SAFE public projects git trees. - safe/jmp/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
  43 #define NF_CONNTRACK_VERSION    "0.5.0"
  44
  45 unsigned int
  46 (*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 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         nf_conntrack_event(IPCT_DESTROY, ct);
 185         set_bit(IPS_DYING_BIT, &ct->status);
 186
 187         /* To make sure we don't get any weird locking issues here:
 188          * destroy_conntrack() MUST NOT be called with a write lock
 189          * to nf_conntrack_lock!!! -HW */
 190         rcu_read_lock();
 191         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 192         if (l4proto && l4proto->destroy)
 193                 l4proto->destroy(ct);
 194
 195         rcu_read_unlock();
 196
 197         spin_lock_bh(&nf_conntrack_lock);
 198         /* Expectations will have been removed in clean_from_lists,
 199          * except TFTP can create an expectation on the first packet,
 200          * before connection is in the list, so we need to clean here,
 201          * too. */
 202         nf_ct_remove_expectations(ct);
 203
 204         /* We overload first tuple to link into unconfirmed list. */
 205         if (!nf_ct_is_confirmed(ct)) {
 206                 BUG_ON(hlist_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode));
 207                 hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 208         }
 209
 210         NF_CT_STAT_INC(net, delete);
 211         spin_unlock_bh(&nf_conntrack_lock);
 212
 213         if (ct->master)
 214                 nf_ct_put(ct->master);
 215
 216         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 217         nf_conntrack_free(ct);
 218 }
 219
 220 static void death_by_timeout(unsigned long ul_conntrack)
 221 {
 222         struct nf_conn *ct = (void *)ul_conntrack;
 223         struct net *net = nf_ct_net(ct);
 224         struct nf_conn_help *help = nfct_help(ct);
 225         struct nf_conntrack_helper *helper;
 226
 227         if (help) {
 228                 rcu_read_lock();
 229                 helper = rcu_dereference(help->helper);
 230                 if (helper && helper->destroy)
 231                         helper->destroy(ct);
 232                 rcu_read_unlock();
 233         }
 234
 235         spin_lock_bh(&nf_conntrack_lock);
 236         /* Inside lock so preempt is disabled on module removal path.
 237          * Otherwise we can get spurious warnings. */
 238         NF_CT_STAT_INC(net, delete_list);
 239         clean_from_lists(ct);
 240         spin_unlock_bh(&nf_conntrack_lock);
 241         nf_ct_put(ct);
 242 }
 243
 244 struct nf_conntrack_tuple_hash *
 245 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 246 {
 247         struct nf_conntrack_tuple_hash *h;
 248         struct hlist_node *n;
 249         unsigned int hash = hash_conntrack(tuple);
 250
 251         /* Disable BHs the entire time since we normally need to disable them
 252          * at least once for the stats anyway.
 253          */
 254         local_bh_disable();
 255         hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
 256                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 257                         NF_CT_STAT_INC(net, found);
 258                         local_bh_enable();
 259                         return h;
 260                 }
 261                 NF_CT_STAT_INC(net, searched);
 262         }
 263         local_bh_enable();
 264
 265         return NULL;
 266 }
 267 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 268
 269 /* Find a connection corresponding to a tuple. */
 270 struct nf_conntrack_tuple_hash *
 271 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
 272 {
 273         struct nf_conntrack_tuple_hash *h;
 274         struct nf_conn *ct;
 275
 276         rcu_read_lock();
 277         h = __nf_conntrack_find(net, tuple);
 278         if (h) {
 279                 ct = nf_ct_tuplehash_to_ctrack(h);
 280                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 281                         h = NULL;
 282         }
 283         rcu_read_unlock();
 284
 285         return h;
 286 }
 287 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 288
 289 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 290                                        unsigned int hash,
 291                                        unsigned int repl_hash)
 292 {
 293         struct net *net = nf_ct_net(ct);
 294
 295         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 296                            &net->ct.hash[hash]);
 297         hlist_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnode,
 298                            &net->ct.hash[repl_hash]);
