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 #include <net/netfilter/nf_conntrack_acct.h>
  41
  42 #define NF_CONNTRACK_VERSION    "0.5.0"
  43
  44 DEFINE_SPINLOCK(nf_conntrack_lock);
  45 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  46
  47 /* nf_conntrack_standalone needs this */
  48 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
  49 EXPORT_SYMBOL_GPL(nf_conntrack_count);
  50
  51 unsigned int nf_conntrack_htable_size __read_mostly;
  52 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  53
  54 int nf_conntrack_max __read_mostly;
  55 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  56
  57 struct hlist_head *nf_conntrack_hash __read_mostly;
  58 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
  59
  60 struct nf_conn nf_conntrack_untracked __read_mostly;
  61 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  62
  63 unsigned int nf_ct_log_invalid __read_mostly;
  64 HLIST_HEAD(unconfirmed);
  65 static int nf_conntrack_vmalloc __read_mostly;
  66 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  67
  68 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
  69 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
  70
  71 static int nf_conntrack_hash_rnd_initted;
  72 static unsigned int nf_conntrack_hash_rnd;
  73
  74 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  75                                   unsigned int size, unsigned int rnd)
  76 {
  77         unsigned int n;
  78         u_int32_t h;
  79
  80         /* The direction must be ignored, so we hash everything up to the
  81          * destination ports (which is a multiple of 4) and treat the last
  82          * three bytes manually.
  83          */
  84         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  85         h = jhash2((u32 *)tuple, n,
  86                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
  87                           tuple->dst.protonum));
  88
  89         return ((u64)h * size) >> 32;
  90 }
  91
  92 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  93 {
  94         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  95                                 nf_conntrack_hash_rnd);
  96 }
  97
  98 bool
  99 nf_ct_get_tuple(const struct sk_buff *skb,
 100                 unsigned int nhoff,
 101                 unsigned int dataoff,
 102                 u_int16_t l3num,
 103                 u_int8_t protonum,
 104                 struct nf_conntrack_tuple *tuple,
 105                 const struct nf_conntrack_l3proto *l3proto,
 106                 const struct nf_conntrack_l4proto *l4proto)
 107 {
 108         memset(tuple, 0, sizeof(*tuple));
 109
 110         tuple->src.l3num = l3num;
 111         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 112                 return false;
 113
 114         tuple->dst.protonum = protonum;
 115         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 116
 117         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 118 }
 119 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 120
 121 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 122                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 123 {
 124         struct nf_conntrack_l3proto *l3proto;
 125         struct nf_conntrack_l4proto *l4proto;
 126         unsigned int protoff;
 127         u_int8_t protonum;
 128         int ret;
 129
 130         rcu_read_lock();
 131
 132         l3proto = __nf_ct_l3proto_find(l3num);
 133         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 134         if (ret != NF_ACCEPT) {
 135                 rcu_read_unlock();
 136                 return false;
 137         }
 138
 139         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 140
 141         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 142                               l3proto, l4proto);
 143
 144         rcu_read_unlock();
 145         return ret;
 146 }
 147 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 148
 149 bool
 150 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 151                    const struct nf_conntrack_tuple *orig,
 152                    const struct nf_conntrack_l3proto *l3proto,
 153                    const struct nf_conntrack_l4proto *l4proto)
 154 {
 155         memset(inverse, 0, sizeof(*inverse));
 156
 157         inverse->src.l3num = orig->src.l3num;
 158         if (l3proto->invert_tuple(inverse, orig) == 0)
 159                 return false;
 160
 161         inverse->dst.dir = !orig->dst.dir;
 162
 163         inverse->dst.protonum = orig->dst.protonum;
 164         return l4proto->invert_tuple(inverse, orig);
 165 }
 166 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 167
 168 static void
 169 clean_from_lists(struct nf_conn *ct)
 170 {
 171         pr_debug("clean_from_lists(%p)\n", ct);
 172         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 173         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 174
 175         /* Destroy all pending expectations */
 176         nf_ct_remove_expectations(ct);
 177 }
 178
 179 static void
 180 destroy_conntrack(struct nf_conntrack *nfct)
 181 {
 182         struct nf_conn *ct = (struct nf_conn *)nfct;
 183         struct nf_conntrack_l4proto *l4proto;
 184
 185         pr_debug("destroy_conntrack(%p)\n", ct);
 186         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 187         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 188
 189         nf_conntrack_event(IPCT_DESTROY, ct);
 190         set_bit(IPS_DYING_BIT, &ct->status);
 191
 192         /* To make sure we don't get any weird locking issues here:
 193          * destroy_conntrack() MUST NOT be called with a write lock
 194          * to nf_conntrack_lock!!! -HW */
 195         rcu_read_lock();
 196         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 197         if (l4proto && l4proto->destroy)
 198                 l4proto->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         /* Disable BHs the entire time since we normally need to disable them
 256          * at least once for the stats anyway.
