nf_conntrack_core.c revision 74c51a1497033e6ff7b8096797daca233a4a30df
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#include <net/netfilter/nf_conntrack_acct.h> 41 42#define NF_CONNTRACK_VERSION "0.5.0" 43 44DEFINE_SPINLOCK(nf_conntrack_lock); 45EXPORT_SYMBOL_GPL(nf_conntrack_lock); 46 47unsigned int nf_conntrack_htable_size __read_mostly; 48EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); 49 50int nf_conntrack_max __read_mostly; 51EXPORT_SYMBOL_GPL(nf_conntrack_max); 52 53struct nf_conn nf_conntrack_untracked __read_mostly; 54EXPORT_SYMBOL_GPL(nf_conntrack_untracked); 55 56unsigned int nf_ct_log_invalid __read_mostly; 57static struct kmem_cache *nf_conntrack_cachep __read_mostly; 58 59DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat); 60EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat); 61 62static int nf_conntrack_hash_rnd_initted; 63static unsigned int nf_conntrack_hash_rnd; 64 65static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 66 unsigned int size, unsigned int rnd) 67{ 68 unsigned int n; 69 u_int32_t h; 70 71 /* The direction must be ignored, so we hash everything up to the 72 * destination ports (which is a multiple of 4) and treat the last 73 * three bytes manually. 74 */ 75 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); 76 h = jhash2((u32 *)tuple, n, 77 rnd ^ (((__force __u16)tuple->dst.u.all << 16) | 78 tuple->dst.protonum)); 79 80 return ((u64)h * size) >> 32; 81} 82 83static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple) 84{ 85 return __hash_conntrack(tuple, nf_conntrack_htable_size, 86 nf_conntrack_hash_rnd); 87} 88 89bool 90nf_ct_get_tuple(const struct sk_buff *skb, 91 unsigned int nhoff, 92 unsigned int dataoff, 93 u_int16_t l3num, 94 u_int8_t protonum, 95 struct nf_conntrack_tuple *tuple, 96 const struct nf_conntrack_l3proto *l3proto, 97 const struct nf_conntrack_l4proto *l4proto) 98{ 99 memset(tuple, 0, sizeof(*tuple)); 100 101 tuple->src.l3num = l3num; 102 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 103 return false; 104 105 tuple->dst.protonum = protonum; 106 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 107 108 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 109} 110EXPORT_SYMBOL_GPL(nf_ct_get_tuple); 111 112bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, 113 u_int16_t l3num, struct nf_conntrack_tuple *tuple) 114{ 115 struct nf_conntrack_l3proto *l3proto; 116 struct nf_conntrack_l4proto *l4proto; 117 unsigned int protoff; 118 u_int8_t protonum; 119 int ret; 120 121 rcu_read_lock(); 122 123 l3proto = __nf_ct_l3proto_find(l3num); 124 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); 125 if (ret != NF_ACCEPT) { 126 rcu_read_unlock(); 127 return false; 128 } 129 130 l4proto = __nf_ct_l4proto_find(l3num, protonum); 131 132 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, 133 l3proto, l4proto); 134 135 rcu_read_unlock(); 136 return ret; 137} 138EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); 139 140bool 141nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 142 const struct nf_conntrack_tuple *orig, 143 const struct nf_conntrack_l3proto *l3proto, 144 const struct nf_conntrack_l4proto *l4proto) 145{ 146 memset(inverse, 0, sizeof(*inverse)); 147 148 inverse->src.l3num = orig->src.l3num; 149 if (l3proto->invert_tuple(inverse, orig) == 0) 150 return false; 151 152 inverse->dst.dir = !orig->dst.dir; 153 154 inverse->dst.protonum = orig->dst.protonum; 155 return l4proto->invert_tuple(inverse, orig); 156} 157EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); 158 159static void 160clean_from_lists(struct nf_conn *ct) 161{ 162 pr_debug("clean_from_lists(%p)\n", ct); 163 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 164 hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode); 165 166 /* Destroy all pending expectations */ 167 nf_ct_remove_expectations(ct); 168} 169 170static void 171destroy_conntrack(struct nf_conntrack *nfct) 172{ 173 struct nf_conn *ct = (struct nf_conn *)nfct; 174 struct nf_conntrack_l4proto *l4proto; 175 176 pr_debug("destroy_conntrack(%p)\n", ct); 177 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 178 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 179 180 nf_conntrack_event(IPCT_DESTROY, ct); 181 set_bit(IPS_DYING_BIT, &ct->status); 182 183 /* To make sure we don't get any weird