 299 }
 300
 301 void nf_conntrack_hash_insert(struct nf_conn *ct)
 302 {
 303         unsigned int hash, repl_hash;
 304
 305         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 306         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 307
 308         spin_lock_bh(&nf_conntrack_lock);
 309         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 310         spin_unlock_bh(&nf_conntrack_lock);
 311 }
 312 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 313
 314 /* Confirm a connection given skb; places it in hash table */
 315 int
 316 __nf_conntrack_confirm(struct sk_buff *skb)
 317 {
 318         unsigned int hash, repl_hash;
 319         struct nf_conntrack_tuple_hash *h;
 320         struct nf_conn *ct;
 321         struct nf_conn_help *help;
 322         struct hlist_node *n;
 323         enum ip_conntrack_info ctinfo;
 324         struct net *net;
 325
 326         ct = nf_ct_get(skb, &ctinfo);
 327         net = nf_ct_net(ct);
 328
 329         /* ipt_REJECT uses nf_conntrack_attach to attach related
 330            ICMP/TCP RST packets in other direction.  Actual packet
 331            which created connection will be IP_CT_NEW or for an
 332            expected connection, IP_CT_RELATED. */
 333         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 334                 return NF_ACCEPT;
 335
 336         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 337         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 338
 339         /* We're not in hash table, and we refuse to set up related
 340            connections for unconfirmed conns.  But packet copies and
 341            REJECT will give spurious warnings here. */
 342         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 343
 344         /* No external references means noone else could have
 345            confirmed us. */
 346         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 347         pr_debug("Confirming conntrack %p\n", ct);
 348
 349         spin_lock_bh(&nf_conntrack_lock);
 350
 351         /* See if there's one in the list already, including reverse:
 352            NAT could have grabbed it without realizing, since we're
 353            not in the hash.  If there is, we lost race. */
 354         hlist_for_each_entry(h, n, &net->ct.hash[hash], hnode)
 355                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 356                                       &h->tuple))
 357                         goto out;
 358         hlist_for_each_entry(h, n, &net->ct.hash[repl_hash], hnode)
 359                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 360                                       &h->tuple))
 361                         goto out;
 362
 363         /* Remove from unconfirmed list */
 364         hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
 365
 366         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 367         /* Timer relative to confirmation time, not original
 368            setting time, otherwise we'd get timer wrap in
 369            weird delay cases. */
 370         ct->timeout.expires += jiffies;
 371         add_timer(&ct->timeout);
 372         atomic_inc(&ct->ct_general.use);
 373         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 374         NF_CT_STAT_INC(net, insert);
 375         spin_unlock_bh(&nf_conntrack_lock);
 376         help = nfct_help(ct);
 377         if (help && help->helper)
 378                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 379 #ifdef CONFIG_NF_NAT_NEEDED
 380         if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
 381             test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
 382                 nf_conntrack_event_cache(IPCT_NATINFO, ct);
 383 #endif
 384         nf_conntrack_event_cache(master_ct(ct) ?
 385                                  IPCT_RELATED : IPCT_NEW, ct);
 386         return NF_ACCEPT;
 387
 388 out:
 389         NF_CT_STAT_INC(net, insert_failed);
 390         spin_unlock_bh(&nf_conntrack_lock);
 391         return NF_DROP;
 392 }
 393 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 394
 395 /* Returns true if a connection correspondings to the tuple (required
 396    for NAT). */
 397 int
 398 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 399                          const struct nf_conn *ignored_conntrack)
 400 {
 401         struct net *net = nf_ct_net(ignored_conntrack);
 402         struct nf_conntrack_tuple_hash *h;
 403         struct hlist_node *n;
 404         unsigned int hash = hash_conntrack(tuple);
 405
 406         /* Disable BHs the entire time since we need to disable them at
 407          * least once for the stats anyway.