 257          */
 258         local_bh_disable();
 259         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 260                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 261                         NF_CT_STAT_INC(found);
 262                         local_bh_enable();
 263                         return h;
 264                 }
 265                 NF_CT_STAT_INC(searched);
 266         }
 267         local_bh_enable();
 268
 269         return NULL;
 270 }
 271 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 272
 273 /* Find a connection corresponding to a tuple. */
 274 struct nf_conntrack_tuple_hash *
 275 nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
 276 {
 277         struct nf_conntrack_tuple_hash *h;
 278         struct nf_conn *ct;
 279
 280         rcu_read_lock();
 281         h = __nf_conntrack_find(tuple);
 282         if (h) {
 283                 ct = nf_ct_tuplehash_to_ctrack(h);
 284                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 285                         h = NULL;
 286         }
 287         rcu_read_unlock();
 288
 289         return h;
 290 }
 291 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 292
 293 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 294                                        unsigned int hash,
 295                                        unsigned int repl_hash)
 296 {
 297         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 298                            &nf_conntrack_hash[hash]);
 299         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 300                            &nf_conntrack_hash[repl_hash]);
 301 }
 302
 303 void nf_conntrack_hash_insert(struct nf_conn *ct)
 304 {
 305         unsigned int hash, repl_hash;
 306
 307         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 308         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 309
 310         spin_lock_bh(&nf_conntrack_lock);
 311         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 312         spin_unlock_bh(&nf_conntrack_lock);
 313 }
 314 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 315
 316 /* Confirm a connection given skb; places it in hash table */
 317 int
 318 __nf_conntrack_confirm(struct sk_buff *skb)
 319 {
 320         unsigned int hash, repl_hash;
 321         struct nf_conntrack_tuple_hash *h;
 322         struct nf_conn *ct;
 323         struct nf_conn_help *help;
 324         struct hlist_node *n;
 325         enum ip_conntrack_info ctinfo;
 326
 327         ct = nf_ct_get(skb, &ctinfo);
 328
 329         /* ipt_REJECT uses nf_conntrack_attach to attach related
 330            ICMP/TCP RST packets in other direction.  Actual packet
 331            which created connection will be IP_CT_NEW or for an
 332            expected connection, IP_CT_RELATED. */
 333         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 334                 return NF_ACCEPT;
 335
 336         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 337         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 338
 339         /* We're not in hash table, and we refuse to set up related
 340            connections for unconfirmed conns.  But packet copies and
 341            REJECT will give spurious warnings here. */
 342         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 343
 344         /* No external references means noone else could have
 345            confirmed us. */
 346         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 347         pr_debug("Confirming conntrack %p\n", ct);
 348
 349         spin_lock_bh(&nf_conntrack_lock);
 350
 351         /* See if there's one in the list already, including reverse:
 352            NAT could have grabbed it without realizing, since we're
 353            not in the hash.  If there is, we lost race. */
 354         hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode)
 355                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 356                                       &h->tuple))
 357                         goto out;
 358         hlist_for_each_entry(h, n, &nf_conntrack_hash[repl_hash], hnode)
 359                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 360                                       &h->tuple))
 361                         goto out;
 362
 363         /* Remove from unconfirmed list */
 364         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 365
 366         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 367         /* Timer relative to confirmation time, not original
 368            setting time, otherwise we'd get timer wrap in
 369            weird delay cases. */
 370         ct->timeout.expires += jiffies;
 371         add_timer(&ct->timeout);
 372         atomic_inc(&ct->ct_general.use);
 373         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 374         NF_CT_STAT_INC(insert);
 375         spin_unlock_bh(&nf_conntrack_lock);
 376         help = nfct_help(ct);
 377         if (help && help->helper)
 378                 nf_conntrack_event_cache(IPCT_HELPER, skb);
 379 #ifdef CONFIG_NF_NAT_NEEDED
 380         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 381             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 382                 nf_conntrack_event_cache(IPCT_NATINFO, skb);
 383 #endif
 384         nf_conntrack_event_cache(master_ct(ct) ?
