1073 struct nf_conntrack_tuple_hash, hnode); 1074 hlist_del_rcu(&h->hnode); 1075 bucket = __hash_conntrack(&h->tuple, hashsize, rnd); 1076 hlist_add_head(&h->hnode, &hash[bucket]); 1077 } 1078 } 1079 old_size = nf_conntrack_htable_size; 1080 old_vmalloced = nf_conntrack_vmalloc; 1081 old_hash = nf_conntrack_hash; 1082 1083 nf_conntrack_htable_size = hashsize; 1084 nf_conntrack_vmalloc = vmalloced; 1085 nf_conntrack_hash = hash; 1086 nf_conntrack_hash_rnd = rnd; 1087 spin_unlock_bh(&nf_conntrack_lock); 1088 1089 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); 1090 return 0; 1091} 1092EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); 1093 1094module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, 1095 &nf_conntrack_htable_size, 0600); 1096 1097int __init nf_conntrack_init(void) 1098{ 1099 int max_factor = 8; 1100 int ret; 1101 1102 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1103 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ 1104 if (!nf_conntrack_htable_size) { 1105 nf_conntrack_htable_size 1106 = (((num_physpages << PAGE_SHIFT) / 16384) 1107 / sizeof(struct hlist_head)); 1108 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1109 nf_conntrack_htable_size = 16384; 1110 if (nf_conntrack_htable_size < 32) 1111 nf_conntrack_htable_size = 32; 1112 1113 /* Use a max. factor of four by default to get the same max as 1114 * with the old struct list_heads. When a table size is given 1115 * we use the old value of 8 to avoid reducing the max. 1116 * entries. */ 1117 max_factor = 4; 1118 } 1119 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1120 &nf_conntrack_vmalloc); 1121 if (!nf_conntrack_hash) { 1122 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1123 goto err_out; 1124 } 1125 1126 nf_conntrack_max = max_factor * nf_conntrack_htable_size; 1127 1128 printk("nf_conntrack version %s (%u buckets, %d max)\n", 1129 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1130 nf_conntrack_max); 1131 1132 nf_conntrack_cachep = kmem_cache_create("nf_conntrack", 1133 sizeof(struct nf_conn), 1134 0, 0, NULL); 1135 if (!nf_conntrack_cachep) { 1136 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1137 goto err_free_hash; 1138 } 1139 1140 ret = nf_conntrack_proto_init(); 1141 if (ret < 0) 1142 goto err_free_conntrack_slab; 1143 1144 ret = nf_conntrack_expect_init(); 1145 if (ret < 0) 1146 goto out_fini_proto; 1147 1148 ret = nf_conntrack_helper_init(); 1149 if (ret < 0) 1150 goto out_fini_expect; 1151 1152 /* For use by REJECT target */ 1153 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); 1154 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); 1155 1156 /* Set up fake conntrack: 1157 - to never be deleted, not in any hashes */ 1158 atomic_set(&nf_conntrack_untracked.ct_general.use, 1); 1159 /* - and look it like as a confirmed connection */ 1160 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); 1161 1162 return ret; 1163 1164out_fini_expect: 1165 nf_conntrack_expect_fini(); 1166out_fini_proto: 1167 nf_conntrack_proto_fini(); 1168err_free_conntrack_slab: 1169 kmem_cache_destroy(nf_conntrack_cachep); 1170err_free_hash: 1171 nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc, 1172 nf_conntrack_htable_size); 1173err_out: 1174 return -ENOMEM; 1175} 1176