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 #include <net/netfilter/nf_nat.h>
  42 #include <net/netfilter/nf_nat_core.h>
  43
  44 #define NF_CONNTRACK_VERSION    "0.5.0"
  45
  46 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  47                                       enum nf_nat_manip_type manip,
  48                                       struct nlattr *attr) __read_mostly;
  49 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  50
  51 DEFINE_SPINLOCK(nf_conntrack_lock);
  52 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  53
  54 unsigned int nf_conntrack_htable_size __read_mostly;
  55 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  56
  57 unsigned int nf_conntrack_max __read_mostly;
  58 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  59
  60 struct nf_conn nf_conntrack_untracked __read_mostly;
  61 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  62
  63 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  64
  65 static int nf_conntrack_hash_rnd_initted;
  66 static unsigned int nf_conntrack_hash_rnd;
  67
  68 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  69                                   unsigned int size, unsigned int rnd)
  70 {
  71         unsigned int n;
  72         u_int32_t h;
  73
  74         /* The direction must be ignored, so we hash everything up to the
  75          * destination ports (which is a multiple of 4) and treat the last
  76          * three bytes manually.
  77          */
  78         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  79         h = jhash2((u32 *)tuple, n,
  80                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
  81                           tuple->dst.protonum));
  82
  83         return ((u64)h * size) >> 32;
  84 }
  85
  86 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  87 {
  88         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  89                                 nf_conntrack_hash_rnd);
  90 }
  91
  92 bool
  93 nf_ct_get_tuple(const struct sk_buff *skb,
  94                 unsigned int nhoff,
  95                 unsigned int dataoff,
  96                 u_int16_t l3num,
  97                 u_int8_t protonum,
  98                 struct nf_conntrack_tuple *tuple,
  99                 const struct nf_conntrack_l3proto *l3proto,
 100                 const struct nf_conntrack_l4proto *l4proto)
 101 {
 102         memset(tuple, 0, sizeof(*tuple));
 103
 104         tuple->src.l3num = l3num;
 105         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 106                 return false;
 107
 108         tuple->dst.protonum = protonum;
 109         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 110
 111         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 112 }
 113 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 114
 115 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 116                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 117 {
 118         struct nf_conntrack_l3proto *l3proto;
 119         struct nf_conntrack_l4proto *l4proto;
 120         unsigned int protoff;
 121         u_int8_t protonum;
 122         int ret;
 123
 124         rcu_read_lock();
 125
 126         l3proto = __nf_ct_l3proto_find(l3num);
 127         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 128         if (ret != NF_ACCEPT) {
 129                 rcu_read_unlock();
 130                 return false;
 131         }
 132
 133         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 134
 135         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 136                               l3proto, l4proto);
 137
 138         rcu_read_unlock();
 139         return ret;
 140 }
 141 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 142
 143 bool
 144 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 145                    const struct nf_conntrack_tuple *orig,
 146                    const struct nf_conntrack_l3proto *l3proto,
 147                    const struct nf_conntrack_l4proto *l4proto)
 148 {
 149         memset(inverse, 0, sizeof(*inverse));
 150
 151         inverse->src.l3num = orig->src.l3num;
 152         if (l3proto->invert_tuple(inverse, orig) == 0)
 153                 return false;
 154
 155         inverse->dst.dir = !orig->dst.dir;
 156
 157         inverse->dst.protonum = orig->dst.protonum;
 158         return l4proto->invert_tuple(inverse, orig);
 159 }
 160 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 161
 162 static void
 163 clean_from_lists(struct nf_conn *ct)
 164 {
 165         pr_debug("clean_from_lists(%p)\n", ct);
 166         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 167         hlist_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 168
 169         /* Destroy all pending expectations */
 170         nf_ct_remove_expectations(ct);
 171 }
 172
 173 static void
 174 destroy_conntrack(struct nf_conntrack *nfct)
 175 {
 176         struct nf_conn *ct = (struct nf_conn *)nfct;
 177         struct net *net = nf_ct_net(ct);
 178         struct nf_conntrack_l4proto *l4proto;
 179
 180         pr_debug("destroy_conntrack(%p)\n", ct);
 181         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 182         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 183
 184         if (!test_bit(IPS_DYING_BIT, &ct->status))
 185                 nf_conntrack_event(IPCT_DESTROY, ct);
 186         set_bit(IPS_DYING_BIT, &ct->status);
 187
 188         /* To make sure we don't get any weird locking issues here:
 189          * destroy_conntrack() MUST NOT be called with a write lock
 190          * to nf_conntrack_lock!!! -HW */
 191         rcu_read_lock();
 192         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 193         if (l4proto && l4proto->destroy)
 194                 l4proto->destroy(ct);
