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 n;
  77         u_int32_t h;
  78
  79         /* The direction must be ignored, so we hash everything up to the
  80          * destination ports (which is a multiple of 4) and treat the last
  81          * three bytes manually.
  82          */
  83         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  84         h = jhash2((u32 *)tuple, n,
  85                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
  86                           tuple->dst.protonum));
  87
  88         return ((u64)h * size) >> 32;
  89 }
  90
  91 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  92 {
  93         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  94                                 nf_conntrack_hash_rnd);
  95 }
  96
  97 int
  98 nf_ct_get_tuple(const struct sk_buff *skb,
  99                 unsigned int nhoff,
 100                 unsigned int dataoff,
 101                 u_int16_t l3num,
 102                 u_int8_t protonum,
 103                 struct nf_conntrack_tuple *tuple,
 104                 const struct nf_conntrack_l3proto *l3proto,
 105                 const struct nf_conntrack_l4proto *l4proto)
 106 {
 107         NF_CT_TUPLE_U_BLANK(tuple);
 108
 109         tuple->src.l3num = l3num;
 110         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 111                 return 0;
 112
 113         tuple->dst.protonum = protonum;
 114         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 115
 116         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 117 }
 118 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 119
 120 int nf_ct_get_tuplepr(const struct sk_buff *skb,
 121                       unsigned int nhoff,
 122                       u_int16_t l3num,
 123                       struct nf_conntrack_tuple *tuple)
 124 {
 125         struct nf_conntrack_l3proto *l3proto;
 126         struct nf_conntrack_l4proto *l4proto;
 127         unsigned int protoff;
 128         u_int8_t protonum;
 129         int ret;
 130
 131         rcu_read_lock();
 132
 133         l3proto = __nf_ct_l3proto_find(l3num);
 134         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 135         if (ret != NF_ACCEPT) {
 136                 rcu_read_unlock();
 137                 return 0;
 138         }
 139
 140         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 141
 142         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 143                               l3proto, l4proto);
 144
 145         rcu_read_unlock();
 146         return ret;
 147 }
 148 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 149
 150 int
 151 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 152                    const struct nf_conntrack_tuple *orig,
 153                    const struct nf_conntrack_l3proto *l3proto,
 154                    const struct nf_conntrack_l4proto *l4proto)
 155 {
 156         NF_CT_TUPLE_U_BLANK(inverse);
 157
 158         inverse->src.l3num = orig->src.l3num;
 159         if (l3proto->invert_tuple(inverse, orig) == 0)
 160                 return 0;
 161
 162         inverse->dst.dir = !orig->dst.dir;
 163
 164         inverse->dst.protonum = orig->dst.protonum;
 165         return l4proto->invert_tuple(inverse, orig);
 166 }
 167 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 168
 169 static void
 170 clean_from_lists(struct nf_conn *ct)
 171 {
 172         pr_debug("clean_from_lists(%p)\n", ct);
 173         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 174         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 175
 176         /* Destroy all pending expectations */
 177         nf_ct_remove_expectations(ct);
 178 }
 179
 180 static void
 181 destroy_conntrack(struct nf_conntrack *nfct)
 182 {
 183         struct nf_conn *ct = (struct nf_conn *)nfct;
 184         struct nf_conntrack_l4proto *l4proto;
 185
 186         pr_debug("destroy_conntrack(%p)\n", ct);
 187         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 188         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 189
 190         nf_conntrack_event(IPCT_DESTROY, ct);
 191         set_bit(IPS_DYING_BIT, &ct->status);
 192
 193         /* To make sure we don't get any weird locking issues here:
 194          * destroy_conntrack() MUST NOT be called with a write lock
 195          * to nf_conntrack_lock!!! -HW */
 196         rcu_read_lock();
 197         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 198         if (l4proto && l4proto->destroy)
 199                 l4proto->destroy(ct);
 200
 201         nf_ct_ext_destroy(ct);
 202
 203         rcu_read_unlock();
 204
 205         spin_lock_bh(&nf_conntrack_lock);
 206         /* Expectations will have been removed in clean_from_lists,
 207          * except TFTP can create an expectation on the first packet,
 208          * before connection is in the list, so we need to clean here,
 209          * too. */
 210         nf_ct_remove_expectations(ct);
 211
 212         /* We overload first tuple to link into unconfirmed list. */
 213         if (!nf_ct_is_confirmed(ct)) {
 214                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
 215                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 216         }
 217
 218         NF_CT_STAT_INC(delete);
 219         spin_unlock_bh(&nf_conntrack_lock);
 220
 221         if (ct->master)
 222                 nf_ct_put(ct->master);
 223
 224         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 225         nf_conntrack_free(ct);
 226 }
 227
 228 static void death_by_timeout(unsigned long ul_conntrack)
 229 {
 230         struct nf_conn *ct = (void *)ul_conntrack;
 231         struct nf_conn_help *help = nfct_help(ct);
 232         struct nf_conntrack_helper *helper;
 233
 234         if (help) {
 235                 rcu_read_lock();
 236                 helper = rcu_dereference(help->helper);
 237                 if (helper && helper->destroy)
 238                         helper->destroy(ct);
 239                 rcu_read_unlock();
 240         }
 241
 242         spin_lock_bh(&nf_conntrack_lock);
 243         /* Inside lock so preempt is disabled on module removal path.
