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 #include <linux/rculist_nulls.h>
  33
  34 #include <net/netfilter/nf_conntrack.h>
  35 #include <net/netfilter/nf_conntrack_l3proto.h>
  36 #include <net/netfilter/nf_conntrack_l4proto.h>
  37 #include <net/netfilter/nf_conntrack_expect.h>
  38 #include <net/netfilter/nf_conntrack_helper.h>
  39 #include <net/netfilter/nf_conntrack_core.h>
  40 #include <net/netfilter/nf_conntrack_extend.h>
  41 #include <net/netfilter/nf_conntrack_acct.h>
  42 #include <net/netfilter/nf_nat.h>
  43 #include <net/netfilter/nf_nat_core.h>
  44
  45 #define NF_CONNTRACK_VERSION    "0.5.0"
  46
  47 int (*nfnetlink_parse_nat_setup_hook)(struct nf_conn *ct,
  48                                       enum nf_nat_manip_type manip,
  49                                       struct nlattr *attr) __read_mostly;
  50 EXPORT_SYMBOL_GPL(nfnetlink_parse_nat_setup_hook);
  51
  52 DEFINE_SPINLOCK(nf_conntrack_lock);
  53 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
  54
  55 unsigned int nf_conntrack_htable_size __read_mostly;
  56 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
  57
  58 unsigned int nf_conntrack_max __read_mostly;
  59 EXPORT_SYMBOL_GPL(nf_conntrack_max);
  60
  61 struct nf_conn nf_conntrack_untracked __read_mostly;
  62 EXPORT_SYMBOL_GPL(nf_conntrack_untracked);
  63
  64 static struct kmem_cache *nf_conntrack_cachep __read_mostly;
  65
  66 static int nf_conntrack_hash_rnd_initted;
  67 static unsigned int nf_conntrack_hash_rnd;
  68
  69 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
  70                                   unsigned int size, unsigned int rnd)
  71 {
  72         unsigned int n;
  73         u_int32_t h;
  74
  75         /* The direction must be ignored, so we hash everything up to the
  76          * destination ports (which is a multiple of 4) and treat the last
  77          * three bytes manually.
  78          */
  79         n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
  80         h = jhash2((u32 *)tuple, n,
  81                    rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
  82                           tuple->dst.protonum));
  83
  84         return ((u64)h * size) >> 32;
  85 }
  86
  87 static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
  88 {
  89         return __hash_conntrack(tuple, nf_conntrack_htable_size,
  90                                 nf_conntrack_hash_rnd);
  91 }
  92
  93 bool
  94 nf_ct_get_tuple(const struct sk_buff *skb,
  95                 unsigned int nhoff,
  96                 unsigned int dataoff,
  97                 u_int16_t l3num,
  98                 u_int8_t protonum,
  99                 struct nf_conntrack_tuple *tuple,
 100                 const struct nf_conntrack_l3proto *l3proto,
 101                 const struct nf_conntrack_l4proto *l4proto)
 102 {
 103         memset(tuple, 0, sizeof(*tuple));
 104
 105         tuple->src.l3num = l3num;
 106         if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
 107                 return false;
 108
 109         tuple->dst.protonum = protonum;
 110         tuple->dst.dir = IP_CT_DIR_ORIGINAL;
 111
 112         return l4proto->pkt_to_tuple(skb, dataoff, tuple);
 113 }
 114 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 115
 116 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
 117                        u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 118 {
 119         struct nf_conntrack_l3proto *l3proto;
 120         struct nf_conntrack_l4proto *l4proto;
 121         unsigned int protoff;
 122         u_int8_t protonum;
 123         int ret;
 124
 125         rcu_read_lock();
 126
 127         l3proto = __nf_ct_l3proto_find(l3num);
 128         ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
 129         if (ret != NF_ACCEPT) {
 130                 rcu_read_unlock();
 131                 return false;
 132         }
 133
 134         l4proto = __nf_ct_l4proto_find(l3num, protonum);
 135
 136         ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
 137                               l3proto, l4proto);
 138
 139         rcu_read_unlock();
 140         return ret;
 141 }
 142 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 143
 144 bool
 145 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
 146                    const struct nf_conntrack_tuple *orig,
 147                    const struct nf_conntrack_l3proto *l3proto,
 148                    const struct nf_conntrack_l4proto *l4proto)
 149 {
 150         memset(inverse, 0, sizeof(*inverse));
 151
 152         inverse->src.l3num = orig->src.l3num;
 153         if (l3proto->invert_tuple(inverse, orig) == 0)
 154                 return false;
 155
 156         inverse->dst.dir = !orig->dst.dir;
 157
 158         inverse->dst.protonum = orig->dst.protonum;
 159         return l4proto->invert_tuple(inverse, orig);
 160 }
 161 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 162
 163 static void
 164 clean_from_lists(struct nf_conn *ct)
 165 {
 166         pr_debug("clean_from_lists(%p)\n", ct);
 167         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 168         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
 169
 170         /* Destroy all pending expectations */
 171         nf_ct_remove_expectations(ct);
 172 }
 173
 174 static void
 175 destroy_conntrack(struct nf_conntrack *nfct)
 176 {
 177         struct nf_conn *ct = (struct nf_conn *)nfct;
 178         struct net *net = nf_ct_net(ct);
 179         struct nf_conntrack_l4proto *l4proto;
 180
 181         pr_debug("destroy_conntrack(%p)\n", ct);
 182         NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
 183         NF_CT_ASSERT(!timer_pending(&ct->timeout));
 184
 185         if (!test_bit(IPS_DYING_BIT, &ct->status))
 186                 nf_conntrack_event(IPCT_DESTROY, ct);
 187         set_bit(IPS_DYING_BIT, &ct->status);
 188
 189         /* To make sure we don't get any weird locking issues here:
 190          * destroy_conntrack() MUST NOT be called with a write lock
 191          * to nf_conntrack_lock!!! -HW */
 192         rcu_read_lock();
 193         l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
 194         if (l4proto && l4proto->destroy)
 195                 l4proto->destroy(ct);
 196
 197         rcu_read_unlock();
 198
 199         spin_lock_bh(&nf_conntrack_lock);
 200         /* Expectations will have been removed in clean_from_lists,
 201          * except TFTP can create an expectation on the first packet,
 202          * before connection is in the list, so we need to clean here,
 203          * too. */
 204         nf_ct_remove_expectations(ct);
 205
 206         /* We overload first tuple to link into unconfirmed list. */
 207         if (!nf_ct_is_confirmed(ct)) {
 208                 BUG_ON(hlist_nulls_unhashed(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode));
 209                 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 210         }
 211
 212         NF_CT_STAT_INC(net, delete);
 213         spin_unlock_bh(&nf_conntrack_lock);
 214
 215         if (ct->master)
 216                 nf_ct_put(ct->master);
 217
 218         pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
 219         nf_conntrack_free(ct);
 220 }
 221
 222 static void death_by_timeout(unsigned long ul_conntrack)