locking issues here: 184 * destroy_conntrack() MUST NOT be called with a write lock 185 * to nf_conntrack_lock!!! -HW */ 186 rcu_read_lock(); 187 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); 188 if (l4proto && l4proto->destroy) 189 l4proto->destroy(ct); 190 191 rcu_read_unlock(); 192 193 spin_lock_bh(&nf_conntrack_lock); 194 /* Expectations will have been removed in clean_from_lists, 195 * except TFTP can create an expectation on the first packet, 196 * before connection is in the list, so we need to clean here, 197 * too. */ 198 nf_ct_remove_expectations(ct); 199 200 /* We overload first tuple to link into unconfirmed list. */ 201 if (!nf_ct_is_confirmed(ct)) { 202 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode)); 203 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 204 } 205 206 NF_CT_STAT_INC(delete); 207 spin_unlock_bh(&nf_conntrack_lock); 208 209 if (ct->master) 210 nf_ct_put(ct->master); 211 212 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); 213 nf_conntrack_free(ct); 214} 215 216static void death_by_timeout(unsigned long ul_conntrack) 217{ 218 struct nf_conn *ct = (void *)ul_conntrack; 219 struct nf_conn_help *help = nfct_help(ct); 220 struct nf_conntrack_helper *helper; 221 222 if (help) { 223 rcu_read_lock(); 224 helper = rcu_dereference(help->helper); 225 if (helper && helper->destroy) 226 helper->destroy(ct); 227 rcu_read_unlock(); 228 } 229 230 spin_lock_bh(&nf_conntrack_lock); 231 /* Inside lock so preempt is disabled on module removal path. 232 * Otherwise we can get spurious warnings. */ 233 NF_CT_STAT_INC(delete_list); 234 clean_from_lists(ct); 235 spin_unlock_bh(&nf_conntrack_lock); 236 nf_ct_put(ct); 237} 238 239struct nf_conntrack_tuple_hash * 240__nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) 241{ 242 struct nf_conntrack_tuple_hash *h; 243 struct hlist_node *n; 244 unsigned int hash = hash_conntrack(tuple); 245 246 /* Disable BHs the entire time since we normally need to disable them 247 * at least once for the stats anyway. 248 */ 249 local_bh_disable(); 250 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { 251 if (nf_ct_tuple_equal(tuple, &h->tuple)) { 252 NF_CT_STAT_INC(found); 253 local_bh_enable(); 254 return h; 255 } 256 NF_CT_STAT_INC(searched); 257 } 258 local_bh_enable(); 259 260 return NULL; 261} 262EXPORT_SYMBOL_GPL(__nf_conntrack_find); 263 264/* Find a connection corresponding to a tuple. */ 265struct nf_conntrack_tuple_hash * 266nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) 267{ 268 struct nf_conntrack_tuple_hash *h; 269 struct nf_conn *ct; 270 271 rcu_read_lock(); 272 h = __nf_conntrack_find(net, tuple); 273 if (h) { 274 ct = nf_ct_tuplehash_to_ctrack(h); 275 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) 276 h = NULL; 277 } 278 rcu_read_unlock(); 279 280 return h; 281} 282EXPORT_SYMBOL_GPL(nf_conntrack_find_get); 283 284static void __nf_conntrack_hash_insert(struct nf_conn *ct, 285 unsigned int hash, 286 unsigned int repl_hash) 287{ 288 struct net *net = nf_ct_net(ct); 289 290 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, 291 &net->ct.hash[hash]); 292 hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode, 293 &net->ct.hash[repl_hash]); 294} 295 296void nf_conntrack_hash_insert(struct nf_conn *ct) 297{ 298 unsigned int hash, repl_hash; 299 300 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 301 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 302 303 spin_lock_bh(&nf_conntrack_lock); 304 __nf_conntrack_hash_insert(ct, hash, repl_hash); 305 spin_unlock_bh(&nf_conntrack_lock); 306} 307EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); 308 309/* Confirm a connection given skb; places it in hash table */ 310int 311__nf_conntrack_confirm(struct sk_buff *skb) 312{ 313 unsigned int hash, repl_hash; 314 struct nf_conntrack_tuple_hash *h; 315 struct nf_conn *ct; 316 struct nf_conn_help *help; 317 struct hlist_node *n; 318 enum ip_conntrack_info ctinfo; 319 struct net *net; 320 321 ct = nf_ct_get(skb, &ctinfo); 322 net = nf_ct_net(ct); 323 324 /* ipt_REJECT uses nf_conntrack_attach to attach related 325 ICMP/TCP RST packets in other direction. Actual packet 326 which created connection will be IP_CT_NEW or for an 327 expected connection, IP_CT_RELATED. */ 328 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 329 return