 244          * Otherwise we can get spurious warnings. */
 245         NF_CT_STAT_INC(delete_list);
 246         clean_from_lists(ct);
 247         spin_unlock_bh(&nf_conntrack_lock);
 248         nf_ct_put(ct);
 249 }
 250
 251 struct nf_conntrack_tuple_hash *
 252 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
 253 {
 254         struct nf_conntrack_tuple_hash *h;
 255         struct hlist_node *n;
 256         unsigned int hash = hash_conntrack(tuple);
 257
 258         /* Disable BHs the entire time since we normally need to disable them
 259          * at least once for the stats anyway.
 260          */
 261         local_bh_disable();
 262         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 263                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 264                         NF_CT_STAT_INC(found);
 265                         local_bh_enable();
 266                         return h;
 267                 }
 268                 NF_CT_STAT_INC(searched);
 269         }
 270         local_bh_enable();
 271
 272         return NULL;
 273 }
 274 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 275
 276 /* Find a connection corresponding to a tuple. */
 277 struct nf_conntrack_tuple_hash *
 278 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
 279 {
 280         struct nf_conntrack_tuple_hash *h;
 281         struct nf_conn *ct;
 282
 283         rcu_read_lock();
 284         h = __nf_conntrack_find(tuple);
 285         if (h) {
 286                 ct = nf_ct_tuplehash_to_ctrack(h);
 287                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 288                         h = NULL;
 289         }
 290         rcu_read_unlock();
 291
 292         return h;
 293 }
 294 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 295
 296 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 297                                        unsigned int hash,
 298                                        unsigned int repl_hash)
 299 {
 300         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 301                            &nf_conntrack_hash[hash]);
 302         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 303                            &nf_conntrack_hash[repl_hash]);
 304 }
 305
 306 void nf_conntrack_hash_insert(struct nf_conn *ct)
 307 {
 308         unsigned int hash, repl_hash;
 309
 310         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 311         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 312
 313         spin_lock_bh(&nf_conntrack_lock);
 314         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 315         spin_unlock_bh(&nf_conntrack_lock);
 316 }
 317 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 318
 319 /* Confirm a connection given skb; places it in hash table */
 320 int
 321 __nf_conntrack_confirm(struct sk_buff *skb)
 322 {
 323         unsigned int hash, repl_hash;
 324         struct nf_conntrack_tuple_hash *h;
 325         struct nf_conn *ct;
 326         struct nf_conn_help *help;
 327         struct hlist_node *n;
 328         enum ip_conntrack_info ctinfo;
 329
 330         ct = nf_ct_get(skb, &ctinfo);
 331
 332         /* ipt_REJECT uses nf_conntrack_attach to attach related
 333            ICMP/TCP RST packets in other direction.  Actual packet
 334            which created connection will be IP_CT_NEW or for an
 335            expected connection, IP_CT_RELATED. */
 336         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 337                 return NF_ACCEPT;
 338
 339         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 340         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 341
 342         /* We're not in hash table, and we refuse to set up related
 343            connections for unconfirmed conns.  But packet copies and
 344            REJECT will give spurious warnings here. */
 345         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 346
 347         /* No external references means noone else could have
 348            confirmed us. */
 349         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 350         pr_debug("Confirming conntrack %p\n", ct);
 351
 352         spin_lock_bh(&nf_conntrack_lock);
 353
 354         /* See if there's one in the list already, including reverse:
 355            NAT could have grabbed it without realizing, since we're
 356            not in the hash.  If there is, we lost race. */
 357         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
 358                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 359                                       &h->tuple))
 360                         goto out;
 361         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
 362                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 363                                       &h->tuple))
 364                         goto out;
 365
 366         /* Remove from unconfirmed list */
 367         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 368
 369         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 370         /* Timer relative to confirmation time, not original
 371            setting time, otherwise we'd get timer wrap in
 372            weird delay cases. */
 373         ct->timeout.expires += jiffies;
 374         add_timer(&ct->timeout);
 375         atomic_inc(&ct->ct_general.use);
 376         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 377         NF_CT_STAT_INC(insert);
 378         spin_unlock_bh(&nf_conntrack_lock);
 379         help = nfct_help(ct);
 380         if (help && help->helper)
 381                 nf_conntrack_event_cache(IPCT_HELPER, skb);
 382 #ifdef CONFIG_NF_NAT_NEEDED
 383         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 384             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 385                 nf_conntrack_event_cache(IPCT_NATINFO, skb);
 386 #endif
 387         nf_conntrack_event_cache(master_ct(ct) ?