 195
 196         rcu_read_unlock();
 197
 198         spin_lock_bh(&nf_conntrack_lock);
 199         /* Expectations will have been removed in clean_from_lists,
 200          * except TFTP can create an expectation on the first packet,
 201          * before connection is in the list, so we need to clean here,
 202          * too. */
 203         nf_ct_remove_expectations(ct);
 204
 205         /* We overload first tuple to link into unconfirmed list. */
 206         if (!nf_ct_is_confirmed(ct)) {
 207                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
 208                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 209         }
 210
 211         NF_CT_STAT_INC(net, delete);
 212         spin_unlock_bh(&nf_conntrack_lock);
 213
 214         if (ct->master)
 215                 nf_ct_put(ct->master);
 216
 217         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 218         nf_conntrack_free(ct);
 219 }
 220
 221 static void death_by_timeout(unsigned long ul_conntrack)
 222 {
 223         struct nf_conn *ct = (void *)ul_conntrack;
 224         struct net *net = nf_ct_net(ct);
 225         struct nf_conn_help *help = nfct_help(ct);
 226         struct nf_conntrack_helper *helper;
 227
 228         if (help) {
 229                 rcu_read_lock();
 230                 helper = rcu_dereference(help->helper);
 231                 if (helper && helper->destroy)
 232                         helper->destroy(ct);
 233                 rcu_read_unlock();
 234         }
 235
 236         spin_lock_bh(&nf_conntrack_lock);
 237         /* Inside lock so preempt is disabled on module removal path.
 238          * Otherwise we can get spurious warnings. */
 239         NF_CT_STAT_INC(net, delete_list);
 240         clean_from_lists(ct);
 241         spin_unlock_bh(&nf_conntrack_lock);
 242         nf_ct_put(ct);
 243 }
 244
 245 struct nf_conntrack_tuple_hash *
 246 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 247 {
 248         struct nf_conntrack_tuple_hash *h;
 249         struct hlist_node *n;
 250         unsigned int hash = hash_conntrack(tuple);
 251
 252         /* Disable BHs the entire time since we normally need to disable them
 253          * at least once for the stats anyway.
 254          */
 255         local_bh_disable();
 256         hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
 257                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 258                         NF_CT_STAT_INC(net, found);
 259                         local_bh_enable();
 260                         return h;
 261                 }
 262                 NF_CT_STAT_INC(net, searched);
 263         }
 264         local_bh_enable();
 265
 266         return NULL;
 267 }
 268 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 269
 270 /* Find a connection corresponding to a tuple. */
 271 struct nf_conntrack_tuple_hash *
 272 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
 273 {
 274         struct nf_conntrack_tuple_hash *h;
 275         struct nf_conn *ct;
 276
 277         rcu_read_lock();
 278         h = __nf_conntrack_find(net, tuple);
 279         if (h) {
 280                 ct = nf_ct_tuplehash_to_ctrack(h);
 281                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 282                         h = NULL;
 283         }
 284         rcu_read_unlock();
 285
 286         return h;
 287 }
 288 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 289
 290 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 291                                        unsigned int hash,
 292                                        unsigned int repl_hash)
 293 {
 294         struct net *net = nf_ct_net(ct);
 295
 296         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 297                            &net->ct.hash[hash]);
 298         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 299                            &net->ct.hash[repl_hash]);
 300 }
 301
 302 void nf_conntrack_hash_insert(struct nf_conn *ct)
 303 {
 304         unsigned int hash, repl_hash;
 305
 306         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 307         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 308
 309         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 310 }
 311 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 312
 313 /* Confirm a connection given skb; places it in hash table */
 314 int
 315 __nf_conntrack_confirm(struct sk_buff *skb)
 316 {
 317         unsigned int hash, repl_hash;
 318         struct nf_conntrack_tuple_hash *h;
 319         struct nf_conn *ct;
 320         struct nf_conn_help *help;
 321         struct hlist_node *n;
 322         enum ip_conntrack_info ctinfo;
 323         struct net *net;
 324
 325         ct = nf_ct_get(skb, &ctinfo);
 326         net = nf_ct_net(ct);
 327
 328         /* ipt_REJECT uses nf_conntrack_attach to attach related
 329            ICMP/TCP RST packets in other direction.  Actual packet
 330            which created connection will be IP_CT_NEW or for an
 331            expected connection, IP_CT_RELATED. */
 332         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 333                 return NF_ACCEPT;
 334
 335         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 336         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 337
 338         /* We're not in hash table, and we refuse to set up related
 339            connections for unconfirmed conns.  But packet copies and
 340            REJECT will give spurious warnings here. */
 341         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 342
 343         /* No external references means noone else could have
 344            confirmed us. */
 345         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 346         pr_debug("Confirming conntrack %p\n", ct);
