net/netfilter/nf_conntrack_core.c

   1 /* Connection state tracking for netfilter.  This is separated from,
   2    but required by, the NAT layer; it can also be used by an iptables
   3    extension. */
   4
   5 /* (C) 1999-2001 Paul `Rusty' Russell
   6  * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
   7  * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License version 2 as
  11  * published by the Free Software Foundation.
  12  */
  13
  14 #include <linux/types.h>
  15 #include <linux/netfilter.h>
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/proc_fs.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/stddef.h>
  21 #include <linux/slab.h>
  22 #include <linux/random.h>
  23 #include <linux/jhash.h>
  24 #include <linux/err.h>
  25 #include <linux/percpu.h>
  26 #include <linux/moduleparam.h>
  27 #include <linux/notifier.h>
  28 #include <linux/kernel.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/socket.h>
  31 #include <linux/mm.h>
  32
  33 #include <net/netfilter/nf_conntrack.h>
  34 #include <net/netfilter/nf_conntrack_l3proto.h>
  35 #include <net/netfilter/nf_conntrack_l4proto.h>
  36 #include <net/netfilter/nf_conntrack_expect.h>
  37 #include <net/netfilter/nf_conntrack_helper.h>
  38 #include <net/netfilter/nf_conntrack_core.h>
  39 #include <net/netfilter/nf_conntrack_extend.h>
  40
  41 #define NF_CONNTRACK_VERSION    "0.5.0"
  42
  43 DEFINE_SPINLOCK(nf_conntrack_lock);
  44 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  45
  46 /* nf_conntrack_standalone needs this */
  47 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
  48 EXPORT_SYMBOL_GPL(nf_conntrack_count);
  49
  50 unsigned int nf_conntrack_htable_size __read_mostly;
  51 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  52
  53 int nf_conntrack_max __read_mostly;
  54 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  55
  56 struct hlist_head *nf_conntrack_hash __read_mostly;
  57 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  58
  59 struct nf_conn nf_conntrack_untracked __read_mostly;
  60 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  61
  62 unsigned int nf_ct_log_invalid __read_mostly;
  63 HLIST_HEAD(unconfirmed);
  64 static int nf_conntrack_vmalloc __read_mostly;
  65 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  66
  67 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
  68 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
  69
  70 static int nf_conntrack_hash_rnd_initted;
  71 static unsigned int nf_conntrack_hash_rnd;
  72
  73 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  74                                   unsigned int size, unsigned int rnd)
  75 {
  76         unsigned int a, b;
  77
  78         a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
  79                    (tuple->src.l3num << 16) | tuple->dst.protonum);
  80         b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
  81                    ((__force __u16)tuple->src.u.all << 16) |
  82                     (__force __u16)tuple->dst.u.all);
  83
  84         return ((u64)jhash_2words(a, b, rnd) * size) >> 32;
  85 }
  86
  87 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  88 {
  89         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  90                                 nf_conntrack_hash_rnd);
  91 }
  92
  93 int
  94 nf_ct_get_tuple(const struct sk_buff *skb,
  95                 unsigned int nhoff,
  96                 unsigned int dataoff,
  97                 u_int16_t l3num,
  98                 u_int8_t protonum,
  99                 struct nf_conntrack_tuple *tuple,
 100                 const struct nf_conntrack_l3proto *l3proto,
 101                 const struct nf_conntrack_l4proto *l4proto)
 102 {
 103         NF_CT_TUPLE_U_BLANK(tuple);
 104
 105         tuple->src.l3num = l3num;
 106         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 107                 return 0;
 108
 109         tuple->dst.protonum = protonum;
 110         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 111
 112         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 113 }
 114 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 115
 116 int nf_ct_get_tuplepr(const struct sk_buff *skb,
 117                       unsigned int nhoff,
 118                       u_int16_t l3num,
 119                       struct nf_conntrack_tuple *tuple)
 120 {
 121         struct nf_conntrack_l3proto *l3proto;
 122         struct nf_conntrack_l4proto *l4proto;
 123         unsigned int protoff;
 124         u_int8_t protonum;
 125         int ret;
 126
 127         rcu_read_lock();
 128
 129         l3proto = __nf_ct_l3proto_find(l3num);
 130         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 131         if (ret != NF_ACCEPT) {
 132                 rcu_read_unlock();
 133                 return 0;
 134         }
 135
 136         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 137
 138         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 139                               l3proto, l4proto);
 140
 141         rcu_read_unlock();
 142         return ret;
 143 }
 144 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 145
 146 int
 147 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 148                    const struct nf_conntrack_tuple *orig,
 149                    const struct nf_conntrack_l3proto *l3proto,