 388                                  IPCT_RELATED : IPCT_NEW, skb);
 389         return NF_ACCEPT;
 390
 391 out:
 392         NF_CT_STAT_INC(insert_failed);
 393         spin_unlock_bh(&nf_conntrack_lock);
 394         return NF_DROP;
 395 }
 396 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 397
 398 /* Returns true if a connection correspondings to the tuple (required
 399    for NAT). */
 400 int
 401 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 402                          const struct nf_conn *ignored_conntrack)
 403 {
 404         struct nf_conntrack_tuple_hash *h;
 405         struct hlist_node *n;
 406         unsigned int hash = hash_conntrack(tuple);
 407
 408         /* Disable BHs the entire time since we need to disable them at
 409          * least once for the stats anyway.
 410          */
 411         rcu_read_lock_bh();
 412         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 413                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 414                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 415                         NF_CT_STAT_INC(found);
 416                         rcu_read_unlock_bh();
 417                         return 1;
 418                 }
 419                 NF_CT_STAT_INC(searched);
 420         }
 421         rcu_read_unlock_bh();
 422
 423         return 0;
 424 }
 425 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 426
 427 #define NF_CT_EVICTION_RANGE    8
 428
 429 /* There's a small race here where we may free a just-assured
 430    connection.  Too bad: we're in trouble anyway. */
 431 static noinline int early_drop(unsigned int hash)
 432 {
 433         /* Use oldest entry, which is roughly LRU */
 434         struct nf_conntrack_tuple_hash *h;
 435         struct nf_conn *ct = NULL, *tmp;
 436         struct hlist_node *n;
 437         unsigned int i, cnt = 0;
 438         int dropped = 0;
 439
 440         rcu_read_lock();
 441         for (i = 0; i < nf_conntrack_htable_size; i++) {
 442                 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash],
 443                                          hnode) {
 444                         tmp = nf_ct_tuplehash_to_ctrack(h);
 445                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 446                                 ct = tmp;
 447                         cnt++;
 448                 }
 449
 450                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 451                         ct = NULL;
 452                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 453                         break;
 454                 hash = (hash + 1) % nf_conntrack_htable_size;
 455         }
 456         rcu_read_unlock();
 457
 458         if (!ct)
 459                 return dropped;
 460
 461         if (del_timer(&ct->timeout)) {
 462                 death_by_timeout((unsigned long)ct);
 463                 dropped = 1;
 464                 NF_CT_STAT_INC_ATOMIC(early_drop);
 465         }
 466         nf_ct_put(ct);
 467         return dropped;
 468 }
 469
 470 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 471                                    const struct nf_conntrack_tuple *repl)
 472 {
 473         struct nf_conn *ct = NULL;
 474
 475         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 476                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 477                 nf_conntrack_hash_rnd_initted = 1;
 478         }
 479
 480         /* We don't want any race condition at early drop stage */
 481         atomic_inc(&nf_conntrack_count);
 482
 483         if (nf_conntrack_max &&
 484             unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
 485                 unsigned int hash = hash_conntrack(orig);
 486                 if (!early_drop(hash)) {
 487                         atomic_dec(&nf_conntrack_count);
 488                         if (net_ratelimit())
 489                                 printk(KERN_WARNING
 490                                        "nf_conntrack: table full, dropping"
 491                                        " packet.\n");
 492                         return ERR_PTR(-ENOMEM);
 493                 }
 494         }
 495
 496         ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
 497         if (ct == NULL) {
 498                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 499                 atomic_dec(&nf_conntrack_count);
 500                 return ERR_PTR(-ENOMEM);
 501         }
 502
 503         atomic_set(&ct->ct_general.use, 1);
 504         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 505         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 506         /* Don't set timer yet: wait for confirmation */
 507         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 508         INIT_RCU_HEAD(&ct->rcu);
 509
 510         return ct;
 511 }
 512 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 513
 514 static void nf_conntrack_free_rcu(struct rcu_head *head)
 515 {
 516         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
 517
 518         nf_ct_ext_free(ct);
 519         kmem_cache_free(nf_conntrack_cachep, ct);
 520         atomic_dec(&nf_conntrack_count);
 521 }
 522
 523 void nf_conntrack_free(struct nf_conn *ct)
