git.oblomov.eu Git - linux-2.6/blob - 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         struct net *net = nf_ct_net(ct);
 521
 522         nf_ct_ext_free(ct);
 523         kmem_cache_free(nf_conntrack_cachep, ct);
 524         atomic_dec(&net->ct.count);
 525 }
 526
 527 void nf_conntrack_free(struct nf_conn *ct)
 528 {
 529         nf_ct_ext_destroy(ct);
 530         call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 531 }
 532 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 533
 534 /* Allocate a new conntrack: we return -ENOMEM if classification
 535    failed due to stress.  Otherwise it really is unclassifiable. */
 536 static struct nf_conntrack_tuple_hash *
 537 init_conntrack(struct net *net,
 538                const struct nf_conntrack_tuple *tuple,
 539                struct nf_conntrack_l3proto *l3proto,
 540                struct nf_conntrack_l4proto *l4proto,
 541                struct sk_buff *skb,
 542                unsigned int dataoff)
 543 {
 544         struct nf_conn *ct;
 545         struct nf_conn_help *help;
 546         struct nf_conntrack_tuple repl_tuple;
 547         struct nf_conntrack_expect *exp;
 548
 549         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 550                 pr_debug("Can't invert tuple.\n");
 551                 return NULL;
 552         }
 553
 554         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 555         if (IS_ERR(ct)) {
 556                 pr_debug("Can't allocate conntrack.\n");
 557                 return (struct nf_conntrack_tuple_hash *)ct;
 558         }
 559
 560         if (!l4proto->new(ct, skb, dataoff)) {
 561                 nf_conntrack_free(ct);
 562                 pr_debug("init conntrack: can't track with proto module\n");
 563                 return NULL;
 564         }
 565
 566         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 567
 568         spin_lock_bh(&nf_conntrack_lock);
 569         exp = nf_ct_find_expectation(net, tuple);
 570         if (exp) {
 571                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 572                          ct, exp);
 573                 /* Welcome, Mr. Bond.  We've been expecting you... */
 574                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 575                 ct->master = exp->master;
 576                 if (exp->helper) {
 577                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 578                         if (help)
 579                                 rcu_assign_pointer(help->helper, exp->helper);
 580                 }
 581
 582 #ifdef CONFIG_NF_CONNTRACK_MARK
 583                 ct->mark = exp->master->mark;
 584 #endif
 585 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 586                 ct->secmark = exp->master->secmark;
 587 #endif
 588                 nf_conntrack_get(&ct->master->ct_general);
 589                 NF_CT_STAT_INC(net, expect_new);
 590         } else {
 591                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 592                 NF_CT_STAT_INC(net, new);
 593         }
 594
 595         /* Overload tuple linked list to put us in unconfirmed list. */
 596         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 597                        &net->ct.unconfirmed);
 598
 599         spin_unlock_bh(&nf_conntrack_lock);
 600
 601         if (exp) {
 602                 if (exp->expectfn)
 603                         exp->expectfn(ct, exp);
 604                 nf_ct_expect_put(exp);
 605         }
 606
 607         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 608 }
 609
 610 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 611 static inline struct nf_conn *
 612 resolve_normal_ct(struct net *net,
 613                   struct sk_buff *skb,
 614                   unsigned int dataoff,
 615                   u_int16_t l3num,
 616                   u_int8_t protonum,
 617                   struct nf_conntrack_l3proto *l3proto,
 618                   struct nf_conntrack_l4proto *l4proto,
 619                   int *set_reply,
 620                   enum ip_conntrack_info *ctinfo)
 621 {
 622         struct nf_conntrack_tuple tuple;
 623         struct nf_conntrack_tuple_hash *h;
 624         struct nf_conn *ct;
 625
 626         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 627                              dataoff, l3num, protonum, &tuple, l3proto,
 628                              l4proto)) {
 629                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 630                 return NULL;
 631         }
 632
 633         /* look for tuple match */
 634         h = nf_conntrack_find_get(net, &tuple);
 635         if (!h) {
 636                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 637                 if (!h)
 638                         return NULL;
 639                 if (IS_ERR(h))
 640                         return (void *)h;
 641         }
 642         ct = nf_ct_tuplehash_to_ctrack(h);