 408          */
 409         rcu_read_lock_bh();
 410         hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
 411                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 412                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 413                         NF_CT_STAT_INC(net, found);
 414                         rcu_read_unlock_bh();
 415                         return 1;
 416                 }
 417                 NF_CT_STAT_INC(net, searched);
 418         }
 419         rcu_read_unlock_bh();
 420
 421         return 0;
 422 }
 423 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 424
 425 #define NF_CT_EVICTION_RANGE    8
 426
 427 /* There's a small race here where we may free a just-assured
 428    connection.  Too bad: we're in trouble anyway. */
 429 static noinline int early_drop(struct net *net, unsigned int hash)
 430 {
 431         /* Use oldest entry, which is roughly LRU */
 432         struct nf_conntrack_tuple_hash *h;
 433         struct nf_conn *ct = NULL, *tmp;
 434         struct hlist_node *n;
 435         unsigned int i, cnt = 0;
 436         int dropped = 0;
 437
 438         rcu_read_lock();
 439         for (i = 0; i < nf_conntrack_htable_size; i++) {
 440                 hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 441                                          hnode) {
 442                         tmp = nf_ct_tuplehash_to_ctrack(h);
 443                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 444                                 ct = tmp;
 445                         cnt++;
 446                 }
 447
 448                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 449                         ct = NULL;
 450                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 451                         break;
 452                 hash = (hash + 1) % nf_conntrack_htable_size;
 453         }
 454         rcu_read_unlock();
 455
 456         if (!ct)
 457                 return dropped;
 458
 459         if (del_timer(&ct->timeout)) {
 460                 death_by_timeout((unsigned long)ct);
 461                 dropped = 1;
 462                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 463         }
 464         nf_ct_put(ct);
 465         return dropped;
 466 }
 467
 468 struct nf_conn *nf_conntrack_alloc(struct net *net,
 469                                    const struct nf_conntrack_tuple *orig,
 470                                    const struct nf_conntrack_tuple *repl,
 471                                    gfp_t gfp)
 472 {
 473         struct nf_conn *ct = NULL;
 474
 475         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 476                 get_random_bytes(&nf_conntrack_hash_rnd, 4);
 477                 nf_conntrack_hash_rnd_initted = 1;
 478         }
 479
 480         /* We don't want any race condition at early drop stage */
 481         atomic_inc(&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 (ct == NULL || 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                 struct nf_conntrack_helper *helper;
 592
 593                 helper = __nf_ct_helper_find(&repl_tuple);
 594                 if (helper) {
 595                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 596                         if (help)
 597                                 rcu_assign_pointer(help->helper, helper);
 598                 }
 599                 NF_CT_STAT_INC(net, new);
 600         }
 601
 602         /* Overload tuple linked list to put us in unconfirmed list. */
 603         hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
 604                        &net->ct.unconfirmed);
 605
 606         spin_unlock_bh(&nf_conntrack_lock);
 607
 608         if (exp) {
 609                 if (exp->expectfn)
 610                         exp->expectfn(ct, exp);
 611                 nf_ct_expect_put(exp);
 612         }
 613
 614         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 615 }
 616
 617 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 618 static inline struct nf_conn *
 619 resolve_normal_ct(struct net *net,
 620                   struct sk_buff *skb,
 621                   unsigned int dataoff,
 622                   u_int16_t l3num,
 623                   u_int8_t protonum,
 624                   struct nf_conntrack_l3proto *l3proto,
 625                   struct nf_conntrack_l4proto *l4proto,
 626                   int *set_reply,
 627                   enum ip_conntrack_info *ctinfo)
 628 {
 629         struct nf_conntrack_tuple tuple;
 630         struct nf_conntrack_tuple_hash *h;
 631         struct nf_conn *ct;
 632
 633         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 634                              dataoff, l3num, protonum, &tuple, l3proto,
 635                              l4proto)) {
 636                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 637                 return NULL;
 638         }
 639
 640         /* look for tuple match */
 641         h = nf_conntrack_find_get(net, &tuple);
 642         if (!h) {
 643                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 644                 if (!h)