 347
 348         spin_lock_bh(&nf_conntrack_lock);
 349
 350         /* See if there's one in the list already, including reverse:
 351            NAT could have grabbed it without realizing, since we're
 352            not in the hash.  If there is, we lost race. */
 353         hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
 354                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 355                                       &h->tuple))
 356                         goto out;
 357         hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
 358                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 359                                       &h->tuple))
 360                         goto out;
 361
 362         /* Remove from unconfirmed list */
 363         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 364
 365         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 366         /* Timer relative to confirmation time, not original
 367            setting time, otherwise we'd get timer wrap in
 368            weird delay cases. */
 369         ct->timeout.expires += jiffies;
 370         add_timer(&ct->timeout);
 371         atomic_inc(&ct->ct_general.use);
 372         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 373         NF_CT_STAT_INC(net, insert);
 374         spin_unlock_bh(&nf_conntrack_lock);
 375         help = nfct_help(ct);
 376         if (help && help->helper)
 377                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 378 #ifdef CONFIG_NF_NAT_NEEDED
 379         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 380             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 381                 nf_conntrack_event_cache(IPCT_NATINFO, ct);
 382 #endif
 383         nf_conntrack_event_cache(master_ct(ct) ?
 384                                  IPCT_RELATED : IPCT_NEW, ct);
 385         return NF_ACCEPT;
 386
 387 out:
 388         NF_CT_STAT_INC(net, insert_failed);
 389         spin_unlock_bh(&nf_conntrack_lock);
 390         return NF_DROP;
 391 }
 392 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 393
 394 /* Returns true if a connection correspondings to the tuple (required
 395    for NAT). */
 396 int
 397 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 398                          const struct nf_conn *ignored_conntrack)
 399 {
 400         struct net *net = nf_ct_net(ignored_conntrack);
 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, &net->ct.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(net, found);
 413                         rcu_read_unlock_bh();
 414                         return 1;
 415                 }
 416                 NF_CT_STAT_INC(net, 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(struct net *net, 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, &net->ct.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(net, 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;
 473
 474         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 475                 get_random_bytes(&nf_conntrack_hash_rnd,
 476                                 sizeof(nf_conntrack_hash_rnd));
 477                 nf_conntrack_hash_rnd_initted = 1;
 478         }
 479
 480         /* We don't want any race condition at early drop stage */
 481         atomic_inc(&net->ct.count);
 482
 483         if (nf_conntrack_max &&
 484             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 485                 unsigned int hash = hash_conntrack(orig);
 486                 if (!early_drop(net, hash)) {
 487                         atomic_dec(&net->ct.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);
 497         if (ct == NULL) {
 498                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 499                 atomic_dec(&net->ct.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 #ifdef CONFIG_NET_NS
 509         ct->ct_net = net;
 510 #endif
 511         INIT_RCU_HEAD(&ct->rcu);
 512
 513         return ct;
 514 }
 515 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 516
 517 static void nf_conntrack_free_rcu(struct rcu_head *head)
 518 {
 519         struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
 520
 521         nf_ct_ext_free(ct);
 522         kmem_cache_free(nf_conntrack_cachep, ct);
 523 }
 524
 525 void nf_conntrack_free(struct nf_conn *ct)
 526 {
 527         struct net *net = nf_ct_net(ct);
 528
 529         nf_ct_ext_destroy(ct);
 530         atomic_dec(&net->ct.count);
 531         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 532 }
 533 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 534
 535 /* Allocate a new conntrack: we return -ENOMEM if classification
 536    failed due to stress.  Otherwise it really is unclassifiable. */
 537 static struct nf_conntrack_tuple_hash *
 538 init_conntrack(struct net *net,
 539                const struct nf_conntrack_tuple *tuple,
 540                struct nf_conntrack_l3proto *l3proto,
 541                struct nf_conntrack_l4proto *l4proto,
 542                struct sk_buff *skb,
 543                unsigned int dataoff)
 544 {
 545         struct nf_conn *ct;
 546         struct nf_conn_help *help;
 547         struct nf_conntrack_tuple repl_tuple;
 548         struct nf_conntrack_expect *exp;
 549
 550         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 551                 pr_debug("Can't invert tuple.\n");
 552                 return NULL;
 553         }
 554
 555         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 556         if (IS_ERR(ct)) {
 557                 pr_debug("Can't allocate conntrack.\n");
 558                 return (struct nf_conntrack_tuple_hash *)ct;
 559         }
 560
 561         if (!l4proto->new(ct, skb, dataoff)) {