 223 {
 224         struct nf_conn *ct = (void *)ul_conntrack;
 225         struct net *net = nf_ct_net(ct);
 226         struct nf_conn_help *help = nfct_help(ct);
 227         struct nf_conntrack_helper *helper;
 228
 229         if (help) {
 230                 rcu_read_lock();
 231                 helper = rcu_dereference(help->helper);
 232                 if (helper && helper->destroy)
 233                         helper->destroy(ct);
 234                 rcu_read_unlock();
 235         }
 236
 237         spin_lock_bh(&nf_conntrack_lock);
 238         /* Inside lock so preempt is disabled on module removal path.
 239          * Otherwise we can get spurious warnings. */
 240         NF_CT_STAT_INC(net, delete_list);
 241         clean_from_lists(ct);
 242         spin_unlock_bh(&nf_conntrack_lock);
 243         nf_ct_put(ct);
 244 }
 245
 246 /*
 247  * Warning :
 248  * - Caller must take a reference on returned object
 249  *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
 250  * OR
 251  * - Caller must lock nf_conntrack_lock before calling this function
 252  */
 253 struct nf_conntrack_tuple_hash *
 254 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 255 {
 256         struct nf_conntrack_tuple_hash *h;
 257         struct hlist_nulls_node *n;
 258         unsigned int hash = hash_conntrack(tuple);
 259
 260         /* Disable BHs the entire time since we normally need to disable them
 261          * at least once for the stats anyway.
 262          */
 263         local_bh_disable();
 264 begin:
 265         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 266                 if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 267                         NF_CT_STAT_INC(net, found);
 268                         local_bh_enable();
 269                         return h;
 270                 }
 271                 NF_CT_STAT_INC(net, searched);
 272         }
 273         /*
 274          * if the nulls value we got at the end of this lookup is
 275          * not the expected one, we must restart lookup.
 276          * We probably met an item that was moved to another chain.
 277          */
 278         if (get_nulls_value(n) != hash)
 279                 goto begin;
 280         local_bh_enable();
 281
 282         return NULL;
 283 }
 284 EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 285
 286 /* Find a connection corresponding to a tuple. */
 287 struct nf_conntrack_tuple_hash *
 288 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
 289 {
 290         struct nf_conntrack_tuple_hash *h;
 291         struct nf_conn *ct;
 292
 293         rcu_read_lock();
 294 begin:
 295         h = __nf_conntrack_find(net, tuple);
 296         if (h) {
 297                 ct = nf_ct_tuplehash_to_ctrack(h);
 298                 if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 299                         h = NULL;
 300                 else {
 301                         if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
 302                                 nf_ct_put(ct);
 303                                 goto begin;
 304                         }
 305                 }
 306         }
 307         rcu_read_unlock();
 308
 309         return h;
 310 }
 311 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
 312
 313 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
 314                                        unsigned int hash,
 315                                        unsigned int repl_hash)
 316 {
 317         struct net *net = nf_ct_net(ct);
 318
 319         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 320                            &net->ct.hash[hash]);
 321         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
 322                            &net->ct.hash[repl_hash]);
 323 }
 324
 325 void nf_conntrack_hash_insert(struct nf_conn *ct)
 326 {
 327         unsigned int hash, repl_hash;
 328
 329         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 330         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 331
 332         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 333 }
 334 EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
 335
 336 /* Confirm a connection given skb; places it in hash table */
 337 int
 338 __nf_conntrack_confirm(struct sk_buff *skb)
 339 {
 340         unsigned int hash, repl_hash;
 341         struct nf_conntrack_tuple_hash *h;
 342         struct nf_conn *ct;
 343         struct nf_conn_help *help;
 344         struct hlist_nulls_node *n;
 345         enum ip_conntrack_info ctinfo;
 346         struct net *net;
 347
 348         ct = nf_ct_get(skb, &ctinfo);
 349         net = nf_ct_net(ct);
 350
 351         /* ipt_REJECT uses nf_conntrack_attach to attach related
 352            ICMP/TCP RST packets in other direction.  Actual packet
 353            which created connection will be IP_CT_NEW or for an
 354            expected connection, IP_CT_RELATED. */
 355         if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
 356                 return NF_ACCEPT;
 357
 358         hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
 359         repl_hash = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 360
 361         /* We're not in hash table, and we refuse to set up related
 362            connections for unconfirmed conns.  But packet copies and
 363            REJECT will give spurious warnings here. */
 364         /* NF_CT_ASSERT(atomic_read(&ct->ct_general.use) == 1); */
 365
 366         /* No external references means noone else could have
 367            confirmed us. */
 368         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 369         pr_debug("Confirming conntrack %p\n", ct);
 370
 371         spin_lock_bh(&nf_conntrack_lock);
 372
 373         /* See if there's one in the list already, including reverse:
 374            NAT could have grabbed it without realizing, since we're
 375            not in the hash.  If there is, we lost race. */
 376         hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
 377                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
 378                                       &h->tuple))
 379                         goto out;
 380         hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
 381                 if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
 382                                       &h->tuple))
 383                         goto out;
 384
 385         /* Remove from unconfirmed list */
 386         hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
 387
 388         __nf_conntrack_hash_insert(ct, hash, repl_hash);
 389         /* Timer relative to confirmation time, not original
 390            setting time, otherwise we'd get timer wrap in
 391            weird delay cases. */
 392         ct->timeout.expires += jiffies;
 393         add_timer(&ct->timeout);
 394         atomic_inc(&ct->ct_general.use);
 395         set_bit(IPS_CONFIRMED_BIT, &ct->status);
 396         NF_CT_STAT_INC(net, insert);
 397         spin_unlock_bh(&nf_conntrack_lock);
 398         help = nfct_help(ct);
 399         if (help && help->helper)
 400                 nf_conntrack_event_cache(IPCT_HELPER, ct);
 401
 402         nf_conntrack_event_cache(master_ct(ct) ?