 645                         return NULL;
 646                 if (IS_ERR(h))
 647                         return (void *)h;
 648         }
 649         ct = nf_ct_tuplehash_to_ctrack(h);
 650
 651         /* It exists; we have (non-exclusive) reference. */
 652         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 653                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 654                 /* Please set reply bit if this packet OK */
 655                 *set_reply = 1;
 656         } else {
 657                 /* Once we've had two way comms, always ESTABLISHED. */
 658                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 659                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 660                         *ctinfo = IP_CT_ESTABLISHED;
 661                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 662                         pr_debug("nf_conntrack_in: related packet for %p\n",
 663                                  ct);
 664                         *ctinfo = IP_CT_RELATED;
 665                 } else {
 666                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 667                         *ctinfo = IP_CT_NEW;
 668                 }
 669                 *set_reply = 0;
 670         }
 671         skb->nfct = &ct->ct_general;
 672         skb->nfctinfo = *ctinfo;
 673         return ct;
 674 }
 675
 676 unsigned int
 677 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 678                 struct sk_buff *skb)
 679 {
 680         struct nf_conn *ct;
 681         enum ip_conntrack_info ctinfo;
 682         struct nf_conntrack_l3proto *l3proto;
 683         struct nf_conntrack_l4proto *l4proto;
 684         unsigned int dataoff;
 685         u_int8_t protonum;
 686         int set_reply = 0;
 687         int ret;
 688
 689         /* Previously seen (loopback or untracked)?  Ignore. */
 690         if (skb->nfct) {
 691                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 692                 return NF_ACCEPT;
 693         }
 694
 695         /* rcu_read_lock()ed by nf_hook_slow */
 696         l3proto = __nf_ct_l3proto_find(pf);
 697         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 698                                    &dataoff, &protonum);
 699         if (ret <= 0) {
 700                 pr_debug("not prepared to track yet or error occured\n");
 701                 NF_CT_STAT_INC_ATOMIC(net, error);
 702                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 703                 return -ret;
 704         }
 705
 706         l4proto = __nf_ct_l4proto_find(pf, protonum);
 707
 708         /* It may be an special packet, error, unclean...
 709          * inverse of the return code tells to the netfilter
 710          * core what to do with the packet. */
 711         if (l4proto->error != NULL) {
 712                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 713                 if (ret <= 0) {
 714                         NF_CT_STAT_INC_ATOMIC(net, error);
 715                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 716                         return -ret;
 717                 }
 718         }
 719
 720         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 721                                l3proto, l4proto, &set_reply, &ctinfo);
 722         if (!ct) {
 723                 /* Not valid part of a connection */
 724                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 725                 return NF_ACCEPT;
 726         }
 727
 728         if (IS_ERR(ct)) {
 729                 /* Too stressed to deal. */
 730                 NF_CT_STAT_INC_ATOMIC(net, drop);
 731                 return NF_DROP;
 732         }
 733
 734         NF_CT_ASSERT(skb->nfct);
 735
 736         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 737         if (ret < 0) {
 738                 /* Invalid: inverse of the return code tells
 739                  * the netfilter core what to do */
 740                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 741                 nf_conntrack_put(skb->nfct);
 742                 skb->nfct = NULL;
 743                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 744                 return -ret;
 745         }
 746
 747         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 748                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 749
 750         return ret;
 751 }
 752 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 753
 754 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 755                           const struct nf_conntrack_tuple *orig)
 756 {
 757         bool ret;
 758
 759         rcu_read_lock();
 760         ret = nf_ct_invert_tuple(inverse, orig,
 761                                  __nf_ct_l3proto_find(orig->src.l3num),
 762                                  __nf_ct_l4proto_find(orig->src.l3num,
 763                                                       orig->dst.protonum));
 764         rcu_read_unlock();
 765         return ret;
 766 }
 767 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 768