NF_ACCEPT; 330 331 hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 332 repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple); 333 334 /* We're not in hash table, and we refuse to set up related 335 connections for unconfirmed conns. But packet copies and 336 REJECT will give spurious warnings here. */ 337 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 338 339 /* No external references means noone else could have 340 confirmed us. */ 341 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 342 pr_debug("Confirming conntrack %p\n", ct); 343 344 spin_lock_bh(&nf_conntrack_lock); 345 346 /* See if there's one in the list already, including reverse: 347 NAT could have grabbed it without realizing, since we're 348 not in the hash. If there is, we lost race. */ 349 hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode) 350 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 351 &h->tuple)) 352 goto out; 353 hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode) 354 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 355 &h->tuple)) 356 goto out; 357 358 /* Remove from unconfirmed list */ 359 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode); 360 361 __nf_conntrack_hash_insert(ct, hash, repl_hash); 362 /* Timer relative to confirmation time, not original 363 setting time, otherwise we'd get timer wrap in 364 weird delay cases. */ 365 ct->timeout.expires += jiffies; 366 add_timer(&ct->timeout); 367 atomic_inc(&ct->ct_general.use); 368 set_bit(IPS_CONFIRMED_BIT, &ct->status); 369 NF_CT_STAT_INC(insert); 370 spin_unlock_bh(&nf_conntrack_lock); 371 help = nfct_help(ct); 372 if (help && help->helper) 373 nf_conntrack_event_cache(IPCT_HELPER, skb); 374#ifdef CONFIG_NF_NAT_NEEDED 375 if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) || 376 test_bit(IPS_DST_NAT_DONE_BIT, &ct->status)) 377 nf_conntrack_event_cache(IPCT_NATINFO, skb); 378#endif 379 nf_conntrack_event_cache(master_ct(ct) ? 380 IPCT_RELATED : IPCT_NEW, skb); 381 return NF_ACCEPT; 382 383out: 384 NF_CT_STAT_INC(insert_failed); 385 spin_unlock_bh(&nf_conntrack_lock); 386 return NF_DROP; 387} 388EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); 389 390/* Returns true if a connection correspondings to the tuple (required 391 for NAT). */ 392int 393nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 394 const struct nf_conn *ignored_conntrack) 395{ 396 struct net *net = nf_ct_net(ignored_conntrack); 397 struct nf_conntrack_tuple_hash *h; 398 struct hlist_node *n; 399 unsigned int hash = hash_conntrack(tuple); 400 401 /* Disable BHs the entire time since we need to disable them at 402 * least once for the stats anyway. 403 */ 404 rcu_read_lock_bh(); 405 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) { 406 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack && 407 nf_ct_tuple_equal(tuple, &h->tuple)) { 408 NF_CT_STAT_INC(found); 409 rcu_read_unlock_bh(); 410 return 1; 411 } 412 NF_CT_STAT_INC(searched); 413 } 414 rcu_read_unlock_bh(); 415 416 return 0; 417} 418EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); 419 420#define NF_CT_EVICTION_RANGE 8 421 422/* There's a small race here where we may free a just-assured 423 connection. Too bad: we're in trouble anyway. */ 424static noinline int early_drop(struct net *net, unsigned int hash) 425{ 426 /* Use oldest entry, which is roughly LRU */ 427 struct nf_conntrack_tuple_hash *h; 428 struct nf_conn *ct = NULL, *tmp; 429 struct hlist_node *n; 430 unsigned int i, cnt = 0; 431 int dropped = 0; 432 433 rcu_read_lock(); 434 for (i = 0; i < nf_conntrack_htable_size; i++) { 435 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], 436 hnode) { 437 tmp = nf_ct_tuplehash_to_ctrack(h); 438 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) 439 ct = tmp; 440 cnt++; 441 } 442 443 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use))) 444 ct = NULL; 445 if (ct || cnt >= NF_CT_EVICTION_RANGE) 446 break; 447 hash = (hash + 1) % nf_conntrack_htable_size; 448 } 449 rcu_read_unlock(); 450 451 if (!ct) 452 return dropped; 453 454 if (del_timer(&ct->timeout)) { 455 death_by_timeout((unsigned long)ct); 456 dropped = 1; 457 NF_CT_STAT_INC_ATOMIC(early_drop); 458 } 459 nf_ct_put(ct); 460 return dropped; 461} 462 463struct nf_conn *nf_conntrack_alloc(struct net *net, 464 const struct nf_conntrack_tuple *orig, 465 const struct nf_conntrack_tuple *repl, 466 gfp_t gfp) 467{ 468 struct nf_conn *ct = NULL; 469 470 if (unlikely(!nf_conntrack_hash_rnd_initted)) { 471 get_random_bytes(&nf_conntrack_hash_rnd, 4); 472 nf_conntrack_hash_rnd_initted = 1; 473 } 474 475 /* We don't want any race condition at early drop stage */ 476 atomic_inc(&net->ct.count); 477 478 if (nf_conntrack_max && 479 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { 480 unsigned int hash = hash_conntrack(orig); 481 if (!early_drop(net, hash)) { 482 atomic_dec(&net->ct.count); 483 if (net_ratelimit()) 484 printk(KERN_WARNING 485 "nf_conntrack: table full, dropping" 486 " packet.\n"); 487 return ERR_PTR(-ENOMEM); 488 } 489 } 490 491 ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp); 492 if (ct == NULL) { 493 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); 494 atomic_dec(&net->ct.count); 495 return ERR_PTR(-ENOMEM); 496 } 497 498 atomic_set(&ct->ct_general.use, 1); 499 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 500 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 501 /* Don't set timer yet: wait for confirmation */ 502 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); 503#ifdef CONFIG_NET_NS 504 ct->ct_net = net; 505#endif 506 INIT_RCU_HEAD(&ct->rcu); 507 508 return ct; 509} 510EXPORT_SYMBOL_GPL(nf_conntrack_alloc); 511 512static void nf_conntrack_free_rcu(struct rcu_head *head) 513{ 514 struct nf_conn *ct = container_of(head, struct nf_conn, rcu); 515 struct net *net = nf_ct_net(ct); 516 517 nf_ct_ext_free(ct); 518 kmem_cache_free(nf_conntrack_cachep, ct); 519 atomic_dec(&net->ct.count); 520} 521 522void nf_conntrack_free(struct nf_conn *ct) 523{ 524 nf_ct_ext_destroy(ct); 525 call_rcu(&ct->rcu, nf_conntrack_free_rcu); 526} 527EXPORT_SYMBOL_GPL(nf_conntrack_free); 528 529/* Allocate a new conntrack: we return -ENOMEM if classification 530 failed due to stress. Otherwise it really is unclassifiable. */ 531static struct nf_conntrack_tuple_hash * 532init_conntrack(struct net *net, 533 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(net, tuple, &repl_tuple, GFP_ATOMIC); 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 nf_ct_acct_ext_add(ct, GFP_ATOMIC); 562 563 spin_lock_bh(&nf_conntrack_lock); 564 exp = nf_ct_find_expectation(net, tuple); 565 if (exp) { 566 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", 567 ct, exp); 568 /* Welcome, Mr. Bond. We've been expecting you... */ 569 __set_bit(IPS_EXPECTED_BIT, &ct->status); 570 ct->master = exp->master; 571 if (exp->helper) { 572 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 573 if (help) 574 rcu_assign_pointer(help->helper, exp->helper); 575 } 576 577#ifdef CONFIG_NF_CONNTRACK_MARK 578 ct->mark = exp->master->mark; 579#endif 580#ifdef CONFIG_NF_CONNTRACK_SECMARK 581 ct->secmark = exp->master->secmark; 582#endif 583 nf_conntrack_get(&ct->master->ct_general); 584 NF_CT_STAT_INC(expect_new); 585 } else { 586 struct nf_conntrack_helper *helper; 587 588 helper = __nf_ct_helper_find(&repl_tuple); 589 if (helper) { 590 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 591 if (help) 592 rcu_assign_pointer(help->helper, helper); 593 } 594 NF_CT_STAT_INC(new); 595 } 596 597 /* Overload tuple linked list to put us in unconfirmed list. */ 598 hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, 599 &net->ct.unconfirmed); 600 601 spin_unlock_bh(&nf_conntrack_lock); 602 603 if (exp) { 604 if (exp->expectfn) 605 exp->expectfn(ct, exp); 606 nf_ct_expect_put(exp); 607 } 608 609 return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; 610} 611 612/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 613static inline struct nf_conn * 614resolve_normal_ct(struct net *net, 615 struct sk_buff *skb, 616 unsigned int dataoff, 617 u_int16_t l3num, 618 u_int8_t protonum, 619 struct nf_conntrack_l3proto *l3proto, 620 struct nf_conntrack_l4proto *l4proto, 621 int *set_reply, 622 enum ip_conntrack_info *ctinfo) 623{ 624 struct nf_conntrack_tuple tuple; 625 struct nf_conntrack_tuple_hash *h; 626 struct nf_conn *ct; 627 628 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), 629 dataoff, l3num, protonum, &tuple, l3proto, 630 l4proto)) { 631 pr_debug("resolve_normal_ct: Can't get tuple\n"); 632 return NULL; 633 } 634 635 /* look for tuple match */ 636 h = nf_conntrack_find_get(net, &tuple); 