 385                                  IPCT_RELATED : IPCT_NEW, skb);
 386         return NF_ACCEPT;
 387
 388 out:
 389         NF_CT_STAT_INC(insert_failed);
 390         spin_unlock_bh(&nf_conntrack_lock);
 391         return NF_DROP;
 392 }
 393 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 394
 395 /* Returns true if a connection correspondings to the tuple (required
 396    for NAT). */
 397 int
 398 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 399                          const struct nf_conn *ignored_conntrack)
 400 {
 401         struct nf_conntrack_tuple_hash *h;
 402         struct hlist_node *n;
 403         unsigned int hash = hash_conntrack(tuple);
 404
 405         /* Disable BHs the entire time since we need to disable them at
 406          * least once for the stats anyway.
 407          */
 408         rcu_read_lock_bh();
 409         hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
 410                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 411                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 412                         NF_CT_STAT_INC(found);
 413                         rcu_read_unlock_bh();
 414                         return 1;
 415                 }
 416                 NF_CT_STAT_INC(searched);
 417         }
 418         rcu_read_unlock_bh();
 419
 420         return 0;
 421 }
 422 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 423
 424 #define NF_CT_EVICTION_RANGE    8
 425
 426 /* There's a small race here where we may free a just-assured
 427    connection.  Too bad: we're in trouble anyway. */
 428 static noinline int early_drop(unsigned int hash)
 429 {
 430         /* Use oldest entry, which is roughly LRU */
 431         struct nf_conntrack_tuple_hash *h;
 432         struct nf_conn *ct = NULL, *tmp;
 433         struct hlist_node *n;
 434         unsigned int i, cnt = 0;
 435         int dropped = 0;
 436
 437         rcu_read_lock();
 438         for (i = 0; i < nf_conntrack_htable_size; i++) {
 439                 hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash],
 440                                          hnode) {
 441                         tmp = nf_ct_tuplehash_to_ctrack(h);
 442                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 443                                 ct = tmp;
 444                         cnt++;
 445                 }
 446
 447                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 448                         ct = NULL;
 449                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 450                         break;
 451                 hash = (hash + 1) % nf_conntrack_htable_size;
 452         }
 453         rcu_read_unlock();
 454
 455         if (!ct)
 456                 return dropped;
 457
 458         if (del_timer(&ct->timeout)) {
 459                 death_by_timeout((unsigned long)ct);
 460                 dropped = 1;
 461                 NF_CT_STAT_INC_ATOMIC(early_drop);
 462         }
 463         nf_ct_put(ct);
 464         return dropped;
 465 }
 466
 467 struct nf_conn *nf_conntrack_alloc(struct net *net,
 468                                    const struct nf_conntrack_tuple *orig,
 469                                    const struct nf_conntrack_tuple *repl,
 470                                    gfp_t gfp)
 471 {
 472         struct nf_conn *ct = NULL;
 473
 474         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 475                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 476                 nf_conntrack_hash_rnd_initted = 1;
 477         }
 478
 479         /* We don't want any race condition at early drop stage */
 480         atomic_inc(&nf_conntrack_count);
 481
 482         if (nf_conntrack_max &&
 483             unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
 484                 unsigned int hash = hash_conntrack(orig);
 485                 if (!early_drop(hash)) {
 486                         atomic_dec(&nf_conntrack_count);
 487                         if (net_ratelimit())
 488                                 printk(KERN_WARNING
 489                                        "nf_conntrack: table full, dropping"
 490                                        " packet.\n");
 491                         return ERR_PTR(-ENOMEM);
 492                 }
 493         }
 494
 495         ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
 496         if (ct == NULL) {
 497                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 498                 atomic_dec(&nf_conntrack_count);
 499                 return ERR_PTR(-ENOMEM);
 500         }
 501
 502         atomic_set(&ct->ct_general.use, 1);
 503         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 504         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 505         /* Don't set timer yet: wait for confirmation */
 506         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 507 #ifdef CONFIG_NET_NS
 508         ct->ct_net = net;
 509 #endif
 510         INIT_RCU_HEAD(&ct->rcu);
 511
 512         return ct;
 513 }
 514 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 515
 516 static void nf_conntrack_free_rcu(struct rcu_head *head)
 517 {
 518         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
 519
 520         nf_ct_ext_free(ct);
 521         kmem_cache_free(nf_conntrack_cachep, ct);
 522         atomic_dec(&nf_conntrack_count);
 523 }
 524
 525 void nf_conntrack_free(struct nf_conn *ct)
 526 {
 527         nf_ct_ext_destroy(ct);
