nf_conntrack_core.c revision d862a6622e9db508d4b28cc7c5bc28bd548cc24e
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/sched.h> 18#include <linux/skbuff.h> 19#include <linux/proc_fs.h> 20#include <linux/vmalloc.h> 21#include <linux/stddef.h> 22#include <linux/slab.h> 23#include <linux/random.h> 24#include <linux/jhash.h> 25#include <linux/err.h> 26#include <linux/percpu.h> 27#include <linux/moduleparam.h> 28#include <linux/notifier.h> 29#include <linux/kernel.h> 30#include <linux/netdevice.h> 31#include <linux/socket.h> 32#include <linux/mm.h> 33#include <linux/nsproxy.h> 34#include <linux/rculist_nulls.h> 35 36#include <net/netfilter/nf_conntrack.h> 37#include <net/netfilter/nf_conntrack_l3proto.h> 38#include <net/netfilter/nf_conntrack_l4proto.h> 39#include <net/netfilter/nf_conntrack_expect.h> 40#include <net/netfilter/nf_conntrack_helper.h> 41#include <net/netfilter/nf_conntrack_core.h> 42#include <net/netfilter/nf_conntrack_extend.h> 43#include <net/netfilter/nf_conntrack_acct.h> 44#include <net/netfilter/nf_conntrack_ecache.h> 45#include <net/netfilter/nf_conntrack_zones.h> 46#include <net/netfilter/nf_nat.h> 47#include <net/netfilter/nf_nat_core.h> 48 49#define NF_CONNTRACK_VERSION "0.5.0" 50 51int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, 52 enum nf_nat_manip_type manip, 53 const struct nlattr *attr) __read_mostly; 54EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); 55 56DEFINE_SPINLOCK(nf_conntrack_lock); 57EXPORT_SYMBOL_GPL(nf_conntrack_lock); 58 59unsigned int nf_conntrack_htable_size __read_mostly; 60EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); 61 62unsigned int nf_conntrack_max __read_mostly; 63EXPORT_SYMBOL_GPL(nf_conntrack_max); 64 65DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); 66EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); 67 68unsigned int nf_conntrack_hash_rnd __read_mostly; 69 70static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone) 71{ 72 unsigned int n; 73 74 /* The direction must be ignored, so we hash everything up to the 75 * destination ports (which is a multiple of 4) and treat the last 76 * three bytes manually. 77 */ 78 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); 79 return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^ 80 (((__force __u16)tuple->dst.u.all << 16) | 81 tuple->dst.protonum)); 82} 83 84static u32 __hash_bucket(u32 hash, unsigned int size) 85{ 86 return ((u64)hash * size) >> 32; 87} 88 89static u32 hash_bucket(u32 hash, const struct net *net) 90{ 91 return __hash_bucket(hash, net->ct.htable_size); 92} 93 94static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 95 u16 zone, unsigned int size) 96{ 97 return __hash_bucket(hash_conntrack_raw(tuple, zone), size); 98} 99 100static inline u_int32_t hash_conntrack(const struct net *net, u16 zone, 101 const struct nf_conntrack_tuple *tuple) 102{ 103 return __hash_conntrack(tuple, zone, net->ct.htable_size); 104} 105 106bool 107nf_ct_get_tuple(const struct sk_buff *skb, 108 unsigned int nhoff, 109 unsigned int dataoff, 110 u_int16_t l3num, 111 u_int8_t protonum, 112 struct nf_conntrack_tuple *tuple, 113 const struct nf_conntrack_l3proto *l3proto, 114 const struct nf_conntrack_l4proto *l4proto) 115{ 116 memset(tuple, 0, sizeof(*tuple)); 117 118 tuple->src.l3num = l3num; 119 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 120 return false; 121 122 tuple->dst.protonum = protonum; 123 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 124 125 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 126} 127EXPORT_SYMBOL_GPL(nf_ct_get_tuple); 128 129bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, 130 u_int16_t l3num, struct nf_conntrack_tuple *tuple) 131{ 132 struct nf_conntrack_l3proto *l3proto; 133 struct nf_conntrack_l4proto *l4proto; 134 unsigned int protoff; 135 u_int8_t protonum; 136 int ret; 137 138 rcu_read_lock(); 139 140 l3proto = __nf_ct_l3proto_find(l3num); 141 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); 142 if (ret != NF_ACCEPT) { 143 rcu_read_unlock(); 144 return false; 145 } 146 147 l4proto = __nf_ct_l4proto_find(l3num, protonum); 148 149 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, 150 l3proto, l4proto); 151 152 rcu_read_unlock(); 153 return ret; 154} 155EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); 156 157bool 158nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 159 const struct nf_conntrack_tuple *orig, 160 const struct nf_conntrack_l3proto *l3proto, 161 const struct nf_conntrack_l4proto *l4proto) 162{ 163 memset(inverse, 0, sizeof(*inverse)); 164 165 inverse->src.l3num = orig->src.l3num; 166 if (l3proto->invert_tuple(inverse, orig) == 0) 167 return false; 168 169 inverse->dst.dir = !orig->dst.dir; 170 171 inverse->dst.protonum = orig->dst.protonum; 172 return l4proto->invert_tuple(inverse, orig); 173} 174EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); 175 176static void 177clean_from_lists(struct nf_conn *ct) 178{ 179 pr_debug("clean_from_lists(%p)\n", ct); 180 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 181 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); 182 183 /* Destroy all pending expectations */ 184 nf_ct_remove_expectations(ct); 185} 186 187static void 188destroy_conntrack(struct nf_conntrack *nfct) 189{ 190 struct nf_conn *ct = (struct nf_conn *)nfct; 191 struct net *net = nf_ct_net(ct); 192 struct nf_conntrack_l4proto *l4proto; 193 194 pr_debug("destroy_conntrack(%p)\n", ct); 195 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 196 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 197 198 /* To make sure we don't get any weird locking issues here: 199 * destroy_conntrack() MUST NOT be called with a write lock 200 * to nf_conntrack_lock!!! -HW */ 201 rcu_read_lock(); 202 