 150                    const struct nf_conntrack_l4proto *l4proto)
 151 {
 152         NF_CT_TUPLE_U_BLANK(inverse);
 153
 154         inverse->src.l3num = orig->src.l3num;
 155         if (l3proto->invert_tuple(inverse, orig) == 0)
 156                 return 0;
 157
 158         inverse->dst.dir = !orig->dst.dir;
 159
 160         inverse->dst.protonum = orig->dst.protonum;
 161         return l4proto->invert_tuple(inverse, orig);
 162 }
 163 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 164
 165 static void
 166 clean_from_lists(struct nf_conn *ct)
 167 {
 168         pr_debug("clean_from_lists(%p)\n", ct);
 169         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 170         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 171
 172         /* Destroy all pending expectations */
 173         nf_ct_remove_expectations(ct);
 174 }
 175
 176 static void
 177 destroy_conntrack(struct nf_conntrack *nfct)
 178 {
 179         struct nf_conn *ct = (struct nf_conn *)nfct;
 180         struct nf_conntrack_l4proto *l4proto;
 181
 182         pr_debug("destroy_conntrack(%p)\n", ct);
 183         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 184         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 185
 186         nf_conntrack_event(IPCT_DESTROY, ct);
 187         set_bit(IPS_DYING_BIT, &ct->status);
 188
 189         /* To make sure we don't get any weird locking issues here:
 190          * destroy_conntrack() MUST NOT be called with a write lock
 191          * to nf_conntrack_lock!!! -HW */
 192         rcu_read_lock();
 193         l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
 194                                        ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
 195         if (l4proto && l4proto->destroy)
 196                 l4proto->destroy(ct);
 197
 198         nf_ct_ext_destroy(ct);
 199
 200         rcu_read_unlock();
 201
 202         spin_lock_bh(&nf_conntrack_lock);
 203         /* Expectations will have been removed in clean_from_lists,
 204          * except TFTP can create an expectation on the first packet,
 205          * before connection is in the list, so we need to clean here,
 206          * too. */
 207         nf_ct_remove_expectations(ct);
 208
 209         /* We overload first tuple to link into unconfirmed list. */
 210         if (!nf_ct_is_confirmed(ct)) {
 211                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
 212                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 213         }
 214
 215         NF_CT_STAT_INC(delete);
 216         spin_unlock_bh(&nf_conntrack_lock);
 217
 218         if (ct->master)
 219                 nf_ct_put(ct->master);
 220
 221         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 222         nf_conntrack_free(ct);
 223 }
 224
 225 static void death_by_timeout(unsigned long ul_conntrack)
 226 {
 227         struct nf_conn *ct = (void *)ul_conntrack;
 228         struct nf_conn_help *help = nfct_help(ct);
 229         struct nf_conntrack_helper *helper;
 230
 231         if (help) {
 232                 rcu_read_lock();
 233                 helper = rcu_dereference(help->helper);
 234                 if (helper && helper->destroy)
 235                         helper->destroy(ct);
 236                 rcu_read_unlock();
 237         }
 238
 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(delete_list);
 243         clean_from_lists(ct);
 244         spin_unlock_bh(&nf_conntrack_lock);
 245         nf_ct_put(ct);
 246 }
 247
 248 struct nf_conntrack_tuple_hash *
 249 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
 250 {
 251         struct nf_conntrack_tuple_hash *h;
 252         struct hlist_node *n;
 253         unsigned int hash = hash_conntrack(tuple);
 254
 255         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 256                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 257                         NF_CT_STAT_INC(found);
 258                         return h;
 259                 }
 260                 NF_CT_STAT_INC(searched);
 261         }
 262
 263         return NULL;
 264 }
 265 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 266
 267 /* Find a connection corresponding to a tuple. */
 268 struct nf_conntrack_tuple_hash *
 269 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
 270 {
 271         struct nf_conntrack_tuple_hash *h;
 272         struct nf_conn *ct;
 273
 274         rcu_read_lock();
 275         h = __nf_conntrack_find(tuple);
 276         if (h) {
 277                 ct = nf_ct_tuplehash_to_ctrack(h);
 278                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 279                         h = NULL;
 280         }
 281         rcu_read_unlock();
 282
 283         return h;
 284 }
 285 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 286
 287 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 288                                        unsigned int hash,
 289                                        unsigned int repl_hash)
 290 {
 291         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 292                            &nf_conntrack_hash[hash]);
 293         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 294                            &nf_conntrack_hash[repl_hash]);
 295 }
 296
 297 void nf_conntrack_hash_insert(struct nf_conn *ct)
 298 {
 299         unsigned int hash, repl_hash;