 528         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 529 }
 530 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 531
 532 /* Allocate a new conntrack: we return -ENOMEM if classification
 533    failed due to stress.  Otherwise it really is unclassifiable. */
 534 static struct nf_conntrack_tuple_hash *
 535 init_conntrack(struct net *net,
 536                const struct nf_conntrack_tuple *tuple,
 537                struct nf_conntrack_l3proto *l3proto,
 538                struct nf_conntrack_l4proto *l4proto,
 539                struct sk_buff *skb,
 540                unsigned int dataoff)
 541 {
 542         struct nf_conn *ct;
 543         struct nf_conn_help *help;
 544         struct nf_conntrack_tuple repl_tuple;
 545         struct nf_conntrack_expect *exp;
 546
 547         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 548                 pr_debug("Can't invert tuple.\n");
 549                 return NULL;
 550         }
 551
 552         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 553         if (ct == NULL || IS_ERR(ct)) {
 554                 pr_debug("Can't allocate conntrack.\n");
 555                 return (struct nf_conntrack_tuple_hash *)ct;
 556         }
 557
 558         if (!l4proto->new(ct, skb, dataoff)) {
 559                 nf_conntrack_free(ct);
 560                 pr_debug("init conntrack: can't track with proto module\n");
 561                 return NULL;
 562         }
 563
 564         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 565
 566         spin_lock_bh(&nf_conntrack_lock);
 567         exp = nf_ct_find_expectation(tuple);
 568         if (exp) {
 569                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 570                          ct, exp);
 571                 /* Welcome, Mr. Bond.  We've been expecting you... */
 572                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 573                 ct->master = exp->master;
 574                 if (exp->helper) {
 575                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 576                         if (help)
 577                                 rcu_assign_pointer(help->helper, exp->helper);
 578                 }
 579
 580 #ifdef CONFIG_NF_CONNTRACK_MARK
 581                 ct->mark = exp->master->mark;
 582 #endif
 583 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 584                 ct->secmark = exp->master->secmark;
 585 #endif
 586                 nf_conntrack_get(&ct->master->ct_general);
 587                 NF_CT_STAT_INC(expect_new);
 588         } else {
 589                 struct nf_conntrack_helper *helper;
 590
 591                 helper = __nf_ct_helper_find(&repl_tuple);
 592                 if (helper) {
 593                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 594                         if (help)
 595                                 rcu_assign_pointer(help->helper, helper);
 596                 }
 597                 NF_CT_STAT_INC(new);
 598         }
 599
 600         /* Overload tuple linked list to put us in unconfirmed list. */
 601         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
 602
 603         spin_unlock_bh(&nf_conntrack_lock);
 604
 605         if (exp) {
 606                 if (exp->expectfn)
 607                         exp->expectfn(ct, exp);
 608                 nf_ct_expect_put(exp);
 609         }
 610
 611         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 612 }
 613
 614 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 615 static inline struct nf_conn *
 616 resolve_normal_ct(struct sk_buff *skb,
 617                   unsigned int dataoff,
 618                   u_int16_t l3num,
 619                   u_int8_t protonum,
 620                   struct nf_conntrack_l3proto *l3proto,
 621                   struct nf_conntrack_l4proto *l4proto,
 622                   int *set_reply,
 623                   enum ip_conntrack_info *ctinfo)
 624 {
 625         struct nf_conntrack_tuple tuple;
 626         struct nf_conntrack_tuple_hash *h;
 627         struct nf_conn *ct;
 628
 629         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 630                              dataoff, l3num, protonum, &tuple, l3proto,
 631                              l4proto)) {
 632                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 633                 return NULL;
 634         }
 635
 636         /* look for tuple match */
 637         h = nf_conntrack_find_get(&tuple);
 638         if (!h) {
 639                 h = init_conntrack(&init_net, &tuple, l3proto, l4proto, skb,
 640                                    dataoff);
 641                 if (!h)
 642                         return NULL;
 643                 if (IS_ERR(h))
 644                         return (void *)h;
 645         }
 646         ct = nf_ct_tuplehash_to_ctrack(h);
 647
 648         /* It exists; we have (non-exclusive) reference. */
 649         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 650                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 651                 /* Please set reply bit if this packet OK */
 652                 *set_reply = 1;
 653         } else {
 654                 /* Once we've had two way comms, always ESTABLISHED. */
 655                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 656                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 657                         *ctinfo = IP_CT_ESTABLISHED;
 658                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 659                         pr_debug("nf_conntrack_in: related packet for %p\n",
