nf_conntrack_core.c revision d96fc659aeb27686cef42d305cfd0c9702f8841c
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_conntrack_timestamp.h> 47#include <net/netfilter/nf_conntrack_timeout.h> 48#include <net/netfilter/nf_nat.h> 49#include <net/netfilter/nf_nat_core.h> 50 51#define NF_CONNTRACK_VERSION "0.5.0" 52 53int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct, 54 enum nf_nat_manip_type manip, 55 const struct nlattr *attr) __read_mostly; 56EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook); 57 58DEFINE_SPINLOCK(nf_conntrack_lock); 59EXPORT_SYMBOL_GPL(nf_conntrack_lock); 60 61unsigned int nf_conntrack_htable_size __read_mostly; 62EXPORT_SYMBOL_GPL(nf_conntrack_htable_size); 63 64unsigned int nf_conntrack_max __read_mostly; 65EXPORT_SYMBOL_GPL(nf_conntrack_max); 66 67DEFINE_PER_CPU(struct nf_conn, nf_conntrack_untracked); 68EXPORT_PER_CPU_SYMBOL(nf_conntrack_untracked); 69 70unsigned int nf_conntrack_hash_rnd __read_mostly; 71EXPORT_SYMBOL_GPL(nf_conntrack_hash_rnd); 72 73static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple, u16 zone) 74{ 75 unsigned int n; 76 77 /* The direction must be ignored, so we hash everything up to the 78 * destination ports (which is a multiple of 4) and treat the last 79 * three bytes manually. 80 */ 81 n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32); 82 return jhash2((u32 *)tuple, n, zone ^ nf_conntrack_hash_rnd ^ 83 (((__force __u16)tuple->dst.u.all << 16) | 84 tuple->dst.protonum)); 85} 86 87static u32 __hash_bucket(u32 hash, unsigned int size) 88{ 89 return ((u64)hash * size) >> 32; 90} 91 92static u32 hash_bucket(u32 hash, const struct net *net) 93{ 94 return __hash_bucket(hash, net->ct.htable_size); 95} 96 97static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple, 98 u16 zone, unsigned int size) 99{ 100 return __hash_bucket(hash_conntrack_raw(tuple, zone), size); 101} 102 103static inline u_int32_t hash_conntrack(const struct net *net, u16 zone, 104 const struct nf_conntrack_tuple *tuple) 105{ 106 return __hash_conntrack(tuple, zone, net->ct.htable_size); 107} 108 109bool 110nf_ct_get_tuple(const struct sk_buff *skb, 111 unsigned int nhoff, 112 unsigned int dataoff, 113 u_int16_t l3num, 114 u_int8_t protonum, 115 struct nf_conntrack_tuple *tuple, 116 const struct nf_conntrack_l3proto *l3proto, 117 const struct nf_conntrack_l4proto *l4proto) 118{ 119 memset(tuple, 0, sizeof(*tuple)); 120 121 tuple->src.l3num = l3num; 122 if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0) 123 return false; 124 125 tuple->dst.protonum = protonum; 126 tuple->dst.dir = IP_CT_DIR_ORIGINAL; 127 128 return l4proto->pkt_to_tuple(skb, dataoff, tuple); 129} 130EXPORT_SYMBOL_GPL(nf_ct_get_tuple); 131 132bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff, 133 u_int16_t l3num, struct nf_conntrack_tuple *tuple) 134{ 135 struct nf_conntrack_l3proto *l3proto; 136 struct nf_conntrack_l4proto *l4proto; 137 unsigned int protoff; 138 u_int8_t protonum; 139 int ret; 140 141 rcu_read_lock(); 142 143 l3proto = __nf_ct_l3proto_find(l3num); 144 ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum); 145 if (ret != NF_ACCEPT) { 146 rcu_read_unlock(); 147 return false; 148 } 149 150 l4proto = __nf_ct_l4proto_find(l3num, protonum); 151 152 ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple, 153 l3proto, l4proto); 154 155 rcu_read_unlock(); 156 return ret; 157} 158EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr); 159 160bool 161nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse, 162 const struct nf_conntrack_tuple *orig, 163 const struct nf_conntrack_l3proto *l3proto, 164 const struct nf_conntrack_l4proto *l4proto) 165{ 166 memset(inverse, 0, sizeof(*inverse)); 167 168 inverse->src.l3num = orig->src.l3num; 169 if (l3proto->invert_tuple(inverse, orig) == 0) 170 return false; 171 172 inverse->dst.dir = !orig->dst.dir; 173 174 inverse->dst.protonum = orig->dst.protonum; 175 return l4proto->invert_tuple(inverse, orig); 176} 177EXPORT_SYMBOL_GPL(nf_ct_invert_tuple); 178 179static void 180clean_from_lists(struct nf_conn *ct) 181{ 182 pr_debug("clean_from_lists(%p)\n", ct); 183 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 184 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode); 185 186 /* Destroy all pending expectations */ 187 nf_ct_remove_expectations(ct); 188} 189 190static void 191destroy_conntrack(struct nf_conntrack *nfct) 192{ 193 struct nf_conn *ct = (struct nf_conn *)nfct; 194 struct net *net = nf_ct_net(ct); 195 struct nf_conntrack_l4proto *l4proto; 196 197 pr_debug("destroy_conntrack(%p)\n", ct); 198 NF_CT_ASSERT(atomic_read(&nfct->use) == 0); 199 NF_CT_ASSERT(!timer_pending(&ct->timeout)); 200 201 /* To make sure we don't get any weird locking issues here: 202 * destroy_conntrack() MUST NOT be called with a write lock 203 * to nf_conntrack_lock!!! -HW */ 204 rcu_read_lock(); 205 l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); 206 if (l4proto && l4proto->destroy) 207 l4proto->destroy(ct); 208 209 rcu_read_unlock(); 210 211 spin_lock_bh(&nf_conntrack_lock); 212 /* Expectations will have been removed in clean_from_lists, 213 * except TFTP can create an expectation on the first packet, 214 * before connection is in the list, so we need to clean here, 215 * too. */ 216 nf_ct_remove_expectations(ct); 217 218 /* We overload first tuple to link into unconfirmed list. */ 219 if (!nf_ct_is_confirmed(ct)) { 220 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode)); 221 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 222 } 223 224 NF_CT_STAT_INC(net, delete); 225 spin_unlock_bh(&nf_conntrack_lock); 226 227 if (ct->master) 228 nf_ct_put(ct->master); 229 230 pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct); 231 nf_conntrack_free(ct); 232} 233 234void nf_ct_delete_from_lists(struct nf_conn *ct) 235{ 236 struct net *net = nf_ct_net(ct); 237 238 nf_ct_helper_destroy(ct); 239 spin_lock_bh(&nf_conntrack_lock); 240 /* Inside lock so preempt is disabled on module removal path. 241 * Otherwise we can get spurious warnings. */ 242 NF_CT_STAT_INC(net, delete_list); 