 524 {
 525         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 526 }
 527 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 528
 529 /* Allocate a new conntrack: we return -ENOMEM if classification
 530    failed due to stress.  Otherwise it really is unclassifiable. */
 531 static struct nf_conntrack_tuple_hash *
 532 init_conntrack(const struct nf_conntrack_tuple *tuple,
 533                struct nf_conntrack_l3proto *l3proto,
 534                struct nf_conntrack_l4proto *l4proto,
 535                struct sk_buff *skb,
 536                unsigned int dataoff)
 537 {
 538         struct nf_conn *ct;
 539         struct nf_conn_help *help;
 540         struct nf_conntrack_tuple repl_tuple;
 541         struct nf_conntrack_expect *exp;
 542
 543         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 544                 pr_debug("Can't invert tuple.\n");
 545                 return NULL;
 546         }
 547
 548         ct = nf_conntrack_alloc(tuple, &repl_tuple);
 549         if (ct == NULL || IS_ERR(ct)) {
 550                 pr_debug("Can't allocate conntrack.\n");
 551                 return (struct nf_conntrack_tuple_hash *)ct;
 552         }
 553
 554         if (!l4proto->new(ct, skb, dataoff)) {
 555                 nf_conntrack_free(ct);
 556                 pr_debug("init conntrack: can't track with proto module\n");
 557                 return NULL;
 558         }
 559
 560         spin_lock_bh(&nf_conntrack_lock);
 561         exp = nf_ct_find_expectation(tuple);
 562         if (exp) {
 563                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 564                          ct, exp);
 565                 /* Welcome, Mr. Bond.  We've been expecting you... */
 566                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 567                 ct->master = exp->master;
 568                 if (exp->helper) {
 569                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 570                         if (help)
 571                                 rcu_assign_pointer(help->helper, exp->helper);
 572                 }
 573
 574 #ifdef CONFIG_NF_CONNTRACK_MARK
 575                 ct->mark = exp->master->mark;
 576 #endif
 577 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 578                 ct->secmark = exp->master->secmark;
 579 #endif
 580                 nf_conntrack_get(&ct->master->ct_general);
 581                 NF_CT_STAT_INC(expect_new);
 582         } else {
 583                 struct nf_conntrack_helper *helper;
 584
 585                 helper = __nf_ct_helper_find(&repl_tuple);
 586                 if (helper) {
 587                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 588                         if (help)
 589                                 rcu_assign_pointer(help->helper, helper);
 590                 }
 591                 NF_CT_STAT_INC(new);
 592         }
 593
 594         /* Overload tuple linked list to put us in unconfirmed list. */
 595         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
 596
 597         spin_unlock_bh(&nf_conntrack_lock);
 598
 599         if (exp) {
 600                 if (exp->expectfn)
 601                         exp->expectfn(ct, exp);
 602                 nf_ct_expect_put(exp);
 603         }
 604
 605         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 606 }
 607
 608 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 609 static inline struct nf_conn *
 610 resolve_normal_ct(struct sk_buff *skb,
 611                   unsigned int dataoff,
 612                   u_int16_t l3num,
 613                   u_int8_t protonum,
 614                   struct nf_conntrack_l3proto *l3proto,
 615                   struct nf_conntrack_l4proto *l4proto,
 616                   int *set_reply,
 617                   enum ip_conntrack_info *ctinfo)
 618 {
 619         struct nf_conntrack_tuple tuple;
 620         struct nf_conntrack_tuple_hash *h;
 621         struct nf_conn *ct;
 622
 623         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 624                              dataoff, l3num, protonum, &tuple, l3proto,
 625                              l4proto)) {
 626                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 627                 return NULL;
 628         }
 629
 630         /* look for tuple match */
 631         h = nf_conntrack_find_get(&tuple);
 632         if (!h) {
 633                 h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
 634                 if (!h)
 635                         return NULL;
 636                 if (IS_ERR(h))
 637                         return (void *)h;
 638         }
 639         ct = nf_ct_tuplehash_to_ctrack(h);
 640
 641         /* It exists; we have (non-exclusive) reference. */
 642         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 643                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 644                 /* Please set reply bit if this packet OK */
 645                 *set_reply = 1;
 646         } else {
 647                 /* Once we've had two way comms, always ESTABLISHED. */
 648                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 649                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 650                         *ctinfo = IP_CT_ESTABLISHED;