637 if (!h) { 638 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff); 639 if (!h) 640 return NULL; 641 if (IS_ERR(h)) 642 return (void *)h; 643 } 644 ct = nf_ct_tuplehash_to_ctrack(h); 645 646 /* It exists; we have (non-exclusive) reference. */ 647 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 648 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 649 /* Please set reply bit if this packet OK */ 650 *set_reply = 1; 651 } else { 652 /* Once we've had two way comms, always ESTABLISHED. */ 653 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 654 pr_debug("nf_conntrack_in: normal packet for %p\n", ct); 655 *ctinfo = IP_CT_ESTABLISHED; 656 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 657 pr_debug("nf_conntrack_in: related packet for %p\n", 658 ct); 659 *ctinfo = IP_CT_RELATED; 660 } else { 661 pr_debug("nf_conntrack_in: new packet for %p\n", ct); 662 *ctinfo = IP_CT_NEW; 663 } 664 *set_reply = 0; 665 } 666 skb->nfct = &ct->ct_general; 667 skb->nfctinfo = *ctinfo; 668 return ct; 669} 670 671unsigned int 672nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, 673 struct sk_buff *skb) 674{ 675 struct nf_conn *ct; 676 enum ip_conntrack_info ctinfo; 677 struct nf_conntrack_l3proto *l3proto; 678 struct nf_conntrack_l4proto *l4proto; 679 unsigned int dataoff; 680 u_int8_t protonum; 681 int set_reply = 0; 682 int ret; 683 684 /* Previously seen (loopback or untracked)? Ignore. */ 685 if (skb->nfct) { 686 NF_CT_STAT_INC_ATOMIC(ignore); 687 return NF_ACCEPT; 688 } 689 690 /* rcu_read_lock()ed by nf_hook_slow */ 691 l3proto = __nf_ct_l3proto_find(pf); 692 ret = l3proto->get_l4proto(skb, skb_network_offset(skb), 693 &dataoff, &protonum); 694 if (ret <= 0) { 695 pr_debug("not prepared to track yet or error occured\n"); 696 NF_CT_STAT_INC_ATOMIC(error); 697 NF_CT_STAT_INC_ATOMIC(invalid); 698 return -ret; 699 } 700 701 l4proto = __nf_ct_l4proto_find(pf, protonum); 702 703 /* It may be an special packet, error, unclean... 704 * inverse of the return code tells to the netfilter 705 * core what to do with the packet. */ 706 if (l4proto->error != NULL) { 707 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum); 708 if (ret <= 0) { 709 NF_CT_STAT_INC_ATOMIC(error); 710 NF_CT_STAT_INC_ATOMIC(invalid); 711 return -ret; 712 } 713 } 714 715 ct = resolve_normal_ct(net, skb, dataoff, pf, protonum, 716 l3proto, l4proto, &set_reply, &ctinfo); 717 if (!ct) { 718 /* Not valid part of a connection */ 719 NF_CT_STAT_INC_ATOMIC(invalid); 720 return NF_ACCEPT; 721 } 722 723 if (IS_ERR(ct)) { 724 /* Too stressed to deal. */ 725 NF_CT_STAT_INC_ATOMIC(drop); 726 return NF_DROP; 727 } 728 729 NF_CT_ASSERT(skb->nfct); 730 731 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); 732 if (ret < 0) { 733 /* Invalid: inverse of the return code tells 734 * the netfilter core what to do */ 735 pr_debug("nf_conntrack_in: Can't track with proto module\n"); 736 nf_conntrack_put(skb->nfct); 737 skb->nfct = NULL; 738 NF_CT_STAT_INC_ATOMIC(invalid); 739 return -ret; 740 } 741 742 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 743 nf_conntrack_event_cache(IPCT_STATUS, skb); 744 745 return ret; 746} 747EXPORT_SYMBOL_GPL(nf_conntrack_in); 748 749bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 750 const struct nf_conntrack_tuple *orig) 751{ 752 bool ret; 753 754 rcu_read_lock(); 755 ret = nf_ct_invert_tuple(inverse, orig, 756 __nf_ct_l3proto_find(orig->src.l3num), 757 __nf_ct_l4proto_find(orig->src.l3num, 758 orig->dst.protonum)); 759 rcu_read_unlock(); 760 return ret; 761} 762EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); 763 764/* Alter reply tuple (maybe alter helper). This is for NAT, and is 765 implicitly racy: see __nf_conntrack_confirm */ 766void nf_conntrack_alter_reply(struct nf_conn *ct, 767 const struct nf_conntrack_tuple *newreply) 768{ 769 struct nf_conn_help *help = nfct_help(ct); 770 struct nf_conntrack_helper *helper; 771 772 /* Should be unconfirmed, so not in hash table yet */ 773 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 774 775 pr_debug("Altering reply tuple of %p to ", ct); 776 nf_ct_dump_tuple(newreply); 777 778 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 779 if (ct->master || (help && !hlist_empty(&help->expectations))) 780 return; 781 782 rcu_read_lock(); 783 helper = __nf_ct_helper_find(newreply); 