 403                                  IPCT_RELATED : IPCT_NEW, ct);
 404         return NF_ACCEPT;
 405
 406 out:
 407         NF_CT_STAT_INC(net, insert_failed);
 408         spin_unlock_bh(&nf_conntrack_lock);
 409         return NF_DROP;
 410 }
 411 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
 412
 413 /* Returns true if a connection correspondings to the tuple (required
 414    for NAT). */
 415 int
 416 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 417                          const struct nf_conn *ignored_conntrack)
 418 {
 419         struct net *net = nf_ct_net(ignored_conntrack);
 420         struct nf_conntrack_tuple_hash *h;
 421         struct hlist_nulls_node *n;
 422         unsigned int hash = hash_conntrack(tuple);
 423
 424         /* Disable BHs the entire time since we need to disable them at
 425          * least once for the stats anyway.
 426          */
 427         rcu_read_lock_bh();
 428         hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 429                 if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
 430                     nf_ct_tuple_equal(tuple, &h->tuple)) {
 431                         NF_CT_STAT_INC(net, found);
 432                         rcu_read_unlock_bh();
 433                         return 1;
 434                 }
 435                 NF_CT_STAT_INC(net, searched);
 436         }
 437         rcu_read_unlock_bh();
 438
 439         return 0;
 440 }
 441 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 442
 443 #define NF_CT_EVICTION_RANGE    8
 444
 445 /* There's a small race here where we may free a just-assured
 446    connection.  Too bad: we're in trouble anyway. */
 447 static noinline int early_drop(struct net *net, unsigned int hash)
 448 {
 449         /* Use oldest entry, which is roughly LRU */
 450         struct nf_conntrack_tuple_hash *h;
 451         struct nf_conn *ct = NULL, *tmp;
 452         struct hlist_nulls_node *n;
 453         unsigned int i, cnt = 0;
 454         int dropped = 0;
 455
 456         rcu_read_lock();
 457         for (i = 0; i < nf_conntrack_htable_size; i++) {
 458                 hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
 459                                          hnnode) {
 460                         tmp = nf_ct_tuplehash_to_ctrack(h);
 461                         if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
 462                                 ct = tmp;
 463                         cnt++;
 464                 }
 465
 466                 if (ct && unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 467                         ct = NULL;
 468                 if (ct || cnt >= NF_CT_EVICTION_RANGE)
 469                         break;
 470                 hash = (hash + 1) % nf_conntrack_htable_size;
 471         }
 472         rcu_read_unlock();
 473
 474         if (!ct)
 475                 return dropped;
 476
 477         if (del_timer(&ct->timeout)) {
 478                 death_by_timeout((unsigned long)ct);
 479                 dropped = 1;
 480                 NF_CT_STAT_INC_ATOMIC(net, early_drop);
 481         }
 482         nf_ct_put(ct);
 483         return dropped;
 484 }
 485
 486 struct nf_conn *nf_conntrack_alloc(struct net *net,
 487                                    const struct nf_conntrack_tuple *orig,
 488                                    const struct nf_conntrack_tuple *repl,
 489                                    gfp_t gfp)
 490 {
 491         struct nf_conn *ct;
 492
 493         if (unlikely(!nf_conntrack_hash_rnd_initted)) {
 494                 get_random_bytes(&nf_conntrack_hash_rnd,
 495                                 sizeof(nf_conntrack_hash_rnd));
 496                 nf_conntrack_hash_rnd_initted = 1;
 497         }
 498
 499         /* We don't want any race condition at early drop stage */
 500         atomic_inc(&net->ct.count);
 501
 502         if (nf_conntrack_max &&
 503             unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
 504                 unsigned int hash = hash_conntrack(orig);
 505                 if (!early_drop(net, hash)) {
 506                         atomic_dec(&net->ct.count);
 507                         if (net_ratelimit())
 508                                 printk(KERN_WARNING
 509                                        "nf_conntrack: table full, dropping"
 510                                        " packet.\n");
 511                         return ERR_PTR(-ENOMEM);
 512                 }
 513         }
 514
 515         ct = kmem_cache_zalloc(nf_conntrack_cachep, gfp);
 516         if (ct == NULL) {
 517                 pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
 518                 atomic_dec(&net->ct.count);
 519                 return ERR_PTR(-ENOMEM);
 520         }
 521
 522         spin_lock_init(&ct->lock);
 523         atomic_set(&ct->ct_general.use, 1);
 524         ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
 525         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
 526         /* Don't set timer yet: wait for confirmation */
 527         setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
 528 #ifdef CONFIG_NET_NS
 529         ct->ct_net = net;
 530 #endif
 531
 532         return ct;
 533 }
 534 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 535
 536 void nf_conntrack_free(struct nf_conn *ct)
 537 {
 538         struct net *net = nf_ct_net(ct);
 539
 540         nf_ct_ext_destroy(ct);
 541         atomic_dec(&net->ct.count);
 542         nf_ct_ext_free(ct);
 543         kmem_cache_free(nf_conntrack_cachep, ct);
 544 }
 545 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 546
 547 /* Allocate a new conntrack: we return -ENOMEM if classification