 769 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 770    implicitly racy: see __nf_conntrack_confirm */
 771 void nf_conntrack_alter_reply(struct nf_conn *ct,
 772                               const struct nf_conntrack_tuple *newreply)
 773 {
 774         struct nf_conn_help *help = nfct_help(ct);
 775         struct nf_conntrack_helper *helper;
 776
 777         /* Should be unconfirmed, so not in hash table yet */
 778         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 779
 780         pr_debug("Altering reply tuple of %p to ", ct);
 781         nf_ct_dump_tuple(newreply);
 782
 783         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 784         if (ct->master || (help && !hlist_empty(&help->expectations)))
 785                 return;
 786
 787         rcu_read_lock();
 788         helper = __nf_ct_helper_find(newreply);
 789         if (helper == NULL) {
 790                 if (help)
 791                         rcu_assign_pointer(help->helper, NULL);
 792                 goto out;
 793         }
 794
 795         if (help == NULL) {
 796                 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 797                 if (help == NULL)
 798                         goto out;
 799         } else {
 800                 memset(&help->help, 0, sizeof(help->help));
 801         }
 802
 803         rcu_assign_pointer(help->helper, helper);
 804 out:
 805         rcu_read_unlock();
 806 }
 807 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 808
 809 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 810 void __nf_ct_refresh_acct(struct nf_conn *ct,
 811                           enum ip_conntrack_info ctinfo,
 812                           const struct sk_buff *skb,
 813                           unsigned long extra_jiffies,
 814                           int do_acct)
 815 {
 816         int event = 0;
 817
 818         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 819         NF_CT_ASSERT(skb);
 820
 821         spin_lock_bh(&nf_conntrack_lock);
 822
 823         /* Only update if this is not a fixed timeout */
 824         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 825                 goto acct;
 826
 827         /* If not in hash table, timer will not be active yet */
 828         if (!nf_ct_is_confirmed(ct)) {
 829                 ct->timeout.expires = extra_jiffies;
 830                 event = IPCT_REFRESH;
 831         } else {
 832                 unsigned long newtime = jiffies + extra_jiffies;
 833
 834                 /* Only update the timeout if the new timeout is at least
 835                    HZ jiffies from the old timeout. Need del_timer for race
 836                    avoidance (may already be dying). */
 837                 if (newtime - ct->timeout.expires >= HZ
 838                     && del_timer(&ct->timeout)) {
 839                         ct->timeout.expires = newtime;
 840                         add_timer(&ct->timeout);
 841                         event = IPCT_REFRESH;
 842                 }
 843         }
 844
 845 acct:
 846         if (do_acct) {
 847                 struct nf_conn_counter *acct;
 848
 849                 acct = nf_conn_acct_find(ct);
 850                 if (acct) {
 851                         acct[CTINFO2DIR(ctinfo)].packets++;
 852                         acct[CTINFO2DIR(ctinfo)].bytes +=
 853                                 skb->len - skb_network_offset(skb);
 854                 }
 855         }
 856
 857         spin_unlock_bh(&nf_conntrack_lock);
 858
 859         /* must be unlocked when calling event cache */
 860         if (event)
 861                 nf_conntrack_event_cache(event, ct);
 862 }
 863 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 864
 865 bool __nf_ct_kill_acct(struct nf_conn *ct,
 866                        enum ip_conntrack_info ctinfo,
 867                        const struct sk_buff *skb,
 868                        int do_acct)
 869 {
 870         if (do_acct) {
 871                 struct nf_conn_counter *acct;
 872
 873                 spin_lock_bh(&nf_conntrack_lock);
 874                 acct = nf_conn_acct_find(ct);
 875                 if (acct) {
 876                         acct[CTINFO2DIR(ctinfo)].packets++;
 877                         acct[CTINFO2DIR(ctinfo)].bytes +=
 878                                 skb->len - skb_network_offset(skb);
 879                 }
 880                 spin_unlock_bh(&nf_conntrack_lock);
 881         }
 882
 883         if (del_timer(&ct->timeout)) {
 884                 ct->timeout.function((unsigned long)ct);
 885                 return true;
 886         }
 887         return false;
 888 }
 889 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 890
 891 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 892
 893 #include <linux/netfilter/nfnetlink.h>
 894 #include <linux/netfilter/nfnetlink_conntrack.h>
 895 #include <linux/mutex.h>
 896
 897 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 898  * in ip_conntrack_core, since we don't want the protocols to autoload
 899  * or depend on ctnetlink */
 900 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 901                                const struct nf_conntrack_tuple *tuple)