784 if (helper == NULL) { 785 if (help) 786 rcu_assign_pointer(help->helper, NULL); 787 goto out; 788 } 789 790 if (help == NULL) { 791 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 792 if (help == NULL) 793 goto out; 794 } else { 795 memset(&help->help, 0, sizeof(help->help)); 796 } 797 798 rcu_assign_pointer(help->helper, helper); 799out: 800 rcu_read_unlock(); 801} 802EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); 803 804/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 805void __nf_ct_refresh_acct(struct nf_conn *ct, 806 enum ip_conntrack_info ctinfo, 807 const struct sk_buff *skb, 808 unsigned long extra_jiffies, 809 int do_acct) 810{ 811 int event = 0; 812 813 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 814 NF_CT_ASSERT(skb); 815 816 spin_lock_bh(&nf_conntrack_lock); 817 818 /* Only update if this is not a fixed timeout */ 819 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) 820 goto acct; 821 822 /* If not in hash table, timer will not be active yet */ 823 if (!nf_ct_is_confirmed(ct)) { 824 ct->timeout.expires = extra_jiffies; 825 event = IPCT_REFRESH; 826 } else { 827 unsigned long newtime = jiffies + extra_jiffies; 828 829 /* Only update the timeout if the new timeout is at least 830 HZ jiffies from the old timeout. Need del_timer for race 831 avoidance (may already be dying). */ 832 if (newtime - ct->timeout.expires >= HZ 833 && del_timer(&ct->timeout)) { 834 ct->timeout.expires = newtime; 835 add_timer(&ct->timeout); 836 event = IPCT_REFRESH; 837 } 838 } 839 840acct: 841 if (do_acct) { 842 struct nf_conn_counter *acct; 843 844 acct = nf_conn_acct_find(ct); 845 if (acct) { 846 acct[CTINFO2DIR(ctinfo)].packets++; 847 acct[CTINFO2DIR(ctinfo)].bytes += 848 skb->len - skb_network_offset(skb); 849 } 850 } 851 852 spin_unlock_bh(&nf_conntrack_lock); 853 854 /* must be unlocked when calling event cache */ 855 if (event) 856 nf_conntrack_event_cache(event, skb); 857} 858EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); 859 860bool __nf_ct_kill_acct(struct nf_conn *ct, 861 enum ip_conntrack_info ctinfo, 862 const struct sk_buff *skb, 863 int do_acct) 864{ 865 if (do_acct) { 866 struct nf_conn_counter *acct; 867 868 spin_lock_bh(&nf_conntrack_lock); 869 acct = nf_conn_acct_find(ct); 870 if (acct) { 871 acct[CTINFO2DIR(ctinfo)].packets++; 872 acct[CTINFO2DIR(ctinfo)].bytes += 873 skb->len - skb_network_offset(skb); 874 } 875 spin_unlock_bh(&nf_conntrack_lock); 876 } 877 878 if (del_timer(&ct->timeout)) { 879 ct->timeout.function((unsigned long)ct); 880 return true; 881 } 882 return false; 883} 884EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); 885 886#if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) 887 888#include <linux/netfilter/nfnetlink.h> 889#include <linux/netfilter/nfnetlink_conntrack.h> 890#include <linux/mutex.h> 891 892/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 893 * in ip_conntrack_core, since we don't want the protocols to autoload 894 * or depend on ctnetlink */ 895int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, 896 const struct nf_conntrack_tuple *tuple) 897{ 898 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); 899 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); 900 return 0; 901 902nla_put_failure: 903 return -1; 904} 905EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); 906 907const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { 908 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, 909 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, 910}; 911EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); 912 913int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], 914 struct nf_conntrack_tuple *t) 915{ 916 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) 917 return -EINVAL; 918 919 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); 920 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); 921 922 return 0; 923} 924EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); 925#endif 926 927/* Used by ipt_REJECT and ip6t_REJECT. */ 928static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 929{ 930 struct nf_conn *ct; 931 enum ip_conntrack_info ctinfo; 932 933 /* This ICMP is in reverse direction to the packet which caused it */ 934 ct = nf_ct_get(skb, &ctinfo); 935 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 