243 clean_from_lists(ct); 244 spin_unlock_bh(&nf_conntrack_lock); 245} 246EXPORT_SYMBOL_GPL(nf_ct_delete_from_lists); 247 248static void death_by_event(unsigned long ul_conntrack) 249{ 250 struct nf_conn *ct = (void *)ul_conntrack; 251 struct net *net = nf_ct_net(ct); 252 253 if (nf_conntrack_event(IPCT_DESTROY, ct) < 0) { 254 /* bad luck, let's retry again */ 255 ct->timeout.expires = jiffies + 256 (random32() % net->ct.sysctl_events_retry_timeout); 257 add_timer(&ct->timeout); 258 return; 259 } 260 /* we've got the event delivered, now it's dying */ 261 set_bit(IPS_DYING_BIT, &ct->status); 262 spin_lock(&nf_conntrack_lock); 263 hlist_nulls_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 264 spin_unlock(&nf_conntrack_lock); 265 nf_ct_put(ct); 266} 267 268void nf_ct_insert_dying_list(struct nf_conn *ct) 269{ 270 struct net *net = nf_ct_net(ct); 271 272 /* add this conntrack to the dying list */ 273 spin_lock_bh(&nf_conntrack_lock); 274 hlist_nulls_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 275 &net->ct.dying); 276 spin_unlock_bh(&nf_conntrack_lock); 277 /* set a new timer to retry event delivery */ 278 setup_timer(&ct->timeout, death_by_event, (unsigned long)ct); 279 ct->timeout.expires = jiffies + 280 (random32() % net->ct.sysctl_events_retry_timeout); 281 add_timer(&ct->timeout); 282} 283EXPORT_SYMBOL_GPL(nf_ct_insert_dying_list); 284 285static void death_by_timeout(unsigned long ul_conntrack) 286{ 287 struct nf_conn *ct = (void *)ul_conntrack; 288 struct nf_conn_tstamp *tstamp; 289 290 tstamp = nf_conn_tstamp_find(ct); 291 if (tstamp && tstamp->stop == 0) 292 tstamp->stop = ktime_to_ns(ktime_get_real()); 293 294 if (!test_bit(IPS_DYING_BIT, &ct->status) && 295 unlikely(nf_conntrack_event(IPCT_DESTROY, ct) < 0)) { 296 /* destroy event was not delivered */ 297 nf_ct_delete_from_lists(ct); 298 nf_ct_insert_dying_list(ct); 299 return; 300 } 301 set_bit(IPS_DYING_BIT, &ct->status); 302 nf_ct_delete_from_lists(ct); 303 nf_ct_put(ct); 304} 305 306/* 307 * Warning : 308 * - Caller must take a reference on returned object 309 * and recheck nf_ct_tuple_equal(tuple, &h->tuple) 310 * OR 311 * - Caller must lock nf_conntrack_lock before calling this function 312 */ 313static struct nf_conntrack_tuple_hash * 314____nf_conntrack_find(struct net *net, u16 zone, 315 const struct nf_conntrack_tuple *tuple, u32 hash) 316{ 317 struct nf_conntrack_tuple_hash *h; 318 struct hlist_nulls_node *n; 319 unsigned int bucket = hash_bucket(hash, net); 320 321 /* Disable BHs the entire time since we normally need to disable them 322 * at least once for the stats anyway. 323 */ 324 local_bh_disable(); 325begin: 326 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) { 327 if (nf_ct_tuple_equal(tuple, &h->tuple) && 328 nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) { 329 NF_CT_STAT_INC(net, found); 330 local_bh_enable(); 331 return h; 332 } 333 NF_CT_STAT_INC(net, searched); 334 } 335 /* 336 * if the nulls value we got at the end of this lookup is 337 * not the expected one, we must restart lookup. 338 * We probably met an item that was moved to another chain. 339 */ 340 if (get_nulls_value(n) != bucket) { 341 NF_CT_STAT_INC(net, search_restart); 342 goto begin; 343 } 344 local_bh_enable(); 345 346 return NULL; 347} 348 349struct nf_conntrack_tuple_hash * 350__nf_conntrack_find(struct net *net, u16 zone, 351 const struct nf_conntrack_tuple *tuple) 352{ 353 return ____nf_conntrack_find(net, zone, tuple, 354 hash_conntrack_raw(tuple, zone)); 355} 356EXPORT_SYMBOL_GPL(__nf_conntrack_find); 357 358/* Find a connection corresponding to a tuple. */ 359static struct nf_conntrack_tuple_hash * 360__nf_conntrack_find_get(struct net *net, u16 zone, 361 const struct nf_conntrack_tuple *tuple, u32 hash) 362{ 363 struct nf_conntrack_tuple_hash *h; 364 struct nf_conn *ct; 365 366 rcu_read_lock(); 367begin: 368 h = ____nf_conntrack_find(net, zone, tuple, hash); 369 if (h) { 370 ct = nf_ct_tuplehash_to_ctrack(h); 371 if (unlikely(nf_ct_is_dying(ct) || 372 !atomic_inc_not_zero(&ct->ct_general.use))) 373 h = NULL; 374 else { 375 if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) || 376 nf_ct_zone(ct) != zone)) { 377 nf_ct_put(ct); 378 goto begin; 379 } 380 } 381 } 382 rcu_read_unlock(); 383 384 return h; 385} 386 387struct nf_conntrack_tuple_hash * 388nf_conntrack_find_get(struct net *net, u16 zone, 389 const struct nf_conntrack_tuple *tuple) 390{ 391 return __nf_conntrack_find_get(net, zone, tuple, 392 hash_conntrack_raw(tuple, zone)); 393} 394EXPORT_SYMBOL_GPL(nf_conntrack_find_get); 395 396static void __nf_conntrack_hash_insert(struct nf_conn *ct, 397 unsigned int hash, 398 unsigned int repl_hash) 399{ 400 struct net *net = nf_ct_net(ct); 401 402 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 403 &net->ct.hash[hash]); 404 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode, 405 &net->ct.hash[repl_hash]); 406} 407 408int 409nf_conntrack_hash_check_insert(struct nf_conn *ct) 410{ 411 struct net *net = nf_ct_net(ct); 412 unsigned int hash, repl_hash; 413 struct nf_conntrack_tuple_hash *h; 414 struct hlist_nulls_node *n; 415 u16 zone; 416 417 zone = nf_ct_zone(ct); 418 hash = hash_conntrack(net, zone, 419 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple); 420 repl_hash = hash_conntrack(net, zone, 421 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 422 423 spin_lock_bh(&nf_conntrack_lock); 424 425 /* See if there's one in the list already, including reverse */ 426 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) 427 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 428 &h->tuple) && 429 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 430 goto out; 431 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) 432 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 433 &h->tuple) && 434 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 435 goto out; 436 437 add_timer(&ct->timeout); 438 nf_conntrack_get(&ct->ct_general); 439 __nf_conntrack_hash_insert(ct, hash, repl_hash); 440 NF_CT_STAT_INC(net, insert); 441 