 902 {
 903         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 904         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 905         return 0;
 906
 907 nla_put_failure:
 908         return -1;
 909 }
 910 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 911
 912 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 913         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 914         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 915 };
 916 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 917
 918 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 919                                struct nf_conntrack_tuple *t)
 920 {
 921         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 922                 return -EINVAL;
 923
 924         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 925         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 926
 927         return 0;
 928 }
 929 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 930 #endif
 931
 932 /* Used by ipt_REJECT and ip6t_REJECT. */
 933 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 934 {
 935         struct nf_conn *ct;
 936         enum ip_conntrack_info ctinfo;
 937
 938         /* This ICMP is in reverse direction to the packet which caused it */
 939         ct = nf_ct_get(skb, &ctinfo);
 940         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 941                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 942         else
 943                 ctinfo = IP_CT_RELATED;
 944
 945         /* Attach to new skbuff, and increment count */
 946         nskb->nfct = &ct->ct_general;
 947         nskb->nfctinfo = ctinfo;
 948         nf_conntrack_get(nskb->nfct);
 949 }
 950
 951 /* Bring out ya dead! */
 952 static struct nf_conn *
 953 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 954                 void *data, unsigned int *bucket)
 955 {
 956         struct nf_conntrack_tuple_hash *h;
 957         struct nf_conn *ct;
 958         struct hlist_node *n;
 959
 960         spin_lock_bh(&nf_conntrack_lock);
 961         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 962                 hlist_for_each_entry(h, n, &net->ct.hash[*bucket], hnode) {
 963                         ct = nf_ct_tuplehash_to_ctrack(h);
 964                         if (iter(ct, data))
 965                                 goto found;
 966                 }
 967         }
 968         hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode) {
 969                 ct = nf_ct_tuplehash_to_ctrack(h);
 970                 if (iter(ct, data))
 971                         set_bit(IPS_DYING_BIT, &ct->status);
 972         }
 973         spin_unlock_bh(&nf_conntrack_lock);
 974         return NULL;
 975 found:
 976         atomic_inc(&ct->ct_general.use);
 977         spin_unlock_bh(&nf_conntrack_lock);
 978         return ct;
 979 }
 980
 981 void nf_ct_iterate_cleanup(struct net *net,
 982                            int (*iter)(struct nf_conn *i, void *data),
 983                            void *data)
 984 {
 985         struct nf_conn *ct;
 986         unsigned int bucket = 0;
 987
 988         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
 989                 /* Time to push up daises... */
 990                 if (del_timer(&ct->timeout))
 991                         death_by_timeout((unsigned long)ct);
 992                 /* ... else the timer will get him soon. */
 993
 994                 nf_ct_put(ct);
 995         }
 996 }
 997 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 998
 999 static int kill_all(struct nf_conn *i, void *data)
1000 {
1001         return 1;
1002 }
1003
1004 void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, unsigned int size)
1005 {
1006         if (vmalloced)
1007                 vfree(hash);
1008         else
1009                 free_pages((unsigned long)hash,
1010                            get_order(sizeof(struct hlist_head) * size));
1011 }
1012 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1013
1014 void nf_conntrack_flush(struct net *net)
1015 {
1016         nf_ct_iterate_cleanup(net, kill_all, NULL);
1017 }
1018 EXPORT_SYMBOL_GPL(nf_conntrack_flush);
1019
1020 static void nf_conntrack_cleanup_init_net(void)
1021 {
1022         nf_conntrack_helper_fini();
1023         nf_conntrack_proto_fini();
1024         kmem_cache_destroy(nf_conntrack_cachep);
1025 }
1026
1027 static void nf_conntrack_cleanup_net(struct net *net)
1028 {
1029         nf_ct_event_cache_flush(net);
1030         nf_conntrack_ecache_fini(net);
1031  i_see_dead_people:
1032         nf_conntrack_flush(net);
1033         if (atomic_read(&net->ct.count) != 0) {
1034                 schedule();
1035                 goto i_see_dead_people;
1036         }
1037         /* wait until all references to nf_conntrack_untracked are dropped */
1038         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1039                 schedule();
1040
1041         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1042                              nf_conntrack_htable_size);