 643
 644         /* It exists; we have (non-exclusive) reference. */
 645         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 646                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 647                 /* Please set reply bit if this packet OK */
 648                 *set_reply = 1;
 649         } else {
 650                 /* Once we've had two way comms, always ESTABLISHED. */
 651                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 652                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 653                         *ctinfo = IP_CT_ESTABLISHED;
 654                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 655                         pr_debug("nf_conntrack_in: related packet for %p\n",
 656                                  ct);
 657                         *ctinfo = IP_CT_RELATED;
 658                 } else {
 659                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 660                         *ctinfo = IP_CT_NEW;
 661                 }
 662                 *set_reply = 0;
 663         }
 664         skb->nfct = &ct->ct_general;
 665         skb->nfctinfo = *ctinfo;
 666         return ct;
 667 }
 668
 669 unsigned int
 670 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 671                 struct sk_buff *skb)
 672 {
 673         struct nf_conn *ct;
 674         enum ip_conntrack_info ctinfo;
 675         struct nf_conntrack_l3proto *l3proto;
 676         struct nf_conntrack_l4proto *l4proto;
 677         unsigned int dataoff;
 678         u_int8_t protonum;
 679         int set_reply = 0;
 680         int ret;
 681
 682         /* Previously seen (loopback or untracked)?  Ignore. */
 683         if (skb->nfct) {
 684                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 685                 return NF_ACCEPT;
 686         }
 687
 688         /* rcu_read_lock()ed by nf_hook_slow */
 689         l3proto = __nf_ct_l3proto_find(pf);
 690         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 691                                    &dataoff, &protonum);
 692         if (ret <= 0) {
 693                 pr_debug("not prepared to track yet or error occured\n");
 694                 NF_CT_STAT_INC_ATOMIC(net, error);
 695                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 696                 return -ret;
 697         }
 698
 699         l4proto = __nf_ct_l4proto_find(pf, protonum);
 700
 701         /* It may be an special packet, error, unclean...
 702          * inverse of the return code tells to the netfilter
 703          * core what to do with the packet. */
 704         if (l4proto->error != NULL) {
 705                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 706                 if (ret <= 0) {
 707                         NF_CT_STAT_INC_ATOMIC(net, error);
 708                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 709                         return -ret;
 710                 }
 711         }
 712
 713         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 714                                l3proto, l4proto, &set_reply, &ctinfo);
 715         if (!ct) {
 716                 /* Not valid part of a connection */
 717                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 718                 return NF_ACCEPT;
 719         }
 720
 721         if (IS_ERR(ct)) {
 722                 /* Too stressed to deal. */
 723                 NF_CT_STAT_INC_ATOMIC(net, drop);
 724                 return NF_DROP;
 725         }
 726
 727         NF_CT_ASSERT(skb->nfct);
 728
 729         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 730         if (ret < 0) {
 731                 /* Invalid: inverse of the return code tells
 732                  * the netfilter core what to do */
 733                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 734                 nf_conntrack_put(skb->nfct);
 735                 skb->nfct = NULL;
 736                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 737                 if (ret == -NF_DROP)
 738                         NF_CT_STAT_INC_ATOMIC(net, drop);
 739                 return -ret;
 740         }
 741
 742         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 743                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 744
 745         return ret;
 746 }
 747 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 748
 749 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 750                           const struct nf_conntrack_tuple *orig)
 751 {
 752         bool ret;
 753
 754         rcu_read_lock();
 755         ret = nf_ct_invert_tuple(inverse, orig,
 756                                  __nf_ct_l3proto_find(orig->src.l3num),
 757                                  __nf_ct_l4proto_find(orig->src.l3num,
 758                                                       orig->dst.protonum));
 759         rcu_read_unlock();