 660                                  ct);
 661                         *ctinfo = IP_CT_RELATED;
 662                 } else {
 663                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 664                         *ctinfo = IP_CT_NEW;
 665                 }
 666                 *set_reply = 0;
 667         }
 668         skb->nfct = &ct->ct_general;
 669         skb->nfctinfo = *ctinfo;
 670         return ct;
 671 }
 672
 673 unsigned int
 674 nf_conntrack_in(u_int8_t pf, unsigned int hooknum, struct sk_buff *skb)
 675 {
 676         struct nf_conn *ct;
 677         enum ip_conntrack_info ctinfo;
 678         struct nf_conntrack_l3proto *l3proto;
 679         struct nf_conntrack_l4proto *l4proto;
 680         unsigned int dataoff;
 681         u_int8_t protonum;
 682         int set_reply = 0;
 683         int ret;
 684
 685         /* Previously seen (loopback or untracked)?  Ignore. */
 686         if (skb->nfct) {
 687                 NF_CT_STAT_INC_ATOMIC(ignore);
 688                 return NF_ACCEPT;
 689         }
 690
 691         /* rcu_read_lock()ed by nf_hook_slow */
 692         l3proto = __nf_ct_l3proto_find(pf);
 693         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 694                                    &dataoff, &protonum);
 695         if (ret <= 0) {
 696                 pr_debug("not prepared to track yet or error occured\n");
 697                 NF_CT_STAT_INC_ATOMIC(error);
 698                 NF_CT_STAT_INC_ATOMIC(invalid);
 699                 return -ret;
 700         }
 701
 702         l4proto = __nf_ct_l4proto_find(pf, protonum);
 703
 704         /* It may be an special packet, error, unclean...
 705          * inverse of the return code tells to the netfilter
 706          * core what to do with the packet. */
 707         if (l4proto->error != NULL &&
 708             (ret = l4proto->error(skb, dataoff, &ctinfo, pf, hooknum)) <= 0) {
 709                 NF_CT_STAT_INC_ATOMIC(error);
 710                 NF_CT_STAT_INC_ATOMIC(invalid);
 711                 return -ret;
 712         }
 713
 714         ct = resolve_normal_ct(skb, dataoff, pf, protonum, l3proto, l4proto,
 715                                &set_reply, &ctinfo);
 716         if (!ct) {
 717                 /* Not valid part of a connection */
 718                 NF_CT_STAT_INC_ATOMIC(invalid);
 719                 return NF_ACCEPT;
 720         }
 721
 722         if (IS_ERR(ct)) {
 723                 /* Too stressed to deal. */
 724                 NF_CT_STAT_INC_ATOMIC(drop);
 725                 return NF_DROP;
 726         }
 727
 728         NF_CT_ASSERT(skb->nfct);
 729
 730         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 731         if (ret < 0) {
 732                 /* Invalid: inverse of the return code tells
 733                  * the netfilter core what to do */
 734                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 735                 nf_conntrack_put(skb->nfct);
 736                 skb->nfct = NULL;
 737                 NF_CT_STAT_INC_ATOMIC(invalid);
 738                 return -ret;
 739         }
 740
 741         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 742                 nf_conntrack_event_cache(IPCT_STATUS, skb);
 743
 744         return ret;
 745 }
 746 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 747
 748 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 749                           const struct nf_conntrack_tuple *orig)
 750 {
 751         bool ret;
 752
 753         rcu_read_lock();
 754         ret = nf_ct_invert_tuple(inverse, orig,
 755                                  __nf_ct_l3proto_find(orig->src.l3num),
 756                                  __nf_ct_l4proto_find(orig->src.l3num,
 757                                                       orig->dst.protonum));
 758         rcu_read_unlock();
 759         return ret;
 760 }
 761 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 762
 763 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 764    implicitly racy: see __nf_conntrack_confirm */
 765 void nf_conntrack_alter_reply(struct nf_conn *ct,
 766                               const struct nf_conntrack_tuple *newreply)
 767 {
 768         struct nf_conn_help *help = nfct_help(ct);
 769         struct nf_conntrack_helper *helper;
 770
 771         /* Should be unconfirmed, so not in hash table yet */
 772         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 773
 774         pr_debug("Altering reply tuple of %p to ", ct);
 775         nf_ct_dump_tuple(newreply);
 776
 777         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 778         if (ct->master || (help && !hlist_empty(&help->expectations)))
 779                 return;
 780
 781         rcu_read_lock();
 782         helper = __nf_ct_helper_find(newreply);
 783         if (helper == NULL) {
 784                 if (help)
 785                         rcu_assign_pointer(help->helper, NULL);
 786                 goto out;
 787         }
 788
 789         if (help == NULL) {
 790                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 791                 if (help == NULL)
 792                         goto out;
 793         } else {
 794                 memset(&help->help, 0, sizeof(help->help));
 795         }
 796
 797         rcu_assign_pointer(help->helper, helper);
 798 out:
 799         rcu_read_unlock();
 800 }
 801 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 802