 562                 nf_conntrack_free(ct);
 563                 pr_debug("init conntrack: can't track with proto module\n");
 564                 return NULL;
 565         }
 566
 567         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 568
 569         spin_lock_bh(&nf_conntrack_lock);
 570         exp = nf_ct_find_expectation(net, tuple);
 571         if (exp) {
 572                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 573                          ct, exp);
 574                 /* Welcome, Mr. Bond.  We've been expecting you... */
 575                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 576                 ct->master = exp->master;
 577                 if (exp->helper) {
 578                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 579                         if (help)
 580                                 rcu_assign_pointer(help->helper, exp->helper);
 581                 }
 582
 583 #ifdef CONFIG_NF_CONNTRACK_MARK
 584                 ct->mark = exp->master->mark;
 585 #endif
 586 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 587                 ct->secmark = exp->master->secmark;
 588 #endif
 589                 nf_conntrack_get(&ct->master->ct_general);
 590                 NF_CT_STAT_INC(net, expect_new);
 591         } else {
 592                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 593                 NF_CT_STAT_INC(net, new);
 594         }
 595
 596         /* Overload tuple linked list to put us in unconfirmed list. */
 597         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 598                        &net->ct.unconfirmed);
 599
 600         spin_unlock_bh(&nf_conntrack_lock);
 601
 602         if (exp) {
 603                 if (exp->expectfn)
 604                         exp->expectfn(ct, exp);
 605                 nf_ct_expect_put(exp);
 606         }
 607
 608         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 609 }
 610
 611 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 612 static inline struct nf_conn *
 613 resolve_normal_ct(struct net *net,
 614                   struct sk_buff *skb,
 615                   unsigned int dataoff,
 616                   u_int16_t l3num,
 617                   u_int8_t protonum,
 618                   struct nf_conntrack_l3proto *l3proto,
 619                   struct nf_conntrack_l4proto *l4proto,
 620                   int *set_reply,
 621                   enum ip_conntrack_info *ctinfo)
 622 {
 623         struct nf_conntrack_tuple tuple;
 624         struct nf_conntrack_tuple_hash *h;
 625         struct nf_conn *ct;
 626
 627         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 628                              dataoff, l3num, protonum, &tuple, l3proto,
 629                              l4proto)) {
 630                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 631                 return NULL;
 632         }
 633
 634         /* look for tuple match */
 635         h = nf_conntrack_find_get(net, &tuple);
 636         if (!h) {
 637                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 638                 if (!h)
 639                         return NULL;
 640                 if (IS_ERR(h))
 641                         return (void *)h;
 642         }
 643         ct = nf_ct_tuplehash_to_ctrack(h);
 644
 645         /* It exists; we have (non-exclusive) reference. */
 646         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 647                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 648                 /* Please set reply bit if this packet OK */
 649                 *set_reply = 1;
 650         } else {
 651                 /* Once we've had two way comms, always ESTABLISHED. */
 652                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 653                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 654                         *ctinfo = IP_CT_ESTABLISHED;
 655                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 656                         pr_debug("nf_conntrack_in: related packet for %p\n",
 657                                  ct);
 658                         *ctinfo = IP_CT_RELATED;
 659                 } else {
 660                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 661                         *ctinfo = IP_CT_NEW;
 662                 }
 663                 *set_reply = 0;
 664         }
 665         skb->nfct = &ct->ct_general;
 666         skb->nfctinfo = *ctinfo;
 667         return ct;
 668 }
 669
 670 unsigned int
 671 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 672                 struct sk_buff *skb)
 673 {
 674         struct nf_conn *ct;
 675         enum ip_conntrack_info ctinfo;
 676         struct nf_conntrack_l3proto *l3proto;
 677         struct nf_conntrack_l4proto *l4proto;
 678         unsigned int dataoff;
 679         u_int8_t protonum;
 680         int set_reply = 0;
 681         int ret;
 682
 683         /* Previously seen (loopback or untracked)?  Ignore. */
 684         if (skb->nfct) {
 685                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 686                 return NF_ACCEPT;
 687         }
 688
 689         /* rcu_read_lock()ed by nf_hook_slow */
 690         l3proto = __nf_ct_l3proto_find(pf);
 691         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 692                                    &dataoff, &protonum);
 693         if (ret <= 0) {
 694                 pr_debug("not prepared to track yet or error occured\n");
 695                 NF_CT_STAT_INC_ATOMIC(net, error);
 696                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 697                 return -ret;
 698         }
 699
 700         l4proto = __nf_ct_l4proto_find(pf, protonum);
 701
 702         /* It may be an special packet, error, unclean...