 651                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 652                         pr_debug("nf_conntrack_in: related packet for %p\n",
 653                                  ct);
 654                         *ctinfo = IP_CT_RELATED;
 655                 } else {
 656                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 657                         *ctinfo = IP_CT_NEW;
 658                 }
 659                 *set_reply = 0;
 660         }
 661         skb->nfct = &ct->ct_general;
 662         skb->nfctinfo = *ctinfo;
 663         return ct;
 664 }
 665
 666 unsigned int
 667 nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
 668 {
 669         struct nf_conn *ct;
 670         enum ip_conntrack_info ctinfo;
 671         struct nf_conntrack_l3proto *l3proto;
 672         struct nf_conntrack_l4proto *l4proto;
 673         unsigned int dataoff;
 674         u_int8_t protonum;
 675         int set_reply = 0;
 676         int ret;
 677
 678         /* Previously seen (loopback or untracked)?  Ignore. */
 679         if (skb->nfct) {
 680                 NF_CT_STAT_INC_ATOMIC(ignore);
 681                 return NF_ACCEPT;
 682         }
 683
 684         /* rcu_read_lock()ed by nf_hook_slow */
 685         l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
 686         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 687                                    &dataoff, &protonum);
 688         if (ret <= 0) {
 689                 pr_debug("not prepared to track yet or error occured\n");
 690                 NF_CT_STAT_INC_ATOMIC(error);
 691                 NF_CT_STAT_INC_ATOMIC(invalid);
 692                 return -ret;
 693         }
 694
 695         l4proto = __nf_ct_l4proto_find((u_int16_t)pf, protonum);
 696
 697         /* It may be an special packet, error, unclean...
 698          * inverse of the return code tells to the netfilter
 699          * core what to do with the packet. */
 700         if (l4proto->error != NULL &&
 701             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 702                 NF_CT_STAT_INC_ATOMIC(error);
 703                 NF_CT_STAT_INC_ATOMIC(invalid);
 704                 return -ret;
 705         }
 706
 707         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
 708                                &set_reply, &ctinfo);
 709         if (!ct) {
 710                 /* Not valid part of a connection */
 711                 NF_CT_STAT_INC_ATOMIC(invalid);
 712                 return NF_ACCEPT;
 713         }
 714
 715         if (IS_ERR(ct)) {
 716                 /* Too stressed to deal. */
 717                 NF_CT_STAT_INC_ATOMIC(drop);
 718                 return NF_DROP;
 719         }
 720
 721         NF_CT_ASSERT(skb->nfct);
 722
 723         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 724         if (ret < 0) {
 725                 /* Invalid: inverse of the return code tells
 726                  * the netfilter core what to do */
 727                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 728                 nf_conntrack_put(skb->nfct);
 729                 skb->nfct = NULL;
 730                 NF_CT_STAT_INC_ATOMIC(invalid);
 731                 return -ret;
 732         }
 733
 734         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 735                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 736
 737         return ret;
 738 }
 739 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 740
 741 int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 742                          const struct nf_conntrack_tuple *orig)
 743 {
 744         int ret;
 745
 746         rcu_read_lock();
 747         ret = nf_ct_invert_tuple(inverse, orig,
 748                                  __nf_ct_l3proto_find(orig->src.l3num),
 749                                  __nf_ct_l4proto_find(orig->src.l3num,
 750                                                       orig->dst.protonum));
 751         rcu_read_unlock();
 752         return ret;
 753 }
 754 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 755
 756 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 757    implicitly racy: see __nf_conntrack_confirm */
 758 void nf_conntrack_alter_reply(struct nf_conn *ct,
 759                               const struct nf_conntrack_tuple *newreply)
 760 {
 761         struct nf_conn_help *help = nfct_help(ct);
 762         struct nf_conntrack_helper *helper;
 763
 764         /* Should be unconfirmed, so not in hash table yet */
 765         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 766
 767         pr_debug("Altering reply tuple of %p to ", ct);
 768         NF_CT_DUMP_TUPLE(newreply);
 769
 770         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 771         if (ct->master || (help && help->expecting != 0))
 772                 return;
 773
 774         rcu_read_lock();
 775         helper = __nf_ct_helper_find(newreply);
 776         if (helper == NULL) {
 777                 if (help)
 778                         rcu_assign_pointer(help->helper, NULL);
 779                 goto out;
 780         }
 781
 782         if (help == NULL) {
 783                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 784                 if (help == NULL)