spin_unlock_bh(&nf_conntrack_lock); 442 443 return 0; 444 445out: 446 NF_CT_STAT_INC(net, insert_failed); 447 spin_unlock_bh(&nf_conntrack_lock); 448 return -EEXIST; 449} 450EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert); 451 452/* Confirm a connection given skb; places it in hash table */ 453int 454__nf_conntrack_confirm(struct sk_buff *skb) 455{ 456 unsigned int hash, repl_hash; 457 struct nf_conntrack_tuple_hash *h; 458 struct nf_conn *ct; 459 struct nf_conn_help *help; 460 struct nf_conn_tstamp *tstamp; 461 struct hlist_nulls_node *n; 462 enum ip_conntrack_info ctinfo; 463 struct net *net; 464 u16 zone; 465 466 ct = nf_ct_get(skb, &ctinfo); 467 net = nf_ct_net(ct); 468 469 /* ipt_REJECT uses nf_conntrack_attach to attach related 470 ICMP/TCP RST packets in other direction. Actual packet 471 which created connection will be IP_CT_NEW or for an 472 expected connection, IP_CT_RELATED. */ 473 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) 474 return NF_ACCEPT; 475 476 zone = nf_ct_zone(ct); 477 /* reuse the hash saved before */ 478 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev; 479 hash = hash_bucket(hash, net); 480 repl_hash = hash_conntrack(net, zone, 481 &ct->tuplehash[IP_CT_DIR_REPLY].tuple); 482 483 /* We're not in hash table, and we refuse to set up related 484 connections for unconfirmed conns. But packet copies and 485 REJECT will give spurious warnings here. */ 486 /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */ 487 488 /* No external references means no one else could have 489 confirmed us. */ 490 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 491 pr_debug("Confirming conntrack %p\n", ct); 492 493 spin_lock_bh(&nf_conntrack_lock); 494 495 /* We have to check the DYING flag inside the lock to prevent 496 a race against nf_ct_get_next_corpse() possibly called from 497 user context, else we insert an already 'dead' hash, blocking 498 further use of that particular connection -JM */ 499 500 if (unlikely(nf_ct_is_dying(ct))) { 501 spin_unlock_bh(&nf_conntrack_lock); 502 return NF_ACCEPT; 503 } 504 505 /* See if there's one in the list already, including reverse: 506 NAT could have grabbed it without realizing, since we're 507 not in the hash. If there is, we lost race. */ 508 hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode) 509 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, 510 &h->tuple) && 511 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 512 goto out; 513 hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode) 514 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple, 515 &h->tuple) && 516 zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h))) 517 goto out; 518 519 /* Remove from unconfirmed list */ 520 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode); 521 522 /* Timer relative to confirmation time, not original 523 setting time, otherwise we'd get timer wrap in 524 weird delay cases. */ 525 ct->timeout.expires += jiffies; 526 add_timer(&ct->timeout); 527 atomic_inc(&ct->ct_general.use); 528 ct->status |= IPS_CONFIRMED; 529 530 /* set conntrack timestamp, if enabled. */ 531 tstamp = nf_conn_tstamp_find(ct); 532 if (tstamp) { 533 if (skb->tstamp.tv64 == 0) 534 __net_timestamp((struct sk_buff *)skb); 535 536 tstamp->start = ktime_to_ns(skb->tstamp); 537 } 538 /* Since the lookup is lockless, hash insertion must be done after 539 * starting the timer and setting the CONFIRMED bit. The RCU barriers 540 * guarantee that no other CPU can find the conntrack before the above 541 * stores are visible. 542 */ 543 __nf_conntrack_hash_insert(ct, hash, repl_hash); 544 NF_CT_STAT_INC(net, insert); 545 spin_unlock_bh(&nf_conntrack_lock); 546 547 help = nfct_help(ct); 548 if (help && help->helper) 549 nf_conntrack_event_cache(IPCT_HELPER, ct); 550 551 nf_conntrack_event_cache(master_ct(ct) ? 552 IPCT_RELATED : IPCT_NEW, ct); 553 return NF_ACCEPT; 554 555out: 556 NF_CT_STAT_INC(net, insert_failed); 557 spin_unlock_bh(&nf_conntrack_lock); 558 return NF_DROP; 559} 560EXPORT_SYMBOL_GPL(__nf_conntrack_confirm); 561 562/* Returns true if a connection correspondings to the tuple (required 563 for NAT). */ 564int 565nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple, 566 const struct nf_conn *ignored_conntrack) 567{ 568 struct net *net = nf_ct_net(ignored_conntrack); 569 struct nf_conntrack_tuple_hash *h; 570 struct hlist_nulls_node *n; 571 struct nf_conn *ct; 572 u16 zone = nf_ct_zone(ignored_conntrack); 573 unsigned int hash = hash_conntrack(net, zone, tuple); 574 575 /* Disable BHs the entire time since we need to disable them at 576 * least once for the stats anyway. 577 */ 578 rcu_read_lock_bh(); 579 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) { 580 ct = nf_ct_tuplehash_to_ctrack(h); 581 if (ct != ignored_conntrack && 582 nf_ct_tuple_equal(tuple, &h->tuple) && 583 nf_ct_zone(ct) == zone) { 584 NF_CT_STAT_INC(net, found); 585 rcu_read_unlock_bh(); 586 return 1; 587 } 588 NF_CT_STAT_INC(net, searched); 589 } 590 rcu_read_unlock_bh(); 591 592 return 0; 593} 594EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken); 595 596#define NF_CT_EVICTION_RANGE 8 597 598/* There's a small race here where we may free a just-assured 599 connection. Too bad: we're in trouble anyway. */ 600static noinline int early_drop(struct net *net, unsigned int hash) 601{ 602 /* Use oldest entry, which is roughly LRU */ 603 struct nf_conntrack_tuple_hash *h; 604 struct nf_conn *ct = NULL, *tmp; 605 struct hlist_nulls_node *n; 606 unsigned int i, cnt = 0; 607 int dropped = 0; 608 609 rcu_read_lock(); 610 for (i = 0; i < net->ct.htable_size; i++) { 611 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], 612 hnnode) { 613 tmp = nf_ct_tuplehash_to_ctrack(h); 614 if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) 615 ct = tmp; 616 cnt++; 617 } 618 619 if (ct != NULL) { 620 if (likely(!nf_ct_is_dying(ct) && 621 atomic_inc_not_zero(&ct->ct_general.use))) 622 break; 623 else 624 ct = NULL; 625 } 626 627 if (cnt >= NF_CT_EVICTION_RANGE) 628 break; 629 630 hash = (hash + 1) % net->ct.htable_size; 631 } 632 rcu_read_unlock(); 633 634 if (!ct) 635 return dropped; 636 637 if (del_timer(&ct->timeout)) { 638 death_by_timeout((unsigned long)ct); 639 /* Check if we indeed killed this entry. Reliable event 640 delivery may have inserted it into the dying list. */ 641 if (test_bit(IPS_DYING_BIT, &ct->status)) { 642 dropped = 1; 643 NF_CT_STAT_INC_ATOMIC(net, early_drop); 644 } 645 } 646 nf_ct_put(ct); 647 return dropped; 648} 649 650void init_nf_conntrack_hash_rnd(void) 651{ 652 unsigned int rand; 653 654 /* 655 * Why not initialize nf_conntrack_rnd in a "init()" function ? 