 300
 301         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 302         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 303
 304         spin_lock_bh(&nf_conntrack_lock);
 305         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 306         spin_unlock_bh(&nf_conntrack_lock);
 307 }
 308 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 309
 310 /* Confirm a connection given skb; places it in hash table */
 311 int
 312 __nf_conntrack_confirm(struct sk_buff *skb)
 313 {
 314         unsigned int hash, repl_hash;
 315         struct nf_conntrack_tuple_hash *h;
 316         struct nf_conn *ct;
 317         struct nf_conn_help *help;
 318         struct hlist_node *n;
 319         enum ip_conntrack_info ctinfo;
 320
 321         ct = nf_ct_get(skb, &ctinfo);
 322
 323         /* ipt_REJECT uses nf_conntrack_attach to attach related
 324            ICMP/TCP RST packets in other direction.  Actual packet
 325            which created connection will be IP_CT_NEW or for an
 326            expected connection, IP_CT_RELATED. */
 327         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 328                 return NF_ACCEPT;
 329
 330         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 331         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 332
 333         /* We're not in hash table, and we refuse to set up related
 334            connections for unconfirmed conns.  But packet copies and
 335            REJECT will give spurious warnings here. */
 336         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 337
 338         /* No external references means noone else could have
 339            confirmed us. */
 340         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 341         pr_debug("Confirming conntrack %p\n", ct);
 342
 343         spin_lock_bh(&nf_conntrack_lock);
 344
 345         /* See if there's one in the list already, including reverse:
 346            NAT could have grabbed it without realizing, since we're
 347            not in the hash.  If there is, we lost race. */
 348         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
 349                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 350                                       &h->tuple))
 351                         goto out;
 352         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
 353                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 354                                       &h->tuple))
 355                         goto out;
 356
 357         /* Remove from unconfirmed list */
 358         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 359
 360         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 361         /* Timer relative to confirmation time, not original
 362            setting time, otherwise we'd get timer wrap in
 363            weird delay cases. */
 364         ct->timeout.expires += jiffies;
 365         add_timer(&ct->timeout);
 366         atomic_inc(&ct->ct_general.use);
 367         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 368         NF_CT_STAT_INC(insert);
 369         spin_unlock_bh(&nf_conntrack_lock);
 370         help = nfct_help(ct);
 371         if (help && help->helper)
 372                 nf_conntrack_event_cache(IPCT_HELPER, skb);
 373 #ifdef CONFIG_NF_NAT_NEEDED
 374         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 375             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 376                 nf_conntrack_event_cache(IPCT_NATINFO, skb);
 377 #endif
 378         nf_conntrack_event_cache(master_ct(ct) ?
 379                                  IPCT_RELATED : IPCT_NEW, skb);
 380         return NF_ACCEPT;
 381
 382 out:
 383         NF_CT_STAT_INC(insert_failed);
 384         spin_unlock_bh(&nf_conntrack_lock);
 385         return NF_DROP;
 386 }
 387 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 388
 389 /* Returns true if a connection correspondings to the tuple (required
 390    for NAT). */
 391 int
 392 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 393                          const struct nf_conn *ignored_conntrack)
 394 {
 395         struct nf_conntrack_tuple_hash *h;
 396         struct hlist_node *n;
 397         unsigned int hash = hash_conntrack(tuple);
 398
 399         rcu_read_lock();
 400         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 401                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 402                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 403                         NF_CT_STAT_INC(found);
 404                         rcu_read_unlock();
 405                         return 1;
 406                 }
 407                 NF_CT_STAT_INC(searched);
 408         }
 409         rcu_read_unlock();
 410
 411         return 0;
 412 }
 413 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 414
 415 #define NF_CT_EVICTION_RANGE    8
 416
 417 /* There's a small race here where we may free a just-assured
 418    connection.  Too bad: we're in trouble anyway. */
 419 static int early_drop(unsigned int hash)
 420 {
 421         /* Use oldest entry, which is roughly LRU */
 422         struct nf_conntrack_tuple_hash *h;
 423         struct nf_conn *ct = NULL, *tmp;
 424         struct hlist_node *n;
 425         unsigned int i, cnt = 0;
 426         int dropped = 0;
 427