 703          * inverse of the return code tells to the netfilter
 704          * core what to do with the packet. */
 705         if (l4proto->error != NULL) {
 706                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 707                 if (ret <= 0) {
 708                         NF_CT_STAT_INC_ATOMIC(net, error);
 709                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 710                         return -ret;
 711                 }
 712         }
 713
 714         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 715                                l3proto, l4proto, &set_reply, &ctinfo);
 716         if (!ct) {
 717                 /* Not valid part of a connection */
 718                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 719                 return NF_ACCEPT;
 720         }
 721
 722         if (IS_ERR(ct)) {
 723                 /* Too stressed to deal. */
 724                 NF_CT_STAT_INC_ATOMIC(net, 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(net, invalid);
 738                 if (ret == -NF_DROP)
 739                         NF_CT_STAT_INC_ATOMIC(net, drop);
 740                 return -ret;
 741         }
 742
 743         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 744                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 745
 746         return ret;
 747 }
 748 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 749
 750 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 751                           const struct nf_conntrack_tuple *orig)
 752 {
 753         bool ret;
 754
 755         rcu_read_lock();
 756         ret = nf_ct_invert_tuple(inverse, orig,
 757                                  __nf_ct_l3proto_find(orig->src.l3num),
 758                                  __nf_ct_l4proto_find(orig->src.l3num,
 759                                                       orig->dst.protonum));
 760         rcu_read_unlock();
 761         return ret;
 762 }
 763 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 764
 765 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 766    implicitly racy: see __nf_conntrack_confirm */
 767 void nf_conntrack_alter_reply(struct nf_conn *ct,
 768                               const struct nf_conntrack_tuple *newreply)
 769 {
 770         struct nf_conn_help *help = nfct_help(ct);
 771
 772         /* Should be unconfirmed, so not in hash table yet */
 773         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 774
 775         pr_debug("Altering reply tuple of %p to ", ct);
 776         nf_ct_dump_tuple(newreply);
 777
 778         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 779         if (ct->master || (help && !hlist_empty(&help->expectations)))
 780                 return;
 781
 782         rcu_read_lock();
 783         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 784         rcu_read_unlock();
 785 }
 786 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 787
 788 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 789 void __nf_ct_refresh_acct(struct nf_conn *ct,
 790                           enum ip_conntrack_info ctinfo,
 791                           const struct sk_buff *skb,
 792                           unsigned long extra_jiffies,
 793                           int do_acct)
 794 {
 795         int event = 0;
 796
 797         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 798         NF_CT_ASSERT(skb);
 799
 800         spin_lock_bh(&nf_conntrack_lock);
 801
 802         /* Only update if this is not a fixed timeout */
 803         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 804                 goto acct;
 805
 806         /* If not in hash table, timer will not be active yet */
 807         if (!nf_ct_is_confirmed(ct)) {
 808                 ct->timeout.expires = extra_jiffies;
 809                 event = IPCT_REFRESH;
 810         } else {
 811                 unsigned long newtime = jiffies + extra_jiffies;
 812
 813                 /* Only update the timeout if the new timeout is at least
 814                    HZ jiffies from the old timeout. Need del_timer for race
 815                    avoidance (may already be dying). */
 816                 if (newtime - ct->timeout.expires >= HZ
 817                     && del_timer(&ct->timeout)) {
 818                         ct->timeout.expires = newtime;
 819                         add_timer(&ct->timeout);
 820                         event = IPCT_REFRESH;
 821                 }
 822         }
 823
 824 acct:
 825         if (do_acct) {
 826                 struct nf_conn_counter *acct;
 827
 828                 acct = nf_conn_acct_find(ct);
 829                 if (acct) {
 830                         acct[CTINFO2DIR(ctinfo)].packets++;
 831                         acct[CTINFO2DIR(ctinfo)].bytes +=
 832                                 skb->len - skb_network_offset(skb);
 833                 }
 834         }
 835
 836         spin_unlock_bh(&nf_conntrack_lock);
 837
 838         /* must be unlocked when calling event cache */
 839         if (event)