 548    failed due to stress.  Otherwise it really is unclassifiable. */
 549 static struct nf_conntrack_tuple_hash *
 550 init_conntrack(struct net *net,
 551                const struct nf_conntrack_tuple *tuple,
 552                struct nf_conntrack_l3proto *l3proto,
 553                struct nf_conntrack_l4proto *l4proto,
 554                struct sk_buff *skb,
 555                unsigned int dataoff)
 556 {
 557         struct nf_conn *ct;
 558         struct nf_conn_help *help;
 559         struct nf_conntrack_tuple repl_tuple;
 560         struct nf_conntrack_expect *exp;
 561
 562         if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
 563                 pr_debug("Can't invert tuple.\n");
 564                 return NULL;
 565         }
 566
 567         ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
 568         if (IS_ERR(ct)) {
 569                 pr_debug("Can't allocate conntrack.\n");
 570                 return (struct nf_conntrack_tuple_hash *)ct;
 571         }
 572
 573         if (!l4proto->new(ct, skb, dataoff)) {
 574                 nf_conntrack_free(ct);
 575                 pr_debug("init conntrack: can't track with proto module\n");
 576                 return NULL;
 577         }
 578
 579         nf_ct_acct_ext_add(ct, GFP_ATOMIC);
 580
 581         spin_lock_bh(&nf_conntrack_lock);
 582         exp = nf_ct_find_expectation(net, tuple);
 583         if (exp) {
 584                 pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
 585                          ct, exp);
 586                 /* Welcome, Mr. Bond.  We've been expecting you... */
 587                 __set_bit(IPS_EXPECTED_BIT, &ct->status);
 588                 ct->master = exp->master;
 589                 if (exp->helper) {
 590                         help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
 591                         if (help)
 592                                 rcu_assign_pointer(help->helper, exp->helper);
 593                 }
 594
 595 #ifdef CONFIG_NF_CONNTRACK_MARK
 596                 ct->mark = exp->master->mark;
 597 #endif
 598 #ifdef CONFIG_NF_CONNTRACK_SECMARK
 599                 ct->secmark = exp->master->secmark;
 600 #endif
 601                 nf_conntrack_get(&ct->master->ct_general);
 602                 NF_CT_STAT_INC(net, expect_new);
 603         } else {
 604                 __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 605                 NF_CT_STAT_INC(net, new);
 606         }
 607
 608         /* Overload tuple linked list to put us in unconfirmed list. */
 609         hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
 610                        &net->ct.unconfirmed);
 611
 612         spin_unlock_bh(&nf_conntrack_lock);
 613
 614         if (exp) {
 615                 if (exp->expectfn)
 616                         exp->expectfn(ct, exp);
 617                 nf_ct_expect_put(exp);
 618         }
 619
 620         return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 621 }
 622
 623 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
 624 static inline struct nf_conn *
 625 resolve_normal_ct(struct net *net,
 626                   struct sk_buff *skb,
 627                   unsigned int dataoff,
 628                   u_int16_t l3num,
 629                   u_int8_t protonum,
 630                   struct nf_conntrack_l3proto *l3proto,
 631                   struct nf_conntrack_l4proto *l4proto,
 632                   int *set_reply,
 633                   enum ip_conntrack_info *ctinfo)
 634 {
 635         struct nf_conntrack_tuple tuple;
 636         struct nf_conntrack_tuple_hash *h;
 637         struct nf_conn *ct;
 638
 639         if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
 640                              dataoff, l3num, protonum, &tuple, l3proto,
 641                              l4proto)) {
 642                 pr_debug("resolve_normal_ct: Can't get tuple\n");
 643                 return NULL;
 644         }
 645
 646         /* look for tuple match */
 647         h = nf_conntrack_find_get(net, &tuple);
 648         if (!h) {
 649                 h = init_conntrack(net, &tuple, l3proto, l4proto, skb, dataoff);
 650                 if (!h)
 651                         return NULL;
 652                 if (IS_ERR(h))
 653                         return (void *)h;
 654         }
 655         ct = nf_ct_tuplehash_to_ctrack(h);
 656
 657         /* It exists; we have (non-exclusive) reference. */
 658         if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
 659                 *ctinfo = IP_CT_ESTABLISHED + IP_CT_IS_REPLY;
 660                 /* Please set reply bit if this packet OK */
 661                 *set_reply = 1;
 662         } else {
 663                 /* Once we've had two way comms, always ESTABLISHED. */
 664                 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 665                         pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
 666                         *ctinfo = IP_CT_ESTABLISHED;
 667                 } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
 668                         pr_debug("nf_conntrack_in: related packet for %p\n",
 669                                  ct);
 670                         *ctinfo = IP_CT_RELATED;
 671                 } else {
 672                         pr_debug("nf_conntrack_in: new packet for %p\n", ct);
 673                         *ctinfo = IP_CT_NEW;
 674                 }
 675                 *set_reply = 0;
 676         }
 677         skb->nfct = &ct->ct_general;
 678         skb->nfctinfo = *ctinfo;
 679         return ct;
 680 }
 681
 682 unsigned int
 683 nf_conntrack_in(struct net *net, u_int8_t pf, unsigned int hooknum,
 684                 struct sk_buff *skb)
 685 {
 686         struct nf_conn *ct;