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); 203 if (l4proto && l4proto->destroy) 204 l4proto->destroy(ct); 205 206 rcu_read_unlock(); 207 208 spin_lock_bh(&nf_conntrack_lock); 209 /* Expectations will have been removed in clean_from_lists, 210 * except TFTP can create an expectation on the first packet, 211 * before connection is in the list, so we need to clean here, 212 * too. */ 213 nf_ct_remove_expectations(ct); 214 215 /* We overload first tuple to link into unconfirmed list. */ 216 if (!nf_ct_is_confirmed(ct)) { 217 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); 218 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 219 } 220 221 NF_CT_STAT_INC(net, delete); 222 spin_unlock_bh(&nf_conntrack_lock); 223 224 if (ct->master) 225 nf_ct_put(ct->master); 226 227 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); 228 nf_conntrack_free(ct); 229} 230 231void nf_ct_delete_from_lists(struct nf_conn *ct) 232{ 233 struct net *net = nf_ct_net(ct); 234 235 nf_ct_helper_destroy(ct); 236 spin_lock_bh(&nf_conntrack_lock); 237 /* Inside lock so preempt is disabled on module removal path. 238 * Otherwise we can get spurious warnings. */ 239 NF_CT_STAT_INC(net, delete_list); 240 clean_from_lists(ct); 241 spin_unlock_bh(&nf_conntrack_lock); 242} 243EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); 244 245static void death_by_event(unsigned long ul_conntrack) 246{ 247 struct nf_conn *ct = (void *)ul_conntrack; 248 struct net *net = nf_ct_net(ct); 249 250 if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { 251 /* bad luck, let's retry again */ 252 ct->timeout.expires = jiffies + 253 (random32() % net->ct.sysctl_events_retry_timeout); 254 add_timer(&ct->timeout); 255 return; 256 } 257 /* we've got the event delivered, now it's dying */ 258 set_bit(IPS_DYING_BIT, &ct->status); 259 spin_lock(&nf_conntrack_lock); 260 hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 261 spin_unlock(&nf_conntrack_lock); 262 nf_ct_put(ct); 263} 264 265void nf_ct_insert_dying_list(struct nf_conn *ct) 266{ 267 struct net *net = nf_ct_net(ct); 268 269 /* add this conntrack to the dying list */ 270 spin_lock_bh(&nf_conntrack_lock); 271 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 272 &net->ct.dying); 273 spin_unlock_bh(&nf_conntrack_lock); 274 /* set a new timer to retry event delivery */ 275 setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); 276 ct->timeout.expires = jiffies + 277 (random32() % net->ct.sysctl_events_retry_timeout); 278 add_timer(&ct->timeout); 279} 280EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); 281 282static void death_by_timeout(unsigned long ul_conntrack) 283{ 284 struct nf_conn *ct = (void *)ul_conntrack; 285 286 if (!test_bit(IPS_DYING_BIT, &ct->status) && 287 unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { 288 /* destroy event was not delivered */ 289 nf_ct_delete_from_lists(ct); 290 nf_ct_insert_dying_list(ct); 291 return; 292 } 293 set_bit(IPS_DYING_BIT, &ct->status); 294 nf_ct_delete_from_lists(ct); 295 nf_ct_put(ct); 296} 297 298/* 299 * Warning : 300 * - Caller must take a reference on returned object 301 * and recheck nf_ct_tuple_equal(tuple, &h->tuple) 302 * OR 303 * - Caller must lock nf_conntrack_lock before calling this function 304 */ 305static struct nf_conntrack_tuple_hash * 306____nf_conntrack_find(struct net *net, u16 zone, 307 const struct nf_conntrack_tuple *tuple, u32 hash) 308{ 309 struct nf_conntrack_tuple_hash *h; 310 struct hlist_nulls_node *n; 311 unsigned int bucket = hash_bucket(hash, net); 312 313 /* Disable BHs the entire time since we normally need to disable them 314 * at least once for the stats anyway. 315 */ 316 local_bh_disable(); 317begin: 318 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) { 319 if (nf_ct_tuple_equal(tuple, &h->tuple) && 320 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) { 321 NF_CT_STAT_INC(net, found); 322 local_bh_enable(); 323 return h; 324 } 325 NF_CT_STAT_INC(net, searched); 326 } 327 /* 328 * if the nulls value we got at the end of this lookup is 329 * not the expected one, we must restart lookup. 330 * We probably met an item that was moved to another chain. 331 */ 332 if (get_nulls_value(n) != bucket) { 333 NF_CT_STAT_INC(net, search_restart); 334 goto begin; 335 } 336 local_bh_enable(); 337 338 return NULL; 339} 340 341struct nf_conntrack_tuple_hash * 342__nf_conntrack_find(struct net *net, u16 zone, 343 const struct nf_conntrack_tuple *tuple) 344{ 345 return ____nf_conntrack_find(net, zone, tuple, 346 hash_conntrack_raw(tuple, zone)); 347} 348EXPORT_SYMBOL_GPL(__nf_conntrack_find); 349 350/* Find a connection corresponding to a tuple. */ 351static struct nf_conntrack_tuple_hash * 352__nf_conntrack_find_get(struct net *net, u16 zone, 353 const struct nf_conntrack_tuple *tuple, u32 hash) 354{ 355 struct nf_conntrack_tuple_hash *h; 356 struct nf_conn *ct; 357 358 rcu_read_lock(); 359begin: 360 h = ____nf_conntrack_find(net, zone, tuple, hash); 361 if (h) { 362 ct = nf_ct_tuplehash_to_ctrack(h); 363 if (unlikely(nf_ct_is_dying(ct) || 364 !atomic_inc_not_zero(&ct->ct_general.use))) 365 h = NULL; 366 else { 367 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) || 368 nf_ct_zone(ct) != zone)) { 369 nf_ct_put(ct); 370 goto begin; 371 } 372 } 373 } 374 rcu_read_unlock(); 375 376 return h; 377} 378 379struct nf_conntrack_tuple_hash * 380nf_conntrack_find_get(struct net *net, u16 zone, 381 const struct nf_conntrack_tuple *tuple) 382{ 383 return __nf_conntrack_find_get(net, zone, tuple, 384 hash_conntrack_raw(tuple, zone)); 385} 386EXPORT_SYMBOL_GPL(nf_conntrack_find_get); 387 388static void __nf_conntrack_hash_insert(struct nf_conn *ct, 389 unsigned int hash, 390 unsigned int repl_hash) 391{ 392 struct net *net = nf_ct_net(ct); 393 394 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 