 428         rcu_read_lock();
 429         for (i = 0; i < nf_conntrack_htable_size; i++) {
 430                 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash],
 431                                          hnode) {
 432                         tmp = nf_ct_tuplehash_to_ctrack(h);
 433                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 434                                 ct = tmp;
 435                         cnt++;
 436                 }
 437
 438                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 439                         ct = NULL;
 440                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 441                         break;
 442                 hash = (hash + 1) % nf_conntrack_htable_size;
 443         }
 444         rcu_read_unlock();
 445
 446         if (!ct)
 447                 return dropped;
 448
 449         if (del_timer(&ct->timeout)) {
 450                 death_by_timeout((unsigned long)ct);
 451                 dropped = 1;
 452                 NF_CT_STAT_INC_ATOMIC(early_drop);
 453         }
 454         nf_ct_put(ct);
 455         return dropped;
 456 }
 457
 458 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 459                                    const struct nf_conntrack_tuple *repl)
 460 {
 461         struct nf_conn *conntrack = NULL;
 462
 463         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 464                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 465                 nf_conntrack_hash_rnd_initted = 1;
 466         }
 467
 468         /* We don't want any race condition at early drop stage */
 469         atomic_inc(&nf_conntrack_count);
 470
 471         if (nf_conntrack_max
 472             && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
 473                 unsigned int hash = hash_conntrack(orig);
 474                 if (!early_drop(hash)) {
 475                         atomic_dec(&nf_conntrack_count);
 476                         if (net_ratelimit())
 477                                 printk(KERN_WARNING
 478                                        "nf_conntrack: table full, dropping"
 479                                        " packet.\n");
 480                         return ERR_PTR(-ENOMEM);
 481                 }
 482         }
 483
 484         conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
 485         if (conntrack == NULL) {
 486                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 487                 atomic_dec(&nf_conntrack_count);
 488                 return ERR_PTR(-ENOMEM);
 489         }
 490
 491         atomic_set(&conntrack->ct_general.use, 1);
 492         conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 493         conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 494         /* Don't set timer yet: wait for confirmation */
 495         setup_timer(&conntrack->timeout, death_by_timeout,
 496                     (unsigned long)conntrack);
 497         INIT_RCU_HEAD(&conntrack->rcu);
 498
 499         return conntrack;
 500 }
 501 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 502
 503 static void nf_conntrack_free_rcu(struct rcu_head *head)
 504 {
 505         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
 506
 507         nf_ct_ext_free(ct);
 508         kmem_cache_free(nf_conntrack_cachep, ct);
 509         atomic_dec(&nf_conntrack_count);
 510 }
 511
 512 void nf_conntrack_free(struct nf_conn *conntrack)
 513 {
 514         call_rcu(&conntrack->rcu, nf_conntrack_free_rcu);
 515 }
 516 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 517
 518 /* Allocate a new conntrack: we return -ENOMEM if classification
 519    failed due to stress.  Otherwise it really is unclassifiable. */
 520 static struct nf_conntrack_tuple_hash *
 521 init_conntrack(const struct nf_conntrack_tuple *tuple,
 522                struct nf_conntrack_l3proto *l3proto,
 523                struct nf_conntrack_l4proto *l4proto,
 524                struct sk_buff *skb,
 525                unsigned int dataoff)
 526 {
 527         struct nf_conn *conntrack;
 528         struct nf_conn_help *help;
 529         struct nf_conntrack_tuple repl_tuple;
 530         struct nf_conntrack_expect *exp;
 531
 532         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 533                 pr_debug("Can't invert tuple.\n");
 534                 return NULL;
 535         }
 536
 537         conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
 538         if (conntrack == NULL || IS_ERR(conntrack)) {
 539                 pr_debug("Can't allocate conntrack.\n");
 540                 return (struct nf_conntrack_tuple_hash *)conntrack;
 541         }
 542
 543         if (!l4proto->new(conntrack, skb, dataoff)) {
 544                 nf_conntrack_free(conntrack);
 545                 pr_debug("init conntrack: can't track with proto module\n");
 546                 return NULL;
 547         }
 548
 549         spin_lock_bh(&nf_conntrack_lock);
 550         exp = nf_ct_find_expectation(tuple);
 551         if (exp) {
 552                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 553                          conntrack, exp);
 554                 /* Welcome, Mr. Bond.  We've been expecting you... */
 555                 __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
 556                 conntrack->master = exp->master;