 785                         goto out;
 786         } else {
 787                 memset(&help->help, 0, sizeof(help->help));
 788         }
 789
 790         rcu_assign_pointer(help->helper, helper);
 791 out:
 792         rcu_read_unlock();
 793 }
 794 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 795
 796 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 797 void __nf_ct_refresh_acct(struct nf_conn *ct,
 798                           enum ip_conntrack_info ctinfo,
 799                           const struct sk_buff *skb,
 800                           unsigned long extra_jiffies,
 801                           int do_acct)
 802 {
 803         int event = 0;
 804
 805         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 806         NF_CT_ASSERT(skb);
 807
 808         spin_lock_bh(&nf_conntrack_lock);
 809
 810         /* Only update if this is not a fixed timeout */
 811         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 812                 goto acct;
 813
 814         /* If not in hash table, timer will not be active yet */
 815         if (!nf_ct_is_confirmed(ct)) {
 816                 ct->timeout.expires = extra_jiffies;
 817                 event = IPCT_REFRESH;
 818         } else {
 819                 unsigned long newtime = jiffies + extra_jiffies;
 820
 821                 /* Only update the timeout if the new timeout is at least
 822                    HZ jiffies from the old timeout. Need del_timer for race
 823                    avoidance (may already be dying). */
 824                 if (newtime - ct->timeout.expires >= HZ
 825                     && del_timer(&ct->timeout)) {
 826                         ct->timeout.expires = newtime;
 827                         add_timer(&ct->timeout);
 828                         event = IPCT_REFRESH;
 829                 }
 830         }
 831
 832 acct:
 833 #ifdef CONFIG_NF_CT_ACCT
 834         if (do_acct) {
 835                 ct->counters[CTINFO2DIR(ctinfo)].packets++;
 836                 ct->counters[CTINFO2DIR(ctinfo)].bytes +=
 837                         skb->len - skb_network_offset(skb);
 838
 839                 if ((ct->counters[CTINFO2DIR(ctinfo)].packets & 0x80000000)
 840                     || (ct->counters[CTINFO2DIR(ctinfo)].bytes & 0x80000000))
 841                         event |= IPCT_COUNTER_FILLING;
 842         }
 843 #endif
 844
 845         spin_unlock_bh(&nf_conntrack_lock);
 846
 847         /* must be unlocked when calling event cache */
 848         if (event)
 849                 nf_conntrack_event_cache(event, skb);
 850 }
 851 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 852
 853 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 854
 855 #include <linux/netfilter/nfnetlink.h>
 856 #include <linux/netfilter/nfnetlink_conntrack.h>
 857 #include <linux/mutex.h>
 858
 859 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 860  * in ip_conntrack_core, since we don't want the protocols to autoload
 861  * or depend on ctnetlink */
 862 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 863                                const struct nf_conntrack_tuple *tuple)
 864 {
 865         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 866         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 867         return 0;
 868
 869 nla_put_failure:
 870         return -1;
 871 }
 872 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 873
 874 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 875         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 876         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 877 };
 878 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 879
 880 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 881                                struct nf_conntrack_tuple *t)
 882 {
 883         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 884                 return -EINVAL;
 885
 886         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 887         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 888
 889         return 0;
 890 }
 891 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 892 #endif
 893
 894 /* Used by ipt_REJECT and ip6t_REJECT. */
 895 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 896 {
 897         struct nf_conn *ct;
 898         enum ip_conntrack_info ctinfo;
 899
 900         /* This ICMP is in reverse direction to the packet which caused it */
 901         ct = nf_ct_get(skb, &ctinfo);
 902         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 903                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 904         else
 905                 ctinfo = IP_CT_RELATED;
 906
 907         /* Attach to new skbuff, and increment count */
 908         nskb->nfct = &ct->ct_general;
 909         nskb->nfctinfo = ctinfo;
 910         nf_conntrack_get(nskb->nfct);
 911 }
 912
 913 /* Bring out ya dead! */
 914 static struct nf_conn *
 915 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 916                 void *data, unsigned int *bucket)
 917 {
 918         struct nf_conntrack_tuple_hash *h;
 919         struct nf_conn *ct;
 920         struct hlist_node *n;
 921
 922         spin_lock_bh(&nf_conntrack_lock);