656 * Because there isn't enough entropy when system initializing, 657 * and we initialize it as late as possible. 658 */ 659 do { 660 get_random_bytes(&rand, sizeof(rand)); 661 } while (!rand); 662 cmpxchg(&nf_conntrack_hash_rnd, 0, rand); 663} 664 665static struct nf_conn * 666__nf_conntrack_alloc(struct net *net, u16 zone, 667 const struct nf_conntrack_tuple *orig, 668 const struct nf_conntrack_tuple *repl, 669 gfp_t gfp, u32 hash) 670{ 671 struct nf_conn *ct; 672 673 if (unlikely(!nf_conntrack_hash_rnd)) { 674 init_nf_conntrack_hash_rnd(); 675 /* recompute the hash as nf_conntrack_hash_rnd is initialized */ 676 hash = hash_conntrack_raw(orig, zone); 677 } 678 679 /* We don't want any race condition at early drop stage */ 680 atomic_inc(&net->ct.count); 681 682 if (nf_conntrack_max && 683 unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) { 684 if (!early_drop(net, hash_bucket(hash, net))) { 685 atomic_dec(&net->ct.count); 686 if (net_ratelimit()) 687 printk(KERN_WARNING 688 "nf_conntrack: table full, dropping" 689 " packet.\n"); 690 return ERR_PTR(-ENOMEM); 691 } 692 } 693 694 /* 695 * Do not use kmem_cache_zalloc(), as this cache uses 696 * SLAB_DESTROY_BY_RCU. 697 */ 698 ct = kmem_cache_alloc(net->ct.nf_conntrack_cachep, gfp); 699 if (ct == NULL) { 700 atomic_dec(&net->ct.count); 701 return ERR_PTR(-ENOMEM); 702 } 703 /* 704 * Let ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.next 705 * and ct->tuplehash[IP_CT_DIR_REPLY].hnnode.next unchanged. 706 */ 707 memset(&ct->tuplehash[IP_CT_DIR_MAX], 0, 708 offsetof(struct nf_conn, proto) - 709 offsetof(struct nf_conn, tuplehash[IP_CT_DIR_MAX])); 710 spin_lock_init(&ct->lock); 711 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig; 712 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL; 713 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl; 714 /* save hash for reusing when confirming */ 715 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash; 716 /* Don't set timer yet: wait for confirmation */ 717 setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct); 718 write_pnet(&ct->ct_net, net); 719#ifdef CONFIG_NF_CONNTRACK_ZONES 720 if (zone) { 721 struct nf_conntrack_zone *nf_ct_zone; 722 723 nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC); 724 if (!nf_ct_zone) 725 goto out_free; 726 nf_ct_zone->id = zone; 727 } 728#endif 729 /* 730 * changes to lookup keys must be done before setting refcnt to 1 731 */ 732 smp_wmb(); 733 atomic_set(&ct->ct_general.use, 1); 734 return ct; 735 736#ifdef CONFIG_NF_CONNTRACK_ZONES 737out_free: 738 atomic_dec(&net->ct.count); 739 kmem_cache_free(net->ct.nf_conntrack_cachep, ct); 740 return ERR_PTR(-ENOMEM); 741#endif 742} 743 744struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone, 745 const struct nf_conntrack_tuple *orig, 746 const struct nf_conntrack_tuple *repl, 747 gfp_t gfp) 748{ 749 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0); 750} 751EXPORT_SYMBOL_GPL(nf_conntrack_alloc); 752 753void nf_conntrack_free(struct nf_conn *ct) 754{ 755 struct net *net = nf_ct_net(ct); 756 757 nf_ct_ext_destroy(ct); 758 atomic_dec(&net->ct.count); 759 nf_ct_ext_free(ct); 760 kmem_cache_free(net->ct.nf_conntrack_cachep, ct); 761} 762EXPORT_SYMBOL_GPL(nf_conntrack_free); 763 764/* Allocate a new conntrack: we return -ENOMEM if classification 765 failed due to stress. Otherwise it really is unclassifiable. */ 766static struct nf_conntrack_tuple_hash * 767init_conntrack(struct net *net, struct nf_conn *tmpl, 768 const struct nf_conntrack_tuple *tuple, 769 struct nf_conntrack_l3proto *l3proto, 770 struct nf_conntrack_l4proto *l4proto, 771 struct sk_buff *skb, 772 unsigned int dataoff, u32 hash) 773{ 774 struct nf_conn *ct; 775 struct nf_conn_help *help; 776 struct nf_conntrack_tuple repl_tuple; 777 struct nf_conntrack_ecache *ecache; 778 struct nf_conntrack_expect *exp; 779 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; 780 struct nf_conn_timeout *timeout_ext; 781 unsigned int *timeouts; 782 783 if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) { 784 pr_debug("Can't invert tuple.\n"); 785 return NULL; 786 } 787 788 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC, 789 hash); 790 if (IS_ERR(ct)) 791 return (struct nf_conntrack_tuple_hash *)ct; 792 793 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL; 794 if (timeout_ext) 795 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext); 796 else 797 timeouts = l4proto->get_timeouts(net); 798 799 if (!l4proto->new(ct, skb, dataoff, timeouts)) { 800 nf_conntrack_free(ct); 801 pr_debug("init conntrack: can't track with proto module\n"); 802 return NULL; 803 } 804 805 if (timeout_ext) 806 nf_ct_timeout_ext_add(ct, timeout_ext->timeout, GFP_ATOMIC); 807 808 nf_ct_acct_ext_add(ct, GFP_ATOMIC); 809 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC); 810 811 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL; 812 nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0, 813 ecache ? ecache->expmask : 0, 814 GFP_ATOMIC); 815 816 spin_lock_bh(&nf_conntrack_lock); 817 exp = nf_ct_find_expectation(net, zone, tuple); 818 if (exp) { 819 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n", 820 ct, exp); 821 /* Welcome, Mr. Bond. We've been expecting you... */ 822 __set_bit(IPS_EXPECTED_BIT, &ct->status); 823 ct->master = exp->master; 824 if (exp->helper) { 825 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC); 826 if (help) 827 rcu_assign_pointer(help->helper, exp->helper); 828 } 829 830#ifdef