 557                 if (exp->helper) {
 558                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
 559                         if (help)
 560                                 rcu_assign_pointer(help->helper, exp->helper);
 561                 }
 562
 563 #ifdef CONFIG_NF_CONNTRACK_MARK
 564                 conntrack->mark = exp->master->mark;
 565 #endif
 566 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 567                 conntrack->secmark = exp->master->secmark;
 568 #endif
 569                 nf_conntrack_get(&conntrack->master->ct_general);
 570                 NF_CT_STAT_INC(expect_new);
 571         } else {
 572                 struct nf_conntrack_helper *helper;
 573
 574                 helper = __nf_ct_helper_find(&repl_tuple);
 575                 if (helper) {
 576                         help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
 577                         if (help)
 578                                 rcu_assign_pointer(help->helper, helper);
 579                 }
 580                 NF_CT_STAT_INC(new);
 581         }
 582
 583         /* Overload tuple linked list to put us in unconfirmed list. */
 584         hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 585                        &unconfirmed);
 586
 587         spin_unlock_bh(&nf_conntrack_lock);
 588
 589         if (exp) {
 590                 if (exp->expectfn)
 591                         exp->expectfn(conntrack, exp);
 592                 nf_ct_expect_put(exp);
 593         }
 594
 595         return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
 596 }
 597
 598 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 599 static inline struct nf_conn *
 600 resolve_normal_ct(struct sk_buff *skb,
 601                   unsigned int dataoff,
 602                   u_int16_t l3num,
 603                   u_int8_t protonum,
 604                   struct nf_conntrack_l3proto *l3proto,
 605                   struct nf_conntrack_l4proto *l4proto,
 606                   int *set_reply,
 607                   enum ip_conntrack_info *ctinfo)
 608 {
 609         struct nf_conntrack_tuple tuple;
 610         struct nf_conntrack_tuple_hash *h;
 611         struct nf_conn *ct;
 612
 613         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 614                              dataoff, l3num, protonum, &tuple, l3proto,
 615                              l4proto)) {
 616                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 617                 return NULL;
 618         }
 619
 620         /* look for tuple match */
 621         h = nf_conntrack_find_get(&tuple);
 622         if (!h) {
 623                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
 624                 if (!h)
 625                         return NULL;
 626                 if (IS_ERR(h))
 627                         return (void *)h;
 628         }
 629         ct = nf_ct_tuplehash_to_ctrack(h);
 630
 631         /* It exists; we have (non-exclusive) reference. */
 632         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 633                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 634                 /* Please set reply bit if this packet OK */
 635                 *set_reply = 1;
 636         } else {
 637                 /* Once we've had two way comms, always ESTABLISHED. */
 638                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 639                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 640                         *ctinfo = IP_CT_ESTABLISHED;
 641                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 642                         pr_debug("nf_conntrack_in: related packet for %p\n",
 643                                  ct);
 644                         *ctinfo = IP_CT_RELATED;
 645                 } else {
 646                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 647                         *ctinfo = IP_CT_NEW;
 648                 }
 649                 *set_reply = 0;
 650         }
 651         skb->nfct = &ct->ct_general;
 652         skb->nfctinfo = *ctinfo;
 653         return ct;
 654 }
 655
 656 unsigned int
 657 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
 658 {
 659         struct nf_conn *ct;
 660         enum ip_conntrack_info ctinfo;
 661         struct nf_conntrack_l3proto *l3proto;
 662         struct nf_conntrack_l4proto *l4proto;
 663         unsigned int dataoff;
 664         u_int8_t protonum;
 665         int set_reply = 0;
 666         int ret;
 667
 668         /* Previously seen (loopback or untracked)?  Ignore. */
 669         if (skb->nfct) {
 670                 NF_CT_STAT_INC_ATOMIC(ignore);
 671                 return NF_ACCEPT;
 672         }
 673
 674         /* rcu_read_lock()ed by nf_hook_slow */
 675         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 676         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 677                                    &dataoff, &protonum);
 678         if (ret <= 0) {
 679                 pr_debug("not prepared to track yet or error occured\n");
 680                 NF_CT_STAT_INC_ATOMIC(error);
 681                 NF_CT_STAT_INC_ATOMIC(invalid);
 682                 return -ret;
 683         }
 684
 685         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
 686
 687         /* It may be an special packet, error, unclean...