 687         enum ip_conntrack_info ctinfo;
 688         struct nf_conntrack_l3proto *l3proto;
 689         struct nf_conntrack_l4proto *l4proto;
 690         unsigned int dataoff;
 691         u_int8_t protonum;
 692         int set_reply = 0;
 693         int ret;
 694
 695         /* Previously seen (loopback or untracked)?  Ignore. */
 696         if (skb->nfct) {
 697                 NF_CT_STAT_INC_ATOMIC(net, ignore);
 698                 return NF_ACCEPT;
 699         }
 700
 701         /* rcu_read_lock()ed by nf_hook_slow */
 702         l3proto = __nf_ct_l3proto_find(pf);
 703         ret = l3proto->get_l4proto(skb, skb_network_offset(skb),
 704                                    &dataoff, &protonum);
 705         if (ret <= 0) {
 706                 pr_debug("not prepared to track yet or error occured\n");
 707                 NF_CT_STAT_INC_ATOMIC(net, error);
 708                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 709                 return -ret;
 710         }
 711
 712         l4proto = __nf_ct_l4proto_find(pf, protonum);
 713
 714         /* It may be an special packet, error, unclean...
 715          * inverse of the return code tells to the netfilter
 716          * core what to do with the packet. */
 717         if (l4proto->error != NULL) {
 718                 ret = l4proto->error(net, skb, dataoff, &ctinfo, pf, hooknum);
 719                 if (ret <= 0) {
 720                         NF_CT_STAT_INC_ATOMIC(net, error);
 721                         NF_CT_STAT_INC_ATOMIC(net, invalid);
 722                         return -ret;
 723                 }
 724         }
 725
 726         ct = resolve_normal_ct(net, skb, dataoff, pf, protonum,
 727                                l3proto, l4proto, &set_reply, &ctinfo);
 728         if (!ct) {
 729                 /* Not valid part of a connection */
 730                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 731                 return NF_ACCEPT;
 732         }
 733
 734         if (IS_ERR(ct)) {
 735                 /* Too stressed to deal. */
 736                 NF_CT_STAT_INC_ATOMIC(net, drop);
 737                 return NF_DROP;
 738         }
 739
 740         NF_CT_ASSERT(skb->nfct);
 741
 742         ret = l4proto->packet(ct, skb, dataoff, ctinfo, pf, hooknum);
 743         if (ret <= 0) {
 744                 /* Invalid: inverse of the return code tells
 745                  * the netfilter core what to do */
 746                 pr_debug("nf_conntrack_in: Can't track with proto module\n");
 747                 nf_conntrack_put(skb->nfct);
 748                 skb->nfct = NULL;
 749                 NF_CT_STAT_INC_ATOMIC(net, invalid);
 750                 if (ret == -NF_DROP)
 751                         NF_CT_STAT_INC_ATOMIC(net, drop);
 752                 return -ret;
 753         }
 754
 755         if (set_reply && !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
 756                 nf_conntrack_event_cache(IPCT_STATUS, ct);
 757
 758         return ret;
 759 }
 760 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 761
 762 bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
 763                           const struct nf_conntrack_tuple *orig)
 764 {
 765         bool ret;
 766
 767         rcu_read_lock();
 768         ret = nf_ct_invert_tuple(inverse, orig,
 769                                  __nf_ct_l3proto_find(orig->src.l3num),
 770                                  __nf_ct_l4proto_find(orig->src.l3num,
 771                                                       orig->dst.protonum));
 772         rcu_read_unlock();
 773         return ret;
 774 }
 775 EXPORT_SYMBOL_GPL(nf_ct_invert_tuplepr);
 776
 777 /* Alter reply tuple (maybe alter helper).  This is for NAT, and is
 778    implicitly racy: see __nf_conntrack_confirm */
 779 void nf_conntrack_alter_reply(struct nf_conn *ct,
 780                               const struct nf_conntrack_tuple *newreply)
 781 {
 782         struct nf_conn_help *help = nfct_help(ct);
 783
 784         /* Should be unconfirmed, so not in hash table yet */
 785         NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 786
 787         pr_debug("Altering reply tuple of %p to ", ct);
 788         nf_ct_dump_tuple(newreply);
 789
 790         ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
 791         if (ct->master || (help && !hlist_empty(&help->expectations)))
 792                 return;
 793
 794         rcu_read_lock();
 795         __nf_ct_try_assign_helper(ct, GFP_ATOMIC);
 796         rcu_read_unlock();
 797 }
 798 EXPORT_SYMBOL_GPL(nf_conntrack_alter_reply);
 799
 800 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 801 void __nf_ct_refresh_acct(struct nf_conn *ct,
 802                           enum ip_conntrack_info ctinfo,
 803                           const struct sk_buff *skb,
 804                           unsigned long extra_jiffies,
 805                           int do_acct)
 806 {
 807         NF_CT_ASSERT(ct->timeout.data == (unsigned long)ct);
 808         NF_CT_ASSERT(skb);
 809
 810         spin_lock_bh(&nf_conntrack_lock);
 811
 812         /* Only update if this is not a fixed timeout */
 813         if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
 814                 goto acct;
 815
 816         /* If not in hash table, timer will not be active yet */
 817         if (!nf_ct_is_confirmed(ct)) {
 818                 ct->timeout.expires = extra_jiffies;
 819         } else {
 820                 unsigned long newtime = jiffies + extra_jiffies;
 821
 822                 /* Only update the timeout if the new timeout is at least
 823                    HZ jiffies from the old timeout. Need del_timer for race
 824                    avoidance (may already be dying). */