 803 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 804 void __nf_ct_refresh_acct(struct nf_conn *ct,
 805                           enum ip_conntrack_info ctinfo,
 806                           const struct sk_buff *skb,
 807                           unsigned long extra_jiffies,
 808                           int do_acct)
 809 {
 810         int event = 0;
 811
 812         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 813         NF_CT_ASSERT(skb);
 814
 815         spin_lock_bh(&nf_conntrack_lock);
 816
 817         /* Only update if this is not a fixed timeout */
 818         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 819                 goto acct;
 820
 821         /* If not in hash table, timer will not be active yet */
 822         if (!nf_ct_is_confirmed(ct)) {
 823                 ct->timeout.expires = extra_jiffies;
 824                 event = IPCT_REFRESH;
 825         } else {
 826                 unsigned long newtime = jiffies + extra_jiffies;
 827
 828                 /* Only update the timeout if the new timeout is at least
 829                    HZ jiffies from the old timeout. Need del_timer for race
 830                    avoidance (may already be dying). */
 831                 if (newtime - ct->timeout.expires >= HZ
 832                     && del_timer(&ct->timeout)) {
 833                         ct->timeout.expires = newtime;
 834                         add_timer(&ct->timeout);
 835                         event = IPCT_REFRESH;
 836                 }
 837         }
 838
 839 acct:
 840         if (do_acct) {
 841                 struct nf_conn_counter *acct;
 842
 843                 acct = nf_conn_acct_find(ct);
 844                 if (acct) {
 845                         acct[CTINFO2DIR(ctinfo)].packets++;
 846                         acct[CTINFO2DIR(ctinfo)].bytes +=
 847                                 skb->len - skb_network_offset(skb);
 848                 }
 849         }
 850
 851         spin_unlock_bh(&nf_conntrack_lock);
 852
 853         /* must be unlocked when calling event cache */
 854         if (event)
 855                 nf_conntrack_event_cache(event, skb);
 856 }
 857 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 858
 859 bool __nf_ct_kill_acct(struct nf_conn *ct,
 860                        enum ip_conntrack_info ctinfo,
 861                        const struct sk_buff *skb,
 862                        int do_acct)
 863 {
 864         if (do_acct) {
 865                 struct nf_conn_counter *acct;
 866
 867                 spin_lock_bh(&nf_conntrack_lock);
 868                 acct = nf_conn_acct_find(ct);
 869                 if (acct) {
 870                         acct[CTINFO2DIR(ctinfo)].packets++;
 871                         acct[CTINFO2DIR(ctinfo)].bytes +=
 872                                 skb->len - skb_network_offset(skb);
 873                 }
 874                 spin_unlock_bh(&nf_conntrack_lock);
 875         }
 876
 877         if (del_timer(&ct->timeout)) {
 878                 ct->timeout.function((unsigned long)ct);
 879                 return true;
 880         }
 881         return false;
 882 }
 883 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 884
 885 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 886
 887 #include <linux/netfilter/nfnetlink.h>
 888 #include <linux/netfilter/nfnetlink_conntrack.h>
 889 #include <linux/mutex.h>
 890
 891 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 892  * in ip_conntrack_core, since we don't want the protocols to autoload
 893  * or depend on ctnetlink */
 894 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 895                                const struct nf_conntrack_tuple *tuple)
 896 {
 897         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 898         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 899         return 0;
 900
 901 nla_put_failure:
 902         return -1;
 903 }
 904 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 905
 906 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 907         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 908         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 909 };
 910 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 911
 912 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 913                                struct nf_conntrack_tuple *t)
 914 {
 915         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 916                 return -EINVAL;
 917
 918         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 919         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 920
 921         return 0;
 922 }
 923 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 924 #endif
 925
 926 /* Used by ipt_REJECT and ip6t_REJECT. */
 927 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 928 {
 929         struct nf_conn *ct;
 930         enum ip_conntrack_info ctinfo;
 931
 932         /* This ICMP is in reverse direction to the packet which caused it */
 933         ct = nf_ct_get(skb, &ctinfo);
 934         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 935                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 936         else
 937                 ctinfo = IP_CT_RELATED;
 938
 939         /* Attach to new skbuff, and increment count */
 940         nskb->nfct = &ct->ct_general;
 941         nskb->nfctinfo = ctinfo;
 942         nf_conntrack_get(nskb->nfct);
 943 }
 944
 945 /* Bring out ya dead! */
 946 static struct nf_conn *
 947 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