CONFIG_NF_CONNTRACK_MARK 831 ct->mark = exp->master->mark; 832#endif 833#ifdef CONFIG_NF_CONNTRACK_SECMARK 834 ct->secmark = exp->master->secmark; 835#endif 836 nf_conntrack_get(&ct->master->ct_general); 837 NF_CT_STAT_INC(net, expect_new); 838 } else { 839 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC); 840 NF_CT_STAT_INC(net, new); 841 } 842 843 /* Overload tuple linked list to put us in unconfirmed list. */ 844 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode, 845 &net->ct.unconfirmed); 846 847 spin_unlock_bh(&nf_conntrack_lock); 848 849 if (exp) { 850 if (exp->expectfn) 851 exp->expectfn(ct, exp); 852 nf_ct_expect_put(exp); 853 } 854 855 return &ct->tuplehash[IP_CT_DIR_ORIGINAL]; 856} 857 858/* On success, returns conntrack ptr, sets skb->nfct and ctinfo */ 859static inline struct nf_conn * 860resolve_normal_ct(struct net *net, struct nf_conn *tmpl, 861 struct sk_buff *skb, 862 unsigned int dataoff, 863 u_int16_t l3num, 864 u_int8_t protonum, 865 struct nf_conntrack_l3proto *l3proto, 866 struct nf_conntrack_l4proto *l4proto, 867 int *set_reply, 868 enum ip_conntrack_info *ctinfo) 869{ 870 struct nf_conntrack_tuple tuple; 871 struct nf_conntrack_tuple_hash *h; 872 struct nf_conn *ct; 873 u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE; 874 u32 hash; 875 876 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), 877 dataoff, l3num, protonum, &tuple, l3proto, 878 l4proto)) { 879 pr_debug("resolve_normal_ct: Can't get tuple\n"); 880 return NULL; 881 } 882 883 /* look for tuple match */ 884 hash = hash_conntrack_raw(&tuple, zone); 885 h = __nf_conntrack_find_get(net, zone, &tuple, hash); 886 if (!h) { 887 h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto, 888 skb, dataoff, hash); 889 if (!h) 890 return NULL; 891 if (IS_ERR(h)) 892 return (void *)h; 893 } 894 ct = nf_ct_tuplehash_to_ctrack(h); 895 896 /* It exists; we have (non-exclusive) reference. */ 897 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) { 898 *ctinfo = IP_CT_ESTABLISHED_REPLY; 899 /* Please set reply bit if this packet OK */ 900 *set_reply = 1; 901 } else { 902 /* Once we've had two way comms, always ESTABLISHED. */ 903 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) { 904 pr_debug("nf_conntrack_in: normal packet for %p\n", ct); 905 *ctinfo = IP_CT_ESTABLISHED; 906 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) { 907 pr_debug("nf_conntrack_in: related packet for %p\n", 908 ct); 909 *ctinfo = IP_CT_RELATED; 910 } else { 911 pr_debug("nf_conntrack_in: new packet for %p\n", ct); 912 *ctinfo = IP_CT_NEW; 913 } 914 *set_reply = 0; 915 } 916 skb->nfct = &ct->ct_general; 917 skb->nfctinfo = *ctinfo; 918 return ct; 919} 920 921unsigned int 922nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum, 923 struct sk_buff *skb) 924{ 925 struct nf_conn *ct, *tmpl = NULL; 926 enum ip_conntrack_info ctinfo; 927 struct nf_conntrack_l3proto *l3proto; 928 struct nf_conntrack_l4proto *l4proto; 929 struct nf_conn_timeout *timeout_ext; 930 unsigned int *timeouts; 931 unsigned int dataoff; 932 u_int8_t protonum; 933 int set_reply = 0; 934 int ret; 935 936 if (skb->nfct) { 937 /* Previously seen (loopback or untracked)? Ignore. */ 938 tmpl = (struct nf_conn *)skb->nfct; 939 if (!nf_ct_is_template(tmpl)) { 940 NF_CT_STAT_INC_ATOMIC(net, ignore); 941 return NF_ACCEPT; 942 } 943 skb->nfct = NULL; 944 } 945 946 /* rcu_read_lock()ed by nf_hook_slow */ 947 l3proto = __nf_ct_l3proto_find(pf); 948 ret = l3proto->get_l4proto(skb, skb_network_offset(skb), 949 &dataoff, &protonum); 950 if (ret <= 0) { 951 pr_debug("not prepared to track yet or error occurred\n"); 952 NF_CT_STAT_INC_ATOMIC(net, error); 953 NF_CT_STAT_INC_ATOMIC(net, invalid); 954 ret = -ret; 955 goto out; 956 } 957 958 l4proto = __nf_ct_l4proto_find(pf, protonum); 959 960 /* It may be an special packet, error, unclean... 961 * inverse of the return code tells to the netfilter 962 * core what to do with the packet. */ 963 if (l4proto->error != NULL) { 964 ret = l4proto->error(net, tmpl, skb, dataoff, &ctinfo, 965 pf, hooknum); 966 if (ret <= 0) { 967 NF_CT_STAT_INC_ATOMIC(net, error); 968 NF_CT_STAT_INC_ATOMIC(net, invalid); 969 ret = -ret; 970 goto out; 971 } 972 /* ICMP[v6] protocol trackers may assign one conntrack. */ 973 if (skb->nfct) 974 goto out; 975 } 976 977 ct = resolve_normal_ct(net, tmpl, skb, dataoff, pf, protonum, 978 l3proto, l4proto, &set_reply, &ctinfo); 979 if (!ct) { 980 /* Not valid part of a connection */ 981 NF_CT_STAT_INC_ATOMIC(net, invalid); 982 ret = NF_ACCEPT; 983 goto out; 984 } 985 986 if (IS_ERR(ct)) { 987 /* Too stressed to deal. */ 988 NF_CT_STAT_INC_ATOMIC(net, drop); 989 ret = NF_DROP; 990 goto out; 991 } 992 993 NF_CT_ASSERT(skb->nfct); 994 995 /* Decide what timeout policy we want to apply to this flow. */ 996 timeout_ext = nf_ct_timeout_find(ct); 997 if (timeout_ext) 998 timeouts = NF_CT_TIMEOUT_EXT_DATA(timeout_ext); 999 else 1000 timeouts = l4proto->get_timeouts(net); 1001 1002 ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum, timeouts); 1003 if (ret <= 0) { 1004 /* Invalid: inverse of the return code tells 1005 * the netfilter core what to do */ 1006 pr_debug("nf_conntrack_in: Can't track with proto module\n"); 1007 nf_conntrack_put(skb->nfct); 1008 skb->nfct = NULL; 1009 NF_CT_STAT_INC_ATOMIC(net, invalid); 1010 if (ret == -NF_DROP) 1011 NF_CT_STAT_INC_ATOMIC(net, drop); 1012 ret = -ret; 1013 goto out; 1014 } 1015 1016 if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status)) 1017 nf_conntrack_event_cache(IPCT_REPLY, ct); 1018out: 1019 if (tmpl) { 1020 /* Special case: we have to repeat this hook, assign the 1021 * template again to this packet. We assume that this packet 1022 * has no conntrack assigned. This is used by nf_ct_tcp. */ 1023 if (ret == NF_REPEAT) 1024 skb->nfct = (struct nf_conntrack *)tmpl; 1025 else 1026 nf_ct_put(tmpl); 1027 } 1028 1029 return ret; 1030} 1031EXPORT_SYMBOL_GPL(nf_conntrack_in); 1032 1033bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse, 1034 const struct nf_conntrack_tuple *orig) 1035{ 1036 bool ret; 1037 