395 &net->ct.hash[hash]); 396 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, 397 &net->ct.hash[repl_hash]); 398} 399 400void nf_conntrack_hash_insert(struct nf_conn *ct) 401{ 402 struct net *net = nf_ct_net(ct); 403 unsigned int hash, repl_hash; 404 u16 zone; 405 406 zone = nf_ct_zone(ct); 407 hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 408 repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 409 410 __nf_conntrack_hash_insert(ct, hash, repl_hash); 411} 412EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert); 413 414/* Confirm a connection given skb; places it in hash table */ 415int 416__nf_conntrack_confirm(struct sk_buff *skb) 417{ 418 unsigned int hash, repl_hash; 419 struct nf_conntrack_tuple_hash *h; 420 struct nf_conn *ct; 421 struct nf_conn_help *help; 422 struct hlist_nulls_node *n; 423 enum ip_conntrack_info ctinfo; 424 struct net *net; 425 u16 zone; 426 427 ct = nf_ct_get(skb, &ctinfo); 428 net = nf_ct_net(ct); 429 430 /* ipt_REJECT uses nf_conntrack_attach to attach related 431 ICMP/TCP RST packets in other direction. Actual packet 432 which created connection will be IP_CT_NEW or for an 433 expected connection, IP_CT_RELATED. */ 434 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 435 return NF_ACCEPT; 436 437 zone = nf_ct_zone(ct); 438 /* reuse the hash saved before */ 439 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; 440 hash = hash_bucket(hash, net); 441 repl_hash = hash_conntrack(net, zone, 442 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 443 444 /* We're not in hash table, and we refuse to set up related 445 connections for unconfirmed conns. But packet copies and 446 REJECT will give spurious warnings here. */ 447 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 448 449 /* No external references means noone else could have 450 confirmed us. */ 451 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 452 pr_debug("Confirming conntrack %p\n", ct); 453 454 spin_lock_bh(&nf_conntrack_lock); 455 456 /* We have to check the DYING flag inside the lock to prevent 457 a race against nf_ct_get_next_corpse() possibly called from 458 user context, else we insert an already 'dead' hash, blocking 459 further use of that particular connection -JM */ 460 461 if (unlikely(nf_ct_is_dying(ct))) { 462 spin_unlock_bh(&nf_conntrack_lock); 463 return NF_ACCEPT; 464 } 465 466 /* See if there's one in the list already, including reverse: 467 NAT could have grabbed it without realizing, since we're 468 not in the hash. If there is, we lost race. */ 469 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) 470 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 471 &h->tuple) && 472 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 473 goto out; 474 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) 475 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 476 &h->tuple) && 477 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 478 goto out; 479 480 /* Remove from unconfirmed list */ 481 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 482 483 /* Timer relative to confirmation time, not original 484 setting time, otherwise we'd get timer wrap in 485 weird delay cases. */ 486 ct->timeout.expires += jiffies; 487 add_timer(&ct->timeout); 488 atomic_inc(&ct->ct_general.use); 489 set_bit(IPS_CONFIRMED_BIT, &ct->status); 490 491 /* Since the lookup is lockless, hash insertion must be done after 492 * starting the timer and setting the CONFIRMED bit. The RCU barriers 493 * guarantee that no other CPU can find the conntrack before the above 494 * stores are visible. 495 */ 496 __nf_conntrack_hash_insert(ct, hash, repl_hash); 497 NF_CT_STAT_INC(net, insert); 498 spin_unlock_bh(&nf_conntrack_lock); 499 500 help = nfct_help(ct); 501 if (help && help->helper) 502 nf_conntrack_event_cache(IPCT_HELPER, ct); 503 504 nf_conntrack_event_cache(master_ct(ct) ? 505 IPCT_RELATED : IPCT_NEW, ct); 506 return NF_ACCEPT; 507 508out: 509 NF_CT_STAT_INC(net, insert_failed); 510 spin_unlock_bh(&nf_conntrack_lock); 511 return NF_DROP; 512} 513EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); 514 515/* Returns true if a connection correspondings to the tuple (required 516 for NAT). */ 517int 518nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 519 const struct nf_conn *ignored_conntrack) 520{ 521 struct net *net = nf_ct_net(ignored_conntrack); 522 struct nf_conntrack_tuple_hash *h; 523 struct hlist_nulls_node *n; 524 struct nf_conn *ct; 525 u16 zone = nf_ct_zone(ignored_conntrack); 526 unsigned int hash = hash_conntrack(net, zone, tuple); 527 528 /* Disable BHs the entire time since we need to disable them at 529 * least once for the stats anyway. 530 */ 531 rcu_read_lock_bh(); 532 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { 533 ct = nf_ct_tuplehash_to_ctrack(h); 534 if (ct != ignored_conntrack && 535 nf_ct_tuple_equal(tuple, &h->tuple) && 536 nf_ct_zone(ct) == zone) { 537 NF_CT_STAT_INC(net, found); 538 rcu_read_unlock_bh(); 539 return 1; 540 } 541 NF_CT_STAT_INC(net, searched); 542 } 543 rcu_read_unlock_bh(); 544 545 return 0; 546} 547EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); 548 549#define NF_CT_EVICTION_RANGE 8 550 551/* There's a small race here where we may free a just-assured 552 connection. Too bad: we're in trouble anyway. */ 553static noinline int early_drop(struct net *net, unsigned int hash) 554{ 555 /* Use oldest entry, which is roughly LRU */ 556 struct nf_conntrack_tuple_hash *h; 557 struct nf_conn *ct = NULL, *tmp; 558 struct hlist_nulls_node *n; 559 unsigned int i, cnt = 0; 560 int dropped = 0; 561 562 rcu_read_lock(); 563 for (i = 0; i < net->ct.htable_size; i++) { 564 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], 565 hnnode) { 566 tmp = nf_ct_tuplehash_to_ctrack(h); 567 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) 568 ct = tmp; 569 cnt++; 570 } 571 572 if (ct != NULL) { 573 if (likely(!nf_ct_is_dying(ct) && 574 atomic_inc_not_zero(&ct->ct_general.use))) 575 break; 576 else 577 ct = NULL; 578 } 579 580 if (cnt >= NF_CT_EVICTION_RANGE) 581 break; 582 583 hash = (hash + 1) % net->ct.htable_size; 584 } 585 rcu_read_unlock(); 586 587 if (!ct) 588 return dropped; 589 590 if (del_timer(&ct->timeout)) { 591 death_by_timeout((unsigned long)ct); 592 dropped = 1; 593 NF_CT_STAT_INC_ATOMIC(net, early_drop); 594 } 595 nf_ct_put(ct); 596 return dropped; 597} 598 599void init_nf_conntrack_hash_rnd(void) 600{ 601 unsigned int rand; 602 603 /* 604 * Why not initialize nf_conntrack_rnd in a "init()" function ? 