 923         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 924                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
 925                         ct = nf_ct_tuplehash_to_ctrack(h);
 926                         if (iter(ct, data))
 927                                 goto found;
 928                 }
 929         }
 930         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
 931                 ct = nf_ct_tuplehash_to_ctrack(h);
 932                 if (iter(ct, data))
 933                         set_bit(IPS_DYING_BIT, &ct->status);
 934         }
 935         spin_unlock_bh(&nf_conntrack_lock);
 936         return NULL;
 937 found:
 938         atomic_inc(&ct->ct_general.use);
 939         spin_unlock_bh(&nf_conntrack_lock);
 940         return ct;
 941 }
 942
 943 void
 944 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 945 {
 946         struct nf_conn *ct;
 947         unsigned int bucket = 0;
 948
 949         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 950                 /* Time to push up daises... */
 951                 if (del_timer(&ct->timeout))
 952                         death_by_timeout((unsigned long)ct);
 953                 /* ... else the timer will get him soon. */
 954
 955                 nf_ct_put(ct);
 956         }
 957 }
 958 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 959
 960 static int kill_all(struct nf_conn *i, void *data)
 961 {
 962         return 1;
 963 }
 964
 965 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
 966 {
 967         if (vmalloced)
 968                 vfree(hash);
 969         else
 970                 free_pages((unsigned long)hash,
 971                            get_order(sizeof(struct hlist_head) * size));
 972 }
 973 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
 974
 975 void nf_conntrack_flush(void)
 976 {
 977         nf_ct_iterate_cleanup(kill_all, NULL);
 978 }
 979 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
 980
 981 /* Mishearing the voices in his head, our hero wonders how he's
 982    supposed to kill the mall. */
 983 void nf_conntrack_cleanup(void)
 984 {
 985         rcu_assign_pointer(ip_ct_attach, NULL);
 986
 987         /* This makes sure all current packets have passed through
 988            netfilter framework.  Roll on, two-stage module
 989            delete... */
 990         synchronize_net();
 991
 992         nf_ct_event_cache_flush();
 993  i_see_dead_people:
 994         nf_conntrack_flush();
 995         if (atomic_read(&nf_conntrack_count) != 0) {
 996                 schedule();
 997                 goto i_see_dead_people;
 998         }
 999         /* wait until all references to nf_conntrack_untracked are dropped */
1000         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1001                 schedule();
1002
1003         rcu_assign_pointer(nf_ct_destroy, NULL);
1004
1005         kmem_cache_destroy(nf_conntrack_cachep);
1006         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1007                              nf_conntrack_htable_size);
1008
1009         nf_conntrack_proto_fini();
1010         nf_conntrack_helper_fini();
1011         nf_conntrack_expect_fini();
1012 }
1013
1014 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1015 {
1016         struct hlist_head *hash;
1017         unsigned int size, i;
1018
1019         *vmalloced = 0;
1020
1021         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1022         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1023                                        get_order(sizeof(struct hlist_head)
1024                                                  * size));
1025         if (!hash) {
1026                 *vmalloced = 1;
1027                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1028                 hash = vmalloc(sizeof(struct hlist_head) * size);
1029         }
1030
1031         if (hash)
1032                 for (i = 0; i < size; i++)
1033                         INIT_HLIST_HEAD(&hash[i]);
1034
1035         return hash;
1036 }
1037 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1038
1039 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1040 {
1041         int i, bucket, vmalloced, old_vmalloced;
1042         unsigned int hashsize, old_size;
1043         int rnd;
1044         struct hlist_head *hash, *old_hash;
1045         struct nf_conntrack_tuple_hash *h;
1046
1047         /* On boot, we can set this without any fancy locking. */
1048         if (!nf_conntrack_htable_size)
1049                 return param_set_uint(val, kp);
1050
1051         hashsize = simple_strtoul(val, NULL, 0);
1052         if (!hashsize)
1053                 return -EINVAL;
1054
1055         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1056         if (!hash)
1057                 return -ENOMEM;
1058
1059         /* We have to rehahs for the new table anyway, so we also can
1060          * use a newrandom seed */
1061         get_random_bytes(&rnd, 4);
1062
1063         /* Lookups in the old hash might happen in parallel, which means we
1064          * might get false negatives during connection lookup. New connections
1065          * created because of a false negative won't make it into the hash
1066          * though since that required taking the lock.