 948                 void *data, unsigned int *bucket)
 949 {
 950         struct nf_conntrack_tuple_hash *h;
 951         struct nf_conn *ct;
 952         struct hlist_node *n;
 953
 954         spin_lock_bh(&nf_conntrack_lock);
 955         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 956                 hlist_for_each_entry(h, n, &nf_conntrack_hash[*bucket], hnode) {
 957                         ct = nf_ct_tuplehash_to_ctrack(h);
 958                         if (iter(ct, data))
 959                                 goto found;
 960                 }
 961         }
 962         hlist_for_each_entry(h, n, &unconfirmed, hnode) {
 963                 ct = nf_ct_tuplehash_to_ctrack(h);
 964                 if (iter(ct, data))
 965                         set_bit(IPS_DYING_BIT, &ct->status);
 966         }
 967         spin_unlock_bh(&nf_conntrack_lock);
 968         return NULL;
 969 found:
 970         atomic_inc(&ct->ct_general.use);
 971         spin_unlock_bh(&nf_conntrack_lock);
 972         return ct;
 973 }
 974
 975 void
 976 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 977 {
 978         struct nf_conn *ct;
 979         unsigned int bucket = 0;
 980
 981         while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
 982                 /* Time to push up daises... */
 983                 if (del_timer(&ct->timeout))
 984                         death_by_timeout((unsigned long)ct);
 985                 /* ... else the timer will get him soon. */
 986
 987                 nf_ct_put(ct);
 988         }
 989 }
 990 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 991
 992 static int kill_all(struct nf_conn *i, void *data)
 993 {
 994         return 1;
 995 }
 996
 997 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
 998 {
 999         if (vmalloced)
1000                 vfree(hash);
1001         else
1002                 free_pages((unsigned long)hash,
1003                            get_order(sizeof(struct hlist_head) * size));
1004 }
1005 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1006
1007 void nf_conntrack_flush(void)
1008 {
1009         nf_ct_iterate_cleanup(kill_all, NULL);
1010 }
1011 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1012
1013 /* Mishearing the voices in his head, our hero wonders how he's
1014    supposed to kill the mall. */
1015 void nf_conntrack_cleanup(struct net *net)
1016 {
1017         rcu_assign_pointer(ip_ct_attach, NULL);
1018
1019         /* This makes sure all current packets have passed through
1020            netfilter framework.  Roll on, two-stage module
1021            delete... */
1022         synchronize_net();
1023
1024         nf_ct_event_cache_flush();
1025  i_see_dead_people:
1026         nf_conntrack_flush();
1027         if (atomic_read(&nf_conntrack_count) != 0) {
1028                 schedule();
1029                 goto i_see_dead_people;
1030         }
1031         /* wait until all references to nf_conntrack_untracked are dropped */
1032         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1033                 schedule();
1034
1035         rcu_assign_pointer(nf_ct_destroy, NULL);
1036
1037         kmem_cache_destroy(nf_conntrack_cachep);
1038         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1039                              nf_conntrack_htable_size);
1040
1041         nf_conntrack_acct_fini();
1042         nf_conntrack_expect_fini();
1043         nf_conntrack_helper_fini();
1044         nf_conntrack_proto_fini();
1045 }
1046
1047 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1048 {
1049         struct hlist_head *hash;
1050         unsigned int size, i;
1051
1052         *vmalloced = 0;
1053
1054         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1055         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1056                                        get_order(sizeof(struct hlist_head)
1057                                                  * size));
1058         if (!hash) {
1059                 *vmalloced = 1;
1060                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1061                 hash = vmalloc(sizeof(struct hlist_head) * size);
1062         }
1063
1064         if (hash)
1065                 for (i = 0; i < size; i++)
1066                         INIT_HLIST_HEAD(&hash[i]);
1067
1068         return hash;
1069 }
1070 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1071
1072 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1073 {
1074         int i, bucket, vmalloced, old_vmalloced;
1075         unsigned int hashsize, old_size;
1076         int rnd;
1077         struct hlist_head *hash, *old_hash;
1078         struct nf_conntrack_tuple_hash *h;
1079
1080         /* On boot, we can set this without any fancy locking. */
1081         if (!nf_conntrack_htable_size)
1082                 return param_set_uint(val, kp);
1083
1084         hashsize = simple_strtoul(val, NULL, 0);
1085         if (!hashsize)
1086                 return -EINVAL;
1087
1088         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1089         if (!hash)
1090                 return -ENOMEM;
1091
1092         /* We have to rehahs for the new table anyway, so we also can
1093          * use a newrandom seed */
1094         get_random_bytes(&rnd, 4);
1095
1096         /* Lookups in the old hash might happen in parallel, which means we
1097          * might get false negatives during connection lookup. New connections
1098          * created because of a false negative won't make it into the hash
1099          * though since that required taking the lock.