605 * Because there isn't enough entropy when system initializing, 606 * and we initialize it as late as possible. 607 */ 608 do { 609 get_random_bytes(&rand, sizeof(rand)); 610 } while (!rand); 611 cmpxchg(&nf_conntrack_hash_rnd, 0, rand); 612} 613 614static struct nf_conn * 615__nf_conntrack_alloc(struct net *net, u16 zone, 616 const struct nf_conntrack_tuple *orig, 617 const struct nf_conntrack_tuple *repl, 618 gfp_t gfp, u32 hash) 619{ 620 struct nf_conn *ct; 621 622 if (unlikely(!nf_conntrack_hash_rnd)) { 623 init_nf_conntrack_hash_rnd(); 624 /* recompute the hash as nf_conntrack_hash_rnd is initialized */ 625 hash = hash_conntrack_raw(orig, zone); 626 } 627 628 /* We don't want any race condition at early drop stage */ 629 atomic_inc(&net->ct.count); 630 631 if (nf_conntrack_max && 632 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { 633 if (!early_drop(net, hash_bucket(hash, net))) { 634 atomic_dec(&net->ct.count); 635 if (net_ratelimit()) 636 printk(KERN_WARNING 637 "nf_conntrack: table full, dropping" 638 " packet.\n"); 639 return ERR_PTR(-ENOMEM); 640 } 641 } 642 643 /* 644 * Do not use kmem_cache_zalloc(), as this cache uses 645 * SLAB_DESTROY_BY_RCU. 646 */ 647 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); 648 if (ct == NULL) { 649 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n"); 650 atomic_dec(&net->ct.count); 651 return ERR_PTR(-ENOMEM); 652 } 653 /* 654 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next 655 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. 656 */ 657 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, 658 offsetof(struct nf_conn, proto) - 659 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); 660 spin_lock_init(&ct->lock); 661 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 662 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; 663 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 664 /* save hash for reusing when confirming */ 665 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; 666 /* Don't set timer yet: wait for confirmation */ 667 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); 668 write_pnet(&ct->ct_net, net); 669#ifdef CONFIG_NF_CONNTRACK_ZONES 670 if (zone) { 671 struct nf_conntrack_zone *nf_ct_zone; 672 673 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC); 674 if (!nf_ct_zone) 675 goto out_free; 676 nf_ct_zone->id = zone; 677 } 678#endif 679 /* 680 * changes to lookup keys must be done before setting refcnt to 1 681 */ 682 smp_wmb(); 683 atomic_set(&ct->ct_general.use, 1); 684 return ct; 685 686#ifdef CONFIG_NF_CONNTRACK_ZONES 687out_free: 688 kmem_cache_free(net->ct.nf_conntrack_cachep, ct); 689 return ERR_PTR(-ENOMEM); 690#endif 691} 692 693struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone, 694 const struct nf_conntrack_tuple *orig, 695 const struct nf_conntrack_tuple *repl, 696 gfp_t gfp) 697{ 698 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); 699} 700EXPORT_SYMBOL_GPL(nf_conntrack_alloc); 701 702void nf_conntrack_free(struct nf_conn *ct) 703{ 704 struct net *net = nf_ct_net(ct); 705 706 nf_ct_ext_destroy(ct); 707 atomic_dec(&net->ct.count); 708 nf_ct_ext_free(ct); 709 kmem_cache_free(net->ct.nf_conntrack_cachep, ct); 710} 711EXPORT_SYMBOL_GPL(nf_conntrack_free); 712 713/* Allocate a new conntrack: we return -ENOMEM if classification 714 failed due to stress. Otherwise it really is unclassifiable. */ 715static struct nf_conntrack_tuple_hash * 716init_conntrack(struct net *net, struct nf_conn *tmpl, 717 const struct nf_conntrack_tuple *tuple, 718 struct nf_conntrack_l3proto *l3proto, 719 struct nf_conntrack_l4proto *l4proto, 720 struct sk_buff *skb, 721 unsigned int dataoff, u32 hash) 722{ 723 struct nf_conn *ct; 724 struct nf_conn_help *help; 725 struct nf_conntrack_tuple repl_tuple; 726 struct nf_conntrack_ecache *ecache; 727 struct nf_conntrack_expect *exp; 728 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; 729 730 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 731 pr_debug("Can't invert tuple.\n"); 732 return NULL; 733 } 734 735 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, 736 hash); 737 if (IS_ERR(ct)) { 738 pr_debug("Can't allocate conntrack.