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