1100          */
1101         spin_lock_bh(&nf_conntrack_lock);
1102         for (i = 0; i < nf_conntrack_htable_size; i++) {
1103                 while (!hlist_empty(&nf_conntrack_hash[i])) {
1104                         h = hlist_entry(nf_conntrack_hash[i].first,
1105                                         struct nf_conntrack_tuple_hash, hnode);
1106                         hlist_del_rcu(&h->hnode);
1107                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1108                         hlist_add_head(&h->hnode, &hash[bucket]);
1109                 }
1110         }
1111         old_size = nf_conntrack_htable_size;
1112         old_vmalloced = nf_conntrack_vmalloc;
1113         old_hash = nf_conntrack_hash;
1114
1115         nf_conntrack_htable_size = hashsize;
1116         nf_conntrack_vmalloc = vmalloced;
1117         nf_conntrack_hash = hash;
1118         nf_conntrack_hash_rnd = rnd;
1119         spin_unlock_bh(&nf_conntrack_lock);
1120
1121         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1122         return 0;
1123 }
1124 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1125
1126 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1127                   &nf_conntrack_htable_size, 0600);
1128
1129 int nf_conntrack_init(struct net *net)
1130 {
1131         int max_factor = 8;
1132         int ret;
1133
1134         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1135          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1136         if (!nf_conntrack_htable_size) {
1137                 nf_conntrack_htable_size
1138                         = (((num_physpages << PAGE_SHIFT) / 16384)
1139                            / sizeof(struct hlist_head));
1140                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1141                         nf_conntrack_htable_size = 16384;
1142                 if (nf_conntrack_htable_size < 32)
1143                         nf_conntrack_htable_size = 32;
1144
1145                 /* Use a max. factor of four by default to get the same max as
1146                  * with the old struct list_heads. When a table size is given
1147                  * we use the old value of 8 to avoid reducing the max.
1148                  * entries. */
1149                 max_factor = 4;
1150         }
1151         nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1152                                                   &nf_conntrack_vmalloc);
1153         if (!nf_conntrack_hash) {
1154                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1155                 goto err_out;
1156         }
1157
1158         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1159
1160         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1161                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1162                nf_conntrack_max);
1163
1164         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1165                                                 sizeof(struct nf_conn),
1166                                                 0, 0, NULL);
1167         if (!nf_conntrack_cachep) {
1168                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1169                 goto err_free_hash;
1170         }
1171
1172         ret = nf_conntrack_proto_init();
1173         if (ret < 0)
1174                 goto err_free_conntrack_slab;
1175
1176         ret = nf_conntrack_expect_init();
1177         if (ret < 0)
1178                 goto out_fini_proto;
1179
1180         ret = nf_conntrack_helper_init();
1181         if (ret < 0)
1182                 goto out_fini_expect;
1183
1184         ret = nf_conntrack_acct_init();
1185         if (ret < 0)
1186                 goto out_fini_helper;
1187
1188         /* For use by REJECT target */
1189         rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1190         rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1191
1192         /* Set up fake conntrack:
1193             - to never be deleted, not in any hashes */
1194 #ifdef CONFIG_NET_NS
1195         nf_conntrack_untracked.ct_net = &init_net;
1196 #endif
1197         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1198         /*  - and look it like as a confirmed connection */
1199         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1200
1201         return ret;
1202
1203 out_fini_helper:
1204         nf_conntrack_helper_fini();
1205 out_fini_expect:
1206         nf_conntrack_expect_fini();
1207 out_fini_proto:
1208         nf_conntrack_proto_fini();
1209 err_free_conntrack_slab:
1210         kmem_cache_destroy(nf_conntrack_cachep);
1211 err_free_hash:
1212         nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
1213                              nf_conntrack_htable_size);
1214 err_out:
1215         return -ENOMEM;
1216 }