1043         nf_conntrack_acct_fini(net);
1044         nf_conntrack_expect_fini(net);
1045         free_percpu(net->ct.stat);
1046 }
1047
1048 /* Mishearing the voices in his head, our hero wonders how he's
1049    supposed to kill the mall. */
1050 void nf_conntrack_cleanup(struct net *net)
1051 {
1052         if (net_eq(net, &init_net))
1053                 rcu_assign_pointer(ip_ct_attach, NULL);
1054
1055         /* This makes sure all current packets have passed through
1056            netfilter framework.  Roll on, two-stage module
1057            delete... */
1058         synchronize_net();
1059
1060         nf_conntrack_cleanup_net(net);
1061
1062         if (net_eq(net, &init_net)) {
1063                 rcu_assign_pointer(nf_ct_destroy, NULL);
1064                 nf_conntrack_cleanup_init_net();
1065         }
1066 }
1067
1068 struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced)
1069 {
1070         struct hlist_head *hash;
1071         unsigned int size, i;
1072
1073         *vmalloced = 0;
1074
1075         size = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_head));
1076         hash = (void*)__get_free_pages(GFP_KERNEL|__GFP_NOWARN,
1077                                        get_order(sizeof(struct hlist_head)
1078                                                  * size));
1079         if (!hash) {
1080                 *vmalloced = 1;
1081                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1082                 hash = vmalloc(sizeof(struct hlist_head) * size);
1083         }
1084
1085         if (hash)
1086                 for (i = 0; i < size; i++)
1087                         INIT_HLIST_HEAD(&hash[i]);
1088
1089         return hash;
1090 }
1091 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1092
1093 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1094 {
1095         int i, bucket, vmalloced, old_vmalloced;
1096         unsigned int hashsize, old_size;
1097         int rnd;
1098         struct hlist_head *hash, *old_hash;
1099         struct nf_conntrack_tuple_hash *h;
1100
1101         /* On boot, we can set this without any fancy locking. */
1102         if (!nf_conntrack_htable_size)
1103                 return param_set_uint(val, kp);
1104
1105         hashsize = simple_strtoul(val, NULL, 0);
1106         if (!hashsize)
1107                 return -EINVAL;
1108
1109         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
1110         if (!hash)
1111                 return -ENOMEM;
1112
1113         /* We have to rehahs for the new table anyway, so we also can
1114          * use a newrandom seed */
1115         get_random_bytes(&rnd, 4);
1116
1117         /* Lookups in the old hash might happen in parallel, which means we
1118          * might get false negatives during connection lookup. New connections
1119          * created because of a false negative won't make it into the hash
1120          * though since that required taking the lock.
1121          */
1122         spin_lock_bh(&nf_conntrack_lock);
1123         for (i = 0; i < nf_conntrack_htable_size; i++) {
1124                 while (!hlist_empty(&init_net.ct.hash[i])) {
1125                         h = hlist_entry(init_net.ct.hash[i].first,
1126                                         struct nf_conntrack_tuple_hash, hnode);
1127                         hlist_del_rcu(&h->hnode);
1128                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1129                         hlist_add_head(&h->hnode, &hash[bucket]);
1130                 }
1131         }
1132         old_size = nf_conntrack_htable_size;
1133         old_vmalloced = init_net.ct.hash_vmalloc;
1134         old_hash = init_net.ct.hash;
1135
1136         nf_conntrack_htable_size = hashsize;
1137         init_net.ct.hash_vmalloc = vmalloced;
1138         init_net.ct.hash = hash;
1139         nf_conntrack_hash_rnd = rnd;
1140         spin_unlock_bh(&nf_conntrack_lock);
1141
1142         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1143         return 0;
1144 }
1145 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1146
1147 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1148                   &nf_conntrack_htable_size, 0600);
1149
1150 static int nf_conntrack_init_init_net(void)
1151 {
1152         int max_factor = 8;
1153         int ret;
1154
1155         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1156          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1157         if (!nf_conntrack_htable_size) {
1158                 nf_conntrack_htable_size
1159                         = (((num_physpages << PAGE_SHIFT) / 16384)
1160                            / sizeof(struct hlist_head));
1161                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1162                         nf_conntrack_htable_size = 16384;
1163                 if (nf_conntrack_htable_size < 32)
1164                         nf_conntrack_htable_size = 32;
1165
1166                 /* Use a max. factor of four by default to get the same max as
1167                  * with the old struct list_heads. When a table size is given
1168                  * we use the old value of 8 to avoid reducing the max.