936 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 937 else 938 ctinfo = IP_CT_RELATED; 939 940 /* Attach to new skbuff, and increment count */ 941 nskb->nfct = &ct->ct_general; 942 nskb->nfctinfo = ctinfo; 943 nf_conntrack_get(nskb->nfct); 944} 945 946/* Bring out ya dead! */ 947static struct nf_conn * 948get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), 949 void *data, unsigned int *bucket) 950{ 951 struct nf_conntrack_tuple_hash *h; 952 struct nf_conn *ct; 953 struct hlist_node *n; 954 955 spin_lock_bh(&nf_conntrack_lock); 956 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) { 957 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) { 958 ct = nf_ct_tuplehash_to_ctrack(h); 959 if (iter(ct, data)) 960 goto found; 961 } 962 } 963 hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) { 964 ct = nf_ct_tuplehash_to_ctrack(h); 965 if (iter(ct, data)) 966 set_bit(IPS_DYING_BIT, &ct->status); 967 } 968 spin_unlock_bh(&nf_conntrack_lock); 969 return NULL; 970found: 971 atomic_inc(&ct->ct_general.use); 972 spin_unlock_bh(&nf_conntrack_lock); 973 return ct; 974} 975 976void nf_ct_iterate_cleanup(struct net *net, 977 int (*iter)(struct nf_conn *i, void *data), 978 void *data) 979{ 980 struct nf_conn *ct; 981 unsigned int bucket = 0; 982 983 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { 984 /* Time to push up daises... */ 985 if (del_timer(&ct->timeout)) 986 death_by_timeout((unsigned long)ct); 987 /* ... else the timer will get him soon. */ 988 989 nf_ct_put(ct); 990 } 991} 992EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); 993 994static int kill_all(struct nf_conn *i, void *data) 995{ 996 return 1; 997} 998 999void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size) 1000{ 1001 if (vmalloced) 1002 vfree(hash); 1003 else 1004 free_pages((unsigned long)hash, 1005 get_order(sizeof(struct hlist_head) * size)); 1006} 1007EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); 1008 1009void nf_conntrack_flush(struct net *net) 1010{ 1011 nf_ct_iterate_cleanup(net, kill_all, NULL); 1012} 1013EXPORT_SYMBOL_GPL(nf_conntrack_flush); 1014 1015/* Mishearing the voices in his head, our hero wonders how he's 1016 supposed to kill the mall. */ 1017void nf_conntrack_cleanup(struct net *net) 1018{ 1019 rcu_assign_pointer(ip_ct_attach, NULL); 1020 1021 /* This makes sure all current packets have passed through 1022 netfilter framework. Roll on, two-stage module 1023 delete... */ 1024 synchronize_net(); 1025 1026 nf_ct_event_cache_flush(); 1027 i_see_dead_people: 1028 nf_conntrack_flush(net); 1029 if (atomic_read(&net->ct.count) != 0) { 1030 schedule(); 1031 goto i_see_dead_people; 1032 } 1033 /* wait until all references to nf_conntrack_untracked are dropped */ 1034 while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1) 1035 schedule(); 1036 1037 rcu_assign_pointer(nf_ct_destroy, NULL); 1038 1039 kmem_cache_destroy(nf_conntrack_cachep); 1040 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, 1041 nf_conntrack_htable_size); 1042 1043 nf_conntrack_acct_fini(); 1044 nf_conntrack_expect_fini(net); 1045 nf_conntrack_helper_fini(); 1046 nf_conntrack_proto_fini(); 1047} 1048 1049struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced) 1050{ 1051 struct hlist_head *hash; 1052 unsigned int size, i; 1053 1054 *vmalloced = 0; 1055 1056 size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head)); 1057 hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN, 1058 get_order(sizeof(struct hlist_head) 1059 * size)); 1060 if (!hash) { 1061 *vmalloced = 1; 1062 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1063 hash = vmalloc(sizeof(struct hlist_head) * size); 1064 } 1065 1066 if (hash) 1067 for (i = 0; i < size; i++) 1068 INIT_HLIST_HEAD(&hash[i]); 1069 1070 return hash; 1071} 1072EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); 1073 1074int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) 1075{ 1076 int i, bucket, vmalloced, old_vmalloced; 1077 unsigned int hashsize, old_size; 1078 int rnd; 1079 struct hlist_head *hash, *old_hash; 1080 struct nf_conntrack_tuple_hash *h; 1081 1082 /* On boot, we can set this without any fancy locking. */ 1083 if (!nf_conntrack_htable_size) 1084 return param_set_uint(val, kp); 1085 1086 hashsize = simple_strtoul(val, NULL, 0); 1087 if (!hashsize) 