 688          * inverse of the return code tells to the netfilter
 689          * core what to do with the packet. */
 690         if (l4proto->error != NULL &&
 691             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 692                 NF_CT_STAT_INC_ATOMIC(error);
 693                 NF_CT_STAT_INC_ATOMIC(invalid);
 694                 return -ret;
 695         }
 696
 697         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
 698                                &set_reply, &ctinfo);
 699         if (!ct) {
 700                 /* Not valid part of a connection */
 701                 NF_CT_STAT_INC_ATOMIC(invalid);
 702                 return NF_ACCEPT;
 703         }
 704
 705         if (IS_ERR(ct)) {
 706                 /* Too stressed to deal. */
 707                 NF_CT_STAT_INC_ATOMIC(drop);
 708                 return NF_DROP;
 709         }
 710
 711         NF_CT_ASSERT(skb->nfct);
 712
 713         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 714         if (ret < 0) {
 715                 /* Invalid: inverse of the return code tells
 716                  * the netfilter core what to do */
 717                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 718                 nf_conntrack_put(skb->nfct);
 719                 skb->nfct = NULL;
 720                 NF_CT_STAT_INC_ATOMIC(invalid);
 721                 return -ret;
 722         }
 723
 724         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 725                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 726
 727         return ret;
 728 }
 729 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 730
 731 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 732                          const struct nf_conntrack_tuple *orig)
 733 {
 734         int ret;
 735
 736         rcu_read_lock();
 737         ret = nf_ct_invert_tuple(inverse, orig,
 738                                  __nf_ct_l3proto_find(orig->src.l3num),
 739                                  __nf_ct_l4proto_find(orig->src.l3num,
 740                                                       orig->dst.protonum));
 741         rcu_read_unlock();
 742         return ret;
 743 }
 744 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 745
 746 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 747    implicitly racy: see __nf_conntrack_confirm */
 748 void nf_conntrack_alter_reply(struct nf_conn *ct,
 749                               const struct nf_conntrack_tuple *newreply)
 750 {
 751         struct nf_conn_help *help = nfct_help(ct);
 752         struct nf_conntrack_helper *helper;
 753
 754         /* Should be unconfirmed, so not in hash table yet */
 755         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 756
 757         pr_debug("Altering reply tuple of %p to ", ct);
 758         NF_CT_DUMP_TUPLE(newreply);
 759
 760         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 761         if (ct->master || (help && help->expecting != 0))
 762                 return;
 763
 764         rcu_read_lock();
 765         helper = __nf_ct_helper_find(newreply);
 766         if (helper == NULL) {
 767                 if (help)
 768                         rcu_assign_pointer(help->helper, NULL);
 769                 goto out;
 770         }
 771
 772         if (help == NULL) {
 773                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 774                 if (help == NULL)
 775                         goto out;
 776         } else {
 777                 memset(&help->help, 0, sizeof(help->help));
 778         }
 779
 780         rcu_assign_pointer(help->helper, helper);
 781 out:
 782         rcu_read_unlock();
 783 }
 784 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 785
 786 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 787 void __nf_ct_refresh_acct(struct nf_conn *ct,
 788                           enum ip_conntrack_info ctinfo,
 789                           const struct sk_buff *skb,
 790                           unsigned long extra_jiffies,
 791                           int do_acct)
 792 {
 793         int event = 0;
 794
 795         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 796         NF_CT_ASSERT(skb);
 797
 798         spin_lock_bh(&nf_conntrack_lock);
 799
 800         /* Only update if this is not a fixed timeout */
 801         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 802                 goto acct;
 803
 804         /* If not in hash table, timer will not be active yet */
 805         if (!nf_ct_is_confirmed(ct)) {
 806                 ct->timeout.expires = extra_jiffies;
 807                 event = IPCT_REFRESH;
 808         } else {
 809                 unsigned long newtime = jiffies + extra_jiffies;
 810
 811                 /* Only update the timeout if the new timeout is at least
 812                    HZ jiffies from the old timeout. Need del_timer for race
 813                    avoidance (may already be dying). */
 814                 if (newtime - ct->timeout.expires >= HZ
 815                     && del_timer(&ct->timeout)) {
 816                         ct->timeout.expires = newtime;
 817                         add_timer(&ct->timeout);
 818                         event = IPCT_REFRESH;
 819                 }
 820         }
 821
 822 acct:
 823 #ifdef CONFIG_NF_CT_ACCT
 824         if (do_acct) {
 825                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
 826                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
 827                         skb->len - skb_network_offset(skb);
 828
 829                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
 830                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
 831                         event |= IPCT_COUNTER_FILLING;
 832         }
 833 #endif
 834
 835         spin_unlock_bh(&nf_conntrack_lock);
 836
 837         /* must be unlocked when calling event cache */
 838         if (event)
 839                 nf_conntrack_event_cache(event, skb);
 840 }
 841 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 842
 843 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 844
 845 #include <linux/netfilter/nfnetlink.h>
 846 #include <linux/netfilter/nfnetlink_conntrack.h>
 847 #include <linux/mutex.h>
 848
 849 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 850  * in ip_conntrack_core, since we don't want the protocols to autoload
 851  * or depend on ctnetlink */
 852 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 853                                const struct nf_conntrack_tuple *tuple)
 854 {
 855         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 856         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 857         return 0;
 858
 859 nla_put_failure:
 860         return -1;
 861 }
 862 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 863
 864 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 865         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 866         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 867 };
 868 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 869