\n"); 739 return (struct nf_conntrack_tuple_hash *)ct; 740 } 741 742 if (!l4proto->new(ct, skb, dataoff)) { 743 nf_conntrack_free(ct); 744 pr_debug("init conntrack: can't track with proto module\n"); 745 return NULL; 746 } 747 748 nf_ct_acct_ext_add(ct, GFP_ATOMIC); 749 750 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; 751 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, 752 ecache ? ecache->expmask : 0, 753 GFP_ATOMIC); 754 755 spin_lock_bh(&nf_conntrack_lock); 756 exp = nf_ct_find_expectation(net, zone, tuple); 757 if (exp) { 758 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", 759 ct, exp); 760 /* Welcome, Mr. Bond. We've been expecting you... */ 761 __set_bit(IPS_EXPECTED_BIT, &ct->status); 762 ct->master = exp->master; 763 if (exp->helper) { 764 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 765 if (help) 766 rcu_assign_pointer(help->helper, exp->helper); 767 } 768 769#ifdef CONFIG_NF_CONNTRACK_MARK 770 ct->mark = exp->master->mark; 771#endif 772#ifdef CONFIG_NF_CONNTRACK_SECMARK 773 ct->secmark = exp->master->secmark; 774#endif 775 nf_conntrack_get(&ct->master->ct_general); 776 NF_CT_STAT_INC(net, expect_new); 777 } else { 778 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); 779 NF_CT_STAT_INC(net, new); 780 } 781 782 /* Overload tuple linked list to put us in unconfirmed list. */ 783 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 784 &net->ct.unconfirmed); 785 786 spin_unlock_bh(&nf_conntrack_lock); 787 788 if (exp) { 789 if (exp->expectfn) 790 exp->expectfn(ct, exp); 791 nf_ct_expect_put(exp); 792 } 793 794 return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; 795} 796 797/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 798static inline struct nf_conn * 799resolve_normal_ct(struct net *net, struct nf_conn *tmpl, 800 struct sk_buff *skb, 801 unsigned int dataoff, 802 u_int16_t l3num, 803 u_int8_t protonum, 804 struct nf_conntrack_l3proto *l3proto, 805 struct nf_conntrack_l4proto *l4proto, 806 int *set_reply, 807 enum ip_conntrack_info *ctinfo) 808{ 809 struct nf_conntrack_tuple tuple; 810 struct nf_conntrack_tuple_hash *h; 811 struct nf_conn *ct; 812 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; 813 u32 hash; 814 815 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), 816 dataoff, l3num, protonum, &tuple, l3proto, 817 l4proto)) { 818 pr_debug("resolve_normal_ct: Can't get tuple\n"); 819 return NULL; 820 } 821 822 /* look for tuple match */ 823 hash = hash_conntrack_raw(&tuple, zone); 824 h = __nf_conntrack_find_get(net, zone, &tuple, hash); 825 if (!h) { 826 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, 827 skb, dataoff, hash); 828 if (!h) 829 return NULL; 830 if (IS_ERR(h)) 831 return (void *)h; 832 } 833 ct = nf_ct_tuplehash_to_ctrack(h); 834 835 /* It exists; we have (non-exclusive) reference. */ 836 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 837 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY; 838 /* Please set reply bit if this packet OK */ 839 *set_reply = 1; 840 } else { 841 /* Once we've had two way comms, always ESTABLISHED. */ 842 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 843 pr_debug("nf_conntrack_in: normal packet for %p\n", ct); 844 *ctinfo = IP_CT_ESTABLISHED; 845 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 846 pr_debug("nf_conntrack_in: related packet for %p\n", 847 ct); 848 *ctinfo = IP_CT_RELATED; 849 } else { 850 pr_debug("nf_conntrack_in: new packet for %p\n", ct); 851 *ctinfo = IP_CT_NEW; 852 } 853 *set_reply = 0; 854 } 855 skb->nfct = &ct->ct_general; 856 skb->nfctinfo = *ctinfo; 857 return ct; 858} 859 860unsigned int 861nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, 862 struct sk_buff *skb) 863{ 864 struct nf_conn *ct, *tmpl = NULL; 865 enum ip_conntrack_info ctinfo; 866 struct nf_conntrack_l3proto *l3proto; 867 struct nf_conntrack_l4proto *l4proto; 868 unsigned int dataoff; 869 u_int8_t protonum; 870 int set_reply = 0; 871 int ret; 872 873 if (skb->nfct) { 874 /* Previously seen (loopback or untracked)? Ignore. */ 875 tmpl = (struct nf_conn *)skb->nfct; 876 if (!nf_ct_is_template(tmpl)) { 877 NF_CT_STAT_INC_ATOMIC(net, ignore); 878 return NF_ACCEPT; 879 } 880 skb->nfct = NULL; 881 } 882 883 /* rcu_read_lock()ed by nf_hook_slow */ 884 l3proto = __nf_ct_l3proto_find(pf); 885 ret = l3proto->get_l4proto(skb, skb_network_offset(skb), 886 &dataoff, &protonum); 887 if (ret <= 0) { 888 pr_debug("not prepared to track yet or error occured\n"); 889 NF_CT_STAT_INC_ATOMIC(net, error); 890 NF_CT_STAT_INC_ATOMIC(net, invalid); 891 ret = -ret; 892 goto out; 893 } 894 895 l4proto = __nf_ct_l4proto_find(pf, protonum); 896 897 /* It may be an special packet, error, unclean... 898 * inverse of the return code tells to the netfilter 899 * core what to do with the packet. */ 900 if (l4proto->error != NULL) { 901 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, 902 pf, hooknum); 903 if (ret <= 0) { 904 NF_CT_STAT_INC_ATOMIC(net, error); 905 NF_CT_STAT_INC_ATOMIC(net, invalid); 906 ret = -ret; 907 goto out; 908 } 909 } 910 911 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, 912 l3proto, l4proto, &set_reply, &ctinfo); 913 if (!ct) { 914 /* Not valid part of a connection */ 915 NF_CT_STAT_INC_ATOMIC(net, invalid); 916 ret = NF_ACCEPT; 917 goto out; 918 } 919 920 if (IS_ERR(ct)) { 921 /* Too stressed to deal. */ 922 NF_CT_STAT_INC_ATOMIC(net, drop); 923 ret = NF_DROP; 924 goto out; 925 } 926 927 NF_CT_ASSERT(skb->nfct); 928 929 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum); 930 if (ret <= 0) { 931 /* Invalid: inverse of the return code tells 932 * the netfilter core what to do */ 933 pr_debug("nf_conntrack_in: Can't track with proto module\n"); 934 nf_conntrack_put(skb->nfct); 935 skb->nfct = NULL; 936 NF_CT_STAT_INC_ATOMIC(net, invalid); 937 if (ret == -NF_DROP) 938 NF_CT_STAT_INC_ATOMIC(net, drop); 939 ret = -ret; 940 goto out; 941 } 942 943 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 944 nf_conntrack_event_cache(IPCT_REPLY, ct); 945out: 946 if (tmpl) 947 nf_ct_put(tmpl); 948 949 return ret; 950} 951EXPORT_SYMBOL_GPL(nf_conntrack_in); 952 953bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 954 const struct nf_conntrack_tuple *orig) 955{ 956 bool ret; 957 958 rcu_read_lock(); 959 ret = nf_ct_invert_tuple(inverse, orig, 960 __nf_ct_l3proto_find(orig->src.l3num), 961 __nf_ct_l4proto_find(orig->src.l3num, 962 orig->dst.protonum)); 963 rcu_read_unlock(); 964 return ret; 965} 966EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); 967 968/* Alter reply tuple (maybe alter helper). This is for NAT, and is 969 implicitly