 760         return ret;
 761 }
 762 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 763
 764 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 765    implicitly racy: see __nf_conntrack_confirm */
 766 void nf_conntrack_alter_reply(struct nf_conn *ct,
 767                               const struct nf_conntrack_tuple *newreply)
 768 {
 769         struct nf_conn_help *help = nfct_help(ct);
 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         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 783         rcu_read_unlock();
 784 }
 785 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 786
 787 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 788 void __nf_ct_refresh_acct(struct nf_conn *ct,
 789                           enum ip_conntrack_info ctinfo,
 790                           const struct sk_buff *skb,
 791                           unsigned long extra_jiffies,
 792                           int do_acct)
 793 {
 794         int event = 0;
 795
 796         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 797         NF_CT_ASSERT(skb);
 798
 799         spin_lock_bh(&nf_conntrack_lock);
 800
 801         /* Only update if this is not a fixed timeout */
 802         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 803                 goto acct;
 804
 805         /* If not in hash table, timer will not be active yet */
 806         if (!nf_ct_is_confirmed(ct)) {
 807                 ct->timeout.expires = extra_jiffies;
 808                 event = IPCT_REFRESH;
 809         } else {
 810                 unsigned long newtime = jiffies + extra_jiffies;
 811
 812                 /* Only update the timeout if the new timeout is at least
 813                    HZ jiffies from the old timeout. Need del_timer for race
 814                    avoidance (may already be dying). */
 815                 if (newtime - ct->timeout.expires >= HZ
 816                     && del_timer(&ct->timeout)) {
 817                         ct->timeout.expires = newtime;
 818                         add_timer(&ct->timeout);
 819                         event = IPCT_REFRESH;
 820                 }
 821         }
 822
 823 acct:
 824         if (do_acct) {
 825                 struct nf_conn_counter *acct;
 826
 827                 acct = nf_conn_acct_find(ct);
 828                 if (acct) {
 829                         acct[CTINFO2DIR(ctinfo)].packets++;
 830                         acct[CTINFO2DIR(ctinfo)].bytes +=
 831                                 skb->len - skb_network_offset(skb);
 832                 }
 833         }
 834
 835         spin_unlock_bh(&nf_conntrack_lock);
 836
 837         /* must be unlocked when calling event cache */
 838         if (event)
 839                 nf_conntrack_event_cache(event, ct);
 840 }
 841 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 842
 843 bool __nf_ct_kill_acct(struct nf_conn *ct,
 844                        enum ip_conntrack_info ctinfo,
 845                        const struct sk_buff *skb,
 846                        int do_acct)
 847 {
 848         if (do_acct) {
 849                 struct nf_conn_counter *acct;
 850
 851                 spin_lock_bh(&nf_conntrack_lock);
 852                 acct = nf_conn_acct_find(ct);
 853                 if (acct) {
 854                         acct[CTINFO2DIR(ctinfo)].packets++;
 855                         acct[CTINFO2DIR(ctinfo)].bytes +=
 856                                 skb->len - skb_network_offset(skb);
 857                 }
 858                 spin_unlock_bh(&nf_conntrack_lock);
 859         }
 860
 861         if (del_timer(&ct->timeout)) {
 862                 ct->timeout.function((unsigned long)ct);
 863                 return true;
 864         }
 865         return false;
 866 }
 867 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 868
 869 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 870
 871 #include <linux/netfilter/nfnetlink.h>
 872 #include <linux/netfilter/nfnetlink_conntrack.h>
 873 #include <linux/mutex.h>
 874
 875 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 876  * in ip_conntrack_core, since we don't want the protocols to autoload
 877  * or depend on ctnetlink */
 878 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 879                                const struct nf_conntrack_tuple *tuple)
 880 {
 881         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 882         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 883         return 0;
 884
 885 nla_put_failure:
 886         return -1;
 887 }
 888 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 889
 890 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 891         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 892         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 893 };
 894 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 895
 896 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 897                                struct nf_conntrack_tuple *t)