 870 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 871                                struct nf_conntrack_tuple *t)
 872 {
 873         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 874                 return -EINVAL;
 875
 876         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 877         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 878
 879         return 0;
 880 }
 881 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 882 #endif
 883
 884 /* Used by ipt_REJECT and ip6t_REJECT. */
 885 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 886 {
 887         struct nf_conn *ct;
 888         enum ip_conntrack_info ctinfo;
 889
 890         /* This ICMP is in reverse direction to the packet which caused it */
 891         ct = nf_ct_get(skb, &ctinfo);
 892         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 893                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 894         else
 895                 ctinfo = IP_CT_RELATED;
 896
 897         /* Attach to new skbuff, and increment count */
 898         nskb->nfct = &ct->ct_general;
 899         nskb->nfctinfo = ctinfo;
 900         nf_conntrack_get(nskb->nfct);
 901 }
 902
 903 static inline int
 904 do_iter(const struct nf_conntrack_tuple_hash *i,
 905         int (*iter)(struct nf_conn *i, void *data),
 906         void *data)
 907 {
 908         return iter(nf_ct_tuplehash_to_ctrack(i), data);
 909 }
 910
 911 /* Bring out ya dead! */
 912 static struct nf_conn *
 913 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 914                 void *data, unsigned int *bucket)
 915 {
 916         struct nf_conntrack_tuple_hash *h;
 917         struct nf_conn *ct;
 918         struct hlist_node *n;
 919
 920         spin_lock_bh(&nf_conntrack_lock);
 921         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 922                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
 923                         ct = nf_ct_tuplehash_to_ctrack(h);
 924                         if (iter(ct, data))
 925                                 goto found;
 926                 }
 927         }
 928         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
 929                 ct = nf_ct_tuplehash_to_ctrack(h);
 930                 if (iter(ct, data))
 931                         set_bit(IPS_DYING_BIT, &ct->status);
 932         }
 933         spin_unlock_bh(&nf_conntrack_lock);
 934         return NULL;
 935 found:
 936         atomic_inc(&ct->ct_general.use);
 937         spin_unlock_bh(&nf_conntrack_lock);
 938         return ct;
 939 }
 940
 941 void
 942 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 943 {
 944         struct nf_conn *ct;
 945         unsigned int bucket = 0;
 946
 947         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 948                 /* Time to push up daises... */
 949                 if (del_timer(&ct->timeout))
 950                         death_by_timeout((unsigned long)ct);
 951                 /* ... else the timer will get him soon. */
 952
 953                 nf_ct_put(ct);
 954         }
 955 }
 956 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 957
 958 static int kill_all(struct nf_conn *i, void *data)
 959 {
 960         return 1;
 961 }
 962
 963 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
 964 {
 965         if (vmalloced)
 966                 vfree(hash);
 967         else
 968                 free_pages((unsigned long)hash,
 969                            get_order(sizeof(struct hlist_head) * size));
 970 }
 971 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
 972
 973 void nf_conntrack_flush(void)
 974 {
 975         nf_ct_iterate_cleanup(kill_all, NULL);
 976 }
 977 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 978
 979 /* Mishearing the voices in his head, our hero wonders how he's
 980    supposed to kill the mall. */
 981 void nf_conntrack_cleanup(void)
 982 {
 983         rcu_assign_pointer(ip_ct_attach, NULL);
 984
 985         /* This makes sure all current packets have passed through
 986            netfilter framework.  Roll on, two-stage module
 987            delete... */
 988         synchronize_net();
 989
 990         nf_ct_event_cache_flush();
 991  i_see_dead_people:
 992         nf_conntrack_flush();
 993         if (atomic_read(&nf_conntrack_count) != 0) {
 994                 schedule();
 995                 goto i_see_dead_people;
 996         }
 997         /* wait until all references to nf_conntrack_untracked are dropped */
 998         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
 999                 schedule();
1000
1001         rcu_assign_pointer(nf_ct_destroy, NULL);
1002
1003         kmem_cache_destroy(nf_conntrack_cachep);
1004         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1005                              nf_conntrack_htable_size);
1006
1007         nf_conntrack_proto_fini();
1008         nf_conntrack_helper_fini();
1009         nf_conntrack_expect_fini();
1010 }
1011
1012 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1013 {
1014         struct hlist_head *hash;
1015         unsigned int size, i;
1016
1017         *vmalloced = 0;
1018
1019         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1020         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1021                                        get_order(sizeof(struct hlist_head)
1022                                                  * size));
1023         if (!hash) {
1024                 *vmalloced = 1;
1025                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1026                 hash = vmalloc(sizeof(struct hlist_head) * size);
1027         }
1028
1029         if (hash)
1030                 for (i = 0; i < size; i++)
1031                         INIT_HLIST_HEAD(&hash[i]);
1032
1033         return hash;
1034 }
1035 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1036
1037 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1038 {
1039         int i, bucket, vmalloced, old_vmalloced;
1040         unsigned int hashsize, old_size;
1041         int rnd;
1042         struct hlist_head *hash, *old_hash;
1043         struct nf_conntrack_tuple_hash *h;
1044
1045         /* On boot, we can set this without any fancy locking. */
1046         if (!nf_conntrack_htable_size)
1047                 return param_set_uint(val, kp);
1048
1049         hashsize = simple_strtoul(val, NULL, 0);
1050         if (!hashsize)
1051                 return -EINVAL;
1052
1053         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1054         if (!hash)
1055                 return -ENOMEM;
1056
1057         /* We have to rehahs for the new table anyway, so we also can
1058          * use a newrandom seed */
1059         get_random_bytes(&rnd, 4);
1060
1061         /* Lookups in the old hash might happen in parallel, which means we
1062          * might get false negatives during connection lookup. New connections
1063          * created because of a false negative won't make it into the hash
1064          * though since that required taking the lock.