1067          */
1068         spin_lock_bh(&nf_conntrack_lock);
1069         for (i = 0; i < nf_conntrack_htable_size; i++) {
1070                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1071                         h = hlist_entry(nf_conntrack_hash[i].first,
1072                                         struct nf_conntrack_tuple_hash, hnode);
1073                         hlist_del_rcu(&h->hnode);
1074                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1075                         hlist_add_head(&h->hnode, &hash[bucket]);
1076                 }
1077         }
1078         old_size = nf_conntrack_htable_size;
1079         old_vmalloced = nf_conntrack_vmalloc;
1080         old_hash = nf_conntrack_hash;
1081
1082         nf_conntrack_htable_size = hashsize;
1083         nf_conntrack_vmalloc = vmalloced;
1084         nf_conntrack_hash = hash;
1085         nf_conntrack_hash_rnd = rnd;
1086         spin_unlock_bh(&nf_conntrack_lock);
1087
1088         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1089         return 0;
1090 }
1091 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1092
1093 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1094                   &nf_conntrack_htable_size, 0600);
1095
1096 int __init nf_conntrack_init(void)
1097 {
1098         int max_factor = 8;
1099         int ret;
1100
1101         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1102          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1103         if (!nf_conntrack_htable_size) {
1104                 nf_conntrack_htable_size
1105                         = (((num_physpages << PAGE_SHIFT) / 16384)
1106                            / sizeof(struct hlist_head));
1107                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1108                         nf_conntrack_htable_size = 16384;
1109                 if (nf_conntrack_htable_size < 32)
1110                         nf_conntrack_htable_size = 32;
1111
1112                 /* Use a max. factor of four by default to get the same max as
1113                  * with the old struct list_heads. When a table size is given
1114                  * we use the old value of 8 to avoid reducing the max.
1115                  * entries. */
1116                 max_factor = 4;
1117         }
1118         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1119                                                   &nf_conntrack_vmalloc);
1120         if (!nf_conntrack_hash) {
1121                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1122                 goto err_out;
1123         }
1124
1125         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1126
1127         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1128                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1129                nf_conntrack_max);
1130
1131         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1132                                                 sizeof(struct nf_conn),
1133                                                 0, 0, NULL);
1134         if (!nf_conntrack_cachep) {
1135                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1136                 goto err_free_hash;
1137         }
1138
1139         ret = nf_conntrack_proto_init();
1140         if (ret < 0)
1141                 goto err_free_conntrack_slab;
1142
1143         ret = nf_conntrack_expect_init();
1144         if (ret < 0)
1145                 goto out_fini_proto;
1146
1147         ret = nf_conntrack_helper_init();
1148         if (ret < 0)
1149                 goto out_fini_expect;
1150
1151         /* For use by REJECT target */
1152         rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1153         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1154
1155         /* Set up fake conntrack:
1156             - to never be deleted, not in any hashes */
1157         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1158         /*  - and look it like as a confirmed connection */
1159         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1160
1161         return ret;
1162
1163 out_fini_expect:
1164         nf_conntrack_expect_fini();
1165 out_fini_proto:
1166         nf_conntrack_proto_fini();
1167 err_free_conntrack_slab:
1168         kmem_cache_destroy(nf_conntrack_cachep);
1169 err_free_hash:
1170         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1171                              nf_conntrack_htable_size);
1172 err_out:
1173         return -ENOMEM;
1174 }