 825                 if (newtime - ct->timeout.expires >= HZ
 826                     && del_timer(&ct->timeout)) {
 827                         ct->timeout.expires = newtime;
 828                         add_timer(&ct->timeout);
 829                 }
 830         }
 831
 832 acct:
 833         if (do_acct) {
 834                 struct nf_conn_counter *acct;
 835
 836                 acct = nf_conn_acct_find(ct);
 837                 if (acct) {
 838                         acct[CTINFO2DIR(ctinfo)].packets++;
 839                         acct[CTINFO2DIR(ctinfo)].bytes +=
 840                                 skb->len - skb_network_offset(skb);
 841                 }
 842         }
 843
 844         spin_unlock_bh(&nf_conntrack_lock);
 845 }
 846 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 847
 848 bool __nf_ct_kill_acct(struct nf_conn *ct,
 849                        enum ip_conntrack_info ctinfo,
 850                        const struct sk_buff *skb,
 851                        int do_acct)
 852 {
 853         if (do_acct) {
 854                 struct nf_conn_counter *acct;
 855
 856                 spin_lock_bh(&nf_conntrack_lock);
 857                 acct = nf_conn_acct_find(ct);
 858                 if (acct) {
 859                         acct[CTINFO2DIR(ctinfo)].packets++;
 860                         acct[CTINFO2DIR(ctinfo)].bytes +=
 861                                 skb->len - skb_network_offset(skb);
 862                 }
 863                 spin_unlock_bh(&nf_conntrack_lock);
 864         }
 865
 866         if (del_timer(&ct->timeout)) {
 867                 ct->timeout.function((unsigned long)ct);
 868                 return true;
 869         }
 870         return false;
 871 }
 872 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 873
 874 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 875
 876 #include <linux/netfilter/nfnetlink.h>
 877 #include <linux/netfilter/nfnetlink_conntrack.h>
 878 #include <linux/mutex.h>
 879
 880 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
 881  * in ip_conntrack_core, since we don't want the protocols to autoload
 882  * or depend on ctnetlink */
 883 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
 884                                const struct nf_conntrack_tuple *tuple)
 885 {
 886         NLA_PUT_BE16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port);
 887         NLA_PUT_BE16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port);
 888         return 0;
 889
 890 nla_put_failure:
 891         return -1;
 892 }
 893 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 894
 895 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
 896         [CTA_PROTO_SRC_PORT]  = { .type = NLA_U16 },
 897         [CTA_PROTO_DST_PORT]  = { .type = NLA_U16 },
 898 };
 899 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
 900
 901 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
 902                                struct nf_conntrack_tuple *t)
 903 {
 904         if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
 905                 return -EINVAL;
 906
 907         t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
 908         t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
 909
 910         return 0;
 911 }
 912 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 913
 914 int nf_ct_port_nlattr_tuple_size(void)
 915 {
 916         return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
 917 }
 918 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
 919 #endif
 920
 921 /* Used by ipt_REJECT and ip6t_REJECT. */
 922 static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
 923 {
 924         struct nf_conn *ct;
 925         enum ip_conntrack_info ctinfo;
 926
 927         /* This ICMP is in reverse direction to the packet which caused it */
 928         ct = nf_ct_get(skb, &ctinfo);
 929         if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
 930                 ctinfo = IP_CT_RELATED + IP_CT_IS_REPLY;
 931         else
 932                 ctinfo = IP_CT_RELATED;
 933
 934         /* Attach to new skbuff, and increment count */
 935         nskb->nfct = &ct->ct_general;
 936         nskb->nfctinfo = ctinfo;
 937         nf_conntrack_get(nskb->nfct);
 938 }
 939
 940 /* Bring out ya dead! */
 941 static struct nf_conn *
 942 get_next_corpse(struct net *net, int (*iter)(struct nf_conn *i, void *data),
 943                 void *data, unsigned int *bucket)
 944 {
 945         struct nf_conntrack_tuple_hash *h;
 946         struct nf_conn *ct;
 947         struct hlist_nulls_node *n;
 948
 949         spin_lock_bh(&nf_conntrack_lock);
 950         for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
 951                 hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
 952                         ct = nf_ct_tuplehash_to_ctrack(h);
 953                         if (iter(ct, data))
 954                                 goto found;
 955                 }
 956         }
 957         hlist_nulls_for_each_entry(h, n, &net->ct.unconfirmed, hnnode) {
 958                 ct = nf_ct_tuplehash_to_ctrack(h);
 959                 if (iter(ct, data))
 960                         set_bit(IPS_DYING_BIT, &ct->status);
 961         }
 962         spin_unlock_bh(&nf_conntrack_lock);
 963         return NULL;
 964 found:
 965         atomic_inc(&ct->ct_general.use);
 966         spin_unlock_bh(&nf_conntrack_lock);
 967         return ct;
 968 }
 969
 970 void nf_ct_iterate_cleanup(struct net *net,
 971                            int (*iter)(struct nf_conn *i, void *data),
 972                            void *data)
 973 {
 974         struct nf_conn *ct;
 975         unsigned int bucket = 0;
 976
 977         while ((ct = get_next_corpse(net, iter, data, &bucket)) != NULL) {