racy: see __nf_conntrack_confirm */ 970void nf_conntrack_alter_reply(struct nf_conn *ct, 971 const struct nf_conntrack_tuple *newreply) 972{ 973 struct nf_conn_help *help = nfct_help(ct); 974 975 /* Should be unconfirmed, so not in hash table yet */ 976 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 977 978 pr_debug("Altering reply tuple of %p to ", ct); 979 nf_ct_dump_tuple(newreply); 980 981 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 982 if (ct->master || (help && !hlist_empty(&help->expectations))) 983 return; 984 985 rcu_read_lock(); 986 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); 987 rcu_read_unlock(); 988} 989EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); 990 991/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 992void __nf_ct_refresh_acct(struct nf_conn *ct, 993 enum ip_conntrack_info ctinfo, 994 const struct sk_buff *skb, 995 unsigned long extra_jiffies, 996 int do_acct) 997{ 998 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 999 NF_CT_ASSERT(skb); 1000 1001 /* Only update if this is not a fixed timeout */ 1002 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) 1003 goto acct; 1004 1005 /* If not in hash table, timer will not be active yet */ 1006 if (!nf_ct_is_confirmed(ct)) { 1007 ct->timeout.expires = extra_jiffies; 1008 } else { 1009 unsigned long newtime = jiffies + extra_jiffies; 1010 1011 /* Only update the timeout if the new timeout is at least 1012 HZ jiffies from the old timeout. Need del_timer for race 1013 avoidance (may already be dying). */ 1014 if (newtime - ct->timeout.expires >= HZ) 1015 mod_timer_pending(&ct->timeout, newtime); 1016 } 1017 1018acct: 1019 if (do_acct) { 1020 struct nf_conn_counter *acct; 1021 1022 acct = nf_conn_acct_find(ct); 1023 if (acct) { 1024 spin_lock_bh(&ct->lock); 1025 acct[CTINFO2DIR(ctinfo)].packets++; 1026 acct[CTINFO2DIR(ctinfo)].bytes += skb->len; 1027 spin_unlock_bh(&ct->lock); 1028 } 1029 } 1030} 1031EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); 1032 1033bool __nf_ct_kill_acct(struct nf_conn *ct, 1034 enum ip_conntrack_info ctinfo, 1035 const struct sk_buff *skb, 1036 int do_acct) 1037{ 1038 if (do_acct) { 1039 struct nf_conn_counter *acct; 1040 1041 acct = nf_conn_acct_find(ct); 1042 if (acct) { 1043 spin_lock_bh(&ct->lock); 1044 acct[CTINFO2DIR(ctinfo)].packets++; 1045 acct[CTINFO2DIR(ctinfo)].bytes += 1046 skb->len - skb_network_offset(skb); 1047 spin_unlock_bh(&ct->lock); 1048 } 1049 } 1050 1051 if (del_timer(&ct->timeout)) { 1052 ct->timeout.function((unsigned long)ct); 1053 return true; 1054 } 1055 return false; 1056} 1057EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); 1058 1059#ifdef CONFIG_NF_CONNTRACK_ZONES 1060static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { 1061 .len = sizeof(struct nf_conntrack_zone), 1062 .align = __alignof__(struct nf_conntrack_zone), 1063 .id = NF_CT_EXT_ZONE, 1064}; 1065#endif 1066 1067#if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE) 1068 1069#include <linux/netfilter/nfnetlink.h> 1070#include <linux/netfilter/nfnetlink_conntrack.h> 1071#include <linux/mutex.h> 1072 1073/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 1074 * in ip_conntrack_core, since we don't want the protocols to autoload 1075 * or depend on ctnetlink */ 1076int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, 1077 const struct nf_conntrack_tuple *tuple) 1078{ 1079 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); 1080 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); 1081 return 0; 1082 1083nla_put_failure: 1084 return -1; 1085} 1086EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); 1087 1088const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { 1089 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, 1090 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, 1091}; 1092EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); 1093 1094int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], 1095 struct nf_conntrack_tuple *t) 1096{ 1097 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) 1098 return -EINVAL; 1099 1100 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); 1101 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); 1102 1103 return 0; 1104} 1105EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); 1106 1107int nf_ct_port_nlattr_tuple_size(void) 1108{ 1109 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); 1110} 1111EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); 1112#endif 1113 1114/* Used by ipt_REJECT and ip6t_REJECT. */ 1115static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 1116{ 1117 struct nf_conn *ct; 1118 enum ip_conntrack_info ctinfo; 1119 1120 /* This ICMP is in reverse direction to the packet which caused it */ 1121 ct = nf_ct_get(skb, &ctinfo); 1122 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 1123 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY; 1124 else 1125 ctinfo = IP_CT_RELATED; 1126 1127 /* Attach to new skbuff, and increment count */ 1128 nskb->nfct = &ct->ct_general; 1129 nskb->nfctinfo = ctinfo; 1130 nf_conntrack_get(nskb->nfct); 1131} 1132 1133/* Bring out ya dead! */ 1134static struct nf_conn * 1135get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), 1136 void *data, unsigned int *bucket) 1137{ 1138 struct nf_conntrack_tuple_hash *h; 1139 struct nf_conn *ct; 1140 struct hlist_nulls_node *n; 1141 1142 spin_lock_bh(&nf_conntrack_lock); 1143 for (; *bucket < net->ct.htable_size; (*bucket)++) { 1144 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { 1145 ct = nf_ct_tuplehash_to_ctrack(h); 1146 if (iter(ct, data)) 1147 goto found; 1148 } 1149 } 1150 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { 1151 ct = nf_ct_tuplehash_to_ctrack(h); 1152 if (iter(ct, data)) 1153 