 898 {
 899         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 900                 return -EINVAL;
 901
 902         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 903         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 904
 905         return 0;
 906 }
 907 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 908 #endif
 909
 910 /* Used by ipt_REJECT and ip6t_REJECT. */
 911 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 912 {
 913         struct nf_conn *ct;
 914         enum ip_conntrack_info ctinfo;
 915
 916         /* This ICMP is in reverse direction to the packet which caused it */
 917         ct = nf_ct_get(skb, &ctinfo);
 918         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 919                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 920         else
 921                 ctinfo = IP_CT_RELATED;
 922
 923         /* Attach to new skbuff, and increment count */
 924         nskb->nfct = &ct->ct_general;
 925         nskb->nfctinfo = ctinfo;
 926         nf_conntrack_get(nskb->nfct);
 927 }
 928
 929 /* Bring out ya dead! */
 930 static struct nf_conn *
 931 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 932                 void *data, unsigned int *bucket)
 933 {
 934         struct nf_conntrack_tuple_hash *h;
 935         struct nf_conn *ct;
 936         struct hlist_node *n;
 937
 938         spin_lock_bh(&nf_conntrack_lock);
 939         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 940                 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
 941                         ct = nf_ct_tuplehash_to_ctrack(h);
 942                         if (iter(ct, data))
 943                                 goto found;
 944                 }
 945         }
 946         hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
 947                 ct = nf_ct_tuplehash_to_ctrack(h);
 948                 if (iter(ct, data))
 949                         set_bit(IPS_DYING_BIT, &ct->status);
 950         }
 951         spin_unlock_bh(&nf_conntrack_lock);
 952         return NULL;
 953 found:
 954         atomic_inc(&ct->ct_general.use);
 955         spin_unlock_bh(&nf_conntrack_lock);
 956         return ct;
 957 }
 958
 959 void nf_ct_iterate_cleanup(struct net *net,
 960                            int (*iter)(struct nf_conn *i, void *data),
 961                            void *data)
 962 {
 963         struct nf_conn *ct;
 964         unsigned int bucket = 0;
 965
 966         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
 967                 /* Time to push up daises... */
 968                 if (del_timer(&ct->timeout))
 969                         death_by_timeout((unsigned long)ct);
 970                 /* ... else the timer will get him soon. */
 971
 972                 nf_ct_put(ct);
 973         }
 974 }
 975 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 976
 977 struct __nf_ct_flush_report {
 978         u32 pid;
 979         int report;
 980 };
 981
 982 static int kill_all(struct nf_conn *i, void *data)
 983 {
 984         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
 985
 986         /* get_next_corpse sets the dying bit for us */
 987         nf_conntrack_event_report(IPCT_DESTROY,
 988                                   i,
 989                                   fr->pid,
 990                                   fr->report);
 991         return 1;
 992 }
 993
 994 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
 995 {
 996         if (vmalloced)
 997                 vfree(hash);
 998         else
 999                 free_pages((unsigned long)hash,
1000                            get_order(sizeof(struct hlist_head) * size));
1001 }
1002 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1003
1004 void nf_conntrack_flush(struct net *net, u32 pid, int report)
1005 {
1006         struct __nf_ct_flush_report fr = {
1007                 .pid    = pid,
1008                 .report = report,
1009         };
1010         nf_ct_iterate_cleanup(net, kill_all, &fr);
1011 }
1012 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1013
1014 static void nf_conntrack_cleanup_init_net(void)
1015 {
1016         nf_conntrack_helper_fini();
1017         nf_conntrack_proto_fini();
1018         kmem_cache_destroy(nf_conntrack_cachep);
1019 }
1020
1021 static void nf_conntrack_cleanup_net(struct net *net)
1022 {
1023         nf_ct_event_cache_flush(net);
1024         nf_conntrack_ecache_fini(net);
1025  i_see_dead_people:
1026         nf_conntrack_flush(net, 0, 0);
1027         if (atomic_read(&net->ct.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         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1036                              nf_conntrack_htable_size);
1037         nf_conntrack_acct_fini(net);
1038         nf_conntrack_expect_fini(net);
1039         free_percpu(net->ct.stat);
1040 }
1041
1042 /* Mishearing the voices in his head, our hero wonders how he's