1038 rcu_read_lock(); 1039 ret = nf_ct_invert_tuple(inverse, orig, 1040 __nf_ct_l3proto_find(orig->src.l3num), 1041 __nf_ct_l4proto_find(orig->src.l3num, 1042 orig->dst.protonum)); 1043 rcu_read_unlock(); 1044 return ret; 1045} 1046EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr); 1047 1048/* Alter reply tuple (maybe alter helper). This is for NAT, and is 1049 implicitly racy: see __nf_conntrack_confirm */ 1050void nf_conntrack_alter_reply(struct nf_conn *ct, 1051 const struct nf_conntrack_tuple *newreply) 1052{ 1053 struct nf_conn_help *help = nfct_help(ct); 1054 1055 /* Should be unconfirmed, so not in hash table yet */ 1056 NF_CT_ASSERT(!nf_ct_is_confirmed(ct)); 1057 1058 pr_debug("Altering reply tuple of %p to ", ct); 1059 nf_ct_dump_tuple(newreply); 1060 1061 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply; 1062 if (ct->master || (help && !hlist_empty(&help->expectations))) 1063 return; 1064 1065 rcu_read_lock(); 1066 __nf_ct_try_assign_helper(ct, NULL, GFP_ATOMIC); 1067 rcu_read_unlock(); 1068} 1069EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply); 1070 1071/* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */ 1072void __nf_ct_refresh_acct(struct nf_conn *ct, 1073 enum ip_conntrack_info ctinfo, 1074 const struct sk_buff *skb, 1075 unsigned long extra_jiffies, 1076 int do_acct) 1077{ 1078 NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct); 1079 NF_CT_ASSERT(skb); 1080 1081 /* Only update if this is not a fixed timeout */ 1082 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) 1083 goto acct; 1084 1085 /* If not in hash table, timer will not be active yet */ 1086 if (!nf_ct_is_confirmed(ct)) { 1087 ct->timeout.expires = extra_jiffies; 1088 } else { 1089 unsigned long newtime = jiffies + extra_jiffies; 1090 1091 /* Only update the timeout if the new timeout is at least 1092 HZ jiffies from the old timeout. Need del_timer for race 1093 avoidance (may already be dying). */ 1094 if (newtime - ct->timeout.expires >= HZ) 1095 mod_timer_pending(&ct->timeout, newtime); 1096 } 1097 1098acct: 1099 if (do_acct) { 1100 struct nf_conn_counter *acct; 1101 1102 acct = nf_conn_acct_find(ct); 1103 if (acct) { 1104 atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets); 1105 atomic64_add(skb->len, &acct[CTINFO2DIR(ctinfo)].bytes); 1106 } 1107 } 1108} 1109EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct); 1110 1111bool __nf_ct_kill_acct(struct nf_conn *ct, 1112 enum ip_conntrack_info ctinfo, 1113 const struct sk_buff *skb, 1114 int do_acct) 1115{ 1116 if (do_acct) { 1117 struct nf_conn_counter *acct; 1118 1119 acct = nf_conn_acct_find(ct); 1120 if (acct) { 1121 atomic64_inc(&acct[CTINFO2DIR(ctinfo)].packets); 1122 atomic64_add(skb->len - skb_network_offset(skb), 1123 &acct[CTINFO2DIR(ctinfo)].bytes); 1124 } 1125 } 1126 1127 if (del_timer(&ct->timeout)) { 1128 ct->timeout.function((unsigned long)ct); 1129 return true; 1130 } 1131 return false; 1132} 1133EXPORT_SYMBOL_GPL(__nf_ct_kill_acct); 1134 1135#ifdef CONFIG_NF_CONNTRACK_ZONES 1136static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = { 1137 .len = sizeof(struct nf_conntrack_zone), 1138 .align = __alignof__(struct nf_conntrack_zone), 1139 .id = NF_CT_EXT_ZONE, 1140}; 1141#endif 1142 1143#if IS_ENABLED(CONFIG_NF_CT_NETLINK) 1144 1145#include <linux/netfilter/nfnetlink.h> 1146#include <linux/netfilter/nfnetlink_conntrack.h> 1147#include <linux/mutex.h> 1148 1149/* Generic function for tcp/udp/sctp/dccp and alike. This needs to be 1150 * in ip_conntrack_core, since we don't want the protocols to autoload 1151 * or depend on ctnetlink */ 1152int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb, 1153 const struct nf_conntrack_tuple *tuple) 1154{ 1155 NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port); 1156 NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port); 1157 return 0; 1158 1159nla_put_failure: 1160 return -1; 1161} 1162EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr); 1163 1164const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = { 1165 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 }, 1166 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 }, 1167}; 1168EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy); 1169 1170int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[], 1171 struct nf_conntrack_tuple *t) 1172{ 1173 if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT]) 1174 return -EINVAL; 1175 1176 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]); 1177 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]); 1178 1179 return 0; 1180} 1181EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple); 1182 1183int nf_ct_port_nlattr_tuple_size(void) 1184{ 1185 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1); 1186} 1187EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size); 1188#endif 1189 1190/* Used by ipt_REJECT and ip6t_REJECT. */ 1191static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb) 1192{ 1193 struct nf_conn *ct; 1194 enum ip_conntrack_info ctinfo; 1195 1196 /* This ICMP is in reverse direction to the packet which caused it */ 1197 ct = nf_ct_get(skb, &ctinfo); 1198 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) 1199 ctinfo = IP_CT_RELATED_REPLY; 1200 else 1201 ctinfo = IP_CT_RELATED; 1202 1203 /* Attach to new skbuff, and increment count */ 1204 nskb->nfct = &ct->ct_general; 1205 nskb->nfctinfo = ctinfo; 1206 nf_conntrack_get(nskb->nfct); 1207} 1208 1209/* Bring out ya dead! */ 1210static struct nf_conn * 1211get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data), 1212 void *data, unsigned int *bucket) 1213{ 1214 struct nf_conntrack_tuple_hash *h; 1215 struct nf_conn *ct; 1216 struct hlist_nulls_node *n; 1217 1218 spin_lock_bh(&nf_conntrack_lock); 1219 for (; *bucket < net->ct.htable_size; (*bucket)++) { 1220 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) { 1221 ct = nf_ct_tuplehash_to_ctrack(h); 1222 if (iter(ct, data)) 1223 goto found; 1224 } 1225 } 1226 hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) { 1227 ct = nf_ct_tuplehash_to_ctrack(h); 