set_bit(IPS_DYING_BIT, &ct->status); 1154 } 1155 spin_unlock_bh(&nf_conntrack_lock); 1156 return NULL; 1157found: 1158 atomic_inc(&ct->ct_general.use); 1159 spin_unlock_bh(&nf_conntrack_lock); 1160 return ct; 1161} 1162 1163void nf_ct_iterate_cleanup(struct net *net, 1164 int (*iter)(struct nf_conn *i, void *data), 1165 void *data) 1166{ 1167 struct nf_conn *ct; 1168 unsigned int bucket = 0; 1169 1170 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { 1171 /* Time to push up daises... */ 1172 if (del_timer(&ct->timeout)) 1173 death_by_timeout((unsigned long)ct); 1174 /* ... else the timer will get him soon. */ 1175 1176 nf_ct_put(ct); 1177 } 1178} 1179EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); 1180 1181struct __nf_ct_flush_report { 1182 u32 pid; 1183 int report; 1184}; 1185 1186static int kill_report(struct nf_conn *i, void *data) 1187{ 1188 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; 1189 1190 /* If we fail to deliver the event, death_by_timeout() will retry */ 1191 if (nf_conntrack_event_report(IPCT_DESTROY, i, 1192 fr->pid, fr->report) < 0) 1193 return 1; 1194 1195 /* Avoid the delivery of the destroy event in death_by_timeout(). */ 1196 set_bit(IPS_DYING_BIT, &i->status); 1197 return 1; 1198} 1199 1200static int kill_all(struct nf_conn *i, void *data) 1201{ 1202 return 1; 1203} 1204 1205void nf_ct_free_hashtable(void *hash, unsigned int size) 1206{ 1207 if (is_vmalloc_addr(hash)) 1208 vfree(hash); 1209 else 1210 free_pages((unsigned long)hash, 1211 get_order(sizeof(struct hlist_head) * size)); 1212} 1213EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); 1214 1215void nf_conntrack_flush_report(struct net *net, u32 pid, int report) 1216{ 1217 struct __nf_ct_flush_report fr = { 1218 .pid = pid, 1219 .report = report, 1220 }; 1221 nf_ct_iterate_cleanup(net, kill_report, &fr); 1222} 1223EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); 1224 1225static void nf_ct_release_dying_list(struct net *net) 1226{ 1227 struct nf_conntrack_tuple_hash *h; 1228 struct nf_conn *ct; 1229 struct hlist_nulls_node *n; 1230 1231 spin_lock_bh(&nf_conntrack_lock); 1232 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { 1233 ct = nf_ct_tuplehash_to_ctrack(h); 1234 /* never fails to remove them, no listeners at this point */ 1235 nf_ct_kill(ct); 1236 } 1237 spin_unlock_bh(&nf_conntrack_lock); 1238} 1239 1240static int untrack_refs(void) 1241{ 1242 int cnt = 0, cpu; 1243 1244 for_each_possible_cpu(cpu) { 1245 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1246 1247 cnt += atomic_read(&ct->ct_general.use) - 1; 1248 } 1249 return cnt; 1250} 1251 1252static void nf_conntrack_cleanup_init_net(void) 1253{ 1254 while (untrack_refs() > 0) 1255 schedule(); 1256 1257 nf_conntrack_helper_fini(); 1258 nf_conntrack_proto_fini(); 1259#ifdef CONFIG_NF_CONNTRACK_ZONES 1260 nf_ct_extend_unregister(&nf_ct_zone_extend); 1261#endif 1262} 1263 1264static void nf_conntrack_cleanup_net(struct net *net) 1265{ 1266 i_see_dead_people: 1267 nf_ct_iterate_cleanup(net, kill_all, NULL); 1268 nf_ct_release_dying_list(net); 1269 if (atomic_read(&net->ct.count) != 0) { 1270 schedule(); 1271 goto i_see_dead_people; 1272 } 1273 1274 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); 1275 nf_conntrack_ecache_fini(net); 1276 nf_conntrack_acct_fini(net); 1277 nf_conntrack_expect_fini(net); 1278 kmem_cache_destroy(net->ct.nf_conntrack_cachep); 1279 kfree(net->ct.slabname); 1280 free_percpu(net->ct.stat); 1281} 1282 1283/* Mishearing the voices in his head, our hero wonders how he's 1284 supposed to kill the mall. */ 1285void nf_conntrack_cleanup(struct net *net) 1286{ 1287 if (net_eq(net, &init_net)) 1288 rcu_assign_pointer(ip_ct_attach, NULL); 1289 1290 /* This makes sure all current packets have passed through 1291 netfilter framework. Roll on, two-stage module 1292 delete... */ 1293 synchronize_net(); 1294 1295 nf_conntrack_cleanup_net(net); 1296 1297 if (net_eq(net, &init_net)) { 1298 rcu_assign_pointer(nf_ct_destroy, NULL); 1299 nf_conntrack_cleanup_init_net(); 1300 } 1301} 1302 1303void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) 1304{ 1305 struct hlist_nulls_head *hash; 1306 unsigned int nr_slots, i; 1307 size_t sz; 1308 1309 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); 1310 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); 1311 sz = nr_slots * sizeof(struct hlist_nulls_head); 1312 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, 1313 get_order(sz)); 1314 if (!hash) { 1315 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1316 hash = __vmalloc(sz, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 1317 PAGE_KERNEL); 1318 } 1319 1320 if (hash && nulls) 1321 for (i = 0; i < nr_slots; i++) 1322 INIT_HLIST_NULLS_HEAD(&hash[i], i); 1323 1324 return hash; 1325} 1326EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); 1327 1328int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) 1329{ 1330 int i, bucket; 1331 unsigned int hashsize, old_size; 1332 struct hlist_nulls_head *hash, *old_hash; 1333 struct nf_conntrack_tuple_hash *h; 1334 struct nf_conn *ct; 1335 1336 if (current->nsproxy->net_ns != &init_net) 1337 return -EOPNOTSUPP; 1338 1339 /* On boot, we can set this without any fancy locking. */ 1340 if (!nf_conntrack_htable_size) 1341 return param_set_uint(val, kp); 1342 1343 hashsize = simple_strtoul(val, NULL, 0); 1344 if (!hashsize) 1345 return -EINVAL; 1346 1347 hash = nf_ct_alloc_hashtable(&hashsize, 1); 1348 if (!hash) 1349 return -ENOMEM; 1350 1351 /* Lookups in the old hash might happen in parallel, which means we 1352 * might get false negatives during connection lookup. New connections 1353 * created because of a false negative won't make it into the hash 1354 * though since that required taking the lock. 1355 */ 1356 spin_lock_bh(&nf_conntrack_lock); 