1043    supposed to kill the mall. */
1044 void nf_conntrack_cleanup(struct net *net)
1045 {
1046         if (net_eq(net, &init_net))
1047                 rcu_assign_pointer(ip_ct_attach, NULL);
1048
1049         /* This makes sure all current packets have passed through
1050            netfilter framework.  Roll on, two-stage module
1051            delete... */
1052         synchronize_net();
1053
1054         nf_conntrack_cleanup_net(net);
1055
1056         if (net_eq(net, &init_net)) {
1057                 rcu_assign_pointer(nf_ct_destroy, NULL);
1058                 nf_conntrack_cleanup_init_net();
1059         }
1060 }
1061
1062 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1063 {
1064         struct hlist_head *hash;
1065         unsigned int size, i;
1066
1067         *vmalloced = 0;
1068
1069         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1070         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1071                                        get_order(sizeof(struct hlist_head)
1072                                                  * size));
1073         if (!hash) {
1074                 *vmalloced = 1;
1075                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1076                 hash = vmalloc(sizeof(struct hlist_head) * size);
1077         }
1078
1079         if (hash)
1080                 for (i = 0; i < size; i++)
1081                         INIT_HLIST_HEAD(&hash[i]);
1082
1083         return hash;
1084 }
1085 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1086
1087 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1088 {
1089         int i, bucket, vmalloced, old_vmalloced;
1090         unsigned int hashsize, old_size;
1091         int rnd;
1092         struct hlist_head *hash, *old_hash;
1093         struct nf_conntrack_tuple_hash *h;
1094
1095         /* On boot, we can set this without any fancy locking. */
1096         if (!nf_conntrack_htable_size)
1097                 return param_set_uint(val, kp);
1098
1099         hashsize = simple_strtoul(val, NULL, 0);
1100         if (!hashsize)
1101                 return -EINVAL;
1102
1103         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1104         if (!hash)
1105                 return -ENOMEM;
1106
1107         /* We have to rehahs for the new table anyway, so we also can
1108          * use a newrandom seed */
1109         get_random_bytes(&rnd, sizeof(rnd));
1110
1111         /* Lookups in the old hash might happen in parallel, which means we
1112          * might get false negatives during connection lookup. New connections
1113          * created because of a false negative won't make it into the hash
1114          * though since that required taking the lock.
1115          */
1116         spin_lock_bh(&nf_conntrack_lock);
1117         for (i = 0; i < nf_conntrack_htable_size; i++) {
1118                 while (!hlist_empty(&init_net.ct.hash[i])) {
1119                         h = hlist_entry(init_net.ct.hash[i].first,
1120                                         struct nf_conntrack_tuple_hash, hnode);
1121                         hlist_del_rcu(&h->hnode);
1122                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1123                         hlist_add_head(&h->hnode, &hash[bucket]);
1124                 }
1125         }
1126         old_size = nf_conntrack_htable_size;
1127         old_vmalloced = init_net.ct.hash_vmalloc;
1128         old_hash = init_net.ct.hash;
1129
1130         nf_conntrack_htable_size = hashsize;
1131         init_net.ct.hash_vmalloc = vmalloced;
1132         init_net.ct.hash = hash;
1133         nf_conntrack_hash_rnd = rnd;
1134         spin_unlock_bh(&nf_conntrack_lock);
1135
1136         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1137         return 0;
1138 }
1139 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1140
1141 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1142                   &nf_conntrack_htable_size, 0600);
1143
1144 static int nf_conntrack_init_init_net(void)
1145 {
1146         int max_factor = 8;
1147         int ret;
1148
1149         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1150          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1151         if (!nf_conntrack_htable_size) {
1152                 nf_conntrack_htable_size
1153                         = (((num_physpages << PAGE_SHIFT) / 16384)
1154                            / sizeof(struct hlist_head));
1155                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1156                         nf_conntrack_htable_size = 16384;
1157                 if (nf_conntrack_htable_size < 32)
1158                         nf_conntrack_htable_size = 32;
1159
1160                 /* Use a max. factor of four by default to get the same max as
1161                  * with the old struct list_heads. When a table size is given
1162                  * we use the old value of 8 to avoid reducing the max.