 840                 nf_conntrack_event_cache(event, ct);
 841 }
 842 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 843
 844 bool __nf_ct_kill_acct(struct nf_conn *ct,
 845                        enum ip_conntrack_info ctinfo,
 846                        const struct sk_buff *skb,
 847                        int do_acct)
 848 {
 849         if (do_acct) {
 850                 struct nf_conn_counter *acct;
 851
 852                 spin_lock_bh(&nf_conntrack_lock);
 853                 acct = nf_conn_acct_find(ct);
 854                 if (acct) {
 855                         acct[CTINFO2DIR(ctinfo)].packets++;
 856                         acct[CTINFO2DIR(ctinfo)].bytes +=
 857                                 skb->len - skb_network_offset(skb);
 858                 }
 859                 spin_unlock_bh(&nf_conntrack_lock);
 860         }
 861
 862         if (del_timer(&ct->timeout)) {
 863                 ct->timeout.function((unsigned long)ct);
 864                 return true;
 865         }
 866         return false;
 867 }
 868 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 869
 870 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 871
 872 #include <linux/netfilter/nfnetlink.h>
 873 #include <linux/netfilter/nfnetlink_conntrack.h>
 874 #include <linux/mutex.h>
 875
 876 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 877  * in ip_conntrack_core, since we don't want the protocols to autoload
 878  * or depend on ctnetlink */
 879 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 880                                const struct nf_conntrack_tuple *tuple)
 881 {
 882         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 883         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 884         return 0;
 885
 886 nla_put_failure:
 887         return -1;
 888 }
 889 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 890
 891 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 892         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 893         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 894 };
 895 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 896
 897 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 898                                struct nf_conntrack_tuple *t)
 899 {
 900         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 901                 return -EINVAL;
 902
 903         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 904         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 905
 906         return 0;
 907 }
 908 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 909 #endif
 910
 911 /* Used by ipt_REJECT and ip6t_REJECT. */
 912 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 913 {
 914         struct nf_conn *ct;
 915         enum ip_conntrack_info ctinfo;
 916
 917         /* This ICMP is in reverse direction to the packet which caused it */
 918         ct = nf_ct_get(skb, &ctinfo);
 919         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 920                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 921         else
 922                 ctinfo = IP_CT_RELATED;
 923
 924         /* Attach to new skbuff, and increment count */
 925         nskb->nfct = &ct->ct_general;
 926         nskb->nfctinfo = ctinfo;
 927         nf_conntrack_get(nskb->nfct);
 928 }
 929
 930 /* Bring out ya dead! */
 931 static struct nf_conn *
 932 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 933                 void *data, unsigned int *bucket)
 934 {
 935         struct nf_conntrack_tuple_hash *h;
 936         struct nf_conn *ct;
 937         struct hlist_node *n;
 938
 939         spin_lock_bh(&nf_conntrack_lock);
 940         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 941                 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
 942                         ct = nf_ct_tuplehash_to_ctrack(h);
 943                         if (iter(ct, data))
 944                                 goto found;
 945                 }
 946         }
 947         hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
 948                 ct = nf_ct_tuplehash_to_ctrack(h);
 949                 if (iter(ct, data))
 950                         set_bit(IPS_DYING_BIT, &ct->status);
 951         }
 952         spin_unlock_bh(&nf_conntrack_lock);
 953         return NULL;
 954 found:
 955         atomic_inc(&ct->ct_general.use);
 956         spin_unlock_bh(&nf_conntrack_lock);
 957         return ct;
 958 }
 959
 960 void nf_ct_iterate_cleanup(struct net *net,
 961                            int (*iter)(struct nf_conn *i, void *data),
 962                            void *data)
 963 {
 964         struct nf_conn *ct;
 965         unsigned int bucket = 0;
 966
 967         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
 968                 /* Time to push up daises... */
 969                 if (del_timer(&ct->timeout))
 970                         death_by_timeout((unsigned long)ct);
 971                 /* ... else the timer will get him soon. */
 972
 973                 nf_ct_put(ct);
 974         }