1065          */
1066         spin_lock_bh(&nf_conntrack_lock);
1067         for (i = 0; i < nf_conntrack_htable_size; i++) {
1068                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1069                         h = hlist_entry(nf_conntrack_hash[i].first,
1070                                         struct nf_conntrack_tuple_hash, hnode);
1071                         hlist_del_rcu(&h->hnode);
1072                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1073                         hlist_add_head(&h->hnode, &hash[bucket]);
1074                 }
1075         }
1076         old_size = nf_conntrack_htable_size;
1077         old_vmalloced = nf_conntrack_vmalloc;
1078         old_hash = nf_conntrack_hash;
1079
1080         nf_conntrack_htable_size = hashsize;
1081         nf_conntrack_vmalloc = vmalloced;
1082         nf_conntrack_hash = hash;
1083         nf_conntrack_hash_rnd = rnd;
1084         spin_unlock_bh(&nf_conntrack_lock);
1085
1086         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1087         return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1090
1091 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1092                   &nf_conntrack_htable_size, 0600);
1093
1094 int __init nf_conntrack_init(void)
1095 {
1096         int max_factor = 8;
1097         int ret;
1098
1099         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1100          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1101         if (!nf_conntrack_htable_size) {
1102                 nf_conntrack_htable_size
1103                         = (((num_physpages << PAGE_SHIFT) / 16384)
1104                            / sizeof(struct hlist_head));
1105                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1106                         nf_conntrack_htable_size = 16384;
1107                 if (nf_conntrack_htable_size < 32)
1108                         nf_conntrack_htable_size = 32;
1109
1110                 /* Use a max. factor of four by default to get the same max as
1111                  * with the old struct list_heads. When a table size is given
1112                  * we use the old value of 8 to avoid reducing the max.
1113                  * entries. */
1114                 max_factor = 4;
1115         }
1116         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1117                                                   &nf_conntrack_vmalloc);
1118         if (!nf_conntrack_hash) {
1119                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1120                 goto err_out;
1121         }
1122
1123         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1124
1125         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1126                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1127                nf_conntrack_max);
1128
1129         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1130                                                 sizeof(struct nf_conn),
1131                                                 0, 0, NULL);
1132         if (!nf_conntrack_cachep) {
1133                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1134                 goto err_free_hash;
1135         }
1136
1137         ret = nf_conntrack_proto_init();
1138         if (ret < 0)
1139                 goto err_free_conntrack_slab;
1140
1141         ret = nf_conntrack_expect_init();
1142         if (ret < 0)
1143                 goto out_fini_proto;
1144
1145         ret = nf_conntrack_helper_init();
1146         if (ret < 0)
1147                 goto out_fini_expect;
1148
1149         /* For use by REJECT target */
1150         rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1151         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1152
1153         /* Set up fake conntrack:
1154             - to never be deleted, not in any hashes */
1155         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1156         /*  - and look it like as a confirmed connection */
1157         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1158
1159         return ret;
1160
1161 out_fini_expect:
1162         nf_conntrack_expect_fini();
1163 out_fini_proto:
1164         nf_conntrack_proto_fini();
1165 err_free_conntrack_slab:
1166         kmem_cache_destroy(nf_conntrack_cachep);
1167 err_free_hash:
1168         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1169                              nf_conntrack_htable_size);
1170 err_out:
1171         return -ENOMEM;
1172 }