1228 if (iter(ct, data)) 1229 set_bit(IPS_DYING_BIT, &ct->status); 1230 } 1231 spin_unlock_bh(&nf_conntrack_lock); 1232 return NULL; 1233found: 1234 atomic_inc(&ct->ct_general.use); 1235 spin_unlock_bh(&nf_conntrack_lock); 1236 return ct; 1237} 1238 1239void nf_ct_iterate_cleanup(struct net *net, 1240 int (*iter)(struct nf_conn *i, void *data), 1241 void *data) 1242{ 1243 struct nf_conn *ct; 1244 unsigned int bucket = 0; 1245 1246 while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) { 1247 /* Time to push up daises... */ 1248 if (del_timer(&ct->timeout)) 1249 death_by_timeout((unsigned long)ct); 1250 /* ... else the timer will get him soon. */ 1251 1252 nf_ct_put(ct); 1253 } 1254} 1255EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup); 1256 1257struct __nf_ct_flush_report { 1258 u32 pid; 1259 int report; 1260}; 1261 1262static int kill_report(struct nf_conn *i, void *data) 1263{ 1264 struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data; 1265 struct nf_conn_tstamp *tstamp; 1266 1267 tstamp = nf_conn_tstamp_find(i); 1268 if (tstamp && tstamp->stop == 0) 1269 tstamp->stop = ktime_to_ns(ktime_get_real()); 1270 1271 /* If we fail to deliver the event, death_by_timeout() will retry */ 1272 if (nf_conntrack_event_report(IPCT_DESTROY, i, 1273 fr->pid, fr->report) < 0) 1274 return 1; 1275 1276 /* Avoid the delivery of the destroy event in death_by_timeout(). */ 1277 set_bit(IPS_DYING_BIT, &i->status); 1278 return 1; 1279} 1280 1281static int kill_all(struct nf_conn *i, void *data) 1282{ 1283 return 1; 1284} 1285 1286void nf_ct_free_hashtable(void *hash, unsigned int size) 1287{ 1288 if (is_vmalloc_addr(hash)) 1289 vfree(hash); 1290 else 1291 free_pages((unsigned long)hash, 1292 get_order(sizeof(struct hlist_head) * size)); 1293} 1294EXPORT_SYMBOL_GPL(nf_ct_free_hashtable); 1295 1296void nf_conntrack_flush_report(struct net *net, u32 pid, int report) 1297{ 1298 struct __nf_ct_flush_report fr = { 1299 .pid = pid, 1300 .report = report, 1301 }; 1302 nf_ct_iterate_cleanup(net, kill_report, &fr); 1303} 1304EXPORT_SYMBOL_GPL(nf_conntrack_flush_report); 1305 1306static void nf_ct_release_dying_list(struct net *net) 1307{ 1308 struct nf_conntrack_tuple_hash *h; 1309 struct nf_conn *ct; 1310 struct hlist_nulls_node *n; 1311 1312 spin_lock_bh(&nf_conntrack_lock); 1313 hlist_nulls_for_each_entry(h, n, &net->ct.dying, hnnode) { 1314 ct = nf_ct_tuplehash_to_ctrack(h); 1315 /* never fails to remove them, no listeners at this point */ 1316 nf_ct_kill(ct); 1317 } 1318 spin_unlock_bh(&nf_conntrack_lock); 1319} 1320 1321static int untrack_refs(void) 1322{ 1323 int cnt = 0, cpu; 1324 1325 for_each_possible_cpu(cpu) { 1326 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1327 1328 cnt += atomic_read(&ct->ct_general.use) - 1; 1329 } 1330 return cnt; 1331} 1332 1333static void nf_conntrack_cleanup_init_net(void) 1334{ 1335 while (untrack_refs() > 0) 1336 schedule(); 1337 1338 nf_conntrack_helper_fini(); 1339 nf_conntrack_proto_fini(); 1340#ifdef CONFIG_NF_CONNTRACK_ZONES 1341 nf_ct_extend_unregister(&nf_ct_zone_extend); 1342#endif 1343} 1344 1345static void nf_conntrack_cleanup_net(struct net *net) 1346{ 1347 i_see_dead_people: 1348 nf_ct_iterate_cleanup(net, kill_all, NULL); 1349 nf_ct_release_dying_list(net); 1350 if (atomic_read(&net->ct.count) != 0) { 1351 schedule(); 1352 goto i_see_dead_people; 1353 } 1354 1355 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); 1356 nf_conntrack_timeout_fini(net); 1357 nf_conntrack_ecache_fini(net); 1358 nf_conntrack_tstamp_fini(net); 1359 nf_conntrack_acct_fini(net); 1360 nf_conntrack_expect_fini(net); 1361 kmem_cache_destroy(net->ct.nf_conntrack_cachep); 1362 kfree(net->ct.slabname); 1363 free_percpu(net->ct.stat); 1364} 1365 1366/* Mishearing the voices in his head, our hero wonders how he's 1367 supposed to kill the mall. */ 1368void nf_conntrack_cleanup(struct net *net) 1369{ 1370 if (net_eq(net, &init_net)) 1371 RCU_INIT_POINTER(ip_ct_attach, NULL); 1372 1373 /* This makes sure all current packets have passed through 1374 netfilter framework. Roll on, two-stage module 1375 delete... */ 1376 synchronize_net(); 1377 1378 nf_conntrack_cleanup_net(net); 1379 1380 if (net_eq(net, &init_net)) { 1381 RCU_INIT_POINTER(nf_ct_destroy, NULL); 1382 nf_conntrack_cleanup_init_net(); 1383 } 1384} 1385 1386void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls) 1387{ 1388 struct hlist_nulls_head *hash; 1389 unsigned int nr_slots, i; 1390 size_t sz; 1391 1392 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head)); 1393 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head)); 1394 sz = nr_slots * sizeof(struct hlist_nulls_head); 1395 hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO, 1396 get_order(sz)); 1397 if (!hash) { 1398 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n"); 1399 hash = vzalloc(sz); 1400 } 1401 1402 if (hash && nulls) 1403 for (i = 0; i < nr_slots; i++) 1404 INIT_HLIST_NULLS_HEAD(&hash[i], i); 1405 1406 return hash; 1407} 1408EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable); 1409 1410int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp) 1411{ 1412 int i, bucket; 1413 unsigned int hashsize, old_size; 1414 struct hlist_nulls_head *hash, *old_hash; 1415 struct nf_conntrack_tuple_hash *h; 1416 struct nf_conn *ct; 1417 1418 if (current->nsproxy->net_ns != &init_net) 1419 return -EOPNOTSUPP; 1420 1421 /* On boot, we can set this without any fancy locking. */ 1422 if (!nf_conntrack_htable_size) 1423 return param_set_uint(val, kp); 1424 1425 hashsize = simple_strtoul(val, NULL, 0); 1426 if (!hashsize) 1427 return -EINVAL; 1428 1429 hash = nf_ct_alloc_hashtable(&hashsize, 1); 1430 if (!hash) 1431 return -ENOMEM; 1432 1433 /* Lookups in the old hash might happen in parallel, which means we 1434 * might get false negatives during connection lookup. New connections 1435 * created because of a false negative won't make it into the hash 1436 * though since that required taking the lock. 1437 */ 1438 spin_lock_bh(&nf_conntrack_lock); 1439 for (i = 0; i < init_net.ct.htable_size; i++) { 1440 while (!hlist_nulls_empty(&init_net.ct.hash[i])) { 1441 h = hlist_nulls_entry(init_net.ct.hash[i].first, 1442 struct nf_conntrack_tuple_hash, hnnode); 1443 ct = nf_ct_tuplehash_to_ctrack(h); 1444 hlist_nulls_del_rcu(&h->hnnode); 1445 bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct), 1446 hashsize); 1447 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]); 1448 } 1449 } 1450 old_size = init_net.ct.htable_size; 1451 old_hash = init_net.ct.hash; 1452 1453 init_net.ct.htable_size = nf_conntrack_htable_size = hashsize; 1454 init_net.ct.hash = hash; 1455 spin_unlock_bh(&nf_conntrack_lock); 1456 1457 nf_ct_free_hashtable(old_hash, old_size); 1458 return 0; 1459} 1460EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize); 1461 1462module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint, 1463 &nf_conntrack_htable_size, 0600); 1464 1465void nf_ct_untracked_status_or(unsigned long bits) 1466{ 1467 int cpu; 1468 1469 for_each_possible_cpu(cpu) 1470 per_cpu(nf_conntrack_untracked, cpu).status |= bits; 1471} 1472EXPORT_SYMBOL_GPL(nf_ct_untracked_status_or); 1473 1474static int nf_conntrack_init_init_net(void) 1475{ 1476 int max_factor = 8; 1477 int ret, cpu; 1478 1479 /* Idea from tcp.c: use 1/16384 of memory. On i386: 32MB 1480 * machine has 512 buckets. >= 1GB machines have 16384 buckets. */ 1481 if (!nf_conntrack_htable_size) { 1482 nf_conntrack_htable_size 1483 = (((totalram_pages << PAGE_SHIFT) / 16384) 1484 / sizeof(struct hlist_head)); 1485 if (totalram_pages > (1024 * 1024 * 1024 / PAGE_SIZE)) 1486 nf_conntrack_htable_size = 16384; 1487 if (nf_conntrack_htable_size < 32) 1488 nf_conntrack_htable_size = 32; 1489 1490 /* Use a max. factor of four by default to get the same max as 1491 * with the old struct list_heads. When a table size is given 1492 * we use the old value of 8 to avoid reducing the max. 1493 * entries. */ 1494 max_factor = 4; 1495 } 1496 nf_conntrack_max = max_factor * nf_conntrack_htable_size; 1497 1498 printk(KERN_INFO "nf_conntrack version %s (%u buckets, %d max)\n", 1499 NF_CONNTRACK_VERSION, nf_conntrack_htable_size, 1500 nf_conntrack_max); 1501 1502 ret = nf_conntrack_proto_init(); 1503 if (ret < 0) 1504 goto err_proto; 1505 1506 ret = nf_conntrack_helper_init(); 1507 if (ret < 0) 1508 goto err_helper; 1509 1510#ifdef CONFIG_NF_CONNTRACK_ZONES 1511 ret = nf_ct_extend_register(&nf_ct_zone_extend); 1512 if (ret < 0) 1513 goto err_extend; 1514#endif 1515 /* Set up fake conntrack: to never be deleted, not in any hashes */ 1516 for_each_possible_cpu(cpu) { 1517 struct nf_conn *ct = &per_cpu(nf_conntrack_untracked, cpu); 1518 write_pnet(&ct->ct_net, &init_net); 1519 atomic_set(&ct->ct_general.use, 1); 1520 } 1521 /* - and look it like as a confirmed connection */ 1522 nf_ct_untracked_status_or(IPS_CONFIRMED | IPS_UNTRACKED); 1523 return 0; 1524 1525#ifdef CONFIG_NF_CONNTRACK_ZONES 1526err_extend: 1527 nf_conntrack_helper_fini(); 1528#endif 1529err_helper: 1530 nf_conntrack_proto_fini(); 1531err_proto: 1532 return ret; 1533} 1534 1535/* 1536 * We need to use special "null" values, not used in hash table 1537 */ 1538#define UNCONFIRMED_NULLS_VAL ((1<<30)+0) 1539#define DYING_NULLS_VAL ((1<<30)+1) 1540 1541static int nf_conntrack_init_net(struct net *net) 1542{ 1543 int ret; 1544 1545 atomic_set(&net->ct.count, 0); 1546 INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL); 1547 INIT_HLIST_NULLS_HEAD(&net->ct.dying, DYING_NULLS_VAL); 1548 net->ct.stat = alloc_percpu(struct ip_conntrack_stat); 1549 if (!net->ct.stat) { 1550 ret = -ENOMEM; 1551 goto err_stat; 1552 } 1553 1554 net->ct.slabname = kasprintf(GFP_KERNEL, "nf_conntrack_%p", net); 1555 if (!net->ct.slabname) { 1556 ret = -ENOMEM; 1557 goto err_slabname; 1558 } 1559 1560 net->ct.nf_conntrack_cachep = kmem_cache_create(net->ct.slabname, 1561 sizeof(struct nf_conn), 0, 1562 SLAB_DESTROY_BY_RCU, NULL); 1563 if (!net->ct.nf_conntrack_cachep) { 1564 printk(KERN_ERR "Unable to create nf_conn slab cache\n"); 1565 ret = -ENOMEM; 1566 goto err_cache; 1567 } 1568 1569 net->ct.htable_size = nf_conntrack_htable_size; 1570 net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1); 1571 if (!net->ct.hash) { 1572 ret = -ENOMEM; 1573 printk(KERN_ERR "Unable to create nf_conntrack_hash\n"); 1574 goto err_hash; 1575 } 1576 ret = nf_conntrack_expect_init(net); 1577 if (ret < 0) 1578 goto err_expect; 1579 ret = nf_conntrack_acct_init(net); 1580 if (ret < 0) 1581 goto err_acct; 1582 ret = nf_conntrack_tstamp_init(net); 1583 if (ret < 0) 1584 goto err_tstamp; 1585 ret = nf_conntrack_ecache_init(net); 1586 if (ret < 0) 1587 goto err_ecache; 1588 ret = nf_conntrack_timeout_init(net); 1589 if (ret < 0) 1590 goto err_timeout; 1591 1592 return 0; 1593 1594err_timeout: 1595 nf_conntrack_timeout_fini(net); 1596err_ecache: 1597 nf_conntrack_tstamp_fini(net); 1598err_tstamp: 1599 nf_conntrack_acct_fini(net); 1600err_acct: 1601 nf_conntrack_expect_fini(net); 1602err_expect: 1603 nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size); 1604err_hash: 1605 kmem_cache_destroy(net->ct.nf_conntrack_cachep); 1606err_cache: 1607 kfree(net->ct.slabname); 1608err_slabname: 1609 free_percpu(net->ct.stat); 1610err_stat: 1611 return ret; 1612} 1613 1614s16 (*nf_ct_nat_offset)(const struct nf_conn *ct, 1615 enum ip_conntrack_dir dir, 1616 u32 seq); 1617EXPORT_SYMBOL_GPL(nf_ct_nat_offset); 1618 1619int nf_conntrack_init(struct net *net) 1620{ 1621 int ret; 1622 1623 if (net_eq(net, &init_net)) { 1624 ret = nf_conntrack_init_init_net(); 1625 if (ret < 0) 1626 goto out_init_net; 1627 } 1628 ret = nf_conntrack_init_net(net); 1629 if (ret < 0) 1630 goto out_net; 1631 1632 if (net_eq(net, &init_net)) { 1633 /* For use by REJECT target */ 1634 RCU_INIT_POINTER(ip_ct_attach, nf_conntrack_attach); 1635 RCU_INIT_POINTER(nf_ct_destroy, destroy_conntrack); 1636 1637 /* Howto get NAT offsets */ 1638 RCU_INIT_POINTER(nf_ct_nat_offset, NULL); 1639 } 1640 return 0; 1641 1642out_net: 1643 if (net_eq(net, &init_net)) 1644 nf_conntrack_cleanup_init_net(); 1645out_init_net: 1646 return ret; 1647} 1648