 975 }
 976 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 977
 978 struct __nf_ct_flush_report {
 979         u32 pid;
 980         int report;
 981 };
 982
 983 static int kill_all(struct nf_conn *i, void *data)
 984 {
 985         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
 986
 987         /* get_next_corpse sets the dying bit for us */
 988         nf_conntrack_event_report(IPCT_DESTROY,
 989                                   i,
 990                                   fr->pid,
 991                                   fr->report);
 992         return 1;
 993 }
 994
 995 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
 996 {
 997         if (vmalloced)
 998                 vfree(hash);
 999         else
1000                 free_pages((unsigned long)hash,
1001                            get_order(sizeof(struct hlist_head) * size));
1002 }
1003 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1004
1005 void nf_conntrack_flush(struct net *net, u32 pid, int report)
1006 {
1007         struct __nf_ct_flush_report fr = {
1008                 .pid    = pid,
1009                 .report = report,
1010         };
1011         nf_ct_iterate_cleanup(net, kill_all, &fr);
1012 }
1013 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1014
1015 static void nf_conntrack_cleanup_init_net(void)
1016 {
1017         nf_conntrack_helper_fini();
1018         nf_conntrack_proto_fini();
1019         kmem_cache_destroy(nf_conntrack_cachep);
1020 }
1021
1022 static void nf_conntrack_cleanup_net(struct net *net)
1023 {
1024         nf_ct_event_cache_flush(net);
1025         nf_conntrack_ecache_fini(net);
1026  i_see_dead_people:
1027         nf_conntrack_flush(net, 0, 0);
1028         if (atomic_read(&net->ct.count) != 0) {
1029                 schedule();
1030                 goto i_see_dead_people;
1031         }
1032         /* wait until all references to nf_conntrack_untracked are dropped */
1033         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1034                 schedule();
1035
1036         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1037                              nf_conntrack_htable_size);
1038         nf_conntrack_acct_fini(net);
1039         nf_conntrack_expect_fini(net);
1040         free_percpu(net->ct.stat);
1041 }
1042
1043 /* Mishearing the voices in his head, our hero wonders how he's
1044    supposed to kill the mall. */
1045 void nf_conntrack_cleanup(struct net *net)
1046 {
1047         if (net_eq(net, &init_net))
1048                 rcu_assign_pointer(ip_ct_attach, NULL);
1049
1050         /* This makes sure all current packets have passed through
1051            netfilter framework.  Roll on, two-stage module
1052            delete... */
1053         synchronize_net();
1054
1055         nf_conntrack_cleanup_net(net);
1056
1057         if (net_eq(net, &init_net)) {
1058                 rcu_assign_pointer(nf_ct_destroy, NULL);
1059                 nf_conntrack_cleanup_init_net();
1060         }
1061 }
1062
1063 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1064 {
1065         struct hlist_head *hash;
1066         unsigned int size, i;
1067
1068         *vmalloced = 0;
1069
1070         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1071         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1072                                        get_order(sizeof(struct hlist_head)
1073                                                  * size));
1074         if (!hash) {
1075                 *vmalloced = 1;
1076                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1077                 hash = vmalloc(sizeof(struct hlist_head) * size);
1078         }
1079
1080         if (hash)
1081                 for (i = 0; i < size; i++)
1082                         INIT_HLIST_HEAD(&hash[i]);
1083
1084         return hash;
1085 }
1086 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1087
1088 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1089 {
1090         int i, bucket, vmalloced, old_vmalloced;
1091         unsigned int hashsize, old_size;
1092         int rnd;
1093         struct hlist_head *hash, *old_hash;
1094         struct nf_conntrack_tuple_hash *h;
1095
1096         /* On boot, we can set this without any fancy locking. */
1097         if (!nf_conntrack_htable_size)
1098                 return param_set_uint(val, kp);
1099
1100         hashsize = simple_strtoul(val, NULL, 0);
1101         if (!hashsize)
1102                 return -EINVAL;
1103
1104         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1105         if (!hash)
1106                 return -ENOMEM;
1107
1108         /* We have to rehahs for the new table anyway, so we also can
1109          * use a newrandom seed */
1110         get_random_bytes(&rnd, sizeof(rnd));
1111
1112         /* Lookups in the old hash might happen in parallel, which means we
1113          * might get false negatives during connection lookup. New connections
1114          * created because of a false negative won't make it into the hash
1115          * though since that required taking the lock.