 978                 /* Time to push up daises... */
 979                 if (del_timer(&ct->timeout))
 980                         death_by_timeout((unsigned long)ct);
 981                 /* ... else the timer will get him soon. */
 982
 983                 nf_ct_put(ct);
 984         }
 985 }
 986 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup);
 987
 988 struct __nf_ct_flush_report {
 989         u32 pid;
 990         int report;
 991 };
 992
 993 static int kill_report(struct nf_conn *i, void *data)
 994 {
 995         struct __nf_ct_flush_report *fr = (struct __nf_ct_flush_report *)data;
 996
 997         /* get_next_corpse sets the dying bit for us */
 998         nf_conntrack_event_report(IPCT_DESTROY,
 999                                   i,
1000                                   fr->pid,
1001                                   fr->report);
1002         return 1;
1003 }
1004
1005 static int kill_all(struct nf_conn *i, void *data)
1006 {
1007         return 1;
1008 }
1009
1010 void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size)
1011 {
1012         if (vmalloced)
1013                 vfree(hash);
1014         else
1015                 free_pages((unsigned long)hash,
1016                            get_order(sizeof(struct hlist_head) * size));
1017 }
1018 EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
1019
1020 void nf_conntrack_flush_report(struct net *net, u32 pid, int report)
1021 {
1022         struct __nf_ct_flush_report fr = {
1023                 .pid    = pid,
1024                 .report = report,
1025         };
1026         nf_ct_iterate_cleanup(net, kill_report, &fr);
1027 }
1028 EXPORT_SYMBOL_GPL(nf_conntrack_flush_report);
1029
1030 static void nf_conntrack_cleanup_init_net(void)
1031 {
1032         nf_conntrack_helper_fini();
1033         nf_conntrack_proto_fini();
1034         kmem_cache_destroy(nf_conntrack_cachep);
1035 }
1036
1037 static void nf_conntrack_cleanup_net(struct net *net)
1038 {
1039         nf_ct_event_cache_flush(net);
1040         nf_conntrack_ecache_fini(net);
1041  i_see_dead_people:
1042         nf_ct_iterate_cleanup(net, kill_all, NULL);
1043         if (atomic_read(&net->ct.count) != 0) {
1044                 schedule();
1045                 goto i_see_dead_people;
1046         }
1047         /* wait until all references to nf_conntrack_untracked are dropped */
1048         while (atomic_read(&nf_conntrack_untracked.ct_general.use) > 1)
1049                 schedule();
1050
1051         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1052                              nf_conntrack_htable_size);
1053         nf_conntrack_acct_fini(net);
1054         nf_conntrack_expect_fini(net);
1055         free_percpu(net->ct.stat);
1056 }
1057
1058 /* Mishearing the voices in his head, our hero wonders how he's
1059    supposed to kill the mall. */
1060 void nf_conntrack_cleanup(struct net *net)
1061 {
1062         if (net_eq(net, &init_net))
1063                 rcu_assign_pointer(ip_ct_attach, NULL);
1064
1065         /* This makes sure all current packets have passed through
1066            netfilter framework.  Roll on, two-stage module
1067            delete... */
1068         synchronize_net();
1069
1070         nf_conntrack_cleanup_net(net);
1071
1072         if (net_eq(net, &init_net)) {
1073                 rcu_assign_pointer(nf_ct_destroy, NULL);
1074                 nf_conntrack_cleanup_init_net();
1075         }
1076 }
1077
1078 void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls)
1079 {
1080         struct hlist_nulls_head *hash;
1081         unsigned int nr_slots, i;
1082         size_t sz;
1083
1084         *vmalloced = 0;
1085
1086         BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
1087         nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
1088         sz = nr_slots * sizeof(struct hlist_nulls_head);
1089         hash = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO,
1090                                         get_order(sz));
1091         if (!hash) {
1092                 *vmalloced = 1;
1093                 printk(KERN_WARNING "nf_conntrack: falling back to vmalloc.\n");
1094                 hash = __vmalloc(sz, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
1095         }
1096
1097         if (hash && nulls)
1098                 for (i = 0; i < nr_slots; i++)
1099                         INIT_HLIST_NULLS_HEAD(&hash[i], i);
1100
1101         return hash;
1102 }
1103 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
1104
1105 int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
1106 {
1107         int i, bucket, vmalloced, old_vmalloced;
1108         unsigned int hashsize, old_size;
1109         int rnd;
1110         struct hlist_nulls_head *hash, *old_hash;
1111         struct nf_conntrack_tuple_hash *h;
1112
1113         /* On boot, we can set this without any fancy locking. */
1114         if (!nf_conntrack_htable_size)
1115                 return param_set_uint(val, kp);
1116
1117         hashsize = simple_strtoul(val, NULL, 0);
1118         if (!hashsize)
1119                 return -EINVAL;
1120
1121         hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced, 1);
1122         if (!hash)
1123                 return -ENOMEM;
1124
1125         /* We have to rehahs for the new table anyway, so we also can
1126          * use a newrandom seed */
1127         get_random_bytes(&rnd, sizeof(rnd));
1128
1129         /* Lookups in the old hash might happen in parallel, which means we
1130          * might get false negatives during connection lookup. New connections
1131          * created because of a false negative won't make it into the hash
1132          * though since that required taking the lock.