1357 for (i = 0; i < init_net.ct.htable_size; i++) { 1358 while (!hlist_nulls_empty(&init_net.ct.hash[i])) { 1359 h = hlist_nulls_entry(init_net.ct.hash[i].first, 1360 struct nf_conntrack_tuple_hash, hnnode); 1361 ct = nf_ct_tuplehash_to_ctrack(h); 1362 hlist_nulls_del_rcu(&h->hnnode); 1363 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct), 1364 hashsize); 1365 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); 1366 } 1367 } 1368 old_size = init_net.ct.htable_size; 1369 old_hash = init_net.ct.hash; 1370 1371 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize; 1372 init_net.ct.hash = hash; 1373 spin_unlock_bh(&nf_conntrack_lock); 1374 1375 nf_ct_free_hashtable(old_hash, old_size); 1376 return 0; 1377} 1378EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); 1379 1380module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, 1381 &nf_conntrack_htable_size, 0600); 1382 1383void nf_ct_untracked_status_or(unsigned long bits) 1384{ 1385 int cpu; 1386 1387 for_each_possible_cpu(cpu) 1388 per_cpu(nf_conntrack_untracked, cpu).status |= bits; 1389} 1390EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); 1391 1392static int nf_conntrack_init_init_net(void) 1393{ 1394 int max_factor = 8; 1395 int ret, cpu; 1396 1397 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1398 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ 1399 if (!nf_conntrack_htable_size) { 1400 nf_conntrack_htable_size 1401 = (((totalram_pages << PAGE_SHIFT) / 16384) 1402 / sizeof(struct hlist_head)); 1403 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1404 nf_conntrack_htable_size = 16384; 1405 if (nf_conntrack_htable_size < 32) 1406 nf_conntrack_htable_size = 32; 1407 1408 /* Use a max. factor of four by default to get the same max as 1409 * with the old struct list_heads. When a table size is given 1410 * we use the old value of 8 to avoid reducing the max. 1411 * entries. */ 1412 max_factor = 4; 1413 } 1414 nf_conntrack_max = max_factor * nf_conntrack_htable_size; 1415 1416 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", 1417 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1418 nf_conntrack_max); 1419 1420 ret = nf_conntrack_proto_init(); 1421 if (ret < 0) 1422 goto err_proto; 1423 1424 ret = nf_conntrack_helper_init(); 1425 if (ret < 0) 1426 goto err_helper; 1427 1428#ifdef CONFIG_NF_CONNTRACK_ZONES 1429 ret = nf_ct_extend_register(&nf_ct_zone_extend); 1430 if (ret < 0) 1431 goto err_extend; 1432#endif 1433 /* Set up fake conntrack: to never be deleted, not in any hashes */ 1434 for_each_possible_cpu(cpu) { 1435 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1436 write_pnet(&ct->ct_net, &init_net); 1437 atomic_set(&ct->ct_general.use, 1); 1438 } 1439 /* - and look it like as a confirmed connection */ 1440 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); 1441 return 0; 1442 1443#ifdef CONFIG_NF_CONNTRACK_ZONES 1444err_extend: 1445 nf_conntrack_helper_fini(); 1446#endif 1447err_helper: 1448 nf_conntrack_proto_fini(); 1449err_proto: 1450 return ret; 1451} 1452 1453/* 1454 * We need to use special "null" values, not used in hash table 1455 */ 1456#define UNCONFIRMED_NULLS_VAL ((1<<30)+0) 1457#define DYING_NULLS_VAL ((1<<30)+1) 1458 1459static int nf_conntrack_init_net(struct net *net) 1460{ 1461 int ret; 1462 1463 atomic_set(&net->ct.count, 0); 1464 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); 1465 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); 1466 net->ct.stat = alloc_percpu(struct ip_conntrack_stat); 1467 if (!net->ct.stat) { 1468 ret = -ENOMEM; 1469 goto err_stat; 1470 } 1471 1472 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); 1473 if (!net->ct.slabname) { 1474 ret = -ENOMEM; 1475 goto err_slabname; 1476 } 1477 1478 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, 1479 sizeof(struct nf_conn), 0, 1480 SLAB_DESTROY_BY_RCU, NULL); 1481 if (!net->ct.nf_conntrack_cachep) { 1482 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1483 ret = -ENOMEM; 1484 goto err_cache; 1485 } 1486 1487 net->ct.htable_size = nf_conntrack_htable_size; 1488 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1); 1489 if (!net->ct.hash) { 1490 ret = -ENOMEM; 1491 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1492 goto err_hash; 1493 } 1494 ret = nf_conntrack_expect_init(net); 1495 if (ret < 0) 1496 goto err_expect; 1497 ret = nf_conntrack_acct_init(net); 1498 if (ret < 0) 1499 goto err_acct; 1500 ret = nf_conntrack_ecache_init(net); 1501 if (ret < 0) 1502 goto err_ecache; 1503 1504 return 0; 1505 1506err_ecache: 1507 nf_conntrack_acct_fini(net); 1508err_acct: 1509 nf_conntrack_expect_fini(net); 1510err_expect: 1511 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); 1512err_hash: 1513 kmem_cache_destroy(net->ct.nf_conntrack_cachep); 1514err_cache: 1515 kfree(net->ct.slabname); 1516err_slabname: 1517 free_percpu(net->ct.stat); 1518err_stat: 1519 return ret; 1520} 1521 1522s16 (*nf_ct_nat_offset)(const struct nf_conn *ct, 1523 enum ip_conntrack_dir dir, 1524 u32 seq); 1525EXPORT_SYMBOL_GPL(nf_ct_nat_offset); 1526 1527int nf_conntrack_init(struct net *net) 1528{ 1529 int ret; 1530 1531 if (net_eq(net, &init_net)) { 1532 ret = nf_conntrack_init_init_net(); 1533 if (ret < 0) 1534 goto out_init_net; 1535 } 1536 ret = nf_conntrack_init_net(net); 1537 if (ret < 0) 1538 goto out_net; 1539 1540 if (net_eq(net, &init_net)) { 1541 /* For use by REJECT target */ 1542 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach); 1543 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack); 1544 1545 /* Howto get NAT offsets */ 1546 rcu_assign_pointer(nf_ct_nat_offset, NULL); 1547 } 1548 return 0; 1549 1550out_net: 1551 if (net_eq(net, &init_net)) 1552 nf_conntrack_cleanup_init_net(); 1553out_init_net: 1554 return ret; 1555} 1556