1116          */
1117         spin_lock_bh(&nf_conntrack_lock);
1118         for (i = 0; i < nf_conntrack_htable_size; i++) {
1119                 while (!hlist_empty(&init_net.ct.hash[i])) {
1120                         h = hlist_entry(init_net.ct.hash[i].first,
1121                                         struct nf_conntrack_tuple_hash, hnode);
1122                         hlist_del_rcu(&h->hnode);
1123                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1124                         hlist_add_head(&h->hnode, &hash[bucket]);
1125                 }
1126         }
1127         old_size = nf_conntrack_htable_size;
1128         old_vmalloced = init_net.ct.hash_vmalloc;
1129         old_hash = init_net.ct.hash;
1130
1131         nf_conntrack_htable_size = hashsize;
1132         init_net.ct.hash_vmalloc = vmalloced;
1133         init_net.ct.hash = hash;
1134         nf_conntrack_hash_rnd = rnd;
1135         spin_unlock_bh(&nf_conntrack_lock);
1136
1137         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1138         return 0;
1139 }
1140 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1141
1142 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1143                   &nf_conntrack_htable_size, 0600);
1144
1145 static int nf_conntrack_init_init_net(void)
1146 {
1147         int max_factor = 8;
1148         int ret;
1149
1150         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1151          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1152         if (!nf_conntrack_htable_size) {
1153                 nf_conntrack_htable_size
1154                         = (((num_physpages << PAGE_SHIFT) / 16384)
1155                            / sizeof(struct hlist_head));
1156                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1157                         nf_conntrack_htable_size = 16384;
1158                 if (nf_conntrack_htable_size < 32)
1159                         nf_conntrack_htable_size = 32;
1160
1161                 /* Use a max. factor of four by default to get the same max as
1162                  * with the old struct list_heads. When a table size is given
1163                  * we use the old value of 8 to avoid reducing the max.
1164                  * entries. */
1165                 max_factor = 4;
1166         }
1167         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1168
1169         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1170                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1171                nf_conntrack_max);
1172
1173         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1174                                                 sizeof(struct nf_conn),
1175                                                 0, 0, NULL);
1176         if (!nf_conntrack_cachep) {
1177                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1178                 ret = -ENOMEM;
1179                 goto err_cache;
1180         }
1181
1182         ret = nf_conntrack_proto_init();
1183         if (ret < 0)
1184                 goto err_proto;
1185
1186         ret = nf_conntrack_helper_init();
1187         if (ret < 0)
1188                 goto err_helper;
1189
1190         return 0;
1191
1192 err_helper:
1193         nf_conntrack_proto_fini();
1194 err_proto:
1195         kmem_cache_destroy(nf_conntrack_cachep);
1196 err_cache:
1197         return ret;
1198 }
1199
1200 static int nf_conntrack_init_net(struct net *net)
1201 {
1202         int ret;
1203
1204         atomic_set(&net->ct.count, 0);
1205         INIT_HLIST_HEAD(&net->ct.unconfirmed);
1206         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1207         if (!net->ct.stat) {
1208                 ret = -ENOMEM;
1209                 goto err_stat;
1210         }
1211         ret = nf_conntrack_ecache_init(net);
1212         if (ret < 0)
1213                 goto err_ecache;
1214         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1215                                                   &net->ct.hash_vmalloc);
1216         if (!net->ct.hash) {
1217                 ret = -ENOMEM;
1218                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1219                 goto err_hash;
1220         }
1221         ret = nf_conntrack_expect_init(net);
1222         if (ret < 0)
1223                 goto err_expect;
1224         ret = nf_conntrack_acct_init(net);
1225         if (ret < 0)
1226                 goto err_acct;
1227
1228         /* Set up fake conntrack:
1229             - to never be deleted, not in any hashes */
1230 #ifdef CONFIG_NET_NS
1231         nf_conntrack_untracked.ct_net = &init_net;
1232 #endif
1233         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1234         /*  - and look it like as a confirmed connection */
1235         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1236
1237         return 0;
1238
1239 err_acct:
1240         nf_conntrack_expect_fini(net);
1241 err_expect:
1242         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1243                              nf_conntrack_htable_size);
1244 err_hash:
1245         nf_conntrack_ecache_fini(net);
1246 err_ecache:
1247         free_percpu(net->ct.stat);
1248 err_stat:
1249         return ret;
1250 }
1251
1252 int nf_conntrack_init(struct net *net)
1253 {
1254         int ret;
1255
1256         if (net_eq(net, &init_net)) {
1257                 ret = nf_conntrack_init_init_net();
1258                 if (ret < 0)
1259                         goto out_init_net;
1260         }
1261         ret = nf_conntrack_init_net(net);
1262         if (ret < 0)
1263                 goto out_net;
1264
1265         if (net_eq(net, &init_net)) {
1266                 /* For use by REJECT target */
1267                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1268                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1269         }
1270         return 0;
1271
1272 out_net:
1273         if (net_eq(net, &init_net))
1274                 nf_conntrack_cleanup_init_net();
1275 out_init_net:
1276         return ret;
1277 }