1133          */
1134         spin_lock_bh(&nf_conntrack_lock);
1135         for (i = 0; i < nf_conntrack_htable_size; i++) {
1136                 while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
1137                         h = hlist_nulls_entry(init_net.ct.hash[i].first,
1138                                         struct nf_conntrack_tuple_hash, hnnode);
1139                         hlist_nulls_del_rcu(&h->hnnode);
1140                         bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
1141                         hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
1142                 }
1143         }
1144         old_size = nf_conntrack_htable_size;
1145         old_vmalloced = init_net.ct.hash_vmalloc;
1146         old_hash = init_net.ct.hash;
1147
1148         nf_conntrack_htable_size = hashsize;
1149         init_net.ct.hash_vmalloc = vmalloced;
1150         init_net.ct.hash = hash;
1151         nf_conntrack_hash_rnd = rnd;
1152         spin_unlock_bh(&nf_conntrack_lock);
1153
1154         nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
1155         return 0;
1156 }
1157 EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
1158
1159 module_param_call(hashsize, nf_conntrack_set_hashsize, param_get_uint,
1160                   &nf_conntrack_htable_size, 0600);
1161
1162 static int nf_conntrack_init_init_net(void)
1163 {
1164         int max_factor = 8;
1165         int ret;
1166
1167         /* Idea from tcp.c: use 1/16384 of memory.  On i386: 32MB
1168          * machine has 512 buckets. >= 1GB machines have 16384 buckets. */
1169         if (!nf_conntrack_htable_size) {
1170                 nf_conntrack_htable_size
1171                         = (((num_physpages << PAGE_SHIFT) / 16384)
1172                            / sizeof(struct hlist_head));
1173                 if (num_physpages > (1024 * 1024 * 1024 / PAGE_SIZE))
1174                         nf_conntrack_htable_size = 16384;
1175                 if (nf_conntrack_htable_size < 32)
1176                         nf_conntrack_htable_size = 32;
1177
1178                 /* Use a max. factor of four by default to get the same max as
1179                  * with the old struct list_heads. When a table size is given
1180                  * we use the old value of 8 to avoid reducing the max.
1181                  * entries. */
1182                 max_factor = 4;
1183         }
1184         nf_conntrack_max = max_factor * nf_conntrack_htable_size;
1185
1186         printk("nf_conntrack version %s (%u buckets, %d max)\n",
1187                NF_CONNTRACK_VERSION, nf_conntrack_htable_size,
1188                nf_conntrack_max);
1189
1190         nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
1191                                                 sizeof(struct nf_conn),
1192                                                 0, SLAB_DESTROY_BY_RCU, NULL);
1193         if (!nf_conntrack_cachep) {
1194                 printk(KERN_ERR "Unable to create nf_conn slab cache\n");
1195                 ret = -ENOMEM;
1196                 goto err_cache;
1197         }
1198
1199         ret = nf_conntrack_proto_init();
1200         if (ret < 0)
1201                 goto err_proto;
1202
1203         ret = nf_conntrack_helper_init();
1204         if (ret < 0)
1205                 goto err_helper;
1206
1207         return 0;
1208
1209 err_helper:
1210         nf_conntrack_proto_fini();
1211 err_proto:
1212         kmem_cache_destroy(nf_conntrack_cachep);
1213 err_cache:
1214         return ret;
1215 }
1216
1217 static int nf_conntrack_init_net(struct net *net)
1218 {
1219         int ret;
1220
1221         atomic_set(&net->ct.count, 0);
1222         INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, 0);
1223         net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
1224         if (!net->ct.stat) {
1225                 ret = -ENOMEM;
1226                 goto err_stat;
1227         }
1228         ret = nf_conntrack_ecache_init(net);
1229         if (ret < 0)
1230                 goto err_ecache;
1231         net->ct.hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
1232                                              &net->ct.hash_vmalloc, 1);
1233         if (!net->ct.hash) {
1234                 ret = -ENOMEM;
1235                 printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
1236                 goto err_hash;
1237         }
1238         ret = nf_conntrack_expect_init(net);
1239         if (ret < 0)
1240                 goto err_expect;
1241         ret = nf_conntrack_acct_init(net);
1242         if (ret < 0)
1243                 goto err_acct;
1244
1245         /* Set up fake conntrack:
1246             - to never be deleted, not in any hashes */
1247 #ifdef CONFIG_NET_NS
1248         nf_conntrack_untracked.ct_net = &init_net;
1249 #endif
1250         atomic_set(&nf_conntrack_untracked.ct_general.use, 1);
1251         /*  - and look it like as a confirmed connection */
1252         set_bit(IPS_CONFIRMED_BIT, &nf_conntrack_untracked.status);
1253
1254         return 0;
1255
1256 err_acct:
1257         nf_conntrack_expect_fini(net);
1258 err_expect:
1259         nf_ct_free_hashtable(net->ct.hash, net->ct.hash_vmalloc,
1260                              nf_conntrack_htable_size);
1261 err_hash:
1262         nf_conntrack_ecache_fini(net);
1263 err_ecache:
1264         free_percpu(net->ct.stat);
1265 err_stat:
1266         return ret;
1267 }
1268
1269 int nf_conntrack_init(struct net *net)
1270 {
1271         int ret;
1272
1273         if (net_eq(net, &init_net)) {
1274                 ret = nf_conntrack_init_init_net();
1275                 if (ret < 0)
1276                         goto out_init_net;
1277         }
1278         ret = nf_conntrack_init_net(net);
1279         if (ret < 0)
1280                 goto out_net;
1281
1282         if (net_eq(net, &init_net)) {
1283                 /* For use by REJECT target */
1284                 rcu_assign_pointer(ip_ct_attach, nf_conntrack_attach);
1285                 rcu_assign_pointer(nf_ct_destroy, destroy_conntrack);
1286         }
1287         return 0;
1288
1289 out_net:
1290         if (net_eq(net, &init_net))
1291                 nf_conntrack_cleanup_init_net();
1292 out_init_net:
1293         return ret;
1294 }