1088 return -EINVAL; 1089 1090 hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced); 1091 if (!hash) 1092 return -ENOMEM; 1093 1094 /* We have to rehahs for the new table anyway, so we also can 1095 * use a newrandom seed */ 1096 get_random_bytes(&rnd, 4); 1097 1098 /* Lookups in the old hash might happen in parallel, which means we 1099 * might get false negatives during connection lookup. New connections 1100 * created because of a false negative won't make it into the hash 1101 * though since that required taking the lock. 1102 */ 1103 spin_lock_bh(&nf_conntrack_lock); 1104 for (i = 0; i < nf_conntrack_htable_size; i++) { 1105 while (!hlist_empty(&init_net.ct.hash[i])) { 1106 h = hlist_entry(init_net.ct.hash[i].first, 1107 struct nf_conntrack_tuple_hash, hnode); 1108 hlist_del_rcu(&h->hnode); 1109 bucket = __hash_conntrack(&h->tuple, hashsize, rnd); 1110 hlist_add_head(&h->hnode, &hash[bucket]); 1111 } 1112 } 1113 old_size = nf_conntrack_htable_size; 1114 old_vmalloced = init_net.ct.hash_vmalloc; 1115 old_hash = init_net.ct.hash; 1116 1117 nf_conntrack_htable_size = hashsize; 1118 init_net.ct.hash_vmalloc = vmalloced; 1119 init_net.ct.hash = hash; 1120 nf_conntrack_hash_rnd = rnd; 1121 spin_unlock_bh(&nf_conntrack_lock); 1122 1123 nf_ct_free_hashtable(old_hash, old_vmalloced, old_size); 1124 return 0; 1125} 1126EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); 1127 1128module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, 1129 &nf_conntrack_htable_size, 0600); 1130 1131int nf_conntrack_init(struct net *net) 1132{ 1133 int max_factor = 8; 1134 int ret; 1135 1136 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1137 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ 1138 if (!nf_conntrack_htable_size) { 1139 nf_conntrack_htable_size 1140 = (((num_physpages << PAGE_SHIFT) / 16384) 1141 / sizeof(struct hlist_head)); 1142 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1143 nf_conntrack_htable_size = 16384; 1144 if (nf_conntrack_htable_size < 32) 1145 nf_conntrack_htable_size = 32; 1146 1147 /* Use a max. factor of four by default to get the same max as 1148 * with the old struct list_heads. When a table size is given 1149 * we use the old value of 8 to avoid reducing the max. 1150 * entries. */ 1151 max_factor = 4; 1152 } 1153 atomic_set(&net->ct.count, 0); 1154 net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1155 &net->ct.hash_vmalloc); 1156 if (!net->ct.hash) { 1157 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1158 goto err_out; 1159 } 1160 INIT_HLIST_HEAD(&net->ct.unconfirmed); 1161 1162 nf_conntrack_max = max_factor * nf_conntrack_htable_size; 1163 1164 printk("nf_conntrack version %s (%u buckets, %d max)\n", 1165 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1166 nf_conntrack_max); 1167 1168 nf_conntrack_cachep = kmem_cache_create("nf_conntrack", 1169 sizeof(struct nf_conn), 1170 0, 0, NULL); 1171 if (!nf_conntrack_cachep) { 1172 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1173 goto err_free_hash; 1174 } 1175 1176 ret = nf_conntrack_proto_init(); 1177 if (ret < 0) 1178 goto err_free_conntrack_slab; 1179 1180 ret = nf_conntrack_expect_init(net); 1181 if (ret < 0) 1182 goto out_fini_proto; 1183 1184 ret = nf_conntrack_helper_init(); 1185 if (ret < 0) 1186 goto out_fini_expect; 1187 1188 ret = nf_conntrack_acct_init(); 1189 if (ret < 0) 1190 goto out_fini_helper; 1191 1192 /* For use by REJECT target */ 1193 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); 1194 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); 1195 1196 /* Set up fake conntrack: 1197 - to never be deleted, not in any hashes */ 1198#ifdef CONFIG_NET_NS 1199 nf_conntrack_untracked.ct_net = &init_net; 1200#endif 1201 atomic_set(&nf_conntrack_untracked.ct_general.use, 1); 1202 /* - and look it like as a confirmed connection */ 1203 set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status); 1204 1205 return ret; 1206 1207out_fini_helper: 1208 nf_conntrack_helper_fini(); 1209out_fini_expect: 1210 nf_conntrack_expect_fini(net); 1211out_fini_proto: 1212 nf_conntrack_proto_fini(); 1213err_free_conntrack_slab: 1214 kmem_cache_destroy(nf_conntrack_cachep); 1215err_free_hash: 1216 nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc, 1217 nf_conntrack_htable_size); 1218err_out: 1219 return -ENOMEM; 1220} 1221