1169                  * entries. */
1170                 max_factor = 4;
1171         }
1172         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1173
1174         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1175                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1176                nf_conntrack_max);
1177
1178         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1179                                                 sizeof(struct nf_conn),
1180                                                 0, 0, NULL);
1181         if (!nf_conntrack_cachep) {
1182                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1183                 ret = -ENOMEM;
1184                 goto err_cache;
1185         }
1186
1187         ret = nf_conntrack_proto_init();
1188         if (ret < 0)
1189                 goto err_proto;
1190
1191         ret = nf_conntrack_helper_init();
1192         if (ret < 0)
1193                 goto err_helper;
1194
1195         return 0;
1196
1197 err_helper:
1198         nf_conntrack_proto_fini();
1199 err_proto:
1200         kmem_cache_destroy(nf_conntrack_cachep);
1201 err_cache:
1202         return ret;
1203 }
1204
1205 static int nf_conntrack_init_net(struct net *net)
1206 {
1207         int ret;
1208
1209         atomic_set(&net->ct.count, 0);
1210         INIT_HLIST_HEAD(&net->ct.unconfirmed);
1211         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1212         if (!net->ct.stat) {
1213                 ret = -ENOMEM;
1214                 goto err_stat;
1215         }
1216         ret = nf_conntrack_ecache_init(net);
1217         if (ret < 0)
1218                 goto err_ecache;
1219         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1220                                                   &net->ct.hash_vmalloc);
1221         if (!net->ct.hash) {
1222                 ret = -ENOMEM;
1223                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1224                 goto err_hash;
1225         }
1226         ret = nf_conntrack_expect_init(net);
1227         if (ret < 0)
1228                 goto err_expect;
1229         ret = nf_conntrack_acct_init(net);
1230         if (ret < 0)
1231                 goto err_acct;
1232
1233         /* Set up fake conntrack:
1234             - to never be deleted, not in any hashes */
1235 #ifdef CONFIG_NET_NS
1236         nf_conntrack_untracked.ct_net = &init_net;
1237 #endif
1238         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1239         /*  - and look it like as a confirmed connection */
1240         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1241
1242         return 0;
1243
1244 err_acct:
1245         nf_conntrack_expect_fini(net);
1246 err_expect:
1247         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1248                              nf_conntrack_htable_size);
1249 err_hash:
1250         nf_conntrack_ecache_fini(net);
1251 err_ecache:
1252         free_percpu(net->ct.stat);
1253 err_stat:
1254         return ret;
1255 }
1256
1257 int nf_conntrack_init(struct net *net)
1258 {
1259         int ret;
1260
1261         if (net_eq(net, &init_net)) {
1262                 ret = nf_conntrack_init_init_net();
1263                 if (ret < 0)
1264                         goto out_init_net;
1265         }
1266         ret = nf_conntrack_init_net(net);
1267         if (ret < 0)
1268                 goto out_net;
1269
1270         if (net_eq(net, &init_net)) {
1271                 /* For use by REJECT target */
1272                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1273                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1274         }
1275         return 0;
1276
1277 out_net:
1278         if (net_eq(net, &init_net))
1279                 nf_conntrack_cleanup_init_net();
1280 out_init_net:
1281         return ret;
1282 }