1163                  * entries. */
1164                 max_factor = 4;
1165         }
1166         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1167
1168         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1169                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1170                nf_conntrack_max);
1171
1172         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1173                                                 sizeof(struct nf_conn),
1174                                                 0, 0, NULL);
1175         if (!nf_conntrack_cachep) {
1176                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1177                 ret = -ENOMEM;
1178                 goto err_cache;
1179         }
1180
1181         ret = nf_conntrack_proto_init();
1182         if (ret < 0)
1183                 goto err_proto;
1184
1185         ret = nf_conntrack_helper_init();
1186         if (ret < 0)
1187                 goto err_helper;
1188
1189         return 0;
1190
1191 err_helper:
1192         nf_conntrack_proto_fini();
1193 err_proto:
1194         kmem_cache_destroy(nf_conntrack_cachep);
1195 err_cache:
1196         return ret;
1197 }
1198
1199 static int nf_conntrack_init_net(struct net *net)
1200 {
1201         int ret;
1202
1203         atomic_set(&net->ct.count, 0);
1204         INIT_HLIST_HEAD(&net->ct.unconfirmed);
1205         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1206         if (!net->ct.stat) {
1207                 ret = -ENOMEM;
1208                 goto err_stat;
1209         }
1210         ret = nf_conntrack_ecache_init(net);
1211         if (ret < 0)
1212                 goto err_ecache;
1213         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1214                                                   &net->ct.hash_vmalloc);
1215         if (!net->ct.hash) {
1216                 ret = -ENOMEM;
1217                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1218                 goto err_hash;
1219         }
1220         ret = nf_conntrack_expect_init(net);
1221         if (ret < 0)
1222                 goto err_expect;
1223         ret = nf_conntrack_acct_init(net);
1224         if (ret < 0)
1225                 goto err_acct;
1226
1227         /* Set up fake conntrack:
1228             - to never be deleted, not in any hashes */
1229 #ifdef CONFIG_NET_NS
1230         nf_conntrack_untracked.ct_net = &init_net;
1231 #endif
1232         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1233         /*  - and look it like as a confirmed connection */
1234         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1235
1236         return 0;
1237
1238 err_acct:
1239         nf_conntrack_expect_fini(net);
1240 err_expect:
1241         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1242                              nf_conntrack_htable_size);
1243 err_hash:
1244         nf_conntrack_ecache_fini(net);
1245 err_ecache:
1246         free_percpu(net->ct.stat);
1247 err_stat:
1248         return ret;
1249 }
1250
1251 int nf_conntrack_init(struct net *net)
1252 {
1253         int ret;
1254
1255         if (net_eq(net, &init_net)) {
1256                 ret = nf_conntrack_init_init_net();
1257                 if (ret < 0)
1258                         goto out_init_net;
1259         }
1260         ret = nf_conntrack_init_net(net);
1261         if (ret < 0)
1262                 goto out_net;
1263
1264         if (net_eq(net, &init_net)) {
1265                 /* For use by REJECT target */
1266                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1267                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1268         }
1269         return 0;
1270
1271 out_net:
1272         if (net_eq(net, &init_net))
1273                 nf_conntrack_cleanup_init_net();
1274 out_init_net:
1275         return ret;
1276 }